1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc -verify-machineinstrs -mtriple=powerpc64le-unknown-linux-gnu \
3 ; RUN: -mcpu=pwr10 -ppc-asm-full-reg-names \
4 ; RUN: -ppc-vsr-nums-as-vr < %s | FileCheck %s
5 ; RUN: llc -verify-machineinstrs -mtriple=powerpc64-unknown-linux-gnu \
6 ; RUN: -mcpu=pwr10 -ppc-asm-full-reg-names \
7 ; RUN: -ppc-vsr-nums-as-vr < %s | FileCheck %s --check-prefix=CHECK-BE
10 declare <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8>, <16 x i8>, <16 x i8>, <16 x i8>)
11 define void @ass_acc(ptr %ptr, <16 x i8> %vc) {
12 ; CHECK-LABEL: ass_acc:
13 ; CHECK: # %bb.0: # %entry
14 ; CHECK-NEXT: xxlor vs3, v2, v2
15 ; CHECK-NEXT: xxlor vs2, v2, v2
16 ; CHECK-NEXT: xxlor vs0, vs2, vs2
17 ; CHECK-NEXT: xxlor vs1, vs3, vs3
18 ; CHECK-NEXT: stxv vs0, 48(r3)
19 ; CHECK-NEXT: stxv vs1, 32(r3)
20 ; CHECK-NEXT: stxv vs2, 16(r3)
21 ; CHECK-NEXT: stxv vs3, 0(r3)
24 ; CHECK-BE-LABEL: ass_acc:
25 ; CHECK-BE: # %bb.0: # %entry
26 ; CHECK-BE-NEXT: xxlor vs3, v2, v2
27 ; CHECK-BE-NEXT: xxlor vs2, v2, v2
28 ; CHECK-BE-NEXT: xxlor vs0, vs2, vs2
29 ; CHECK-BE-NEXT: xxlor vs1, vs3, vs3
30 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
31 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
32 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
33 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
36 %0 = tail call <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8> %vc, <16 x i8> %vc, <16 x i8> %vc, <16 x i8> %vc)
37 store <512 x i1> %0, ptr %ptr, align 64
42 declare <512 x i1> @llvm.ppc.mma.xxmtacc(<512 x i1>)
43 define void @int_xxmtacc(ptr %ptr, <16 x i8> %vc) {
44 ; CHECK-LABEL: int_xxmtacc:
45 ; CHECK: # %bb.0: # %entry
46 ; CHECK-NEXT: xxlor vs3, v2, v2
47 ; CHECK-NEXT: xxlor vs2, v2, v2
48 ; CHECK-NEXT: xxlor vs0, vs2, vs2
49 ; CHECK-NEXT: xxlor vs1, vs3, vs3
50 ; CHECK-NEXT: xxmtacc acc0
51 ; CHECK-NEXT: stxv vs0, 48(r3)
52 ; CHECK-NEXT: stxv vs1, 32(r3)
53 ; CHECK-NEXT: stxv vs2, 16(r3)
54 ; CHECK-NEXT: stxv vs3, 0(r3)
57 ; CHECK-BE-LABEL: int_xxmtacc:
58 ; CHECK-BE: # %bb.0: # %entry
59 ; CHECK-BE-NEXT: xxlor vs3, v2, v2
60 ; CHECK-BE-NEXT: xxlor vs2, v2, v2
61 ; CHECK-BE-NEXT: xxlor vs0, vs2, vs2
62 ; CHECK-BE-NEXT: xxlor vs1, vs3, vs3
63 ; CHECK-BE-NEXT: xxmtacc acc0
64 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
65 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
66 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
67 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
70 ; One xxmtacc is generated from the call to assemble.acc then one xxmtacc is
71 ; generated from the call to xxmtacc then one xxmfacc is generated for the store
72 %0 = tail call <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8> %vc, <16 x i8> %vc, <16 x i8> %vc, <16 x i8> %vc)
73 %1 = tail call <512 x i1> @llvm.ppc.mma.xxmtacc(<512 x i1> %0)
74 store <512 x i1> %1, ptr %ptr, align 64
79 declare <512 x i1> @llvm.ppc.mma.xxmfacc(<512 x i1>)
80 define void @int_xxmfacc(ptr %ptr, <16 x i8> %vc) {
81 ; CHECK-LABEL: int_xxmfacc:
82 ; CHECK: # %bb.0: # %entry
83 ; CHECK-NEXT: xxlor vs3, v2, v2
84 ; CHECK-NEXT: xxlor vs2, v2, v2
85 ; CHECK-NEXT: xxlor vs0, vs2, vs2
86 ; CHECK-NEXT: xxlor vs1, vs3, vs3
87 ; CHECK-NEXT: stxv vs0, 48(r3)
88 ; CHECK-NEXT: stxv vs1, 32(r3)
89 ; CHECK-NEXT: stxv vs2, 16(r3)
90 ; CHECK-NEXT: stxv vs3, 0(r3)
93 ; CHECK-BE-LABEL: int_xxmfacc:
94 ; CHECK-BE: # %bb.0: # %entry
