1; RUN: opt < %s -msan -msan-check-access-address=0 -S | FileCheck %s
2
3; Test instrumentation of vector shift instructions.
4
5target 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"
6target triple = "x86_64-unknown-linux-gnu"
7
8declare x86_mmx @llvm.x86.mmx.psll.d(x86_mmx, x86_mmx)
9declare <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32>, <8 x i32>)
10declare <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32>, <4 x i32>)
11declare <8 x i16> @llvm.x86.sse2.psrl.w(<8 x i16>, <8 x i16>)
12declare <8 x i16> @llvm.x86.sse2.pslli.w(<8 x i16>, i32)
13declare <2 x i64> @llvm.x86.sse2.psll.dq(<2 x i64>, i32)
14declare <2 x i64> @llvm.x86.sse2.psll.dq.bs(<2 x i64>, i32)
15
16define i64 @test_mmx(i64 %x.coerce, i64 %y.coerce) sanitize_memory {
17entry:
18  %0 = bitcast i64 %x.coerce to <2 x i32>
19  %1 = bitcast <2 x i32> %0 to x86_mmx
20  %2 = bitcast i64 %y.coerce to x86_mmx
21  %3 = tail call x86_mmx @llvm.x86.mmx.psll.d(x86_mmx %1, x86_mmx %2)
22  %4 = bitcast x86_mmx %3 to <2 x i32>
23  %5 = bitcast <2 x i32> %4 to <1 x i64>
24  %6 = extractelement <1 x i64> %5, i32 0
25  ret i64 %6
26}
27
28; CHECK: @test_mmx
29; CHECK: = icmp ne i64 {{.*}}, 0
30; CHECK: [[C:%.*]] = sext i1 {{.*}} to i64
31; CHECK: [[A:%.*]] = call x86_mmx @llvm.x86.mmx.psll.d(
32; CHECK: [[B:%.*]] = bitcast x86_mmx {{.*}}[[A]] to i64
33; CHECK: = or i64 {{.*}}[[B]], {{.*}}[[C]]
34; CHECK: call x86_mmx @llvm.x86.mmx.psll.d(
35; CHECK: ret i64
36
37
38define <8 x i16> @test_sse2_scalar(<8 x i16> %x, i32 %y) sanitize_memory {
39entry:
40  %0 = tail call <8 x i16> @llvm.x86.sse2.pslli.w(<8 x i16> %x, i32 %y)
41  ret <8 x i16> %0
42}
43
44; CHECK: @test_sse2_scalar
45; CHECK: = icmp ne i32 {{.*}}, 0
46; CHECK: = sext i1 {{.*}} to i128
47; CHECK: = bitcast i128 {{.*}} to <8 x i16>
48; CHECK: = call <8 x i16> @llvm.x86.sse2.pslli.w(
49; CHECK: = or <8 x i16>
50; CHECK: call <8 x i16> @llvm.x86.sse2.pslli.w(
51; CHECK: ret <8 x i16>
52
53
54define <8 x i16> @test_sse2(<8 x i16> %x, <8 x i16> %y) sanitize_memory {
55entry:
56  %0 = tail call <8 x i16> @llvm.x86.sse2.psrl.w(<8 x i16> %x, <8 x i16> %y)
57  ret <8 x i16> %0
58}
59
60; CHECK: @test_sse2
61; CHECK: = bitcast <8 x i16> {{.*}} to i128
62; CHECK: = trunc i128 {{.*}} to i64
63; CHECK: = icmp ne i64 {{.*}}, 0
64; CHECK: = sext i1 {{.*}} to i128
65; CHECK: = bitcast i128 {{.*}} to <8 x i16>
66; CHECK: = call <8 x i16> @llvm.x86.sse2.psrl.w(
67; CHECK: = or <8 x i16>
68; CHECK: call <8 x i16> @llvm.x86.sse2.psrl.w(
69; CHECK: ret <8 x i16>
70
71
72; Test variable shift (i.e. vector by vector).
73
74define <4 x i32> @test_avx2(<4 x i32> %x, <4 x i32> %y) sanitize_memory {
75entry:
76  %0 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %x, <4 x i32> %y)
77  ret <4 x i32> %0
78}
79
80; CHECK: @test_avx2
81; CHECK: = icmp ne <4 x i32> {{.*}}, zeroinitializer
82; CHECK: = sext <4 x i1> {{.*}} to <4 x i32>
83; CHECK: = call <4 x i32> @llvm.x86.avx2.psllv.d(
84; CHECK: = or <4 x i32>
85; CHECK: = tail call <4 x i32> @llvm.x86.avx2.psllv.d(
86; CHECK: ret <4 x i32>
87
88define <8 x i32> @test_avx2_256(<8 x i32> %x, <8 x i32> %y) sanitize_memory {
89entry:
90  %0 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %x, <8 x i32> %y)
91  ret <8 x i32> %0
92}
93
94; CHECK: @test_avx2_256
95; CHECK: = icmp ne <8 x i32> {{.*}}, zeroinitializer
96; CHECK: = sext <8 x i1> {{.*}} to <8 x i32>
97; CHECK: = call <8 x i32> @llvm.x86.avx2.psllv.d.256(
98; CHECK: = or <8 x i32>
99; CHECK: = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(
100; CHECK: ret <8 x i32>
101