[SDAG] Fold umul_lohi with 0 or 1 multiplicand

[nikic]

Summary

These can turn up during multiplication legalization. In principle these should also apply to smul_lohi, but I wasn't able to figure out how to produce those.

[RKSimon]

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

4292 ↗

(On Diff #215748)

Do we need a canonicalization of constants to the RHS?

[nikic]

llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

4292 ↗

(On Diff #215748)

Possibly, but no test case where this would be needed comes to mind. If we start from canonical IR we have constants on the RHS anyway, but even if we start from non-canonical IR, the constant will be moved to the RHS for ISD::MUL, before we get here. Possibly it could happen via UMULO, though in that case we should probably already canonicalize while it is a UMULO.

[RKSimon]

OK, thanks for checking - LGTM

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

 
 SDValue DAGCombiner::visitUMUL_LOHI(SDNode *N) {
   if (SDValue Res = SimplifyNodeWithTwoResults(N, ISD::MUL, ISD::MULHU))
     return Res;
 
   EVT VT = N->getValueType(0);
   SDLoc DL(N);
 
+  // (umul_lohi N0, 0) -> (0, 0)
+  if (isNullConstant(N->getOperand(1))) {
+    SDValue Zero = DAG.getConstant(0, DL, VT);
+    return CombineTo(N, Zero, Zero);
+  }
+
+  // (umul_lohi N0, 1) -> (N0, 0)
+  if (isOneConstant(N->getOperand(1))) {
+    SDValue Zero = DAG.getConstant(0, DL, VT);
+    return CombineTo(N, N->getOperand(0), Zero);
+  }
+
   // If the type is twice as wide is legal, transform the mulhu to a wider
   // multiply plus a shift.
   if (VT.isSimple() && !VT.isVector()) {
     MVT Simple = VT.getSimpleVT();
     unsigned SimpleSize = Simple.getSizeInBits();
     EVT NewVT = EVT::getIntegerVT(*DAG.getContext(), SimpleSize*2);
     if (TLI.isOperationLegal(ISD::MUL, NewVT)) {
       SDValue Lo = DAG.getNode(ISD::ZERO_EXTEND, DL, NewVT, N->getOperand(0));

llvm/trunk/test/CodeGen/X86/i128-mul.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefixes=X86,X86-NOBMI
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+bmi2 | FileCheck %s --check-prefixes=X86,X86-BMI
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefixes=X64,X64-NOBMI
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+bmi2 | FileCheck %s --check-prefixes=X64,X64-BMI
 
 ; PR1198
 
 define i64 @foo(i64 %x, i64 %y) nounwind {
 ; X86-LABEL: foo:
 ; X86:       # %bb.0:
 ; X86-NEXT:    pushl %ebp
 ; X86-NEXT:    pushl %ebx
 ; X86-NEXT:    pushl %edi
 ; X86-NEXT:    pushl %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    movl %ecx, %eax
-; X86-NEXT:    mull %ebp
+; X86-NEXT:    mull %ebx
+; X86-NEXT:    movl %edx, %edi
+; X86-NEXT:    movl %ebp, %eax
+; X86-NEXT:    mull %ebx
 ; X86-NEXT:    movl %edx, %ebx
-; X86-NEXT:    movl %esi, %eax
-; X86-NEXT:    mull %ebp
-; X86-NEXT:    movl %edx, %ebp
-; X86-NEXT:    movl %eax, %esi
-; X86-NEXT:    addl %ebx, %esi
-; X86-NEXT:    adcl $0, %ebp
+; X86-NEXT:    movl %eax, %ebp
+; X86-NEXT:    addl %edi, %ebp
+; X86-NEXT:    adcl $0, %ebx
 ; X86-NEXT:    movl %ecx, %eax
-; X86-NEXT:    mull %edi
-; X86-NEXT:    movl %edx, %ebx
-; X86-NEXT:    addl %esi, %eax
-; X86-NEXT:    adcl %ebp, %ebx
+; X86-NEXT:    mull %esi
+; X86-NEXT:    movl %edx, %ecx
+; X86-NEXT:    addl %ebp, %eax
+; X86-NEXT:    adcl %ebx, %ecx
 ; X86-NEXT:    setb %al
-; X86-NEXT:    movzbl %al, %ecx
+; X86-NEXT:    movzbl %al, %edi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %edi
-; X86-NEXT:    movl %edx, %esi
-; X86-NEXT:    movl %eax, %ebp
-; X86-NEXT:    addl %ebx, %ebp
-; X86-NEXT:    adcl %ecx, %esi
-; X86-NEXT:    xorl %ecx, %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %ecx
-; X86-NEXT:    movl %edx, %edi
-; X86-NEXT:    movl %eax, %ebx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %ecx
-; X86-NEXT:    addl %ebx, %eax
+; X86-NEXT:    mull %esi
+; X86-NEXT:    addl %ecx, %eax
 ; X86-NEXT:    adcl %edi, %edx
-; X86-NEXT:    addl %ebp, %eax
-; X86-NEXT:    adcl %esi, %edx
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: foo:
 ; X64:       # %bb.0:

 
 define i64 @mul1(i64 %n, i64* nocapture %z, i64* nocapture %x, i64 %y) nounwind {
 ; X86-LABEL: mul1:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    pushl %ebp
 ; X86-NEXT:    pushl %ebx
 ; X86-NEXT:    pushl %edi
 ; X86-NEXT:    pushl %esi
-; X86-NEXT:    subl $28, %esp
+; X86-NEXT:    subl $24, %esp
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    orl %ecx, %eax
 ; X86-NEXT:    je .LBB1_3
 ; X86-NEXT:  # %bb.1: # %for.body.preheader
 ; X86-NEXT:    xorl %eax, %eax
 ; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    xorl %ebx, %ebx
-; X86-NEXT:    movl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X86-NEXT:    xorl %ebp, %ebp
+; X86-NEXT:    movl $0, (%esp) # 4-byte Folded Spill
 ; X86-NEXT:    .p2align 4, 0x90
 ; X86-NEXT:  .LBB1_2: # %for.body
 ; X86-NEXT:    # =>This Inner Loop Header: Depth=1
 ; X86-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    movl (%eax,%ebx,8), %ebp
-; X86-NEXT:    movl 4(%eax,%ebx,8), %esi
-; X86-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-NEXT:    movl %ebp, %eax
-; X86-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    mull %ecx
+; X86-NEXT:    movl (%eax,%ebp,8), %esi
+; X86-NEXT:    movl 4(%eax,%ebp,8), %ecx
+; X86-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-NEXT:    movl %esi, %eax
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NEXT:    movl %edx, %edi
+; X86-NEXT:    mull %edx
+; X86-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-NEXT:    movl %esi, %eax
-; X86-NEXT:    mull %ecx
+; X86-NEXT:    movl %ecx, %eax
+; X86-NEXT:    mull %edi
 ; X86-NEXT:    movl %edx, %ecx
-; X86-NEXT:    movl %eax, %esi
-; X86-NEXT:    addl %edi, %esi
+; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X86-NEXT:    adcl $0, %ecx
-; X86-NEXT:    movl %ebp, %eax
+; X86-NEXT:    movl %esi, %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edx
 ; X86-NEXT:    mull %edx
-; X86-NEXT:    movl %edx, %ebp
+; X86-NEXT:    movl %edx, %esi
 ; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    addl %esi, %edi
-; X86-NEXT:    adcl %ecx, %ebp
-; X86-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X86-NEXT:    addl %ebx, %edi
+; X86-NEXT:    adcl %ecx, %esi
+; X86-NEXT:    setb %bl
 ; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X86-NEXT:    mull {{[0-9]+}}(%esp)
-; X86-NEXT:    movl %edx, %ecx
-; X86-NEXT:    movl %eax, %esi
-; X86-NEXT:    addl %ebp, %esi
-; X86-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X86-NEXT:    adcl %eax, %ecx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    mull %edx
-; X86-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-NEXT:    movl %eax, %ebp
-; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    mull %edx
-; X86-NEXT:    addl %ebp, %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X86-NEXT:    addl %esi, %eax
-; X86-NEXT:    adcl %ecx, %edx
-; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X86-NEXT:    movzbl %bl, %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X86-NEXT:    adcl %esi, %edx
+; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X86-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X86-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X86-NEXT:    adcl $0, %eax
 ; X86-NEXT:    adcl $0, %edx
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %esi, (%ecx,%ebx,8)
-; X86-NEXT:    movl %edi, 4(%ecx,%ebx,8)
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl %ecx, %edi
-; X86-NEXT:    addl $1, %ebx
-; X86-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X86-NEXT:    adcl $0, %esi
-; X86-NEXT:    movl %ebx, %ecx
-; X86-NEXT:    xorl %ebp, %ecx
-; X86-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-NEXT:    xorl %edi, %esi
-; X86-NEXT:    orl %ecx, %esi
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    movl %ecx, (%esi,%ebp,8)
+; X86-NEXT:    movl %edi, 4(%esi,%ebp,8)
+; X86-NEXT:    addl $1, %ebp
+; X86-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X86-NEXT:    adcl $0, %edi
+; X86-NEXT:    movl %ebp, %esi
+; X86-NEXT:    xorl %ebx, %esi
+; X86-NEXT:    movl %edi, (%esp) # 4-byte Spill
+; X86-NEXT:    xorl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    orl %esi, %edi
 ; X86-NEXT:    jne .LBB1_2
 ; X86-NEXT:  .LBB1_3: # %for.end
 ; X86-NEXT:    xorl %eax, %eax
 ; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    addl $28, %esp
+; X86-NEXT:    addl $24, %esp
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    popl %ebp
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul1:
 ; X64:       # %bb.0: # %entry

llvm/trunk/test/CodeGen/X86/imul.ll

 ; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    movq %rdi, %rcx
 ; X64-NEXT:    shlq $63, %rcx
 ; X64-NEXT:    subq %rcx, %rax
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: testNegOverflow:
 ; X86:       # %bb.0: # %entry
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X86-NEXT:    movl $1, %edx
-; X86-NEXT:    movl %ecx, %eax
-; X86-NEXT:    mull %edx
-; X86-NEXT:    shll $31, %ecx
-; X86-NEXT:    addl %ecx, %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT:    movl %eax, %edx
+; X86-NEXT:    shll $31, %edx
 ; X86-NEXT:    addl {{[0-9]+}}(%esp), %edx
 ; X86-NEXT:    retl
 entry:
   %tmp3 = mul i64 %a, -9223372036854775807
   ret i64 %tmp3
 }

llvm/trunk/test/CodeGen/X86/mul-i1024.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i386-unknown | FileCheck %s --check-prefix=X32
 ; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefix=X64
 
