[llvm] [FPEnv][AMDGPU] Correct AMDGPUSimplifyLibCalls handling of strictfp attribute. (PR #86705)

[Kevin P. Neal via llvm-commits]

https://github.com/kpneal created https://github.com/llvm/llvm-project/pull/86705

The AMDGPUSimplifyLibCalls pass was lowering function calls with the strictfp attribute to sequences that included function calls incorrectly lacking the attribute. This patch corrects that.

The pass now also emits the correct constrained fp call instead of normal FP instructions when in a function with the strictfp attribute.

>From fb50c36f08bf632d768f00586de7b71da4e13449 Mon Sep 17 00:00:00 2001
From: "Kevin P. Neal" <kevin.neal at sas.com>
Date: Tue, 26 Mar 2024 13:46:17 -0400
Subject: [PATCH] [FPEnv][AMDGPU] Correct AMDGPUSimplifyLibCalls handling of
 strictfp attribute.

The AMDGPUSimplifyLibCalls pass was lowering function calls with
the strictfp attribute to sequences that included function calls
incorrectly lacking the attribute. This patch corrects that.

The pass now also emits the correct constrained fp call instead of
normal FP instructions when in a function with the strictfp attribute.
---
 llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp            |  2 ++
 .../CodeGen/AMDGPU/amdgpu-simplify-libcall-pown.ll   | 12 ++++++------
 .../CodeGen/AMDGPU/amdgpu-simplify-libcall-rootn.ll  |  2 +-
 3 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
index 84b4ccc1ae7ba7..5aa35becd842c3 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
@@ -657,6 +657,8 @@ bool AMDGPULibCalls::fold(CallInst *CI) {
     return true;
 
   IRBuilder<> B(CI);
+  if (CI->isStrictFP())
+    B.setIsFPConstrained(true);
 
   if (FPMathOperator *FPOp = dyn_cast<FPMathOperator>(CI)) {
     // Under unsafe-math, evaluate calls if possible.
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pown.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pown.ll
index 942f459ea6b8ca..8ddaf243db92c4 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pown.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pown.ll
@@ -808,7 +808,7 @@ define float @test_pown_fast_f32_nobuiltin(float %x, i32 %y) {
 ; CHECK-LABEL: define float @test_pown_fast_f32_nobuiltin
 ; CHECK-SAME: (float [[X:%.*]], i32 [[Y:%.*]]) {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[CALL:%.*]] = tail call fast float @_Z4pownfi(float [[X]], i32 [[Y]]) #[[ATTR3:[0-9]+]]
+; CHECK-NEXT:    [[CALL:%.*]] = tail call fast float @_Z4pownfi(float [[X]], i32 [[Y]]) #[[ATTR4:[0-9]+]]
 ; CHECK-NEXT:    ret float [[CALL]]
 ;
 entry:
@@ -820,11 +820,11 @@ define float @test_pown_fast_f32_strictfp(float %x, i32 %y) #1 {
 ; CHECK-LABEL: define float @test_pown_fast_f32_strictfp
 ; CHECK-SAME: (float [[X:%.*]], i32 [[Y:%.*]]) #[[ATTR0:[0-9]+]] {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[__FABS:%.*]] = call fast float @llvm.fabs.f32(float [[X]])
-; CHECK-NEXT:    [[__LOG2:%.*]] = call fast float @llvm.log2.f32(float [[__FABS]])
-; CHECK-NEXT:    [[POWNI2F:%.*]] = sitofp i32 [[Y]] to float
-; CHECK-NEXT:    [[__YLOGX:%.*]] = fmul fast float [[__LOG2]], [[POWNI2F]]
-; CHECK-NEXT:    [[__EXP2:%.*]] = call fast float @llvm.exp2.f32(float [[__YLOGX]])
+; CHECK-NEXT:    [[__FABS:%.*]] = call fast float @llvm.fabs.f32(float [[X]]) #[[ATTR0]]
+; CHECK-NEXT:    [[__LOG2:%.*]] = call fast float @llvm.log2.f32(float [[__FABS]]) #[[ATTR0]]
+; CHECK-NEXT:    [[POWNI2F:%.*]] = call fast float @llvm.experimental.constrained.sitofp.f32.i32(i32 [[Y]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR0]]
+; CHECK-NEXT:    [[__YLOGX:%.*]] = call fast float @llvm.experimental.constrained.fmul.f32(float [[POWNI2F]], float [[__LOG2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR0]]
+; CHECK-NEXT:    [[__EXP2:%.*]] = call fast float @llvm.exp2.f32(float [[__YLOGX]]) #[[ATTR0]]
 ; CHECK-NEXT:    [[__YEVEN:%.*]] = shl i32 [[Y]], 31
 ; CHECK-NEXT:    [[TMP0:%.*]] = bitcast float [[X]] to i32
 ; CHECK-NEXT:    [[__POW_SIGN:%.*]] = and i32 [[__YEVEN]], [[TMP0]]
diff --git a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-rootn.ll b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-rootn.ll
index 2ffa647d1869a5..2e64a3456c2427 100644
--- a/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-rootn.ll
+++ b/llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-rootn.ll
@@ -896,7 +896,7 @@ define float @test_rootn_f32__y_neg2__strictfp(float %x) #1 {
 ; CHECK-LABEL: define float @test_rootn_f32__y_neg2__strictfp(
 ; CHECK-SAME: float [[X:%.*]]) #[[ATTR0]] {
 ; CHECK-NEXT:  entry:
-; CHECK-NEXT:    [[__ROOTN2RSQRT:%.*]] = call float @_Z5rsqrtf(float [[X]])
+; CHECK-NEXT:    [[__ROOTN2RSQRT:%.*]] = call float @_Z5rsqrtf(float [[X]]) #[[ATTR0]]
 ; CHECK-NEXT:    ret float [[__ROOTN2RSQRT]]
 ;
 entry: