[X86][GlobalISel] Support fp80 for G_FPTRUNC and G_FPEXT #141611
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-llvm-globalisel Author: Evgenii Kudriashov (e-kud) ChangesWe intentionally omit Full diff: https://github.com/llvm/llvm-project/pull/141611.diff 4 Files Affected:
diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
index f008cb1bea839..58215d4e00202 100644
--- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
+++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
@@ -376,9 +376,15 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
Action.legalForTypesWithMemDesc({{s8, p0, s8, 1},
{s16, p0, s16, 1},
{s32, p0, s32, 1},
- {s80, p0, s80, 1},
{p0, p0, p0, 1},
{v4s8, p0, v4s8, 1}});
+
+ if (UseX87)
+ Action.legalForTypesWithMemDesc({{s80, p0, s32, 1},
+ {s80, p0, s64, 1},
+ {s32, p0, s80, 1},
+ {s64, p0, s80, 1},
+ {s80, p0, s80, 1}});
if (Is64Bit)
Action.legalForTypesWithMemDesc(
{{s64, p0, s64, 1}, {v2s32, p0, v2s32, 1}});
@@ -476,18 +482,17 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
.widenScalarToNextPow2(1);
// fp conversions
- getActionDefinitionsBuilder(G_FPEXT).legalIf([=](const LegalityQuery &Query) {
- return (HasSSE2 && typePairInSet(0, 1, {{s64, s32}})(Query)) ||
- (HasAVX && typePairInSet(0, 1, {{v4s64, v4s32}})(Query)) ||
- (HasAVX512 && typePairInSet(0, 1, {{v8s64, v8s32}})(Query));
- });
-
- getActionDefinitionsBuilder(G_FPTRUNC).legalIf(
- [=](const LegalityQuery &Query) {
- return (HasSSE2 && typePairInSet(0, 1, {{s32, s64}})(Query)) ||
- (HasAVX && typePairInSet(0, 1, {{v4s32, v4s64}})(Query)) ||
- (HasAVX512 && typePairInSet(0, 1, {{v8s32, v8s64}})(Query));
- });
+ getActionDefinitionsBuilder(G_FPEXT)
+ .legalFor(HasSSE2, {{s64, s32}})
+ .legalFor(HasAVX, {{v4s64, v4s32}})
+ .legalFor(HasAVX512, {{v8s64, v8s32}})
+ .customFor(UseX87, {{s64, s32}, {s80, s32}, {s80, s64}});
+
+ getActionDefinitionsBuilder(G_FPTRUNC)
+ .legalFor(HasSSE2, {{s32, s64}})
+ .legalFor(HasAVX, {{v4s32, v4s64}})
+ .legalFor(HasAVX512, {{v8s32, v8s64}})
+ .customFor(UseX87, {{s32, s64}, {s32, s80}, {s64, s80}});
getActionDefinitionsBuilder(G_SITOFP)
.legalIf([=](const LegalityQuery &Query) {
@@ -671,6 +676,9 @@ bool X86LegalizerInfo::legalizeCustom(LegalizerHelper &Helper, MachineInstr &MI,
return legalizeUITOFP(MI, MRI, Helper);
case TargetOpcode::G_STORE:
return legalizeNarrowingStore(MI, MRI, Helper);
+ case TargetOpcode::G_FPEXT:
+ case TargetOpcode::G_FPTRUNC:
+ return legalizeFPExtAndTrunc(MI, MRI, Helper);
}
llvm_unreachable("expected switch to return");
}
@@ -781,6 +789,33 @@ bool X86LegalizerInfo::legalizeNarrowingStore(MachineInstr &MI,
return true;
}
+bool X86LegalizerInfo::legalizeFPExtAndTrunc(MachineInstr &MI,
+ MachineRegisterInfo &MRI,
+ LegalizerHelper &Helper) const {
+ assert((MI.getOpcode() == TargetOpcode::G_FPEXT ||
+ MI.getOpcode() == TargetOpcode::G_FPTRUNC) &&
+ "Only G_FPEXT and G_FPTRUNC are expected");
+ auto [DstReg, DstTy, SrcReg, SrcTy] = MI.getFirst2RegLLTs();
+ MachinePointerInfo PtrInfo;
+ LLT StackTy = MI.getOpcode() == TargetOpcode::G_FPEXT ? SrcTy : DstTy;
+ Align StackTyAlign = Helper.getStackTemporaryAlignment(StackTy);
+ auto StackTemp = Helper.createStackTemporary(StackTy.getSizeInBytes(),
+ StackTyAlign, PtrInfo);
+
+ MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
+ MachineFunction &MF = MIRBuilder.getMF();
+ auto *StoreMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
+ StackTy, StackTyAlign);
+ MIRBuilder.buildStore(SrcReg, StackTemp, *StoreMMO);
+
+ auto *LoadMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad,
+ StackTy, StackTyAlign);
+ MIRBuilder.buildLoad(DstReg, StackTemp, *LoadMMO);
+
+ MI.eraseFromParent();
+ return true;
+}
+
bool X86LegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
MachineInstr &MI) const {
return true;
diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
