1 ; RUN: llc -mtriple=aarch64-linux-gnu -mcpu=cortex-a57 -verify-machineinstrs < %s | FileCheck %s
3 ; This file check a bug in MachineCopyPropagation pass. The last COPY will be
4 ; incorrectly removed if the machine instructions are as follows:
9 ; This is caused by a bug in function SourceNoLongerAvailable(), which fails to
10 ; remove the relationship of D6 and "%q5_q6 = COPY %q2_q3".
12 @failed = internal unnamed_addr global i1 false
16 ; CHECK-NOT: // kill: def D{{[0-9]+}} killed D{{[0-9]+}}
17 define void @foo(<2 x i32> %shuffle251, <8 x i8> %vtbl1.i, i8* %t2, <2 x i32> %vrsubhn_v2.i1364) {
19 %val0 = alloca [2 x i64], align 8
20 %val1 = alloca <2 x i64>, align 16
21 %vmull = tail call <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32> <i32 -1, i32 -1>, <2 x i32> %shuffle251)
22 %vgetq_lane = extractelement <2 x i64> %vmull, i32 0
23 %cmp = icmp eq i64 %vgetq_lane, 1
24 br i1 %cmp, label %if.end, label %if.then
26 if.then: ; preds = %entry
27 store i1 true, i1* @failed, align 1
30 if.end: ; preds = %if.then, %entry
32 %sqdmull = tail call <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16> <i16 1, i16 0, i16 0, i16 0>, <4 x i16> <i16 2, i16 0, i16 0, i16 0>)
33 %sqadd = tail call <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32> zeroinitializer, <4 x i32> %sqdmull)
34 %shuffle = shufflevector <4 x i32> %sqadd, <4 x i32> undef, <2 x i32> zeroinitializer
35 %0 = mul <2 x i32> %shuffle, <i32 -1, i32 0>
36 %sub = add <2 x i32> %0, <i32 1, i32 0>
37 %sext = sext <2 x i32> %sub to <2 x i64>
38 %vset_lane603 = shufflevector <2 x i64> %sext, <2 x i64> undef, <1 x i32> zeroinitializer
39 %t1 = bitcast [2 x i64]* %val0 to i8*
40 call void @llvm.aarch64.neon.st2lane.v2i64.p0i8(<2 x i64> zeroinitializer, <2 x i64> zeroinitializer, i64 1, i8* %t1)
41 call void @llvm.aarch64.neon.st2lane.v1i64.p0i8(<1 x i64> <i64 4096>, <1 x i64> <i64 -1>, i64 0, i8* %t2)
42 %vld2_lane = call { <1 x i64>, <1 x i64> } @llvm.aarch64.neon.ld2lane.v1i64.p0i8(<1 x i64> <i64 11>, <1 x i64> <i64 11>, i64 0, i8* %t2)
43 %vld2_lane.0.extract = extractvalue { <1 x i64>, <1 x i64> } %vld2_lane, 0
44 %vld2_lane.1.extract = extractvalue { <1 x i64>, <1 x i64> } %vld2_lane, 1
45 %vld2_lane1 = call { <1 x i64>, <1 x i64> } @llvm.aarch64.neon.ld2lane.v1i64.p0i8(<1 x i64> %vld2_lane.0.extract, <1 x i64> %vld2_lane.1.extract, i64 0, i8* %t1)
46 %vld2_lane1.0.extract = extractvalue { <1 x i64>, <1 x i64> } %vld2_lane1, 0
47 %vld2_lane1.1.extract = extractvalue { <1 x i64>, <1 x i64> } %vld2_lane1, 1
48 %t3 = bitcast <2 x i64>* %val1 to i8*
49 call void @llvm.aarch64.neon.st2.v1i64.p0i8(<1 x i64> %vld2_lane1.0.extract, <1 x i64> %vld2_lane1.1.extract, i8* %t3)
50 %t4 = load <2 x i64>, <2 x i64>* %val1, align 16
51 %vsubhn = sub <2 x i64> <i64 11, i64 0>, %t4
52 %vsubhn1 = lshr <2 x i64> %vsubhn, <i64 32, i64 32>
53 %vsubhn2 = trunc <2 x i64> %vsubhn1 to <2 x i32>
54 %neg = xor <2 x i32> %vsubhn2, <i32 -1, i32 -1>
55 %sqadd1 = call <1 x i64> @llvm.aarch64.neon.usqadd.v1i64(<1 x i64> <i64 -1>, <1 x i64> <i64 1>)
56 %sqadd2 = call <1 x i64> @llvm.aarch64.neon.usqadd.v1i64(<1 x i64> %vset_lane603, <1 x i64> %sqadd1)
57 %sqadd3 = call <1 x i64> @llvm.aarch64.neon.usqadd.v1i64(<1 x i64> <i64 1>, <1 x i64> %sqadd2)
58 %shuffle.i = shufflevector <2 x i32> <i32 undef, i32 0>, <2 x i32> %vrsubhn_v2.i1364, <2 x i32> <i32 1, i32 3>
59 %cmp.i = icmp uge <2 x i32> %shuffle.i, %neg
60 %sext.i = sext <2 x i1> %cmp.i to <2 x i32>
61 %vpadal = call <1 x i64> @llvm.aarch64.neon.uaddlp.v1i64.v2i32(<2 x i32> %sext.i)
62 %t5 = sub <1 x i64> %vpadal, %sqadd3
63 %vget_lane1 = extractelement <1 x i64> %t5, i32 0
64 %cmp2 = icmp eq i64 %vget_lane1, 15
65 br i1 %cmp2, label %if.end2, label %if.then2
67 if.then2: ; preds = %if.end
68 store i1 true, i1* @failed, align 1
71 if.end2: ; preds = %if.then682, %if.end
73 %vext = shufflevector <8 x i8> <i8 undef, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1, i8 -1>, <8 x i8> %vtbl1.i, <8 x i32> <i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8>
74 %t6 = bitcast <8 x i8> %vext to <2 x i32>
75 call void @f0(<2 x i32> %t6)
79 declare void @f0(<2 x i32>)
81 declare <8 x i8> @f1()
85 declare <4 x i32> @llvm.aarch64.neon.sqdmull.v4i32(<4 x i16>, <4 x i16>)
87 declare void @llvm.aarch64.neon.st2lane.v2i64.p0i8(<2 x i64>, <2 x i64>, i64, i8* nocapture)
89 declare void @llvm.aarch64.neon.st2lane.v1i64.p0i8(<1 x i64>, <1 x i64>, i64, i8* nocapture)
91 declare { <1 x i64>, <1 x i64> } @llvm.aarch64.neon.ld2lane.v1i64.p0i8(<1 x i64>, <1 x i64>, i64, i8*)
93 declare void @llvm.aarch64.neon.st2.v1i64.p0i8(<1 x i64>, <1 x i64>, i8* nocapture)
95 declare <1 x i64> @llvm.aarch64.neon.usqadd.v1i64(<1 x i64>, <1 x i64>)
97 declare <1 x i64> @llvm.aarch64.neon.uaddlp.v1i64.v2i32(<2 x i32>)
99 declare <4 x i32> @llvm.aarch64.neon.sqadd.v4i32(<4 x i32>, <4 x i32>)
101 declare <2 x i64> @llvm.aarch64.neon.umull.v2i64(<2 x i32>, <2 x i32>)