xref: /llvm-project/llvm/test/Instrumentation/MemorySanitizer/vector_shift.ll (revision fe7f5f9126cea9ceba703d5bd07b766181f2bd72)
1; RUN: opt < %s -msan-check-access-address=0 -S -passes=msan 2>&1 | FileCheck %s
2; REQUIRES: x86-registered-target
3
4; Test instrumentation of vector shift instructions.
5
6target 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"
7target triple = "x86_64-unknown-linux-gnu"
8
9declare <1 x i64> @llvm.x86.mmx.psll.d(<1 x i64>, <1 x i64>)
10declare <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32>, <16 x i32>)
11declare <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32>, <8 x i32>)
12declare <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32>, <4 x i32>)
13declare <8 x i16> @llvm.x86.sse2.psrl.w(<8 x i16>, <8 x i16>)
14declare <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16>, <8 x i16>)
15declare <8 x i16> @llvm.x86.sse2.pslli.w(<8 x i16>, i32)
16declare <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16>, i32)
17
18define i64 @test_mmx(i64 %x.coerce, i64 %y.coerce) sanitize_memory {
19entry:
20  %0 = bitcast i64 %x.coerce to <2 x i32>
21  %1 = bitcast <2 x i32> %0 to <1 x i64>
22  %2 = bitcast i64 %y.coerce to <1 x i64>
23  %3 = tail call <1 x i64> @llvm.x86.mmx.psll.d(<1 x i64> %1, <1 x i64> %2)
24  %4 = bitcast <1 x i64> %3 to <2 x i32>
25  %5 = bitcast <2 x i32> %4 to <1 x i64>
26  %6 = extractelement <1 x i64> %5, i32 0
27  ret i64 %6
28}
29
30; CHECK-LABEL: @test_mmx
31; CHECK: = icmp ne i64 {{.*}}, 0
32; CHECK: [[B:%.*]] = sext i1 {{.*}} to i64
33; CHECK: [[C:%.*]] = bitcast i64 [[B]] to <1 x i64>
34; CHECK: [[A:%.*]] = call <1 x i64> @llvm.x86.mmx.psll.d(
35; CHECK: = or <1 x i64> {{.*}}[[A]], {{.*}}[[C]]
36; CHECK: call <1 x i64> @llvm.x86.mmx.psll.d(
37; CHECK: ret i64
38
39
40define <8 x i16> @test_sse2_scalar(<8 x i16> %x, i32 %y) sanitize_memory {
41entry:
42  %0 = tail call <8 x i16> @llvm.x86.sse2.pslli.w(<8 x i16> %x, i32 %y)
43  ret <8 x i16> %0
44}
45
46; CHECK-LABEL: @test_sse2_scalar
47; CHECK: = icmp ne i32 {{.*}}, 0
48; CHECK: = sext i1 {{.*}} to i128
49; CHECK: = bitcast i128 {{.*}} to <8 x i16>
50; CHECK: = call <8 x i16> @llvm.x86.sse2.pslli.w(
51; CHECK: = or <8 x i16>
52; CHECK: call <8 x i16> @llvm.x86.sse2.pslli.w(
53; CHECK: ret <8 x i16>
54
55
56define <32 x i16> @test_avx512_scalar(<32 x i16> %x, i32 %y) sanitize_memory {
57entry:
58  %0 = tail call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %x, i32 %y)
59  ret <32 x i16> %0
60}
61
62; CHECK-LABEL: @test_avx512_scalar
63; CHECK: = icmp ne i32 {{.*}}, 0
64; CHECK: = sext i1 {{.*}} to i512
65; CHECK: = bitcast i512 {{.*}} to <32 x i16>
66; CHECK: = call <32 x i16> @llvm.x86.avx512.pslli.w.512(
67; CHECK: = or <32 x i16>
68; CHECK: call <32 x i16> @llvm.x86.avx512.pslli.w.512(
69; CHECK: ret <32 x i16>
70
71
