1 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s
3 declare i32 @llvm.x86.avx512.kortestz.w(i16, i16) nounwind readnone
4 ; CHECK-LABEL: test_kortestz
7 define i32 @test_kortestz(i16 %a0, i16 %a1) {
8 %res = call i32 @llvm.x86.avx512.kortestz.w(i16 %a0, i16 %a1)
12 declare i32 @llvm.x86.avx512.kortestc.w(i16, i16) nounwind readnone
13 ; CHECK-LABEL: test_kortestc
16 define i32 @test_kortestc(i16 %a0, i16 %a1) {
17 %res = call i32 @llvm.x86.avx512.kortestc.w(i16 %a0, i16 %a1)
21 declare i16 @llvm.x86.avx512.kand.w(i16, i16) nounwind readnone
22 ; CHECK-LABEL: test_kand
25 define i16 @test_kand(i16 %a0, i16 %a1) {
26 %t1 = call i16 @llvm.x86.avx512.kand.w(i16 %a0, i16 8)
27 %t2 = call i16 @llvm.x86.avx512.kand.w(i16 %t1, i16 %a1)
31 declare i16 @llvm.x86.avx512.knot.w(i16) nounwind readnone
32 ; CHECK-LABEL: test_knot
34 define i16 @test_knot(i16 %a0) {
35 %res = call i16 @llvm.x86.avx512.knot.w(i16 %a0)
39 declare i16 @llvm.x86.avx512.kunpck.bw(i16, i16) nounwind readnone
41 ; CHECK-LABEL: unpckbw_test
44 define i16 @unpckbw_test(i16 %a0, i16 %a1) {
45 %res = call i16 @llvm.x86.avx512.kunpck.bw(i16 %a0, i16 %a1)
49 define <16 x float> @test_rcp_ps_512(<16 x float> %a0) {
51 %res = call <16 x float> @llvm.x86.avx512.rcp14.ps.512(<16 x float> %a0) ; <<16 x float>> [#uses=1]
54 declare <16 x float> @llvm.x86.avx512.rcp14.ps.512(<16 x float>) nounwind readnone
56 define <8 x double> @test_rcp_pd_512(<8 x double> %a0) {
58 %res = call <8 x double> @llvm.x86.avx512.rcp14.pd.512(<8 x double> %a0) ; <<8 x double>> [#uses=1]
61 declare <8 x double> @llvm.x86.avx512.rcp14.pd.512(<8 x double>) nounwind readnone
63 define <16 x float> @test_rcp28_ps_512(<16 x float> %a0) {
65 %res = call <16 x float> @llvm.x86.avx512.rcp28.ps.512(<16 x float> %a0) ; <<16 x float>> [#uses=1]
68 declare <16 x float> @llvm.x86.avx512.rcp28.ps.512(<16 x float>) nounwind readnone
70 define <8 x double> @test_rcp28_pd_512(<8 x double> %a0) {
72 %res = call <8 x double> @llvm.x86.avx512.rcp28.pd.512(<8 x double> %a0) ; <<8 x double>> [#uses=1]
75 declare <8 x double> @llvm.x86.avx512.rcp28.pd.512(<8 x double>) nounwind readnone
77 define <8 x double> @test_rndscale_pd_512(<8 x double> %a0) {
79 %res = call <8 x double> @llvm.x86.avx512.rndscale.pd.512(<8 x double> %a0, i32 7) ; <<8 x double>> [#uses=1]
82 declare <8 x double> @llvm.x86.avx512.rndscale.pd.512(<8 x double>, i32) nounwind readnone
85 define <16 x float> @test_rndscale_ps_512(<16 x float> %a0) {
87 %res = call <16 x float> @llvm.x86.avx512.rndscale.ps.512(<16 x float> %a0, i32 7) ; <<16 x float>> [#uses=1]
90 declare <16 x float> @llvm.x86.avx512.rndscale.ps.512(<16 x float>, i32) nounwind readnone
93 define <16 x float> @test_rsqrt_ps_512(<16 x float> %a0) {
95 %res = call <16 x float> @llvm.x86.avx512.rsqrt14.ps.512(<16 x float> %a0) ; <<16 x float>> [#uses=1]
98 declare <16 x float> @llvm.x86.avx512.rsqrt14.ps.512(<16 x float>) nounwind readnone
100 define <16 x float> @test_rsqrt28_ps_512(<16 x float> %a0) {
102 %res = call <16 x float> @llvm.x86.avx512.rsqrt28.ps.512(<16 x float> %a0) ; <<16 x float>> [#uses=1]
