1 ; RUN: llc -march=mips -mattr=+msa,+fp64 < %s | FileCheck %s
3 define void @add_v4f32(<4 x float>* %c, <4 x float>* %a, <4 x float>* %b) nounwind {
6 %1 = load <4 x float>* %a
7 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
8 %2 = load <4 x float>* %b
9 ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
10 %3 = fadd <4 x float> %1, %2
11 ; CHECK-DAG: fadd.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
12 store <4 x float> %3, <4 x float>* %c
13 ; CHECK-DAG: st.w [[R3]], 0($4)
16 ; CHECK: .size add_v4f32
19 define void @add_v2f64(<2 x double>* %c, <2 x double>* %a, <2 x double>* %b) nounwind {
22 %1 = load <2 x double>* %a
23 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
24 %2 = load <2 x double>* %b
25 ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
26 %3 = fadd <2 x double> %1, %2
27 ; CHECK-DAG: fadd.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
28 store <2 x double> %3, <2 x double>* %c
29 ; CHECK-DAG: st.d [[R3]], 0($4)
32 ; CHECK: .size add_v2f64
35 define void @sub_v4f32(<4 x float>* %c, <4 x float>* %a, <4 x float>* %b) nounwind {
38 %1 = load <4 x float>* %a
39 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
40 %2 = load <4 x float>* %b
41 ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
42 %3 = fsub <4 x float> %1, %2
43 ; CHECK-DAG: fsub.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
44 store <4 x float> %3, <4 x float>* %c
45 ; CHECK-DAG: st.w [[R3]], 0($4)
48 ; CHECK: .size sub_v4f32
51 define void @sub_v2f64(<2 x double>* %c, <2 x double>* %a, <2 x double>* %b) nounwind {
54 %1 = load <2 x double>* %a
55 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
56 %2 = load <2 x double>* %b
57 ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
58 %3 = fsub <2 x double> %1, %2
59 ; CHECK-DAG: fsub.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
60 store <2 x double> %3, <2 x double>* %c
61 ; CHECK-DAG: st.d [[R3]], 0($4)
64 ; CHECK: .size sub_v2f64
67 define void @mul_v4f32(<4 x float>* %c, <4 x float>* %a, <4 x float>* %b) nounwind {
70 %1 = load <4 x float>* %a
71 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
72 %2 = load <4 x float>* %b
73 ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
74 %3 = fmul <4 x float> %1, %2
75 ; CHECK-DAG: fmul.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
76 store <4 x float> %3, <4 x float>* %c
77 ; CHECK-DAG: st.w [[R3]], 0($4)
80 ; CHECK: .size mul_v4f32
83 define void @mul_v2f64(<2 x double>* %c, <2 x double>* %a, <2 x double>* %b) nounwind {
86 %1 = load <2 x double>* %a
87 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
88 %2 = load <2 x double>* %b
89 ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
90 %3 = fmul <2 x double> %1, %2
91 ; CHECK-DAG: fmul.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
92 store <2 x double> %3, <2 x double>* %c
93 ; CHECK-DAG: st.d [[R3]], 0($4)
96 ; CHECK: .size mul_v2f64
99 define void @fma_v4f32(<4 x float>* %d, <4 x float>* %a, <4 x float>* %b,
100 <4 x float>* %c) nounwind {
103 %1 = load <4 x float>* %a
104 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
105 %2 = load <4 x float>* %b
106 ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
107 %3 = load <4 x float>* %c
108 ; CHECK-DAG: ld.w [[R3:\$w[0-9]+]], 0($7)
109 %4 = tail call <4 x float> @llvm.fma.v4f32 (<4 x float> %1, <4 x float> %2,
111 ; CHECK-DAG: fmadd.w [[R1]], [[R2]], [[R3]]
112 store <4 x float> %4, <4 x float>* %d
113 ; CHECK-DAG: st.w [[R1]], 0($4)
116 ; CHECK: .size fma_v4f32
119 define void @fma_v2f64(<2 x double>* %d, <2 x double>* %a, <2 x double>* %b,
120 <2 x double>* %c) nounwind {
123 %1 = load <2 x double>* %a
124 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
125 %2 = load <2 x double>* %b
126 ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
