[DAGCombiner] allow narrowing of add followed by truncate

[spatel]

Summary

trunc (add X, C ) --> add (trunc X), C'

If we're throwing away the top bits of an 'add' instruction, do it in the narrow destination type.
This makes the truncate-able opcode list identical to the sibling transform done in IR (in instcombine).

This change used to show regressions for x86, but those are gone after D55494. This gets us closer to deleting the x86 custom function (combineTruncatedArithmetic) that does almost the same thing.

[andreadb]

test/CodeGen/X86/pr32345.ll

122–126

Not sure what is going on here..

We fail to realize that XOR is both commutative and associative.
We are essentially doing (%eax XOR %ecx) XOR %ecx.
That is equivalent to %eax XOR (%ecx XOR %ecx). The second XOR is a zero-idiom.
So, it becomes %eax XOR 0, which is %eax.

That entire computation could be folded away.

We should generate just this:

movzwl var_27, %ecx
movzwl var_22, %eax

As a side note: we even fail to realize that the last zero-extending move is redundant! The upper half of EAX is already zero, because it already comes from a zero-extending move. The XOR is done with ECX, which also is zero-extended! So, the upper bits cannot possibly be anything else other than zero.
So, the zero-extending move is completely redundant here, because the result is already zero.

[andreadb]

From an x86 point of view, this change looks good.

At some point, we should address the poor codegen from test/CodeGen/X86/pr32345.ll. Maybe you could raise a bug for it.

[spatel]

In D55866#1336120, @andreadb wrote:

From an x86 point of view, this change looks good.

At some point, we should address the poor codegen from test/CodeGen/X86/pr32345.ll. Maybe you could raise a bug for it.

I don't think it's worth the effort if we look at the background for this test; the intent was to check for a crash with *unoptimized* code:
https://bugs.llvm.org/show_bug.cgi?id=32345
rL298923
...so the test includes CSE values that the DAG never expects to encounter. I think the optimized RUN was just added for completeness/sanity. The whole thing is folded to 'unreachable' in IR.

[andreadb]

In D55866#1336263, @spatel wrote:

In D55866#1336120, @andreadb wrote:

From an x86 point of view, this change looks good.

At some point, we should address the poor codegen from test/CodeGen/X86/pr32345.ll. Maybe you could raise a bug for it.

I don't think it's worth the effort if we look at the background for this test; the intent was to check for a crash with *unoptimized* code:
https://bugs.llvm.org/show_bug.cgi?id=32345
rL298923
...so the test includes CSE values that the DAG never expects to encounter. I think the optimized RUN was just added for completeness/sanity. The whole thing is folded to 'unreachable' in IR.

Fair enough. I didn't check if it was folded by the optimizers at IR level.
Given that the whole sequence is optimized out in IR, then I agree with you: it is not not worthy to look into it.

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

SDValue DAGCombiner::visitTRUNCATE(SDNode *N) {
   if (SDValue NewVSel = matchVSelectOpSizesWithSetCC(N))
     return NewVSel;
 
   // Narrow a suitable binary operation with a non-opaque constant operand by
   // moving it ahead of the truncate. This is limited to pre-legalization
   // because targets may prefer a wider type during later combines and invert
   // this transform.
   switch (N0.getOpcode()) {
-  // TODO: Add case for ADD - that will likely require a change in logic here
-  // or target-specific changes to avoid regressions.
+  case ISD::ADD:
   case ISD::SUB:
   case ISD::MUL:
   case ISD::AND:
   case ISD::OR:
   case ISD::XOR:
     if (!LegalOperations && N0.hasOneUse() &&
         (isConstantOrConstantVector(N0.getOperand(0), true) ||
          isConstantOrConstantVector(N0.getOperand(1), true))) {

test/CodeGen/AMDGPU/imm16.ll

 ; VI: buffer_store_short [[REG]]
 define amdgpu_kernel void @add_inline_imm_16_f16(half addrspace(1)* %out, half %x) {
   %y = fadd half %x, 0xH0010
   store half %y, half addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}add_inline_imm_neg_1_f16:
-; VI: v_add_u32_e32 [[REG:v[0-9]+]], vcc, -1
+; VI: v_add_u16_e32 [[REG:v[0-9]+]], -1, [[REG:v[0-9]+]]
 ; VI: buffer_store_short [[REG]]
 define amdgpu_kernel void @add_inline_imm_neg_1_f16(half addrspace(1)* %out, i16 addrspace(1)* %in) {
   %x = load i16, i16 addrspace(1)* %in
   %y = add i16 %x, -1
   %ybc = bitcast i16 %y to half
   store half %ybc, half addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}add_inline_imm_neg_2_f16:
-; VI: v_add_u32_e32 [[REG:v[0-9]+]], vcc, 0xfffe
+; VI: v_add_u16_e32 [[REG:v[0-9]+]], -2, [[REG:v[0-9]+]]
 ; VI: buffer_store_short [[REG]]
 define amdgpu_kernel void @add_inline_imm_neg_2_f16(half addrspace(1)* %out, i16 addrspace(1)* %in) {
   %x = load i16, i16 addrspace(1)* %in
   %y = add i16 %x, -2
   %ybc = bitcast i16 %y to half
   store half %ybc, half addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}add_inline_imm_neg_16_f16:
-; VI: v_add_u32_e32 [[REG:v[0-9]+]], vcc, 0xfff0
+; VI: v_add_u16_e32 [[REG:v[0-9]+]], -16, [[REG:v[0-9]+]]
 ; VI: buffer_store_short [[REG]]
 define amdgpu_kernel void @add_inline_imm_neg_16_f16(half addrspace(1)* %out, i16 addrspace(1)* %in) {
   %x = load i16, i16 addrspace(1)* %in
   %y = add i16 %x, -16
   %ybc = bitcast i16 %y to half
   store half %ybc, half addrspace(1)* %out
   ret void
 }

test/CodeGen/AMDGPU/trunc-combine.ll

define i32 @trunc_bitcast_i64_lshr_32_i32(i64 %bar) {
   %trunc = trunc i64 %srl to i32
   ret i32 %trunc
 }
 
 ; GCN-LABEL: {{^}}trunc_bitcast_v2i32_to_i16:
 ; GCN: _load_dword
 ; GCN-NOT: _load_dword
 ; GCN-NOT: v_mov_b32
-; GCN: v_add_u32_e32 v0, vcc, 4, v0
+; GCN: v_add_u16_e32 v0, 4, v0
 define i16 @trunc_bitcast_v2i32_to_i16(<2 x i32> %bar) {
   %load0 = load i32, i32 addrspace(1)* undef
   %load1 = load i32, i32 addrspace(1)* null
   %insert.0 = insertelement <2 x i32> undef, i32 %load0, i32 0
   %insert.1 = insertelement <2 x i32> %insert.0, i32 99, i32 1
   %bc = bitcast <2 x i32> %insert.1 to i64
   %trunc = trunc i64 %bc to i16
   %add = add i16 %trunc, 4
   ret i16 %add
 }
 
 ; Make sure there's no crash if the source vector type is FP
 ; GCN-LABEL: {{^}}trunc_bitcast_v2f32_to_i16:
 ; GCN: _load_dword
 ; GCN-NOT: _load_dword
 ; GCN-NOT: v_mov_b32
-; GCN: v_add_u32_e32 v0, vcc, 4, v0
+; GCN: v_add_u16_e32 v0, 4, v0
 define i16 @trunc_bitcast_v2f32_to_i16(<2 x float> %bar) {
   %load0 = load float, float addrspace(1)* undef
   %load1 = load float, float addrspace(1)* null
   %insert.0 = insertelement <2 x float> undef, float %load0, i32 0
   %insert.1 = insertelement <2 x float> %insert.0, float 4.0, i32 1
   %bc = bitcast <2 x float> %insert.1 to i64
   %trunc = trunc i64 %bc to i16
   %add = add i16 %trunc, 4