index 54f776456397b..b224f3f46a2d5 100644
--- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
+++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
@@ -48,6 +48,9 @@ class X86LegalizerInfo : public LegalizerInfo {
bool legalizeNarrowingStore(MachineInstr &MI, MachineRegisterInfo &MRI,
LegalizerHelper &Helper) const;
+
+ bool legalizeFPExtAndTrunc(MachineInstr &MI, MachineRegisterInfo &MRI,
+ LegalizerHelper &Helper) const;
};
} // namespace llvm
#endif
diff --git a/llvm/test/CodeGen/X86/GlobalISel/fpext-scalar.ll b/llvm/test/CodeGen/X86/GlobalISel/fpext-scalar.ll
deleted file mode 100644
index 8501009e2915a..0000000000000
--- a/llvm/test/CodeGen/X86/GlobalISel/fpext-scalar.ll
+++ /dev/null
@@ -1,12 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=x86_64-linux-gnu -global-isel -verify-machineinstrs < %s -o - | FileCheck %s --check-prefix=CHECK
-
-define double @test(float %a) {
-; CHECK-LABEL: test:
-; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: cvtss2sd %xmm0, %xmm0
-; CHECK-NEXT: retq
-entry:
- %conv = fpext float %a to double
- ret double %conv
-}
diff --git a/llvm/test/CodeGen/X86/isel-fptrunc-fpext.ll b/llvm/test/CodeGen/X86/isel-fptrunc-fpext.ll
new file mode 100644
index 0000000000000..0ad9b90806ce9
--- /dev/null
+++ b/llvm/test/CodeGen/X86/isel-fptrunc-fpext.ll
@@ -0,0 +1,279 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes X86,FASTSDAG-X86,SDAG-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes SSE,FASTSDAG-SSE,SDAG-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,SDAG-AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,SDAG-AVX
+; COMM: FastISel has troubles with fp80 type
+; RUN: llc < %s -mtriple=i686-unknown-unknown -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes X86,FASTSDAG-X86,FAST-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes SSE,FASTSDAG-SSE,FAST-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,FAST-AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,FAST-AVX
+; COMM: GlobalISel can't legalize double stores on 32bit platform due to lack of double/integer distinguish during legalization
+; RUN: llc < %s -mtriple=i686-unknown-unknown -fast-isel=0 -global-isel=1 -global-isel-abort=2 | FileCheck %s --check-prefixes X86,GLOBAL-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel=0 -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes SSE,GLOBAL-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel=0 -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes AVX,GLOBAL-AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel=0 -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes AVX,GLOBAL-AVX
+
+define double @fpext_float_to_double(float %f) {
+; X86-LABEL: fpext_float_to_double:
+; X86: # %bb.0:
+; X86-NEXT: flds {{[0-9]+}}(%esp)
+; X86-NEXT: retl
+;
+; SSE-LABEL: fpext_float_to_double:
+; SSE: # %bb.0:
+; SSE-NEXT: cvtss2sd %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: fpext_float_to_double:
+; AVX: # %bb.0:
+; AVX-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX-NEXT: retq
+ %1 = fpext float %f to double
+ ret double %1
+}
+
+define x86_fp80 @fpext_float_to_x86_fp80(float %f) {
+; FASTSDAG-X86-LABEL: fpext_float_to_x86_fp80:
+; FASTSDAG-X86: # %bb.0:
+; FASTSDAG-X86-NEXT: flds {{[0-9]+}}(%esp)
+; FASTSDAG-X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fpext_float_to_x86_fp80:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: movss %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: flds -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fpext_float_to_x86_fp80:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: vmovss %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: flds -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-X86-LABEL: fpext_float_to_x86_fp80:
+; GLOBAL-X86: # %bb.0:
+; GLOBAL-X86-NEXT: pushl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 8