95 ; CHECK-BE-NEXT: xxlor vs3, v2, v2
96 ; CHECK-BE-NEXT: xxlor vs2, v2, v2
97 ; CHECK-BE-NEXT: xxlor vs0, vs2, vs2
98 ; CHECK-BE-NEXT: xxlor vs1, vs3, vs3
99 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
100 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
101 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
102 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
105 ; One xxmtacc is generated from the call to assemble.acc then one xxmfacc is
106 ; generated from the call to xxmfacc then one xxmfacc is generated for the store
107 %0 = tail call <512 x i1> @llvm.ppc.mma.assemble.acc(<16 x i8> %vc, <16 x i8> %vc, <16 x i8> %vc, <16 x i8> %vc)
108 %1 = tail call <512 x i1> @llvm.ppc.mma.xxmfacc(<512 x i1> %0)
109 store <512 x i1> %1, ptr %ptr, align 64
114 declare <512 x i1> @llvm.ppc.mma.xxsetaccz()
115 define void @int_xxsetaccz(ptr %ptr) {
116 ; CHECK-LABEL: int_xxsetaccz:
117 ; CHECK: # %bb.0: # %entry
118 ; CHECK-NEXT: xxsetaccz acc0
119 ; CHECK-NEXT: xxmfacc acc0
120 ; CHECK-NEXT: stxv vs0, 48(r3)
121 ; CHECK-NEXT: stxv vs1, 32(r3)
122 ; CHECK-NEXT: stxv vs2, 16(r3)
123 ; CHECK-NEXT: stxv vs3, 0(r3)
126 ; CHECK-BE-LABEL: int_xxsetaccz:
127 ; CHECK-BE: # %bb.0: # %entry
128 ; CHECK-BE-NEXT: xxsetaccz acc0
129 ; CHECK-BE-NEXT: xxmfacc acc0
130 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
131 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
132 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
133 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
136 %0 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
137 store <512 x i1> %0, ptr %ptr, align 64
142 declare { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } @llvm.ppc.mma.disassemble.acc(<512 x i1>)
143 define void @disass_acc(ptr %ptr1, ptr %ptr2, ptr %ptr3, ptr %ptr4) {
144 ; CHECK-LABEL: disass_acc:
145 ; CHECK: # %bb.0: # %entry
146 ; CHECK-NEXT: xxsetaccz acc0
147 ; CHECK-NEXT: xxmfacc acc0
148 ; CHECK-NEXT: stxv vs3, 0(r3)
149 ; CHECK-NEXT: stxv vs2, 0(r4)
150 ; CHECK-NEXT: stxv vs1, 0(r5)
151 ; CHECK-NEXT: stxv vs0, 0(r6)
154 ; CHECK-BE-LABEL: disass_acc:
155 ; CHECK-BE: # %bb.0: # %entry
156 ; CHECK-BE-NEXT: xxsetaccz acc0
157 ; CHECK-BE-NEXT: xxmfacc acc0
158 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
159 ; CHECK-BE-NEXT: stxv vs1, 0(r4)
160 ; CHECK-BE-NEXT: stxv vs2, 0(r5)
161 ; CHECK-BE-NEXT: stxv vs3, 0(r6)
164 %0 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
165 %1 = tail call { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } @llvm.ppc.mma.disassemble.acc(<512 x i1> %0)
166 %2 = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } %1, 0
167 %3 = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } %1, 1
168 %4 = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } %1, 2
169 %5 = extractvalue { <16 x i8>, <16 x i8>, <16 x i8>, <16 x i8> } %1, 3
170 store <16 x i8> %2, ptr %ptr1, align 16
171 store <16 x i8> %3, ptr %ptr2, align 16
172 store <16 x i8> %4, ptr %ptr3, align 16
173 store <16 x i8> %5, ptr %ptr4, align 16
177 declare <512 x i1> @llvm.ppc.mma.xvi4ger8pp(<512 x i1>, <16 x i8>, <16 x i8>)
178 define void @testBranch(ptr %ptr, <16 x i8> %vc, i32 %val) {
179 ; CHECK-LABEL: testBranch:
180 ; CHECK: # %bb.0: # %entry
181 ; CHECK-NEXT: cmplwi r7, 0
182 ; CHECK-NEXT: beq cr0, .LBB5_2
183 ; CHECK-NEXT: # %bb.1: # %if.then
184 ; CHECK-NEXT: xxsetaccz acc0
185 ; CHECK-NEXT: b .LBB5_3
