[FPEnv][AMDGPU] Correct AMDGPUSimplifyLibCalls handling of strictfp attribute.

[kpneal]

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.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Kevin P. Neal (kpneal)

Changes

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.

---

Full diff: https://github.com/llvm/llvm-project/pull/86705.diff

3 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp (+2)
• (modified) llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-pown.ll (+6-6)
• (modified) llvm/test/CodeGen/AMDGPU/amdgpu-simplify-libcall-rootn.ll (+1-1)

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:

[arsenm]

this is still emitting an unconstrained log2

[arsenm]

unconstrained exp2

[kpneal]

The way to fix those would be to add code to the IRBuilder in CreateCall to check for strictfp mode and switch over to the constrained call. That code hasn't been written yet. Is that something I need to do before I can move forward here?