 define void @test_1024(i1024* %a, i1024* %b, i1024* %out) nounwind {
 ; X32-LABEL: test_1024:
 ; X32:       # %bb.0:
 ; X32-NEXT:    pushl %ebp
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    subl $1000, %esp # imm = 0x3E8
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    subl $400, %esp # imm = 0x190
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 48(%eax), %ecx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 32(%edx), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 60(%ecx), %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl 56(%ecx), %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl (%eax), %ebp
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edi, %edi
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 4(%eax), %ecx
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, %edi
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 32(%esi), %eax
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    adcl %ebp, %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl 48(%edi), %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 36(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 36(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 52(%edi), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    leal (%ebx,%eax), %eax
-; X32-NEXT:    leal (%ecx,%ebp), %edx
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebp, %ecx
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl (%eax), %eax
+; X32-NEXT:    movl 8(%eax), %ebp
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 12(%eax), %edi
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-NEXT:    movl 16(%ebp), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl %ebx, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl (%ebp), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %ebp
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edi, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebx, %edi
+; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl 4(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl %ebp, %ecx
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    movl %ebx, %ebp
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb %cl
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %edi, (%esp) # 4-byte Spill
-; X32-NEXT:    movzbl %cl, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 8(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    addl %edi, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-NEXT:    movl 52(%ebp), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %ebx
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl 56(%ebp), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %ebp
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 1-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl (%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebp, %ecx
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 40(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 40(%esi), %edi
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 16(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ebx, %ebx
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl 20(%ecx), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 24(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %ebx
-; X32-NEXT:    addl %eax, %ebx
+; X32-NEXT:    movl 44(%esi), %ebp
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl 20(%edi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %esi
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl 24(%edi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %edi
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-NEXT:    movl 32(%ebp), %edi
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 4(%ecx), %eax
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 36(%ebp), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %esi
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl %cl, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 8(%eax), %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
+; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %ecx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movl %esi, %edx
+; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl %esi, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %edx
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebp, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %eax, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ebx, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl (%esp), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ecx, %edx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %edx, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %edx, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload
 ; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 40(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 16(%eax), %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %edx
-; X32-NEXT:    addl %eax, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %ebp
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    adcl %ebx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl 48(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl 52(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %edi, %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 56(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
+; X32-NEXT:    movl 20(%eax), %edx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    mull %edx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %ebx
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
+; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %edi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl 64(%edi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 80(%edi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %ebp, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 80(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edi, %edi
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 64(%ecx), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %ebx, %esi
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 24(%eax), %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 28(%eax), %ecx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movb {{[-0-9]+}}(%e{{[sb]}}p), %al # 1-byte Reload
-; X32-NEXT:    addb $255, %al
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebp, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 68(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 72(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %edx
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movzbl %bl, %edi
+; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl (%esp), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 84(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 88(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %ebx
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %ebp, %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 84(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 88(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %ebx
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-NEXT:    movl 68(%ebp), %eax
+; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl 72(%ebp), %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %edx
-; X32-NEXT:    movl %ebx, %ebp
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl %edi, %ebx
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %ecx
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %ecx
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 12(%eax), %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    movl %ecx, %edx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    movl %edi, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %edi, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl %ebx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    setb %dl
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movzbl %dl, %edx
-; X32-NEXT:    adcl %ebx, %edx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl (%esp), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 44(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebx, %edi
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %edx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %edi, %ecx
 ; X32-NEXT:    movzbl %bl, %eax
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %edi, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %edi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl %ebp, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    adcl %edi, %ebx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 1-byte Folded Reload
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edx
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addb $255, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 12(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %edx
-; X32-NEXT:    movzbl %bl, %ebp
-; X32-NEXT:    adcl %ecx, %ebp
-; X32-NEXT:    movl %esi, %ecx
-; X32-NEXT:    addl %esi, %edx
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edi, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %edi
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl %ebx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addb $255, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 44(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    setb %cl
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    movzbl %cl, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebx, %ecx
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %esi, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %edx
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
+; X32-NEXT:    adcl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb %dl
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movzbl %dl, %eax
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl %ebp, %esi
-; X32-NEXT:    addb $255, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    addb $255, {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    adcl (%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 24(%esi), %ebp
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 60(%eax), %esi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 28(%esi), %edi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl 16(%edi), %esi
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 20(%edi), %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    adcl %ebx, %ebp
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebp, %ecx
 ; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %esi, %ebx
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edi, %ebp
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    addl %ebx, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl (%esp), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebp, %ecx
 ; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl (%esp), %edi # 1-byte Folded Reload
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 8(%esi), %ecx
 ; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl 12(%esi), %ebp
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl (%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X32-NEXT:    movl (%ebp), %edi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 4(%ebp), %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl (%esp), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %esi, %ebx
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebp, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebx, %edi
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl (%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    addl %edi, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload
+; X32-NEXT:    adcl %edi, %eax
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebp
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 28(%eax), %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %ebx
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebp, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %ebx, %edi
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movzbl %bl, %edi
+; X32-NEXT:    adcl %edi, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    setb %cl
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %edi, %ebx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl %eax, %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl 32(%ebx), %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 28(%eax), %ecx
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl 36(%eax), %ecx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, %ebx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 40(%eax), %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 44(%eax), %ebx
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %esi, %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    addl %esi, %ebp
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ebx, %ecx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    adcl %edx, %esi
 ; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %ebp
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %esi, %edi
 ; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %esi
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebx, %esi
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ebx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 48(%eax), %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %ecx
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 52(%eax), %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %edx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl %ebp, %esi
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ecx, %esi
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 56(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebp, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 60(%eax), %ecx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
 ; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %ebx, %edi
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl %ebx, %edi
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    addl %edi, %ebx
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movl %ecx, %edi
 ; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %ebx
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %esi, %ebp
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    addl %ebx, %ebp
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    movl %ecx, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl (%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    movzbl %cl, %eax
+; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ebx, %ecx
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
 ; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %edi, %ecx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %esi, %edi
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %ebx
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebp, %ecx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 60(%eax), %ebp
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebp, %esi
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ecx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movl %ebp, %esi
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ebx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %ebp
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl %ecx, %edi
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl (%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl (%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl %ebp, %edx
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl (%esp), %ebx # 4-byte Reload
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %ebx, %ecx
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ecx, %esi
 ; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ebx, %ebp
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %esi, %ebp
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
+; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %edx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %edx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %edi, %ecx
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %edi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %ebx, %edi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %edi, %ebx
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    addl %ebx, %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, %ebp
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebx
+; X32-NEXT:    adcl %eax, %ebp
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 64(%eax), %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 68(%eax), %ecx
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    movl %ecx, %esi
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 76(%eax), %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebp, %esi
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %ecx, %edi
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ecx, %esi
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebp, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 72(%eax), %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 76(%eax), %esi
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl (%esp), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    movl %esi, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebx, %esi
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    addl %ecx, %ebp
 ; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    adcl %ebx, %esi
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %ebp
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ebx
+; X32-NEXT:    adcl %ebp, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebx, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl (%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
-; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload
+; X32-NEXT:    adcl %edi, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 80(%eax), %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 84(%eax), %ecx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ecx, %ebp
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %edi, %ecx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %edi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 88(%eax), %edi
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 92(%eax), %edi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 92(%eax), %ebp
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %edi, %ecx
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %esi, %edi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %edx
 ; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl $0, %eax
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl %ecx, %edi
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl %esi, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    addl %ebx, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebx, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %esi, %ebp
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, %ebp
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
+; X32-NEXT:    adcl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, (%esp) # 4-byte Folded Spill
+; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
+; X32-NEXT:    adcl %ebx, %ebp
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %esi, %edi
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebp, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl %ebp, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebx, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %ebp
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %ebp, %edi
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %esi, %edi
 ; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    imull %ebp, %edi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %esi, %ecx
+; X32-NEXT:    imull %eax, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ecx, %edx
+; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %esi
+; X32-NEXT:    movl %esi, %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    imull %ebx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    addl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %edi, %ecx
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl %edi, %esi
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    imull %edi, %esi
+; X32-NEXT:    addl %edx, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, %edi
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl %ebx, %ecx
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    adcl %esi, %ebp
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    imull %ebp, %esi
+; X32-NEXT:    imull %ebp, %edi
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    addl %edi, %edx
 ; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    addl %edx, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %ebx, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    imull %edi, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    addl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %eax, %ecx
+; X32-NEXT:    imull %edi, %ecx
 ; X32-NEXT:    addl %edx, %ecx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    setb %bl
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movzbl %cl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl 104(%esi), %ebp
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl 104(%esi), %ebx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl 108(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl 108(%esi), %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl 96(%edi), %ebx
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl 100(%edi), %edi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %edi, %ecx
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 96(%esi), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 100(%esi), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %ebp, %esi
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %ebx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebp, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %ebx, %ecx
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %edi, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl 112(%esi), %edi
-; X32-NEXT:    imull %edi, %ebp
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 112(%ecx), %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %edi, %ebx
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl 116(%esi), %eax
+; X32-NEXT:    addl %ebx, %edx
+; X32-NEXT:    movl 116(%ecx), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    imull %eax, %ecx
-; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    movl 120(%esi), %eax
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    imull %esi, %ecx
+; X32-NEXT:    imull %eax, %esi
+; X32-NEXT:    addl %edx, %esi
+; X32-NEXT:    movl %esi, %ebx
+; X32-NEXT:    movl 120(%ecx), %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    imull %ecx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ecx, %edx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 124(%ecx), %ecx
-; X32-NEXT:    imull %ebp, %ecx
-; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 124(%esi), %esi
+; X32-NEXT:    imull %ebp, %esi
+; X32-NEXT:    addl %edx, %esi
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    adcl %ebx, %esi
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl %ebp, %ebx
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    adcl %ecx, %ebp
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    imull %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    imull %eax, %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %esi, %edx
-; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %ebx
+; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl %edx, %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    imull %ebp, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %ebx, %edi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %edx
 ; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    imull %eax, %ecx
 ; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebp, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ebx, %esi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    adcl %ebp, %ebx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movzbl %bl, %edi
+; X32-NEXT:    adcl %edi, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, %ecx
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, %edi
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, %ebx
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload

 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 88(%esi), %ebp
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 92(%eax), %esi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 92(%esi), %ebx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl (%esp), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %ebp, %ebx
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 80(%esi), %edi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 84(%esi), %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %ebx, %ecx
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    adcl %esi, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ebx
+; X32-NEXT:    adcl %esi, %ebp
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 72(%esi), %ebp
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl 76(%esi), %ebx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl (%esp), %ebp # 4-byte Reload
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 76(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl 64(%ebx), %ecx
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl 68(%ebx), %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %ebx, %ecx
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ebx, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ebx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, %ecx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ecx, %esi
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %edi, %ebp
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %ebx, %esi
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movzbl %bl, %esi
 ; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %edi, %ebx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %ebx, %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    movl %edi, %ebp
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebp
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ebx, %ebp
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %ebp, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl %eax, %ebx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 96(%ecx), %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
 ; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 100(%eax), %esi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %esi, %ecx
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl 96(%eax), %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 100(%eax), %ecx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %ecx, %esi
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    movzbl %bl, %eax
-; X32-NEXT:    adcl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl %eax, %ebx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl (%esp), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 104(%eax), %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl 104(%eax), %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl 108(%eax), %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %edi, %esi
-; X32-NEXT:    movzbl %bl, %eax
-; X32-NEXT:    adcl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %ecx, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %ebp
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ebx, %ebp
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    imull %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    imull %eax, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %esi
-; X32-NEXT:    movl %esi, %ebx
+; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %ebx, %ecx
+; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %eax, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    imull %edi, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    addl %esi, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    imull %ecx, %esi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    imull %ebp, %esi
 ; X32-NEXT:    addl %edx, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebx, %edi
+; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %ebp
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl 124(%edi), %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    imull %eax, %ecx
 ; X32-NEXT:    movl 120(%edi), %esi
+; X32-NEXT:    movl %edi, %ebx
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    addl %ecx, %edx
-; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    imull (%esp), %esi # 4-byte Folded Reload
 ; X32-NEXT:    addl %edx, %esi
-; X32-NEXT:    movl 112(%edi), %ebp
-; X32-NEXT:    movl 116(%edi), %ebx
+; X32-NEXT:    movl 112(%edi), %edi
+; X32-NEXT:    movl 116(%ebx), %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    imull %ebx, %edi
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    imull %ebp, %ebx
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %ebx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %ebp, %ecx
+; X32-NEXT:    imull %edi, %ecx
 ; X32-NEXT:    addl %edx, %ecx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ebx, %ebp
-; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    adcl %ebx, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl %esi, %ebx
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ecx, %ebp
+; X32-NEXT:    adcl %ebx, %ebp
+; X32-NEXT:    setb %cl
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %bl, %edi
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    movzbl %cl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %ebp, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %ebx, %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    adcl %edx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl (%esp), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %edi, %ecx
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl %ebp, %edi
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ecx, %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    imull %eax, %edi
+; X32-NEXT:    movl %edi, %ecx
+; X32-NEXT:    imull %eax, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    addl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    imull %ebp, %esi
 ; X32-NEXT:    addl %edx, %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    imull %ebx, %edi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    addl %edi, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    imull %esi, %edi
 ; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %ecx

 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %edi, %ebp
 ; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movzbl %cl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    imull %eax, %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %esi, %ebx
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    imull %ebx, %ecx
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    addl %ebx, %edx
+; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ecx, %ebx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    imull %esi, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    imull %edi, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    addl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %edi, %ecx
+; X32-NEXT:    imull %ebp, %ecx
 ; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl %ebx, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebp, %ebx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl %edi, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    addl %ebp, %ecx
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl %esi, %edi
 ; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edx
-; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movzbl %cl, %ecx
 ; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl (%esp), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, %edx
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, %edi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, %ebp
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl (%esp), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl (%esp), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, (%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, 4(%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, 8(%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload

 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, 48(%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, 52(%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, 56(%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, 60(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 64(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 68(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 72(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 76(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 80(%ecx)
-; X32-NEXT:    movl %ebp, 84(%ecx)
-; X32-NEXT:    movl %edi, 88(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 92(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 96(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 100(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 104(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 108(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, 112(%ecx)
-; X32-NEXT:    movl %ebx, 116(%ecx)
-; X32-NEXT:    movl %edx, 120(%ecx)
-; X32-NEXT:    movl %eax, 124(%ecx)
-; X32-NEXT:    addl $1000, %esp # imm = 0x3E8
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 64(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 68(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 72(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 76(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 80(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 84(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 88(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 92(%ecx)
+; X32-NEXT:    movl %ebp, 96(%ecx)
+; X32-NEXT:    movl %ebx, 100(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 104(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 108(%ecx)
+; X32-NEXT:    movl %edi, 112(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 116(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %eax, 120(%ecx)
+; X32-NEXT:    movl %edx, 124(%ecx)
+; X32-NEXT:    addl $400, %esp # imm = 0x190
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    popl %edi
 ; X32-NEXT:    popl %ebx
 ; X32-NEXT:    popl %ebp
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_1024:
 ; X64:       # %bb.0:
 ; X64-NEXT:    pushq %rbp
 ; X64-NEXT:    pushq %r15
 ; X64-NEXT:    pushq %r14
 ; X64-NEXT:    pushq %r13
 ; X64-NEXT:    pushq %r12
 ; X64-NEXT:    pushq %rbx
-; X64-NEXT:    subq $352, %rsp # imm = 0x160
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq 48(%rdi), %r9
-; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq 40(%rdi), %rbp
-; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq 32(%rdi), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdi, %r10
-; X64-NEXT:    xorl %r8d, %r8d
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rdi, %rbx
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    addq %rcx, %rbx
-; X64-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rcx, %r11
-; X64-NEXT:    adcq %rdi, %rbp
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movzbl %bl, %ebx
-; X64-NEXT:    addq %rax, %rbp
-; X64-NEXT:    adcq %rdx, %rbx
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    subq $240, %rsp
 ; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    movq %rdi, %r14
-; X64-NEXT:    movq %rdi, %r8
 ; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rdx, %r14
-; X64-NEXT:    addq %rbp, %rcx
-; X64-NEXT:    movq %rcx, %r12
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbx, %r14
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq (%rsi), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    xorl %ebp, %ebp
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq 8(%rsi), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    xorl %r9d, %r9d
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    addq %rcx, %r15
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    addq %rdi, %r15
-; X64-NEXT:    adcq %rcx, %rbp
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    setb %bl
-; X64-NEXT:    addq %rax, %rbp
-; X64-NEXT:    movzbl %bl, %ebx
-; X64-NEXT:    adcq %rdx, %rbx
-; X64-NEXT:    movq 16(%rsi), %rax
+; X64-NEXT:    movq 40(%rdi), %r15
+; X64-NEXT:    movq 32(%rdi), %r9
+; X64-NEXT:    movq 56(%rdi), %r8
+; X64-NEXT:    movq 48(%rdi), %rbx
+; X64-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq (%rsi), %rdi
+; X64-NEXT:    movq 8(%rsi), %r11
 ; X64-NEXT:    movq %rsi, %r13
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdi, %r9
-; X64-NEXT:    addq %rax, %r9
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    adcq %rdx, %rax
-; X64-NEXT:    addq %rbp, %r9
-; X64-NEXT:    adcq %rbx, %rax
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    adcq %rcx, %r8
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    movq %r8, %rax
 ; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq (%r10), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    xorl %r8d, %r8d
-; X64-NEXT:    mulq %r8
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    movq %rdx, %rax
-; X64-NEXT:    adcq %rcx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq 32(%r13), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rsi
 ; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    movq %rbx, %r8
-; X64-NEXT:    addq %r13, %rax
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    adcq %rdx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rbp, %r10
+; X64-NEXT:    adcq %rsi, %rcx
+; X64-NEXT:    setb %al
+; X64-NEXT:    movzbl %al, %esi
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %rcx, %r14
+; X64-NEXT:    adcq %rsi, %r8
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    adcq %r15, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    adcq %r9, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %r14
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %rdi
-; X64-NEXT:    movq 8(%r10), %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    addq %rsi, %r12
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rcx, %rdi
 ; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    adcq %rsi, %rbp
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    setb %bl
-; X64-NEXT:    addq %rax, %rbp
-; X64-NEXT:    movzbl %bl, %ebx
-; X64-NEXT:    adcq %rdx, %rbx
-; X64-NEXT:    movq 16(%r10), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rcx
-; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    movq %rsi, %r10
-; X64-NEXT:    adcq %rdx, %r10
-; X64-NEXT:    addq %rbp, %rcx
-; X64-NEXT:    adcq %rbx, %r10
-; X64-NEXT:    movq %r8, %rdx
-; X64-NEXT:    movq %r8, %r14
-; X64-NEXT:    movq %r8, (%rsp) # 8-byte Spill
-; X64-NEXT:    addq %r11, %rdx
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r12, %rsi
-; X64-NEXT:    adcq %r12, %r15
-; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rcx, %r9
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rcx, %r8
-; X64-NEXT:    adcq %r10, %rdi
-; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    movq 40(%rdi), %rax
+; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    addq %rdi, %rax
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    xorl %r9d, %r9d
-; X64-NEXT:    mulq %r9
+; X64-NEXT:    adcq %rbp, %rbx
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %r15, %rbp
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %r15
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    addq %r11, %rcx
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    addq %r13, %rcx
-; X64-NEXT:    adcq %r11, %rbp
-; X64-NEXT:    setb %bl
-; X64-NEXT:    addq %rax, %rbp
-; X64-NEXT:    movzbl %bl, %ebx
-; X64-NEXT:    adcq %rdx, %rbx
-; X64-NEXT:    movq 48(%rdi), %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r13, %r12
-; X64-NEXT:    addq %rax, %r12
-; X64-NEXT:    movq %r11, %rdi
-; X64-NEXT:    adcq %rdx, %rdi
-; X64-NEXT:    addq %rbp, %r12
-; X64-NEXT:    adcq %rbx, %rdi
+; X64-NEXT:    addq %rbx, %rcx
+; X64-NEXT:    movzbl %sil, %eax
+; X64-NEXT:    adcq %rax, %r15
+; X64-NEXT:    addq %r12, %rcx
+; X64-NEXT:    adcq %r10, %r15
+; X64-NEXT:    adcq $0, %r14
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    movq %r13, %rdi
 ; X64-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    addq %r13, %r14
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rcx, %rsi
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r12, %r8
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rdi, %r10
-; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    addq %r13, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    adcq %r11, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    movq %r8, %r10
-; X64-NEXT:    addq %r13, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Folded Reload
-; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq 56(%rax), %r11
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rsi, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rbx, %r8
-; X64-NEXT:    adcq %rbp, %rsi
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    movq 16(%r13), %r10
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, %r11
+; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %r12, %rsi
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq 24(%rdi), %rdi
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdi, %r13
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movzbl %cl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Folded Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %r15
-; X64-NEXT:    adcq %rdx, %r12
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdi, %r12
 ; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rsi, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r11
-; X64-NEXT:    mulq %r13
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rcx
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r13
-; X64-NEXT:    movq %rdx, %r13
+; X64-NEXT:    addq %rsi, %rax
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rcx, %rsi
-; X64-NEXT:    movzbl %bl, %eax
-; X64-NEXT:    adcq %rax, %r13
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
-; X64-NEXT:    addq %r14, %rsi
-; X64-NEXT:    adcq %r8, %r13
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    adcq $0, %r12
-; X64-NEXT:    movq %r11, %rbx
+; X64-NEXT:    adcq %rbx, %rbp
+; X64-NEXT:    setb %r9b
 ; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, %r11
-; X64-NEXT:    mulq %rbp
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq 24(%rax), %r9
-; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rbp, %r8
-; X64-NEXT:    adcq %rdi, %rcx
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %bl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq %r10, %rbp
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    addq %r11, %rbp
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    adcq %r10, %rbx
-; X64-NEXT:    addq %rax, %rbp
-; X64-NEXT:    adcq %rdx, %rbx
-; X64-NEXT:    addq %rsi, %r14
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r13, %r8
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rbp, %rbx
+; X64-NEXT:    movzbl %r9b, %eax
+; X64-NEXT:    adcq %rax, %rdi
+; X64-NEXT:    addq %rcx, %r13
+; X64-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r15, %rsi
+; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    addq %r15, %rbp
-; X64-NEXT:    adcq %r12, %rbx
-; X64-NEXT:    setb %r15b
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    addq %r14, %rbx
+; X64-NEXT:    adcq %r8, %rdi
+; X64-NEXT:    setb %r11b
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
 ; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    mulq %rsi
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %r14, %rdi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %r14, %rbp
 ; X64-NEXT:    adcq $0, %rsi
 ; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r12
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rdi, %r14
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, %rbp
 ; X64-NEXT:    adcq %rsi, %rcx
 ; X64-NEXT:    setb %sil
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %r9, %r12
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r12
 ; X64-NEXT:    addq %rcx, %rax
 ; X64-NEXT:    movzbl %sil, %ecx
 ; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    addq %r11, %rcx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    adcq %r10, %r9
-; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    adcq %rdx, %r9
-; X64-NEXT:    addq %rbp, %r13
-; X64-NEXT:    adcq %rbx, %r14
-; X64-NEXT:    movzbl %r15b, %eax
-; X64-NEXT:    adcq %rax, %rcx
-; X64-NEXT:    adcq $0, %r9
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
-; X64-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq 24(%rax), %rcx
+; X64-NEXT:    addq %rbx, %r15
+; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rdi, %rbp
+; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movzbl %r11b, %ecx
+; X64-NEXT:    adcq %rcx, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    movq 16(%rsi), %rcx
 ; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    movq 24(%rsi), %rbx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rbx, %rbp
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    addq %r9, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, %r15
 ; X64-NEXT:    addq %rbp, %r15
-; X64-NEXT:    adcq %rsi, %rbx
-; X64-NEXT:    setb %sil
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movzbl %sil, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %r8
-; X64-NEXT:    adcq %rdx, %r10
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    setb %dil
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rcx, %r12
+; X64-NEXT:    movzbl %dil, %eax
+; X64-NEXT:    adcq %rax, %r11
+; X64-NEXT:    movq (%rsi), %rdi
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq 8(%rsi), %r9
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rbp
 ; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rdi, %rbx
+; X64-NEXT:    addq %rcx, %rbx
 ; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r9
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rsi
 ; X64-NEXT:    addq %rbx, %rax
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rdi
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    movq %rsi, %rbp
-; X64-NEXT:    mulq %r11
+; X64-NEXT:    adcq %rbp, %rsi
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %r13
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rsi, %rcx
+; X64-NEXT:    movzbl %bl, %eax
+; X64-NEXT:    adcq %rax, %r13
+; X64-NEXT:    addq %r14, %rcx
+; X64-NEXT:    adcq %r15, %r13
+; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    adcq $0, %r11
+; X64-NEXT:    movq %rdi, %rbp
+; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r10
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rsi, %rdi
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    adcq %rbx, %rbp
+; X64-NEXT:    setb %r15b
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %r8
 ; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rdi, %rbx
-; X64-NEXT:    movzbl %cl, %eax
-; X64-NEXT:    adcq %rax, %rsi
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
-; X64-NEXT:    addq %r14, %rbx
-; X64-NEXT:    adcq %r15, %rsi
+; X64-NEXT:    addq %rbp, %rbx
+; X64-NEXT:    movzbl %r15b, %eax
+; X64-NEXT:    adcq %rax, %r8
+; X64-NEXT:    addq %rcx, %r14
+; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r13, %rdi
+; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rbx
 ; X64-NEXT:    adcq $0, %r8
-; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    mulq %rdi
+; X64-NEXT:    addq %r12, %rbx
+; X64-NEXT:    adcq %r11, %r8
+; X64-NEXT:    setb %r14b
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %r10, (%rsp) # 8-byte Spill
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %r14
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdi, %r15
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbp
 ; X64-NEXT:    addq %rcx, %rbp
 ; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq %r12
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rsi, %r15
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    mulq %rsi
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rbp, %r10
 ; X64-NEXT:    adcq %rdi, %rcx
 ; X64-NEXT:    setb %dil
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    mulq %r12
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rsi
 ; X64-NEXT:    addq %rcx, %rax
 ; X64-NEXT:    movzbl %dil, %ecx
 ; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq (%rsp), %rdi # 8-byte Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
-; X64-NEXT:    addq %r13, %rdi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
-; X64-NEXT:    adcq %r14, %rbp
-; X64-NEXT:    addq %rax, %rdi
-; X64-NEXT:    adcq %rdx, %rbp
 ; X64-NEXT:    addq %rbx, %r11
-; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rsi, %r9
-; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    addq %r8, %rdi
-; X64-NEXT:    adcq %r10, %rbp
-; X64-NEXT:    setb %r10b
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r8, %rbx
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rbx, %r8
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq %r12
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %bl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    movq %r9, %r15
-; X64-NEXT:    addq %r13, %r15
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
-; X64-NEXT:    movq %r12, %r13
-; X64-NEXT:    adcq %r14, %r13
-; X64-NEXT:    addq %rax, %r15
-; X64-NEXT:    adcq %rdx, %r13
-; X64-NEXT:    addq %rdi, %r11
-; X64-NEXT:    adcq %rbp, %r8
-; X64-NEXT:    movzbl %r10b, %eax
-; X64-NEXT:    adcq %rax, %r15
-; X64-NEXT:    adcq $0, %r13
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
+; X64-NEXT:    adcq %r8, %r10
+; X64-NEXT:    movzbl %r14b, %ecx
+; X64-NEXT:    adcq %rcx, %rax
+; X64-NEXT:    adcq $0, %rdx
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Folded Reload
 ; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Folded Reload
+; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
+; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    adcq $0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; X64-NEXT:    adcq $0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; X64-NEXT:    adcq $0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; X64-NEXT:    adcq $0, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    movq 32(%rsi), %rdi
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %r9, %rbx
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq 40(%rsi), %rcx
+; X64-NEXT:    movq %rsi, %r13
+; X64-NEXT:    movq %r15, %rax
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rcx, %r8
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %rbp, %r9
+; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %r10
 ; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %rbp
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %rsi, %rcx
-; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rcx, %r8
-; X64-NEXT:    adcq %rbx, %rsi
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movzbl %cl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movq %r12, %r10
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %r9
-; X64-NEXT:    adcq %rdx, %r10
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rdx, %r12
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r12, %rbx
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r12
+; X64-NEXT:    addq %rcx, %r14
+; X64-NEXT:    movzbl %sil, %eax
+; X64-NEXT:    adcq %rax, %r10
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
+; X64-NEXT:    movq %r12, %rax
+; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbx, %rax
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %sil
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, %rbp
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
+; X64-NEXT:    movq %r15, %rax
 ; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %rcx, %rsi
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %r12, %rax
+; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rdi, %rbp
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rbp, %rcx
 ; X64-NEXT:    movzbl %sil, %eax
-; X64-NEXT:    adcq %rax, %rdi
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload
-; X64-NEXT:    addq %r14, %rbx
-; X64-NEXT:    adcq %r8, %rdi
-; X64-NEXT:    adcq $0, %r9
+; X64-NEXT:    adcq %rax, %r8
+; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
+; X64-NEXT:    adcq %r9, %r8
+; X64-NEXT:    adcq $0, %r14
 ; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    movq %r12, %r11
+; X64-NEXT:    movq 48(%r13), %r11
+; X64-NEXT:    movq %r12, %rdi
 ; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %r8
-; X64-NEXT:    mulq %rcx
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %r14, %rcx
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %r12, %rbx
 ; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq 56(%rax), %rsi
-; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    movq 56(%r13), %rsi
+; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %r11
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rbx, %r13
+; X64-NEXT:    adcq %rbp, %rdi
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rsi, %r15
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rcx, %r14
-; X64-NEXT:    adcq %rbp, %rsi
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    movzbl %cl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq (%rsp), %rcx # 8-byte Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    addq %r8, %rcx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    adcq %r11, %rsi
-; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    adcq %rdx, %rsi
-; X64-NEXT:    addq %rbx, %r12
-; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rdi, %r14
-; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rdi, %r12
+; X64-NEXT:    movzbl %bl, %eax
+; X64-NEXT:    adcq %rax, %rsi
+; X64-NEXT:    addq %rcx, %r9
+; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r8, %r13
+; X64-NEXT:    adcq $0, %r12
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    addq %r9, %rcx
+; X64-NEXT:    addq %r14, %r12
 ; X64-NEXT:    adcq %r10, %rsi
 ; X64-NEXT:    setb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rax, %r8
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
 ; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %r12
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r9, %rbx
-; X64-NEXT:    adcq $0, %r12
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %r9
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rbx, %rbp
-; X64-NEXT:    adcq %r12, %r9
-; X64-NEXT:    setb %bl
+; X64-NEXT:    addq %r9, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    addq %rbp, %r9
+; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    setb %dil
 ; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    addq %r9, %rax
-; X64-NEXT:    movzbl %bl, %edi
-; X64-NEXT:    adcq %rdi, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
-; X64-NEXT:    addq %r8, %r12
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    adcq %r11, %r10
-; X64-NEXT:    addq %rax, %r12
-; X64-NEXT:    adcq %rdx, %r10
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
-; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    adcq %rsi, %rbp
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %rcx, %r14
+; X64-NEXT:    movzbl %dil, %eax
+; X64-NEXT:    adcq %rax, %r10
+; X64-NEXT:    addq %r12, %r8
+; X64-NEXT:    adcq %rsi, %r9
 ; X64-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
-; X64-NEXT:    adcq %rax, %r12
+; X64-NEXT:    adcq %rax, %r14
 ; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Folded Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
 ; X64-NEXT:    addq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
 ; X64-NEXT:    adcq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; X64-NEXT:    adcq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; X64-NEXT:    adcq %r13, %r14
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rdx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    adcq $0, %r12
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; X64-NEXT:    adcq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
+; X64-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    adcq $0, %r9
+; X64-NEXT:    adcq $0, %r14
 ; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Folded Reload
+; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
+; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
+; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Folded Reload
 ; X64-NEXT:    setb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r11, %rbx
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
+; X64-NEXT:    movq %rdi, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
 ; X64-NEXT:    mulq %rsi
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rbx, %r9
-; X64-NEXT:    adcq %rdi, %rcx
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %r13
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %bl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %r8
-; X64-NEXT:    adcq %rdx, %r15
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %r11
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rsi, %r8
+; X64-NEXT:    movq %rdx, %rsi
 ; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r11, %rbx
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r14
-; X64-NEXT:    mulq %r13
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rbx, %r12
+; X64-NEXT:    adcq %rsi, %rcx
 ; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r9
 ; X64-NEXT:    movq %rdx, %r11
 ; X64-NEXT:    movq %rax, %r13
 ; X64-NEXT:    addq %rcx, %r13
 ; X64-NEXT:    movzbl %bl, %eax
 ; X64-NEXT:    adcq %rax, %r11
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Folded Reload
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Folded Reload
-; X64-NEXT:    adcq %r9, %r11
-; X64-NEXT:    adcq $0, %r8
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rcx, %rdi
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, %r8
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    addq %rdi, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rbx, %rbp
+; X64-NEXT:    setb %bl
 ; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    movq %rsi, %r15
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %r15
 ; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %rbx, %rcx
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    addq %rbp, %rcx
+; X64-NEXT:    movzbl %bl, %eax
+; X64-NEXT:    adcq %rax, %r15
+; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
+; X64-NEXT:    adcq %r12, %r15
+; X64-NEXT:    adcq $0, %r13
+; X64-NEXT:    adcq $0, %r11
+; X64-NEXT:    movq %r8, %rbx
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rsi, %r9
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %r8, %rsi
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rbx
 ; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rcx, %r8
