| blob | e444bd0f8f212f590cd2a09730f0e30b2fc833cd |
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 -mattr=-bmi,+sse2 | FileCheck %s -check-prefixes=NOBMI
3 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+sse2 | FileCheck %s -check-prefixes=BMI,SSE,SSE2
4 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+sse4.2 | FileCheck %s -check-prefixes=BMI,SSE,SSE42
5 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+avx | FileCheck %s -check-prefixes=BMI,AVX1
6 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+avx2 | FileCheck %s -check-prefixes=BMI,AVX2
7 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+avx512f | FileCheck %s -check-prefixes=BMI,AVX512,AVX512F
8 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+avx512vl,+avx512bw,+avx512dq | FileCheck %s -check-prefixes=BMI,AVX512,AVX512BW
9 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+avx512vl,+avx512vpopcntdq | FileCheck %s -check-prefixes=BMI,AVX512,AVX512VPOPCNT
10 ; RUN: opt < %s -mtriple=x86_64-unknown-linux-gnu -passes="print<cost-model>" 2>&1 -disable-output -mattr=+bmi,+avx512vl,+avx512bitalg | FileCheck %s -check-prefixes=BMI,AVX512,AVX512BITALG
12 ; Verify the cost of scalar trailing zero count instructions.
14 declare i64 @llvm.cttz.i64(i64, i1)
15 declare i32 @llvm.cttz.i32(i32, i1)
16 declare i16 @llvm.cttz.i16(i16, i1)
17 declare i8 @llvm.cttz.i8(i8, i1)
19 define i64 @var_cttz_i64(i64 %a) {
20 ; NOBMI-LABEL: 'var_cttz_i64'
21 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call i64 @llvm.cttz.i64(i64 %a, i1 false)
22 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %cttz
23 ;
24 ; BMI-LABEL: 'var_cttz_i64'
25 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i64 @llvm.cttz.i64(i64 %a, i1 false)
26 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %cttz
27 ;
28 %cttz = call i64 @llvm.cttz.i64(i64 %a, i1 0)
29 ret i64 %cttz
30 }
32 define i64 @var_cttz_i64u(i64 %a) {
33 ; NOBMI-LABEL: 'var_cttz_i64u'
34 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i64 @llvm.cttz.i64(i64 %a, i1 true)
35 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %cttz
36 ;
37 ; BMI-LABEL: 'var_cttz_i64u'
38 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i64 @llvm.cttz.i64(i64 %a, i1 true)
39 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i64 %cttz
40 ;
41 %cttz = call i64 @llvm.cttz.i64(i64 %a, i1 1)
42 ret i64 %cttz
43 }
45 define i32 @var_cttz_i32(i32 %a) {
46 ; NOBMI-LABEL: 'var_cttz_i32'
47 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call i32 @llvm.cttz.i32(i32 %a, i1 false)
48 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %cttz
49 ;
50 ; BMI-LABEL: 'var_cttz_i32'
51 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i32 @llvm.cttz.i32(i32 %a, i1 false)
52 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %cttz
53 ;
54 %cttz = call i32 @llvm.cttz.i32(i32 %a, i1 0)
55 ret i32 %cttz
56 }
58 define i32 @var_cttz_i32u(i32 %a) {
59 ; NOBMI-LABEL: 'var_cttz_i32u'
60 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i32 @llvm.cttz.i32(i32 %a, i1 true)
61 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %cttz
62 ;
63 ; BMI-LABEL: 'var_cttz_i32u'
64 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i32 @llvm.cttz.i32(i32 %a, i1 true)
65 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i32 %cttz
66 ;
67 %cttz = call i32 @llvm.cttz.i32(i32 %a, i1 1)
68 ret i32 %cttz
69 }
71 define i16 @var_cttz_i16(i16 %a) {
72 ; NOBMI-LABEL: 'var_cttz_i16'
73 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call i16 @llvm.cttz.i16(i16 %a, i1 false)
74 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %cttz
75 ;
76 ; BMI-LABEL: 'var_cttz_i16'
77 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i16 @llvm.cttz.i16(i16 %a, i1 false)
78 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %cttz
79 ;
80 %cttz = call i16 @llvm.cttz.i16(i16 %a, i1 0)
81 ret i16 %cttz
82 }
84 define i16 @var_cttz_i16u(i16 %a) {
85 ; NOBMI-LABEL: 'var_cttz_i16u'
86 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call i16 @llvm.cttz.i16(i16 %a, i1 true)
87 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %cttz
88 ;
89 ; BMI-LABEL: 'var_cttz_i16u'
90 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i16 @llvm.cttz.i16(i16 %a, i1 true)
91 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i16 %cttz
92 ;
93 %cttz = call i16 @llvm.cttz.i16(i16 %a, i1 1)
94 ret i16 %cttz
95 }
97 define i8 @var_cttz_i8(i8 %a) {
98 ; NOBMI-LABEL: 'var_cttz_i8'
99 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call i8 @llvm.cttz.i8(i8 %a, i1 false)
100 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i8 %cttz
101 ;
102 ; BMI-LABEL: 'var_cttz_i8'
103 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i8 @llvm.cttz.i8(i8 %a, i1 false)
104 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i8 %cttz
105 ;
106 %cttz = call i8 @llvm.cttz.i8(i8 %a, i1 0)
107 ret i8 %cttz
108 }
110 define i8 @var_cttz_i8u(i8 %a) {
111 ; NOBMI-LABEL: 'var_cttz_i8u'