103 ret <16 x float> %res
105 declare <16 x float> @llvm.x86.avx512.rsqrt28.ps.512(<16 x float>) nounwind readnone
107 define <4 x float> @test_rsqrt14_ss(<4 x float> %a0) {
109 %res = call <4 x float> @llvm.x86.avx512.rsqrt14.ss(<4 x float> %a0) ; <<4 x float>> [#uses=1]
112 declare <4 x float> @llvm.x86.avx512.rsqrt14.ss(<4 x float>) nounwind readnone
114 define <4 x float> @test_rsqrt28_ss(<4 x float> %a0) {
116 %res = call <4 x float> @llvm.x86.avx512.rsqrt28.ss(<4 x float> %a0) ; <<4 x float>> [#uses=1]
119 declare <4 x float> @llvm.x86.avx512.rsqrt28.ss(<4 x float>) nounwind readnone
121 define <4 x float> @test_rcp14_ss(<4 x float> %a0) {
123 %res = call <4 x float> @llvm.x86.avx512.rcp14.ss(<4 x float> %a0) ; <<4 x float>> [#uses=1]
126 declare <4 x float> @llvm.x86.avx512.rcp14.ss(<4 x float>) nounwind readnone
128 define <4 x float> @test_rcp28_ss(<4 x float> %a0) {
130 %res = call <4 x float> @llvm.x86.avx512.rcp28.ss(<4 x float> %a0) ; <<4 x float>> [#uses=1]
133 declare <4 x float> @llvm.x86.avx512.rcp28.ss(<4 x float>) nounwind readnone
135 define <8 x double> @test_sqrt_pd_512(<8 x double> %a0) {
137 %res = call <8 x double> @llvm.x86.avx512.sqrt.pd.512(<8 x double> %a0) ; <<8 x double>> [#uses=1]
138 ret <8 x double> %res
140 declare <8 x double> @llvm.x86.avx512.sqrt.pd.512(<8 x double>) nounwind readnone
142 define <16 x float> @test_sqrt_ps_512(<16 x float> %a0) {
144 %res = call <16 x float> @llvm.x86.avx512.sqrt.ps.512(<16 x float> %a0) ; <<16 x float>> [#uses=1]
145 ret <16 x float> %res
147 declare <16 x float> @llvm.x86.avx512.sqrt.ps.512(<16 x float>) nounwind readnone
149 define <4 x float> @test_sqrt_ss(<4 x float> %a0, <4 x float> %a1) {
151 %res = call <4 x float> @llvm.x86.avx512.sqrt.ss(<4 x float> %a0, <4 x float> %a1) ; <<4 x float>> [#uses=1]
154 declare <4 x float> @llvm.x86.avx512.sqrt.ss(<4 x float>, <4 x float>) nounwind readnone
156 define <2 x double> @test_sqrt_sd(<2 x double> %a0, <2 x double> %a1) {
158 %res = call <2 x double> @llvm.x86.avx512.sqrt.sd(<2 x double> %a0, <2 x double> %a1) ; <<2 x double>> [#uses=1]
159 ret <2 x double> %res
161 declare <2 x double> @llvm.x86.avx512.sqrt.sd(<2 x double>, <2 x double>) nounwind readnone
163 define i64 @test_x86_sse2_cvtsd2si64(<2 x double> %a0) {
165 %res = call i64 @llvm.x86.sse2.cvtsd2si64(<2 x double> %a0) ; <i64> [#uses=1]
168 declare i64 @llvm.x86.sse2.cvtsd2si64(<2 x double>) nounwind readnone
170 define <2 x double> @test_x86_sse2_cvtsi642sd(<2 x double> %a0, i64 %a1) {
172 %res = call <2 x double> @llvm.x86.sse2.cvtsi642sd(<2 x double> %a0, i64 %a1) ; <<2 x double>> [#uses=1]
173 ret <2 x double> %res
175 declare <2 x double> @llvm.x86.sse2.cvtsi642sd(<2 x double>, i64) nounwind readnone
177 define <2 x double> @test_x86_avx512_cvtusi642sd(<2 x double> %a0, i64 %a1) {
178 ; CHECK: vcvtusi2sdqz
179 %res = call <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double> %a0, i64 %a1) ; <<2 x double>> [#uses=1]
180 ret <2 x double> %res
182 declare <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double>, i64) nounwind readnone
