[llvm] [DAG] Make strictfp attribute only restricts for libm and make non-math optimizations possible (PR #165464)

[zhijian lin via llvm-commits]

https://github.com/diggerlin updated https://github.com/llvm/llvm-project/pull/165464

>From a40b5e1546f5330372ff1907dc81101a96d23284 Mon Sep 17 00:00:00 2001
From: zhijian <zhijian at ca.ibm.com>
Date: Tue, 28 Oct 2025 19:12:38 +0000
Subject: [PATCH 1/2] let strictfp only work for libm

---
 .../CodeGen/SelectionDAG/SelectionDAGBuilder.cpp  | 15 +++++++++------
 llvm/test/CodeGen/PowerPC/milicode32.ll           |  3 ++-
 2 files changed, 11 insertions(+), 7 deletions(-)

diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index a52265055c88a..37108aa3a56d8 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -9392,7 +9392,9 @@ bool SelectionDAGBuilder::visitStrNLenCall(const CallInst &I) {
 bool SelectionDAGBuilder::visitUnaryFloatCall(const CallInst &I,
                                               unsigned Opcode) {
   // We already checked this call's prototype; verify it doesn't modify errno.
-  if (!I.onlyReadsMemory())
+  // Do not perform optimizations for call sites that require strict
+  // floating-point semantics.
+  if (!I.onlyReadsMemory() || I.isStrictFP())
     return false;
 
   SDNodeFlags Flags;
@@ -9412,7 +9414,9 @@ bool SelectionDAGBuilder::visitUnaryFloatCall(const CallInst &I,
 bool SelectionDAGBuilder::visitBinaryFloatCall(const CallInst &I,
                                                unsigned Opcode) {
   // We already checked this call's prototype; verify it doesn't modify errno.
-  if (!I.onlyReadsMemory())
+  // Do not perform optimizations for call sites that require strict
+  // floating-point semantics.
+  if (!I.onlyReadsMemory() || I.isStrictFP())
     return false;
 
   SDNodeFlags Flags;
@@ -9445,11 +9449,10 @@ void SelectionDAGBuilder::visitCall(const CallInst &I) {
 
     // Check for well-known libc/libm calls.  If the function is internal, it
     // can't be a library call.  Don't do the check if marked as nobuiltin for
-    // some reason or the call site requires strict floating point semantics.
+    // some reason.
     LibFunc Func;
-    if (!I.isNoBuiltin() && !I.isStrictFP() && !F->hasLocalLinkage() &&
-        F->hasName() && LibInfo->getLibFunc(*F, Func) &&
-        LibInfo->hasOptimizedCodeGen(Func)) {
+    if (!I.isNoBuiltin() && !F->hasLocalLinkage() && F->hasName() &&
+        LibInfo->getLibFunc(*F, Func) && LibInfo->hasOptimizedCodeGen(Func)) {
       switch (Func) {
       default: break;
       case LibFunc_bcmp:
diff --git a/llvm/test/CodeGen/PowerPC/milicode32.ll b/llvm/test/CodeGen/PowerPC/milicode32.ll
index 78d036202fe4e..09239fb9d1b09 100644
--- a/llvm/test/CodeGen/PowerPC/milicode32.ll
+++ b/llvm/test/CodeGen/PowerPC/milicode32.ll
@@ -64,8 +64,9 @@ entry:
   %str.addr = alloca ptr, align 4
   store ptr %str, ptr %str.addr, align 4
   %0 = load ptr, ptr %str.addr, align 4
-  %call = call i32 @strlen(ptr noundef %0)
+  %call = call i32 @strlen(ptr noundef %0) #0
   ret i32 %call
 }
 
 declare i32 @strlen(ptr noundef) nounwind
+attributes #0 = { strictfp }

>From 44c7bad497286f0fb7ecd2f1be2215afba5fb3e7 Mon Sep 17 00:00:00 2001
From: zhijian <zhijian at ca.ibm.com>
Date: Tue, 4 Nov 2025 14:55:02 +0000
Subject: [PATCH 2/2] add comment to add test caee for fp-strict

---
 llvm/test/CodeGen/PowerPC/milicode32.ll | 33 +++++++++++++++++++++++++
 1 file changed, 33 insertions(+)

diff --git a/llvm/test/CodeGen/PowerPC/milicode32.ll b/llvm/test/CodeGen/PowerPC/milicode32.ll
index 09239fb9d1b09..c81b518ef8dbf 100644
--- a/llvm/test/CodeGen/PowerPC/milicode32.ll
+++ b/llvm/test/CodeGen/PowerPC/milicode32.ll
@@ -60,6 +60,39 @@ define i32 @strlen_test(ptr noundef %str) nounwind {
 ; CHECK-LINUX32-P9-NEXT:    addi r1, r1, 16
 ; CHECK-LINUX32-P9-NEXT:    mtlr r0
 ; CHECK-LINUX32-P9-NEXT:    blr
+entry:
+  %str.addr = alloca ptr, align 4
+  store ptr %str, ptr %str.addr, align 4
+  %0 = load ptr, ptr %str.addr, align 4
+  %call = call i32 @strlen(ptr noundef %0)
+  ret i32 %call
+}
+
+define i32 @strlen_test_fp_strict(ptr noundef %str) nounwind {
+; CHECK-AIX-32-P9-LABEL: strlen_test_fp_strict:
+; CHECK-AIX-32-P9:       # %bb.0: # %entry
+; CHECK-AIX-32-P9-NEXT:    mflr r0
+; CHECK-AIX-32-P9-NEXT:    stwu r1, -64(r1)
+; CHECK-AIX-32-P9-NEXT:    stw r0, 72(r1)
+; CHECK-AIX-32-P9-NEXT:    stw r3, 60(r1)
+; CHECK-AIX-32-P9-NEXT:    bl .___strlen[PR]
+; CHECK-AIX-32-P9-NEXT:    nop
+; CHECK-AIX-32-P9-NEXT:    addi r1, r1, 64
+; CHECK-AIX-32-P9-NEXT:    lwz r0, 8(r1)
+; CHECK-AIX-32-P9-NEXT:    mtlr r0
+; CHECK-AIX-32-P9-NEXT:    blr
+;
+; CHECK-LINUX32-P9-LABEL: strlen_test_fp_strict:
+; CHECK-LINUX32-P9:       # %bb.0: # %entry
+; CHECK-LINUX32-P9-NEXT:    mflr r0
+; CHECK-LINUX32-P9-NEXT:    stwu r1, -16(r1)
+; CHECK-LINUX32-P9-NEXT:    stw r0, 20(r1)
+; CHECK-LINUX32-P9-NEXT:    stw r3, 12(r1)
+; CHECK-LINUX32-P9-NEXT:    bl strlen
+; CHECK-LINUX32-P9-NEXT:    lwz r0, 20(r1)
+; CHECK-LINUX32-P9-NEXT:    addi r1, r1, 16
+; CHECK-LINUX32-P9-NEXT:    mtlr r0
+; CHECK-LINUX32-P9-NEXT:    blr
 entry:
   %str.addr = alloca ptr, align 4
   store ptr %str, ptr %str.addr, align 4