[DAGCombiner] fold assertzexts separated by trunc

[spatel]

Summary

If we have an AssertZext of a truncated value that has already been AssertZext'ed, we can assert on the wider source op to improve the zext-y knowledge.

This is an implementation of the change suggested in D36890 (and I think would obsolete that patch). Actually, I don't think the suggestion #1 that I made there is valid, but this achieves the same result in a safe way by matching the a larger pattern.

The fact that this only affects x86 makes me suspicious that x86 is doing something suboptimal to create these AssertZexts in the first place, but I'm not sure.

[spatel]

Here's the effect on "select_C1_C2_zeroext" nodes in case that makes the transform clearer:

      t4: i32 = AssertZext t2, ValueType:ch:i8
    t5: i8 = truncate t4
  t7: i8 = AssertZext t5, ValueType:ch:i1
t8: i1 = truncate t7

-->

  t16: i32 = AssertZext t2, ValueType:ch:i1
t18: i1 = truncate t16

[aaboud]

I have one comment below.
By the way, I noticed that the double AssertZero occur only for the x86_64 (in i386 it does not happen).
It might be worth checking where it comes from, regardless of this patch.

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

7993

What about the other case:
assert (trunc (assert X, i1) to iN), i8 --> trunc (assert X, i1) to iN

Do not we want to cover this as well?
Or we are sure that such case will never occur?

[aivchenk]

That's certainly a much better approach than the original. Couple of comments/questions:

• We eliminate the "outer" assert by making "inner" stronger. In a sense, this essentially moves the information about the extension further away from the use. Is there some sort of canonicalization or a convention for that? The other way of removing redundancy would be just to eliminate the weaker "inner" assert, though it would not solve the initial problem for handling AssertZext in the backend. But may be the problem is that we don't handle AssertZext properly?
• Should we implement the same idea for AssertSext?

[spatel]

In D37017#849197, @aaboud wrote:

I have one comment below.
By the way, I noticed that the double AssertZero occur only for the x86_64 (in i386 it does not happen).
It might be worth checking where it comes from, regardless of this patch.

i386 is (mostly?) immune from this because it passes function arguments on the stack. The first assertzext (the one with i32/i8 types in most of these examples) is generated by X86TargetLowering::LowerFormalArguments(). We always get the zext assert source value type as i8 for an i1 arg because that's what is specified in "RegisterTypeForVT", but I think that's the root cause of the bug. Should we be using the IR type (i1) at this stage when creating assertzext nodes and the subsequent truncate node?

Adding some more potential reviewers. This is my first look at function arg lowering. Is the intermediate trunc to i8 of the formal arg necessary for other reasons? The naive change to X86TargetLowering::LowerFormalArguments() doesn't work for weird vector types or AVX512 masks:

@@ -3015,7 +3015,7 @@
           // Promoting a mask type (v*i1) into a register of type i64/i32/i16/i8
           ArgValue = lowerRegToMasks(ArgValue, VA.getValVT(), RegVT, dl, DAG);
         } else
-          ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
+          ArgValue = DAG.getNode(ISD::TRUNCATE, dl, Ins[InsIndex].ArgVT, ArgValue);
       }
     } else {
       assert(VA.isMemLoc());

We're trying to avoid adding a DAG combine for this pattern by not creating the pattern in the first place:

      t4: i32 = AssertZext t2, ValueType:ch:i8
    t5: i8 = truncate t4
  t7: i8 = AssertZext t5, ValueType:ch:i1
t8: i1 = truncate t7

-->

  t16: i32 = AssertZext t2, ValueType:ch:i1
t18: i1 = truncate t16

[spatel]

In D37017#849272, @aivchenk wrote:

• Should we implement the same idea for AssertSext?

Yes, we should. But after discovering that this code almost already exists in the Mips backend, I'd like to make that a small follow-up just to reduce the patch size and risk.

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

7993

I haven't seen that occur yet, but since it's easy to handle, I'll make the fold more general.

