xref: /llvm-project/llvm/test/Analysis/CostModel/X86/ctlz-codesize.ll (revision dec47b76f406242dfb9d36da4d7adfb171c71104) 
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=-lzcnt,+sse2 | FileCheck %s -check-prefixes=CHECK,NOLZCNT
3; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+sse2 | FileCheck %s -check-prefixes=CHECK,LZCNT,SSE2
4; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+sse4.2 | FileCheck %s -check-prefixes=CHECK,LZCNT,SSE42
5; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+avx | FileCheck %s -check-prefixes=CHECK,LZCNT,AVX,AVX1
6; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+avx2 | FileCheck %s -check-prefixes=CHECK,LZCNT,AVX,AVX2
7; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+avx512f | FileCheck %s -check-prefixes=CHECK,LZCNT,AVX512,AVX512F
8; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+avx512vl,+avx512bw,+avx512dq | FileCheck %s -check-prefixes=CHECK,LZCNT,AVX512,AVX512BW
9; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -cost-kind=code-size -mattr=+lzcnt,+avx512vl,+avx512bw,+avx512dq,+avx512cd | FileCheck %s -check-prefixes=CHECK,LZCNT,AVX512CD
10
11; Verify the cost of scalar leading zero count instructions.
12
13declare i64 @llvm.ctlz.i64(i64, i1)
14declare i32 @llvm.ctlz.i32(i32, i1)
15declare i16 @llvm.ctlz.i16(i16, i1)
16declare  i8 @llvm.ctlz.i8(i8, i1)
17
18define i64 @var_ctlz_i64(i64 %a) {
19; NOLZCNT-LABEL: 'var_ctlz_i64'
20; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %ctlz = call i64 @llvm.ctlz.i64(i64 %a, i1 false)
21; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i64 %ctlz
22;
23; LZCNT-LABEL: 'var_ctlz_i64'
24; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i64 @llvm.ctlz.i64(i64 %a, i1 false)
25; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i64 %ctlz
26;
27  %ctlz = call i64 @llvm.ctlz.i64(i64 %a, i1 0)
28  ret i64 %ctlz
29}
30
31define i64 @var_ctlz_i64u(i64 %a) {
32; NOLZCNT-LABEL: 'var_ctlz_i64u'
33; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %ctlz = call i64 @llvm.ctlz.i64(i64 %a, i1 true)
34; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i64 %ctlz
35;
36; LZCNT-LABEL: 'var_ctlz_i64u'
37; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i64 @llvm.ctlz.i64(i64 %a, i1 true)
38; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i64 %ctlz
39;
40  %ctlz = call i64 @llvm.ctlz.i64(i64 %a, i1 1)
41  ret i64 %ctlz
42}
43
44define i32 @var_ctlz_i32(i32 %a) {
45; NOLZCNT-LABEL: 'var_ctlz_i32'
46; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %ctlz = call i32 @llvm.ctlz.i32(i32 %a, i1 false)
47; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %ctlz
48;
49; LZCNT-LABEL: 'var_ctlz_i32'
50; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i32 @llvm.ctlz.i32(i32 %a, i1 false)
51; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %ctlz
52;
53  %ctlz = call i32 @llvm.ctlz.i32(i32 %a, i1 0)
54  ret i32 %ctlz
55}
56
57define i32 @var_ctlz_i32u(i32 %a) {
58; NOLZCNT-LABEL: 'var_ctlz_i32u'
59; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %ctlz = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
60; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %ctlz
61;
62; LZCNT-LABEL: 'var_ctlz_i32u'
63; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i32 @llvm.ctlz.i32(i32 %a, i1 true)
64; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i32 %ctlz
65;
66  %ctlz = call i32 @llvm.ctlz.i32(i32 %a, i1 1)
67  ret i32 %ctlz
68}
69
70define i16 @var_ctlz_i16(i16 %a) {
71; NOLZCNT-LABEL: 'var_ctlz_i16'
72; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %ctlz = call i16 @llvm.ctlz.i16(i16 %a, i1 false)
73; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i16 %ctlz
74;
75; LZCNT-LABEL: 'var_ctlz_i16'
76; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i16 @llvm.ctlz.i16(i16 %a, i1 false)
77; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i16 %ctlz
78;
79  %ctlz = call i16 @llvm.ctlz.i16(i16 %a, i1 0)
80  ret i16 %ctlz
81}
82
83define i16 @var_ctlz_i16u(i16 %a) {
84; NOLZCNT-LABEL: 'var_ctlz_i16u'
85; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 2 for instruction: %ctlz = call i16 @llvm.ctlz.i16(i16 %a, i1 true)
86; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i16 %ctlz
87;
88; LZCNT-LABEL: 'var_ctlz_i16u'
89; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i16 @llvm.ctlz.i16(i16 %a, i1 true)
90; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i16 %ctlz
91;
92  %ctlz = call i16 @llvm.ctlz.i16(i16 %a, i1 1)
93  ret i16 %ctlz
94}
95
96define i8 @var_ctlz_i8(i8 %a) {
97; NOLZCNT-LABEL: 'var_ctlz_i8'
98; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %ctlz = call i8 @llvm.ctlz.i8(i8 %a, i1 false)
99; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i8 %ctlz
100;
101; LZCNT-LABEL: 'var_ctlz_i8'
102; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i8 @llvm.ctlz.i8(i8 %a, i1 false)
103; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i8 %ctlz
104;
105  %ctlz = call i8 @llvm.ctlz.i8(i8 %a, i1 0)
