1 ; RUN: llc -verify-machineinstrs -mcpu=pwr8 -mtriple=powerpc64le-unknown-linux-gnu < %s | FileCheck %s 2 3 ; This code causes an assertion failure if dereferenceable flag is not properly set in the load generated for memcpy 4 5 ; CHECK-LABEL: @func 6 ; CHECK: lxvd2x [[VREG:[0-9]+]], {{[0-9]+}}, {{[0-9]+}} 7 ; CHECK-NOT: lxvd2x 8 ; CHECK: stxvd2x [[VREG:[0-9]+]], {{[0-9]+}}, {{[0-9]+}} 9 ; CHECK: stxvd2x [[VREG:[0-9]+]], {{[0-9]+}}, {{[0-9]+}} 10 ; CHECK: blr 11 12 define void @func(i1 %flag) { 13 entry: 14 %pairs = alloca [4 x <2 x i64>], align 8 15 %pair1 = getelementptr inbounds [4 x <2 x i64>], [4 x <2 x i64>]* %pairs, i64 0, i64 1 16 %pair2 = getelementptr inbounds [4 x <2 x i64>], [4 x <2 x i64>]* %pairs, i64 0, i64 2 17 %pvec1 = bitcast <2 x i64>* %pair1 to <2 x i64>* 18 %pvec2 = bitcast <2 x i64>* %pair2 to <2 x i64>* 19 %dst = bitcast [4 x <2 x i64>]* %pairs to i8* 20 %src = bitcast <2 x i64>* %pair2 to i8* 21 br i1 %flag, label %end, label %dummy 22 23 end: 24 ; copy third element into first element by memcpy 25 call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 nonnull %dst, i8* align 8 %src, i64 16, i1 false) 26 ; copy third element into second element by LD/ST 27 %vec2 = load <2 x i64>, <2 x i64>* %pvec2, align 8 28 store <2 x i64> %vec2, <2 x i64>* %pvec1, align 8 29 ret void 30 31 dummy: 32 ; to make use of %src in another BB 33 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %src, i8* %src, i64 0, i1 false) 34 br label %end 35 } 36 37 38 ; CHECK-LABEL: @func2 39 ; CHECK: lxvd2x [[VREG:[0-9]+]], {{[0-9]+}}, {{[0-9]+}} 40 ; CHECK-NOT: lxvd2x 41 ; CHECK: stxvd2x [[VREG:[0-9]+]], {{[0-9]+}}, {{[0-9]+}} 42 ; CHECK: stxvd2x [[VREG:[0-9]+]], {{[0-9]+}}, {{[0-9]+}} 43 ; CHECK: blr 44 45 define void @func2(i1 %flag) { 46 entry: 47 %pairs = alloca [4 x <2 x i64>], align 8 48 %pair1 = getelementptr inbounds [4 x <2 x i64>], [4 x <2 x i64>]* %pairs, i64 0, i64 1 49 %pair2 = getelementptr inbounds [4 x <2 x i64>], [4 x <2 x i64>]* %pairs, i64 0, i64 2 50 %pvec1 = bitcast <2 x i64>* %pair1 to <2 x i64>* 51 %pvec2 = bitcast <2 x i64>* %pair2 to <2 x i64>* 52 %dst = bitcast [4 x <2 x i64>]* %pairs to i8* 53 %src = bitcast <2 x i64>* %pair2 to i8* 54 br i1 %flag, label %end, label %dummy 55 56 end: 57 ; copy third element into first element by memcpy 58 call void @llvm.memmove.p0i8.p0i8.i64(i8* align 8 nonnull %dst, i8* align 8 %src, i64 16, i1 false) 59 ; copy third element into second element by LD/ST 60 %vec2 = load <2 x i64>, <2 x i64>* %pvec2, align 8 61 store <2 x i64> %vec2, <2 x i64>* %pvec1, align 8 62 ret void 63 64 dummy: 65 ; to make use of %src in another BB 66 call void @llvm.memcpy.p0i8.p0i8.i64(i8* %src, i8* %src, i64 0, i1 false) 67 br label %end 68 } 69 70 ; Function Attrs: argmemonly nounwind 71 declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1 72 declare void @llvm.memmove.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i1) #1 73 74 attributes #1 = { argmemonly nounwind } 75
