xref: /llvm-project/llvm/test/Transforms/SLPVectorizer/X86/blending-shuffle-inseltpoison.ll (revision a98882112387b6bc585e758a8139fd06ac41f448) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -passes=slp-vectorizer,instcombine -S -o - -mtriple=x86_64-unknown-linux -mcpu=bdver2 | FileCheck %s
3
4define <2 x i8> @g(<2 x i8> %x, <2 x i8> %y) {
5; CHECK-LABEL: @g(
6; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <2 x i8> [[X:%.*]], <2 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 3>
7; CHECK-NEXT:    [[TMP2:%.*]] = mul <2 x i8> [[TMP1]], [[TMP1]]
8; CHECK-NEXT:    ret <2 x i8> [[TMP2]]
9;
10  %x0 = extractelement <2 x i8> %x, i32 0
11  %y1 = extractelement <2 x i8> %y, i32 1
12  %x0x0 = mul i8 %x0, %x0
13  %y1y1 = mul i8 %y1, %y1
14  %ins1 = insertelement <2 x i8> poison, i8 %x0x0, i32 0
15  %ins2 = insertelement <2 x i8> %ins1, i8 %y1y1, i32 1
16  ret <2 x i8> %ins2
17}
18
19define <4 x i8> @h(<4 x i8> %x, <4 x i8> %y) {
20; CHECK-LABEL: @h(
21; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <4 x i32> <i32 0, i32 3, i32 5, i32 6>
22; CHECK-NEXT:    [[TMP2:%.*]] = mul <4 x i8> [[TMP1]], [[TMP1]]
23; CHECK-NEXT:    ret <4 x i8> [[TMP2]]
24;
25  %x0 = extractelement <4 x i8> %x, i32 0
26  %x3 = extractelement <4 x i8> %x, i32 3
27  %y1 = extractelement <4 x i8> %y, i32 1
28  %y2 = extractelement <4 x i8> %y, i32 2
29  %x0x0 = mul i8 %x0, %x0
30  %x3x3 = mul i8 %x3, %x3
31  %y1y1 = mul i8 %y1, %y1
32  %y2y2 = mul i8 %y2, %y2
33  %ins1 = insertelement <4 x i8> poison, i8 %x0x0, i32 0
34  %ins2 = insertelement <4 x i8> %ins1, i8 %x3x3, i32 1
35  %ins3 = insertelement <4 x i8> %ins2, i8 %y1y1, i32 2
36  %ins4 = insertelement <4 x i8> %ins3, i8 %y2y2, i32 3
37  ret <4 x i8> %ins4
38}
39
40define <4 x i8> @h_undef(<4 x i8> %x, <4 x i8> %y) {
41; CHECK-LABEL: @h_undef(
42; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <4 x i32> <i32 poison, i32 3, i32 5, i32 6>
43; CHECK-NEXT:    [[TMP2:%.*]] = mul <4 x i8> [[TMP1]], [[TMP1]]
44; CHECK-NEXT:    ret <4 x i8> [[TMP2]]
45;
46  %x0 = extractelement <4 x i8> poison, i32 0
47  %x3 = extractelement <4 x i8> %x, i32 3
48  %y1 = extractelement <4 x i8> %y, i32 1
49  %y2 = extractelement <4 x i8> %y, i32 2
50  %x0x0 = mul i8 %x0, %x0
51  %x3x3 = mul i8 %x3, %x3
52  %y1y1 = mul i8 %y1, %y1
53  %y2y2 = mul i8 %y2, %y2
54  %ins1 = insertelement <4 x i8> poison, i8 %x0x0, i32 0
55  %ins2 = insertelement <4 x i8> %ins1, i8 %x3x3, i32 1
56  %ins3 = insertelement <4 x i8> %ins2, i8 %y1y1, i32 2
57  %ins4 = insertelement <4 x i8> %ins3, i8 %y2y2, i32 3
58  ret <4 x i8> %ins4
59}
60
61define i8 @i(<4 x i8> %x, <4 x i8> %y) {
62; CHECK-LABEL: @i(
63; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <4 x i32> <i32 0, i32 3, i32 5, i32 6>
64; CHECK-NEXT:    [[TMP2:%.*]] = mul <4 x i8> [[TMP1]], [[TMP1]]
65; CHECK-NEXT:    [[TMP3:%.*]] = call i8 @llvm.vector.reduce.add.v4i8(<4 x i8> [[TMP2]])
66; CHECK-NEXT:    ret i8 [[TMP3]]
67;
68  %x0 = extractelement <4 x i8> %x, i32 0
69  %x3 = extractelement <4 x i8> %x, i32 3
70  %y1 = extractelement <4 x i8> %y, i32 1
71  %y2 = extractelement <4 x i8> %y, i32 2
72  %x0x0 = mul i8 %x0, %x0
73  %x3x3 = mul i8 %x3, %x3
74  %y1y1 = mul i8 %y1, %y1
75  %y2y2 = mul i8 %y2, %y2
76  %1 = add i8 %x0x0, %x3x3
77  %2 = add i8 %y1y1, %y2y2
78  %3 = add i8 %1, %2
79  ret i8 %3
80}
81
82define i8 @j(<4 x i8> %x, <4 x i8> %y) {
83; CHECK-LABEL: @j(
84; CHECK-NEXT:    [[TMP1:%.*]] = shufflevector <4 x i8> [[X:%.*]], <4 x i8> [[Y:%.*]], <2 x i32> <i32 0, i32 5>