test/CodeGen/Hexagon/vect/vect-vaslw.ll

 ; RUN: llc -march=hexagon < %s | FileCheck %s
-; CHECK: vaslw
+; CHECK: vaslh
 
 target datalayout = "e-p:32:32:32-i64:64:64-i32:32:32-i16:16:16-i1:32:32-f64:64:64-f32:32:32-v64:64:64-v32:32:32-a0:0-n16:32"
 target triple = "hexagon-unknown-linux-gnu"
 
 define void @foo(i16* nocapture %v) nounwind {
 entry:
   %p_arrayidx = getelementptr i16, i16* %v, i32 4
   %vector_ptr = bitcast i16* %p_arrayidx to <4 x i16>*

test/CodeGen/X86/load-combine.ll

 ; i8* arg; i32 i;
 ; p = arg + 12;
 ; (i32) p[i] | ((i32) p[i + 1] << 8) | ((i32) p[i + 2] << 16) | ((i32) p[i + 3] << 24)
 define i32 @load_i32_by_i8_base_offset_index(i8* %arg, i32 %i) {
 ; CHECK-LABEL: load_i32_by_i8_base_offset_index:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT:    movl 12(%ecx,%eax), %eax
+; CHECK-NEXT:    movl 12(%eax,%ecx), %eax
 ; CHECK-NEXT:    retl
 ;
 ; CHECK64-LABEL: load_i32_by_i8_base_offset_index:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    movl %esi, %eax
 ; CHECK64-NEXT:    movl 12(%rdi,%rax), %eax
 ; CHECK64-NEXT:    retq
   %tmp = add nuw nsw i32 %i, 3

 ; i8* arg; i32 i;
 ; p = arg + 12;
 ; (i32) p[i + 1] | ((i32) p[i + 2] << 8) | ((i32) p[i + 3] << 16) | ((i32) p[i + 4] << 24)
 define i32 @load_i32_by_i8_base_offset_index_2(i8* %arg, i32 %i) {
 ; CHECK-LABEL: load_i32_by_i8_base_offset_index_2:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT:    movl 13(%ecx,%eax), %eax
+; CHECK-NEXT:    movl 13(%eax,%ecx), %eax
 ; CHECK-NEXT:    retl
 ;
 ; CHECK64-LABEL: load_i32_by_i8_base_offset_index_2:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    movl %esi, %eax
 ; CHECK64-NEXT:    movl 13(%rdi,%rax), %eax
 ; CHECK64-NEXT:    retq
   %tmp = add nuw nsw i32 %i, 4

 ; In order to fold the pattern above we need to reassociate the address computation
 ; first. By the time the address computation is reassociated loads are combined to
 ; to zext and aext loads.
 define i32 @load_i32_by_i8_zaext_loads(i8* %arg, i32 %arg1) {
 ; CHECK-LABEL: load_i32_by_i8_zaext_loads:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT:    movl 12(%ecx,%eax), %eax
+; CHECK-NEXT:    movl 12(%eax,%ecx), %eax
 ; CHECK-NEXT:    retl
 ;
 ; CHECK64-LABEL: load_i32_by_i8_zaext_loads:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    movl %esi, %eax
 ; CHECK64-NEXT:    movl 12(%rdi,%rax), %eax
 ; CHECK64-NEXT:    retq
   %tmp = add nuw nsw i32 %arg1, 3