72define <8 x i16> @test_sse2(<8 x i16> %x, <8 x i16> %y) sanitize_memory {
73entry:
74  %0 = tail call <8 x i16> @llvm.x86.sse2.psrl.w(<8 x i16> %x, <8 x i16> %y)
75  ret <8 x i16> %0
76}
77
78; CHECK-LABEL: @test_sse2
79; CHECK: = bitcast <8 x i16> {{.*}} to i128
80; CHECK: = trunc i128 {{.*}} to i64
81; CHECK: = icmp ne i64 {{.*}}, 0
82; CHECK: = sext i1 {{.*}} to i128
83; CHECK: = bitcast i128 {{.*}} to <8 x i16>
84; CHECK: = call <8 x i16> @llvm.x86.sse2.psrl.w(
85; CHECK: = or <8 x i16>
86; CHECK: call <8 x i16> @llvm.x86.sse2.psrl.w(
87; CHECK: ret <8 x i16>
88
89
90define <32 x i16> @test_avx512(<32 x i16> %x, <8 x i16> %y) sanitize_memory {
91entry:
92  %0 = tail call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %x, <8 x i16> %y)
93  ret <32 x i16> %0
94}
95
96; CHECK-LABEL: @test_avx512
97; CHECK: = bitcast <8 x i16> {{.*}} to i128
98; CHECK: = trunc i128 {{.*}} to i64
99; CHECK: = icmp ne i64 {{.*}}, 0
100; CHECK: = sext i1 {{.*}} to i512
101; CHECK: = bitcast i512 {{.*}} to <32 x i16>
102; CHECK: = call <32 x i16> @llvm.x86.avx512.psrl.w.512(
103; CHECK: = or <32 x i16>
104; CHECK: call <32 x i16> @llvm.x86.avx512.psrl.w.512(
105; CHECK: ret <32 x i16>
106
107
108; Test variable shift (i.e. vector by vector).
109
110define <4 x i32> @test_avx2(<4 x i32> %x, <4 x i32> %y) sanitize_memory {
111entry:
112  %0 = tail call <4 x i32> @llvm.x86.avx2.psllv.d(<4 x i32> %x, <4 x i32> %y)
113  ret <4 x i32> %0
114}
115
116; CHECK-LABEL: @test_avx2
117; CHECK: = icmp ne <4 x i32> {{.*}}, zeroinitializer
118; CHECK: = sext <4 x i1> {{.*}} to <4 x i32>
119; CHECK: = call <4 x i32> @llvm.x86.avx2.psllv.d(
120; CHECK: = or <4 x i32>
121; CHECK: = tail call <4 x i32> @llvm.x86.avx2.psllv.d(
122; CHECK: ret <4 x i32>
123
124define <8 x i32> @test_avx2_256(<8 x i32> %x, <8 x i32> %y) sanitize_memory {
125entry:
126  %0 = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(<8 x i32> %x, <8 x i32> %y)
127  ret <8 x i32> %0
128}
129
130; CHECK-LABEL: @test_avx2_256
131; CHECK: = icmp ne <8 x i32> {{.*}}, zeroinitializer
132; CHECK: = sext <8 x i1> {{.*}} to <8 x i32>
133; CHECK: = call <8 x i32> @llvm.x86.avx2.psllv.d.256(
134; CHECK: = or <8 x i32>
135; CHECK: = tail call <8 x i32> @llvm.x86.avx2.psllv.d.256(
136; CHECK: ret <8 x i32>
137
138define <16 x i32> @test_avx512_512(<16 x i32> %x, <16 x i32> %y) sanitize_memory {
139entry:
140  %0 = tail call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %x, <16 x i32> %y)
141  ret <16 x i32> %0
142}
143
144; CHECK-LABEL: @test_avx512_512
145; CHECK: = icmp ne <16 x i32> {{.*}}, zeroinitializer
146; CHECK: = sext <16 x i1> {{.*}} to <16 x i32>
147; CHECK: = call <16 x i32> @llvm.x86.avx512.psllv.d.512(
148; CHECK: = or <16 x i32>
149; CHECK: = tail call <16 x i32> @llvm.x86.avx512.psllv.d.512(
150; CHECK: ret <16 x i32>
151