186 ; CHECK-NEXT: .LBB5_2: # %if.else
187 ; CHECK-NEXT: lxv vs1, 32(r3)
188 ; CHECK-NEXT: lxv vs0, 48(r3)
189 ; CHECK-NEXT: lxv vs3, 0(r3)
190 ; CHECK-NEXT: lxv vs2, 16(r3)
191 ; CHECK-NEXT: xxmtacc acc0
192 ; CHECK-NEXT: xvi4ger8pp acc0, v2, v2
193 ; CHECK-NEXT: .LBB5_3: # %if.end
194 ; CHECK-NEXT: xxmfacc acc0
195 ; CHECK-NEXT: stxv vs0, 48(r3)
196 ; CHECK-NEXT: stxv vs1, 32(r3)
197 ; CHECK-NEXT: stxv vs2, 16(r3)
198 ; CHECK-NEXT: stxv vs3, 0(r3)
201 ; CHECK-BE-LABEL: testBranch:
202 ; CHECK-BE: # %bb.0: # %entry
203 ; CHECK-BE-NEXT: cmplwi r7, 0
204 ; CHECK-BE-NEXT: beq cr0, .LBB5_2
205 ; CHECK-BE-NEXT: # %bb.1: # %if.then
206 ; CHECK-BE-NEXT: xxsetaccz acc0
207 ; CHECK-BE-NEXT: b .LBB5_3
208 ; CHECK-BE-NEXT: .LBB5_2: # %if.else
209 ; CHECK-BE-NEXT: lxv vs1, 16(r3)
210 ; CHECK-BE-NEXT: lxv vs0, 0(r3)
211 ; CHECK-BE-NEXT: lxv vs3, 48(r3)
212 ; CHECK-BE-NEXT: lxv vs2, 32(r3)
213 ; CHECK-BE-NEXT: xxmtacc acc0
214 ; CHECK-BE-NEXT: xvi4ger8pp acc0, v2, v2
215 ; CHECK-BE-NEXT: .LBB5_3: # %if.end
216 ; CHECK-BE-NEXT: xxmfacc acc0
217 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
218 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
219 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
220 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
223 %tobool = icmp eq i32 %val, 0
224 br i1 %tobool, label %if.else, label %if.then
227 %0 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
231 %1 = load <512 x i1>, ptr %ptr, align 64
232 %2 = tail call <512 x i1> @llvm.ppc.mma.xvi4ger8pp(<512 x i1> %1, <16 x i8> %vc, <16 x i8> %vc)
236 %vq1.0 = phi <512 x i1> [ %0, %if.then ], [ %2, %if.else ]
237 store <512 x i1> %vq1.0, ptr %ptr, align 64
241 ; The following test cases check that the xxsetaccz instruction is correctly rematerialized
242 declare <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1>, <16 x i8>, <16 x i8>)
243 declare <512 x i1> @llvm.ppc.mma.xvf32gerpn(<512 x i1>, <16 x i8>, <16 x i8>)
244 declare <512 x i1> @llvm.ppc.mma.xvf32gernp(<512 x i1>, <16 x i8>, <16 x i8>)
246 define void @testcse(ptr %res, <16 x i8> %vc) {
247 ; CHECK-LABEL: testcse:
248 ; CHECK: # %bb.0: # %entry
249 ; CHECK-NEXT: xxsetaccz acc0
250 ; CHECK-NEXT: xvf32gerpp acc0, v2, v2
251 ; CHECK-NEXT: xxmfacc acc0
252 ; CHECK-NEXT: stxv vs0, 48(r3)
253 ; CHECK-NEXT: stxv vs1, 32(r3)
254 ; CHECK-NEXT: stxv vs2, 16(r3)
255 ; CHECK-NEXT: stxv vs3, 0(r3)
256 ; CHECK-NEXT: stxv vs0, 112(r3)
257 ; CHECK-NEXT: stxv vs1, 96(r3)
258 ; CHECK-NEXT: stxv vs2, 80(r3)
259 ; CHECK-NEXT: stxv vs3, 64(r3)
262 ; CHECK-BE-LABEL: testcse:
263 ; CHECK-BE: # %bb.0: # %entry
264 ; CHECK-BE-NEXT: xxsetaccz acc0
265 ; CHECK-BE-NEXT: xvf32gerpp acc0, v2, v2
266 ; CHECK-BE-NEXT: xxmfacc acc0
267 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
268 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
269 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
270 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
271 ; CHECK-BE-NEXT: stxv vs1, 80(r3)
272 ; CHECK-BE-NEXT: stxv vs0, 64(r3)
273 ; CHECK-BE-NEXT: stxv vs3, 112(r3)
274 ; CHECK-BE-NEXT: stxv vs2, 96(r3)
277 %0 = call <512 x i1> @llvm.ppc.mma.xxsetaccz()
278 %1 = call <512 x i1> @llvm.ppc.mma.xxsetaccz()
279 %2 = call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> %0, <16 x i8> %vc, <16 x i8> %vc)
280 %3 = call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> %1, <16 x i8> %vc, <16 x i8> %vc)