 ; p3 = arg + i + 3;
 ;
 ; (i32) p0[12] | ((i32) p1[12] << 8) | ((i32) p2[12] << 16) | ((i32) p3[12] << 24)
 define i32 @load_i32_by_i8_zsext_loads(i8* %arg, i32 %arg1) {
 ; CHECK-LABEL: load_i32_by_i8_zsext_loads:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
 ; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %ecx
-; CHECK-NEXT:    movl 12(%ecx,%eax), %eax
+; CHECK-NEXT:    movl 12(%eax,%ecx), %eax
 ; CHECK-NEXT:    retl
 ;
 ; CHECK64-LABEL: load_i32_by_i8_zsext_loads:
 ; CHECK64:       # %bb.0:
 ; CHECK64-NEXT:    movl %esi, %eax
 ; CHECK64-NEXT:    movl 12(%rdi,%rax), %eax
 ; CHECK64-NEXT:    retq
   %tmp = add nuw nsw i32 %arg1, 3

test/CodeGen/X86/pr32329.ll

 ; X86-NEXT:    imull %eax, %ecx
 ; X86-NEXT:    addl var_24, %ecx
 ; X86-NEXT:    andl $4194303, %edx # imm = 0x3FFFFF
 ; X86-NEXT:    leal (%edx,%edx), %ebx
 ; X86-NEXT:    subl %eax, %ebx
 ; X86-NEXT:    movl %ebx, %edi
 ; X86-NEXT:    subl %esi, %edi
 ; X86-NEXT:    imull %edi, %ecx
-; X86-NEXT:    addl $-1437483407, %ecx # imm = 0xAA51BE71
+; X86-NEXT:    addb $113, %cl
 ; X86-NEXT:    movl $9, %esi
 ; X86-NEXT:    xorl %ebp, %ebp
 ; X86-NEXT:    shldl %cl, %esi, %ebp
 ; X86-NEXT:    shll %cl, %esi
 ; X86-NEXT:    testb $32, %cl
 ; X86-NEXT:    cmovnel %esi, %ebp
 ; X86-NEXT:    movl $0, %ecx
 ; X86-NEXT:    cmovnel %ecx, %esi

 ; X64-NEXT:    imull %r9d, %ecx
 ; X64-NEXT:    addl {{.*}}(%rip), %ecx
 ; X64-NEXT:    andl $4194303, %eax # imm = 0x3FFFFF
 ; X64-NEXT:    leal (%rax,%rax), %edi
 ; X64-NEXT:    subl %r9d, %edi
 ; X64-NEXT:    movl %edi, %esi
 ; X64-NEXT:    subl %r8d, %esi
 ; X64-NEXT:    imull %esi, %ecx
-; X64-NEXT:    addl $-1437483407, %ecx # imm = 0xAA51BE71
+; X64-NEXT:    addb $113, %cl
 ; X64-NEXT:    movl $9, %edx
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-NEXT:    shlq %cl, %rdx
 ; X64-NEXT:    movq %rdx, {{.*}}(%rip)
 ; X64-NEXT:    cmpl %eax, %esi
 ; X64-NEXT:    setge {{.*}}(%rip)
 ; X64-NEXT:    imull %r9d, %edi
 ; X64-NEXT:    movb %dil, {{.*}}(%rip)

test/CodeGen/X86/pr32345.ll

 ; 6860-NEXT:    movzwl var_27, %ecx
 ; 6860-NEXT:    movw %cx, %dx
 ; 6860-NEXT:    xorw %dx, %ax
 ; 6860-NEXT:    # implicit-def: $esi
 ; 6860-NEXT:    movw %ax, %si
 ; 6860-NEXT:    xorl %ecx, %esi
 ; 6860-NEXT:    movw %si, %ax
 ; 6860-NEXT:    movzwl %ax, %esi
-; 6860-NEXT:    addl $-16610, %ecx # imm = 0xBF1E
 ; 6860-NEXT:    movb %cl, %bl
+; 6860-NEXT:    addb $30, %bl
 ; 6860-NEXT:    xorl %ecx, %ecx
 ; 6860-NEXT:    movl %ecx, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; 6860-NEXT:    movb %bl, %cl
 ; 6860-NEXT:    movl {{[-0-9]+}}(%e{{[sb]}}p), %edi # 4-byte Reload
 ; 6860-NEXT:    shrdl %cl, %edi, %esi
 ; 6860-NEXT:    testb $32, %bl
 ; 6860-NEXT:    movl %esi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill
 ; 6860-NEXT:    movl %edi, {{[-0-9]+}}(%e{{[sb]}}p) # 4-byte Spill