-; X64-NEXT:    adcq %rsi, %rbx
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    addq %rsi, %r8
+; X64-NEXT:    adcq %rbp, %rbx
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    mulq %rdi
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movzbl %cl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %r14
-; X64-NEXT:    adcq %rdx, %rcx
-; X64-NEXT:    addq %r13, %r9
-; X64-NEXT:    movq %r9, %r13
-; X64-NEXT:    adcq %r11, %r8
-; X64-NEXT:    movq %r8, %r15
-; X64-NEXT:    adcq $0, %r14
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rbx, %rbp
+; X64-NEXT:    movzbl %sil, %eax
+; X64-NEXT:    adcq %rax, %rdi
+; X64-NEXT:    addq %rcx, %r12
+; X64-NEXT:    adcq %r15, %r8
+; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    addq %r13, %rbp
+; X64-NEXT:    adcq %r11, %rdi
 ; X64-NEXT:    setb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
-; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
 ; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    movq %rdx, %r15
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    mulq %rsi
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %rcx, %rdi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %r15, %rbx
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rdi, %r9
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %sil
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %sil, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %rsi
-; X64-NEXT:    adcq %rdx, %rcx
-; X64-NEXT:    addq %r14, %r11
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
-; X64-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
-; X64-NEXT:    adcq %rax, %rsi
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    adcq %rsi, %r11
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    addq %r11, %rax
+; X64-NEXT:    movzbl %bl, %ecx
+; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    addq %rbp, %r13
+; X64-NEXT:    adcq %rdi, %r15
+; X64-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X64-NEXT:    adcq %rcx, %rax
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    adcq $0, %rdx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
 ; X64-NEXT:    addq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; X64-NEXT:    adcq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; X64-NEXT:    adcq %r12, %r13
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; X64-NEXT:    adcq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
+; X64-NEXT:    adcq %r14, %r12
+; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r10, %r8
+; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
+; X64-NEXT:    adcq %rax, %r13
 ; X64-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r10, %r15
+; X64-NEXT:    adcq $0, %r15
 ; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
-; X64-NEXT:    adcq %rax, %r11
-; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %r9
-; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    adcq $0, %rcx
 ; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    movq 64(%rsi), %rdi
+; X64-NEXT:    movq (%rsp), %rbx # 8-byte Reload
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r11
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    movq 64(%r9), %r11
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rsi, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq 72(%r9), %rsi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %rcx
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rbx, %r8
-; X64-NEXT:    adcq %rbp, %rsi
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdi, %r8
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq 72(%rsi), %rcx
+; X64-NEXT:    movq %rsi, %r13
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %r13
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rcx, %rsi
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %rsi, %rdi
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %rbp, %r14
+; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rsi, %r12
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    addq %rcx, %r15
 ; X64-NEXT:    movzbl %bl, %eax
-; X64-NEXT:    adcq %rax, %rcx
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    mulq %rdx
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    addq %rax, %r10
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    adcq %rdx, %r15
-; X64-NEXT:    addq %rdi, %r10
-; X64-NEXT:    adcq %rcx, %r15
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r12 # 8-byte Reload
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    movq %r11, %rsi
-; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    adcq %rax, %r10
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
 ; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r8
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    mulq %rsi
 ; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rdi, %rsi
+; X64-NEXT:    setb %cl
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r12
+; X64-NEXT:    movq %rdx, %r12
 ; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rsi, %rdi
+; X64-NEXT:    movzbl %cl, %eax
+; X64-NEXT:    adcq %rax, %r12
 ; X64-NEXT:    addq %r11, %rdi
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    mulq %r13
+; X64-NEXT:    adcq %r14, %r12
+; X64-NEXT:    adcq $0, %r15
+; X64-NEXT:    adcq $0, %r10
+; X64-NEXT:    movq 80(%r13), %r11
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %rax, %r14
 ; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %r11
+; X64-NEXT:    movq %rbp, %rsi
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq 88(%r13), %r13
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    mulq %r13
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %sil, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r13 # 8-byte Reload
-; X64-NEXT:    adcq %r13, %r14
-; X64-NEXT:    addq %rax, %rbx
-; X64-NEXT:    adcq %rdx, %r14
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
-; X64-NEXT:    adcq %r8, %r14
-; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    movq 80(%r9), %rdi
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %r8
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %r8, %rcx
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq 88(%r9), %r9
-; X64-NEXT:    movq %r12, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %rcx, %r8
-; X64-NEXT:    adcq %rsi, %rbp
-; X64-NEXT:    setb %r12b
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %r9
 ; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    adcq %rbx, %rcx
+; X64-NEXT:    setb %r9b
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rbx
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rbp, %rsi
-; X64-NEXT:    movzbl %r12b, %eax
-; X64-NEXT:    adcq %rax, %rcx
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    mulq %rdx
-; X64-NEXT:    movq %rdx, %r12
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    addq %r11, %rax
-; X64-NEXT:    adcq %rdx, %r13
-; X64-NEXT:    addq %rsi, %rax
-; X64-NEXT:    adcq %rcx, %r13
-; X64-NEXT:    addq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
-; X64-NEXT:    adcq %r14, %r8
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rax
-; X64-NEXT:    adcq $0, %r13
-; X64-NEXT:    addq %r10, %rax
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    adcq %r15, %r13
-; X64-NEXT:    setb %r14b
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    addq %rcx, %rsi
+; X64-NEXT:    movzbl %r9b, %eax
+; X64-NEXT:    adcq %rax, %rbx
+; X64-NEXT:    addq %rdi, %r14
+; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r12, %rbp
+; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    mulq %r9
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    addq %r15, %rsi
+; X64-NEXT:    adcq %r10, %rbx
+; X64-NEXT:    setb %r9b
+; X64-NEXT:    movq (%rsp), %r14 # 8-byte Reload
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %r9
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %bl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    addq %r11, %rsi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    adcq %r12, %rcx
-; X64-NEXT:    addq %rax, %rsi
-; X64-NEXT:    adcq %rdx, %rcx
-; X64-NEXT:    addq %r8, %r15
-; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r13, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rdi
+; X64-NEXT:    movq %r14, %rax
+; X64-NEXT:    movq %r14, %r15
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    adcq %rdi, %rcx
+; X64-NEXT:    setb %dil
+; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rcx, %r12
+; X64-NEXT:    movzbl %dil, %eax
+; X64-NEXT:    adcq %rax, %rdx
+; X64-NEXT:    addq %rsi, %r10
+; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rbx, %rbp
 ; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movzbl %r14b, %eax
-; X64-NEXT:    adcq %rax, %rsi
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movzbl %r9b, %eax
+; X64-NEXT:    adcq %rax, %r12
+; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rax, %r9
+; X64-NEXT:    imulq %rax, %r13
 ; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    mulq %rdi
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    addq %r9, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    imulq %rbp, %rdi
-; X64-NEXT:    addq %rdx, %rdi
+; X64-NEXT:    addq %r13, %rdx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
+; X64-NEXT:    imulq %rdi, %r11
+; X64-NEXT:    addq %rdx, %r11
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    imulq %r11, %rsi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
+; X64-NEXT:    imulq %rbx, %rsi
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    mulq %rbp
 ; X64-NEXT:    movq %rax, %r9
 ; X64-NEXT:    addq %rsi, %rdx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rcx, %rax
+; X64-NEXT:    imulq %rbp, %rax
 ; X64-NEXT:    addq %rdx, %rax
 ; X64-NEXT:    addq %r8, %r9
-; X64-NEXT:    adcq %rdi, %rax
+; X64-NEXT:    adcq %r11, %rax
 ; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %rdi
+; X64-NEXT:    movq %rbp, %rax
 ; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rbx, %r11
 ; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rsi
 ; X64-NEXT:    movq %rax, %rbx
 ; X64-NEXT:    addq %rcx, %rbx
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    addq %rbx, %r15
-; X64-NEXT:    adcq %rsi, %rdi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %rbx, %r14
+; X64-NEXT:    adcq %rsi, %rbp
 ; X64-NEXT:    setb %cl
 ; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rdx, %r12
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    addq %rdi, %r13
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r13
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    addq %rbp, %r11
 ; X64-NEXT:    movzbl %cl, %eax
-; X64-NEXT:    adcq %rax, %r12
-; X64-NEXT:    addq %r9, %r13
-; X64-NEXT:    adcq %r8, %r12
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    movq 120(%rbp), %rcx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    imulq %r10, %rcx
-; X64-NEXT:    movq 112(%rbp), %rsi
-; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    adcq %rax, %r13
+; X64-NEXT:    addq %r9, %r11
+; X64-NEXT:    adcq %r8, %r13
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
+; X64-NEXT:    movq 120(%rdx), %rcx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
+; X64-NEXT:    imulq %rax, %rcx
+; X64-NEXT:    movq 112(%rdx), %rsi
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rax, %rbx
 ; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %rax, %r10
 ; X64-NEXT:    addq %rcx, %rdx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
 ; X64-NEXT:    imulq %r8, %rsi
 ; X64-NEXT:    addq %rdx, %rsi
-; X64-NEXT:    movq 96(%rbp), %rdi
-; X64-NEXT:    movq 104(%rbp), %rbx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    imulq %rbx, %rcx
-; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq 96(%rdi), %rcx
+; X64-NEXT:    movq 104(%rdi), %rbp
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    movq %r15, %rdi
+; X64-NEXT:    imulq %rbp, %rdi
+; X64-NEXT:    mulq %rcx
 ; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rcx, %rdx
+; X64-NEXT:    addq %rdi, %rdx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rdi, %rax
+; X64-NEXT:    imulq %rcx, %rax
 ; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    addq %r11, %r9
+; X64-NEXT:    addq %r10, %r9
 ; X64-NEXT:    adcq %rsi, %rax
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    mulq %r10
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    mulq %rbx
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rsi, %rdi
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %rbp, %rdi
-; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %sil
-; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    adcq %rbx, %rcx
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %rbp, %rax
 ; X64-NEXT:    mulq %r8
 ; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %sil, %ecx
+; X64-NEXT:    movzbl %bl, %ecx
 ; X64-NEXT:    adcq %rcx, %rdx
 ; X64-NEXT:    addq %r9, %rax
-; X64-NEXT:    adcq %r11, %rdx
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
-; X64-NEXT:    adcq %r15, %rdi
-; X64-NEXT:    adcq %r13, %rax
-; X64-NEXT:    adcq %r12, %rdx
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
-; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
+; X64-NEXT:    adcq %r10, %rdx
+; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Folded Reload
+; X64-NEXT:    adcq %r14, %rsi
+; X64-NEXT:    adcq %r11, %rax
+; X64-NEXT:    adcq %r13, %rdx
+; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Folded Reload
+; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
+; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r12, %rax
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
 ; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    movq 80(%r9), %rsi
-; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq 80(%r9), %rbp
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq 88(%r9), %rbx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rcx, %rdi
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rbp, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq 88(%r9), %r8
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %r11
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rdi, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rbx, %r14
-; X64-NEXT:    adcq %rbp, %rcx
-; X64-NEXT:    setb %r10b
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %r15
 ; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rcx, %rbx
-; X64-NEXT:    movzbl %r10b, %eax
-; X64-NEXT:    adcq %rax, %rbp
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    xorl %ecx, %ecx
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rdi, %r10
+; X64-NEXT:    adcq %rsi, %rbp
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq (%rsp), %r12 # 8-byte Reload
-; X64-NEXT:    addq %r12, %rsi
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    adcq %r8, %r10
-; X64-NEXT:    addq %rbx, %rsi
-; X64-NEXT:    adcq %rbp, %r10
+; X64-NEXT:    movq %rcx, %rbx
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    addq %rbp, %r11
+; X64-NEXT:    movzbl %sil, %eax
+; X64-NEXT:    adcq %rax, %r8
 ; X64-NEXT:    movq 64(%r9), %r13
 ; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq 72(%r9), %rdi
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rcx, %rbx
-; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rcx
-; X64-NEXT:    setb %r11b
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r15
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    movzbl %r11b, %eax
-; X64-NEXT:    adcq %rax, %rbx
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    movq %r12, %rcx
-; X64-NEXT:    addq %rax, %rcx
-; X64-NEXT:    adcq %rdx, %r8
-; X64-NEXT:    addq %rbp, %rcx
-; X64-NEXT:    adcq %rbx, %r8
-; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    movq %rcx, (%rsp) # 8-byte Spill
-; X64-NEXT:    adcq %r14, %r8
-; X64-NEXT:    movq %r8, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    adcq $0, %r10
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, %r8
-; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %r14, %rdi
 ; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rax, (%rsp) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    movq 72(%r9), %rax
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    addq %r14, %rbp
 ; X64-NEXT:    adcq $0, %rdi
 ; X64-NEXT:    movq %r13, %rax
+; X64-NEXT:    movq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %rcx
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rdi, %rsi
+; X64-NEXT:    setb %cl
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rsi, %r12
+; X64-NEXT:    movzbl %cl, %eax
+; X64-NEXT:    adcq %rax, %r14
+; X64-NEXT:    addq %r15, %r12
+; X64-NEXT:    adcq %r10, %r14
+; X64-NEXT:    adcq $0, %r11
+; X64-NEXT:    adcq $0, %r8
+; X64-NEXT:    movq %r13, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
 ; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    adcq %rdi, %rcx
-; X64-NEXT:    setb %dil
-; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    movq %rdx, %rbp
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %r9, %r15
+; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    mulq %rbx
-; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %dil, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rbp, %rcx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %r13, %rax
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r9 # 8-byte Reload
-; X64-NEXT:    addq %r9, %r15
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    adcq %r8, %r11
-; X64-NEXT:    addq %rax, %r15
-; X64-NEXT:    adcq %rdx, %r11
-; X64-NEXT:    addq (%rsp), %r12 # 8-byte Folded Reload
-; X64-NEXT:    movq %r12, (%rsp) # 8-byte Spill
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    adcq %rsi, %rdi
+; X64-NEXT:    setb %cl
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %r15
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rdi, %r13
+; X64-NEXT:    movzbl %cl, %eax
+; X64-NEXT:    adcq %rax, %r15
+; X64-NEXT:    addq %r12, %r10
+; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r14, %rbp
 ; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %r13
 ; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    adcq $0, %r11
-; X64-NEXT:    addq %rsi, %r15
-; X64-NEXT:    adcq %r10, %r11
-; X64-NEXT:    setb {{[-0-9]+}}(%r{{[sb]}}p) # 1-byte Folded Spill
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %r14, %rsi
-; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r10, %rbx
-; X64-NEXT:    adcq $0, %rdi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    addq %rbx, %rax
+; X64-NEXT:    addq %r11, %r13
+; X64-NEXT:    adcq %r8, %r15
+; X64-NEXT:    setb %r8b
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %r11
 ; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    adcq %rdi, %r10
-; X64-NEXT:    setb %bl
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rcx, %rdi
-; X64-NEXT:    addq %r10, %rax
-; X64-NEXT:    movzbl %bl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
-; X64-NEXT:    addq %r9, %rbx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    adcq %r8, %rcx
-; X64-NEXT:    addq %rax, %rbx
-; X64-NEXT:    adcq %rdx, %rcx
-; X64-NEXT:    addq %r15, %rbp
-; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r11, %r12
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
+; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %r11, %rcx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    adcq %rsi, %r14
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    mulq %r9
+; X64-NEXT:    addq %r14, %rax
+; X64-NEXT:    movzbl %sil, %esi
+; X64-NEXT:    adcq %rsi, %rdx
+; X64-NEXT:    addq %r13, %r12
 ; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movzbl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 1-byte Folded Reload
-; X64-NEXT:    adcq %rax, %rbx
-; X64-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    adcq %r15, %rcx
 ; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movzbl %r8b, %ecx
+; X64-NEXT:    adcq %rcx, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    movq 96(%rbp), %rcx
-; X64-NEXT:    imulq %rcx, %rdi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rdi, %rdx
-; X64-NEXT:    movq 104(%rbp), %r8
-; X64-NEXT:    imulq %r8, %rsi
-; X64-NEXT:    addq %rdx, %rsi
-; X64-NEXT:    movq %rsi, %r11
+; X64-NEXT:    movq 96(%rbp), %rsi
+; X64-NEXT:    imulq %rsi, %r9
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    mulq %rbx
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    addq %r9, %rdx
+; X64-NEXT:    movq 104(%rbp), %r15
+; X64-NEXT:    imulq %r15, %rbx
+; X64-NEXT:    addq %rdx, %rbx
+; X64-NEXT:    movq %rbx, %r9
 ; X64-NEXT:    movq 112(%rbp), %rax
 ; X64-NEXT:    movq %rbp, %rdi
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    imulq %rbp, %rsi
+; X64-NEXT:    movq %rax, %rcx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    addq %rsi, %rdx
-; X64-NEXT:    movq 120(%rdi), %rdi
-; X64-NEXT:    imulq %rbx, %rdi
-; X64-NEXT:    addq %rdx, %rdi
-; X64-NEXT:    addq %r9, %r10
-; X64-NEXT:    adcq %r11, %rdi
-; X64-NEXT:    movq %rbx, %rax
-; X64-NEXT:    movq %rbx, %rsi
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    imulq %rbx, %rcx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rcx, %rdx
+; X64-NEXT:    movq 120(%rdi), %r8
+; X64-NEXT:    imulq %rbp, %r8
+; X64-NEXT:    addq %rdx, %r8
+; X64-NEXT:    addq %r14, %r13
+; X64-NEXT:    adcq %r9, %r8
 ; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %r9
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rbx, %rbp
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    movq %rsi, %rax
-; X64-NEXT:    mulq %r8
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rdx, %r14
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %rsi
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    addq %rbp, %r12
-; X64-NEXT:    adcq %rcx, %rsi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %r14, %rcx
+; X64-NEXT:    adcq $0, %rsi
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %rsi, %rdi
 ; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r9, %rax
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rsi, %rbp
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r15
+; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    movq %rax, %r15
+; X64-NEXT:    addq %rdi, %r15
 ; X64-NEXT:    movzbl %cl, %eax
-; X64-NEXT:    adcq %rax, %rbx
-; X64-NEXT:    addq %r10, %rbp
-; X64-NEXT:    adcq %rdi, %rbx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    imulq %r13, %rsi
-; X64-NEXT:    movq %r13, %rax
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
-; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rsi, %rdx
+; X64-NEXT:    adcq %rax, %r12
+; X64-NEXT:    addq %r13, %r15
+; X64-NEXT:    adcq %r8, %r12
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r11 # 8-byte Reload
-; X64-NEXT:    imulq %r11, %r8
-; X64-NEXT:    addq %rdx, %r8
+; X64-NEXT:    imulq %r11, %rcx
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
+; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rax, %r8
+; X64-NEXT:    addq %rcx, %rdx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
+; X64-NEXT:    imulq %r14, %rsi
+; X64-NEXT:    addq %rdx, %rsi
+; X64-NEXT:    movq %rsi, %r9
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    imulq %r15, %rcx
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    imulq %rbp, %rsi
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    imulq %rdi, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    addq %r9, %r10
-; X64-NEXT:    adcq %r8, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rsi, %rdx
+; X64-NEXT:    imulq %rdi, %r10
+; X64-NEXT:    addq %rdx, %r10
+; X64-NEXT:    addq %r8, %r13
+; X64-NEXT:    adcq %r9, %r10
+; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, %r14
-; X64-NEXT:    mulq %r13
-; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %rdi, %r10
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rsi
 ; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %r13
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r11
 ; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    addq %rdi, %rcx
+; X64-NEXT:    movq %rax, %rdi
+; X64-NEXT:    addq %rsi, %rdi
 ; X64-NEXT:    adcq $0, %r9
-; X64-NEXT:    movq %r14, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    movq %rdx, %rdi
+; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rcx
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rcx, %rsi
-; X64-NEXT:    adcq %r9, %rdi
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %r11
-; X64-NEXT:    addq %rdi, %rax
-; X64-NEXT:    movzbl %cl, %ecx
+; X64-NEXT:    addq %rdi, %rsi
+; X64-NEXT:    adcq %r9, %rcx
+; X64-NEXT:    setb %dil
+; X64-NEXT:    movq %rbp, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    movzbl %dil, %ecx
 ; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    addq %r10, %rax
+; X64-NEXT:    addq %r13, %rax
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
-; X64-NEXT:    adcq %r12, %rsi
-; X64-NEXT:    adcq %rbp, %rax
-; X64-NEXT:    adcq %rbx, %rdx
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
+; X64-NEXT:    adcq %r15, %rax
+; X64-NEXT:    adcq %r12, %rdx
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
+; X64-NEXT:    movq (%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload
-; X64-NEXT:    movq (%rsp), %rbp # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
 ; X64-NEXT:    movq %rcx, %r9
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload
 ; X64-NEXT:    movq %rdi, %r10
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Folded Reload
+; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, (%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 8(%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 16(%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 24(%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 32(%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 40(%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 48(%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Reload
 ; X64-NEXT:    movq %rdi, 56(%rcx)
 ; X64-NEXT:    movq %r9, 64(%rcx)
 ; X64-NEXT:    movq %r10, 72(%rcx)
-; X64-NEXT:    movq %rbp, 80(%rcx)
-; X64-NEXT:    movq %rbx, 88(%rcx)
+; X64-NEXT:    movq %rbx, 80(%rcx)
+; X64-NEXT:    movq %rbp, 88(%rcx)
 ; X64-NEXT:    movq %r8, 96(%rcx)
 ; X64-NEXT:    movq %rsi, 104(%rcx)
 ; X64-NEXT:    movq %rax, 112(%rcx)
 ; X64-NEXT:    movq %rdx, 120(%rcx)
-; X64-NEXT:    addq $352, %rsp # imm = 0x160
+; X64-NEXT:    addq $240, %rsp
 ; X64-NEXT:    popq %rbx
 ; X64-NEXT:    popq %r12
 ; X64-NEXT:    popq %r13
 ; X64-NEXT:    popq %r14
 ; X64-NEXT:    popq %r15
 ; X64-NEXT:    popq %rbp
 ; X64-NEXT:    retq
   %av = load i1024, i1024* %a
   %bv = load i1024, i1024* %b
   %r = mul i1024 %av, %bv
   store i1024 %r, i1024* %out
   ret void
 }