112 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call i8 @llvm.cttz.i8(i8 %a, i1 true)
113 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i8 %cttz
114 ;
115 ; BMI-LABEL: 'var_cttz_i8u'
116 ; BMI-NEXT: Cost Model: Found an estimated cost of 1 for instruction: %cttz = call i8 @llvm.cttz.i8(i8 %a, i1 true)
117 ; BMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret i8 %cttz
118 ;
119 %cttz = call i8 @llvm.cttz.i8(i8 %a, i1 1)
120 ret i8 %cttz
121 }
123 ; Verify the cost of vector trailing zero count instructions.
125 declare <2 x i64> @llvm.cttz.v2i64(<2 x i64>, i1)
126 declare <4 x i32> @llvm.cttz.v4i32(<4 x i32>, i1)
127 declare <8 x i16> @llvm.cttz.v8i16(<8 x i16>, i1)
128 declare <16 x i8> @llvm.cttz.v16i8(<16 x i8>, i1)
130 declare <4 x i64> @llvm.cttz.v4i64(<4 x i64>, i1)
131 declare <8 x i32> @llvm.cttz.v8i32(<8 x i32>, i1)
132 declare <16 x i16> @llvm.cttz.v16i16(<16 x i16>, i1)
133 declare <32 x i8> @llvm.cttz.v32i8(<32 x i8>, i1)
135 declare <8 x i64> @llvm.cttz.v8i64(<8 x i64>, i1)
136 declare <16 x i32> @llvm.cttz.v16i32(<16 x i32>, i1)
137 declare <32 x i16> @llvm.cttz.v32i16(<32 x i16>, i1)
138 declare <64 x i8> @llvm.cttz.v64i8(<64 x i8>, i1)
140 define <2 x i64> @var_cttz_v2i64(<2 x i64> %a) {
141 ; NOBMI-LABEL: 'var_cttz_v2i64'
142 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
143 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
144 ;
145 ; SSE2-LABEL: 'var_cttz_v2i64'
146 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
147 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
148 ;
149 ; SSE42-LABEL: 'var_cttz_v2i64'
150 ; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
151 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
152 ;
153 ; AVX1-LABEL: 'var_cttz_v2i64'
154 ; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
155 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
156 ;
157 ; AVX2-LABEL: 'var_cttz_v2i64'
158 ; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
159 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
160 ;
161 ; AVX512-LABEL: 'var_cttz_v2i64'
162 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 false)
163 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
164 ;
165 %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 0)
166 ret <2 x i64> %cttz
167 }
169 define <2 x i64> @var_cttz_v2i64u(<2 x i64> %a) {
170 ; NOBMI-LABEL: 'var_cttz_v2i64u'
171 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
172 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
173 ;
174 ; SSE2-LABEL: 'var_cttz_v2i64u'
175 ; SSE2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
176 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
177 ;
178 ; SSE42-LABEL: 'var_cttz_v2i64u'
179 ; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
180 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
181 ;
182 ; AVX1-LABEL: 'var_cttz_v2i64u'
183 ; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
184 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
185 ;
186 ; AVX2-LABEL: 'var_cttz_v2i64u'
187 ; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
188 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
189 ;
190 ; AVX512-LABEL: 'var_cttz_v2i64u'
191 ; AVX512-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 true)
192 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <2 x i64> %cttz
193 ;
194 %cttz = call <2 x i64> @llvm.cttz.v2i64(<2 x i64> %a, i1 1)
195 ret <2 x i64> %cttz
196 }
198 define <4 x i64> @var_cttz_v4i64(<4 x i64> %a) {
199 ; NOBMI-LABEL: 'var_cttz_v4i64'
200 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
201 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
202 ;
203 ; SSE2-LABEL: 'var_cttz_v4i64'
204 ; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
205 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
206 ;
207 ; SSE42-LABEL: 'var_cttz_v4i64'
208 ; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
209 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
210 ;
211 ; AVX1-LABEL: 'var_cttz_v4i64'
212 ; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
213 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
214 ;
215 ; AVX2-LABEL: 'var_cttz_v4i64'
216 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
217 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
218 ;
219 ; AVX512-LABEL: 'var_cttz_v4i64'
220 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 false)
221 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
222 ;
223 %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 0)
224 ret <4 x i64> %cttz
225 }
227 define <4 x i64> @var_cttz_v4i64u(<4 x i64> %a) {
228 ; NOBMI-LABEL: 'var_cttz_v4i64u'
229 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
230 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
231 ;
232 ; SSE2-LABEL: 'var_cttz_v4i64u'
233 ; SSE2-NEXT: Cost Model: Found an estimated cost of 28 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
234 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
235 ;
236 ; SSE42-LABEL: 'var_cttz_v4i64u'