 ; 6860-NEXT:    popl %edi
 ; 6860-NEXT:    popl %ebx
 ; 6860-NEXT:    popl %ebp
 ; 6860-NEXT:    .cfi_def_cfa %esp, 4
 ; 6860-NEXT:    retl
 ;
 ; X64-LABEL: foo:
 ; X64:       # %bb.0: # %bb
-; X64-NEXT:    movzwl {{.*}}(%rip), %eax
 ; X64-NEXT:    movzwl {{.*}}(%rip), %ecx
-; X64-NEXT:    movl %ecx, %edx
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    xorl %eax, %edx
-; X64-NEXT:    movzwl %dx, %eax
+; X64-NEXT:    movzwl {{.*}}(%rip), %eax
+; X64-NEXT:    xorw %cx, %ax
+; X64-NEXT:    xorl %ecx, %eax
+; X64-NEXT:    movzwl %ax, %eax
 ; X64-NEXT:    movq %rax, -{{[0-9]+}}(%rsp)
-; X64-NEXT:    addl $-16610, %ecx # imm = 0xBF1E
+; X64-NEXT:    addb $30, %cl
 ; X64-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; X64-NEXT:    shrq %cl, %rax
 ; X64-NEXT:    movb %al, (%rax)
 ; X64-NEXT:    retq
 ;
 ; 686-LABEL: foo:
 ; 686:       # %bb.0: # %bb
 ; 686-NEXT:    pushl %ebp
 ; 686-NEXT:    .cfi_def_cfa_offset 8
 ; 686-NEXT:    .cfi_offset %ebp, -8
 ; 686-NEXT:    movl %esp, %ebp
 ; 686-NEXT:    .cfi_def_cfa_register %ebp
 ; 686-NEXT:    andl $-8, %esp
 ; 686-NEXT:    subl $8, %esp
-; 686-NEXT:    movzwl var_22, %eax
 ; 686-NEXT:    movzwl var_27, %ecx
-; 686-NEXT:    movl %ecx, %edx
-; 686-NEXT:    xorl %ecx, %edx
-; 686-NEXT:    xorl %eax, %edx
-; 686-NEXT:    movzwl %dx, %eax
+; 686-NEXT:    movzwl var_22, %eax
+; 686-NEXT:    xorw %cx, %ax
+; 686-NEXT:    xorl %ecx, %eax
+; 686-NEXT:    movzwl %ax, %eax
 ; 686-NEXT:    movl %eax, (%esp)
 ; 686-NEXT:    movl $0, {{[0-9]+}}(%esp)
-; 686-NEXT:    addl $-16610, %ecx # imm = 0xBF1E
+; 686-NEXT:    addb $30, %cl
 ; 686-NEXT:    xorl %edx, %edx
 ; 686-NEXT:    shrdl %cl, %edx, %eax
 ; 686-NEXT:    testb $32, %cl
 ; 686-NEXT:    jne .LBB0_2
 ; 686-NEXT:  # %bb.1: # %bb
 ; 686-NEXT:    movl %eax, %edx
 ; 686-NEXT:  .LBB0_2: # %bb
 ; 686-NEXT:    movb %dl, (%eax)

test/CodeGen/X86/pr33290.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s --check-prefix=X86
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s --check-prefix=X64
 