[spatel]

Patch updated:
After finding that the same functionality exists in trunk, I think it makes sense to go ahead with this patch. Effectively, we're just taking a general fold that was hiding in the Mips backend and applying it to all targets.

Also, given that there is already another fold for assertzext(assertzext(x)), I assume there's already some scenario under which it makes sense to have folds for these nodes besides the one we're looking at. Therefore, there's not much benefit to trying to stub this pattern out at the source. We just account for the potential inefficiency in argument lowering with this fold.

[spatel]

Ping.

[efriedma]

test/CodeGen/X86/negate-i1.ll

135

This seems kind of scary... is an "i1 zeroext" actually guaranteed to be zero-extended to 64 bits?

[craig.topper]

test/CodeGen/X86/negate-i1.ll

135

Nope. It hit PR28540 which should be fixed by D37729

[spatel]

test/CodeGen/X86/negate-i1.ll

135

Should I make this patch dependent on that fix? Are we correct when we go from:

        t4: i32 = AssertZext t2, ValueType:ch:i8
      t5: i8 = truncate t4
    t7: i8 = AssertZext t5, ValueType:ch:i1
  t8: i1 = truncate t7
t9: i64 = sign_extend t8

to:

    t16: i32 = AssertZext t2, ValueType:ch:i1
  t18: i64 = any_extend t16
t15: i64 = sign_extend_inreg t18, ValueType:ch:i1

[craig.topper]

PR28540 is fixed now.

[spatel]

In D37017#874229, @craig.topper wrote:

PR28540 is fixed now.

Thanks - so now for the questionable test, we'll get movl+negq rather than negq+movq (no changes to the output from this patch).

Any other known problems?

[spatel]

Patch updated:
No code changes, but test output updated after the fix from rL313563.

[spatel]

In D37017#874263, @spatel wrote:

In D37017#874229, @craig.topper wrote:

PR28540 is fixed now.

Thanks - so now for the questionable test, we'll get movl+negq rather than negq+movq (no changes to the output from this patch).

That comment wasn't clear. I meant we still get the same DAG nodes for the questionable test from this combine:

    t16: i32 = AssertZext t2, ValueType:ch:i1
  t18: i64 = any_extend t16
t15: i64 = sign_extend_inreg t18, ValueType:ch:i1

[craig.topper]

LGTM

lib/CodeGen/SelectionDAG/DAGCombiner.cpp

if (SDValue SCC = SimplifySelectCC(
             DAG.getConstant(0, DL, VT),
             cast<CondCodeSDNode>(N0.getOperand(2))->get(), true))
       return SCC;
   }
 
   return SDValue();
 }
 
+// TODO: These transforms should work with AssertSext too.
+// Change the function name, comments, opcode references, and caller.
 SDValue DAGCombiner::visitAssertZext(SDNode *N) {
   SDValue N0 = N->getOperand(0);
   SDValue N1 = N->getOperand(1);
-  EVT EVT = cast<VTSDNode>(N1)->getVT();
+  EVT AssertVT = cast<VTSDNode>(N1)->getVT();
 
   // fold (assertzext (assertzext x, vt), vt) -> (assertzext x, vt)
   if (N0.getOpcode() == ISD::AssertZext &&
-      EVT == cast<VTSDNode>(N0.getOperand(1))->getVT())
+      AssertVT == cast<VTSDNode>(N0.getOperand(1))->getVT())
     return N0;
 