+; GLOBAL-X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; GLOBAL-X86-NEXT: movl %eax, (%esp)
+; GLOBAL-X86-NEXT: flds (%esp)
+; GLOBAL-X86-NEXT: popl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 4
+; GLOBAL-X86-NEXT: retl
+;
+; GLOBAL-SSE-LABEL: fpext_float_to_x86_fp80:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: movd %xmm0, %eax
+; GLOBAL-SSE-NEXT: movl %eax, -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: flds -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fpext_float_to_x86_fp80:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: vmovd %xmm0, %eax
+; GLOBAL-AVX-NEXT: movl %eax, -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: flds -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: retq
+ %1 = fpext float %f to x86_fp80
+ ret x86_fp80 %1
+}
+
+define x86_fp80 @fpext_double_to_x86_fp80(double %d) {
+; FASTSDAG-X86-LABEL: fpext_double_to_x86_fp80:
+; FASTSDAG-X86: # %bb.0:
+; FASTSDAG-X86-NEXT: fldl {{[0-9]+}}(%esp)
+; FASTSDAG-X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fpext_double_to_x86_fp80:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: movsd %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: fldl -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fpext_double_to_x86_fp80:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: vmovsd %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: fldl -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-X86-LABEL: fpext_double_to_x86_fp80:
+; GLOBAL-X86: # %bb.0:
+; GLOBAL-X86-NEXT: pushl %ebp
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 8
+; GLOBAL-X86-NEXT: .cfi_offset %ebp, -8
+; GLOBAL-X86-NEXT: movl %esp, %ebp
+; GLOBAL-X86-NEXT: .cfi_def_cfa_register %ebp
+; GLOBAL-X86-NEXT: andl $-8, %esp
+; GLOBAL-X86-NEXT: subl $8, %esp
+; GLOBAL-X86-NEXT: leal 8(%ebp), %eax
+; GLOBAL-X86-NEXT: movl 8(%ebp), %ecx
+; GLOBAL-X86-NEXT: movl 4(%eax), %eax
+; GLOBAL-X86-NEXT: movl %esp, %edx
+; GLOBAL-X86-NEXT: movl %ecx, (%esp)
+; GLOBAL-X86-NEXT: movl %eax, 4(%edx)
+; GLOBAL-X86-NEXT: fldl (%esp)
+; GLOBAL-X86-NEXT: movl %ebp, %esp
+; GLOBAL-X86-NEXT: popl %ebp
+; GLOBAL-X86-NEXT: .cfi_def_cfa %esp, 4
+; GLOBAL-X86-NEXT: retl
+;
+; GLOBAL-SSE-LABEL: fpext_double_to_x86_fp80:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: movq %xmm0, %rax
+; GLOBAL-SSE-NEXT: movq %rax, -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: fldl -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fpext_double_to_x86_fp80:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: vmovq %xmm0, %rax
+; GLOBAL-AVX-NEXT: movq %rax, -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: fldl -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: retq
+ %1 = fpext double %d to x86_fp80
+ ret x86_fp80 %1
+}
+
+define float @fptrunc_double_to_float(double %d) {
+; FASTSDAG-X86-LABEL: fptrunc_double_to_float:
+; FASTSDAG-X86: # %bb.0:
+; FASTSDAG-X86-NEXT: pushl %eax
+; FASTSDAG-X86-NEXT: .cfi_def_cfa_offset 8
+; FASTSDAG-X86-NEXT: fldl {{[0-9]+}}(%esp)
+; FASTSDAG-X86-NEXT: fstps (%esp)
+; FASTSDAG-X86-NEXT: flds (%esp)
+; FASTSDAG-X86-NEXT: popl %eax
+; FASTSDAG-X86-NEXT: .cfi_def_cfa_offset 4
+; FASTSDAG-X86-NEXT: retl
+;
+; SSE-LABEL: fptrunc_double_to_float:
+; SSE: # %bb.0:
+; SSE-NEXT: cvtsd2ss %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: fptrunc_double_to_float:
+; AVX: # %bb.0:
+; AVX-NEXT: vcvtsd2ss %xmm0, %xmm0, %xmm0
+; AVX-NEXT: retq
+;
+; GLOBAL-X86-LABEL: fptrunc_double_to_float:
+; GLOBAL-X86: # %bb.0:
+; GLOBAL-X86-NEXT: pushl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 8
+; GLOBAL-X86-NEXT: leal {{[0-9]+}}(%esp), %eax
+; GLOBAL-X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; GLOBAL-X86-NEXT: movl 4(%eax), %eax
+; GLOBAL-X86-NEXT: movl %esp, %edx
+; GLOBAL-X86-NEXT: movl %ecx, (%esp)
+; GLOBAL-X86-NEXT: movl %eax, 4(%edx)