237 ; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
238 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
239 ;
240 ; AVX1-LABEL: 'var_cttz_v4i64u'
241 ; AVX1-NEXT: Cost Model: Found an estimated cost of 17 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
242 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
243 ;
244 ; AVX2-LABEL: 'var_cttz_v4i64u'
245 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
246 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
247 ;
248 ; AVX512-LABEL: 'var_cttz_v4i64u'
249 ; AVX512-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 true)
250 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i64> %cttz
251 ;
252 %cttz = call <4 x i64> @llvm.cttz.v4i64(<4 x i64> %a, i1 1)
253 ret <4 x i64> %cttz
254 }
256 define <8 x i64> @var_cttz_v8i64(<8 x i64> %a) {
257 ; NOBMI-LABEL: 'var_cttz_v8i64'
258 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
259 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
260 ;
261 ; SSE2-LABEL: 'var_cttz_v8i64'
262 ; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
263 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
264 ;
265 ; SSE42-LABEL: 'var_cttz_v8i64'
266 ; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
267 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
268 ;
269 ; AVX1-LABEL: 'var_cttz_v8i64'
270 ; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
271 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
272 ;
273 ; AVX2-LABEL: 'var_cttz_v8i64'
274 ; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
275 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
276 ;
277 ; AVX512-LABEL: 'var_cttz_v8i64'
278 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 false)
279 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
280 ;
281 %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 0)
282 ret <8 x i64> %cttz
283 }
285 define <8 x i64> @var_cttz_v8i64u(<8 x i64> %a) {
286 ; NOBMI-LABEL: 'var_cttz_v8i64u'
287 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
288 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
289 ;
290 ; SSE2-LABEL: 'var_cttz_v8i64u'
291 ; SSE2-NEXT: Cost Model: Found an estimated cost of 56 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
292 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
293 ;
294 ; SSE42-LABEL: 'var_cttz_v8i64u'
295 ; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
296 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
297 ;
298 ; AVX1-LABEL: 'var_cttz_v8i64u'
299 ; AVX1-NEXT: Cost Model: Found an estimated cost of 34 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
300 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
301 ;
302 ; AVX2-LABEL: 'var_cttz_v8i64u'
303 ; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
304 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
305 ;
306 ; AVX512-LABEL: 'var_cttz_v8i64u'
307 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 true)
308 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i64> %cttz
309 ;
310 %cttz = call <8 x i64> @llvm.cttz.v8i64(<8 x i64> %a, i1 1)
311 ret <8 x i64> %cttz
312 }
314 define <4 x i32> @var_cttz_v4i32(<4 x i32> %a) {
315 ; NOBMI-LABEL: 'var_cttz_v4i32'
316 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
317 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
318 ;
319 ; SSE-LABEL: 'var_cttz_v4i32'
320 ; SSE-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
321 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
322 ;
323 ; AVX1-LABEL: 'var_cttz_v4i32'
324 ; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
325 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
326 ;
327 ; AVX2-LABEL: 'var_cttz_v4i32'
328 ; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
329 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
330 ;
331 ; AVX512-LABEL: 'var_cttz_v4i32'
332 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 false)
333 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
334 ;
335 %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 0)
336 ret <4 x i32> %cttz
337 }
339 define <4 x i32> @var_cttz_v4i32u(<4 x i32> %a) {
340 ; NOBMI-LABEL: 'var_cttz_v4i32u'
341 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
342 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
343 ;
344 ; SSE-LABEL: 'var_cttz_v4i32u'
345 ; SSE-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
346 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
347 ;
348 ; AVX1-LABEL: 'var_cttz_v4i32u'
349 ; AVX1-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
350 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
351 ;
352 ; AVX2-LABEL: 'var_cttz_v4i32u'
353 ; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
354 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
355 ;
356 ; AVX512-LABEL: 'var_cttz_v4i32u'
357 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 true)
358 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <4 x i32> %cttz
359 ;
360 %cttz = call <4 x i32> @llvm.cttz.v4i32(<4 x i32> %a, i1 1)
361 ret <4 x i32> %cttz
