[llvm] [DAGCombiner] Simplifying `{si|ui}tofp` when only signbit is needed (PR #85445)
via llvm-commits
llvm-commits at lists.llvm.org
Sun Mar 17 11:09:35 PDT 2024
https://github.com/goldsteinn updated https://github.com/llvm/llvm-project/pull/85445
>From 37e6b341fc75f23c980fa3c584bda9b6a4d05636 Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Fri, 15 Mar 2024 12:44:15 -0500
Subject: [PATCH 1/2] [DAGCombiner] Add tests for simplifying `{si|ui}tofp`;
NFC
---
llvm/test/CodeGen/X86/int-to-fp-demanded.ll | 382 ++++++++++++++++++++
1 file changed, 382 insertions(+)
create mode 100644 llvm/test/CodeGen/X86/int-to-fp-demanded.ll
diff --git a/llvm/test/CodeGen/X86/int-to-fp-demanded.ll b/llvm/test/CodeGen/X86/int-to-fp-demanded.ll
new file mode 100644
index 00000000000000..cdde03fb0534b9
--- /dev/null
+++ b/llvm/test/CodeGen/X86/int-to-fp-demanded.ll
@@ -0,0 +1,382 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 4
+; 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
+
+declare void @use.i1(i1)
+declare void @use.i32(i32)
+define i32 @sitofp_signbit_only(i32 %i_in) nounwind {
+; X86-LABEL: sitofp_signbit_only:
+; X86: # %bb.0:
+; X86-NEXT: subl $8, %esp
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl %eax, (%esp)
+; X86-NEXT: fildl (%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movl $-2147483648, %eax # imm = 0x80000000
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl $8, %esp
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_signbit_only:
+; X64: # %bb.0:
+; X64-NEXT: cvtsi2ss %edi, %xmm0
+; X64-NEXT: movmskps %xmm0, %eax
+; X64-NEXT: shll $31, %eax
+; X64-NEXT: retq
+ %f = sitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @sitofp_signbit_only_okay_width(i16 %i_in) nounwind {
+; X86-LABEL: sitofp_signbit_only_okay_width:
+; X86: # %bb.0:
+; X86-NEXT: subl $8, %esp
+; X86-NEXT: movzwl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movw %ax, {{[0-9]+}}(%esp)
+; X86-NEXT: filds {{[0-9]+}}(%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movl $-2147483648, %eax # imm = 0x80000000
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl $8, %esp
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_signbit_only_okay_width:
+; X64: # %bb.0:
+; X64-NEXT: movswl %di, %eax
+; X64-NEXT: cvtsi2ss %eax, %xmm0
+; X64-NEXT: movmskps %xmm0, %eax
+; X64-NEXT: shll $31, %eax
+; X64-NEXT: retq
+ %f = sitofp i16 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @sitofp_signbit_only_fail_bad_width1(i64 %i_in) nounwind {
+; X86-LABEL: sitofp_signbit_only_fail_bad_width1:
+; X86: # %bb.0:
+; X86-NEXT: pushl %eax
+; X86-NEXT: fildll {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: movl $-2147483648, %eax # imm = 0x80000000
+; X86-NEXT: andl (%esp), %eax
+; X86-NEXT: popl %ecx
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_signbit_only_fail_bad_width1:
+; X64: # %bb.0:
+; X64-NEXT: cvtsi2ss %rdi, %xmm0
+; X64-NEXT: movmskps %xmm0, %eax
+; X64-NEXT: shll $31, %eax
+; X64-NEXT: retq
+ %f = sitofp i64 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define <2 x i16> @sitofp_signbit_only_fail_bad_width2(i32 %i_in) nounwind {
+; X86-LABEL: sitofp_signbit_only_fail_bad_width2:
+; X86: # %bb.0:
+; X86-NEXT: subl $8, %esp
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl %eax, (%esp)
+; X86-NEXT: fildl (%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl %eax, %edx
+; X86-NEXT: shrl $16, %edx
+; X86-NEXT: andl $32768, %eax # imm = 0x8000
+; X86-NEXT: andl $32768, %edx # imm = 0x8000
+; X86-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-NEXT: # kill: def $dx killed $dx killed $edx
+; X86-NEXT: addl $8, %esp
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_signbit_only_fail_bad_width2:
+; X64: # %bb.0:
+; X64-NEXT: cvtsi2ss %edi, %xmm0
+; X64-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT: retq
+ %f = sitofp i32 %i_in to float
+ %i2xi16 = bitcast float %f to <2 x i16>
+ %r = and <2 x i16> %i2xi16, <i16 32768, i16 32768>