 @a = common global i32 0, align 4
 @c = common local_unnamed_addr global i8 0, align 1
 @b = common local_unnamed_addr global i32* null, align 8
 
 define void @e() {
 ; X86-LABEL: e:
 ; X86:       # %bb.0: # %entry
 ; X86-NEXT:    movl b, %eax
 ; X86-NEXT:    .p2align 4, 0x90
 ; X86-NEXT:  .LBB0_1: # %for.cond
 ; X86-NEXT:    # =>This Inner Loop Header: Depth=1
 ; X86-NEXT:    movzbl c, %ecx
-; X86-NEXT:    leal a+2(%ecx), %ecx
 ; X86-NEXT:    movb $0, c
+; X86-NEXT:    leal a+2(%ecx), %ecx
 ; X86-NEXT:    movl %ecx, (%eax)
 ; X86-NEXT:    jmp .LBB0_1
 ;
 ; X64-LABEL: e:
 ; X64:       # %bb.0: # %entry
 ; X64-NEXT:    movq {{.*}}(%rip), %rax
 ; X64-NEXT:    movl $a, %esi
 ; X64-NEXT:    .p2align 4, 0x90

test/CodeGen/X86/pr34381.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ;RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=slow-incdec | FileCheck %s
 
 @var_21 = external constant i32, align 4
 @var_29 = external constant i8, align 1
 @var_390 = external global i32, align 4
 @var_11 = external constant i8, align 1
 @var_370 = external global i8, align 1
 
 ; Function Attrs: noinline nounwind optnone uwtable
 define void @_Z3foov() {
 ; CHECK-LABEL: _Z3foov:
 ; CHECK:       # %bb.0: # %entry
 ; CHECK-NEXT:    movsbl {{.*}}(%rip), %eax
 ; CHECK-NEXT:    negl %eax
-; CHECK-NEXT:    cmpl %eax, {{.*}}(%rip)
-; CHECK-NEXT:    setb %al
 ; CHECK-NEXT:    xorl %ecx, %ecx
-; CHECK-NEXT:    addb $-1, %al
-; CHECK-NEXT:    sete %cl
+; CHECK-NEXT:    cmpl %eax, {{.*}}(%rip)
+; CHECK-NEXT:    setb %cl
 ; CHECK-NEXT:    movl %ecx, {{.*}}(%rip)
 ; CHECK-NEXT:    movb {{.*}}(%rip), %al
 ; CHECK-NEXT:    movb %al, {{.*}}(%rip)
 ; CHECK-NEXT:    retq
 entry:
   %0 = load i32, i32* @var_21, align 4
   %1 = load i8, i8* @var_29, align 1
   %conv = sext i8 %1 to i32

test/CodeGen/X86/pr35765.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=x86_64-unknown-linux-gnu %s -o - | FileCheck %s
 
 @ll = local_unnamed_addr global i64 0, align 8
 @x = local_unnamed_addr global i64 2651237805702985558, align 8
 @s1 = local_unnamed_addr global { i8, i8 } { i8 123, i8 5 }, align 2
 @s2 = local_unnamed_addr global { i8, i8 } { i8 -122, i8 3 }, align 2
 
 define void @PR35765() {
 ; CHECK-LABEL: PR35765:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movzwl {{.*}}(%rip), %ecx
-; CHECK-NEXT:    addl $-1398, %ecx # imm = 0xFA8A
+; CHECK-NEXT:    movb {{.*}}(%rip), %cl
+; CHECK-NEXT:    addb $-118, %cl
 ; CHECK-NEXT:    movl $4, %eax
-; CHECK-NEXT:    # kill: def $cl killed $cl killed $ecx
 ; CHECK-NEXT:    shll %cl, %eax
 ; CHECK-NEXT:    movzwl {{.*}}(%rip), %ecx
 ; CHECK-NEXT:    movzwl {{.*}}(%rip), %edx
 ; CHECK-NEXT:    notl %edx
 ; CHECK-NEXT:    orl $63488, %edx # imm = 0xF800
 ; CHECK-NEXT:    movzwl %dx, %edx
 ; CHECK-NEXT:    orl %ecx, %edx
 ; CHECK-NEXT:    xorl %eax, %edx