+  if (N0.getOpcode() == ISD::TRUNCATE && N0.hasOneUse() &&
+      N0.getOperand(0).getOpcode() == ISD::AssertZext) {
+    // We have an assert, truncate, assert sandwich. Make one stronger assert
+    // by asserting on the smallest asserted type to the larger source type.
+    // This eliminates the later assert:
+    // assert (trunc (assert X, i8) to iN), i1 --> trunc (assert X, i1) to iN
+    // assert (trunc (assert X, i1) to iN), i8 --> trunc (assert X, i1) to iN
+    SDValue BigA = N0.getOperand(0);
+    EVT BigA_AssertVT = cast<VTSDNode>(BigA.getOperand(1))->getVT();
+    assert(BigA_AssertVT.bitsLE(N0.getValueType()) &&
+           "Asserting zero/sign-extended bits from a type larger than the "
+           "truncated destination does not provide information");
+
+    SDLoc DL(N);
+    EVT MinAssertVT = AssertVT.bitsLT(BigA_AssertVT) ? AssertVT : BigA_AssertVT;
+    SDValue MinAssertVTVal = DAG.getValueType(MinAssertVT);
+    SDValue NewAssert = DAG.getNode(ISD::AssertZext, DL, BigA.getValueType(),
+                                    BigA.getOperand(0), MinAssertVTVal);
+    return DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), NewAssert);
+  }
+
   return SDValue();
 }
 
 /// If the result of a wider load is shifted to right of N  bits and then
 /// truncated to a narrower type and where N is a multiple of number of bits of
 /// the narrower type, transform it to a narrower load from address + N / num of
 /// bits of new type. If the result is to be extended, also fold the extension
 /// to form a extending load.

lib/Target/Mips/MipsISelLowering.cpp

static SDValue performADDCombine(SDNode *N, SelectionDAG &DAG,
   EVT ValTy = N->getValueType(0);
   SDLoc DL(N);
 
   SDValue Add1 = DAG.getNode(ISD::ADD, DL, ValTy, N->getOperand(0),
                              Add.getOperand(0));
   return DAG.getNode(ISD::ADD, DL, ValTy, Add1, Lo);
 }
 
-static SDValue performAssertZextCombine(SDNode *N, SelectionDAG &DAG,
-                                        TargetLowering::DAGCombinerInfo &DCI,
-                                        const MipsSubtarget &Subtarget) {
-  SDValue N0 = N->getOperand(0);
-  EVT NarrowerVT = cast<VTSDNode>(N->getOperand(1))->getVT();
-
-  if (N0.getOpcode() != ISD::TRUNCATE)
-    return SDValue();
-
-  if (N0.getOperand(0).getOpcode() != ISD::AssertZext)
-    return SDValue();
-
-  // fold (AssertZext (trunc (AssertZext x))) -> (trunc (AssertZext x))
-  // if the type of the extension of the innermost AssertZext node is
-  // smaller from that of the outermost node, eg:
-  // (AssertZext:i32 (trunc:i32 (AssertZext:i64 X, i32)), i8)
-  //   -> (trunc:i32 (AssertZext X, i8))
-  SDValue WiderAssertZext = N0.getOperand(0);
-  EVT WiderVT = cast<VTSDNode>(WiderAssertZext->getOperand(1))->getVT();
-
-  if (NarrowerVT.bitsLT(WiderVT)) {
-    SDValue NewAssertZext = DAG.getNode(
-        ISD::AssertZext, SDLoc(N), WiderAssertZext.getValueType(),
-        WiderAssertZext.getOperand(0), DAG.getValueType(NarrowerVT));
-    return DAG.getNode(ISD::TRUNCATE, SDLoc(N), N->getValueType(0),
-                       NewAssertZext);
-  }
-
-  return SDValue();
-}
-
 static SDValue performSHLCombine(SDNode *N, SelectionDAG &DAG,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const MipsSubtarget &Subtarget) {
   // Pattern match CINS.
   //  $dst = shl (and $src , imm), pos
   //  => cins $dst, $src, pos, size
 
   if (DCI.isBeforeLegalizeOps() || !Subtarget.hasCnMips())
SDValue  MipsTargetLowering::PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI)
   case MipsISD::CMovFP_T:
     return performCMovFPCombine(N, DAG, DCI, Subtarget);
   case ISD::AND:
     return performANDCombine(N, DAG, DCI, Subtarget);
   case ISD::OR:
     return performORCombine(N, DAG, DCI, Subtarget);
   case ISD::ADD:
     return performADDCombine(N, DAG, DCI, Subtarget);
-  case ISD::AssertZext:
-    return performAssertZextCombine(N, DAG, DCI, Subtarget);
   case ISD::SHL:
     return performSHLCombine(N, DAG, DCI, Subtarget);
   case ISD::SUB:
     return performSUBCombine(N, DAG, DCI, Subtarget);
   }
 
   return SDValue();
 }

test/CodeGen/X86/bool-zext.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=i686-unknown-unknown | FileCheck %s -check-prefix=X32
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s -check-prefix=X64
 
-; Check that the argument gets zero-extended before calling.
+; It's not necessary to zero-extend the arg because it is specified 'zeroext'. 
 define void @bar1(i1 zeroext %v1) nounwind ssp {
 ; X32-LABEL: bar1:
 ; X32:       # BB#0:
 ; X32-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    pushl %eax
 ; X32-NEXT:    calll foo1
 ; X32-NEXT:    addl $4, %esp
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: bar1:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %dil, %edi
 ; X64-NEXT:    xorl %eax, %eax
 ; X64-NEXT:    jmp foo1 # TAILCALL
   %conv = zext i1 %v1 to i32
   %call = tail call i32 (...) @foo1(i32 %conv) nounwind
   ret void
 }
 
 ; Check that on x86-64 the arguments are simply forwarded.

test/CodeGen/X86/critical-edge-split-2.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown | FileCheck %s
 
 %0 = type <{ %1, %1 }>
 %1 = type { i8, i8, i8, i8 }
 
 @g_2 = global %0 zeroinitializer
 @g_4 = global %1 zeroinitializer, align 4
 
 ; PR8642
 define i16 @test1(i1 zeroext %C, i8** nocapture %argv) nounwind ssp {
 ; CHECK-LABEL: test1:
 ; CHECK:       # BB#0: # %entry
 ; CHECK-NEXT:    movw $1, %ax
-; CHECK-NEXT:    testb %dil, %dil
+; CHECK-NEXT:    testl %edi, %edi
 ; CHECK-NEXT:    jne .LBB0_2
 ; CHECK-NEXT:  # BB#1: # %cond.false.i
 ; CHECK-NEXT:    movl $g_4, %eax
 ; CHECK-NEXT:    movl $g_2+4, %ecx
 ; CHECK-NEXT:    xorl %esi, %esi
 ; CHECK-NEXT:    cmpq %rax, %rcx
 ; CHECK-NEXT:    sete %sil
 ; CHECK-NEXT:    movl $1, %eax

test/CodeGen/X86/fp128-select.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -O2 -mtriple=x86_64-linux-android -mattr=+mmx \
 ; RUN:     -enable-legalize-types-checking | FileCheck %s --check-prefix=MMX
 ; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu -mattr=+mmx \
 ; RUN:     -enable-legalize-types-checking | FileCheck %s --check-prefix=MMX
 ; RUN: llc < %s -O2 -mtriple=x86_64-linux-android \
 ; RUN:     -enable-legalize-types-checking | FileCheck %s
 ; RUN: llc < %s -O2 -mtriple=x86_64-linux-gnu \
 ; RUN:     -enable-legalize-types-checking | FileCheck %s
 
