1/*
2 * QEMU TCG support -- s390x vector integer instruction support
3 *
4 * Copyright (C) 2019 Red Hat Inc
5 *
6 * Authors:
7 * David Hildenbrand <david@redhat.com>
8 *
9 * This work is licensed under the terms of the GNU GPL, version 2 or later.
10 * See the COPYING file in the top-level directory.
11 */
12#include "qemu/osdep.h"
13#include "qemu-common.h"
14#include "cpu.h"
15#include "vec.h"
16#include "exec/helper-proto.h"
17#include "tcg/tcg-gvec-desc.h"
18
19static bool s390_vec_is_zero(const S390Vector *v)
20{
21 return !v->doubleword[0] && !v->doubleword[1];
22}
23
24static void s390_vec_xor(S390Vector *res, const S390Vector *a,
25 const S390Vector *b)
26{
27 res->doubleword[0] = a->doubleword[0] ^ b->doubleword[0];
28 res->doubleword[1] = a->doubleword[1] ^ b->doubleword[1];
29}
30
31static void s390_vec_and(S390Vector *res, const S390Vector *a,
32 const S390Vector *b)
33{
34 res->doubleword[0] = a->doubleword[0] & b->doubleword[0];
35 res->doubleword[1] = a->doubleword[1] & b->doubleword[1];
36}
37
38static bool s390_vec_equal(const S390Vector *a, const S390Vector *b)
39{
40 return a->doubleword[0] == b->doubleword[0] &&
41 a->doubleword[1] == b->doubleword[1];
42}
43
44static void s390_vec_shl(S390Vector *d, const S390Vector *a, uint64_t count)
45{
46 uint64_t tmp;
47
48 g_assert(count < 128);
49 if (count == 0) {
50 d->doubleword[0] = a->doubleword[0];
51 d->doubleword[1] = a->doubleword[1];
52 } else if (count == 64) {
53 d->doubleword[0] = a->doubleword[1];
54 d->doubleword[1] = 0;
55 } else if (count < 64) {
56 tmp = extract64(a->doubleword[1], 64 - count, count);
57 d->doubleword[1] = a->doubleword[1] << count;
58 d->doubleword[0] = (a->doubleword[0] << count) | tmp;
59 } else {
60 d->doubleword[0] = a->doubleword[1] << (count - 64);
61 d->doubleword[1] = 0;
62 }
63}
64
65static void s390_vec_sar(S390Vector *d, const S390Vector *a, uint64_t count)
66{
67 uint64_t tmp;
68
69 if (count == 0) {
70 d->doubleword[0] = a->doubleword[0];
71 d->doubleword[1] = a->doubleword[1];
72 } else if (count == 64) {
73 d->doubleword[1] = a->doubleword[0];
74 d->doubleword[0] = 0;
75 } else if (count < 64) {
76 tmp = a->doubleword[1] >> count;
77 d->doubleword[1] = deposit64(tmp, 64 - count, count, a->doubleword[0]);
78 d->doubleword[0] = (int64_t)a->doubleword[0] >> count;
79 } else {
80 d->doubleword[1] = (int64_t)a->doubleword[0] >> (count - 64);
81 d->doubleword[0] = 0;
82 }
83}
84
85static void s390_vec_shr(S390Vector *d, const S390Vector *a, uint64_t count)
86{
87 uint64_t tmp;
88
89 g_assert(count < 128);
90 if (count == 0) {
91 d->doubleword[0] = a->doubleword[0];
92 d->doubleword[1] = a->doubleword[1];
93 } else if (count == 64) {
94 d->doubleword[1] = a->doubleword[0];
95 d->doubleword[0] = 0;
96 } else if (count < 64) {
97 tmp = a->doubleword[1] >> count;
98 d->doubleword[1] = deposit64(tmp, 64 - count, count, a->doubleword[0]);
99 d->doubleword[0] = a->doubleword[0] >> count;
100 } else {
101 d->doubleword[1] = a->doubleword[0] >> (count - 64);
102 d->doubleword[0] = 0;
103 }
104}
105#define DEF_VAVG(BITS) \
106void HELPER(gvec_vavg##BITS)(void *v1, const void *v2, const void *v3, \
107 uint32_t desc) \
108{ \
109 int i; \
110 \
111 for (i = 0; i < (128 / BITS); i++) { \
112 const int32_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, i); \
113 const int32_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, i); \
114 \