281 %4 = getelementptr inbounds <512 x i1>, ptr %res, i64 1
282 store <512 x i1> %2, ptr %res, align 64
283 store <512 x i1> %3, ptr %4, align 64
287 define void @testcse2(ptr %res, <16 x i8> %vc) {
288 ; CHECK-LABEL: testcse2:
289 ; CHECK: # %bb.0: # %entry
290 ; CHECK-NEXT: xxsetaccz acc0
291 ; CHECK-NEXT: xxsetaccz acc1
292 ; CHECK-NEXT: xvf32gerpp acc1, v2, v2
293 ; CHECK-NEXT: xvf32gerpn acc0, v2, v2
294 ; CHECK-NEXT: xxmfacc acc1
295 ; CHECK-NEXT: xxmfacc acc0
296 ; CHECK-NEXT: stxv vs4, 48(r3)
297 ; CHECK-NEXT: stxv vs5, 32(r3)
298 ; CHECK-NEXT: stxv vs6, 16(r3)
299 ; CHECK-NEXT: stxv vs7, 0(r3)
300 ; CHECK-NEXT: stxv vs0, 112(r3)
301 ; CHECK-NEXT: stxv vs1, 96(r3)
302 ; CHECK-NEXT: stxv vs2, 80(r3)
303 ; CHECK-NEXT: stxv vs3, 64(r3)
306 ; CHECK-BE-LABEL: testcse2:
307 ; CHECK-BE: # %bb.0: # %entry
308 ; CHECK-BE-NEXT: xxsetaccz acc0
309 ; CHECK-BE-NEXT: xxsetaccz acc1
310 ; CHECK-BE-NEXT: xvf32gerpp acc1, v2, v2
311 ; CHECK-BE-NEXT: xvf32gerpn acc0, v2, v2
312 ; CHECK-BE-NEXT: xxmfacc acc1
313 ; CHECK-BE-NEXT: xxmfacc acc0
314 ; CHECK-BE-NEXT: stxv vs5, 16(r3)
315 ; CHECK-BE-NEXT: stxv vs4, 0(r3)
316 ; CHECK-BE-NEXT: stxv vs7, 48(r3)
317 ; CHECK-BE-NEXT: stxv vs6, 32(r3)
318 ; CHECK-BE-NEXT: stxv vs1, 80(r3)
319 ; CHECK-BE-NEXT: stxv vs0, 64(r3)
320 ; CHECK-BE-NEXT: stxv vs3, 112(r3)
321 ; CHECK-BE-NEXT: stxv vs2, 96(r3)
324 %0 = call <512 x i1> @llvm.ppc.mma.xxsetaccz()
325 %1 = call <512 x i1> @llvm.ppc.mma.xxsetaccz()
326 %2 = call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> %0, <16 x i8> %vc, <16 x i8> %vc)
327 %3 = call <512 x i1> @llvm.ppc.mma.xvf32gerpn(<512 x i1> %1, <16 x i8> %vc, <16 x i8> %vc)
328 %4 = getelementptr inbounds <512 x i1>, ptr %res, i64 1
329 store <512 x i1> %2, ptr %res, align 64
330 store <512 x i1> %3, ptr %4, align 64
334 define void @testcse3(ptr %res, <16 x i8> %vc) {
335 ; CHECK-LABEL: testcse3:
336 ; CHECK: # %bb.0: # %entry
337 ; CHECK-NEXT: xxsetaccz acc0
338 ; CHECK-NEXT: xxsetaccz acc1
339 ; CHECK-NEXT: xvf32gerpp acc1, v2, v2
340 ; CHECK-NEXT: xvf32gerpn acc0, v2, v2
341 ; CHECK-NEXT: xxmfacc acc1
342 ; CHECK-NEXT: xxmfacc acc0
343 ; CHECK-NEXT: stxv vs4, 48(r3)
344 ; CHECK-NEXT: stxv vs5, 32(r3)
345 ; CHECK-NEXT: stxv vs6, 16(r3)
346 ; CHECK-NEXT: stxv vs7, 0(r3)
347 ; CHECK-NEXT: stxv vs0, 112(r3)
348 ; CHECK-NEXT: stxv vs1, 96(r3)
349 ; CHECK-NEXT: stxv vs2, 80(r3)
350 ; CHECK-NEXT: stxv vs3, 64(r3)
353 ; CHECK-BE-LABEL: testcse3:
354 ; CHECK-BE: # %bb.0: # %entry
355 ; CHECK-BE-NEXT: xxsetaccz acc0
356 ; CHECK-BE-NEXT: xxsetaccz acc1
357 ; CHECK-BE-NEXT: xvf32gerpp acc1, v2, v2
358 ; CHECK-BE-NEXT: xvf32gerpn acc0, v2, v2
359 ; CHECK-BE-NEXT: xxmfacc acc1
360 ; CHECK-BE-NEXT: xxmfacc acc0
361 ; CHECK-BE-NEXT: stxv vs5, 16(r3)
362 ; CHECK-BE-NEXT: stxv vs4, 0(r3)
363 ; CHECK-BE-NEXT: stxv vs7, 48(r3)
364 ; CHECK-BE-NEXT: stxv vs6, 32(r3)
365 ; CHECK-BE-NEXT: stxv vs1, 80(r3)
366 ; CHECK-BE-NEXT: stxv vs0, 64(r3)
367 ; CHECK-BE-NEXT: stxv vs3, 112(r3)
368 ; CHECK-BE-NEXT: stxv vs2, 96(r3)
371 %0 = call <512 x i1> @llvm.ppc.mma.xxsetaccz()
372 %1 = call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> %0, <16 x i8> %vc, <16 x i8> %vc)
373 %2 = call <512 x i1> @llvm.ppc.mma.xvf32gerpn(<512 x i1> %0, <16 x i8> %vc, <16 x i8> %vc)
374 %3 = getelementptr inbounds <512 x i1>, ptr %res, i64 1
375 store <512 x i1> %1, ptr %res, align 64