 define void @test_select(fp128* %p, fp128* %q, i1 zeroext %c) {
 ; MMX-LABEL: test_select:
 ; MMX:       # BB#0:
-; MMX-NEXT:    testb %dl, %dl
+; MMX-NEXT:    testl %edx, %edx
 ; MMX-NEXT:    jne .LBB0_1
 ; MMX-NEXT:  # BB#2:
 ; MMX-NEXT:    movaps {{.*}}(%rip), %xmm0
 ; MMX-NEXT:    movaps %xmm0, (%rsi)
 ; MMX-NEXT:    retq
 ; MMX-NEXT:  .LBB0_1:
 ; MMX-NEXT:    movaps (%rdi), %xmm0
 ; MMX-NEXT:    movaps %xmm0, (%rsi)
 ; MMX-NEXT:    retq
 ;
 ; CHECK-LABEL: test_select:
 ; CHECK:       # BB#0:
 ; CHECK-NEXT:    xorl %eax, %eax
-; CHECK-NEXT:    testb %dl, %dl
+; CHECK-NEXT:    testl %edx, %edx
 ; CHECK-NEXT:    cmovneq (%rdi), %rax
 ; CHECK-NEXT:    movabsq $9223231299366420480, %rcx # imm = 0x7FFF800000000000
 ; CHECK-NEXT:    cmovneq 8(%rdi), %rcx
 ; CHECK-NEXT:    movq %rcx, 8(%rsi)
 ; CHECK-NEXT:    movq %rax, (%rsi)
 ; CHECK-NEXT:    retq
   %a = load fp128, fp128* %p, align 2
   %r = select i1 %c, fp128 %a, fp128 0xL00000000000000007FFF800000000000
   store fp128 %r, fp128* %q
   ret void
 }

test/CodeGen/X86/illegal-bitfield-loadstore.ll

 ; X86-NEXT:    andl $16769023, %eax # imm = 0xFFDFFF
 ; X86-NEXT:    orl %edx, %eax
 ; X86-NEXT:    movw %ax, (%ecx)
 ; X86-NEXT:    popl %esi
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: i24_insert_bit:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %sil, %eax
-; X64-NEXT:    movzwl (%rdi), %ecx
-; X64-NEXT:    movzbl 2(%rdi), %edx
-; X64-NEXT:    movb %dl, 2(%rdi)
-; X64-NEXT:    shll $16, %edx
-; X64-NEXT:    orl %ecx, %edx
-; X64-NEXT:    shll $13, %eax
-; X64-NEXT:    andl $16769023, %edx # imm = 0xFFDFFF
-; X64-NEXT:    orl %eax, %edx
-; X64-NEXT:    movw %dx, (%rdi)
+; X64-NEXT:    movzwl (%rdi), %eax
+; X64-NEXT:    movzbl 2(%rdi), %ecx
+; X64-NEXT:    movb %cl, 2(%rdi)
+; X64-NEXT:    shll $16, %ecx
+; X64-NEXT:    orl %eax, %ecx
+; X64-NEXT:    shll $13, %esi
+; X64-NEXT:    andl $16769023, %ecx # imm = 0xFFDFFF
+; X64-NEXT:    orl %esi, %ecx
+; X64-NEXT:    movw %cx, (%rdi)
 ; X64-NEXT:    retq
   %extbit = zext i1 %bit to i24
   %b = load i24, i24* %a, align 1
   %extbit.shl = shl nuw nsw i24 %extbit, 13
   %c = and i24 %b, -8193
   %d = or i24 %c, %extbit.shl
   store i24 %d, i24* %a, align 1
   ret void

 ; X86-NEXT:    movl $-8193, %edx # imm = 0xDFFF
 ; X86-NEXT:    andl (%eax), %edx
 ; X86-NEXT:    orl %ecx, %edx
 ; X86-NEXT:    movl %edx, (%eax)
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: i56_insert_bit:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %sil, %eax
+; X64-NEXT:    movl %esi, %eax
 ; X64-NEXT:    movzwl 4(%rdi), %ecx
 ; X64-NEXT:    movzbl 6(%rdi), %edx
 ; X64-NEXT:    movb %dl, 6(%rdi)
 ; X64-NEXT:    # kill: %EDX<def> %EDX<kill> %RDX<kill> %RDX<def>
 ; X64-NEXT:    shll $16, %edx
 ; X64-NEXT:    orl %ecx, %edx
 ; X64-NEXT:    shlq $32, %rdx
 ; X64-NEXT:    movl (%rdi), %ecx

