[llvm] c9821ce - [AMDGPU] Mark sin/cos load folding as modifying the function.

[Stanislav Mekhanoshin via llvm-commits]

Author: Stanislav Mekhanoshin
Date: 2020-11-13T14:49:33-08:00
New Revision: c9821cec7492f9ffb4586ece4a82fda1e5fe8e1e

URL: https://github.com/llvm/llvm-project/commit/c9821cec7492f9ffb4586ece4a82fda1e5fe8e1e
DIFF: https://github.com/llvm/llvm-project/commit/c9821cec7492f9ffb4586ece4a82fda1e5fe8e1e.diff

LOG: [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

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp b/llvm/lib/Target/AMDGPU/AMDGPULibCalls.cpp
index 538a22df514f..a72f16956e72 100644
--- 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_");