+; GLOBAL-X86-NEXT: flds (%esp)
+; GLOBAL-X86-NEXT: popl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 4
+; GLOBAL-X86-NEXT: retl
+ %1 = fptrunc double %d to float
+ ret float %1
+}
+
+define float @fptrunc_x86_fp80_to_float(x86_fp80 %x) {
+; X86-LABEL: fptrunc_x86_fp80_to_float:
+; X86: # %bb.0:
+; X86-NEXT: pushl %eax
+; X86-NEXT: .cfi_def_cfa_offset 8
+; X86-NEXT: fldt {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: flds (%esp)
+; X86-NEXT: popl %eax
+; X86-NEXT: .cfi_def_cfa_offset 4
+; X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fptrunc_x86_fp80_to_float:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: fstps -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fptrunc_x86_fp80_to_float:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: fstps -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-SSE-LABEL: fptrunc_x86_fp80_to_float:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: fstps -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: movl -{{[0-9]+}}(%rsp), %eax
+; GLOBAL-SSE-NEXT: movd %eax, %xmm0
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fptrunc_x86_fp80_to_float:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: fstps -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: movl -{{[0-9]+}}(%rsp), %eax
+; GLOBAL-AVX-NEXT: vmovd %eax, %xmm0
+; GLOBAL-AVX-NEXT: retq
+ %1 = fptrunc x86_fp80 %x to float
+ ret float %1
+}
+
+define double @fptrunc_x86_fp80_to_double(x86_fp80 %x) {
+; X86-LABEL: fptrunc_x86_fp80_to_double:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebp
+; X86-NEXT: .cfi_def_cfa_offset 8
+; X86-NEXT: .cfi_offset %ebp, -8
+; X86-NEXT: movl %esp, %ebp
+; X86-NEXT: .cfi_def_cfa_register %ebp
+; X86-NEXT: andl $-8, %esp
+; X86-NEXT: subl $8, %esp
+; X86-NEXT: fldt 8(%ebp)
+; X86-NEXT: fstpl (%esp)
+; X86-NEXT: fldl (%esp)
+; X86-NEXT: movl %ebp, %esp
+; X86-NEXT: popl %ebp
+; X86-NEXT: .cfi_def_cfa %esp, 4
+; X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fptrunc_x86_fp80_to_double:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fptrunc_x86_fp80_to_double:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-SSE-LABEL: fptrunc_x86_fp80_to_double:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: movq -{{[0-9]+}}(%rsp), %rax
+; GLOBAL-SSE-NEXT: movq %rax, %xmm0
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fptrunc_x86_fp80_to_double:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: movq -{{[0-9]+}}(%rsp), %rax
+; GLOBAL-AVX-NEXT: vmovq %rax, %xmm0
+; GLOBAL-AVX-NEXT: retq
+ %1 = fptrunc x86_fp80 %x to double
+ ret double %1
+}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; FAST-AVX: {{.*}}
+; FAST-SSE: {{.*}}
+; FAST-X86: {{.*}}
+; SDAG-AVX: {{.*}}
+; SDAG-SSE: {{.*}}
+; SDAG-X86: {{.*}}
|
|
@llvm/pr-subscribers-backend-x86 Author: Evgenii Kudriashov (e-kud) ChangesWe intentionally omit Full diff: https://github.com/llvm/llvm-project/pull/141611.diff 4 Files Affected:
diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
index f008cb1bea839..58215d4e00202 100644
--- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
+++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.cpp
@@ -376,9 +376,15 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
Action.legalForTypesWithMemDesc({{s8, p0, s8, 1},
{s16, p0, s16, 1},
{s32, p0, s32, 1},
- {s80, p0, s80, 1},
{p0, p0, p0, 1},
{v4s8, p0, v4s8, 1}});
+
+ if (UseX87)
+ Action.legalForTypesWithMemDesc({{s80, p0, s32, 1},
+ {s80, p0, s64, 1},
+ {s32, p0, s80, 1},
+ {s64, p0, s80, 1},
+ {s80, p0, s80, 1}});
if (Is64Bit)
Action.legalForTypesWithMemDesc(
{{s64, p0, s64, 1}, {v2s32, p0, v2s32, 1}});
@@ -476,18 +482,17 @@ X86LegalizerInfo::X86LegalizerInfo(const X86Subtarget &STI,
.widenScalarToNextPow2(1);
// fp conversions
- getActionDefinitionsBuilder(G_FPEXT).legalIf([=](const LegalityQuery &Query) {
- return (HasSSE2 && typePairInSet(0, 1, {{s64, s32}})(Query)) ||