test/CodeGen/X86/mask-negated-bool.ll

; CHECK-NEXT:    retq
   %neg = sub i32 0, %ext
   %and = and i32 %neg, 1
   ret i32 %and
 }
 
 define i32 @mask_negated_zext_bool2(i1 zeroext %x) {
 ; CHECK-LABEL: mask_negated_zext_bool2:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    retq
   %ext = zext i1 %x to i32
   %neg = sub i32 0, %ext
   %and = and i32 %neg, 1
   ret i32 %and
 }
 
 define <4 x i32> @mask_negated_zext_bool_vec(<4 x i1> %x) {
; CHECK-NEXT:    retq
   %neg = sub i32 0, %ext
   %and = and i32 %neg, 1
   ret i32 %and
 }
 
 define i32 @mask_negated_sext_bool2(i1 zeroext %x) {
 ; CHECK-LABEL: mask_negated_sext_bool2:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    retq
   %ext = sext i1 %x to i32
   %neg = sub i32 0, %ext
   %and = and i32 %neg, 1
   ret i32 %and
 }
 
 define <4 x i32> @mask_negated_sext_bool_vec(<4 x i1> %x) {

test/CodeGen/X86/negate-i1.ll

 ; X32-NEXT:    retl
   %b = sext i1 %a to i16
   ret i16 %b
 }
 
 define i16 @select_i16_neg1_or_0_zeroext(i1 zeroext %a) {
 ; X64-LABEL: select_i16_neg1_or_0_zeroext:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    negl %eax
-; X64-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
+; X64-NEXT:    negl %edi
+; X64-NEXT:    movl %edi, %eax
 ; X64-NEXT:    retq
 ;
 ; X32-LABEL: select_i16_neg1_or_0_zeroext:
 ; X32:       # BB#0:
 ; X32-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    negl %eax
 ; X32-NEXT:    # kill: %AX<def> %AX<kill> %EAX<kill>
 ; X32-NEXT:    retl

 ; X32-NEXT:    retl
   %b = sext i1 %a to i32
   ret i32 %b
 }
 
 define i32 @select_i32_neg1_or_0_zeroext(i1 zeroext %a) {
 ; X64-LABEL: select_i32_neg1_or_0_zeroext:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    negl %eax
+; X64-NEXT:    negl %edi
+; X64-NEXT:    movl %edi, %eax
 ; X64-NEXT:    retq
 ;
 ; X32-LABEL: select_i32_neg1_or_0_zeroext:
 ; X32:       # BB#0:
 ; X32-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    negl %eax
 ; X32-NEXT:    retl
   %b = sext i1 %a to i32

 ; X32-NEXT:    retl
   %b = sext i1 %a to i64
   ret i64 %b
 }
 
 define i64 @select_i64_neg1_or_0_zeroext(i1 zeroext %a) {
 ; X64-LABEL: select_i64_neg1_or_0_zeroext:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %dil, %eax
+; X64-NEXT:    movl %edi, %eax
 ; X64-NEXT:    negq %rax
 ; X64-NEXT:    retq
 ;
 ; X32-LABEL: select_i64_neg1_or_0_zeroext:
 ; X32:       # BB#0:
 ; X32-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    negl %eax
 ; X32-NEXT:    movl %eax, %edx
 ; X32-NEXT:    retl
   %b = sext i1 %a to i64
   ret i64 %b
 }
 

test/CodeGen/X86/select_const.ll

 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 1, i32 0
   ret i32 %sel
 }
 
 define i32 @select_1_or_0_zeroext(i1 zeroext %cond) {
 ; CHECK-LABEL: select_1_or_0_zeroext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %eax
+; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 1, i32 0
   ret i32 %sel
 }
 
 define i32 @select_1_or_0_signext(i1 signext %cond) {
 ; CHECK-LABEL: select_1_or_0_signext:
 ; CHECK:       # BB#0:

 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 0, i32 -1
   ret i32 %sel
 }
 
 define i32 @select_0_or_neg1_zeroext(i1 zeroext %cond) {
 ; CHECK-LABEL: select_0_or_neg1_zeroext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %eax
-; CHECK-NEXT:    decl %eax
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal -1(%rdi), %eax
 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 0, i32 -1
   ret i32 %sel
 }
 
 define i32 @select_0_or_neg1_signext(i1 signext %cond) {
 ; CHECK-LABEL: select_0_or_neg1_signext:
 ; CHECK:       # BB#0:

 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 -1, i32 0
   ret i32 %sel
 }
 
 define i32 @select_neg1_or_0_zeroext(i1 zeroext %cond) {
 ; CHECK-LABEL: select_neg1_or_0_zeroext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %eax
-; CHECK-NEXT:    negl %eax
+; CHECK-NEXT:    negl %edi
+; CHECK-NEXT:    movl %edi, %eax
 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 -1, i32 0
   ret i32 %sel
 }
 
 define i32 @select_neg1_or_0_signext(i1 signext %cond) {
 ; CHECK-LABEL: select_neg1_or_0_signext:
 ; CHECK:       # BB#0:

 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 42, i32 41
   ret i32 %sel
 }
 
 define i32 @select_Cplus1_C_zeroext(i1 zeroext %cond) {
 ; CHECK-LABEL: select_Cplus1_C_zeroext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %eax
-; CHECK-NEXT:    addl $41, %eax
+; CHECK-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; CHECK-NEXT:    leal 41(%rdi), %eax
 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 42, i32 41
   ret i32 %sel
 }
 
 define i32 @select_Cplus1_C_signext(i1 signext %cond) {
 ; CHECK-LABEL: select_Cplus1_C_signext:
 ; CHECK:       # BB#0:

 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 41, i32 42
   ret i32 %sel
 }
 
 define i32 @select_C_Cplus1_zeroext(i1 zeroext %cond) {
 ; CHECK-LABEL: select_C_Cplus1_zeroext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    movzbl %dil, %ecx
 ; CHECK-NEXT:    movl $42, %eax
-; CHECK-NEXT:    subl %ecx, %eax
+; CHECK-NEXT:    subl %edi, %eax
 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 41, i32 42
   ret i32 %sel
 }
 
 define i32 @select_C_Cplus1_signext(i1 signext %cond) {
 ; CHECK-LABEL: select_C_Cplus1_signext:
 ; CHECK:       # BB#0:

 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 421, i32 42
   ret i32 %sel
 }
 
 define i32 @select_C1_C2_zeroext(i1 zeroext %cond) {
 ; CHECK-LABEL: select_C1_C2_zeroext:
 ; CHECK:       # BB#0:
-; CHECK-NEXT:    testb %dil, %dil
+; CHECK-NEXT:    testl %edi, %edi
 ; CHECK-NEXT:    movl $421, %ecx # imm = 0x1A5
 ; CHECK-NEXT:    movl $42, %eax
 ; CHECK-NEXT:    cmovnel %ecx, %eax
 ; CHECK-NEXT:    retq
   %sel = select i1 %cond, i32 421, i32 42
   ret i32 %sel
 }
 

test/CodeGen/X86/sext-i1.ll

 ; X32-LABEL: select_0_or_1s_zeroext:
 ; X32:       # BB#0:
 ; X32-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
 ; X32-NEXT:    decl %eax
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: select_0_or_1s_zeroext:
 ; X64:       # BB#0:
-; X64-NEXT:    movzbl %dil, %eax
-; X64-NEXT:    decl %eax
+; X64-NEXT:    # kill: %EDI<def> %EDI<kill> %RDI<def>
+; X64-NEXT:    leal -1(%rdi), %eax
 ; X64-NEXT:    retq
   %not = xor i1 %cond, 1
   %sext = sext i1 %not to i32
   ret i32 %sext
 }
 
 ; sext (xor Bool, -1) --> sub (zext Bool), 1