184 define i64 @test_x86_sse2_cvttsd2si64(<2 x double> %a0) {
186 %res = call i64 @llvm.x86.sse2.cvttsd2si64(<2 x double> %a0) ; <i64> [#uses=1]
189 declare i64 @llvm.x86.sse2.cvttsd2si64(<2 x double>) nounwind readnone
192 define i64 @test_x86_sse_cvtss2si64(<4 x float> %a0) {
194 %res = call i64 @llvm.x86.sse.cvtss2si64(<4 x float> %a0) ; <i64> [#uses=1]
197 declare i64 @llvm.x86.sse.cvtss2si64(<4 x float>) nounwind readnone
200 define <4 x float> @test_x86_sse_cvtsi642ss(<4 x float> %a0, i64 %a1) {
202 %res = call <4 x float> @llvm.x86.sse.cvtsi642ss(<4 x float> %a0, i64 %a1) ; <<4 x float>> [#uses=1]
205 declare <4 x float> @llvm.x86.sse.cvtsi642ss(<4 x float>, i64) nounwind readnone
208 define i64 @test_x86_sse_cvttss2si64(<4 x float> %a0) {
210 %res = call i64 @llvm.x86.sse.cvttss2si64(<4 x float> %a0) ; <i64> [#uses=1]
213 declare i64 @llvm.x86.sse.cvttss2si64(<4 x float>) nounwind readnone
215 define i64 @test_x86_avx512_cvtsd2usi64(<2 x double> %a0) {
217 %res = call i64 @llvm.x86.avx512.cvtsd2usi64(<2 x double> %a0) ; <i64> [#uses=1]
220 declare i64 @llvm.x86.avx512.cvtsd2usi64(<2 x double>) nounwind readnone
222 define <16 x float> @test_x86_vcvtph2ps_512(<16 x i16> %a0) {
224 %res = call <16 x float> @llvm.x86.avx512.vcvtph2ps.512(<16 x i16> %a0)
225 ret <16 x float> %res
227 declare <16 x float> @llvm.x86.avx512.vcvtph2ps.512(<16 x i16>) nounwind readonly
230 define <16 x i16> @test_x86_vcvtps2ph_256(<16 x float> %a0) {
232 %res = call <16 x i16> @llvm.x86.avx512.vcvtps2ph.512(<16 x float> %a0, i32 0)
235 declare <16 x i16> @llvm.x86.avx512.vcvtps2ph.512(<16 x float>, i32) nounwind readonly
237 define <16 x float> @test_x86_vbroadcast_ss_512(i8* %a0) {
238 ; CHECK: vbroadcastss
239 %res = call <16 x float> @llvm.x86.avx512.vbroadcast.ss.512(i8* %a0) ; <<16 x float>> [#uses=1]
240 ret <16 x float> %res
242 declare <16 x float> @llvm.x86.avx512.vbroadcast.ss.512(i8*) nounwind readonly
244 define <8 x double> @test_x86_vbroadcast_sd_512(i8* %a0) {
245 ; CHECK: vbroadcastsd
246 %res = call <8 x double> @llvm.x86.avx512.vbroadcast.sd.512(i8* %a0) ; <<8 x double>> [#uses=1]
247 ret <8 x double> %res
249 declare <8 x double> @llvm.x86.avx512.vbroadcast.sd.512(i8*) nounwind readonly
251 define <16 x float> @test_x86_vbroadcast_ss_ps_512(<4 x float> %a0) {
252 ; CHECK: vbroadcastss
253 %res = call <16 x float> @llvm.x86.avx512.vbroadcast.ss.ps.512(<4 x float> %a0) ; <<16 x float>> [#uses=1]
254 ret <16 x float> %res
256 declare <16 x float> @llvm.x86.avx512.vbroadcast.ss.ps.512(<4 x float>) nounwind readonly
258 define <8 x double> @test_x86_vbroadcast_sd_pd_512(<2 x double> %a0) {
259 ; CHECK: vbroadcastsd
260 %res = call <8 x double> @llvm.x86.avx512.vbroadcast.sd.pd.512(<2 x double> %a0) ; <<8 x double>> [#uses=1]
261 ret <8 x double> %res
263 declare <8 x double> @llvm.x86.avx512.vbroadcast.sd.pd.512(<2 x double>) nounwind readonly
265 define <16 x i32> @test_x86_pbroadcastd_512(<4 x i32> %a0) {
266 ; CHECK: vpbroadcastd
267 %res = call <16 x i32> @llvm.x86.avx512.pbroadcastd.512(<4 x i32> %a0) ; <<16 x i32>> [#uses=1]