376 store <512 x i1> %2, ptr %3, align 64
380 define void @testcse4(ptr %res, i32 %lim, ptr %vc) {
381 ; CHECK-LABEL: testcse4:
382 ; CHECK: # %bb.0: # %entry
383 ; CHECK-NEXT: cmpwi r4, 1
384 ; CHECK-NEXT: bltlr cr0
385 ; CHECK-NEXT: # %bb.1: # %for.body.preheader
386 ; CHECK-NEXT: clrldi r4, r4, 32
387 ; CHECK-NEXT: li r6, 0
388 ; CHECK-NEXT: mtctr r4
389 ; CHECK-NEXT: li r4, 0
390 ; CHECK-NEXT: .p2align 4
391 ; CHECK-NEXT: .LBB9_2: # %for.body
393 ; CHECK-NEXT: rldic r7, r6, 4, 28
394 ; CHECK-NEXT: xxsetaccz acc2
395 ; CHECK-NEXT: xxsetaccz acc1
396 ; CHECK-NEXT: addi r6, r6, 6
397 ; CHECK-NEXT: lxvx vs0, r5, r7
398 ; CHECK-NEXT: add r7, r5, r7
399 ; CHECK-NEXT: lxv vs1, 16(r7)
400 ; CHECK-NEXT: xvf32gerpp acc2, vs0, vs1
401 ; CHECK-NEXT: lxv vs0, 32(r7)
402 ; CHECK-NEXT: lxv vs1, 48(r7)
403 ; CHECK-NEXT: xvf32gerpn acc1, vs0, vs1
404 ; CHECK-NEXT: lxv vs12, 64(r7)
405 ; CHECK-NEXT: lxv vs13, 80(r7)
406 ; CHECK-NEXT: xxsetaccz acc0
407 ; CHECK-NEXT: rldic r7, r4, 6, 26
408 ; CHECK-NEXT: addi r4, r4, 3
409 ; CHECK-NEXT: add r8, r3, r7
410 ; CHECK-NEXT: xxmfacc acc2
411 ; CHECK-NEXT: xvf32gernp acc0, vs12, vs13
412 ; CHECK-NEXT: stxvx vs11, r3, r7
413 ; CHECK-NEXT: stxv vs8, 48(r8)
414 ; CHECK-NEXT: xxmfacc acc1
415 ; CHECK-NEXT: stxv vs9, 32(r8)
416 ; CHECK-NEXT: stxv vs10, 16(r8)
417 ; CHECK-NEXT: stxv vs4, 112(r8)
418 ; CHECK-NEXT: stxv vs5, 96(r8)
419 ; CHECK-NEXT: xxmfacc acc0
420 ; CHECK-NEXT: stxv vs6, 80(r8)
421 ; CHECK-NEXT: stxv vs7, 64(r8)
422 ; CHECK-NEXT: stxv vs0, 176(r8)
423 ; CHECK-NEXT: stxv vs1, 160(r8)
424 ; CHECK-NEXT: stxv vs2, 144(r8)
425 ; CHECK-NEXT: stxv vs3, 128(r8)
426 ; CHECK-NEXT: bdnz .LBB9_2
427 ; CHECK-NEXT: # %bb.3: # %for.cond.cleanup
430 ; CHECK-BE-LABEL: testcse4:
431 ; CHECK-BE: # %bb.0: # %entry
432 ; CHECK-BE-NEXT: cmpwi r4, 1
433 ; CHECK-BE-NEXT: bltlr cr0
434 ; CHECK-BE-NEXT: # %bb.1: # %for.body.preheader
435 ; CHECK-BE-NEXT: clrldi r4, r4, 32
436 ; CHECK-BE-NEXT: li r6, 0
437 ; CHECK-BE-NEXT: mtctr r4
438 ; CHECK-BE-NEXT: li r4, 0
439 ; CHECK-BE-NEXT: .p2align 4
440 ; CHECK-BE-NEXT: .LBB9_2: # %for.body
442 ; CHECK-BE-NEXT: rldic r7, r6, 4, 28
443 ; CHECK-BE-NEXT: xxsetaccz acc2
444 ; CHECK-BE-NEXT: xxsetaccz acc1
445 ; CHECK-BE-NEXT: addi r6, r6, 6
446 ; CHECK-BE-NEXT: lxvx vs0, r5, r7
447 ; CHECK-BE-NEXT: add r7, r5, r7
448 ; CHECK-BE-NEXT: lxv vs1, 16(r7)
449 ; CHECK-BE-NEXT: xvf32gerpp acc2, vs0, vs1
450 ; CHECK-BE-NEXT: lxv vs0, 32(r7)
451 ; CHECK-BE-NEXT: lxv vs1, 48(r7)
452 ; CHECK-BE-NEXT: xvf32gerpn acc1, vs0, vs1
453 ; CHECK-BE-NEXT: lxv vs12, 64(r7)
454 ; CHECK-BE-NEXT: lxv vs13, 80(r7)
455 ; CHECK-BE-NEXT: xxsetaccz acc0
456 ; CHECK-BE-NEXT: rldic r7, r4, 6, 26
457 ; CHECK-BE-NEXT: addi r4, r4, 3
458 ; CHECK-BE-NEXT: add r8, r3, r7
459 ; CHECK-BE-NEXT: xxmfacc acc2
460 ; CHECK-BE-NEXT: xvf32gernp acc0, vs12, vs13
461 ; CHECK-BE-NEXT: stxvx vs8, r3, r7
462 ; CHECK-BE-NEXT: stxv vs9, 16(r8)
463 ; CHECK-BE-NEXT: xxmfacc acc1
464 ; CHECK-BE-NEXT: stxv vs11, 48(r8)
465 ; CHECK-BE-NEXT: stxv vs10, 32(r8)
466 ; CHECK-BE-NEXT: stxv vs5, 80(r8)
467 ; CHECK-BE-NEXT: stxv vs4, 64(r8)
468 ; CHECK-BE-NEXT: xxmfacc acc0
469 ; CHECK-BE-NEXT: stxv vs7, 112(r8)
470 ; CHECK-BE-NEXT: stxv vs6, 96(r8)
471 ; CHECK-BE-NEXT: stxv vs1, 144(r8)
472 ; CHECK-BE-NEXT: stxv vs0, 128(r8)
473 ; CHECK-BE-NEXT: stxv vs3, 176(r8)
474 ; CHECK-BE-NEXT: stxv vs2, 160(r8)
475 ; CHECK-BE-NEXT: bdnz .LBB9_2
476 ; CHECK-BE-NEXT: # %bb.3: # %for.cond.cleanup
479 %cmp55 = icmp sgt i32 %lim, 0
480 br i1 %cmp55, label %for.body.preheader, label %for.cond.cleanup
482 for.body.preheader: ; preds = %entry
483 %wide.trip.count = zext i32 %lim to i64
486 for.cond.cleanup: ; preds = %for.body, %entry
489 for.body: ; preds = %for.body, %for.body.preheader
490 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ]
491 %0 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
492 %1 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
493 %2 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
494 %3 = trunc i64 %indvars.iv to i32
495 %mul = mul nsw i32 %3, 6
496 %idxprom = zext i32 %mul to i64
497 %arrayidx = getelementptr inbounds <16 x i8>, ptr %vc, i64 %idxprom
498 %4 = load <16 x i8>, ptr %arrayidx, align 16
499 %add2 = or i32 %mul, 1
500 %idxprom3 = zext i32 %add2 to i64
501 %arrayidx4 = getelementptr inbounds <16 x i8>, ptr %vc, i64 %idxprom3
502 %5 = load <16 x i8>, ptr %arrayidx4, align 16
503 %6 = tail call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> %0, <16 x i8> %4, <16 x i8> %5)
504 %add6 = add nuw nsw i32 %mul, 2
505 %idxprom7 = zext i32 %add6 to i64
506 %arrayidx8 = getelementptr inbounds <16 x i8>, ptr %vc, i64 %idxprom7
507 %7 = load <16 x i8>, ptr %arrayidx8, align 16
508 %add10 = add nuw nsw i32 %mul, 3
509 %idxprom11 = zext i32 %add10 to i64
510 %arrayidx12 = getelementptr inbounds <16 x i8>, ptr %vc, i64 %idxprom11
511 %8 = load <16 x i8>, ptr %arrayidx12, align 16
512 %9 = tail call <512 x i1> @llvm.ppc.mma.xvf32gerpn(<512 x i1> %1, <16 x i8> %7, <16 x i8> %8)
513 %add14 = add nuw nsw i32 %mul, 4
514 %idxprom15 = zext i32 %add14 to i64
515 %arrayidx16 = getelementptr inbounds <16 x i8>, ptr %vc, i64 %idxprom15
516 %10 = load <16 x i8>, ptr %arrayidx16, align 16
517 %add18 = add nuw nsw i32 %mul, 5
518 %idxprom19 = zext i32 %add18 to i64
519 %arrayidx20 = getelementptr inbounds <16 x i8>, ptr %vc, i64 %idxprom19
520 %11 = load <16 x i8>, ptr %arrayidx20, align 16
521 %12 = tail call <512 x i1> @llvm.ppc.mma.xvf32gernp(<512 x i1> %2, <16 x i8> %10, <16 x i8> %11)
522 %mul21 = mul i64 %indvars.iv, 3
523 %idx.ext = and i64 %mul21, 4294967295
524 %add.ptr = getelementptr inbounds <512 x i1>, ptr %res, i64 %idx.ext
525 store <512 x i1> %6, ptr %add.ptr, align 64
526 %add.ptr26 = getelementptr inbounds <512 x i1>, ptr %add.ptr, i64 1
527 store <512 x i1> %9, ptr %add.ptr26, align 64
528 %add.ptr30 = getelementptr inbounds <512 x i1>, ptr %add.ptr, i64 2
529 store <512 x i1> %12, ptr %add.ptr30, align 64
530 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
531 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count
532 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body
535 declare i32 @testRedundantPrimeUnprimeF()
536 define void @testRedundantPrimeUnprime(ptr %dst, <16 x i8> %vc) nounwind {
537 ; CHECK-LABEL: testRedundantPrimeUnprime:
538 ; CHECK: # %bb.0: # %entry
539 ; CHECK-NEXT: mflr r0
540 ; CHECK-NEXT: std r30, -16(r1) # 8-byte Folded Spill
541 ; CHECK-NEXT: std r0, 16(r1)
542 ; CHECK-NEXT: stdu r1, -112(r1)
543 ; CHECK-NEXT: xxsetaccz acc0
544 ; CHECK-NEXT: xxsetaccz acc1
545 ; CHECK-NEXT: mr r30, r3
546 ; CHECK-NEXT: xxmfacc acc0
547 ; CHECK-NEXT: stxv vs0, 48(r3)
548 ; CHECK-NEXT: stxv vs1, 32(r3)
549 ; CHECK-NEXT: stxv vs2, 16(r3)
550 ; CHECK-NEXT: stxv vs3, 0(r3)
551 ; CHECK-NEXT: xvf32gerpp acc1, v2, v2
