AMDGPU: Don't try to use LDS/vector for private if pointer value stored

[arsenm]

Summary

If the pointer is the store's value operand, this would produce
a broken module. Make sure the use is actually for the pointer operand.

[arsenm]

Simplify tests

[arsenm]

r243462

lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp

static Value* GEPToVectorIndex(GetElementPtrInst *GEP) {
 
   return GEP->getOperand(2);
 }
 
 // Not an instruction handled below to turn into a vector.
 //
 // TODO: Check isTriviallyVectorizable for calls and handle other
 // instructions.
-static bool canVectorizeInst(Instruction *Inst) {
+static bool canVectorizeInst(Instruction *Inst, User *User) {
   switch (Inst->getOpcode()) {
   case Instruction::Load:
-  case Instruction::Store:
   case Instruction::BitCast:
   case Instruction::AddrSpaceCast:
     return true;
+  case Instruction::Store: {
+    // Must be the stored pointer operand, not a stored value.
+    StoreInst *SI = cast<StoreInst>(Inst);
+    return SI->getPointerOperand() == User;
+  }
   default:
     return false;
   }
 }
 
 static bool tryPromoteAllocaToVector(AllocaInst *Alloca) {
   Type *AllocaTy = Alloca->getAllocatedType();
 
if (!AllocaTy->isArrayTy() ||
     return false;
   }
 
   std::map<GetElementPtrInst*, Value*> GEPVectorIdx;
   std::vector<Value*> WorkList;
   for (User *AllocaUser : Alloca->users()) {
     GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(AllocaUser);
     if (!GEP) {
-      if (!canVectorizeInst(cast<Instruction>(AllocaUser)))
+      if (!canVectorizeInst(cast<Instruction>(AllocaUser), Alloca))
         return false;
 
       WorkList.push_back(AllocaUser);
       continue;
     }
 
     Value *Index = GEPToVectorIndex(GEP);
 
     // If we can't compute a vector index from this GEP, then we can't
     // promote this alloca to vector.
     if (!Index) {
       DEBUG(dbgs() << "  Cannot compute vector index for GEP " << *GEP << '\n');
       return false;
     }
 
     GEPVectorIdx[GEP] = Index;
     for (User *GEPUser : AllocaUser->users()) {
-      if (!canVectorizeInst(cast<Instruction>(GEPUser)))
+      if (!canVectorizeInst(cast<Instruction>(GEPUser), AllocaUser))
         return false;
 
       WorkList.push_back(GEPUser);
     }
   }
 
   VectorType *VectorTy = arrayTypeToVecType(AllocaTy);
 
if (CallInst *CI = dyn_cast<CallInst>(User)) {
       continue;
     }
 
     // FIXME: Correctly handle ptrtoint instructions.
     Instruction *UseInst = dyn_cast<Instruction>(User);
     if (UseInst && UseInst->getOpcode() == Instruction::PtrToInt)
       return false;
 
+    if (StoreInst *SI = dyn_cast_or_null<StoreInst>(UseInst)) {
+      // Reject if the stored value is not the pointer operand.
+      if (SI->getPointerOperand() != User)
+        return false;
+    }
+
     if (!User->getType()->isPointerTy())
       continue;
 
     WorkList.push_back(User);
 
     Success &= collectUsesWithPtrTypes(User, WorkList);
   }
   return Success;

test/CodeGen/AMDGPU/promote-alloca-stored-pointer-value.ll

+; RUN: llc -march=amdgcn < %s | FileCheck -check-prefix=GCN %s
+
+; Pointer value is stored in a candidate for LDS usage.
+
+; GCN-LABEL: {{^}}stored_lds_pointer_value:
+; GCN: buffer_store_dword v
+define void @stored_lds_pointer_value(float* addrspace(1)* %ptr) #0 {
+  %tmp = alloca float
+  store float 0.0, float *%tmp
+  store float* %tmp, float* addrspace(1)* %ptr
+  ret void
+}
+
+; GCN-LABEL: {{^}}stored_lds_pointer_value_gep:
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
+; GCN: buffer_store_dword v
+; GCN: buffer_store_dword v
+define void @stored_lds_pointer_value_gep(float* addrspace(1)* %ptr, i32 %idx) #0 {
+bb:
+  %tmp = alloca float, i32 16
+  store float 0.0, float* %tmp
+  %tmp2 = getelementptr inbounds float, float* %tmp, i32 %idx
+  store float* %tmp2, float* addrspace(1)* %ptr
+  ret void
+}
+
+; Pointer value is stored in a candidate for vector usage
+; GCN-LABEL: {{^}}stored_vector_pointer_value:
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD0
+; GCN-DAG: s_mov_b32 s{{[0-9]+}}, SCRATCH_RSRC_DWORD1
+; GCN: buffer_store_dword
+; GCN: buffer_store_dword
+; GCN: buffer_store_dword
+; GCN: buffer_store_dword
+define void @stored_vector_pointer_value(i32* addrspace(1)* %out, i32 %index) {
+entry:
+  %tmp0 = alloca [4 x i32]
+  %x = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 0
+  %y = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 1
+  %z = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 2
+  %w = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 3
+  store i32 0, i32* %x
+  store i32 1, i32* %y
+  store i32 2, i32* %z
+  store i32 3, i32* %w
+  %tmp1 = getelementptr [4 x i32], [4 x i32]* %tmp0, i32 0, i32 %index
+  store i32* %tmp1, i32* addrspace(1)* %out
+  ret void
+}
+
+attributes #0 = { nounwind }