- (HasAVX && typePairInSet(0, 1, {{v4s64, v4s32}})(Query)) ||
- (HasAVX512 && typePairInSet(0, 1, {{v8s64, v8s32}})(Query));
- });
-
- getActionDefinitionsBuilder(G_FPTRUNC).legalIf(
- [=](const LegalityQuery &Query) {
- return (HasSSE2 && typePairInSet(0, 1, {{s32, s64}})(Query)) ||
- (HasAVX && typePairInSet(0, 1, {{v4s32, v4s64}})(Query)) ||
- (HasAVX512 && typePairInSet(0, 1, {{v8s32, v8s64}})(Query));
- });
+ getActionDefinitionsBuilder(G_FPEXT)
+ .legalFor(HasSSE2, {{s64, s32}})
+ .legalFor(HasAVX, {{v4s64, v4s32}})
+ .legalFor(HasAVX512, {{v8s64, v8s32}})
+ .customFor(UseX87, {{s64, s32}, {s80, s32}, {s80, s64}});
+
+ getActionDefinitionsBuilder(G_FPTRUNC)
+ .legalFor(HasSSE2, {{s32, s64}})
+ .legalFor(HasAVX, {{v4s32, v4s64}})
+ .legalFor(HasAVX512, {{v8s32, v8s64}})
+ .customFor(UseX87, {{s32, s64}, {s32, s80}, {s64, s80}});
getActionDefinitionsBuilder(G_SITOFP)
.legalIf([=](const LegalityQuery &Query) {
@@ -671,6 +676,9 @@ bool X86LegalizerInfo::legalizeCustom(LegalizerHelper &Helper, MachineInstr &MI,
return legalizeUITOFP(MI, MRI, Helper);
case TargetOpcode::G_STORE:
return legalizeNarrowingStore(MI, MRI, Helper);
+ case TargetOpcode::G_FPEXT:
+ case TargetOpcode::G_FPTRUNC:
+ return legalizeFPExtAndTrunc(MI, MRI, Helper);
}
llvm_unreachable("expected switch to return");
}
@@ -781,6 +789,33 @@ bool X86LegalizerInfo::legalizeNarrowingStore(MachineInstr &MI,
return true;
}
+bool X86LegalizerInfo::legalizeFPExtAndTrunc(MachineInstr &MI,
+ MachineRegisterInfo &MRI,
+ LegalizerHelper &Helper) const {
+ assert((MI.getOpcode() == TargetOpcode::G_FPEXT ||
+ MI.getOpcode() == TargetOpcode::G_FPTRUNC) &&
+ "Only G_FPEXT and G_FPTRUNC are expected");
+ auto [DstReg, DstTy, SrcReg, SrcTy] = MI.getFirst2RegLLTs();
+ MachinePointerInfo PtrInfo;
+ LLT StackTy = MI.getOpcode() == TargetOpcode::G_FPEXT ? SrcTy : DstTy;
+ Align StackTyAlign = Helper.getStackTemporaryAlignment(StackTy);
+ auto StackTemp = Helper.createStackTemporary(StackTy.getSizeInBytes(),
+ StackTyAlign, PtrInfo);
+
+ MachineIRBuilder &MIRBuilder = Helper.MIRBuilder;
+ MachineFunction &MF = MIRBuilder.getMF();
+ auto *StoreMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOStore,
+ StackTy, StackTyAlign);
+ MIRBuilder.buildStore(SrcReg, StackTemp, *StoreMMO);
+
+ auto *LoadMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad,
+ StackTy, StackTyAlign);
+ MIRBuilder.buildLoad(DstReg, StackTemp, *LoadMMO);
+
+ MI.eraseFromParent();
+ return true;
+}
+
bool X86LegalizerInfo::legalizeIntrinsic(LegalizerHelper &Helper,
MachineInstr &MI) const {
return true;
diff --git a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
index 54f776456397b..b224f3f46a2d5 100644
--- a/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
+++ b/llvm/lib/Target/X86/GISel/X86LegalizerInfo.h
@@ -48,6 +48,9 @@ class X86LegalizerInfo : public LegalizerInfo {
bool legalizeNarrowingStore(MachineInstr &MI, MachineRegisterInfo &MRI,
LegalizerHelper &Helper) const;
+
+ bool legalizeFPExtAndTrunc(MachineInstr &MI, MachineRegisterInfo &MRI,
+ LegalizerHelper &Helper) const;
};
} // namespace llvm
#endif
diff --git a/llvm/test/CodeGen/X86/GlobalISel/fpext-scalar.ll b/llvm/test/CodeGen/X86/GlobalISel/fpext-scalar.ll
deleted file mode 100644
index 8501009e2915a..0000000000000
--- a/llvm/test/CodeGen/X86/GlobalISel/fpext-scalar.ll
+++ /dev/null
@@ -1,12 +0,0 @@
-; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
-; RUN: llc -mtriple=x86_64-linux-gnu -global-isel -verify-machineinstrs < %s -o - | FileCheck %s --check-prefix=CHECK
-
-define double @test(float %a) {
-; CHECK-LABEL: test:
-; CHECK: # %bb.0: # %entry
-; CHECK-NEXT: cvtss2sd %xmm0, %xmm0
-; CHECK-NEXT: retq
-entry:
- %conv = fpext float %a to double
- ret double %conv
-}
diff --git a/llvm/test/CodeGen/X86/isel-fptrunc-fpext.ll b/llvm/test/CodeGen/X86/isel-fptrunc-fpext.ll