552 ; CHECK-NEXT: xxmfacc acc1
553 ; CHECK-NEXT: stxv vs4, 80(r1)
554 ; CHECK-NEXT: stxv vs5, 64(r1)
555 ; CHECK-NEXT: stxv vs6, 48(r1)
556 ; CHECK-NEXT: stxv vs7, 32(r1)
557 ; CHECK-NEXT: bl testRedundantPrimeUnprimeF@notoc
558 ; CHECK-NEXT: lxvp vsp0, 64(r1)
559 ; CHECK-NEXT: lxvp vsp2, 32(r1)
560 ; CHECK-NEXT: stxv vs0, 112(r30)
561 ; CHECK-NEXT: stxv vs1, 96(r30)
562 ; CHECK-NEXT: stxv vs2, 80(r30)
563 ; CHECK-NEXT: stxv vs3, 64(r30)
564 ; CHECK-NEXT: addi r1, r1, 112
565 ; CHECK-NEXT: ld r0, 16(r1)
566 ; CHECK-NEXT: ld r30, -16(r1) # 8-byte Folded Reload
567 ; CHECK-NEXT: mtlr r0
570 ; CHECK-BE-LABEL: testRedundantPrimeUnprime:
571 ; CHECK-BE: # %bb.0: # %entry
572 ; CHECK-BE-NEXT: mflr r0
573 ; CHECK-BE-NEXT: std r0, 16(r1)
574 ; CHECK-BE-NEXT: stdu r1, -192(r1)
575 ; CHECK-BE-NEXT: xxsetaccz acc0
576 ; CHECK-BE-NEXT: xxsetaccz acc1
577 ; CHECK-BE-NEXT: std r30, 176(r1) # 8-byte Folded Spill
578 ; CHECK-BE-NEXT: mr r30, r3
579 ; CHECK-BE-NEXT: xxmfacc acc0
580 ; CHECK-BE-NEXT: stxv vs1, 16(r3)
581 ; CHECK-BE-NEXT: stxv vs0, 0(r3)
582 ; CHECK-BE-NEXT: stxv vs3, 48(r3)
583 ; CHECK-BE-NEXT: stxv vs2, 32(r3)
584 ; CHECK-BE-NEXT: xvf32gerpp acc1, v2, v2
585 ; CHECK-BE-NEXT: xxmfacc acc1
586 ; CHECK-BE-NEXT: stxv vs4, 112(r1)
587 ; CHECK-BE-NEXT: stxv vs5, 128(r1)
588 ; CHECK-BE-NEXT: stxv vs6, 144(r1)
589 ; CHECK-BE-NEXT: stxv vs7, 160(r1)
590 ; CHECK-BE-NEXT: bl testRedundantPrimeUnprimeF
592 ; CHECK-BE-NEXT: lxvp vsp0, 112(r1)
593 ; CHECK-BE-NEXT: lxvp vsp2, 144(r1)
594 ; CHECK-BE-NEXT: stxv vs3, 112(r30)
595 ; CHECK-BE-NEXT: stxv vs2, 96(r30)
596 ; CHECK-BE-NEXT: stxv vs1, 80(r30)
597 ; CHECK-BE-NEXT: stxv vs0, 64(r30)
598 ; CHECK-BE-NEXT: ld r30, 176(r1) # 8-byte Folded Reload
599 ; CHECK-BE-NEXT: addi r1, r1, 192
600 ; CHECK-BE-NEXT: ld r0, 16(r1)
601 ; CHECK-BE-NEXT: mtlr r0
604 %0 = tail call <512 x i1> @llvm.ppc.mma.xxsetaccz()
605 store <512 x i1> %0, ptr %dst, align 64
606 %1 = tail call <512 x i1> @llvm.ppc.mma.xvf32gerpp(<512 x i1> %0, <16 x i8> %vc, <16 x i8> %vc)
607 %call = tail call signext i32 @testRedundantPrimeUnprimeF()
608 %add.ptr1 = getelementptr inbounds <512 x i1>, ptr %dst, i64 1
609 store <512 x i1> %1, ptr %add.ptr1, align 64
613 declare <256 x i1> @llvm.ppc.vsx.lxvp(ptr)
614 declare void @llvm.ppc.vsx.stxvp(<256 x i1>, ptr)
616 ; Function Attrs: nofree nounwind
617 define void @test_ldst_1(ptr nocapture readonly %vqp, ptr %vpp, <16 x i8> %vc, ptr nocapture %resp) {
618 ; CHECK-LABEL: test_ldst_1:
619 ; CHECK: # %bb.0: # %entry
620 ; CHECK-NEXT: lxv vs1, 32(r3)
621 ; CHECK-NEXT: lxv vs0, 48(r3)
622 ; CHECK-NEXT: lxv vs3, 0(r3)
623 ; CHECK-NEXT: lxv vs2, 16(r3)
624 ; CHECK-NEXT: plxvp vsp36, 8(r4), 0
625 ; CHECK-NEXT: xxmtacc acc0
626 ; CHECK-NEXT: pmxvf64gernn acc0, vsp36, v2, 0, 0
627 ; CHECK-NEXT: xxmfacc acc0
628 ; CHECK-NEXT: stxv vs0, 48(r7)
629 ; CHECK-NEXT: stxv vs1, 32(r7)
630 ; CHECK-NEXT: stxv vs2, 16(r7)
631 ; CHECK-NEXT: stxv vs3, 0(r7)
634 ; CHECK-BE-LABEL: test_ldst_1:
635 ; CHECK-BE: # %bb.0: # %entry
636 ; CHECK-BE-NEXT: lxv vs1, 16(r3)
637 ; CHECK-BE-NEXT: lxv vs0, 0(r3)
638 ; CHECK-BE-NEXT: lxv vs3, 48(r3)
639 ; CHECK-BE-NEXT: lxv vs2, 32(r3)
640 ; CHECK-BE-NEXT: plxvp vsp36, 8(r4), 0
641 ; CHECK-BE-NEXT: xxmtacc acc0
642 ; CHECK-BE-NEXT: pmxvf64gernn acc0, vsp36, v2, 0, 0
643 ; CHECK-BE-NEXT: xxmfacc acc0
644 ; CHECK-BE-NEXT: stxv vs1, 16(r7)
645 ; CHECK-BE-NEXT: stxv vs0, 0(r7)