106  ret i8 %ctlz
107}
108
109define i8 @var_ctlz_i8u(i8 %a) {
110; NOLZCNT-LABEL: 'var_ctlz_i8u'
111; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 3 for instruction: %ctlz = call i8 @llvm.ctlz.i8(i8 %a, i1 true)
112; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i8 %ctlz
113;
114; LZCNT-LABEL: 'var_ctlz_i8u'
115; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call i8 @llvm.ctlz.i8(i8 %a, i1 true)
116; LZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret i8 %ctlz
117;
118  %ctlz = call i8 @llvm.ctlz.i8(i8 %a, i1 1)
119  ret i8 %ctlz
120}
121
122; Verify the cost of vector leading zero count instructions.
123
124declare <2 x i64> @llvm.ctlz.v2i64(<2 x i64>, i1)
125declare <4 x i32> @llvm.ctlz.v4i32(<4 x i32>, i1)
126declare <8 x i16> @llvm.ctlz.v8i16(<8 x i16>, i1)
127declare <16 x i8> @llvm.ctlz.v16i8(<16 x i8>, i1)
128
129declare <4 x i64> @llvm.ctlz.v4i64(<4 x i64>, i1)
130declare <8 x i32> @llvm.ctlz.v8i32(<8 x i32>, i1)
131declare <16 x i16> @llvm.ctlz.v16i16(<16 x i16>, i1)
132declare <32 x i8> @llvm.ctlz.v32i8(<32 x i8>, i1)
133
134declare <8 x i64> @llvm.ctlz.v8i64(<8 x i64>, i1)
135declare <16 x i32> @llvm.ctlz.v16i32(<16 x i32>, i1)
136declare <32 x i16> @llvm.ctlz.v32i16(<32 x i16>, i1)
137declare <64 x i8> @llvm.ctlz.v64i8(<64 x i8>, i1)
138
139define <2 x i64> @var_ctlz_v2i64(<2 x i64> %a) {
140; NOLZCNT-LABEL: 'var_ctlz_v2i64'
141; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 false)
142; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
143;
144; SSE2-LABEL: 'var_ctlz_v2i64'
145; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 false)
146; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
147;
148; SSE42-LABEL: 'var_ctlz_v2i64'
149; SSE42-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 false)
150; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
151;
152; AVX-LABEL: 'var_ctlz_v2i64'
153; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 false)
154; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
155;
156; AVX512-LABEL: 'var_ctlz_v2i64'
157; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 false)
158; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
159;
160; AVX512CD-LABEL: 'var_ctlz_v2i64'
161; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 false)
162; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
163;
164  %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 0)
165  ret <2 x i64> %ctlz
166}
167
168define <2 x i64> @var_ctlz_v2i64u(<2 x i64> %a) {
169; NOLZCNT-LABEL: 'var_ctlz_v2i64u'
170; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
171; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
172;
173; SSE2-LABEL: 'var_ctlz_v2i64u'
174; SSE2-NEXT:  Cost Model: Found an estimated cost of 36 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
175; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
176;
177; SSE42-LABEL: 'var_ctlz_v2i64u'
178; SSE42-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
179; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
180;
181; AVX-LABEL: 'var_ctlz_v2i64u'
182; AVX-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
183; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
184;
185; AVX512-LABEL: 'var_ctlz_v2i64u'
186; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
187; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
188;
189; AVX512CD-LABEL: 'var_ctlz_v2i64u'
190; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 true)
191; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <2 x i64> %ctlz
192;
193  %ctlz = call <2 x i64> @llvm.ctlz.v2i64(<2 x i64> %a, i1 1)
194  ret <2 x i64> %ctlz
195}
196
197define <4 x i64> @var_ctlz_v4i64(<4 x i64> %a) {
198; NOLZCNT-LABEL: 'var_ctlz_v4i64'
199; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
200; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
201;
202; SSE2-LABEL: 'var_ctlz_v4i64'
203; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
204; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
205;
206; SSE42-LABEL: 'var_ctlz_v4i64'
207; SSE42-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
208; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
209;
210; AVX1-LABEL: 'var_ctlz_v4i64'
211; AVX1-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
212; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
213;
214; AVX2-LABEL: 'var_ctlz_v4i64'
215; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
216; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
217;
218; AVX512-LABEL: 'var_ctlz_v4i64'
219; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
220; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
221;
222; AVX512CD-LABEL: 'var_ctlz_v4i64'
223; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 false)
224; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
225;
226  %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 0)
227  ret <4 x i64> %ctlz
228}
229
230define <4 x i64> @var_ctlz_v4i64u(<4 x i64> %a) {