+ ret <2 x i16> %r
+}
+
+define i32 @sitofp_many_bits_fail(i32 %i_in) nounwind {
+; X86-LABEL: sitofp_many_bits_fail:
+; X86: # %bb.0:
+; X86-NEXT: subl $8, %esp
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl %eax, (%esp)
+; X86-NEXT: fildl (%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movl $-2147483647, %eax # imm = 0x80000001
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl $8, %esp
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_many_bits_fail:
+; X64: # %bb.0:
+; X64-NEXT: cvtsi2ss %edi, %xmm0
+; X64-NEXT: movd %xmm0, %eax
+; X64-NEXT: andl $-2147483647, %eax # imm = 0x80000001
+; X64-NEXT: retq
+ %f = sitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483649
+ ret i32 %r
+}
+
+define i32 @sitofp_multiuse_fail(i32 %i_in) nounwind {
+; X86-LABEL: sitofp_multiuse_fail:
+; X86: # %bb.0:
+; X86-NEXT: subl $12, %esp
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT: fildl {{[0-9]+}}(%esp)
+; X86-NEXT: fsts {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: calll use.i32 at PLT
+; X86-NEXT: movl $-2147483648, %eax # imm = 0x80000000
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl $12, %esp
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_multiuse_fail:
+; X64: # %bb.0:
+; X64-NEXT: pushq %rbx
+; X64-NEXT: cvtsi2ss %edi, %xmm0
+; X64-NEXT: movd %xmm0, %ebx
+; X64-NEXT: movl %ebx, %edi
+; X64-NEXT: callq use.i32 at PLT
+; X64-NEXT: andl $-2147483648, %ebx # imm = 0x80000000
+; X64-NEXT: movl %ebx, %eax
+; X64-NEXT: popq %rbx
+; X64-NEXT: retq
+ %f = sitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ call void @use.i32(i32 %i)
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @sitofp_multiuse_okay(i32 %i_in) nounwind {
+; X86-LABEL: sitofp_multiuse_okay:
+; X86: # %bb.0:
+; X86-NEXT: subl $12, %esp
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT: fildl {{[0-9]+}}(%esp)
+; X86-NEXT: fsts {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: calll use.i1 at PLT
+; X86-NEXT: movl $-2147483648, %eax # imm = 0x80000000
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: addl $12, %esp
+; X86-NEXT: retl
+;
+; X64-LABEL: sitofp_multiuse_okay:
+; X64: # %bb.0:
+; X64-NEXT: pushq %rbx
+; X64-NEXT: cvtsi2ss %edi, %xmm0
+; X64-NEXT: movd %xmm0, %ebx
+; X64-NEXT: movl %ebx, %edi
+; X64-NEXT: callq use.i1 at PLT
+; X64-NEXT: andl $-2147483648, %ebx # imm = 0x80000000
+; X64-NEXT: movl %ebx, %eax
+; X64-NEXT: popq %rbx
+; X64-NEXT: retq
+ %f = sitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ %cmp = icmp slt i32 %i, 0
+ call void @use.i1(i32 %i)
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @uitofp_signbit_only(i32 %i_in) nounwind {
+; X86-LABEL: uitofp_signbit_only:
+; X86: # %bb.0:
+; X86-NEXT: xorl %eax, %eax
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_signbit_only:
+; X64: # %bb.0:
+; X64-NEXT: xorl %eax, %eax
+; X64-NEXT: retq
+ %f = uitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @uitofp_signbit_only_okay_width(i16 %i_in) nounwind {
+; X86-LABEL: uitofp_signbit_only_okay_width:
+; X86: # %bb.0:
+; X86-NEXT: xorl %eax, %eax
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_signbit_only_okay_width:
+; X64: # %bb.0:
+; X64-NEXT: xorl %eax, %eax
+; X64-NEXT: retq
+ %f = uitofp i16 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @uitofp_signbit_only_okay_width1(i64 %i_in) nounwind {
+; X86-LABEL: uitofp_signbit_only_okay_width1:
+; X86: # %bb.0:
+; X86-NEXT: xorl %eax, %eax
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_signbit_only_okay_width1:
+; X64: # %bb.0:
+; X64-NEXT: xorl %eax, %eax
+; X64-NEXT: retq
+ %f = uitofp i64 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define <2 x i16> @uitofp_signbit_only_fail_bad_width2(i32 %i_in) nounwind {
+; X86-LABEL: uitofp_signbit_only_fail_bad_width2:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebp
+; X86-NEXT: movl %esp, %ebp
+; X86-NEXT: andl $-8, %esp
+; X86-NEXT: subl $16, %esp
+; X86-NEXT: movl 8(%ebp), %eax
+; X86-NEXT: movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
+; X86-NEXT: fildll {{[0-9]+}}(%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movl $32768, %eax # imm = 0x8000
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: # kill: def $ax killed $ax killed $eax
+; X86-NEXT: xorl %edx, %edx
+; X86-NEXT: movl %ebp, %esp
+; X86-NEXT: popl %ebp
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_signbit_only_fail_bad_width2:
+; X64: # %bb.0:
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: cvtsi2ss %rax, %xmm0
+; X64-NEXT: andps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0
+; X64-NEXT: retq
+ %f = uitofp i32 %i_in to float
+ %i2xi16 = bitcast float %f to <2 x i16>
+ %r = and <2 x i16> %i2xi16, <i16 32768, i16 32768>
+ ret <2 x i16> %r
+}
+
+define i32 @uitofp_many_bits_fail(i32 %i_in) nounwind {
+; X86-LABEL: uitofp_many_bits_fail:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebp
+; X86-NEXT: movl %esp, %ebp
+; X86-NEXT: andl $-8, %esp
+; X86-NEXT: subl $16, %esp
+; X86-NEXT: movl 8(%ebp), %eax
+; X86-NEXT: movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
+; X86-NEXT: fildll {{[0-9]+}}(%esp)
+; X86-NEXT: fstps {{[0-9]+}}(%esp)
+; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; X86-NEXT: andl $1, %eax
+; X86-NEXT: movl %ebp, %esp
+; X86-NEXT: popl %ebp
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_many_bits_fail:
+; X64: # %bb.0:
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: cvtsi2ss %rax, %xmm0
+; X64-NEXT: movd %xmm0, %eax
+; X64-NEXT: andl $1, %eax
+; X64-NEXT: retq
+ %f = uitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ %r = and i32 %i, 2147483649
+ ret i32 %r
+}
+
+define i32 @uitofp_multiuse_fail(i32 %i_in) nounwind {
+; X86-LABEL: uitofp_multiuse_fail:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebp
+; X86-NEXT: movl %esp, %ebp
+; X86-NEXT: andl $-8, %esp
+; X86-NEXT: subl $16, %esp
+; X86-NEXT: movl 8(%ebp), %eax
+; X86-NEXT: movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
+; X86-NEXT: fildll {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: calll use.i32 at PLT
+; X86-NEXT: xorl %eax, %eax
+; X86-NEXT: movl %ebp, %esp
+; X86-NEXT: popl %ebp
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_multiuse_fail:
+; X64: # %bb.0:
+; X64-NEXT: pushq %rax
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: cvtsi2ss %rax, %xmm0
+; X64-NEXT: movd %xmm0, %edi
+; X64-NEXT: callq use.i32 at PLT
+; X64-NEXT: xorl %eax, %eax
+; X64-NEXT: popq %rcx
+; X64-NEXT: retq
+ %f = uitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ call void @use.i32(i32 %i)
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
+
+define i32 @uitofp_multiuse_okay(i32 %i_in) nounwind {
+; X86-LABEL: uitofp_multiuse_okay:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebp
+; X86-NEXT: movl %esp, %ebp
+; X86-NEXT: andl $-8, %esp
+; X86-NEXT: subl $16, %esp
+; X86-NEXT: movl 8(%ebp), %eax
+; X86-NEXT: movl %eax, {{[0-9]+}}(%esp)
+; X86-NEXT: movl $0, {{[0-9]+}}(%esp)
+; X86-NEXT: fildll {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: calll use.i1 at PLT
+; X86-NEXT: xorl %eax, %eax
+; X86-NEXT: movl %ebp, %esp
+; X86-NEXT: popl %ebp
+; X86-NEXT: retl
+;
+; X64-LABEL: uitofp_multiuse_okay:
+; X64: # %bb.0:
+; X64-NEXT: pushq %rax
+; X64-NEXT: movl %edi, %eax
+; X64-NEXT: cvtsi2ss %rax, %xmm0
+; X64-NEXT: movd %xmm0, %edi
+; X64-NEXT: callq use.i1 at PLT
+; X64-NEXT: xorl %eax, %eax
+; X64-NEXT: popq %rcx
+; X64-NEXT: retq
+ %f = uitofp i32 %i_in to float
+ %i = bitcast float %f to i32
+ %cmp = icmp slt i32 %i, 0
+ call void @use.i1(i32 %i)
+ %r = and i32 %i, 2147483648
+ ret i32 %r
+}
>From b8b3e10bc5ba6b8d62a4f2da61f6efa5a302636f Mon Sep 17 00:00:00 2001
From: Noah Goldstein <goldstein.w.n at gmail.com>
Date: Fri, 15 Mar 2024 12:45:31 -0500
Subject: [PATCH 2/2] [DAGCombiner] Simplifying `{si|ui}tofp` when only signbit
is needed
If we only need the signbit `uitofp` simplified to 0, and `sitofp`
simplifies to `bitcast`.