llvm/trunk/test/CodeGen/X86/mul-i256.ll

 ; X32-NEXT:    pushl %ebp
 ; X32-NEXT:    .cfi_def_cfa_offset 8
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    .cfi_def_cfa_offset 12
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    .cfi_def_cfa_offset 16
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    .cfi_def_cfa_offset 20
-; X32-NEXT:    subl $88, %esp
-; X32-NEXT:    .cfi_def_cfa_offset 108
+; X32-NEXT:    subl $72, %esp
+; X32-NEXT:    .cfi_def_cfa_offset 92
 ; X32-NEXT:    .cfi_offset %esi, -20
 ; X32-NEXT:    .cfi_offset %edi, -16
 ; X32-NEXT:    .cfi_offset %ebx, -12
 ; X32-NEXT:    .cfi_offset %ebp, -8
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 12(%ecx), %ebp
-; X32-NEXT:    movl 8(%ecx), %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl (%eax), %ebx
+; X32-NEXT:    movl 12(%ecx), %esi
+; X32-NEXT:    movl 8(%ecx), %ebx
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl (%eax), %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %esi, %ecx
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 4(%eax), %ecx
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %ecx, %edi
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl 4(%eax), %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebx
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    adcl %ebp, %edx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl (%esi), %ebp
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl (%edi), %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %ecx, %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl 4(%edi), %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl 4(%esi), %esi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    adcl %ebx, %ecx
 ; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl %ecx, %ebp
 ; X32-NEXT:    movzbl %bl, %eax
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl (%esp), %ecx # 4-byte Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %ebp, %ecx
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl 8(%edi), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 8(%eax), %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %esi, %ecx
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl 12(%edi), %edi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edi, %ecx
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %edi
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 12(%eax), %ecx
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %ebx, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    xorl %ecx, %ecx
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %ebp, %ebx
-; X32-NEXT:    adcl %esi, %eax
-; X32-NEXT:    movl (%esp), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, %ebx
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ebx, %ebp
-; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %ebx, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edi, %ebx
 ; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
 ; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 16(%ecx), %esi
-; X32-NEXT:    imull %esi, %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 16(%ecx), %edi
+; X32-NEXT:    imull %edi, %ebx
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
 ; X32-NEXT:    addl %ebx, %edx
 ; X32-NEXT:    movl 20(%ecx), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    imull %eax, %edi
-; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    imull %eax, %ebp
+; X32-NEXT:    addl %edx, %ebp
 ; X32-NEXT:    movl 24(%ecx), %eax
-; X32-NEXT:    movl %ecx, %ebp
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %ecx, %edi
+; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    movl 28(%ebp), %ebp
-; X32-NEXT:    imull %ebx, %ebp
-; X32-NEXT:    addl %edx, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %edx, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    imull %ebx, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %ecx, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 28(%ecx), %ecx
+; X32-NEXT:    imull %esi, %ecx
+; X32-NEXT:    addl %edx, %ecx
+; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ebp, %ecx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebx, %edi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    setb %cl
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movzbl %bl, %esi
+; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %ebp, %edx
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X32-NEXT:    movl 28(%ebx), %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    imull %esi, %ecx
 ; X32-NEXT:    movl 24(%ebx), %edi
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %edi

 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, 8(%ecx)
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, 12(%ecx)
 ; X32-NEXT:    movl %ebx, 16(%ecx)
 ; X32-NEXT:    movl %esi, 20(%ecx)
 ; X32-NEXT:    movl %eax, 24(%ecx)
 ; X32-NEXT:    movl %edx, 28(%ecx)
-; X32-NEXT:    addl $88, %esp
+; X32-NEXT:    addl $72, %esp
 ; X32-NEXT:    .cfi_def_cfa_offset 20
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    .cfi_def_cfa_offset 16
 ; X32-NEXT:    popl %edi
 ; X32-NEXT:    .cfi_def_cfa_offset 12
 ; X32-NEXT:    popl %ebx
 ; X32-NEXT:    .cfi_def_cfa_offset 8
 ; X32-NEXT:    popl %ebp

llvm/trunk/test/CodeGen/X86/mul-i512.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i386-unknown | FileCheck %s --check-prefix=X32
 ; RUN: llc < %s -mtriple=x86_64-unknown | FileCheck %s --check-prefix=X64
 
 define void @test_512(i512* %a, i512* %b, i512* %out) nounwind {
 ; X32-LABEL: test_512:
 ; X32:       # %bb.0:
 ; X32-NEXT:    pushl %ebp
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
-; X32-NEXT:    subl $244, %esp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 20(%ecx), %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 16(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %ebp
-; X32-NEXT:    xorl %ebx, %ebx
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %ebx
-; X32-NEXT:    movl %ecx, %edi
+; X32-NEXT:    subl $180, %esp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edx
+; X32-NEXT:    movl 28(%edx), %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb %cl
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl 24(%ebp), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    addl %ebx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %edi
+; X32-NEXT:    movl 24(%edx), %ebp
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl (%eax), %edi
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl (%ecx), %eax
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ebx, %ebx
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 4(%ecx), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %esi
-; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %ebp, %ebx
-; X32-NEXT:    movl %ebx, (%esp) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %ebx
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl 8(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl (%ecx), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ebp, %ebp
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl %eax, %edx
-; X32-NEXT:    addl %esi, %edx
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    adcl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 16(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %edi
+; X32-NEXT:    movl 4(%eax), %ecx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %ecx, %ebp
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    adcl %edx, %eax
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl (%esp), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 4(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edi, %edi
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ebx, %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl %ebp, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %ebx
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 8(%eax), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %cl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %edi
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl 20(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 16(%ecx), %ebx
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movzbl %bl, %ebx
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl 24(%esi), %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %esi
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl %edx, %edi
-; X32-NEXT:    addl %ecx, %esi
-; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    adcl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    addl %edx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 28(%eax), %esi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl 20(%ecx), %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebp, %esi
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    setb %cl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ebp, %esi
+; X32-NEXT:    movzbl %cl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 8(%eax), %ebp
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %ebp
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl 12(%eax), %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %ebp, %edi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
+; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl %esi, %ecx
+; X32-NEXT:    movl 8(%esi), %ebx
 ; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl 12(%ecx), %ecx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl (%ecx), %esi
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 12(%eax), %eax
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl 4(%ecx), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %ebx, %esi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %ecx, %esi
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %ebp, %edi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebx, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    addl %edi, %ebp
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    adcl %ecx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebx
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %edi
+; X32-NEXT:    addl %ebp, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ebp
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl (%esp), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 1-byte Folded Reload
+; X32-NEXT:    adcl %edi, %eax
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 16(%eax), %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %edi
 ; X32-NEXT:    addl %ecx, %edi
 ; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 20(%eax), %ecx
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    mull %ebx
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %ecx, %ebx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl (%esp), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl (%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    addl %ebp, %ecx
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 28(%eax), %ebp
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %ebp, (%esp) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %ebp
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl 24(%eax), %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 28(%eax), %ebx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    adcl %ebp, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %ebp
 ; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %edx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebp, %ecx
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl %bl, %esi
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %ebp
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    addl %ebx, %edi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl (%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %edi, %edx
 ; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %ebx, %ecx
+; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    adcl $0, %eax
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
-; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %ecx
-; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl %bl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ecx, %ebx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %esi, %ebx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl %ebp, %edi
 ; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl %ecx, %ebx
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    movzbl %bl, %eax
 ; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %ebp
-; X32-NEXT:    setb %cl
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl (%esp), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    movzbl %cl, %ecx
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edi, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ecx, %edi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl (%esp), %ebx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %ebx, %ecx
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl %esi, %ecx
+; X32-NEXT:    adcl %esi, %ebx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
+; X32-NEXT:    addl %edi, %ecx
+; X32-NEXT:    adcl (%esp), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 1-byte Folded Reload
+; X32-NEXT:    adcl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 32(%ecx), %edi
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    movl 32(%eax), %edi
 ; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %esi, %ebx
+; X32-NEXT:    addl %ecx, %ebx
 ; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 36(%eax), %esi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %esi, %ecx
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl 36(%eax), %ecx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl %ecx, %esi
+; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %edi, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %esi, %ebp
-; X32-NEXT:    movzbl %bl, %eax
-; X32-NEXT:    adcl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %bl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %eax, %edi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    adcl %edx, %esi
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ebx, %esi
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %edi, %ebp
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    movl %ebx, %eax
 ; X32-NEXT:    mull %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %ecx, %ebp
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    movzbl (%esp), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 40(%eax), %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl 40(%eax), %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl (%esp), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebp
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    movl 44(%eax), %ebx
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    movl 44(%eax), %esi
+; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %ebp, %ecx
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %edi, %esi
+; X32-NEXT:    addl %ecx, %esi
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %ecx
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    mull %edx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    adcl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    adcl $0, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %ebp
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    mull (%esp) # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    adcl %edi, %ecx
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    addl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ebp, %ebx
+; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    adcl %ecx, %eax
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
+; X32-NEXT:    movl %edi, %esi
+; X32-NEXT:    imull %eax, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %ebx, %ecx
+; X32-NEXT:    addl %edx, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    imull %eax, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %ebp, %edx
-; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    addl %edx, %esi
-; X32-NEXT:    movl %esi, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    imull %ebx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    imull %edi, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
+; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %eax, %ecx
 ; X32-NEXT:    addl %esi, %edx
-; X32-NEXT:    movl (%esp), %esi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    imull %edi, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    imull %ebp, %esi
 ; X32-NEXT:    addl %edx, %esi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, %edi
-; X32-NEXT:    adcl %ebp, %esi
-; X32-NEXT:    movl %esi, (%esp) # 4-byte Spill
+; X32-NEXT:    addl (%esp), %ecx # 4-byte Folded Reload
+; X32-NEXT:    movl %ecx, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %ebp, %ebx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
 ; X32-NEXT:    adcl $0, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %ebx, %ebp
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    addl %ebx, %eax
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ebp
 ; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X32-NEXT:    movl 60(%edi), %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    imull %eax, %ecx
 ; X32-NEXT:    movl 56(%edi), %esi
+; X32-NEXT:    movl %edi, %ebx
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    addl %ecx, %edx
 ; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
 ; X32-NEXT:    addl %edx, %esi
-; X32-NEXT:    movl 48(%edi), %ebx
-; X32-NEXT:    movl 52(%edi), %ebp
+; X32-NEXT:    movl 48(%edi), %edi
+; X32-NEXT:    movl 52(%ebx), %ebp
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    imull %ebp, %edi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    imull %ebp, %ebx
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %edi, %edx
+; X32-NEXT:    mull %edi
+; X32-NEXT:    addl %ebx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %ebx, %ecx
+; X32-NEXT:    imull %edi, %ecx
 ; X32-NEXT:    addl %edx, %ecx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %esi, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %esi, %edi
-; X32-NEXT:    adcl $0, %ebp
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl %esi, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %edi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    addl %edi, %ebx
-; X32-NEXT:    adcl %ebp, %ecx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %ebp, %edi
+; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    setb %bl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %esi
 ; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
+; X32-NEXT:    movzbl %bl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl %esi, %ecx
-; X32-NEXT:    movl 40(%esi), %ebx
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X32-NEXT:    movl 40(%edi), %ebp
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl 44(%ecx), %ecx
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebp, %edi
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl 44(%edi), %edi
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %ecx
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %ebx
-; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl %esi, %ecx
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    mull %ebp
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl %ebx, %edi
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %esi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %esi, %edx
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
-; X32-NEXT:    movl 32(%esi), %edi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movl %eax, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    mull %ebp
+; X32-NEXT:    addl %esi, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movzbl %cl, %eax
+; X32-NEXT:    adcl %eax, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl 36(%esi), %esi
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
+; X32-NEXT:    movl 32(%ecx), %esi
 ; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl 36(%ecx), %ebp
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ecx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl $0, %ebx
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    addl %ebp, %eax
+; X32-NEXT:    addl %ecx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl %ebx, %edi
-; X32-NEXT:    setb %bl
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %edi, %esi
-; X32-NEXT:    movzbl %bl, %eax
-; X32-NEXT:    adcl %eax, %ebp
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    adcl %edx, %eax
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl %ebp, %eax
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    setb %cl
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %edi, %ebx
+; X32-NEXT:    movzbl %cl, %eax
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
+; X32-NEXT:    adcl (%esp), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
 ; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl $0, (%esp) # 4-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    adcl $0, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %edi
+; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    adcl %esi, %edi
+; X32-NEXT:    adcl %edi, %esi
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    addl %edi, %eax
-; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %esi
-; X32-NEXT:    adcl %edx, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
-; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    mull %ecx
+; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %eax, %edi
+; X32-NEXT:    addl %esi, %edi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %ecx
+; X32-NEXT:    addl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl (%esp), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl %ebp, (%esp) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edi
 ; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
-; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl (%esp), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    movl %ecx, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, (%esp) # 4-byte Spill
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %esi
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl (%esp), %ebp # 4-byte Folded Reload
-; X32-NEXT:    adcl $0, %esi
-; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %ecx
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, %ebp
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %esi, %eax
+; X32-NEXT:    mull {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Reload
+; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    adcl %esi, %ecx
-; X32-NEXT:    setb (%esp) # 1-byte Folded Spill
-; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebx
-; X32-NEXT:    addl %ecx, %eax
-; X32-NEXT:    movzbl (%esp), %ecx # 1-byte Folded Reload
-; X32-NEXT:    adcl %ecx, %edx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    addl %eax, %edi
-; X32-NEXT:    adcl %edx, %ecx
+; X32-NEXT:    adcl %ebx, %esi
+; X32-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
-; X32-NEXT:    addl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
+; X32-NEXT:    adcl %eax, %edx
+; X32-NEXT:    addl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
+; X32-NEXT:    adcl %ecx, %ebp
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X32-NEXT:    adcl %eax, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl $0, %ecx
-; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl %eax, %esi
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl $0, %edx
+; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl 48(%ecx), %ebp
-; X32-NEXT:    imull %ebp, %ebx
-; X32-NEXT:    movl %ebp, %eax
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    movl 48(%ecx), %edi
+; X32-NEXT:    imull %edi, %ebx
+; X32-NEXT:    movl %edi, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    addl %ebx, %edx
 ; X32-NEXT:    movl 52(%ecx), %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    imull %eax, %edi
-; X32-NEXT:    addl %edx, %edi
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    imull %eax, %esi
+; X32-NEXT:    addl %edx, %esi
+; X32-NEXT:    movl %esi, %ebp
 ; X32-NEXT:    movl 56(%ecx), %eax
-; X32-NEXT:    movl %ecx, %ebx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
-; X32-NEXT:    imull %esi, %edi
+; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    imull %ebx, %esi
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
+; X32-NEXT:    addl %esi, %edx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X32-NEXT:    movl 60(%esi), %esi
+; X32-NEXT:    imull %ecx, %esi
+; X32-NEXT:    addl %edx, %esi
+; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    addl %edi, %edx
-; X32-NEXT:    movl 60(%ebx), %ebx
+; X32-NEXT:    adcl %ebp, %esi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    imull %ecx, %ebx
-; X32-NEXT:    addl %edx, %ebx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    addl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Folded Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    mull %edi
+; X32-NEXT:    movl %edx, %ebp
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    mull %edi
 ; X32-NEXT:    movl %edx, %ecx
-; X32-NEXT:    movl %eax, %edi
-; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
+; X32-NEXT:    movl %eax, %ebx
+; X32-NEXT:    addl %ebp, %ebx
 ; X32-NEXT:    adcl $0, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    addl %edi, %eax