231; NOLZCNT-LABEL: 'var_ctlz_v4i64u'
232; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
233; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
234;
235; SSE2-LABEL: 'var_ctlz_v4i64u'
236; SSE2-NEXT:  Cost Model: Found an estimated cost of 72 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
237; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
238;
239; SSE42-LABEL: 'var_ctlz_v4i64u'
240; SSE42-NEXT:  Cost Model: Found an estimated cost of 56 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
241; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
242;
243; AVX1-LABEL: 'var_ctlz_v4i64u'
244; AVX1-NEXT:  Cost Model: Found an estimated cost of 49 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
245; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
246;
247; AVX2-LABEL: 'var_ctlz_v4i64u'
248; AVX2-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
249; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
250;
251; AVX512-LABEL: 'var_ctlz_v4i64u'
252; AVX512-NEXT:  Cost Model: Found an estimated cost of 24 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
253; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
254;
255; AVX512CD-LABEL: 'var_ctlz_v4i64u'
256; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 true)
257; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i64> %ctlz
258;
259  %ctlz = call <4 x i64> @llvm.ctlz.v4i64(<4 x i64> %a, i1 1)
260  ret <4 x i64> %ctlz
261}
262
263define <8 x i64> @var_ctlz_v8i64(<8 x i64> %a) {
264; NOLZCNT-LABEL: 'var_ctlz_v8i64'
265; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
266; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
267;
268; SSE2-LABEL: 'var_ctlz_v8i64'
269; SSE2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
270; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
271;
272; SSE42-LABEL: 'var_ctlz_v8i64'
273; SSE42-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
274; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
275;
276; AVX1-LABEL: 'var_ctlz_v8i64'
277; AVX1-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
278; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
279;
280; AVX2-LABEL: 'var_ctlz_v8i64'
281; AVX2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
282; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
283;
284; AVX512F-LABEL: 'var_ctlz_v8i64'
285; AVX512F-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
286; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
287;
288; AVX512BW-LABEL: 'var_ctlz_v8i64'
289; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
290; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
291;
292; AVX512CD-LABEL: 'var_ctlz_v8i64'
293; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 false)
294; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
295;
296  %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 0)
297  ret <8 x i64> %ctlz
298}
299
300define <8 x i64> @var_ctlz_v8i64u(<8 x i64> %a) {
301; NOLZCNT-LABEL: 'var_ctlz_v8i64u'
302; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
303; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
304;
305; SSE2-LABEL: 'var_ctlz_v8i64u'
306; SSE2-NEXT:  Cost Model: Found an estimated cost of 144 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
307; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
308;
309; SSE42-LABEL: 'var_ctlz_v8i64u'
310; SSE42-NEXT:  Cost Model: Found an estimated cost of 112 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
311; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
312;
313; AVX1-LABEL: 'var_ctlz_v8i64u'
314; AVX1-NEXT:  Cost Model: Found an estimated cost of 98 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
315; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
316;
317; AVX2-LABEL: 'var_ctlz_v8i64u'
318; AVX2-NEXT:  Cost Model: Found an estimated cost of 48 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
319; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
320;
321; AVX512F-LABEL: 'var_ctlz_v8i64u'
322; AVX512F-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
323; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
324;
325; AVX512BW-LABEL: 'var_ctlz_v8i64u'
326; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
327; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
328;
329; AVX512CD-LABEL: 'var_ctlz_v8i64u'
330; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 true)
331; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i64> %ctlz
332;
333  %ctlz = call <8 x i64> @llvm.ctlz.v8i64(<8 x i64> %a, i1 1)
334  ret <8 x i64> %ctlz
335}
336
337define <4 x i32> @var_ctlz_v4i32(<4 x i32> %a) {
338; NOLZCNT-LABEL: 'var_ctlz_v4i32'
339; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 false)
340; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
341;
342; SSE2-LABEL: 'var_ctlz_v4i32'
343; SSE2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 false)
344; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
