[AMDGPU] Mark sin/cos load folding as modifying the function.

When the load value is folded into the sin/cos operation, the
AMDGPU library calls simplifier could still mark the function
as unmodified. Instead ensure if there is an early return,
return whether the load was folded into the sin/cos call.

Authored by MJDSys

Differential Revision: https://reviews.llvm.org/D91401

diff --git a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
index 538a22d..a72f169 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
@@ -1288,6 +1288,7 @@ bool AMDGPULibCalls::fold_sincos(CallInst *CI, IRBuilder<> &B,
   BasicBlock * const CBB = CI->getParent();
 
   int const MaxScan = 30;
+  bool Changed = false;
 
   { // fold in load value.
     LoadInst *LI = dyn_cast<LoadInst>(CArgVal);
@@ -1295,6 +1296,7 @@ bool AMDGPULibCalls::fold_sincos(CallInst *CI, IRBuilder<> &B,
       BasicBlock::iterator BBI = LI->getIterator();
       Value *AvailableVal = FindAvailableLoadedValue(LI, CBB, BBI, MaxScan, AA);
       if (AvailableVal) {
+        Changed = true;
         CArgVal->replaceAllUsesWith(AvailableVal);
         if (CArgVal->getNumUses() == 0)
           LI->eraseFromParent();
@@ -1330,7 +1332,8 @@ bool AMDGPULibCalls::fold_sincos(CallInst *CI, IRBuilder<> &B,
     if (UI) break;
   }
 
-  if (!UI) return false;
+  if (!UI)
+    return Changed;
 
   // Merge the sin and cos.
 
@@ -1339,7 +1342,8 @@ bool AMDGPULibCalls::fold_sincos(CallInst *CI, IRBuilder<> &B,
   AMDGPULibFunc nf(AMDGPULibFunc::EI_SINCOS, fInfo);
   nf.getLeads()[0].PtrKind = AMDGPULibFunc::getEPtrKindFromAddrSpace(AMDGPUAS::FLAT_ADDRESS);
   FunctionCallee Fsincos = getFunction(M, nf);
-  if (!Fsincos) return false;
+  if (!Fsincos)
+    return Changed;
 
   BasicBlock::iterator ItOld = B.GetInsertPoint();
   AllocaInst *Alloc = insertAlloca(UI, B, "__sincos_");