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    addl %ebx, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ecx, %esi
+; X32-NEXT:    adcl %ecx, %edi
 ; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ebp
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movzbl %cl, %ecx
 ; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl %ebx, %edx
+; X32-NEXT:    adcl %esi, %edx
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
-; X32-NEXT:    imull %ebp, %edi
-; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    imull %esi, %edi
+; X32-NEXT:    movl %esi, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %eax, %esi
+; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl %edi, %edx
 ; X32-NEXT:    imull {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    addl %edx, %ecx
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
-; X32-NEXT:    imull %ebx, %ecx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    mull %edi
+; X32-NEXT:    imull %edi, %ecx
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    addl %ecx, %edx
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
-; X32-NEXT:    imull %edi, %ecx
+; X32-NEXT:    imull %ebx, %ecx
 ; X32-NEXT:    addl %edx, %ecx
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    addl %ebp, %eax
 ; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    movl %ebx, %eax
+; X32-NEXT:    movl %ebx, %ebp
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ebx
+; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl %edi, %eax
-; X32-NEXT:    mull %ebp
+; X32-NEXT:    mull %esi
 ; X32-NEXT:    movl %edx, %esi
-; X32-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    movl %ebx, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    addl %esi, %ecx
-; X32-NEXT:    adcl $0, %edi
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
+; X32-NEXT:    addl %ebx, %ecx
+; X32-NEXT:    adcl $0, %esi
+; X32-NEXT:    movl %ebp, %eax
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Reload
 ; X32-NEXT:    mull %ebx
-; X32-NEXT:    movl %edx, %esi
+; X32-NEXT:    movl %edx, %edi
 ; X32-NEXT:    movl %eax, %ebp
 ; X32-NEXT:    addl %ecx, %ebp
-; X32-NEXT:    adcl %edi, %esi
+; X32-NEXT:    adcl %esi, %edi
 ; X32-NEXT:    setb %cl
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Reload
 ; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %esi, %eax
+; X32-NEXT:    addl %edi, %eax
 ; X32-NEXT:    movzbl %cl, %ecx
 ; X32-NEXT:    adcl %ecx, %ebx
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
+; X32-NEXT:    movl (%esp), %edi # 4-byte Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    movl %ebp, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ecx # 4-byte Folded Reload
 ; X32-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
-; X32-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    movl %edx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Folded Reload
+; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %edx # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebp # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
 ; X32-NEXT:    adcl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, (%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 4(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 8(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 12(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 16(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 20(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 24(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 28(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 32(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 36(%ecx)
-; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
-; X32-NEXT:    movl %edi, 40(%ecx)
-; X32-NEXT:    movl %esi, 44(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, (%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 4(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 8(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 12(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 16(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 20(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 24(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 28(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 32(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 36(%ecx)
+; X32-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %esi # 4-byte Reload
+; X32-NEXT:    movl %esi, 40(%ecx)
+; X32-NEXT:    movl %edi, 44(%ecx)
 ; X32-NEXT:    movl %edx, 48(%ecx)
 ; X32-NEXT:    movl %ebp, 52(%ecx)
 ; X32-NEXT:    movl %eax, 56(%ecx)
 ; X32-NEXT:    movl %ebx, 60(%ecx)
-; X32-NEXT:    addl $244, %esp
+; X32-NEXT:    addl $180, %esp
 ; X32-NEXT:    popl %esi
 ; X32-NEXT:    popl %edi
 ; X32-NEXT:    popl %ebx
 ; X32-NEXT:    popl %ebp
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: test_512:
 ; X64:       # %bb.0:
 ; X64-NEXT:    pushq %rbp
 ; X64-NEXT:    pushq %r15
 ; X64-NEXT:    pushq %r14
 ; X64-NEXT:    pushq %r13
 ; X64-NEXT:    pushq %r12
 ; X64-NEXT:    pushq %rbx
 ; X64-NEXT:    pushq %rax
 ; X64-NEXT:    movq %rdx, (%rsp) # 8-byte Spill
-; X64-NEXT:    movq 24(%rdi), %r11
-; X64-NEXT:    movq 16(%rdi), %r15
+; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq (%rdi), %r9
+; X64-NEXT:    movq 8(%rdi), %r15
+; X64-NEXT:    movq 24(%rdi), %r12
+; X64-NEXT:    movq 16(%rdi), %rax
 ; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq (%rsi), %rdx
-; X64-NEXT:    movq 8(%rsi), %rbp
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    mulq %rdx
-; X64-NEXT:    movq %rdx, %r9
-; X64-NEXT:    movq %rax, %r8
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    movq %r11, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rsi
-; X64-NEXT:    movq %rsi, %r10
-; X64-NEXT:    movq %rdx, %rbx
+; X64-NEXT:    movq (%rsi), %rdi
+; X64-NEXT:    movq 8(%rsi), %r14
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %r9, %rsi
-; X64-NEXT:    adcq $0, %rbx
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    movq %r15, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rsi, %r9
-; X64-NEXT:    adcq %rbx, %rcx
-; X64-NEXT:    setb %al
-; X64-NEXT:    movzbl %al, %ebx
-; X64-NEXT:    movq %r11, %rax
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rbp, %r14
-; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rcx, %rbp
-; X64-NEXT:    adcq %rbx, %rsi
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    movq %r10, %rbx
-; X64-NEXT:    movq %r10, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, %r13
-; X64-NEXT:    movq %rax, %r10
-; X64-NEXT:    movq %r15, %rax
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rax, %r15
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    addq %r10, %r15
-; X64-NEXT:    adcq %r13, %rdx
-; X64-NEXT:    addq %rbp, %r15
-; X64-NEXT:    adcq %rsi, %rdx
-; X64-NEXT:    movq %rdx, %r12
-; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq (%rdi), %rcx
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %rbx
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq 8(%rdi), %rdi
-; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    mulq %rbx
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    movq %r12, %rax
+; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %r11, %rsi
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
 ; X64-NEXT:    adcq $0, %rbp
-; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rsi, %rax
 ; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rbx, %r10
+; X64-NEXT:    adcq %rbp, %rcx
+; X64-NEXT:    setb %al
+; X64-NEXT:    movzbl %al, %esi
+; X64-NEXT:    movq %r12, %rax
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %r8
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rcx, %r13
+; X64-NEXT:    adcq %rsi, %r8
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    addq %rsi, %rax
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
+; X64-NEXT:    adcq $0, %rbx
+; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %r9, %r12
+; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    mulq %r14
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    addq %rbp, %rax
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rbx
-; X64-NEXT:    setb %r11b
-; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    adcq %rbx, %rcx
+; X64-NEXT:    setb %sil
+; X64-NEXT:    movq %r15, %rdi
+; X64-NEXT:    movq %r15, %rax
 ; X64-NEXT:    mulq %r14
-; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rbx, %rbp
-; X64-NEXT:    movzbl %r11b, %eax
-; X64-NEXT:    adcq %rax, %rsi
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    mulq %rdx
-; X64-NEXT:    movq %rdx, %r14
-; X64-NEXT:    movq %rax, %r11
-; X64-NEXT:    addq %rax, %r10
-; X64-NEXT:    adcq %rdx, %r13
-; X64-NEXT:    addq %rbp, %r10
-; X64-NEXT:    adcq %rsi, %r13
-; X64-NEXT:    addq %r8, %r10
-; X64-NEXT:    adcq %r9, %r13
-; X64-NEXT:    adcq $0, %r15
-; X64-NEXT:    adcq $0, %r12
-; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rdx, %r15
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rcx, %rbx
+; X64-NEXT:    movzbl %sil, %eax
+; X64-NEXT:    adcq %rax, %r15
+; X64-NEXT:    addq %r11, %rbx
+; X64-NEXT:    adcq %r10, %r15
+; X64-NEXT:    adcq $0, %r13
+; X64-NEXT:    movq %r8, %r14
+; X64-NEXT:    adcq $0, %r14
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
 ; X64-NEXT:    movq 16(%rsi), %r8
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r9
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %r9, %rax
 ; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rdx, %r10
+; X64-NEXT:    movq %rax, %r9
+; X64-NEXT:    movq %rdi, %rax
+; X64-NEXT:    movq %rdi, %r11
+; X64-NEXT:    movq %rdi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %rbp
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %rdi, %rbx
-; X64-NEXT:    adcq $0, %rbp
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %r10, %rbp
+; X64-NEXT:    adcq $0, %rcx
 ; X64-NEXT:    movq 24(%rsi), %rdi
-; X64-NEXT:    movq %r9, %rax
+; X64-NEXT:    movq %r12, %rax
 ; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %rbp, %rsi
-; X64-NEXT:    setb %bpl
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rdx, %rbx
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rsi, %r9
-; X64-NEXT:    movzbl %bpl, %eax
-; X64-NEXT:    adcq %rax, %rbx
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    xorl %ecx, %ecx
-; X64-NEXT:    mulq %rcx
-; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    addq %rbp, %rax
 ; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rax, %r11
-; X64-NEXT:    adcq %rdx, %r14
-; X64-NEXT:    addq %r9, %r11
-; X64-NEXT:    adcq %rbx, %r14
-; X64-NEXT:    addq %r10, %r12
-; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r13, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Folded Spill
+; X64-NEXT:    adcq %rcx, %rsi
+; X64-NEXT:    setb %cl
+; X64-NEXT:    movq %r11, %rax
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rsi, %r12
+; X64-NEXT:    movzbl %cl, %eax
+; X64-NEXT:    adcq %rax, %r11
+; X64-NEXT:    addq %rbx, %r9
+; X64-NEXT:    movq %r9, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r15, %rbp
+; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %r12
 ; X64-NEXT:    adcq $0, %r11
-; X64-NEXT:    adcq $0, %r14
-; X64-NEXT:    addq %r15, %r11
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Folded Reload
+; X64-NEXT:    addq %r13, %r12
+; X64-NEXT:    adcq %r14, %r11
 ; X64-NEXT:    setb %r9b
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rdx, %r10
-; X64-NEXT:    movq %rax, %r12
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    movq %rdx, %rcx
+; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
+; X64-NEXT:    movq %r10, %rax
 ; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r10, %rbx
+; X64-NEXT:    movq %rax, %rbp
+; X64-NEXT:    addq %rcx, %rbp
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %rcx, %rax
+; X64-NEXT:    movq %rbx, %rax
 ; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    addq %rbx, %rax
-; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rbp, %rax
+; X64-NEXT:    movq %rax, %rbp
 ; X64-NEXT:    adcq %rsi, %rcx
-; X64-NEXT:    setb %sil
-; X64-NEXT:    movq %r15, %rax
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %r10, %rax
 ; X64-NEXT:    mulq %rdi
 ; X64-NEXT:    addq %rcx, %rax
-; X64-NEXT:    movzbl %sil, %ecx
+; X64-NEXT:    movzbl %bl, %ecx
 ; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
-; X64-NEXT:    addq %rbp, %rsi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Folded Reload
-; X64-NEXT:    addq %rax, %rsi
-; X64-NEXT:    adcq %rdx, %rcx
-; X64-NEXT:    addq %r11, %r12
-; X64-NEXT:    movq %r12, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq %r14, %rbx
-; X64-NEXT:    movq %rbx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movzbl %r9b, %eax
-; X64-NEXT:    adcq %rax, %rsi
-; X64-NEXT:    movq %rsi, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    adcq $0, %rcx
-; X64-NEXT:    movq %rcx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    addq %r12, %r14
+; X64-NEXT:    movq %r14, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq %r11, %rbp
+; X64-NEXT:    movq %rbp, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movzbl %r9b, %ecx
+; X64-NEXT:    adcq %rcx, %rax
+; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    adcq $0, %rdx
+; X64-NEXT:    movq %rdx, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    movq 32(%rcx), %rsi
-; X64-NEXT:    imulq %rsi, %rdi
-; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq 32(%rcx), %r10
+; X64-NEXT:    imulq %r10, %rdi
+; X64-NEXT:    movq %r10, %rax
 ; X64-NEXT:    mulq %r8
-; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    movq %rax, %r11
 ; X64-NEXT:    addq %rdi, %rdx
 ; X64-NEXT:    movq 40(%rcx), %r9
 ; X64-NEXT:    imulq %r9, %r8
 ; X64-NEXT:    addq %rdx, %r8
 ; X64-NEXT:    movq 48(%rcx), %rax
-; X64-NEXT:    movq %rcx, %rbx
 ; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rcx # 8-byte Reload
-; X64-NEXT:    imulq %rcx, %rdi
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    mulq %rbp
-; X64-NEXT:    movq %rax, %r14
-; X64-NEXT:    addq %rdi, %rdx
-; X64-NEXT:    movq 56(%rbx), %rbx
-; X64-NEXT:    imulq %rbp, %rbx
-; X64-NEXT:    addq %rdx, %rbx
-; X64-NEXT:    addq %r10, %r14
-; X64-NEXT:    adcq %r8, %rbx
-; X64-NEXT:    movq %rbp, %rax
-; X64-NEXT:    movq %rbp, %r10
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbx # 8-byte Reload
+; X64-NEXT:    imulq %rbx, %rdi
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Reload
 ; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rdi, %rdx
+; X64-NEXT:    movq 56(%rcx), %rbp
+; X64-NEXT:    imulq %rsi, %rbp
+; X64-NEXT:    addq %rdx, %rbp
+; X64-NEXT:    addq %r11, %r12
+; X64-NEXT:    adcq %r8, %rbp
+; X64-NEXT:    movq %rsi, %rax
+; X64-NEXT:    movq %rsi, %rcx
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rdi
 ; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
-; X64-NEXT:    movq %rcx, %rax
-; X64-NEXT:    movq %rcx, %r8
-; X64-NEXT:    mulq %rsi
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    movq %rbx, %r11
+; X64-NEXT:    mulq %r10
 ; X64-NEXT:    movq %rdx, %rsi
-; X64-NEXT:    movq %rax, %rbp
-; X64-NEXT:    addq %rdi, %rbp
+; X64-NEXT:    movq %rax, %rbx
+; X64-NEXT:    addq %rdi, %rbx
 ; X64-NEXT:    adcq $0, %rsi
-; X64-NEXT:    movq %r10, %rax
+; X64-NEXT:    movq %rcx, %rax
 ; X64-NEXT:    mulq %r9
 ; X64-NEXT:    movq %rdx, %rdi
-; X64-NEXT:    movq %rax, %r13
-; X64-NEXT:    addq %rbp, %r13
+; X64-NEXT:    movq %rax, %r10
+; X64-NEXT:    addq %rbx, %r10
 ; X64-NEXT:    adcq %rsi, %rdi
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r8, %rax
+; X64-NEXT:    setb %bl
+; X64-NEXT:    movq %r11, %rax
 ; X64-NEXT:    mulq %r9
-; X64-NEXT:    movq %rdx, %r11
-; X64-NEXT:    movq %rax, %r9
-; X64-NEXT:    addq %rdi, %r9
-; X64-NEXT:    movzbl %cl, %eax
-; X64-NEXT:    adcq %rax, %r11
-; X64-NEXT:    addq %r14, %r9
-; X64-NEXT:    adcq %rbx, %r11
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
-; X64-NEXT:    movq 56(%rbp), %rcx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r10 # 8-byte Reload
-; X64-NEXT:    imulq %r10, %rcx
-; X64-NEXT:    movq 48(%rbp), %rbx
-; X64-NEXT:    movq %r10, %rax
-; X64-NEXT:    mulq %rbx
-; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    addq %rcx, %rdx
-; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r15 # 8-byte Reload
-; X64-NEXT:    imulq %r15, %rbx
-; X64-NEXT:    addq %rdx, %rbx
-; X64-NEXT:    movq 32(%rbp), %rdi
-; X64-NEXT:    movq 40(%rbp), %r8
+; X64-NEXT:    movq %rdx, %r15
+; X64-NEXT:    movq %rax, %r13
+; X64-NEXT:    addq %rdi, %r13
+; X64-NEXT:    movzbl %bl, %eax
+; X64-NEXT:    adcq %rax, %r15
+; X64-NEXT:    addq %r12, %r13
+; X64-NEXT:    adcq %rbp, %r15
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Reload
+; X64-NEXT:    movq 56(%rdx), %rcx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    movq %rax, %rcx
-; X64-NEXT:    imulq %r8, %rcx
-; X64-NEXT:    mulq %rdi
-; X64-NEXT:    movq %rax, %r14
+; X64-NEXT:    imulq %rax, %rcx
+; X64-NEXT:    movq 48(%rdx), %rbp
+; X64-NEXT:    movq %rdx, %rsi
+; X64-NEXT:    movq %rax, %r11
+; X64-NEXT:    mulq %rbp
+; X64-NEXT:    movq %rax, %r12
 ; X64-NEXT:    addq %rcx, %rdx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r8 # 8-byte Reload
+; X64-NEXT:    imulq %r8, %rbp
+; X64-NEXT:    addq %rdx, %rbp
+; X64-NEXT:    movq 32(%rsi), %rdi
+; X64-NEXT:    movq 40(%rsi), %rbx
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Reload
-; X64-NEXT:    imulq %rdi, %rax
-; X64-NEXT:    addq %rdx, %rax
-; X64-NEXT:    addq %rsi, %r14
-; X64-NEXT:    adcq %rbx, %rax
-; X64-NEXT:    movq %rax, {{[-0-9]+}}(%r{{[sb]}}p) # 8-byte Spill
+; X64-NEXT:    movq %rax, %rsi
+; X64-NEXT:    imulq %rbx, %rsi
+; X64-NEXT:    mulq %rdi
+; X64-NEXT:    movq %rax, %rcx
+; X64-NEXT:    addq %rsi, %rdx
+; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %r14 # 8-byte Reload
+; X64-NEXT:    imulq %rdi, %r14
+; X64-NEXT:    addq %rdx, %r14
+; X64-NEXT:    addq %r12, %rcx
+; X64-NEXT:    adcq %rbp, %r14
 ; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %r12
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %rbp
 ; X64-NEXT:    movq %rax, %rsi
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %r10
-; X64-NEXT:    movq %rdx, %rcx
-; X64-NEXT:    movq %rax, %rbx
-; X64-NEXT:    addq %r12, %rbx
-; X64-NEXT:    adcq $0, %rcx
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r11
+; X64-NEXT:    movq %rdx, %r11
+; X64-NEXT:    movq %rax, %r12
+; X64-NEXT:    addq %rbp, %r12
+; X64-NEXT:    adcq $0, %r11
 ; X64-NEXT:    movq %rdi, %rax
-; X64-NEXT:    mulq %r15
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    movq %rdx, %rbp
 ; X64-NEXT:    movq %rax, %rdi
-; X64-NEXT:    addq %rbx, %rdi
-; X64-NEXT:    adcq %rcx, %rbp
-; X64-NEXT:    setb %cl
-; X64-NEXT:    movq %r8, %rax
-; X64-NEXT:    mulq %r15
+; X64-NEXT:    addq %r12, %rdi
+; X64-NEXT:    adcq %r11, %rbp
+; X64-NEXT:    setb %r9b
+; X64-NEXT:    movq %rbx, %rax
+; X64-NEXT:    mulq %r8
 ; X64-NEXT:    addq %rbp, %rax
-; X64-NEXT:    movzbl %cl, %ecx
-; X64-NEXT:    adcq %rcx, %rdx
-; X64-NEXT:    addq %r14, %rax
-; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
+; X64-NEXT:    movzbl %r9b, %ebp
+; X64-NEXT:    adcq %rbp, %rdx
+; X64-NEXT:    addq %rcx, %rax
+; X64-NEXT:    adcq %r14, %rdx
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
-; X64-NEXT:    adcq %r13, %rdi
-; X64-NEXT:    adcq %r9, %rax
-; X64-NEXT:    adcq %r11, %rdx
+; X64-NEXT:    adcq %r10, %rdi
+; X64-NEXT:    adcq %r13, %rax
+; X64-NEXT:    adcq %r15, %rdx
 ; X64-NEXT:    addq {{[-0-9]+}}(%r{{[sb]}}p), %rsi # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdi # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rax # 8-byte Folded Reload
 ; X64-NEXT:    adcq {{[-0-9]+}}(%r{{[sb]}}p), %rdx # 8-byte Folded Reload
 ; X64-NEXT:    movq (%rsp), %rcx # 8-byte Reload
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload
 ; X64-NEXT:    movq %rbp, (%rcx)
 ; X64-NEXT:    movq {{[-0-9]+}}(%r{{[sb]}}p), %rbp # 8-byte Reload