new file mode 100644
index 0000000000000..0ad9b90806ce9
--- /dev/null
+++ b/llvm/test/CodeGen/X86/isel-fptrunc-fpext.ll
@@ -0,0 +1,279 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mtriple=i686-unknown-unknown -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes X86,FASTSDAG-X86,SDAG-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes SSE,FASTSDAG-SSE,SDAG-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,SDAG-AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel=0 -global-isel=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,SDAG-AVX
+; COMM: FastISel has troubles with fp80 type
+; RUN: llc < %s -mtriple=i686-unknown-unknown -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes X86,FASTSDAG-X86,FAST-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes SSE,FASTSDAG-SSE,FAST-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,FAST-AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel=1 -global-isel=0 -fast-isel-abort=0 | FileCheck %s --check-prefixes AVX,FASTSDAG-AVX,FAST-AVX
+; COMM: GlobalISel can't legalize double stores on 32bit platform due to lack of double/integer distinguish during legalization
+; RUN: llc < %s -mtriple=i686-unknown-unknown -fast-isel=0 -global-isel=1 -global-isel-abort=2 | FileCheck %s --check-prefixes X86,GLOBAL-X86
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel=0 -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes SSE,GLOBAL-SSE
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel=0 -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes AVX,GLOBAL-AVX
+; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f -fast-isel=0 -global-isel=1 -global-isel-abort=1 | FileCheck %s --check-prefixes AVX,GLOBAL-AVX
+
+define double @fpext_float_to_double(float %f) {
+; X86-LABEL: fpext_float_to_double:
+; X86: # %bb.0:
+; X86-NEXT: flds {{[0-9]+}}(%esp)
+; X86-NEXT: retl
+;
+; SSE-LABEL: fpext_float_to_double:
+; SSE: # %bb.0:
+; SSE-NEXT: cvtss2sd %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: fpext_float_to_double:
+; AVX: # %bb.0:
+; AVX-NEXT: vcvtss2sd %xmm0, %xmm0, %xmm0
+; AVX-NEXT: retq
+ %1 = fpext float %f to double
+ ret double %1
+}
+
+define x86_fp80 @fpext_float_to_x86_fp80(float %f) {
+; FASTSDAG-X86-LABEL: fpext_float_to_x86_fp80:
+; FASTSDAG-X86: # %bb.0:
+; FASTSDAG-X86-NEXT: flds {{[0-9]+}}(%esp)
+; FASTSDAG-X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fpext_float_to_x86_fp80:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: movss %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: flds -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fpext_float_to_x86_fp80:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: vmovss %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: flds -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-X86-LABEL: fpext_float_to_x86_fp80:
+; GLOBAL-X86: # %bb.0:
+; GLOBAL-X86-NEXT: pushl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 8
+; GLOBAL-X86-NEXT: movl {{[0-9]+}}(%esp), %eax
+; GLOBAL-X86-NEXT: movl %eax, (%esp)
+; GLOBAL-X86-NEXT: flds (%esp)
+; GLOBAL-X86-NEXT: popl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 4
+; GLOBAL-X86-NEXT: retl
+;
+; GLOBAL-SSE-LABEL: fpext_float_to_x86_fp80:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: movd %xmm0, %eax
+; GLOBAL-SSE-NEXT: movl %eax, -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: flds -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fpext_float_to_x86_fp80:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: vmovd %xmm0, %eax
+; GLOBAL-AVX-NEXT: movl %eax, -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: flds -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: retq
+ %1 = fpext float %f to x86_fp80
+ ret x86_fp80 %1
+}
+
+define x86_fp80 @fpext_double_to_x86_fp80(double %d) {
+; FASTSDAG-X86-LABEL: fpext_double_to_x86_fp80:
+; FASTSDAG-X86: # %bb.0:
+; FASTSDAG-X86-NEXT: fldl {{[0-9]+}}(%esp)
+; FASTSDAG-X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fpext_double_to_x86_fp80:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: movsd %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: fldl -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fpext_double_to_x86_fp80:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: vmovsd %xmm0, -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: fldl -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-X86-LABEL: fpext_double_to_x86_fp80:
+; GLOBAL-X86: # %bb.0:
+; GLOBAL-X86-NEXT: pushl %ebp
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 8
+; GLOBAL-X86-NEXT: .cfi_offset %ebp, -8
+; GLOBAL-X86-NEXT: movl %esp, %ebp
+; GLOBAL-X86-NEXT: .cfi_def_cfa_register %ebp
+; GLOBAL-X86-NEXT: andl $-8, %esp
+; GLOBAL-X86-NEXT: subl $8, %esp
+; GLOBAL-X86-NEXT: leal 8(%ebp), %eax
+; GLOBAL-X86-NEXT: movl 8(%ebp), %ecx
+; GLOBAL-X86-NEXT: movl 4(%eax), %eax
+; GLOBAL-X86-NEXT: movl %esp, %edx
+; GLOBAL-X86-NEXT: movl %ecx, (%esp)
+; GLOBAL-X86-NEXT: movl %eax, 4(%edx)
+; GLOBAL-X86-NEXT: fldl (%esp)
+; GLOBAL-X86-NEXT: movl %ebp, %esp
+; GLOBAL-X86-NEXT: popl %ebp
+; GLOBAL-X86-NEXT: .cfi_def_cfa %esp, 4
+; GLOBAL-X86-NEXT: retl
+;
+; GLOBAL-SSE-LABEL: fpext_double_to_x86_fp80:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: movq %xmm0, %rax
+; GLOBAL-SSE-NEXT: movq %rax, -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: fldl -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fpext_double_to_x86_fp80:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: vmovq %xmm0, %rax
+; GLOBAL-AVX-NEXT: movq %rax, -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: fldl -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: retq
+ %1 = fpext double %d to x86_fp80
+ ret x86_fp80 %1
+}
+
+define float @fptrunc_double_to_float(double %d) {
+; FASTSDAG-X86-LABEL: fptrunc_double_to_float:
+; FASTSDAG-X86: # %bb.0:
+; FASTSDAG-X86-NEXT: pushl %eax
+; FASTSDAG-X86-NEXT: .cfi_def_cfa_offset 8
+; FASTSDAG-X86-NEXT: fldl {{[0-9]+}}(%esp)
+; FASTSDAG-X86-NEXT: fstps (%esp)
+; FASTSDAG-X86-NEXT: flds (%esp)
+; FASTSDAG-X86-NEXT: popl %eax
+; FASTSDAG-X86-NEXT: .cfi_def_cfa_offset 4
+; FASTSDAG-X86-NEXT: retl
+;
+; SSE-LABEL: fptrunc_double_to_float:
+; SSE: # %bb.0:
+; SSE-NEXT: cvtsd2ss %xmm0, %xmm0
+; SSE-NEXT: retq
+;
+; AVX-LABEL: fptrunc_double_to_float:
+; AVX: # %bb.0:
+; AVX-NEXT: vcvtsd2ss %xmm0, %xmm0, %xmm0
+; AVX-NEXT: retq
+;
+; GLOBAL-X86-LABEL: fptrunc_double_to_float:
+; GLOBAL-X86: # %bb.0:
+; GLOBAL-X86-NEXT: pushl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 8
+; GLOBAL-X86-NEXT: leal {{[0-9]+}}(%esp), %eax
+; GLOBAL-X86-NEXT: movl {{[0-9]+}}(%esp), %ecx
+; GLOBAL-X86-NEXT: movl 4(%eax), %eax
+; GLOBAL-X86-NEXT: movl %esp, %edx
+; GLOBAL-X86-NEXT: movl %ecx, (%esp)
+; GLOBAL-X86-NEXT: movl %eax, 4(%edx)
+; GLOBAL-X86-NEXT: flds (%esp)
+; GLOBAL-X86-NEXT: popl %eax
+; GLOBAL-X86-NEXT: .cfi_def_cfa_offset 4
+; GLOBAL-X86-NEXT: retl
+ %1 = fptrunc double %d to float
+ ret float %1
+}
+
+define float @fptrunc_x86_fp80_to_float(x86_fp80 %x) {
+; X86-LABEL: fptrunc_x86_fp80_to_float:
+; X86: # %bb.0:
+; X86-NEXT: pushl %eax
+; X86-NEXT: .cfi_def_cfa_offset 8
+; X86-NEXT: fldt {{[0-9]+}}(%esp)
+; X86-NEXT: fstps (%esp)
+; X86-NEXT: flds (%esp)
+; X86-NEXT: popl %eax
+; X86-NEXT: .cfi_def_cfa_offset 4
+; X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fptrunc_x86_fp80_to_float:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: fstps -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fptrunc_x86_fp80_to_float:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: fstps -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-SSE-LABEL: fptrunc_x86_fp80_to_float:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: fstps -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: movl -{{[0-9]+}}(%rsp), %eax