345;
346; SSE42-LABEL: 'var_ctlz_v4i32'
347; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 false)
348; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
349;
350; AVX-LABEL: 'var_ctlz_v4i32'
351; AVX-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 false)
352; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
353;
354; AVX512-LABEL: 'var_ctlz_v4i32'
355; AVX512-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 false)
356; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
357;
358; AVX512CD-LABEL: 'var_ctlz_v4i32'
359; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 false)
360; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
361;
362  %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 0)
363  ret <4 x i32> %ctlz
364}
365
366define <4 x i32> @var_ctlz_v4i32u(<4 x i32> %a) {
367; NOLZCNT-LABEL: 'var_ctlz_v4i32u'
368; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 true)
369; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
370;
371; SSE2-LABEL: 'var_ctlz_v4i32u'
372; SSE2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 true)
373; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
374;
375; SSE42-LABEL: 'var_ctlz_v4i32u'
376; SSE42-NEXT:  Cost Model: Found an estimated cost of 22 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 true)
377; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
378;
379; AVX-LABEL: 'var_ctlz_v4i32u'
380; AVX-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 true)
381; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
382;
383; AVX512-LABEL: 'var_ctlz_v4i32u'
384; AVX512-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 true)
385; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
386;
387; AVX512CD-LABEL: 'var_ctlz_v4i32u'
388; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 true)
389; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <4 x i32> %ctlz
390;
391  %ctlz = call <4 x i32> @llvm.ctlz.v4i32(<4 x i32> %a, i1 1)
392  ret <4 x i32> %ctlz
393}
394
395define <8 x i32> @var_ctlz_v8i32(<8 x i32> %a) {
396; NOLZCNT-LABEL: 'var_ctlz_v8i32'
397; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
398; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
399;
400; SSE2-LABEL: 'var_ctlz_v8i32'
401; SSE2-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
402; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
403;
404; SSE42-LABEL: 'var_ctlz_v8i32'
405; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
406; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
407;
408; AVX1-LABEL: 'var_ctlz_v8i32'
409; AVX1-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
410; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
411;
412; AVX2-LABEL: 'var_ctlz_v8i32'
413; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
414; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
415;
416; AVX512-LABEL: 'var_ctlz_v8i32'
417; AVX512-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
418; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
419;
420; AVX512CD-LABEL: 'var_ctlz_v8i32'
421; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 false)
422; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
423;
424  %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 0)
425  ret <8 x i32> %ctlz
426}
427
428define <8 x i32> @var_ctlz_v8i32u(<8 x i32> %a) {
429; NOLZCNT-LABEL: 'var_ctlz_v8i32u'
430; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
431; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
432;
433; SSE2-LABEL: 'var_ctlz_v8i32u'
434; SSE2-NEXT:  Cost Model: Found an estimated cost of 76 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
435; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
436;
437; SSE42-LABEL: 'var_ctlz_v8i32u'
438; SSE42-NEXT:  Cost Model: Found an estimated cost of 44 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
439; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
440;
441; AVX1-LABEL: 'var_ctlz_v8i32u'
442; AVX1-NEXT:  Cost Model: Found an estimated cost of 39 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
443; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
444;
445; AVX2-LABEL: 'var_ctlz_v8i32u'
446; AVX2-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
447; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
448;
449; AVX512-LABEL: 'var_ctlz_v8i32u'
450; AVX512-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
451; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
452;
453; AVX512CD-LABEL: 'var_ctlz_v8i32u'
454; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 true)
455; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i32> %ctlz
456;
457  %ctlz = call <8 x i32> @llvm.ctlz.v8i32(<8 x i32> %a, i1 1)
458  ret <8 x i32> %ctlz
459}
460
461define <16 x i32> @var_ctlz_v16i32(<16 x i32> %a) {
462; NOLZCNT-LABEL: 'var_ctlz_v16i32'
463; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
464; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
