[InstCombine] Gracefully handle an alloca outside the alloca-AS

[jdoerfert]

Summary

While we might eventually want to disallow allocas that do not have the
alloca-AS set, it seems undesirable to crash on them. Add a cast when
required so that we can support such allocas (at least here).

[ronlieb]

thx for pushing this up. giving it a whirl

[jdoerfert]

Make test more meaningful, insert the cast at the right position

[jdoerfert]

In D104866#2838942, @ronlieb wrote:

thx for pushing this up. giving it a whirl

We'll actually going to create the allocas in heap-2-stack in the right address space later but this is an independent problem we need to fix. The original test exposed another one too, should all be better by tomorrow.

[ronlieb]

gets my unblocked, thx

[ronlieb]

Hi Johannes, can this patch land?
i hit an instance of the issue it fixes, while compiling (attempting to) picongpu using trunk clang for amdgpu.

llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp

if (C->getValue().getActiveBits() <= 64) {
         ++It;
 
       // Now that I is pointing to the first non-allocation-inst in the block,
       // insert our getelementptr instruction...
       //
       Type *IdxTy = IC.getDataLayout().getIntPtrType(AI.getType());
       Value *NullIdx = Constant::getNullValue(IdxTy);
       Value *Idx[2] = {NullIdx, NullIdx};
-      Instruction *GEP = GetElementPtrInst::CreateInBounds(
+      Instruction *NewI = GetElementPtrInst::CreateInBounds(
           NewTy, New, Idx, New->getName() + ".sub");
-      IC.InsertNewInstBefore(GEP, *It);
+      IC.InsertNewInstBefore(NewI, *It);
+
+      // Gracefully handle allocas in other address spaces.
+      if (AI.getType()->getPointerAddressSpace() !=
+          NewI->getType()->getPointerAddressSpace()) {
+        NewI =
+            CastInst::CreatePointerBitCastOrAddrSpaceCast(NewI, AI.getType());
+        IC.InsertNewInstBefore(NewI, *It);
+      }
 
       // Now make everything use the getelementptr instead of the original
       // allocation.
-      return IC.replaceInstUsesWith(AI, GEP);
+      return IC.replaceInstUsesWith(AI, NewI);
     }
   }
 
   if (isa<UndefValue>(AI.getArraySize()))
     return IC.replaceInstUsesWith(AI, Constant::getNullValue(AI.getType()));
 
   // Ensure that the alloca array size argument has type intptr_t, so that
   // any casting is exposed early.

llvm/test/Transforms/InstCombine/alloca-in-non-alloca-as.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+; Gracefully handle the alloca that is not in the alloca AS (=5)
+
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32-p6:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-v96:128-v192:256-v256:256-v512:512-v1024:1024-v2048:2048-n32:64-S32-A5-G1-ni:7"
+target triple = "amdgcn-amd-amdhsa"
+
+declare void @use(i8*, i32**)
+
+define weak amdgpu_kernel void @__omp_offloading_802_ea0109_main_l8(i32* %a) {
+; CHECK-LABEL: @__omp_offloading_802_ea0109_main_l8(
+; CHECK-NEXT:  .master:
+; CHECK-NEXT:    [[TMP0:%.*]] = alloca i32*, align 1, addrspace(5)
+; CHECK-NEXT:    [[DOTSUB:%.*]] = bitcast i32* addrspace(5)* [[TMP0]] to i8 addrspace(5)*
+; CHECK-NEXT:    [[TMP1:%.*]] = addrspacecast i8 addrspace(5)* [[DOTSUB]] to i8*
+; CHECK-NEXT:    [[A_ON_STACK:%.*]] = addrspacecast i32* addrspace(5)* [[TMP0]] to i32**
+; CHECK-NEXT:    call void @use(i8* [[TMP1]], i32** [[A_ON_STACK]])
+; CHECK-NEXT:    ret void
+;
+.master:
+  %0 = alloca i8, i64 8, align 1
+  %a_on_stack = bitcast i8* %0 to i32**
+  store i32* undef, i32** %a_on_stack, align 8
+  call void @use(i8* %0, i32** %a_on_stack)
+  ret void
+}