270 declare <16 x i32> @llvm.x86.avx512.pbroadcastd.512(<4 x i32>) nounwind readonly
272 define <16 x i32> @test_x86_pbroadcastd_i32_512(i32 %a0) {
273 ; CHECK: vpbroadcastd
274 %res = call <16 x i32> @llvm.x86.avx512.pbroadcastd.i32.512(i32 %a0) ; <<16 x i32>> [#uses=1]
277 declare <16 x i32> @llvm.x86.avx512.pbroadcastd.i32.512(i32) nounwind readonly
279 define <8 x i64> @test_x86_pbroadcastq_512(<2 x i64> %a0) {
280 ; CHECK: vpbroadcastq
281 %res = call <8 x i64> @llvm.x86.avx512.pbroadcastq.512(<2 x i64> %a0) ; <<8 x i64>> [#uses=1]
284 declare <8 x i64> @llvm.x86.avx512.pbroadcastq.512(<2 x i64>) nounwind readonly
286 define <8 x i64> @test_x86_pbroadcastq_i64_512(i64 %a0) {
287 ; CHECK: vpbroadcastq
288 %res = call <8 x i64> @llvm.x86.avx512.pbroadcastq.i64.512(i64 %a0) ; <<8 x i64>> [#uses=1]
291 declare <8 x i64> @llvm.x86.avx512.pbroadcastq.i64.512(i64) nounwind readonly
293 define <16 x i32> @test_x86_pmaxu_d(<16 x i32> %a0, <16 x i32> %a1) {
295 %res = call <16 x i32> @llvm.x86.avx512.pmaxu.d(<16 x i32> %a0, <16 x i32> %a1) ; <<16 x i32>> [#uses=1]
298 declare <16 x i32> @llvm.x86.avx512.pmaxu.d(<16 x i32>, <16 x i32>) nounwind readonly
300 define <8 x i64> @test_x86_pmaxu_q(<8 x i64> %a0, <8 x i64> %a1) {
302 %res = call <8 x i64> @llvm.x86.avx512.pmaxu.q(<8 x i64> %a0, <8 x i64> %a1) ; <<8 x i64>> [#uses=1]
305 declare <8 x i64> @llvm.x86.avx512.pmaxu.q(<8 x i64>, <8 x i64>) nounwind readonly
307 define <16 x i32> @test_x86_pmaxs_d(<16 x i32> %a0, <16 x i32> %a1) {
309 %res = call <16 x i32> @llvm.x86.avx512.pmaxs.d(<16 x i32> %a0, <16 x i32> %a1) ; <<16 x i32>> [#uses=1]
312 declare <16 x i32> @llvm.x86.avx512.pmaxs.d(<16 x i32>, <16 x i32>) nounwind readonly
314 define <8 x i64> @test_x86_pmaxs_q(<8 x i64> %a0, <8 x i64> %a1) {
316 %res = call <8 x i64> @llvm.x86.avx512.pmaxs.q(<8 x i64> %a0, <8 x i64> %a1) ; <<8 x i64>> [#uses=1]
319 declare <8 x i64> @llvm.x86.avx512.pmaxs.q(<8 x i64>, <8 x i64>) nounwind readonly
321 define <16 x i32> @test_x86_pminu_d(<16 x i32> %a0, <16 x i32> %a1) {
323 %res = call <16 x i32> @llvm.x86.avx512.pminu.d(<16 x i32> %a0, <16 x i32> %a1) ; <<16 x i32>> [#uses=1]
326 declare <16 x i32> @llvm.x86.avx512.pminu.d(<16 x i32>, <16 x i32>) nounwind readonly
328 define <8 x i64> @test_x86_pminu_q(<8 x i64> %a0, <8 x i64> %a1) {
330 %res = call <8 x i64> @llvm.x86.avx512.pminu.q(<8 x i64> %a0, <8 x i64> %a1) ; <<8 x i64>> [#uses=1]
333 declare <8 x i64> @llvm.x86.avx512.pminu.q(<8 x i64>, <8 x i64>) nounwind readonly
335 define <16 x i32> @test_x86_pmins_d(<16 x i32> %a0, <16 x i32> %a1) {
337 %res = call <16 x i32> @llvm.x86.avx512.pmins.d(<16 x i32> %a0, <16 x i32> %a1) ; <<16 x i32>> [#uses=1]
340 declare <16 x i32> @llvm.x86.avx512.pmins.d(<16 x i32>, <16 x i32>) nounwind readonly
342 define <8 x i64> @test_x86_pmins_q(<8 x i64> %a0, <8 x i64> %a1) {
344 %res = call <8 x i64> @llvm.x86.avx512.pmins.q(<8 x i64> %a0, <8 x i64> %a1) ; <<8 x i64>> [#uses=1]
347 declare <8 x i64> @llvm.x86.avx512.pmins.q(<8 x i64>, <8 x i64>) nounwind readonly
