xref: /llvm-project/llvm/test/Analysis/CostModel/X86/shuffle-broadcast-codesize.ll (revision baecc9e997dd912a6b3589a529394c77570acc19) 
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py
2; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse2 | FileCheck %s -check-prefixes=SSE,SSE2
3; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+ssse3 | FileCheck %s -check-prefixes=SSE,SSSE3
4; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+sse4.2 | FileCheck %s -check-prefixes=SSE,SSE42
5; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx | FileCheck %s -check-prefixes=AVX1
6; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx2 | FileCheck %s -check-prefixes=AVX2
7; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f | FileCheck %s --check-prefixes=AVX512
8; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f,+avx512bw | FileCheck %s --check-prefixes=AVX512
9; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+avx512f,+avx512bw,+avx512vbmi | FileCheck %s --check-prefixes=AVX512
10;
11; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=slm | FileCheck %s --check-prefixes=SSE,SSE42
12; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=goldmont | FileCheck %s --check-prefixes=SSE,SSE42
13; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mcpu=btver2 | FileCheck %s --check-prefixes=AVX1
14
15;
16; Verify the cost model for broadcast shuffles.
17;
18
19define void @test_vXf64(<2 x double> %src128, <4 x double> %src256, <8 x double> %src512) {
20; SSE-LABEL: 'test_vXf64'
21; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> zeroinitializer
22; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> zeroinitializer
23; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> zeroinitializer
24; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
25;
26; AVX1-LABEL: 'test_vXf64'
27; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> zeroinitializer
28; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> zeroinitializer
29; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> zeroinitializer
30; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
31;
32; AVX2-LABEL: 'test_vXf64'
33; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> zeroinitializer
34; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> zeroinitializer
35; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> zeroinitializer
36; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
37;
38; AVX512-LABEL: 'test_vXf64'
39; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> zeroinitializer
40; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> zeroinitializer
41; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> zeroinitializer
42; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
43;
44  %V128 = shufflevector <2 x double> %src128, <2 x double> undef, <2 x i32> zeroinitializer
45  %V256 = shufflevector <4 x double> %src256, <4 x double> undef, <4 x i32> zeroinitializer
46  %V512 = shufflevector <8 x double> %src512, <8 x double> undef, <8 x i32> zeroinitializer
47  ret void
48}
49
50define void @test_vXi64(<2 x i64> %src128, <4 x i64> %src256, <8 x i64> %src512) {
51; SSE-LABEL: 'test_vXi64'
52; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> zeroinitializer
53; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> zeroinitializer
54; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> zeroinitializer
55; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
56;
57; AVX1-LABEL: 'test_vXi64'
58; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> zeroinitializer
59; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> zeroinitializer
60; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> zeroinitializer
61; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
62;
63; AVX2-LABEL: 'test_vXi64'
64; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> zeroinitializer
65; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> zeroinitializer
66; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> zeroinitializer
67; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
68;
69; AVX512-LABEL: 'test_vXi64'
70; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> zeroinitializer
71; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> zeroinitializer
72; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> zeroinitializer
73; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
74;
75  %V128 = shufflevector <2 x i64> %src128, <2 x i64> undef, <2 x i32> zeroinitializer
76  %V256 = shufflevector <4 x i64> %src256, <4 x i64> undef, <4 x i32> zeroinitializer
77  %V512 = shufflevector <8 x i64> %src512, <8 x i64> undef, <8 x i32> zeroinitializer
78  ret void
79}
80
81define void @test_vXf32(<2 x float> %src64, <4 x float> %src128, <8 x float> %src256, <16 x float> %src512) {
82; SSE-LABEL: 'test_vXf32'
83; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> zeroinitializer
84; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> zeroinitializer
85; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> zeroinitializer
86; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> zeroinitializer
87; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
88;
89; AVX1-LABEL: 'test_vXf32'
90; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> zeroinitializer
91; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> zeroinitializer
92; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> zeroinitializer
93; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> zeroinitializer
94; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
95;
96; AVX2-LABEL: 'test_vXf32'
97; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> zeroinitializer
98; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> zeroinitializer
99; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> zeroinitializer
100; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> zeroinitializer
101; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
102;
103; AVX512-LABEL: 'test_vXf32'
104; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> zeroinitializer
105; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> zeroinitializer
106; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> zeroinitializer
107; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> zeroinitializer
108; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
109;
110  %V64 = shufflevector <2 x float> %src64, <2 x float> undef, <2 x i32> zeroinitializer
111  %V128 = shufflevector <4 x float> %src128, <4 x float> undef, <4 x i32> zeroinitializer
112  %V256 = shufflevector <8 x float> %src256, <8 x float> undef, <8 x i32> zeroinitializer
113  %V512 = shufflevector <16 x float> %src512, <16 x float> undef, <16 x i32> zeroinitializer
114  ret void
115}
116
117define void @test_vXi32(<2 x i32> %src64, <4 x i32> %src128, <8 x i32> %src256, <16 x i32> %src512) {
118; SSE-LABEL: 'test_vXi32'
119; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> zeroinitializer
120; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> zeroinitializer
121; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> zeroinitializer
122; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> zeroinitializer
123; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
124;
125; AVX1-LABEL: 'test_vXi32'
126; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> zeroinitializer
127; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> zeroinitializer
128; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> zeroinitializer
129; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> zeroinitializer
130; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
131;
132; AVX2-LABEL: 'test_vXi32'
133; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> zeroinitializer
134; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> zeroinitializer
135; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> zeroinitializer
136; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> zeroinitializer
137; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
138;
139; AVX512-LABEL: 'test_vXi32'
140; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> zeroinitializer
141; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> zeroinitializer
142; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> zeroinitializer
143; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> zeroinitializer
144; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
145;
146  %V64 = shufflevector <2 x i32> %src64, <2 x i32> undef, <2 x i32> zeroinitializer
147  %V128 = shufflevector <4 x i32> %src128, <4 x i32> undef, <4 x i32> zeroinitializer
148  %V256 = shufflevector <8 x i32> %src256, <8 x i32> undef, <8 x i32> zeroinitializer
149  %V512 = shufflevector <16 x i32> %src512, <16 x i32> undef, <16 x i32> zeroinitializer
150  ret void
151}
152
153define void @test_vXf16(<2 x half> %src32, <4 x half> %src64, <8 x half> %src128, <16 x half> %src256, <32 x half> %src512) {
154; SSE2-LABEL: 'test_vXf16'
155; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> zeroinitializer
156; SSE2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> zeroinitializer
157; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> zeroinitializer
158; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> zeroinitializer
159; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> zeroinitializer
160; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
161;
162; SSSE3-LABEL: 'test_vXf16'
163; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> zeroinitializer
164; SSSE3-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> zeroinitializer
165; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> zeroinitializer
166; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> zeroinitializer
167; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> zeroinitializer
168; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
169;
170; SSE42-LABEL: 'test_vXf16'
171; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> zeroinitializer
172; SSE42-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> zeroinitializer
173; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> zeroinitializer
174; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> zeroinitializer
175; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> zeroinitializer
176; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
177;
178; AVX2-LABEL: 'test_vXf16'
179; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> zeroinitializer
180; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> zeroinitializer
181; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> zeroinitializer
182; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> zeroinitializer
183; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> zeroinitializer
184; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
185;
186; AVX512-LABEL: 'test_vXf16'
187; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> zeroinitializer
188; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> zeroinitializer
189; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> zeroinitializer
190; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> zeroinitializer
191; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> zeroinitializer
192; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
193;
194  %V32  = shufflevector <2 x half> %src32, <2 x half> undef, <2 x i32> zeroinitializer
195  %V64  = shufflevector <4 x half> %src64, <4 x half> undef, <4 x i32> zeroinitializer
196  %V128 = shufflevector <8 x half> %src128, <8 x half> undef, <8 x i32> zeroinitializer
197  %V256 = shufflevector <16 x half> %src256, <16 x half> undef, <16 x i32> zeroinitializer
198  %V512 = shufflevector <32 x half> %src512, <32 x half> undef, <32 x i32> zeroinitializer
199  ret void
200}
201
202define void @test_vXbf16(<2 x bfloat> %src32, <4 x bfloat> %src64, <8 x bfloat> %src128, <16 x bfloat> %src256, <32 x bfloat> %src512) {
203; SSE-LABEL: 'test_vXbf16'
204; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x bfloat> %src32, <2 x bfloat> undef, <2 x i32> zeroinitializer
205; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x bfloat> %src64, <4 x bfloat> undef, <4 x i32> zeroinitializer
206; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128 = shufflevector <8 x bfloat> %src128, <8 x bfloat> undef, <8 x i32> zeroinitializer
207; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256 = shufflevector <16 x bfloat> %src256, <16 x bfloat> undef, <16 x i32> zeroinitializer
208; SSE-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512 = shufflevector <32 x bfloat> %src512, <32 x bfloat> undef, <32 x i32> zeroinitializer
209; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
210;
211; AVX1-LABEL: 'test_vXbf16'
212; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x bfloat> %src32, <2 x bfloat> undef, <2 x i32> zeroinitializer
213; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x bfloat> %src64, <4 x bfloat> undef, <4 x i32> zeroinitializer
214; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128 = shufflevector <8 x bfloat> %src128, <8 x bfloat> undef, <8 x i32> zeroinitializer
215; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256 = shufflevector <16 x bfloat> %src256, <16 x bfloat> undef, <16 x i32> zeroinitializer
216; AVX1-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512 = shufflevector <32 x bfloat> %src512, <32 x bfloat> undef, <32 x i32> zeroinitializer
217; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
218;
219; AVX2-LABEL: 'test_vXbf16'
220; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x bfloat> %src32, <2 x bfloat> undef, <2 x i32> zeroinitializer
221; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x bfloat> %src64, <4 x bfloat> undef, <4 x i32> zeroinitializer
222; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128 = shufflevector <8 x bfloat> %src128, <8 x bfloat> undef, <8 x i32> zeroinitializer
223; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256 = shufflevector <16 x bfloat> %src256, <16 x bfloat> undef, <16 x i32> zeroinitializer
224; AVX2-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512 = shufflevector <32 x bfloat> %src512, <32 x bfloat> undef, <32 x i32> zeroinitializer
225; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
226;
227; AVX512-LABEL: 'test_vXbf16'
228; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V32 = shufflevector <2 x bfloat> %src32, <2 x bfloat> undef, <2 x i32> zeroinitializer
229; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V64 = shufflevector <4 x bfloat> %src64, <4 x bfloat> undef, <4 x i32> zeroinitializer
230; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V128 = shufflevector <8 x bfloat> %src128, <8 x bfloat> undef, <8 x i32> zeroinitializer
231; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V256 = shufflevector <16 x bfloat> %src256, <16 x bfloat> undef, <16 x i32> zeroinitializer
232; AVX512-NEXT:  Cost Model: Found an estimated cost of 0 for instruction: %V512 = shufflevector <32 x bfloat> %src512, <32 x bfloat> undef, <32 x i32> zeroinitializer
233; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
234;
235  %V32  = shufflevector <2 x bfloat> %src32, <2 x bfloat> undef, <2 x i32> zeroinitializer
236  %V64  = shufflevector <4 x bfloat> %src64, <4 x bfloat> undef, <4 x i32> zeroinitializer
237  %V128 = shufflevector <8 x bfloat> %src128, <8 x bfloat> undef, <8 x i32> zeroinitializer
238  %V256 = shufflevector <16 x bfloat> %src256, <16 x bfloat> undef, <16 x i32> zeroinitializer
239  %V512 = shufflevector <32 x bfloat> %src512, <32 x bfloat> undef, <32 x i32> zeroinitializer
240  ret void
241}
242
243define void @test_vXi16(<2 x i16> %src32, <4 x i16> %src64, <8 x i16> %src128, <16 x i16> %src256, <32 x i16> %src512) {
244; SSE2-LABEL: 'test_vXi16'
245; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
246; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
247; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
248; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
249; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
250; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
251;
252; SSSE3-LABEL: 'test_vXi16'
253; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
254; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
255; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
256; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
257; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
258; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
259;
260; SSE42-LABEL: 'test_vXi16'
261; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
262; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
263; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
264; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
265; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
266; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
267;
268; AVX1-LABEL: 'test_vXi16'
269; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
270; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
271; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
272; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
273; AVX1-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
274; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
275;
276; AVX2-LABEL: 'test_vXi16'
277; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
278; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
279; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
280; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
281; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
282; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
283;
284; AVX512-LABEL: 'test_vXi16'
285; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
286; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
287; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
288; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
289; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
290; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
291;
292  %V32  = shufflevector <2 x i16> %src32, <2 x i16> undef, <2 x i32> zeroinitializer
293  %V64  = shufflevector <4 x i16> %src64, <4 x i16> undef, <4 x i32> zeroinitializer
294  %V128 = shufflevector <8 x i16> %src128, <8 x i16> undef, <8 x i32> zeroinitializer
295  %V256 = shufflevector <16 x i16> %src256, <16 x i16> undef, <16 x i32> zeroinitializer
296  %V512 = shufflevector <32 x i16> %src512, <32 x i16> undef, <32 x i32> zeroinitializer
297  ret void
298}
299
300define void @test_vXi8(<2 x i8> %src16, <4 x i8> %src32, <8 x i8> %src64, <16 x i8> %src128, <32 x i8> %src256, <64 x i8> %src512) {
301; SSE2-LABEL: 'test_vXi8'
302; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
303; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
304; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
305; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
306; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
307; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
308; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
309;
310; SSSE3-LABEL: 'test_vXi8'
311; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
312; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
313; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
314; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
315; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
316; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
317; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
318;
319; SSE42-LABEL: 'test_vXi8'
320; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
321; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
322; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
323; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
324; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
325; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
326; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
327;
328; AVX1-LABEL: 'test_vXi8'
329; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
330; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
331; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
332; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
333; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
334; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
335; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
336;
337; AVX2-LABEL: 'test_vXi8'
338; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
339; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
340; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
341; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
342; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
343; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
344; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
345;
346; AVX512-LABEL: 'test_vXi8'
347; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
348; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
349; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
350; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
351; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
352; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
353; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
354;
355  %V16  = shufflevector <2 x i8> %src16, <2 x i8> undef, <2 x i32> zeroinitializer
356  %V32  = shufflevector <4 x i8> %src32, <4 x i8> undef, <4 x i32> zeroinitializer
