1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s
4 define <4 x double> @addpd256(<4 x double> %y, <4 x double> %x) nounwind uwtable readnone ssp {
5 ; CHECK-LABEL: addpd256:
6 ; CHECK: ## %bb.0: ## %entry
7 ; CHECK-NEXT: vaddpd %ymm0, %ymm1, %ymm0
10 %add.i = fadd <4 x double> %x, %y
11 ret <4 x double> %add.i
14 define <4 x double> @addpd256fold(<4 x double> %y) nounwind uwtable readnone ssp {
15 ; CHECK-LABEL: addpd256fold:
16 ; CHECK: ## %bb.0: ## %entry
17 ; CHECK-NEXT: vaddpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
20 %add.i = fadd <4 x double> %y, <double 4.500000e+00, double 3.400000e+00, double 2.300000e+00, double 1.200000e+00>
21 ret <4 x double> %add.i
24 define <8 x float> @addps256(<8 x float> %y, <8 x float> %x) nounwind uwtable readnone ssp {
25 ; CHECK-LABEL: addps256:
26 ; CHECK: ## %bb.0: ## %entry
27 ; CHECK-NEXT: vaddps %ymm0, %ymm1, %ymm0
30 %add.i = fadd <8 x float> %x, %y
31 ret <8 x float> %add.i
34 define <8 x float> @addps256fold(<8 x float> %y) nounwind uwtable readnone ssp {
35 ; CHECK-LABEL: addps256fold:
36 ; CHECK: ## %bb.0: ## %entry
37 ; CHECK-NEXT: vaddps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
40 %add.i = fadd <8 x float> %y, <float 4.500000e+00, float 0x400B333340000000, float 0x4002666660000000, float 0x3FF3333340000000, float 4.500000e+00, float 0x400B333340000000, float 0x4002666660000000, float 0x3FF3333340000000>
41 ret <8 x float> %add.i
44 define <4 x double> @subpd256(<4 x double> %y, <4 x double> %x) nounwind uwtable readnone ssp {
45 ; CHECK-LABEL: subpd256:
46 ; CHECK: ## %bb.0: ## %entry
47 ; CHECK-NEXT: vsubpd %ymm0, %ymm1, %ymm0
50 %sub.i = fsub <4 x double> %x, %y
51 ret <4 x double> %sub.i
54 define <4 x double> @subpd256fold(<4 x double> %y, ptr nocapture %x) nounwind uwtable readonly ssp {
55 ; CHECK-LABEL: subpd256fold:
56 ; CHECK: ## %bb.0: ## %entry
57 ; CHECK-NEXT: vsubpd (%rdi), %ymm0, %ymm0
60 %tmp2 = load <4 x double>, ptr %x, align 32
61 %sub.i = fsub <4 x double> %y, %tmp2
62 ret <4 x double> %sub.i
65 define <8 x float> @subps256(<8 x float> %y, <8 x float> %x) nounwind uwtable readnone ssp {
66 ; CHECK-LABEL: subps256:
67 ; CHECK: ## %bb.0: ## %entry
68 ; CHECK-NEXT: vsubps %ymm0, %ymm1, %ymm0
71 %sub.i = fsub <8 x float> %x, %y
72 ret <8 x float> %sub.i
75 define <8 x float> @subps256fold(<8 x float> %y, ptr nocapture %x) nounwind uwtable readonly ssp {
76 ; CHECK-LABEL: subps256fold:
77 ; CHECK: ## %bb.0: ## %entry
78 ; CHECK-NEXT: vsubps (%rdi), %ymm0, %ymm0
81 %tmp2 = load <8 x float>, ptr %x, align 32
82 %sub.i = fsub <8 x float> %y, %tmp2
83 ret <8 x float> %sub.i
86 define <4 x double> @mulpd256(<4 x double> %y, <4 x double> %x) nounwind uwtable readnone ssp {
87 ; CHECK-LABEL: mulpd256:
88 ; CHECK: ## %bb.0: ## %entry
89 ; CHECK-NEXT: vmulpd %ymm0, %ymm1, %ymm0
92 %mul.i = fmul <4 x double> %x, %y
93 ret <4 x double> %mul.i
96 define <4 x double> @mulpd256fold(<4 x double> %y) nounwind uwtable readnone ssp {
97 ; CHECK-LABEL: mulpd256fold:
98 ; CHECK: ## %bb.0: ## %entry
99 ; CHECK-NEXT: vmulpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
102 %mul.i = fmul <4 x double> %y, <double 4.500000e+00, double 3.400000e+00, double 2.300000e+00, double 1.200000e+00>
103 ret <4 x double> %mul.i
106 define <8 x float> @mulps256(<8 x float> %y, <8 x float> %x) nounwind uwtable readnone ssp {
107 ; CHECK-LABEL: mulps256:
108 ; CHECK: ## %bb.0: ## %entry
109 ; CHECK-NEXT: vmulps %ymm0, %ymm1, %ymm0
112 %mul.i = fmul <8 x float> %x, %y
113 ret <8 x float> %mul.i
116 define <8 x float> @mulps256fold(<8 x float> %y) nounwind uwtable readnone ssp {
117 ; CHECK-LABEL: mulps256fold:
118 ; CHECK: ## %bb.0: ## %entry
119 ; CHECK-NEXT: vmulps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
122 %mul.i = fmul <8 x float> %y, <float 4.500000e+00, float 0x400B333340000000, float 0x4002666660000000, float 0x3FF3333340000000, float 4.500000e+00, float 0x400B333340000000, float 0x4002666660000000, float 0x3FF3333340000000>
123 ret <8 x float> %mul.i
126 define <4 x double> @divpd256(<4 x double> %y, <4 x double> %x) nounwind uwtable readnone ssp {
127 ; CHECK-LABEL: divpd256:
128 ; CHECK: ## %bb.0: ## %entry
129 ; CHECK-NEXT: vdivpd %ymm0, %ymm1, %ymm0
132 %div.i = fdiv <4 x double> %x, %y
133 ret <4 x double> %div.i
136 define <4 x double> @divpd256fold(<4 x double> %y) nounwind uwtable readnone ssp {
137 ; CHECK-LABEL: divpd256fold:
138 ; CHECK: ## %bb.0: ## %entry
139 ; CHECK-NEXT: vdivpd {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
142 %div.i = fdiv <4 x double> %y, <double 4.500000e+00, double 3.400000e+00, double 2.300000e+00, double 1.200000e+00>
143 ret <4 x double> %div.i
146 define <8 x float> @divps256(<8 x float> %y, <8 x float> %x) nounwind uwtable readnone ssp {
147 ; CHECK-LABEL: divps256:
148 ; CHECK: ## %bb.0: ## %entry
149 ; CHECK-NEXT: vdivps %ymm0, %ymm1, %ymm0
152 %div.i = fdiv <8 x float> %x, %y
153 ret <8 x float> %div.i
156 define <8 x float> @divps256fold(<8 x float> %y) nounwind uwtable readnone ssp {
157 ; CHECK-LABEL: divps256fold:
158 ; CHECK: ## %bb.0: ## %entry
159 ; CHECK-NEXT: vdivps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm0, %ymm0
162 %div.i = fdiv <8 x float> %y, <float 4.500000e+00, float 0x400B333340000000, float 0x4002666660000000, float 0x3FF3333340000000, float 4.500000e+00, float 0x400B333340000000, float 0x4002666660000000, float 0x3FF3333340000000>
163 ret <8 x float> %div.i
166 define float @sqrtA(float %a) nounwind uwtable readnone ssp {
167 ; CHECK-LABEL: sqrtA:
168 ; CHECK: ## %bb.0: ## %entry
169 ; CHECK-NEXT: vsqrtss %xmm0, %xmm0, %xmm0
172 %conv1 = tail call float @sqrtf(float %a) nounwind readnone
176 declare double @sqrt(double) readnone
178 define double @sqrtB(double %a) nounwind uwtable readnone ssp {
179 ; CHECK-LABEL: sqrtB:
180 ; CHECK: ## %bb.0: ## %entry
181 ; CHECK-NEXT: vsqrtsd %xmm0, %xmm0, %xmm0
184 %call = tail call double @sqrt(double %a) nounwind readnone
188 declare float @sqrtf(float) readnone