362 }
364 define <8 x i32> @var_cttz_v8i32(<8 x i32> %a) {
365 ; NOBMI-LABEL: 'var_cttz_v8i32'
366 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
367 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
368 ;
369 ; SSE-LABEL: 'var_cttz_v8i32'
370 ; SSE-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
371 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
372 ;
373 ; AVX1-LABEL: 'var_cttz_v8i32'
374 ; AVX1-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
375 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
376 ;
377 ; AVX2-LABEL: 'var_cttz_v8i32'
378 ; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
379 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
380 ;
381 ; AVX512-LABEL: 'var_cttz_v8i32'
382 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 false)
383 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
384 ;
385 %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 0)
386 ret <8 x i32> %cttz
387 }
389 define <8 x i32> @var_cttz_v8i32u(<8 x i32> %a) {
390 ; NOBMI-LABEL: 'var_cttz_v8i32u'
391 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
392 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
393 ;
394 ; SSE-LABEL: 'var_cttz_v8i32u'
395 ; SSE-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
396 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
397 ;
398 ; AVX1-LABEL: 'var_cttz_v8i32u'
399 ; AVX1-NEXT: Cost Model: Found an estimated cost of 21 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
400 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
401 ;
402 ; AVX2-LABEL: 'var_cttz_v8i32u'
403 ; AVX2-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
404 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
405 ;
406 ; AVX512-LABEL: 'var_cttz_v8i32u'
407 ; AVX512-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 true)
408 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i32> %cttz
409 ;
410 %cttz = call <8 x i32> @llvm.cttz.v8i32(<8 x i32> %a, i1 1)
411 ret <8 x i32> %cttz
412 }
414 define <16 x i32> @var_cttz_v16i32(<16 x i32> %a) {
415 ; NOBMI-LABEL: 'var_cttz_v16i32'
416 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
417 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
418 ;
419 ; SSE-LABEL: 'var_cttz_v16i32'
420 ; SSE-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
421 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
422 ;
423 ; AVX1-LABEL: 'var_cttz_v16i32'
424 ; AVX1-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
425 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
426 ;
427 ; AVX2-LABEL: 'var_cttz_v16i32'
428 ; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
429 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
430 ;
431 ; AVX512-LABEL: 'var_cttz_v16i32'
432 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 false)
433 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
434 ;
435 %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 0)
436 ret <16 x i32> %cttz
437 }
439 define <16 x i32> @var_cttz_v16i32u(<16 x i32> %a) {
440 ; NOBMI-LABEL: 'var_cttz_v16i32u'
441 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
442 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
443 ;
444 ; SSE-LABEL: 'var_cttz_v16i32u'
445 ; SSE-NEXT: Cost Model: Found an estimated cost of 72 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
446 ; SSE-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
447 ;
448 ; AVX1-LABEL: 'var_cttz_v16i32u'
449 ; AVX1-NEXT: Cost Model: Found an estimated cost of 42 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
450 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
451 ;
452 ; AVX2-LABEL: 'var_cttz_v16i32u'
453 ; AVX2-NEXT: Cost Model: Found an estimated cost of 14 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
454 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
455 ;
456 ; AVX512-LABEL: 'var_cttz_v16i32u'
457 ; AVX512-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 true)
458 ; AVX512-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i32> %cttz
459 ;
460 %cttz = call <16 x i32> @llvm.cttz.v16i32(<16 x i32> %a, i1 1)
461 ret <16 x i32> %cttz
462 }
464 define <8 x i16> @var_cttz_v8i16(<8 x i16> %a) {
465 ; NOBMI-LABEL: 'var_cttz_v8i16'
466 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
467 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
468 ;
469 ; SSE2-LABEL: 'var_cttz_v8i16'
470 ; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
471 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
472 ;
473 ; SSE42-LABEL: 'var_cttz_v8i16'
474 ; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
475 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
476 ;
477 ; AVX1-LABEL: 'var_cttz_v8i16'
478 ; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
479 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
480 ;
481 ; AVX2-LABEL: 'var_cttz_v8i16'
482 ; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
483 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
484 ;