127 %3 = load <2 x double>* %c
128 ; CHECK-DAG: ld.d [[R3:\$w[0-9]+]], 0($7)
129 %4 = tail call <2 x double> @llvm.fma.v2f64 (<2 x double> %1, <2 x double> %2,
131 ; CHECK-DAG: fmadd.d [[R1]], [[R2]], [[R3]]
132 store <2 x double> %4, <2 x double>* %d
133 ; CHECK-DAG: st.d [[R1]], 0($4)
136 ; CHECK: .size fma_v2f64
139 define void @fmsub_v4f32(<4 x float>* %d, <4 x float>* %a, <4 x float>* %b,
140 <4 x float>* %c) nounwind {
141 ; CHECK: fmsub_v4f32:
143 %1 = load <4 x float>* %a
144 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
145 %2 = load <4 x float>* %b
146 ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
147 %3 = load <4 x float>* %c
148 ; CHECK-DAG: ld.w [[R3:\$w[0-9]+]], 0($7)
149 %4 = fmul <4 x float> %2, %3
150 %5 = fsub <4 x float> %1, %4
151 ; CHECK-DAG: fmsub.w [[R1]], [[R2]], [[R3]]
152 store <4 x float> %5, <4 x float>* %d
153 ; CHECK-DAG: st.w [[R1]], 0($4)
156 ; CHECK: .size fmsub_v4f32
159 define void @fmsub_v2f64(<2 x double>* %d, <2 x double>* %a, <2 x double>* %b,
160 <2 x double>* %c) nounwind {
161 ; CHECK: fmsub_v2f64:
163 %1 = load <2 x double>* %a
164 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
165 %2 = load <2 x double>* %b
166 ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
167 %3 = load <2 x double>* %c
168 ; CHECK-DAG: ld.d [[R3:\$w[0-9]+]], 0($7)
169 %4 = fmul <2 x double> %2, %3
170 %5 = fsub <2 x double> %1, %4
171 ; CHECK-DAG: fmsub.d [[R1]], [[R2]], [[R3]]
172 store <2 x double> %5, <2 x double>* %d
173 ; CHECK-DAG: st.d [[R1]], 0($4)
176 ; CHECK: .size fmsub_v2f64
179 define void @fdiv_v4f32(<4 x float>* %c, <4 x float>* %a, <4 x float>* %b) nounwind {
182 %1 = load <4 x float>* %a
183 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
184 %2 = load <4 x float>* %b
185 ; CHECK-DAG: ld.w [[R2:\$w[0-9]+]], 0($6)
186 %3 = fdiv <4 x float> %1, %2
187 ; CHECK-DAG: fdiv.w [[R3:\$w[0-9]+]], [[R1]], [[R2]]
188 store <4 x float> %3, <4 x float>* %c
189 ; CHECK-DAG: st.w [[R3]], 0($4)
192 ; CHECK: .size fdiv_v4f32
195 define void @fdiv_v2f64(<2 x double>* %c, <2 x double>* %a, <2 x double>* %b) nounwind {
198 %1 = load <2 x double>* %a
199 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
200 %2 = load <2 x double>* %b
201 ; CHECK-DAG: ld.d [[R2:\$w[0-9]+]], 0($6)
202 %3 = fdiv <2 x double> %1, %2
203 ; CHECK-DAG: fdiv.d [[R3:\$w[0-9]+]], [[R1]], [[R2]]
204 store <2 x double> %3, <2 x double>* %c
205 ; CHECK-DAG: st.d [[R3]], 0($4)
208 ; CHECK: .size fdiv_v2f64
211 define void @fabs_v4f32(<4 x float>* %c, <4 x float>* %a) nounwind {
214 %1 = load <4 x float>* %a
215 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
216 %2 = tail call <4 x float> @llvm.fabs.v4f32 (<4 x float> %1)
217 ; CHECK-DAG: fmax_a.w [[R3:\$w[0-9]+]], [[R1]], [[R1]]
218 store <4 x float> %2, <4 x float>* %c
219 ; CHECK-DAG: st.w [[R3]], 0($4)
222 ; CHECK: .size fabs_v4f32
225 define void @fabs_v2f64(<2 x double>* %c, <2 x double>* %a) nounwind {
228 %1 = load <2 x double>* %a
229 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
230 %2 = tail call <2 x double> @llvm.fabs.v2f64 (<2 x double> %1)
231 ; CHECK-DAG: fmax_a.d [[R3:\$w[0-9]+]], [[R1]], [[R1]]
232 store <2 x double> %2, <2 x double>* %c
233 ; CHECK-DAG: st.d [[R3]], 0($4)
236 ; CHECK: .size fabs_v2f64
239 define void @fsqrt_v4f32(<4 x float>* %c, <4 x float>* %a) nounwind {
240 ; CHECK: fsqrt_v4f32:
242 %1 = load <4 x float>* %a
243 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
244 %2 = tail call <4 x float> @llvm.sqrt.v4f32 (<4 x float> %1)
245 ; CHECK-DAG: fsqrt.w [[R3:\$w[0-9]+]], [[R1]]
246 store <4 x float> %2, <4 x float>* %c
247 ; CHECK-DAG: st.w [[R3]], 0($4)
250 ; CHECK: .size fsqrt_v4f32
253 define void @fsqrt_v2f64(<2 x double>* %c, <2 x double>* %a) nounwind {
254 ; CHECK: fsqrt_v2f64:
256 %1 = load <2 x double>* %a