test/CodeGen/X86/scheduler-backtracking.ll

 
 define i256 @test1(i256 %a) nounwind {
 ; ILP-LABEL: test1:
 ; ILP:       # %bb.0:
 ; ILP-NEXT:    pushq %r14
 ; ILP-NEXT:    pushq %rbx
 ; ILP-NEXT:    movq %rdi, %rax
 ; ILP-NEXT:    xorl %r8d, %r8d
-; ILP-NEXT:    incl %esi
 ; ILP-NEXT:    addb %sil, %sil
+; ILP-NEXT:    addb $2, %sil
 ; ILP-NEXT:    orb $1, %sil
 ; ILP-NEXT:    movl $1, %r10d
 ; ILP-NEXT:    xorl %r14d, %r14d
 ; ILP-NEXT:    movl %esi, %ecx
 ; ILP-NEXT:    shldq %cl, %r10, %r14
 ; ILP-NEXT:    movl $1, %edx
 ; ILP-NEXT:    shlq %cl, %rdx
-; ILP-NEXT:    leal -128(%rsi), %r9d
 ; ILP-NEXT:    movb $-128, %r11b
-; ILP-NEXT:    xorl %ebx, %ebx
+; ILP-NEXT:    subb %sil, %r11b
+; ILP-NEXT:    leal -128(%rsi), %r9d
+; ILP-NEXT:    xorl %edi, %edi
 ; ILP-NEXT:    movl %r9d, %ecx
-; ILP-NEXT:    shldq %cl, %r10, %rbx
+; ILP-NEXT:    shldq %cl, %r10, %rdi
+; ILP-NEXT:    movl $1, %ebx
+; ILP-NEXT:    shlq %cl, %rbx
+; ILP-NEXT:    movl %r11d, %ecx
+; ILP-NEXT:    shrdq %cl, %r8, %r10
 ; ILP-NEXT:    testb $64, %sil
 ; ILP-NEXT:    cmovneq %rdx, %r14
 ; ILP-NEXT:    cmovneq %r8, %rdx
-; ILP-NEXT:    movl $1, %edi
-; ILP-NEXT:    shlq %cl, %rdi
-; ILP-NEXT:    subb %sil, %r11b
-; ILP-NEXT:    movl %r11d, %ecx
-; ILP-NEXT:    shrdq %cl, %r8, %r10
 ; ILP-NEXT:    testb $64, %r11b
 ; ILP-NEXT:    cmovneq %r8, %r10
 ; ILP-NEXT:    testb $64, %r9b
-; ILP-NEXT:    cmovneq %rdi, %rbx
-; ILP-NEXT:    cmovneq %r8, %rdi
+; ILP-NEXT:    cmovneq %rbx, %rdi
+; ILP-NEXT:    cmovneq %r8, %rbx
 ; ILP-NEXT:    testb %sil, %sil
 ; ILP-NEXT:    cmovsq %r8, %r14
 ; ILP-NEXT:    cmovsq %r8, %rdx
 ; ILP-NEXT:    movq %r14, 8(%rax)
 ; ILP-NEXT:    movq %rdx, (%rax)
-; ILP-NEXT:    cmovnsq %r8, %rbx
-; ILP-NEXT:    cmoveq %r8, %rbx
-; ILP-NEXT:    movq %rbx, 24(%rax)
-; ILP-NEXT:    cmovnsq %r10, %rdi
+; ILP-NEXT:    cmovnsq %r8, %rdi
 ; ILP-NEXT:    cmoveq %r8, %rdi
-; ILP-NEXT:    movq %rdi, 16(%rax)
+; ILP-NEXT:    movq %rdi, 24(%rax)
+; ILP-NEXT:    cmovnsq %r10, %rbx
+; ILP-NEXT:    cmoveq %r8, %rbx
+; ILP-NEXT:    movq %rbx, 16(%rax)
 ; ILP-NEXT:    popq %rbx
 ; ILP-NEXT:    popq %r14
 ; ILP-NEXT:    retq
 ;
 ; HYBRID-LABEL: test1:
 ; HYBRID:       # %bb.0:
 ; HYBRID-NEXT:    movq %rdi, %rax
-; HYBRID-NEXT:    incl %esi
 ; HYBRID-NEXT:    addb %sil, %sil
+; HYBRID-NEXT:    addb $2, %sil
 ; HYBRID-NEXT:    orb $1, %sil
 ; HYBRID-NEXT:    movb $-128, %cl
 ; HYBRID-NEXT:    subb %sil, %cl
 ; HYBRID-NEXT:    xorl %r8d, %r8d
 ; HYBRID-NEXT:    movl $1, %r11d
 ; HYBRID-NEXT:    movl $1, %r9d
 ; HYBRID-NEXT:    shrdq %cl, %r8, %r9
 ; HYBRID-NEXT:    testb $64, %cl