485 ; AVX512F-LABEL: 'var_cttz_v8i16'
486 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
487 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
488 ;
489 ; AVX512BW-LABEL: 'var_cttz_v8i16'
490 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
491 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
492 ;
493 ; AVX512VPOPCNT-LABEL: 'var_cttz_v8i16'
494 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
495 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
496 ;
497 ; AVX512BITALG-LABEL: 'var_cttz_v8i16'
498 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 false)
499 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
500 ;
501 %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 0)
502 ret <8 x i16> %cttz
503 }
505 define <8 x i16> @var_cttz_v8i16u(<8 x i16> %a) {
506 ; NOBMI-LABEL: 'var_cttz_v8i16u'
507 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
508 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
509 ;
510 ; SSE2-LABEL: 'var_cttz_v8i16u'
511 ; SSE2-NEXT: Cost Model: Found an estimated cost of 16 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
512 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
513 ;
514 ; SSE42-LABEL: 'var_cttz_v8i16u'
515 ; SSE42-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
516 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
517 ;
518 ; AVX1-LABEL: 'var_cttz_v8i16u'
519 ; AVX1-NEXT: Cost Model: Found an estimated cost of 9 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
520 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
521 ;
522 ; AVX2-LABEL: 'var_cttz_v8i16u'
523 ; AVX2-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
524 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
525 ;
526 ; AVX512F-LABEL: 'var_cttz_v8i16u'
527 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
528 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
529 ;
530 ; AVX512BW-LABEL: 'var_cttz_v8i16u'
531 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
532 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
533 ;
534 ; AVX512VPOPCNT-LABEL: 'var_cttz_v8i16u'
535 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 4 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
536 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
537 ;
538 ; AVX512BITALG-LABEL: 'var_cttz_v8i16u'
539 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 true)
540 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <8 x i16> %cttz
541 ;
542 %cttz = call <8 x i16> @llvm.cttz.v8i16(<8 x i16> %a, i1 1)
543 ret <8 x i16> %cttz
544 }
546 define <16 x i16> @var_cttz_v16i16(<16 x i16> %a) {
547 ; NOBMI-LABEL: 'var_cttz_v16i16'
548 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
549 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
550 ;
551 ; SSE2-LABEL: 'var_cttz_v16i16'
552 ; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
553 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
554 ;
555 ; SSE42-LABEL: 'var_cttz_v16i16'
556 ; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
557 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
558 ;
559 ; AVX1-LABEL: 'var_cttz_v16i16'
560 ; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
561 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
562 ;
563 ; AVX2-LABEL: 'var_cttz_v16i16'
564 ; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
565 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
566 ;
567 ; AVX512F-LABEL: 'var_cttz_v16i16'
568 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
569 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
570 ;
571 ; AVX512BW-LABEL: 'var_cttz_v16i16'
572 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
573 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
574 ;
575 ; AVX512VPOPCNT-LABEL: 'var_cttz_v16i16'
576 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
577 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
578 ;
579 ; AVX512BITALG-LABEL: 'var_cttz_v16i16'
580 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 false)
581 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
582 ;
583 %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 0)
584 ret <16 x i16> %cttz
585 }
587 define <16 x i16> @var_cttz_v16i16u(<16 x i16> %a) {
588 ; NOBMI-LABEL: 'var_cttz_v16i16u'
589 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
590 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
591 ;
592 ; SSE2-LABEL: 'var_cttz_v16i16u'
593 ; SSE2-NEXT: Cost Model: Found an estimated cost of 32 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
594 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
595 ;
596 ; SSE42-LABEL: 'var_cttz_v16i16u'
597 ; SSE42-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
598 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
599 ;
600 ; AVX1-LABEL: 'var_cttz_v16i16u'
601 ; AVX1-NEXT: Cost Model: Found an estimated cost of 18 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
602 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
603 ;
604 ; AVX2-LABEL: 'var_cttz_v16i16u'