115 s390_vec_write_element##BITS(v1, i, (a + b + 1) >> 1); \
116 } \
117}
118DEF_VAVG(8)
119DEF_VAVG(16)
120
121#define DEF_VAVGL(BITS) \
122void HELPER(gvec_vavgl##BITS)(void *v1, const void *v2, const void *v3, \
123 uint32_t desc) \
124{ \
125 int i; \
126 \
127 for (i = 0; i < (128 / BITS); i++) { \
128 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
129 const uint##BITS##_t b = s390_vec_read_element##BITS(v3, i); \
130 \
131 s390_vec_write_element##BITS(v1, i, (a + b + 1) >> 1); \
132 } \
133}
134DEF_VAVGL(8)
135DEF_VAVGL(16)
136
137#define DEF_VCLZ(BITS) \
138void HELPER(gvec_vclz##BITS)(void *v1, const void *v2, uint32_t desc) \
139{ \
140 int i; \
141 \
142 for (i = 0; i < (128 / BITS); i++) { \
143 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
144 \
145 s390_vec_write_element##BITS(v1, i, clz32(a) - 32 + BITS); \
146 } \
147}
148DEF_VCLZ(8)
149DEF_VCLZ(16)
150
151#define DEF_VCTZ(BITS) \
152void HELPER(gvec_vctz##BITS)(void *v1, const void *v2, uint32_t desc) \
153{ \
154 int i; \
155 \
156 for (i = 0; i < (128 / BITS); i++) { \
157 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
158 \
159 s390_vec_write_element##BITS(v1, i, a ? ctz32(a) : BITS); \
160 } \
161}
162DEF_VCTZ(8)
163DEF_VCTZ(16)
164
165/* like binary multiplication, but XOR instead of addition */
166#define DEF_GALOIS_MULTIPLY(BITS, TBITS) \
167static uint##TBITS##_t galois_multiply##BITS(uint##TBITS##_t a, \
168 uint##TBITS##_t b) \
169{ \
170 uint##TBITS##_t res = 0; \
171 \
172 while (b) { \
173 if (b & 0x1) { \
174 res = res ^ a; \
175 } \
176 a = a << 1; \
177 b = b >> 1; \
178 } \
179 return res; \
180}
181DEF_GALOIS_MULTIPLY(8, 16)
182DEF_GALOIS_MULTIPLY(16, 32)
183DEF_GALOIS_MULTIPLY(32, 64)
184
185static S390Vector galois_multiply64(uint64_t a, uint64_t b)
186{
187 S390Vector res = {};
188 S390Vector va = {
189 .doubleword[1] = a,
190 };
191 S390Vector vb = {
192 .doubleword[1] = b,
193 };
194
195 while (!s390_vec_is_zero(&vb)) {
196 if (vb.doubleword[1] & 0x1) {
197 s390_vec_xor(&res, &res, &va);
198 }
199 s390_vec_shl(&va, &va, 1);
200 s390_vec_shr(&vb, &vb, 1);
201 }
202 return res;
203}
204
205#define DEF_VGFM(BITS, TBITS) \
206void HELPER(gvec_vgfm##BITS)(void *v1, const void *v2, const void *v3, \
207 uint32_t desc) \
208{ \
209 int i; \
210 \
211 for (i = 0; i < (128 / TBITS); i++) { \
212 uint##BITS##_t a = s390_vec_read_element##BITS(v2, i * 2); \
213 uint##BITS##_t b = s390_vec_read_element##BITS(v3, i * 2); \
214 uint##TBITS##_t d = galois_multiply##BITS(a, b); \
215 \
216 a = s390_vec_read_element##BITS(v2, i * 2 + 1); \
217 b = s390_vec_read_element##BITS(v3, i * 2 + 1); \
218 d = d ^ galois_multiply32(a, b); \
219 s390_vec_write_element##TBITS(v1, i, d); \
220 } \
221}
222DEF_VGFM(8, 16)
223DEF_VGFM(16, 32)
224DEF_VGFM(32, 64)
225
226void HELPER(gvec_vgfm64)(void *v1, const void *v2, const void *v3,
227 uint32_t desc)
228{
229 S390Vector tmp1, tmp2;
230 uint64_t a, b;
231
232 a = s390_vec_read_element64(v2, 0);
233 b = s390_vec_read_element64(v3, 0);
234 tmp1 = galois_multiply64(a, b);
235 a = s390_vec_read_element64(v2, 1);
236 b = s390_vec_read_element64(v3, 1);
237 tmp2 = galois_multiply64(a, b);
238 s390_vec_xor(v1, &tmp1, &tmp2);
239}
240
241#define DEF_VGFMA(BITS, TBITS) \
242void HELPER(gvec_vgfma##BITS)(void *v1, const void *v2, const void *v3, \
243 const void *v4, uint32_t desc) \
244{ \
245 int i; \
246 \