349 define <16 x i32> @test_conflict_d(<16 x i32> %a) {
350 ; CHECK: movw $-1, %ax
353 %res = call <16 x i32> @llvm.x86.avx512.mask.conflict.d.512(<16 x i32> %a, <16 x i32> zeroinitializer, i16 -1)
357 declare <16 x i32> @llvm.x86.avx512.mask.conflict.d.512(<16 x i32>, <16 x i32>, i16) nounwind readonly
359 define <8 x i64> @test_conflict_q(<8 x i64> %a) {
360 ; CHECK: movb $-1, %al
363 %res = call <8 x i64> @llvm.x86.avx512.mask.conflict.q.512(<8 x i64> %a, <8 x i64> zeroinitializer, i8 -1)
367 declare <8 x i64> @llvm.x86.avx512.mask.conflict.q.512(<8 x i64>, <8 x i64>, i8) nounwind readonly
370 define <16 x i32> @test_maskz_conflict_d(<16 x i32> %a, i16 %mask) {
372 %res = call <16 x i32> @llvm.x86.avx512.mask.conflict.d.512(<16 x i32> %a, <16 x i32> zeroinitializer, i16 %mask)
376 define <8 x i64> @test_mask_conflict_q(<8 x i64> %a, <8 x i64> %b, i8 %mask) {
378 %res = call <8 x i64> @llvm.x86.avx512.mask.conflict.q.512(<8 x i64> %a, <8 x i64> %b, i8 %mask)
382 define <16 x float> @test_x86_mask_blend_ps_512(i16 %a0, <16 x float> %a1, <16 x float> %a2) {
384 %m0 = bitcast i16 %a0 to <16 x i1>
385 %res = call <16 x float> @llvm.x86.avx512.mask.blend.ps.512(<16 x i1> %m0, <16 x float> %a1, <16 x float> %a2) ; <<16 x float>> [#uses=1]
386 ret <16 x float> %res
389 declare <16 x float> @llvm.x86.avx512.mask.blend.ps.512(<16 x i1> %a0, <16 x float> %a1, <16 x float> %a2) nounwind readonly
391 define <8 x double> @test_x86_mask_blend_pd_512(i8 %a0, <8 x double> %a1, <8 x double> %a2) {
393 %m0 = bitcast i8 %a0 to <8 x i1>
394 %res = call <8 x double> @llvm.x86.avx512.mask.blend.pd.512(<8 x i1> %m0, <8 x double> %a1, <8 x double> %a2) ; <<8 x double>> [#uses=1]
395 ret <8 x double> %res
398 define <8 x double> @test_x86_mask_blend_pd_512_memop(<8 x double> %a, <8 x double>* %ptr, i8 %mask) {
399 ; CHECK-LABEL: test_x86_mask_blend_pd_512_memop
400 ; CHECK: vblendmpd {{.*}}, {{%zmm[0-9]}}, {{%zmm[0-9]}} {%k1}
401 %vmask = bitcast i8 %mask to <8 x i1>
402 %b = load <8 x double>* %ptr
403 %res = call <8 x double> @llvm.x86.avx512.mask.blend.pd.512(<8 x i1> %vmask, <8 x double> %a, <8 x double> %b) ; <<8 x double>> [#uses=1]
404 ret <8 x double> %res
406 declare <8 x double> @llvm.x86.avx512.mask.blend.pd.512(<8 x i1> %a0, <8 x double> %a1, <8 x double> %a2) nounwind readonly
408 define <16 x i32> @test_x86_mask_blend_d_512(i16 %a0, <16 x i32> %a1, <16 x i32> %a2) {
410 %m0 = bitcast i16 %a0 to <16 x i1>
411 %res = call <16 x i32> @llvm.x86.avx512.mask.blend.d.512(<16 x i1> %m0, <16 x i32> %a1, <16 x i32> %a2) ; <<16 x i32>> [#uses=1]
414 declare <16 x i32> @llvm.x86.avx512.mask.blend.d.512(<16 x i1> %a0, <16 x i32> %a1, <16 x i32> %a2) nounwind readonly
416 define <8 x i64> @test_x86_mask_blend_q_512(i8 %a0, <8 x i64> %a1, <8 x i64> %a2) {
418 %m0 = bitcast i8 %a0 to <8 x i1>
419 %res = call <8 x i64> @llvm.x86.avx512.mask.blend.q.512(<8 x i1> %m0, <8 x i64> %a1, <8 x i64> %a2) ; <<8 x i64>> [#uses=1]
422 declare <8 x i64> @llvm.x86.avx512.mask.blend.q.512(<8 x i1> %a0, <8 x i64> %a1, <8 x i64> %a2) nounwind readonly