 ; HYBRID-NEXT:    cmovnsq %r9, %rdx
 ; HYBRID-NEXT:    cmoveq %r8, %rdx
 ; HYBRID-NEXT:    movq %rdx, 16(%rax)
 ; HYBRID-NEXT:    retq
 ;
 ; BURR-LABEL: test1:
 ; BURR:       # %bb.0:
 ; BURR-NEXT:    movq %rdi, %rax
-; BURR-NEXT:    incl %esi
 ; BURR-NEXT:    addb %sil, %sil
+; BURR-NEXT:    addb $2, %sil
 ; BURR-NEXT:    orb $1, %sil
 ; BURR-NEXT:    movb $-128, %cl
 ; BURR-NEXT:    subb %sil, %cl
 ; BURR-NEXT:    xorl %r8d, %r8d
 ; BURR-NEXT:    movl $1, %r11d
 ; BURR-NEXT:    movl $1, %r9d
 ; BURR-NEXT:    shrdq %cl, %r8, %r9
 ; BURR-NEXT:    testb $64, %cl

 ; BURR-NEXT:    cmoveq %r8, %rdx
 ; BURR-NEXT:    movq %rdx, 16(%rax)
 ; BURR-NEXT:    retq
 ;
 ; SRC-LABEL: test1:
 ; SRC:       # %bb.0:
 ; SRC-NEXT:    pushq %rbx
 ; SRC-NEXT:    movq %rdi, %rax
-; SRC-NEXT:    incl %esi
 ; SRC-NEXT:    addb %sil, %sil
+; SRC-NEXT:    addb $2, %sil
 ; SRC-NEXT:    orb $1, %sil
 ; SRC-NEXT:    movb $-128, %cl
 ; SRC-NEXT:    subb %sil, %cl
 ; SRC-NEXT:    xorl %r8d, %r8d
 ; SRC-NEXT:    movl $1, %edi
 ; SRC-NEXT:    movl $1, %r10d
 ; SRC-NEXT:    shrdq %cl, %r8, %r10
 ; SRC-NEXT:    testb $64, %cl

 ; SRC-NEXT:    popq %rbx
 ; SRC-NEXT:    retq
 ;
 ; LIN-LABEL: test1:
 ; LIN:       # %bb.0:
 ; LIN-NEXT:    movq %rdi, %rax
 ; LIN-NEXT:    xorl %r9d, %r9d
 ; LIN-NEXT:    movl $1, %r8d
-; LIN-NEXT:    incl %esi
 ; LIN-NEXT:    addb %sil, %sil
+; LIN-NEXT:    addb $2, %sil
 ; LIN-NEXT:    orb $1, %sil
 ; LIN-NEXT:    movl $1, %edx
 ; LIN-NEXT:    movl %esi, %ecx
 ; LIN-NEXT:    shlq %cl, %rdx
 ; LIN-NEXT:    testb $64, %sil
 ; LIN-NEXT:    movq %rdx, %rcx
 ; LIN-NEXT:    cmovneq %r9, %rcx
 ; LIN-NEXT:    testb %sil, %sil