llvm/trunk/test/CodeGen/X86/mul128.ll

 ; X64-NEXT:    shlq $32, %rcx
 ; X64-NEXT:    addq %rcx, %rdx
 ; X64-NEXT:    movq %rax, {{.*}}(%rip)
 ; X64-NEXT:    movq %rdx, aaa+{{.*}}(%rip)
 ; X64-NEXT:    retq
 ;
 ; X86-LABEL: PR13897:
 ; X86:       # %bb.0: # %"0x0"
-; X86-NEXT:    pushl %ebp
-; X86-NEXT:    pushl %ebx
-; X86-NEXT:    pushl %edi
-; X86-NEXT:    pushl %esi
-; X86-NEXT:    pushl %eax
 ; X86-NEXT:    movl bbb, %eax
-; X86-NEXT:    movl %eax, (%esp) # 4-byte Spill
-; X86-NEXT:    movl $1, %ebx
-; X86-NEXT:    mull %ebx
-; X86-NEXT:    movl %edx, %esi
-; X86-NEXT:    movl %eax, %edi
-; X86-NEXT:    xorl %eax, %eax
-; X86-NEXT:    mull %ebx
-; X86-NEXT:    movl %eax, %ebx
-; X86-NEXT:    addl %esi, %ebx
-; X86-NEXT:    movl %edx, %ebp
-; X86-NEXT:    adcl $0, %ebp
-; X86-NEXT:    addl %edi, %ebx
-; X86-NEXT:    adcl %esi, %ebp
-; X86-NEXT:    setb %cl
-; X86-NEXT:    addl %eax, %ebp
-; X86-NEXT:    movzbl %cl, %eax
-; X86-NEXT:    adcl %edx, %eax
-; X86-NEXT:    addl %edi, %ebp
-; X86-NEXT:    adcl %esi, %eax
-; X86-NEXT:    addl (%esp), %eax # 4-byte Folded Reload
-; X86-NEXT:    movl %edi, aaa
-; X86-NEXT:    movl %ebx, aaa+4
-; X86-NEXT:    movl %ebp, aaa+8
 ; X86-NEXT:    movl %eax, aaa+12
-; X86-NEXT:    addl $4, %esp
-; X86-NEXT:    popl %esi
-; X86-NEXT:    popl %edi
-; X86-NEXT:    popl %ebx
-; X86-NEXT:    popl %ebp
+; X86-NEXT:    movl %eax, aaa+8
+; X86-NEXT:    movl %eax, aaa+4
+; X86-NEXT:    movl %eax, aaa
 ; X86-NEXT:    retl
 "0x0":
   %0 = load i128, i128* @bbb
   %1 = and i128 %0, 4294967295
   %2 = shl i128 %0, 96
   %3 = mul i128 %1, 18446744078004518913
   %4 = add i128 %3, %2
   store i128 %4, i128* @aaa
   ret void
 }

llvm/trunk/test/CodeGen/X86/overflow.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X32
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=X64
 
 define i128 @mulhioverflow(i64 %a, i64 %b, i64 %c) nounwind {
 ; X32-LABEL: mulhioverflow:
 ; X32:       # %bb.0:
 ; X32-NEXT:    pushl %ebp
 ; X32-NEXT:    pushl %ebx
 ; X32-NEXT:    pushl %edi
 ; X32-NEXT:    pushl %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %edi
-; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    movl %esi, %eax
-; X32-NEXT:    mull %ebp
-; X32-NEXT:    movl %edx, %ebp
-; X32-NEXT:    movl %eax, %esi
-; X32-NEXT:    addl %ebx, %esi
-; X32-NEXT:    adcl $0, %ebp
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %ebx
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X32-NEXT:    movl %ecx, %eax
-; X32-NEXT:    mull %edi
+; X32-NEXT:    mull %ebx
+; X32-NEXT:    movl %edx, %edi
+; X32-NEXT:    movl %ebp, %eax
+; X32-NEXT:    mull %ebx
 ; X32-NEXT:    movl %edx, %ebx
-; X32-NEXT:    addl %esi, %eax
-; X32-NEXT:    adcl %ebp, %ebx
-; X32-NEXT:    setb %al
-; X32-NEXT:    movzbl %al, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    mull %edi
-; X32-NEXT:    movl %edx, %esi
 ; X32-NEXT:    movl %eax, %ebp
-; X32-NEXT:    addl %ebx, %ebp
-; X32-NEXT:    adcl %ecx, %esi
-; X32-NEXT:    xorl %ecx, %ecx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    movl %edx, %edi
-; X32-NEXT:    movl %eax, %ebx
-; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-NEXT:    mull %ecx
-; X32-NEXT:    addl %ebx, %eax
-; X32-NEXT:    adcl %edi, %edx
+; X32-NEXT:    addl %edi, %ebp
+; X32-NEXT:    adcl $0, %ebx
+; X32-NEXT:    movl %ecx, %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    movl %edx, %ecx
 ; X32-NEXT:    addl %ebp, %eax
-; X32-NEXT:    adcl %esi, %edx
+; X32-NEXT:    adcl %ebx, %ecx
+; X32-NEXT:    setb %bl
+; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
+; X32-NEXT:    mull %esi
+; X32-NEXT:    addl %ecx, %eax
+; X32-NEXT:    movzbl %bl, %ecx
+; X32-NEXT:    adcl %ecx, %edx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %ecx
 ; X32-NEXT:    andl $1, %ecx
 ; X32-NEXT:    addl %eax, %ecx
 ; X32-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    movl %ecx, (%eax)
 ; X32-NEXT:    adcl $0, %edx
 ; X32-NEXT:    movl %edx, 4(%eax)
 ; X32-NEXT:    setb %cl

llvm/trunk/test/CodeGen/X86/umulo-128-legalisation-lowering.ll

 ; X86-NEXT:    pushl %ebp
 ; X86-NEXT:    .cfi_def_cfa_offset 8
 ; X86-NEXT:    pushl %ebx
 ; X86-NEXT:    .cfi_def_cfa_offset 12
 ; X86-NEXT:    pushl %edi
 ; X86-NEXT:    .cfi_def_cfa_offset 16
 ; X86-NEXT:    pushl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 20
-; X86-NEXT:    subl $28, %esp
-; X86-NEXT:    .cfi_def_cfa_offset 48
+; X86-NEXT:    subl $24, %esp
+; X86-NEXT:    .cfi_def_cfa_offset 44
 ; X86-NEXT:    .cfi_offset %esi, -20
 ; X86-NEXT:    .cfi_offset %edi, -16
 ; X86-NEXT:    .cfi_offset %ebx, -12
 ; X86-NEXT:    .cfi_offset %ebp, -8
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
 ; X86-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    mull %ebx
 ; X86-NEXT:    movl %eax, %ecx
 ; X86-NEXT:    seto {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X86-NEXT:    movl %esi, %eax
 ; X86-NEXT:    mull %edi
 ; X86-NEXT:    movl %eax, %esi
 ; X86-NEXT:    seto {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
 ; X86-NEXT:    addl %ecx, %esi
 ; X86-NEXT:    movl %edi, %eax
 ; X86-NEXT:    mull %ebx
 ; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl %edx, %ecx
 ; X86-NEXT:    addl %esi, %ecx
 ; X86-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %ebp
-; X86-NEXT:    movl %eax, %edi
+; X86-NEXT:    mull %esi
+; X86-NEXT:    movl %esi, %ebx
+; X86-NEXT:    movl %eax, %esi
 ; X86-NEXT:    seto {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebx
-; X86-NEXT:    movl %ebx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %edi
+; X86-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %ebx
-; X86-NEXT:    movl %eax, %esi
+; X86-NEXT:    mull %edi
+; X86-NEXT:    movl %eax, %ebp
 ; X86-NEXT:    seto {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X86-NEXT:    addl %edi, %esi
-; X86-NEXT:    movl %ebx, %eax
-; X86-NEXT:    mull %ebp
-; X86-NEXT:    movl %ebp, %ebx
+; X86-NEXT:    addl %esi, %ebp
+; X86-NEXT:    movl %edi, %eax
+; X86-NEXT:    mull %ebx
+; X86-NEXT:    movl %ebx, %esi
+; X86-NEXT:    movl %eax, %ebx
 ; X86-NEXT:    movl %edx, %edi
-; X86-NEXT:    addl %esi, %edi
+; X86-NEXT:    addl %ebp, %edi
 ; X86-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X86-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
-; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %ebx # 4-byte Folded Reload
 ; X86-NEXT:    adcl %ecx, %edi
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-NEXT:    movl %ebp, %eax
-; X86-NEXT:    movl %ebx, %ecx
-; X86-NEXT:    mull %ebx
-; X86-NEXT:    movl %edx, %ebx
-; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %ecx
+; X86-NEXT:    movl %esi, %ecx
+; X86-NEXT:    mull %esi
 ; X86-NEXT:    movl %edx, %esi
-; X86-NEXT:    movl %eax, %ecx
-; X86-NEXT:    addl %ebx, %ecx
-; X86-NEXT:    adcl $0, %esi
-; X86-NEXT:    movl %ebp, %eax
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
-; X86-NEXT:    mull %ebp
-; X86-NEXT:    movl %edx, %ebx
-; X86-NEXT:    addl %ecx, %eax
 ; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
-; X86-NEXT:    adcl %esi, %ebx
-; X86-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %ebp
-; X86-NEXT:    movl %edx, %ecx
-; X86-NEXT:    movl %eax, %esi
-; X86-NEXT:    addl %ebx, %esi
-; X86-NEXT:    movzbl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 1-byte Folded Reload
-; X86-NEXT:    adcl %eax, %ecx
-; X86-NEXT:    xorl %edx, %edx
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    mull %edx
+; X86-NEXT:    mull %ecx
 ; X86-NEXT:    movl %edx, %ebp
-; X86-NEXT:    movl %eax, %ebx
+; X86-NEXT:    movl %eax, %ecx
+; X86-NEXT:    addl %esi, %ecx
+; X86-NEXT:    adcl $0, %ebp
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT:    xorl %edx, %edx
-; X86-NEXT:    mull %edx
-; X86-NEXT:    addl %ebx, %eax
-; X86-NEXT:    adcl %ebp, %edx
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
+; X86-NEXT:    mull %esi
+; X86-NEXT:    movl %edx, %esi
+; X86-NEXT:    addl %ecx, %eax
+; X86-NEXT:    movl %eax, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
+; X86-NEXT:    adcl %ebp, %esi
+; X86-NEXT:    setb %cl
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %ebp
+; X86-NEXT:    movl %ebp, %eax
+; X86-NEXT:    mull {{[0-9]+}}(%esp)
 ; X86-NEXT:    addl %esi, %eax
+; X86-NEXT:    movzbl %cl, %ecx
 ; X86-NEXT:    adcl %ecx, %edx
-; X86-NEXT:    addl {{[-0-9]+}}(%e{{[sb]}}p), %eax # 4-byte Folded Reload
+; X86-NEXT:    addl %ebx, %eax
 ; X86-NEXT:    adcl %edi, %edx
 ; X86-NEXT:    setb {{[-0-9]+}}(%e{{[sb]}}p) # 1-byte Folded Spill
-; X86-NEXT:    cmpl $0, {{[0-9]+}}(%esp)
+; X86-NEXT:    testl %ebp, %ebp
 ; X86-NEXT:    setne %cl
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %esi
 ; X86-NEXT:    testl %esi, %esi
 ; X86-NEXT:    setne %ch
 ; X86-NEXT:    andb %cl, %ch
 ; X86-NEXT:    movb {{[-0-9]+}}(%e{{[sb]}}p), %cl # 1-byte Reload
 ; X86-NEXT:    orb {{[-0-9]+}}(%e{{[sb]}}p), %cl # 1-byte Folded Reload
 ; X86-NEXT:    orb {{[-0-9]+}}(%e{{[sb]}}p), %cl # 1-byte Folded Reload

 ; X86-NEXT:    andb %bl, %al
 ; X86-NEXT:    orb {{[-0-9]+}}(%e{{[sb]}}p), %al # 1-byte Folded Reload
 ; X86-NEXT:    orb {{[-0-9]+}}(%e{{[sb]}}p), %al # 1-byte Folded Reload
 ; X86-NEXT:    orb {{[-0-9]+}}(%e{{[sb]}}p), %al # 1-byte Folded Reload
 ; X86-NEXT:    orb %bh, %al
 ; X86-NEXT:    andb $1, %al
 ; X86-NEXT:    movb %al, 16(%ecx)
 ; X86-NEXT:    movl %ecx, %eax
-; X86-NEXT:    addl $28, %esp
+; X86-NEXT:    addl $24, %esp
 ; X86-NEXT:    .cfi_def_cfa_offset 20
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    .cfi_def_cfa_offset 16
 ; X86-NEXT:    popl %edi
 ; X86-NEXT:    .cfi_def_cfa_offset 12
 ; X86-NEXT:    popl %ebx
 ; X86-NEXT:    .cfi_def_cfa_offset 8
 ; X86-NEXT:    popl %ebp