[SROA] Fix APInt size when load/store have different address space

Currently there is a bug in SROA::presplitLoadsAndStores which causes assertion in
GEPOperator::accumulateConstantOffset.

Basically it does not consider the situation that the pointer operand of load or store
may be in a non-zero address space and its size may be different from the size of
a pointer in address space 0.

This patch fixes assertion when compiling Blender Cycles kernels for amdgpu backend.

Diffferential Revision: https://reviews.llvm.org/D33298

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@305107 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Transforms/Scalar/SROA.cpp b/lib/Transforms/Scalar/SROA.cpp
index fb1b5813fd79f04fb879bd7eace47138cdaa0579..1527f15f18a3317f992ba322dcc8f04a88455fbd 100644 (file)
--- a/lib/Transforms/Scalar/SROA.cpp
+++ b/lib/Transforms/Scalar/SROA.cpp
@@ -3626,10 +3626,12 @@ bool SROA::presplitLoadsAndStores(AllocaInst &AI, AllocaSlices &AS) {
         auto *PartPtrTy =
             PLoad->getType()->getPointerTo(SI->getPointerAddressSpace());
 
+        auto AS = SI->getPointerAddressSpace();
         StoreInst *PStore = IRB.CreateAlignedStore(
-            PLoad, getAdjustedPtr(IRB, DL, StoreBasePtr,
-                                  APInt(DL.getPointerSizeInBits(), PartOffset),
-                                  PartPtrTy, StoreBasePtr->getName() + "."),
+            PLoad,
+            getAdjustedPtr(IRB, DL, StoreBasePtr,
+                           APInt(DL.getPointerSizeInBits(AS), PartOffset),
+                           PartPtrTy, StoreBasePtr->getName() + "."),
             getAdjustedAlignment(SI, PartOffset, DL), /*IsVolatile*/ false);
         PStore->copyMetadata(*LI, LLVMContext::MD_mem_parallel_loop_access);
         DEBUG(dbgs() << "      +" << PartOffset << ":" << *PStore << "\n");
@@ -3707,9 +3709,10 @@ bool SROA::presplitLoadsAndStores(AllocaInst &AI, AllocaSlices &AS) {
         PLoad = (*SplitLoads)[Idx];
       } else {
         IRB.SetInsertPoint(LI);
+        auto AS = LI->getPointerAddressSpace();
         PLoad = IRB.CreateAlignedLoad(
             getAdjustedPtr(IRB, DL, LoadBasePtr,
-                           APInt(DL.getPointerSizeInBits(), PartOffset),
+                           APInt(DL.getPointerSizeInBits(AS), PartOffset),
                            LoadPartPtrTy, LoadBasePtr->getName() + "."),
             getAdjustedAlignment(LI, PartOffset, DL), /*IsVolatile*/ false,
             LI->getName());
@@ -3717,10 +3720,12 @@ bool SROA::presplitLoadsAndStores(AllocaInst &AI, AllocaSlices &AS) {
 
       // And store this partition.
       IRB.SetInsertPoint(SI);
+      auto AS = SI->getPointerAddressSpace();
       StoreInst *PStore = IRB.CreateAlignedStore(
-          PLoad, getAdjustedPtr(IRB, DL, StoreBasePtr,
-                                APInt(DL.getPointerSizeInBits(), PartOffset),
-                                StorePartPtrTy, StoreBasePtr->getName() + "."),
+          PLoad,
+          getAdjustedPtr(IRB, DL, StoreBasePtr,
+                         APInt(DL.getPointerSizeInBits(AS), PartOffset),
+                         StorePartPtrTy, StoreBasePtr->getName() + "."),
           getAdjustedAlignment(SI, PartOffset, DL), /*IsVolatile*/ false);
 
       // Now build a new slice for the alloca.

diff --git a/test/Transforms/SROA/address-spaces.ll b/test/Transforms/SROA/address-spaces.ll
index 8fba30c2720f5af439063b1c48f0d395c0208e78..a54a3afc79f965c82c6b789eb2082eec9067c24e 100644 (file)
--- a/test/Transforms/SROA/address-spaces.ll
+++ b/test/Transforms/SROA/address-spaces.ll
@@ -101,3 +101,31 @@ entry:
    %ret = fadd float %f1, %f2
    ret float %ret
 }
+
+; Test load from and store to non-zero address space.
+define void @test_load_store_diff_addr_space([2 x float] addrspace(1)* %complex1, [2 x float] addrspace(1)* %complex2) {
+; CHECK-LABEL: @test_load_store_diff_addr_space
+; CHECK-NOT: alloca
+; CHECK: load i32, i32 addrspace(1)*
+; CHECK: load i32, i32 addrspace(1)*
+; CHECK: store i32 %{{.*}}, i32 addrspace(1)*
+; CHECK: store i32 %{{.*}}, i32 addrspace(1)*
+  %a = alloca i64
+  %a.cast = bitcast i64* %a to [2 x float]*
+  %a.gep1 = getelementptr [2 x float], [2 x float]* %a.cast, i32 0, i32 0
+  %a.gep2 = getelementptr [2 x float], [2 x float]* %a.cast, i32 0, i32 1
+  %complex1.gep = getelementptr [2 x float], [2 x float] addrspace(1)* %complex1, i32 0, i32 0
+  %p1 = bitcast float addrspace(1)* %complex1.gep to i64 addrspace(1)*
+  %v1 = load i64, i64 addrspace(1)* %p1
+  store i64 %v1, i64* %a
+  %f1 = load float, float* %a.gep1
+  %f2 = load float, float* %a.gep2
+  %sum = fadd float %f1, %f2
+  store float %sum, float* %a.gep1
+  store float %sum, float* %a.gep2
+  %v2 = load i64, i64* %a
+  %complex2.gep = getelementptr [2 x float], [2 x float] addrspace(1)* %complex2, i32 0, i32 0
+  %p2 = bitcast float addrspace(1)* %complex2.gep to i64 addrspace(1)*
+  store i64 %v2, i64 addrspace(1)* %p2
+  ret void
+}