+; GLOBAL-SSE-NEXT: movd %eax, %xmm0
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fptrunc_x86_fp80_to_float:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: fstps -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: movl -{{[0-9]+}}(%rsp), %eax
+; GLOBAL-AVX-NEXT: vmovd %eax, %xmm0
+; GLOBAL-AVX-NEXT: retq
+ %1 = fptrunc x86_fp80 %x to float
+ ret float %1
+}
+
+define double @fptrunc_x86_fp80_to_double(x86_fp80 %x) {
+; X86-LABEL: fptrunc_x86_fp80_to_double:
+; X86: # %bb.0:
+; X86-NEXT: pushl %ebp
+; X86-NEXT: .cfi_def_cfa_offset 8
+; X86-NEXT: .cfi_offset %ebp, -8
+; X86-NEXT: movl %esp, %ebp
+; X86-NEXT: .cfi_def_cfa_register %ebp
+; X86-NEXT: andl $-8, %esp
+; X86-NEXT: subl $8, %esp
+; X86-NEXT: fldt 8(%ebp)
+; X86-NEXT: fstpl (%esp)
+; X86-NEXT: fldl (%esp)
+; X86-NEXT: movl %ebp, %esp
+; X86-NEXT: popl %ebp
+; X86-NEXT: .cfi_def_cfa %esp, 4
+; X86-NEXT: retl
+;
+; FASTSDAG-SSE-LABEL: fptrunc_x86_fp80_to_double:
+; FASTSDAG-SSE: # %bb.0:
+; FASTSDAG-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; FASTSDAG-SSE-NEXT: movsd {{.*#+}} xmm0 = mem[0],zero
+; FASTSDAG-SSE-NEXT: retq
+;
+; FASTSDAG-AVX-LABEL: fptrunc_x86_fp80_to_double:
+; FASTSDAG-AVX: # %bb.0:
+; FASTSDAG-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; FASTSDAG-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
+; FASTSDAG-AVX-NEXT: retq
+;
+; GLOBAL-SSE-LABEL: fptrunc_x86_fp80_to_double:
+; GLOBAL-SSE: # %bb.0:
+; GLOBAL-SSE-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; GLOBAL-SSE-NEXT: movq -{{[0-9]+}}(%rsp), %rax
+; GLOBAL-SSE-NEXT: movq %rax, %xmm0
+; GLOBAL-SSE-NEXT: retq
+;
+; GLOBAL-AVX-LABEL: fptrunc_x86_fp80_to_double:
+; GLOBAL-AVX: # %bb.0:
+; GLOBAL-AVX-NEXT: fldt {{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: fstpl -{{[0-9]+}}(%rsp)
+; GLOBAL-AVX-NEXT: movq -{{[0-9]+}}(%rsp), %rax
+; GLOBAL-AVX-NEXT: vmovq %rax, %xmm0
+; GLOBAL-AVX-NEXT: retq
+ %1 = fptrunc x86_fp80 %x to double
+ ret double %1
+}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; FAST-AVX: {{.*}}
+; FAST-SSE: {{.*}}
+; FAST-X86: {{.*}}
+; SDAG-AVX: {{.*}}
+; SDAG-SSE: {{.*}}
+; SDAG-X86: {{.*}}
|
arsenm
reviewed
May 27, 2025
There was a problem hiding this comment.
We intentionally omit s32->s64 and s64->s32 variants of stores and loads for X87 because during legalization we can't determine whether it is a floating point store or an integer one.
But you shouldn't need this? The FP-ness should be encoded in the load/store directly. i.e. are we missing FP ext load and store?
|
|
||
| auto *LoadMMO = MF.getMachineMemOperand(PtrInfo, MachineMemOperand::MOLoad, | ||
| StackTy, StackTyAlign); | ||
| MIRBuilder.buildLoad(DstReg, StackTemp, *LoadMMO); |
There was a problem hiding this comment.
Can this be a generic lower action that creates an FP extending load / FP truncating store?
There was a problem hiding this comment.
Yes, it looks quite generic. Let me move it to generic lower actions.
There was a problem hiding this comment.
@arsenm we have a conflicting FP64->FP16 lowering
llvm-project/llvm/lib/CodeGen/GlobalISel/LegalizerHelper.cpp
Lines 8010 to 8019 in 59b7b5b
It doesn't hurt X86 but is it ok for the generic lowering to have FP64->FP16 conversion through scalars but for others using memory?
Yes, we don't have them: IIUC it should be resolved once FP types are added https://discourse.llvm.org/t/rfc-globalisel-adding-fp-type-information-to-llt/83349. After this LLT types will show whether store or load is a float or integer operation. And MMO will show whether it is extending or truncating. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We intentionally omit
s32->s64ands64->s32variants of stores and loads for X87 because during legalization we can't determine whether it is a floating point store or an integer one.