357  %V64  = shufflevector <8 x i8> %src64, <8 x i8> undef, <8 x i32> zeroinitializer
358  %V128 = shufflevector <16 x i8> %src128, <16 x i8> undef, <16 x i32> zeroinitializer
359  %V256 = shufflevector <32 x i8> %src256, <32 x i8> undef, <32 x i32> zeroinitializer
360  %V512 = shufflevector <64 x i8> %src512, <64 x i8> undef, <64 x i32> zeroinitializer
361  ret void
362}
363
364define void @test_vXi1(<2 x i1> %src2, <4 x i1> %src4, <8 x i1> %src8, <16 x i1> %src16, <32 x i1> %src32, <64 x i1> %src64) {
365; SSE2-LABEL: 'test_vXi1'
366; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2 = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
367; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4 = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
368; SSE2-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V8 = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
369; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
370; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
371; SSE2-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
372; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
373;
374; SSSE3-LABEL: 'test_vXi1'
375; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2 = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
376; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4 = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
377; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8 = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
378; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
379; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
380; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
381; SSSE3-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
382;
383; SSE42-LABEL: 'test_vXi1'
384; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2 = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
385; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4 = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
386; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8 = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
387; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
388; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
389; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
390; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
391;
392; AVX1-LABEL: 'test_vXi1'
393; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2 = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
394; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4 = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
395; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8 = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
396; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
397; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
398; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
399; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
400;
401; AVX2-LABEL: 'test_vXi1'
402; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V2 = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
403; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V4 = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
404; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V8 = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
405; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
406; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
407; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
408; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
409;
410; AVX512-LABEL: 'test_vXi1'
411; AVX512-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %V2 = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
412; AVX512-NEXT:  Cost Model: Found an estimated cost of 5 for instruction: %V4 = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
413; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %V8 = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
414; AVX512-NEXT:  Cost Model: Found an estimated cost of 17 for instruction: %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
415; AVX512-NEXT:  Cost Model: Found an estimated cost of 33 for instruction: %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
416; AVX512-NEXT:  Cost Model: Found an estimated cost of 65 for instruction: %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
417; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
418;
419  %V2  = shufflevector <2 x i1> %src2, <2 x i1> undef, <2 x i32> zeroinitializer
420  %V4  = shufflevector <4 x i1> %src4, <4 x i1> undef, <4 x i32> zeroinitializer
421  %V8  = shufflevector <8 x i1> %src8, <8 x i1> undef, <8 x i32> zeroinitializer
422  %V16 = shufflevector <16 x i1> %src16, <16 x i1> undef, <16 x i32> zeroinitializer
423  %V32 = shufflevector <32 x i1> %src32, <32 x i1> undef, <32 x i32> zeroinitializer
424  %V64 = shufflevector <64 x i1> %src64, <64 x i1> undef, <64 x i32> zeroinitializer
425  ret void
426}
427
428;
429; Tests the cost model for broadcast shuffles of second operand.
430;
431
432define void @test_upper_vXf32(<2 x float> %a64, <2 x float> %b64, <4 x float> %a128, <4 x float> %b128, <8 x float> %a256, <8 x float> %b256, <16 x float> %a512, <16 x float> %b512) {
433; SSE-LABEL: 'test_upper_vXf32'
434; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 2, i32 2>
435; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
436; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
437; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
438; SSE-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
439;
440; AVX1-LABEL: 'test_upper_vXf32'
441; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 2, i32 2>
442; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
443; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
444; AVX1-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
445; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
446;
447; AVX2-LABEL: 'test_upper_vXf32'
448; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 2, i32 2>
449; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
450; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
451; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
452; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
453;
454; AVX512-LABEL: 'test_upper_vXf32'
455; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 2, i32 2>
456; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
457; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
458; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
459; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret void
460;
461  %V64 = shufflevector <2 x float> %a64, <2 x float> %b64, <2 x i32> <i32 2, i32 2>
462  %V128 = shufflevector <4 x float> %a128, <4 x float> %b128, <4 x i32> <i32 4, i32 4, i32 4, i32 4>
463  %V256 = shufflevector <8 x float> %a256, <8 x float> %b256, <8 x i32> <i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8, i32 8>
464  %V512 = shufflevector <16 x float> %a512, <16 x float> %b512, <16 x i32> <i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16, i32 16>
465  ret void
466}
467