605 ; AVX2-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
606 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
607 ;
608 ; AVX512F-LABEL: 'var_cttz_v16i16u'
609 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
610 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
611 ;
612 ; AVX512BW-LABEL: 'var_cttz_v16i16u'
613 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
614 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
615 ;
616 ; AVX512VPOPCNT-LABEL: 'var_cttz_v16i16u'
617 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
618 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
619 ;
620 ; AVX512BITALG-LABEL: 'var_cttz_v16i16u'
621 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 true)
622 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i16> %cttz
623 ;
624 %cttz = call <16 x i16> @llvm.cttz.v16i16(<16 x i16> %a, i1 1)
625 ret <16 x i16> %cttz
626 }
628 define <32 x i16> @var_cttz_v32i16(<32 x i16> %a) {
629 ; NOBMI-LABEL: 'var_cttz_v32i16'
630 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
631 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
632 ;
633 ; SSE2-LABEL: 'var_cttz_v32i16'
634 ; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
635 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
636 ;
637 ; SSE42-LABEL: 'var_cttz_v32i16'
638 ; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
639 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
640 ;
641 ; AVX1-LABEL: 'var_cttz_v32i16'
642 ; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
643 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
644 ;
645 ; AVX2-LABEL: 'var_cttz_v32i16'
646 ; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
647 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
648 ;
649 ; AVX512F-LABEL: 'var_cttz_v32i16'
650 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
651 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
652 ;
653 ; AVX512BW-LABEL: 'var_cttz_v32i16'
654 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
655 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
656 ;
657 ; AVX512VPOPCNT-LABEL: 'var_cttz_v32i16'
658 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
659 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
660 ;
661 ; AVX512BITALG-LABEL: 'var_cttz_v32i16'
662 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 false)
663 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
664 ;
665 %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 0)
666 ret <32 x i16> %cttz
667 }
669 define <32 x i16> @var_cttz_v32i16u(<32 x i16> %a) {
670 ; NOBMI-LABEL: 'var_cttz_v32i16u'
671 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
672 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
673 ;
674 ; SSE2-LABEL: 'var_cttz_v32i16u'
675 ; SSE2-NEXT: Cost Model: Found an estimated cost of 64 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
676 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
677 ;
678 ; SSE42-LABEL: 'var_cttz_v32i16u'
679 ; SSE42-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
680 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
681 ;
682 ; AVX1-LABEL: 'var_cttz_v32i16u'
683 ; AVX1-NEXT: Cost Model: Found an estimated cost of 36 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
684 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
685 ;
686 ; AVX2-LABEL: 'var_cttz_v32i16u'
687 ; AVX2-NEXT: Cost Model: Found an estimated cost of 12 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
688 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
689 ;
690 ; AVX512F-LABEL: 'var_cttz_v32i16u'
691 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
692 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
693 ;
694 ; AVX512BW-LABEL: 'var_cttz_v32i16u'
695 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
696 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
697 ;
698 ; AVX512VPOPCNT-LABEL: 'var_cttz_v32i16u'
699 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 7 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
700 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
701 ;
702 ; AVX512BITALG-LABEL: 'var_cttz_v32i16u'
703 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 true)
704 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i16> %cttz
705 ;
706 %cttz = call <32 x i16> @llvm.cttz.v32i16(<32 x i16> %a, i1 1)
707 ret <32 x i16> %cttz
708 }
710 define <16 x i8> @var_cttz_v16i8(<16 x i8> %a) {
711 ; NOBMI-LABEL: 'var_cttz_v16i8'
712 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
713 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
714 ;
715 ; SSE2-LABEL: 'var_cttz_v16i8'
716 ; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
717 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
718 ;
719 ; SSE42-LABEL: 'var_cttz_v16i8'
720 ; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
721 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
722 ;
723 ; AVX1-LABEL: 'var_cttz_v16i8'
724 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
