[llvm] [X86] Remove LowerFCanonicalize and use generic expansion (PR #147730)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Jul 9 06:48:19 PDT 2025
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-selectiondag
@llvm/pr-subscribers-backend-x86
Author: woruyu (woruyu)
<details>
<summary>Changes</summary>
Summary
This PR resolves https://github.com/llvm/llvm-project/issues/143862
Remove LowerFCanonicalizet for x86 to use generic expansion
---
Full diff: https://github.com/llvm/llvm-project/pull/147730.diff
2 Files Affected:
- (modified) llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp (+7)
- (modified) llvm/lib/Target/X86/X86ISelLowering.cpp (+13-16)
``````````diff
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
index f908a66128ec8..f933295d3626c 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp
@@ -41,6 +41,7 @@
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Target/TargetMachine.h"
#include <cassert>
#include <cstdint>
#include <iterator>
@@ -1309,6 +1310,12 @@ void VectorLegalizer::Expand(SDNode *Node, SmallVectorImpl<SDValue> &Results) {
return;
}
break;
+ case ISD::FCANONICALIZE: {
+ const Triple &TT = DAG.getTarget().getTargetTriple();
+ if (TT.isX86()) {
+ return;
+ }
+ }
}
SDValue Unrolled = DAG.UnrollVectorOp(Node);
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 347ba1262b66b..506781694cee5 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -316,8 +316,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::FP_TO_UINT_SAT, VT, Custom);
setOperationAction(ISD::FP_TO_SINT_SAT, VT, Custom);
}
- setOperationAction(ISD::FCANONICALIZE, MVT::f32, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::f64, Custom);
if (Subtarget.is64Bit()) {
setOperationAction(ISD::FP_TO_UINT_SAT, MVT::i64, Custom);
setOperationAction(ISD::FP_TO_SINT_SAT, MVT::i64, Custom);
@@ -348,9 +346,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
// TODO: when we have SSE, these could be more efficient, by using movd/movq.
if (!Subtarget.hasSSE2()) {
setOperationAction(ISD::BITCAST , MVT::f32 , Expand);
- setOperationAction(ISD::BITCAST , MVT::i32 , Expand);
- setOperationAction(ISD::FCANONICALIZE, MVT::f32, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::f64, Custom);
+ setOperationAction(ISD::BITCAST, MVT::i32, Expand);
if (Subtarget.is64Bit()) {
setOperationAction(ISD::BITCAST , MVT::f64 , Expand);
// Without SSE, i64->f64 goes through memory.
@@ -716,7 +712,6 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::STRICT_FROUNDEVEN, MVT::f16, Promote);
setOperationAction(ISD::STRICT_FTRUNC, MVT::f16, Promote);
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f16, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::f16, Custom);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f32, Custom);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f64, Custom);
setOperationAction(ISD::LRINT, MVT::f16, Expand);
@@ -871,7 +866,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::STRICT_FMUL , MVT::f80, Legal);
setOperationAction(ISD::STRICT_FDIV , MVT::f80, Legal);
setOperationAction(ISD::STRICT_FSQRT , MVT::f80, Legal);
- setOperationAction(ISD::FCANONICALIZE , MVT::f80, Custom);
+ setOperationAction(ISD::FCANONICALIZE, MVT::f80, Expand);
if (isTypeLegal(MVT::f16)) {
setOperationAction(ISD::FP_EXTEND, MVT::f80, Custom);
setOperationAction(ISD::STRICT_FP_EXTEND, MVT::f80, Custom);
@@ -934,7 +929,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
if (isTypeLegal(MVT::f80)) {
setOperationAction(ISD::FP_ROUND, MVT::f80, Custom);
setOperationAction(ISD::STRICT_FP_ROUND, MVT::f80, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::f80, Custom);
+ setOperationAction(ISD::FCANONICALIZE, MVT::f80, Expand);
}
setOperationAction(ISD::SETCC, MVT::f128, Custom);
@@ -1070,11 +1065,11 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::VSELECT, MVT::v4f32, Custom);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4f32, Custom);
setOperationAction(ISD::SELECT, MVT::v4f32, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::v4f32, Custom);
+ setOperationAction(ISD::FCANONICALIZE, MVT::v4f32, Expand);
setOperationAction(ISD::LOAD, MVT::v2f32, Custom);
setOperationAction(ISD::STORE, MVT::v2f32, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::v2f32, Custom);
+ setOperationAction(ISD::FCANONICALIZE, MVT::v2f32, Expand);
setOperationAction(ISD::STRICT_FADD, MVT::v4f32, Legal);
setOperationAction(ISD::STRICT_FSUB, MVT::v4f32, Legal);
@@ -1137,7 +1132,8 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::UMULO, MVT::v2i32, Custom);
setOperationAction(ISD::FNEG, MVT::v2f64, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::v2f64, Custom);
+ // woruyu
+ setOperationAction(ISD::FCANONICALIZE, MVT::v2f64, Expand);
setOperationAction(ISD::FABS, MVT::v2f64, Custom);
setOperationAction(ISD::FCOPYSIGN, MVT::v2f64, Custom);
@@ -1473,7 +1469,8 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::FMINIMUM, VT, Custom);
setOperationAction(ISD::FMAXIMUMNUM, VT, Custom);
setOperationAction(ISD::FMINIMUMNUM, VT, Custom);
- setOperationAction(ISD::FCANONICALIZE, VT, Custom);
+ // woruyu
+ setOperationAction(ISD::FCANONICALIZE, VT, Expand);
}
setOperationAction(ISD::LRINT, MVT::v8f32, Custom);
@@ -1741,9 +1738,9 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::FP_TO_UINT, MVT::v2i1, Custom);
setOperationAction(ISD::STRICT_FP_TO_SINT, MVT::v2i1, Custom);
setOperationAction(ISD::STRICT_FP_TO_UINT, MVT::v2i1, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::v8f16, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::v16f16, Custom);
- setOperationAction(ISD::FCANONICALIZE, MVT::v32f16, Custom);
+ setOperationAction(ISD::FCANONICALIZE, MVT::v8f16, Expand);
+ setOperationAction(ISD::FCANONICALIZE, MVT::v16f16, Expand);
+ setOperationAction(ISD::FCANONICALIZE, MVT::v32f16, Expand);
// There is no byte sized k-register load or store without AVX512DQ.
if (!Subtarget.hasDQI()) {
@@ -1825,7 +1822,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
setOperationAction(ISD::FMA, VT, Legal);
setOperationAction(ISD::STRICT_FMA, VT, Legal);
setOperationAction(ISD::FCOPYSIGN, VT, Custom);
- setOperationAction(ISD::FCANONICALIZE, VT, Custom);
+ setOperationAction(ISD::FCANONICALIZE, VT, Expand);
}
setOperationAction(ISD::LRINT, MVT::v16f32,
Subtarget.hasDQI() ? Legal : Custom);
``````````
</details>
https://github.com/llvm/llvm-project/pull/147730
More information about the llvm-commits
mailing list