465;
466; SSE2-LABEL: 'var_ctlz_v16i32'
467; SSE2-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
468; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
469;
470; SSE42-LABEL: 'var_ctlz_v16i32'
471; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
472; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
473;
474; AVX1-LABEL: 'var_ctlz_v16i32'
475; AVX1-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
476; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
477;
478; AVX2-LABEL: 'var_ctlz_v16i32'
479; AVX2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
480; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
481;
482; AVX512F-LABEL: 'var_ctlz_v16i32'
483; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
484; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
485;
486; AVX512BW-LABEL: 'var_ctlz_v16i32'
487; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
488; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
489;
490; AVX512CD-LABEL: 'var_ctlz_v16i32'
491; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 false)
492; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
493;
494  %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 0)
495  ret <16 x i32> %ctlz
496}
497
498define <16 x i32> @var_ctlz_v16i32u(<16 x i32> %a) {
499; NOLZCNT-LABEL: 'var_ctlz_v16i32u'
500; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
501; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
502;
503; SSE2-LABEL: 'var_ctlz_v16i32u'
504; SSE2-NEXT:  Cost Model: Found an estimated cost of 152 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
505; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
506;
507; SSE42-LABEL: 'var_ctlz_v16i32u'
508; SSE42-NEXT:  Cost Model: Found an estimated cost of 88 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
509; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
510;
511; AVX1-LABEL: 'var_ctlz_v16i32u'
512; AVX1-NEXT:  Cost Model: Found an estimated cost of 78 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
513; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
514;
515; AVX2-LABEL: 'var_ctlz_v16i32u'
516; AVX2-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
517; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
518;
519; AVX512F-LABEL: 'var_ctlz_v16i32u'
520; AVX512F-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
521; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
522;
523; AVX512BW-LABEL: 'var_ctlz_v16i32u'
524; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 25 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
525; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
526;
527; AVX512CD-LABEL: 'var_ctlz_v16i32u'
528; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 true)
529; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i32> %ctlz
530;
531  %ctlz = call <16 x i32> @llvm.ctlz.v16i32(<16 x i32> %a, i1 1)
532  ret <16 x i32> %ctlz
533}
534
535define <8 x i16> @var_ctlz_v8i16(<8 x i16> %a) {
536; NOLZCNT-LABEL: 'var_ctlz_v8i16'
537; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 false)
538; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
539;
540; SSE2-LABEL: 'var_ctlz_v8i16'
541; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 false)
542; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
543;
544; SSE42-LABEL: 'var_ctlz_v8i16'
545; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 false)
546; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
547;
548; AVX-LABEL: 'var_ctlz_v8i16'
549; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 false)
550; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
551;
552; AVX512-LABEL: 'var_ctlz_v8i16'
553; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 false)
554; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
555;
556; AVX512CD-LABEL: 'var_ctlz_v8i16'
557; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 false)
558; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
559;
560  %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 0)
561  ret <8 x i16> %ctlz
562}
563
564define <8 x i16> @var_ctlz_v8i16u(<8 x i16> %a) {
565; NOLZCNT-LABEL: 'var_ctlz_v8i16u'
566; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 true)
567; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
568;
569; SSE2-LABEL: 'var_ctlz_v8i16u'
570; SSE2-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 true)
571; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
572;
573; SSE42-LABEL: 'var_ctlz_v8i16u'
574; SSE42-NEXT:  Cost Model: Found an estimated cost of 16 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 true)
575; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
576;
577; AVX-LABEL: 'var_ctlz_v8i16u'
578; AVX-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 true)
579; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
580;
581; AVX512-LABEL: 'var_ctlz_v8i16u'
582; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 true)
583; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
584;
585; AVX512CD-LABEL: 'var_ctlz_v8i16u'
586; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 4 for instruction: %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 true)
587; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <8 x i16> %ctlz
588;
589  %ctlz = call <8 x i16> @llvm.ctlz.v8i16(<8 x i16> %a, i1 1)
590  ret <8 x i16> %ctlz
591}
592
593define <16 x i16> @var_ctlz_v16i16(<16 x i16> %a) {
594; NOLZCNT-LABEL: 'var_ctlz_v16i16'
595; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
596; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
597;
598; SSE2-LABEL: 'var_ctlz_v16i16'
599; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
600; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
601;
602; SSE42-LABEL: 'var_ctlz_v16i16'
603; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
604; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
605;
606; AVX1-LABEL: 'var_ctlz_v16i16'
607; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
608; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
609;
610; AVX2-LABEL: 'var_ctlz_v16i16'
611; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
612; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
613;
614; AVX512-LABEL: 'var_ctlz_v16i16'
615; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
616; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
617;
618; AVX512CD-LABEL: 'var_ctlz_v16i16'
619; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 false)
620; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
621;
622  %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 0)
623  ret <16 x i16> %ctlz
624}
625
626define <16 x i16> @var_ctlz_v16i16u(<16 x i16> %a) {
627; NOLZCNT-LABEL: 'var_ctlz_v16i16u'
628; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
629; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
630;
631; SSE2-LABEL: 'var_ctlz_v16i16u'
632; SSE2-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
633; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
634;
635; SSE42-LABEL: 'var_ctlz_v16i16u'
636; SSE42-NEXT:  Cost Model: Found an estimated cost of 32 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
637; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
638;
639; AVX1-LABEL: 'var_ctlz_v16i16u'
640; AVX1-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
641; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
642;
643; AVX2-LABEL: 'var_ctlz_v16i16u'
644; AVX2-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
645; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
646;
647; AVX512-LABEL: 'var_ctlz_v16i16u'
648; AVX512-NEXT:  Cost Model: Found an estimated cost of 14 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
649; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
650;
651; AVX512CD-LABEL: 'var_ctlz_v16i16u'
652; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 11 for instruction: %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 true)
653; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i16> %ctlz
654;
655  %ctlz = call <16 x i16> @llvm.ctlz.v16i16(<16 x i16> %a, i1 1)
656  ret <16 x i16> %ctlz
657}
658
659define <32 x i16> @var_ctlz_v32i16(<32 x i16> %a) {
660; NOLZCNT-LABEL: 'var_ctlz_v32i16'
661; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
662; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
663;
664; SSE2-LABEL: 'var_ctlz_v32i16'
665; SSE2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
666; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
667;
668; SSE42-LABEL: 'var_ctlz_v32i16'
669; SSE42-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
670; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
671;
672; AVX1-LABEL: 'var_ctlz_v32i16'
673; AVX1-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
674; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
675;
676; AVX2-LABEL: 'var_ctlz_v32i16'
677; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
678; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
679;
680; AVX512F-LABEL: 'var_ctlz_v32i16'
681; AVX512F-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
682; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
683;
684; AVX512BW-LABEL: 'var_ctlz_v32i16'
685; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
686; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
687;
688; AVX512CD-LABEL: 'var_ctlz_v32i16'
689; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 false)
690; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
691;
692  %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 0)
693  ret <32 x i16> %ctlz
694}
695
696define <32 x i16> @var_ctlz_v32i16u(<32 x i16> %a) {
697; NOLZCNT-LABEL: 'var_ctlz_v32i16u'
698; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
699; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
700;
701; SSE2-LABEL: 'var_ctlz_v32i16u'
702; SSE2-NEXT:  Cost Model: Found an estimated cost of 128 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
703; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
704;
705; SSE42-LABEL: 'var_ctlz_v32i16u'
706; SSE42-NEXT:  Cost Model: Found an estimated cost of 64 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
707; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
708;
709; AVX1-LABEL: 'var_ctlz_v32i16u'
710; AVX1-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
711; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
712;
713; AVX2-LABEL: 'var_ctlz_v32i16u'
714; AVX2-NEXT:  Cost Model: Found an estimated cost of 28 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
715; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
716;
717; AVX512F-LABEL: 'var_ctlz_v32i16u'
718; AVX512F-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
719; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
720;
721; AVX512BW-LABEL: 'var_ctlz_v32i16u'
722; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 15 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
723; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
724;
725; AVX512CD-LABEL: 'var_ctlz_v32i16u'
726; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 23 for instruction: %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 true)
727; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i16> %ctlz
728;
729  %ctlz = call <32 x i16> @llvm.ctlz.v32i16(<32 x i16> %a, i1 1)
730  ret <32 x i16> %ctlz
731}
732
733define <16 x i8> @var_ctlz_v16i8(<16 x i8> %a) {
734; NOLZCNT-LABEL: 'var_ctlz_v16i8'
735; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 false)
736; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
737;
738; SSE2-LABEL: 'var_ctlz_v16i8'
739; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 false)
740; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
741;
742; SSE42-LABEL: 'var_ctlz_v16i8'
743; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 false)
744; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
745;
746; AVX-LABEL: 'var_ctlz_v16i8'
747; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 false)
748; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
749;
750; AVX512-LABEL: 'var_ctlz_v16i8'
751; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 false)
752; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
753;
754; AVX512CD-LABEL: 'var_ctlz_v16i8'
755; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 false)
756; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
757;
758  %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 0)
759  ret <16 x i8> %ctlz
760}
761
762define <16 x i8> @var_ctlz_v16i8u(<16 x i8> %a) {
763; NOLZCNT-LABEL: 'var_ctlz_v16i8u'
764; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 true)
765; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
766;
767; SSE2-LABEL: 'var_ctlz_v16i8u'
768; SSE2-NEXT:  Cost Model: Found an estimated cost of 29 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 true)
769; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
770;
771; SSE42-LABEL: 'var_ctlz_v16i8u'
772; SSE42-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 true)
773; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
774;
775; AVX-LABEL: 'var_ctlz_v16i8u'
776; AVX-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 true)
777; AVX-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
778;
779; AVX512-LABEL: 'var_ctlz_v16i8u'
780; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 true)
781; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
782;
783; AVX512CD-LABEL: 'var_ctlz_v16i8u'
784; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 true)
785; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <16 x i8> %ctlz
786;
787  %ctlz = call <16 x i8> @llvm.ctlz.v16i8(<16 x i8> %a, i1 1)
788  ret <16 x i8> %ctlz
789}
790
791define <32 x i8> @var_ctlz_v32i8(<32 x i8> %a) {
792; NOLZCNT-LABEL: 'var_ctlz_v32i8'
793; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
794; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
795;
796; SSE2-LABEL: 'var_ctlz_v32i8'
797; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
798; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
799;
800; SSE42-LABEL: 'var_ctlz_v32i8'
801; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
802; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
803;
804; AVX1-LABEL: 'var_ctlz_v32i8'
805; AVX1-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
806; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
807;
808; AVX2-LABEL: 'var_ctlz_v32i8'
809; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
810; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
811;
812; AVX512-LABEL: 'var_ctlz_v32i8'
813; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
814; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
815;
816; AVX512CD-LABEL: 'var_ctlz_v32i8'
817; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 false)
818; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
819;