646 ; CHECK-BE-NEXT: stxv vs3, 48(r7)
647 ; CHECK-BE-NEXT: stxv vs2, 32(r7)
650 %0 = load <512 x i1>, ptr %vqp, align 64
651 %1 = getelementptr i8, ptr %vpp, i64 8
652 %2 = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr %1)
653 %3 = tail call <512 x i1> @llvm.ppc.mma.pmxvf64gernn(<512 x i1> %0, <256 x i1> %2, <16 x i8> %vc, i32 0, i32 0)
654 store <512 x i1> %3, ptr %resp, align 64
658 ; Function Attrs: nofree nounwind
659 define void @test_ldst_2(ptr nocapture readonly %vqp, ptr %vpp, <16 x i8> %vc, ptr nocapture %resp) {
660 ; CHECK-LABEL: test_ldst_2:
661 ; CHECK: # %bb.0: # %entry
662 ; CHECK-NEXT: lxv vs1, 32(r3)
663 ; CHECK-NEXT: lxv vs0, 48(r3)
664 ; CHECK-NEXT: lxv vs3, 0(r3)
665 ; CHECK-NEXT: lxv vs2, 16(r3)
666 ; CHECK-NEXT: xxmtacc acc0
667 ; CHECK-NEXT: lxvp vsp36, 0(r4)
668 ; CHECK-NEXT: xvf64gernp acc0, vsp36, v2
669 ; CHECK-NEXT: xxmfacc acc0
670 ; CHECK-NEXT: stxv vs0, 48(r7)
671 ; CHECK-NEXT: stxv vs1, 32(r7)
672 ; CHECK-NEXT: stxv vs2, 16(r7)
673 ; CHECK-NEXT: stxv vs3, 0(r7)
676 ; CHECK-BE-LABEL: test_ldst_2:
677 ; CHECK-BE: # %bb.0: # %entry
678 ; CHECK-BE-NEXT: lxv vs1, 16(r3)
679 ; CHECK-BE-NEXT: lxv vs0, 0(r3)
680 ; CHECK-BE-NEXT: lxv vs3, 48(r3)
681 ; CHECK-BE-NEXT: lxv vs2, 32(r3)
682 ; CHECK-BE-NEXT: xxmtacc acc0
683 ; CHECK-BE-NEXT: lxvp vsp36, 0(r4)
684 ; CHECK-BE-NEXT: xvf64gernp acc0, vsp36, v2
685 ; CHECK-BE-NEXT: xxmfacc acc0
686 ; CHECK-BE-NEXT: stxv vs1, 16(r7)
687 ; CHECK-BE-NEXT: stxv vs0, 0(r7)
688 ; CHECK-BE-NEXT: stxv vs3, 48(r7)
689 ; CHECK-BE-NEXT: stxv vs2, 32(r7)
692 %0 = load <512 x i1>, ptr %vqp, align 64
693 %1 = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr %vpp)
694 %2 = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> %0, <256 x i1> %1, <16 x i8> %vc)
695 store <512 x i1> %2, ptr %resp, align 64
699 ; Function Attrs: nofree nounwind
700 define void @test_ldst_3(ptr nocapture readonly %vqp, i64 %offs, ptr %vpp, <16 x i8> %vc, ptr nocapture %resp) {
701 ; CHECK-LABEL: test_ldst_3:
702 ; CHECK: # %bb.0: # %entry
703 ; CHECK-NEXT: lxv vs1, 32(r3)
704 ; CHECK-NEXT: lxv vs0, 48(r3)
705 ; CHECK-NEXT: lxv vs3, 0(r3)
706 ; CHECK-NEXT: lxv vs2, 16(r3)
707 ; CHECK-NEXT: xxmtacc acc0
708 ; CHECK-NEXT: lxvp vsp36, 0(r5)
709 ; CHECK-NEXT: xvf64gernp acc0, vsp36, v2
710 ; CHECK-NEXT: xxmfacc acc0
711 ; CHECK-NEXT: stxv vs0, 48(r9)
712 ; CHECK-NEXT: stxv vs1, 32(r9)
713 ; CHECK-NEXT: stxv vs2, 16(r9)
714 ; CHECK-NEXT: stxv vs3, 0(r9)
717 ; CHECK-BE-LABEL: test_ldst_3:
718 ; CHECK-BE: # %bb.0: # %entry
719 ; CHECK-BE-NEXT: lxv vs1, 16(r3)
720 ; CHECK-BE-NEXT: lxv vs0, 0(r3)
721 ; CHECK-BE-NEXT: lxv vs3, 48(r3)
722 ; CHECK-BE-NEXT: lxv vs2, 32(r3)
723 ; CHECK-BE-NEXT: xxmtacc acc0
724 ; CHECK-BE-NEXT: lxvp vsp36, 0(r5)
725 ; CHECK-BE-NEXT: xvf64gernp acc0, vsp36, v2
726 ; CHECK-BE-NEXT: xxmfacc acc0
727 ; CHECK-BE-NEXT: stxv vs1, 16(r9)
728 ; CHECK-BE-NEXT: stxv vs0, 0(r9)
729 ; CHECK-BE-NEXT: stxv vs3, 48(r9)
730 ; CHECK-BE-NEXT: stxv vs2, 32(r9)
733 %0 = load <512 x i1>, ptr %vqp, align 64
734 %1 = tail call <256 x i1> @llvm.ppc.vsx.lxvp(ptr %vpp)
735 %2 = tail call <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1> %0, <256 x i1> %1, <16 x i8> %vc)
736 store <512 x i1> %2, ptr %resp, align 64
740 declare <512 x i1> @llvm.ppc.mma.pmxvf64gernn(<512 x i1>, <256 x i1>, <16 x i8>, i32, i32)
741 declare <512 x i1> @llvm.ppc.mma.xvf64gernp(<512 x i1>, <256 x i1>, <16 x i8>)