257 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
258 %2 = tail call <2 x double> @llvm.sqrt.v2f64 (<2 x double> %1)
259 ; CHECK-DAG: fsqrt.d [[R3:\$w[0-9]+]], [[R1]]
260 store <2 x double> %2, <2 x double>* %c
261 ; CHECK-DAG: st.d [[R3]], 0($4)
264 ; CHECK: .size fsqrt_v2f64
267 define void @ffint_u_v4f32(<4 x float>* %c, <4 x i32>* %a) nounwind {
268 ; CHECK: ffint_u_v4f32:
270 %1 = load <4 x i32>* %a
271 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
272 %2 = uitofp <4 x i32> %1 to <4 x float>
273 ; CHECK-DAG: ffint_u.w [[R3:\$w[0-9]+]], [[R1]]
274 store <4 x float> %2, <4 x float>* %c
275 ; CHECK-DAG: st.w [[R3]], 0($4)
278 ; CHECK: .size ffint_u_v4f32
281 define void @ffint_u_v2f64(<2 x double>* %c, <2 x i64>* %a) nounwind {
282 ; CHECK: ffint_u_v2f64:
284 %1 = load <2 x i64>* %a
285 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
286 %2 = uitofp <2 x i64> %1 to <2 x double>
287 ; CHECK-DAG: ffint_u.d [[R3:\$w[0-9]+]], [[R1]]
288 store <2 x double> %2, <2 x double>* %c
289 ; CHECK-DAG: st.d [[R3]], 0($4)
292 ; CHECK: .size ffint_u_v2f64
295 define void @ffint_s_v4f32(<4 x float>* %c, <4 x i32>* %a) nounwind {
296 ; CHECK: ffint_s_v4f32:
298 %1 = load <4 x i32>* %a
299 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
300 %2 = sitofp <4 x i32> %1 to <4 x float>
301 ; CHECK-DAG: ffint_s.w [[R3:\$w[0-9]+]], [[R1]]
302 store <4 x float> %2, <4 x float>* %c
303 ; CHECK-DAG: st.w [[R3]], 0($4)
306 ; CHECK: .size ffint_s_v4f32
309 define void @ffint_s_v2f64(<2 x double>* %c, <2 x i64>* %a) nounwind {
310 ; CHECK: ffint_s_v2f64:
312 %1 = load <2 x i64>* %a
313 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
314 %2 = sitofp <2 x i64> %1 to <2 x double>
315 ; CHECK-DAG: ffint_s.d [[R3:\$w[0-9]+]], [[R1]]
316 store <2 x double> %2, <2 x double>* %c
317 ; CHECK-DAG: st.d [[R3]], 0($4)
320 ; CHECK: .size ffint_s_v2f64
323 define void @ftrunc_u_v4f32(<4 x i32>* %c, <4 x float>* %a) nounwind {
324 ; CHECK: ftrunc_u_v4f32:
326 %1 = load <4 x float>* %a
327 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
328 %2 = fptoui <4 x float> %1 to <4 x i32>
329 ; CHECK-DAG: ftrunc_u.w [[R3:\$w[0-9]+]], [[R1]]
330 store <4 x i32> %2, <4 x i32>* %c
331 ; CHECK-DAG: st.w [[R3]], 0($4)
334 ; CHECK: .size ftrunc_u_v4f32
337 define void @ftrunc_u_v2f64(<2 x i64>* %c, <2 x double>* %a) nounwind {
338 ; CHECK: ftrunc_u_v2f64:
340 %1 = load <2 x double>* %a
341 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
342 %2 = fptoui <2 x double> %1 to <2 x i64>
343 ; CHECK-DAG: ftrunc_u.d [[R3:\$w[0-9]+]], [[R1]]
344 store <2 x i64> %2, <2 x i64>* %c
345 ; CHECK-DAG: st.d [[R3]], 0($4)
348 ; CHECK: .size ftrunc_u_v2f64
351 define void @ftrunc_s_v4f32(<4 x i32>* %c, <4 x float>* %a) nounwind {
352 ; CHECK: ftrunc_s_v4f32:
354 %1 = load <4 x float>* %a
355 ; CHECK-DAG: ld.w [[R1:\$w[0-9]+]], 0($5)
356 %2 = fptosi <4 x float> %1 to <4 x i32>
357 ; CHECK-DAG: ftrunc_s.w [[R3:\$w[0-9]+]], [[R1]]
358 store <4 x i32> %2, <4 x i32>* %c
359 ; CHECK-DAG: st.w [[R3]], 0($4)
362 ; CHECK: .size ftrunc_s_v4f32
365 define void @ftrunc_s_v2f64(<2 x i64>* %c, <2 x double>* %a) nounwind {
366 ; CHECK: ftrunc_s_v2f64:
368 %1 = load <2 x double>* %a
369 ; CHECK-DAG: ld.d [[R1:\$w[0-9]+]], 0($5)
370 %2 = fptosi <2 x double> %1 to <2 x i64>
371 ; CHECK-DAG: ftrunc_s.d [[R3:\$w[0-9]+]], [[R1]]
372 store <2 x i64> %2, <2 x i64>* %c
373 ; CHECK-DAG: st.d [[R3]], 0($4)
376 ; CHECK: .size ftrunc_s_v2f64
379 declare <4 x float> @llvm.fabs.v4f32(<4 x float> %Val)
380 declare <2 x double> @llvm.fabs.v2f64(<2 x double> %Val)
381 declare <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b,
383 declare <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b,
385 declare <4 x float> @llvm.sqrt.v4f32(<4 x float> %Val)
386 declare <2 x double> @llvm.sqrt.v2f64(<2 x double> %Val)