820  %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 0)
821  ret <32 x i8> %ctlz
822}
823
824define <32 x i8> @var_ctlz_v32i8u(<32 x i8> %a) {
825; NOLZCNT-LABEL: 'var_ctlz_v32i8u'
826; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
827; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
828;
829; SSE2-LABEL: 'var_ctlz_v32i8u'
830; SSE2-NEXT:  Cost Model: Found an estimated cost of 58 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
831; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
832;
833; SSE42-LABEL: 'var_ctlz_v32i8u'
834; SSE42-NEXT:  Cost Model: Found an estimated cost of 20 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
835; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
836;
837; AVX1-LABEL: 'var_ctlz_v32i8u'
838; AVX1-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
839; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
840;
841; AVX2-LABEL: 'var_ctlz_v32i8u'
842; AVX2-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
843; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
844;
845; AVX512-LABEL: 'var_ctlz_v32i8u'
846; AVX512-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
847; AVX512-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
848;
849; AVX512CD-LABEL: 'var_ctlz_v32i8u'
850; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 true)
851; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <32 x i8> %ctlz
852;
853  %ctlz = call <32 x i8> @llvm.ctlz.v32i8(<32 x i8> %a, i1 1)
854  ret <32 x i8> %ctlz
855}
856
857define <64 x i8> @var_ctlz_v64i8(<64 x i8> %a) {
858; NOLZCNT-LABEL: 'var_ctlz_v64i8'
859; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
860; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
861;
862; SSE2-LABEL: 'var_ctlz_v64i8'
863; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
864; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
865;
866; SSE42-LABEL: 'var_ctlz_v64i8'
867; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
868; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
869;
870; AVX1-LABEL: 'var_ctlz_v64i8'
871; AVX1-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
872; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
873;
874; AVX2-LABEL: 'var_ctlz_v64i8'
875; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
876; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
877;
878; AVX512F-LABEL: 'var_ctlz_v64i8'
879; AVX512F-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
880; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
881;
882; AVX512BW-LABEL: 'var_ctlz_v64i8'
883; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
884; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
885;
886; AVX512CD-LABEL: 'var_ctlz_v64i8'
887; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 false)
888; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
889;
890  %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 0)
891  ret <64 x i8> %ctlz
892}
893
894define <64 x i8> @var_ctlz_v64i8u(<64 x i8> %a) {
895; NOLZCNT-LABEL: 'var_ctlz_v64i8u'
896; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
897; NOLZCNT-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
898;
899; SSE2-LABEL: 'var_ctlz_v64i8u'
900; SSE2-NEXT:  Cost Model: Found an estimated cost of 116 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
901; SSE2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
902;
903; SSE42-LABEL: 'var_ctlz_v64i8u'
904; SSE42-NEXT:  Cost Model: Found an estimated cost of 40 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
905; SSE42-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
906;
907; AVX1-LABEL: 'var_ctlz_v64i8u'
908; AVX1-NEXT:  Cost Model: Found an estimated cost of 38 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
909; AVX1-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
910;
911; AVX2-LABEL: 'var_ctlz_v64i8u'
912; AVX2-NEXT:  Cost Model: Found an estimated cost of 18 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
913; AVX2-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
914;
915; AVX512F-LABEL: 'var_ctlz_v64i8u'
916; AVX512F-NEXT:  Cost Model: Found an estimated cost of 19 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
917; AVX512F-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
918;
919; AVX512BW-LABEL: 'var_ctlz_v64i8u'
920; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 10 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
921; AVX512BW-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
922;
923; AVX512CD-LABEL: 'var_ctlz_v64i8u'
924; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 9 for instruction: %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 true)
925; AVX512CD-NEXT:  Cost Model: Found an estimated cost of 1 for instruction: ret <64 x i8> %ctlz
926;
927  %ctlz = call <64 x i8> @llvm.ctlz.v64i8(<64 x i8> %a, i1 1)
928  ret <64 x i8> %ctlz
929}
930;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
931; CHECK: {{.*}}
932