llvm.org GIT mirror llvm / 7dfcb84
Fix for a regression caused by the LoopVectorizer when vectorizing loops with memory accesses to non-zero address spaces. It simply dropped the AS info. Fixes PR16306. Merged from r184103 Author: Pekka Jaaskelainen <pekka.jaaskelainen@tut.fi> Date: Mon Jun 17 18:49:06 2013 +0000 git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/release_33@185869 91177308-0d34-0410-b5e6-96231b3b80d8 Tom Stellard 7 years ago
2 changed file(s) with 50 addition(s) and 3 deletion(s). Raw diff Collapse all Expand all
963963 Type *DataTy = VectorType::get(ScalarDataTy, VF);
964964 Value *Ptr = LI ? LI->getPointerOperand() : SI->getPointerOperand();
965965 unsigned Alignment = LI ? LI->getAlignment() : SI->getAlignment();
966
966 unsigned AddressSpace = Ptr->getType()->getPointerAddressSpace();
967967 unsigned ScalarAllocatedSize = DL->getTypeAllocSize(ScalarDataTy);
968968 unsigned VectorElementSize = DL->getTypeStoreSize(DataTy)/VF;
969969
10381038 PartPtr = Builder.CreateGEP(PartPtr, Builder.getInt32(1 - VF));
10391039 }
10401040
1041 Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo());
1041 Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo(AddressSpace));
10421042 Builder.CreateStore(StoredVal[Part], VecPtr)->setAlignment(Alignment);
10431043 }
10441044 }
10541054 PartPtr = Builder.CreateGEP(PartPtr, Builder.getInt32(1 - VF));
10551055 }
10561056
1057 Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo());
1057 Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo(AddressSpace));
10581058 Value *LI = Builder.CreateLoad(VecPtr, "wide.load");
10591059 cast(LI)->setAlignment(Alignment);
10601060 Entry[Part] = Reverse ? reverseVector(LI) : LI;
0 ; RUN: opt < %s -loop-vectorize -force-vector-unroll=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s
1
2 ; From a simple program with two address spaces:
3 ; char Y[4*10000] __attribute__((address_space(1)));
4 ; char X[4*10000];
5 ; int main() {
6 ; for (int i = 0; i < 4*10000; ++i)
7 ; X[i] = Y[i] + 1;
8 ; return 0;
9 ;}
10
11
12 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64-S128"
13 target triple = "x86_64-unknown-linux-gnu"
14
15 @Y = common addrspace(1) global [40000 x i8] zeroinitializer, align 16
16 @X = common global [40000 x i8] zeroinitializer, align 16
17
18 ;CHECK: @main
19 ;CHECK: bitcast i8 addrspace(1)* %{{.*}} to <4 x i8> addrspace(1)*
20 ;CHECK: bitcast i8* %{{.*}} to <4 x i8>*
21
22 ; Function Attrs: nounwind uwtable
23 define i32 @main() #0 {
24 entry:
25 br label %for.body
26
27 for.body: ; preds = %for.body, %entry
28 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
29 %arrayidx = getelementptr inbounds [40000 x i8] addrspace(1)* @Y, i64 0, i64 %indvars.iv
30 %0 = load i8 addrspace(1)* %arrayidx, align 1, !tbaa !0
31 %add = add i8 %0, 1
32 %arrayidx3 = getelementptr inbounds [40000 x i8]* @X, i64 0, i64 %indvars.iv
33 store i8 %add, i8* %arrayidx3, align 1, !tbaa !0
34 %indvars.iv.next = add i64 %indvars.iv, 1
35 %lftr.wideiv = trunc i64 %indvars.iv.next to i32
36 %exitcond = icmp eq i32 %lftr.wideiv, 40000
37 br i1 %exitcond, label %for.end, label %for.body
38
39 for.end: ; preds = %for.body
40 ret i32 0
41 }
42
43 attributes #0 = { nounwind uwtable "less-precise-fpmad"="false" "no-frame-pointer-elim"="false" "no-frame-pointer-elim-non-leaf"="false" "no-infs-fp-math"="false" "no-nans-fp-math"="false" "unsafe-fp-math"="false" "use-soft-float"="false" }
44
45 !0 = metadata !{metadata !"omnipotent char", metadata !1}
46 !1 = metadata !{metadata !"Simple C/C++ TBAA"}