725 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
726 ;
727 ; AVX2-LABEL: 'var_cttz_v16i8'
728 ; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
729 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
730 ;
731 ; AVX512F-LABEL: 'var_cttz_v16i8'
732 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
733 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
734 ;
735 ; AVX512BW-LABEL: 'var_cttz_v16i8'
736 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
737 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
738 ;
739 ; AVX512VPOPCNT-LABEL: 'var_cttz_v16i8'
740 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
741 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
742 ;
743 ; AVX512BITALG-LABEL: 'var_cttz_v16i8'
744 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 false)
745 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
746 ;
747 %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 0)
748 ret <16 x i8> %cttz
749 }
751 define <16 x i8> @var_cttz_v16i8u(<16 x i8> %a) {
752 ; NOBMI-LABEL: 'var_cttz_v16i8u'
753 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
754 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
755 ;
756 ; SSE2-LABEL: 'var_cttz_v16i8u'
757 ; SSE2-NEXT: Cost Model: Found an estimated cost of 13 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
758 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
759 ;
760 ; SSE42-LABEL: 'var_cttz_v16i8u'
761 ; SSE42-NEXT: Cost Model: Found an estimated cost of 11 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
762 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
763 ;
764 ; AVX1-LABEL: 'var_cttz_v16i8u'
765 ; AVX1-NEXT: Cost Model: Found an estimated cost of 8 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
766 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
767 ;
768 ; AVX2-LABEL: 'var_cttz_v16i8u'
769 ; AVX2-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
770 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
771 ;
772 ; AVX512F-LABEL: 'var_cttz_v16i8u'
773 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
774 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
775 ;
776 ; AVX512BW-LABEL: 'var_cttz_v16i8u'
777 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
778 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
779 ;
780 ; AVX512VPOPCNT-LABEL: 'var_cttz_v16i8u'
781 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
782 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
783 ;
784 ; AVX512BITALG-LABEL: 'var_cttz_v16i8u'
785 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 true)
786 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <16 x i8> %cttz
787 ;
788 %cttz = call <16 x i8> @llvm.cttz.v16i8(<16 x i8> %a, i1 1)
789 ret <16 x i8> %cttz
790 }
792 define <32 x i8> @var_cttz_v32i8(<32 x i8> %a) {
793 ; NOBMI-LABEL: 'var_cttz_v32i8'
794 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
795 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
796 ;
797 ; SSE2-LABEL: 'var_cttz_v32i8'
798 ; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
799 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
800 ;
801 ; SSE42-LABEL: 'var_cttz_v32i8'
802 ; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
803 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
804 ;
805 ; AVX1-LABEL: 'var_cttz_v32i8'
806 ; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
807 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
808 ;
809 ; AVX2-LABEL: 'var_cttz_v32i8'
810 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
811 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
812 ;
813 ; AVX512F-LABEL: 'var_cttz_v32i8'
814 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
815 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
816 ;
817 ; AVX512BW-LABEL: 'var_cttz_v32i8'
818 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
819 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
820 ;
821 ; AVX512VPOPCNT-LABEL: 'var_cttz_v32i8'
822 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
823 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
824 ;
825 ; AVX512BITALG-LABEL: 'var_cttz_v32i8'
826 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 false)
827 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
828 ;
829 %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 0)
830 ret <32 x i8> %cttz
831 }
833 define <32 x i8> @var_cttz_v32i8u(<32 x i8> %a) {
834 ; NOBMI-LABEL: 'var_cttz_v32i8u'
835 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
836 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
837 ;
838 ; SSE2-LABEL: 'var_cttz_v32i8u'
839 ; SSE2-NEXT: Cost Model: Found an estimated cost of 26 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
840 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
841 ;
842 ; SSE42-LABEL: 'var_cttz_v32i8u'
843 ; SSE42-NEXT: Cost Model: Found an estimated cost of 22 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