191 define <4 x i64> @vpaddq(<4 x i64> %i, <4 x i64> %j) nounwind readnone {
192 ; CHECK-LABEL: vpaddq:
194 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
195 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
196 ; CHECK-NEXT: vpaddq %xmm2, %xmm3, %xmm2
197 ; CHECK-NEXT: vpaddq %xmm1, %xmm0, %xmm0
198 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
200 %x = add <4 x i64> %i, %j
204 define <8 x i32> @vpaddd(<8 x i32> %i, <8 x i32> %j) nounwind readnone {
205 ; CHECK-LABEL: vpaddd:
207 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
208 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
209 ; CHECK-NEXT: vpaddd %xmm2, %xmm3, %xmm2
210 ; CHECK-NEXT: vpaddd %xmm1, %xmm0, %xmm0
211 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
213 %x = add <8 x i32> %i, %j
217 define <16 x i16> @vpaddw(<16 x i16> %i, <16 x i16> %j) nounwind readnone {
218 ; CHECK-LABEL: vpaddw:
220 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
221 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
222 ; CHECK-NEXT: vpaddw %xmm2, %xmm3, %xmm2
223 ; CHECK-NEXT: vpaddw %xmm1, %xmm0, %xmm0
224 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
226 %x = add <16 x i16> %i, %j
230 define <32 x i8> @vpaddb(<32 x i8> %i, <32 x i8> %j) nounwind readnone {
231 ; CHECK-LABEL: vpaddb:
233 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
234 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
235 ; CHECK-NEXT: vpaddb %xmm2, %xmm3, %xmm2
236 ; CHECK-NEXT: vpaddb %xmm1, %xmm0, %xmm0
237 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
239 %x = add <32 x i8> %i, %j
243 define <4 x i64> @vpsubq(<4 x i64> %i, <4 x i64> %j) nounwind readnone {
244 ; CHECK-LABEL: vpsubq:
246 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
247 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
248 ; CHECK-NEXT: vpsubq %xmm2, %xmm3, %xmm2
249 ; CHECK-NEXT: vpsubq %xmm1, %xmm0, %xmm0
250 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
252 %x = sub <4 x i64> %i, %j
256 define <8 x i32> @vpsubd(<8 x i32> %i, <8 x i32> %j) nounwind readnone {
257 ; CHECK-LABEL: vpsubd:
259 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
260 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
261 ; CHECK-NEXT: vpsubd %xmm2, %xmm3, %xmm2
262 ; CHECK-NEXT: vpsubd %xmm1, %xmm0, %xmm0
263 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
265 %x = sub <8 x i32> %i, %j
269 define <16 x i16> @vpsubw(<16 x i16> %i, <16 x i16> %j) nounwind readnone {
270 ; CHECK-LABEL: vpsubw:
272 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
273 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
274 ; CHECK-NEXT: vpsubw %xmm2, %xmm3, %xmm2
275 ; CHECK-NEXT: vpsubw %xmm1, %xmm0, %xmm0
276 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
278 %x = sub <16 x i16> %i, %j
282 define <32 x i8> @vpsubb(<32 x i8> %i, <32 x i8> %j) nounwind readnone {
283 ; CHECK-LABEL: vpsubb:
285 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
286 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