---
.../CodeGen/SelectionDAG/TargetLowering.cpp | 30 +++++++++++++++++++
.../CodeGen/X86/combine-sse41-intrinsics.ll | 3 +-
llvm/test/CodeGen/X86/int-to-fp-demanded.ll | 21 +++++--------
3 files changed, 38 insertions(+), 16 deletions(-)
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 57f8fc409de453..16069c6c0dc317 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -611,6 +611,25 @@ bool TargetLowering::ShrinkDemandedOp(SDValue Op, unsigned BitWidth,
return false;
}
+static SDValue simplifyUseOfIntToFP(SDValue Op, const APInt &DemandedBits,
+ SelectionDAG &DAG) {
+ unsigned Opc = Op.getOpcode();
+ assert((Opc == ISD::SINT_TO_FP || Opc == ISD::UINT_TO_FP) &&
+ "Invalid Int -> FP Opcode");
+ if (!DemandedBits.isSignMask())
+ return SDValue();
+
+ EVT VT = Op.getValueType();
+ if (Opc == ISD::UINT_TO_FP)
+ return DAG.getConstant(0, SDLoc(Op), VT);
+
+ EVT InnerVT = Op.getOperand(0).getValueType();
+ if (VT.getScalarSizeInBits() == InnerVT.getScalarSizeInBits())
+ return DAG.getBitcast(VT, Op.getOperand(0));
+
+ return SDValue();
+}
+
bool TargetLowering::SimplifyDemandedBits(SDValue Op, const APInt &DemandedBits,
DAGCombinerInfo &DCI) const {
SelectionDAG &DAG = DCI.DAG;
@@ -816,6 +835,11 @@ SDValue TargetLowering::SimplifyMultipleUseDemandedBits(
}
break;
}
+ case ISD::UINT_TO_FP:
+ case ISD::SINT_TO_FP:
+ if (SDValue R = simplifyUseOfIntToFP(Op, DemandedBits, DAG))
+ return R;
+ break;
case ISD::SIGN_EXTEND_INREG: {
// If none of the extended bits are demanded, eliminate the sextinreg.