85; CHECK-NEXT:    [[TMP2:%.*]] = mul <2 x i8> [[TMP1]], [[TMP1]]
86; CHECK-NEXT:    [[TMP3:%.*]] = shufflevector <4 x i8> [[X]], <4 x i8> [[Y]], <2 x i32> <i32 3, i32 6>
87; CHECK-NEXT:    [[TMP4:%.*]] = mul <2 x i8> [[TMP3]], [[TMP3]]
88; CHECK-NEXT:    [[TMP5:%.*]] = add <2 x i8> [[TMP2]], [[TMP4]]
89; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <2 x i8> [[TMP5]], i64 0
90; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP5]], i64 1
91; CHECK-NEXT:    [[TMP8:%.*]] = sdiv i8 [[TMP6]], [[TMP7]]
92; CHECK-NEXT:    ret i8 [[TMP8]]
93;
94  %x0 = extractelement <4 x i8> %x, i32 0
95  %x3 = extractelement <4 x i8> %x, i32 3
96  %y1 = extractelement <4 x i8> %y, i32 1
97  %y2 = extractelement <4 x i8> %y, i32 2
98  %x0x0 = mul i8 %x0, %x0
99  %x3x3 = mul i8 %x3, %x3
100  %y1y1 = mul i8 %y1, %y1
101  %y2y2 = mul i8 %y2, %y2
102  %1 = add i8 %x0x0, %x3x3
103  %2 = add i8 %y1y1, %y2y2
104  %3 = sdiv i8 %1, %2
105  ret i8 %3
106}
107
108define i8 @k(<4 x i8> %x) {
109; CHECK-LABEL: @k(
110; CHECK-NEXT:    [[TMP1:%.*]] = mul <4 x i8> [[X:%.*]], [[X]]
111; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>
112; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i8> [[X]], [[X]]
113; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <2 x i32> <i32 2, i32 3>
114; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i8> [[TMP4]], <2 x i8> poison, <2 x i32> <i32 1, i32 0>
115; CHECK-NEXT:    [[TMP6:%.*]] = add <2 x i8> [[TMP2]], [[TMP5]]
116; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP6]], i64 0
117; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i8> [[TMP6]], i64 1
118; CHECK-NEXT:    [[TMP9:%.*]] = sdiv i8 [[TMP7]], [[TMP8]]
119; CHECK-NEXT:    ret i8 [[TMP9]]
120;
121  %x0 = extractelement <4 x i8> %x, i32 0
122  %x3 = extractelement <4 x i8> %x, i32 3
123  %x1 = extractelement <4 x i8> %x, i32 1
124  %x2 = extractelement <4 x i8> %x, i32 2
125  %x0x0 = mul i8 %x0, %x0
126  %x3x3 = mul i8 %x3, %x3
127  %x1x1 = mul i8 %x1, %x1
128  %x2x2 = mul i8 %x2, %x2
129  %1 = add i8 %x0x0, %x3x3
130  %2 = add i8 %x1x1, %x2x2
131  %3 = sdiv i8 %1, %2
132  ret i8 %3
133}
134
135define i8 @k_bb(<4 x i8> %x) {
136; CHECK-LABEL: @k_bb(
137; CHECK-NEXT:    br label [[BB1:%.*]]
138; CHECK:       bb1:
139; CHECK-NEXT:    [[TMP1:%.*]] = mul <4 x i8> [[X:%.*]], [[X]]
140; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <4 x i8> [[TMP1]], <4 x i8> poison, <2 x i32> <i32 0, i32 1>
141; CHECK-NEXT:    [[TMP3:%.*]] = mul <4 x i8> [[X]], [[X]]
142; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <4 x i8> [[TMP3]], <4 x i8> poison, <2 x i32> <i32 2, i32 3>
143; CHECK-NEXT:    [[TMP5:%.*]] = shufflevector <2 x i8> [[TMP4]], <2 x i8> poison, <2 x i32> <i32 1, i32 0>
144; CHECK-NEXT:    [[TMP6:%.*]] = add <2 x i8> [[TMP2]], [[TMP5]]
145; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <2 x i8> [[TMP6]], i64 0
146; CHECK-NEXT:    [[TMP8:%.*]] = extractelement <2 x i8> [[TMP6]], i64 1
147; CHECK-NEXT:    [[TMP9:%.*]] = sdiv i8 [[TMP7]], [[TMP8]]
148; CHECK-NEXT:    ret i8 [[TMP9]]
149;
150  %x0 = extractelement <4 x i8> %x, i32 0
151  br label %bb1
152bb1:
153  %x3 = extractelement <4 x i8> %x, i32 3
154  %x1 = extractelement <4 x i8> %x, i32 1
155  %x2 = extractelement <4 x i8> %x, i32 2
156  %x0x0 = mul i8 %x0, %x0
157  %x3x3 = mul i8 %x3, %x3
158  %x1x1 = mul i8 %x1, %x1
159  %x2x2 = mul i8 %x2, %x2
160  %1 = add i8 %x0x0, %x3x3
161  %2 = add i8 %x1x1, %x2x2
162  %3 = sdiv i8 %1, %2
163  ret i8 %3
164}
165