287 ; CHECK-NEXT: vpsubb %xmm2, %xmm3, %xmm2
288 ; CHECK-NEXT: vpsubb %xmm1, %xmm0, %xmm0
289 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
291 %x = sub <32 x i8> %i, %j
295 define <8 x i32> @vpmulld(<8 x i32> %i, <8 x i32> %j) nounwind readnone {
296 ; CHECK-LABEL: vpmulld:
298 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
299 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
300 ; CHECK-NEXT: vpmulld %xmm2, %xmm3, %xmm2
301 ; CHECK-NEXT: vpmulld %xmm1, %xmm0, %xmm0
302 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
304 %x = mul <8 x i32> %i, %j
308 define <16 x i16> @vpmullw(<16 x i16> %i, <16 x i16> %j) nounwind readnone {
309 ; CHECK-LABEL: vpmullw:
311 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
312 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
313 ; CHECK-NEXT: vpmullw %xmm2, %xmm3, %xmm2
314 ; CHECK-NEXT: vpmullw %xmm1, %xmm0, %xmm0
315 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
317 %x = mul <16 x i16> %i, %j
321 define <4 x i64> @mul_v4i64(<4 x i64> %i, <4 x i64> %j) nounwind readnone {
322 ; CHECK-LABEL: mul_v4i64:
324 ; CHECK-NEXT: vextractf128 $1, %ymm1, %xmm2
325 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm3
326 ; CHECK-NEXT: vpsrlq $32, %xmm3, %xmm4
327 ; CHECK-NEXT: vpmuludq %xmm2, %xmm4, %xmm4
328 ; CHECK-NEXT: vpsrlq $32, %xmm2, %xmm5
329 ; CHECK-NEXT: vpmuludq %xmm5, %xmm3, %xmm5
330 ; CHECK-NEXT: vpaddq %xmm4, %xmm5, %xmm4
331 ; CHECK-NEXT: vpsllq $32, %xmm4, %xmm4
332 ; CHECK-NEXT: vpmuludq %xmm2, %xmm3, %xmm2
333 ; CHECK-NEXT: vpaddq %xmm4, %xmm2, %xmm2
334 ; CHECK-NEXT: vpsrlq $32, %xmm0, %xmm3
335 ; CHECK-NEXT: vpmuludq %xmm1, %xmm3, %xmm3
336 ; CHECK-NEXT: vpsrlq $32, %xmm1, %xmm4
337 ; CHECK-NEXT: vpmuludq %xmm4, %xmm0, %xmm4
338 ; CHECK-NEXT: vpaddq %xmm3, %xmm4, %xmm3
339 ; CHECK-NEXT: vpsllq $32, %xmm3, %xmm3
340 ; CHECK-NEXT: vpmuludq %xmm1, %xmm0, %xmm0
341 ; CHECK-NEXT: vpaddq %xmm3, %xmm0, %xmm0
342 ; CHECK-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0
344 %x = mul <4 x i64> %i, %j
348 declare <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float>) nounwind readnone
350 define <4 x float> @int_sqrt_ss() {
351 ; CHECK-LABEL: int_sqrt_ss:
353 ; CHECK-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
354 ; CHECK-NEXT: vsqrtss %xmm0, %xmm0, %xmm0
356 %x0 = load float, ptr addrspace(1) undef, align 8
357 %x1 = insertelement <4 x float> undef, float %x0, i32 0
358 %x2 = call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> %x1) nounwind
362 define <2 x double> @vector_sqrt_scalar_load(ptr %a0) optsize {
363 ; CHECK-LABEL: vector_sqrt_scalar_load:
365 ; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
366 ; CHECK-NEXT: vsqrtpd %xmm0, %xmm0
368 %a1 = load double, ptr %a0
369 %a2 = insertelement <2 x double> undef, double %a1, i32 0
370 %res = call <2 x double> @llvm.sqrt.v2f64(<2 x double> %a2) ; <<2 x double>> [#uses=1]
371 ret <2 x double> %res
373 declare <2 x double> @llvm.sqrt.v2f64(<2 x double>) nounwind readnone