SDValue Op0 = Op.getOperand(0);
@@ -2313,6 +2337,12 @@ bool TargetLowering::SimplifyDemandedBits(
Known = TLO.DAG.computeKnownBits(Op, DemandedElts, Depth);
break;
}
+ case ISD::UINT_TO_FP:
+ case ISD::SINT_TO_FP:
+ if (SDValue R = simplifyUseOfIntToFP(Op, DemandedBits, TLO.DAG))
+ return TLO.CombineTo(Op, R);
+ Known = TLO.DAG.computeKnownBits(Op, DemandedElts, Depth);
+ break;
case ISD::SIGN_EXTEND_INREG: {
SDValue Op0 = Op.getOperand(0);
EVT ExVT = cast<VTSDNode>(Op.getOperand(1))->getVT();
diff --git a/llvm/test/CodeGen/X86/combine-sse41-intrinsics.ll b/llvm/test/CodeGen/X86/combine-sse41-intrinsics.ll
index cbb5bd09c2399a..a332b3e8908003 100644
--- a/llvm/test/CodeGen/X86/combine-sse41-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/combine-sse41-intrinsics.ll
@@ -164,14 +164,13 @@ define <4 x float> @demandedbits_sitofp_blendvps(<4 x float> %a0, <4 x float> %a
; SSE-LABEL: demandedbits_sitofp_blendvps:
; SSE: # %bb.0:
; SSE-NEXT: movaps %xmm0, %xmm3
-; SSE-NEXT: cvtdq2ps %xmm2, %xmm0
+; SSE-NEXT: movaps %xmm2, %xmm0
; SSE-NEXT: blendvps %xmm0, %xmm1, %xmm3
; SSE-NEXT: movaps %xmm3, %xmm0
; SSE-NEXT: retq
;
; AVX-LABEL: demandedbits_sitofp_blendvps:
; AVX: # %bb.0:
-; AVX-NEXT: vcvtdq2ps %xmm2, %xmm2
; AVX-NEXT: vblendvps %xmm2, %xmm1, %xmm0, %xmm0
; AVX-NEXT: retq
%cvt = sitofp <4 x i32> %a2 to <4 x float>
diff --git a/llvm/test/CodeGen/X86/int-to-fp-demanded.ll b/llvm/test/CodeGen/X86/int-to-fp-demanded.ll
index cdde03fb0534b9..8652136ae5cd94 100644
--- a/llvm/test/CodeGen/X86/int-to-fp-demanded.ll
+++ b/llvm/test/CodeGen/X86/int-to-fp-demanded.ll
@@ -7,19 +7,13 @@ declare void @use.i32(i32)
define i32 @sitofp_signbit_only(i32 %i_in) nounwind {
; X86-LABEL: sitofp_signbit_only:
; X86: # %bb.0:
-; X86-NEXT: subl $8, %esp
-; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: movl %eax, (%esp)
-; X86-NEXT: fildl (%esp)
-; X86-NEXT: fstps {{[0-9]+}}(%esp)
; X86-NEXT: movl $-2147483648, %eax # imm = 0x80000000
; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: addl $8, %esp
; X86-NEXT: retl
;
; X64-LABEL: sitofp_signbit_only:
; X64: # %bb.0:
-; X64-NEXT: cvtsi2ss %edi, %xmm0
+; X64-NEXT: movd %edi, %xmm0
; X64-NEXT: movmskps %xmm0, %eax
; X64-NEXT: shll $31, %eax
; X64-NEXT: retq
@@ -44,8 +38,8 @@ define i32 @sitofp_signbit_only_okay_width(i16 %i_in) nounwind {
;
; X64-LABEL: sitofp_signbit_only_okay_width:
; X64: # %bb.0:
-; X64-NEXT: movswl %di, %eax
-; X64-NEXT: cvtsi2ss %eax, %xmm0
+; X64-NEXT: shll $16, %edi
+; X64-NEXT: movd %edi, %xmm0
; X64-NEXT: movmskps %xmm0, %eax
; X64-NEXT: shll $31, %eax
; X64-NEXT: retq
@@ -82,15 +76,14 @@ define <2 x i16> @sitofp_signbit_only_fail_bad_width2(i32 %i_in) nounwind {
; X86-LABEL: sitofp_signbit_only_fail_bad_width2:
; X86: # %bb.0:
; X86-NEXT: subl $8, %esp
-; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: movl %eax, (%esp)
+; X86-NEXT: movl {{[0-9]+}}(%esp), %edx
+; X86-NEXT: movl %edx, (%esp)
; X86-NEXT: fildl (%esp)
; X86-NEXT: fstps {{[0-9]+}}(%esp)
-; X86-NEXT: movl {{[0-9]+}}(%esp), %eax
-; X86-NEXT: movl %eax, %edx
; X86-NEXT: shrl $16, %edx
-; X86-NEXT: andl $32768, %eax # imm = 0x8000
; X86-NEXT: andl $32768, %edx # imm = 0x8000
+; X86-NEXT: movl $32768, %eax # imm = 0x8000
+; X86-NEXT: andl {{[0-9]+}}(%esp), %eax
; X86-NEXT: # kill: def $ax killed $ax killed $eax
; X86-NEXT: # kill: def $dx killed $dx killed $edx
; X86-NEXT: addl $8, %esp
More information about the llvm-commits
mailing list