844 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
845 ;
846 ; AVX1-LABEL: 'var_cttz_v32i8u'
847 ; AVX1-NEXT: Cost Model: Found an estimated cost of 15 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
848 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
849 ;
850 ; AVX2-LABEL: 'var_cttz_v32i8u'
851 ; AVX2-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
852 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
853 ;
854 ; AVX512F-LABEL: 'var_cttz_v32i8u'
855 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
856 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
857 ;
858 ; AVX512BW-LABEL: 'var_cttz_v32i8u'
859 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
860 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
861 ;
862 ; AVX512VPOPCNT-LABEL: 'var_cttz_v32i8u'
863 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 5 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
864 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
865 ;
866 ; AVX512BITALG-LABEL: 'var_cttz_v32i8u'
867 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 2 for instruction: %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 true)
868 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <32 x i8> %cttz
869 ;
870 %cttz = call <32 x i8> @llvm.cttz.v32i8(<32 x i8> %a, i1 1)
871 ret <32 x i8> %cttz
872 }
874 define <64 x i8> @var_cttz_v64i8(<64 x i8> %a) {
875 ; NOBMI-LABEL: 'var_cttz_v64i8'
876 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
877 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
878 ;
879 ; SSE2-LABEL: 'var_cttz_v64i8'
880 ; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
881 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
882 ;
883 ; SSE42-LABEL: 'var_cttz_v64i8'
884 ; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
885 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
886 ;
887 ; AVX1-LABEL: 'var_cttz_v64i8'
888 ; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
889 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
890 ;
891 ; AVX2-LABEL: 'var_cttz_v64i8'
892 ; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
893 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
894 ;
895 ; AVX512F-LABEL: 'var_cttz_v64i8'
896 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
897 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
898 ;
899 ; AVX512BW-LABEL: 'var_cttz_v64i8'
900 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
901 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
902 ;
903 ; AVX512VPOPCNT-LABEL: 'var_cttz_v64i8'
904 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
905 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
906 ;
907 ; AVX512BITALG-LABEL: 'var_cttz_v64i8'
908 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 false)
909 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
910 ;
911 %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 0)
912 ret <64 x i8> %cttz
913 }
915 define <64 x i8> @var_cttz_v64i8u(<64 x i8> %a) {
916 ; NOBMI-LABEL: 'var_cttz_v64i8u'
917 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
918 ; NOBMI-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
919 ;
920 ; SSE2-LABEL: 'var_cttz_v64i8u'
921 ; SSE2-NEXT: Cost Model: Found an estimated cost of 52 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
922 ; SSE2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
923 ;
924 ; SSE42-LABEL: 'var_cttz_v64i8u'
925 ; SSE42-NEXT: Cost Model: Found an estimated cost of 44 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
926 ; SSE42-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
927 ;
928 ; AVX1-LABEL: 'var_cttz_v64i8u'
929 ; AVX1-NEXT: Cost Model: Found an estimated cost of 30 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
930 ; AVX1-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
931 ;
932 ; AVX2-LABEL: 'var_cttz_v64i8u'
933 ; AVX2-NEXT: Cost Model: Found an estimated cost of 10 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
934 ; AVX2-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
935 ;
936 ; AVX512F-LABEL: 'var_cttz_v64i8u'
937 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
938 ; AVX512F-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
939 ;
940 ; AVX512BW-LABEL: 'var_cttz_v64i8u'
941 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
942 ; AVX512BW-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
943 ;
944 ; AVX512VPOPCNT-LABEL: 'var_cttz_v64i8u'
945 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 6 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
946 ; AVX512VPOPCNT-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
947 ;
948 ; AVX512BITALG-LABEL: 'var_cttz_v64i8u'
949 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 3 for instruction: %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 true)
950 ; AVX512BITALG-NEXT: Cost Model: Found an estimated cost of 0 for instruction: ret <64 x i8> %cttz
951 ;
952 %cttz = call <64 x i8> @llvm.cttz.v64i8(<64 x i8> %a, i1 1)
953 ret <64 x i8> %cttz
954 }