[llvm] AMDGPU: Disable sincos fold for constant inputs (PR #134579)

[Matt Arsenault via llvm-commits]

https://github.com/arsenm created https://github.com/llvm/llvm-project/pull/134579

The use list for ConstantData should not be used. We could do better
by just constant folding here.

>From 0fa970d709b06c689ee4bd1d988c65008f252ef4 Mon Sep 17 00:00:00 2001
From: Matt Arsenault <Matthew.Arsenault at amd.com>
Date: Mon, 7 Apr 2025 08:59:54 +0700
Subject: [PATCH] AMDGPU: Disable sincos fold for constant inputs

The use list for ConstantData should not be used. We could do better
by just constant folding here.
---
 llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp             |  5 +++++
 .../AMDGPU/amdgpu-simplify-libcall-sincos.defined.ll  |  8 +++-----
 .../CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll  | 11 +++++------
 3 files changed, 13 insertions(+), 11 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
index cf8b416d23e50..b65b4d67b3f8c 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
@@ -1361,6 +1361,11 @@ bool AMDGPULibCalls::fold_sincos(FPMathOperator *FPOp, IRBuilder<> &B,
   bool const isSin = fInfo.getId() == AMDGPULibFunc::EI_SIN;
 
   Value *CArgVal = FPOp->getOperand(0);
+
+  // TODO: Constant fold the call
+  if (isa<ConstantData>(CArgVal))
+    return false;
+
   CallInst *CI = cast<CallInst>(FPOp);
 
   Function *F = B.GetInsertBlock()->getParent();
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.defined.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.defined.ll
index ad7e913e6d22c..c8f45fe11390c 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.defined.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.defined.ll
@@ -125,12 +125,10 @@ define void @sincos_f32_value_is_same_constantfp(ptr addrspace(1) nocapture writ
 ; CHECK-LABEL: define void @sincos_f32_value_is_same_constantfp
 ; CHECK-SAME: (ptr addrspace(1) writeonly captures(none) [[SIN_OUT:%.*]], ptr addrspace(1) writeonly captures(none) [[COS_OUT:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT:    [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float 4.200000e+01, ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT:    store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT:    [[CALL:%.*]] = tail call contract float @_Z3sinf(float 4.200000e+01)
+; CHECK-NEXT:    store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
 ; CHECK-NEXT:    [[CALL1:%.*]] = tail call contract float @_Z3cosf(float 4.200000e+01)
-; CHECK-NEXT:    store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT:    store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
 ; CHECK-NEXT:    ret void
 ;
 entry:
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll
index da5686972a86b..03b7d1646eb86 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-sincos.ll
@@ -745,13 +745,12 @@ entry:
 
 define void @sincos_f32_value_is_same_constantfp(ptr addrspace(1) nocapture writeonly %sin_out, ptr addrspace(1) nocapture writeonly %cos_out) {
 ; CHECK-LABEL: define void @sincos_f32_value_is_same_constantfp
-; CHECK-SAME: (ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[SIN_OUT:%.*]], ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR3]] {
+; CHECK-SAME: (ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[SIN_OUT:%.*]], ptr addrspace(1) writeonly captures(none) initializes((0, 4)) [[COS_OUT:%.*]]) local_unnamed_addr #[[ATTR2]] {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[__SINCOS_:%.*]] = alloca float, align 4, addrspace(5)
-; CHECK-NEXT:    [[TMP0:%.*]] = call contract float @_Z6sincosfPU3AS5f(float 4.200000e+01, ptr addrspace(5) [[__SINCOS_]])
-; CHECK-NEXT:    [[TMP1:%.*]] = load float, ptr addrspace(5) [[__SINCOS_]], align 4
-; CHECK-NEXT:    store float [[TMP0]], ptr addrspace(1) [[SIN_OUT]], align 4
-; CHECK-NEXT:    store float [[TMP1]], ptr addrspace(1) [[COS_OUT]], align 4
+; CHECK-NEXT:    [[CALL:%.*]] = tail call contract float @_Z3sinf(float 4.200000e+01)
+; CHECK-NEXT:    store float [[CALL]], ptr addrspace(1) [[SIN_OUT]], align 4
+; CHECK-NEXT:    [[CALL1:%.*]] = tail call contract float @_Z3cosf(float 4.200000e+01)
+; CHECK-NEXT:    store float [[CALL1]], ptr addrspace(1) [[COS_OUT]], align 4
 ; CHECK-NEXT:    ret void
 ;
 entry: