[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

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

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

1

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

3

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

11–14

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

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

[arsenm]

LGTM

lib/Transforms/Vectorize/LoadStoreVectorizer.cpp

bool Vectorizer::areConsecutivePointers(Value *PtrA, Value *PtrB,
                                         const APInt &PtrDelta,
                                         unsigned Depth) const {
   unsigned PtrBitWidth = DL.getPointerTypeSizeInBits(PtrA->getType());
   APInt OffsetA(PtrBitWidth, 0);
   APInt OffsetB(PtrBitWidth, 0);
   PtrA = PtrA->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetA);
   PtrB = PtrB->stripAndAccumulateInBoundsConstantOffsets(DL, OffsetB);
 
+  if (DL.getTypeStoreSizeInBits(PtrA->getType()) != PtrBitWidth ||
+      DL.getTypeStoreSizeInBits(PtrB->getType()) != PtrBitWidth)
+    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

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

+; RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx803 -verify-machineinstrs < %s | FileCheck -check-prefix=GCN %s
+
+%struct._globalized_locals_ty = type { [64 x i32], [64 x i32*], [64 x i32*], [64 x i32*] }
+
+@execution_param = external local_unnamed_addr addrspace(3) externally_initialized global i32, align 4
+
+; GCN-LABEL: {{^}}test:
+; GCN: s_cbranch
+; GCN: s_endpgm
+define amdgpu_kernel void @test() #0 {
+entry:
+  br i1 undef, label %.worker, label %.mastercheck
+
+.worker:                                          ; preds = %entry
+  ret void
+
+.mastercheck:                                     ; preds = %entry
+  %tmp = load i32, i32 addrspace(3)* @execution_param, align 4
+  %tmp3 = and i32 %tmp, 1
+  %tmp4 = icmp eq i32 %tmp3, 0
+  %tmp7 = select i1 %tmp4, i32** null, i32** getelementptr (%struct._globalized_locals_ty, %struct._globalized_locals_ty* null, i64 0, i32 1, i64 undef)
+  %tmp9 = select i1 %tmp4, i32** addrspacecast (i32* addrspace(5)* null to i32**), i32** null
+  store i32* undef, i32** %tmp9, align 8
+  store i32* undef, i32** %tmp7, align 8
+  unreachable
+}