[AMDGPU] Fix for vectorizer crash with pointers of different size

[rampitec]

Summary

When vectorizer strips pointers it can eventually end up with
pointers of two different sizes, then SCEV will crash.

[arsenm]

Was this broken by r365723? Should this use stripPointerCastsSameRepresentation instead?

[arsenm]

lib/Transforms/Vectorize/LoadStoreVectorizer.cpp

347–349 ↗

(On Diff #212455)

Checking that PtrA and PtrB's address space are the same would be simpler

test/CodeGen/AMDGPU/vect-ptr-ptr-size-mismatch.ll

1 ↗

(On Diff #212455)

Test should just run opt and go in test/Transforms/LoadStoreVectorizer

3 ↗

(On Diff #212455)

Should be able to use a kernel argument instead of a global

11–14 ↗

(On Diff #212455)

Control flow should be unnecessary

[rampitec]

In D65480#1607350, @arsenm wrote:

Was this broken by r365723? Should this use stripPointerCastsSameRepresentation instead?

I do not see a version which can accumulate constant offset.

[rampitec]

lib/Transforms/Vectorize/LoadStoreVectorizer.cpp

347–349 ↗

(On Diff #212455)

The also need to be the same as original pointer address space before the strip.

[arsenm]

LGTM

llvm/trunk/lib/Transforms/Vectorize/LoadStoreVectorizer.cpp

bool Vectorizer::isConsecutiveAccess(Value *A, Value *B) {
 
   return areConsecutivePointers(PtrA, PtrB, Size);
 }
 
 bool Vectorizer::areConsecutivePointers(Value *PtrA, Value *PtrB,
                                         const APInt &PtrDelta,
                                         unsigned Depth) const {
   unsigned PtrBitWidth = DL.getPointerTypeSizeInBits(PtrA->getType());
+  unsigned PtrAS = PtrA->getType()->getPointerAddressSpace();
   APInt OffsetA(PtrBitWidth, 0);
   APInt OffsetB(PtrBitWidth, 0);
   PtrA = PtrA->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetA);
   PtrB = PtrB->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetB);
 
+  if (PtrA->getType()->getPointerAddressSpace() != PtrAS ||
+      PtrB->getType()->getPointerAddressSpace() != PtrAS)
+    return false;
+
   APInt OffsetDelta = OffsetB - OffsetA;
 
   // Check if they are based on the same pointer. That makes the offsets
   // sufficient.
   if (PtrA == PtrB)
     return OffsetDelta == PtrDelta;
 
   // Compute the necessary base pointer delta to have the necessary final delta

llvm/trunk/test/Transforms/LoadStoreVectorizer/AMDGPU/vect-ptr-ptr-size-mismatch.ll

+; RUN: opt -mtriple=amdgcn-amd-amdhsa -load-store-vectorizer -S < %s | FileCheck %s
+
+target datalayout = "e-p:64:64-p1:64:64-p2:32:32-p3:32:32-p4:64:64-p5:32:32"
+
+; CHECK-LABEL: @test
+; CHECK: store i32* undef, i32** %tmp9, align 8
+; CHECK: store i32* undef, i32** %tmp7, align 8
+define amdgpu_kernel void @test() {
+entry:
+  %a10.ascast.i = addrspacecast i32* addrspace(5)* null to i32**
+  %tmp4 = icmp eq i32 undef, 0
+  %tmp6 = select i1 false, i32** undef, i32** undef
+  %tmp7 = select i1 %tmp4, i32** null, i32** %tmp6
+  %tmp9 = select i1 %tmp4, i32** %a10.ascast.i, i32** null
+  store i32* undef, i32** %tmp9, align 8
+  store i32* undef, i32** %tmp7, align 8
+  unreachable
+}