247 for (i = 0; i < (128 / TBITS); i++) { \
248 uint##BITS##_t a = s390_vec_read_element##BITS(v2, i * 2); \
249 uint##BITS##_t b = s390_vec_read_element##BITS(v3, i * 2); \
250 uint##TBITS##_t d = galois_multiply##BITS(a, b); \
251 \
252 a = s390_vec_read_element##BITS(v2, i * 2 + 1); \
253 b = s390_vec_read_element##BITS(v3, i * 2 + 1); \
254 d = d ^ galois_multiply32(a, b); \
255 d = d ^ s390_vec_read_element##TBITS(v4, i); \
256 s390_vec_write_element##TBITS(v1, i, d); \
257 } \
258}
259DEF_VGFMA(8, 16)
260DEF_VGFMA(16, 32)
261DEF_VGFMA(32, 64)
262
263void HELPER(gvec_vgfma64)(void *v1, const void *v2, const void *v3,
264 const void *v4, uint32_t desc)
265{
266 S390Vector tmp1, tmp2;
267 uint64_t a, b;
268
269 a = s390_vec_read_element64(v2, 0);
270 b = s390_vec_read_element64(v3, 0);
271 tmp1 = galois_multiply64(a, b);
272 a = s390_vec_read_element64(v2, 1);
273 b = s390_vec_read_element64(v3, 1);
274 tmp2 = galois_multiply64(a, b);
275 s390_vec_xor(&tmp1, &tmp1, &tmp2);
276 s390_vec_xor(v1, &tmp1, v4);
277}
278
279#define DEF_VMAL(BITS) \
280void HELPER(gvec_vmal##BITS)(void *v1, const void *v2, const void *v3, \
281 const void *v4, uint32_t desc) \
282{ \
283 int i; \
284 \
285 for (i = 0; i < (128 / BITS); i++) { \
286 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
287 const uint##BITS##_t b = s390_vec_read_element##BITS(v3, i); \
288 const uint##BITS##_t c = s390_vec_read_element##BITS(v4, i); \
289 \
290 s390_vec_write_element##BITS(v1, i, a * b + c); \
291 } \
292}
293DEF_VMAL(8)
294DEF_VMAL(16)
295
296#define DEF_VMAH(BITS) \
297void HELPER(gvec_vmah##BITS)(void *v1, const void *v2, const void *v3, \
298 const void *v4, uint32_t desc) \
299{ \
300 int i; \
301 \
302 for (i = 0; i < (128 / BITS); i++) { \
303 const int32_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, i); \
304 const int32_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, i); \
305 const int32_t c = (int##BITS##_t)s390_vec_read_element##BITS(v4, i); \
306 \
307 s390_vec_write_element##BITS(v1, i, (a * b + c) >> BITS); \
308 } \
309}
310DEF_VMAH(8)
311DEF_VMAH(16)
312
313#define DEF_VMALH(BITS) \
314void HELPER(gvec_vmalh##BITS)(void *v1, const void *v2, const void *v3, \
315 const void *v4, uint32_t desc) \
316{ \
317 int i; \
318 \
319 for (i = 0; i < (128 / BITS); i++) { \
320 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
321 const uint##BITS##_t b = s390_vec_read_element##BITS(v3, i); \
322 const uint##BITS##_t c = s390_vec_read_element##BITS(v4, i); \
323 \
324 s390_vec_write_element##BITS(v1, i, (a * b + c) >> BITS); \
325 } \
326}
327DEF_VMALH(8)
328DEF_VMALH(16)
329
330#define DEF_VMAE(BITS, TBITS) \
331void HELPER(gvec_vmae##BITS)(void *v1, const void *v2, const void *v3, \
332 const void *v4, uint32_t desc) \
333{ \
334 int i, j; \
335 \
336 for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) { \
337 int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j); \
338 int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j); \
339 int##TBITS##_t c = (int##BITS##_t)s390_vec_read_element##BITS(v4, j); \
340 \
341 s390_vec_write_element##TBITS(v1, i, a * b + c); \
342 } \
343}
344DEF_VMAE(8, 16)
345DEF_VMAE(16, 32)
346DEF_VMAE(32, 64)
347
348#define DEF_VMALE(BITS, TBITS) \
349void HELPER(gvec_vmale##BITS)(void *v1, const void *v2, const void *v3, \
350 const void *v4, uint32_t desc) \
351{ \
352 int i, j; \
353 \
354 for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) { \
355 uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j); \
356 uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j); \
357 uint##TBITS##_t c = s390_vec_read_element##BITS(v4, j); \
358 \
359 s390_vec_write_element##TBITS(v1, i, a * b + c); \
360 } \
361}
362DEF_VMALE(8, 16)
363DEF_VMALE(16, 32)
364DEF_VMALE(32, 64)
365
366#define DEF_VMAO(BITS, TBITS) \
367void HELPER(gvec_vmao##BITS)(void *v1, const void *v2, const void *v3, \
368 const void *v4, uint32_t desc) \
369{ \
370 int i, j; \
371 \
372 for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) { \
373 int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j); \
374 int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j); \
375 int##TBITS##_t c = (int##BITS##_t)s390_vec_read_element##BITS(v4, j); \
376 \
377 s390_vec_write_element##TBITS(v1, i, a * b + c); \
378 } \
379}
380DEF_VMAO(8, 16)
381DEF_VMAO(16, 32)
382DEF_VMAO(32, 64)
383
384#define DEF_VMALO(BITS, TBITS) \
385void HELPER(gvec_vmalo##BITS)(void *v1, const void *v2, const void *v3, \
386 const void *v4, uint32_t desc) \
387{ \
388 int i, j; \
389 \
390 for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) { \
391 uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j); \
392 uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j); \
393 uint##TBITS##_t c = s390_vec_read_element##BITS(v4, j); \
394 \
395 s390_vec_write_element##TBITS(v1, i, a * b + c); \
396 } \
397}
398DEF_VMALO(8, 16)
399DEF_VMALO(16, 32)
400DEF_VMALO(32, 64)
401
402#define DEF_VMH(BITS) \
403void HELPER(gvec_vmh##BITS)(void *v1, const void *v2, const void *v3, \
404 uint32_t desc) \
405{ \
406 int i; \
407 \
408 for (i = 0; i < (128 / BITS); i++) { \
409 const int32_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, i); \
410 const int32_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, i); \
411 \
412 s390_vec_write_element##BITS(v1, i, (a * b) >> BITS); \
413 } \
414}
415DEF_VMH(8)
416DEF_VMH(16)
417
418#define DEF_VMLH(BITS) \
419void HELPER(gvec_vmlh##BITS)(void *v1, const void *v2, const void *v3, \
420 uint32_t desc) \
421{ \
422 int i; \
423 \
424 for (i = 0; i < (128 / BITS); i++) { \
425 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
426 const uint##BITS##_t b = s390_vec_read_element##BITS(v3, i); \
427 \
428 s390_vec_write_element##BITS(v1, i, (a * b) >> BITS); \
429 } \
430}
431DEF_VMLH(8)
432DEF_VMLH(16)
433
434#define DEF_VME(BITS, TBITS) \
435void HELPER(gvec_vme##BITS)(void *v1, const void *v2, const void *v3, \
436 uint32_t desc) \
437{ \
438 int i, j; \
439 \
440 for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) { \
441 int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j); \
442 int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j); \
443 \
444 s390_vec_write_element##TBITS(v1, i, a * b); \
445 } \
446}
447DEF_VME(8, 16)
448DEF_VME(16, 32)
449DEF_VME(32, 64)
450
451#define DEF_VMLE(BITS, TBITS) \
452void HELPER(gvec_vmle##BITS)(void *v1, const void *v2, const void *v3, \
453 uint32_t desc) \
454{ \
455 int i, j; \
456 \
457 for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) { \
458 const uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j); \
459 const uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j); \
460 \
461 s390_vec_write_element##TBITS(v1, i, a * b); \
462 } \
463}
464DEF_VMLE(8, 16)
465DEF_VMLE(16, 32)
466DEF_VMLE(32, 64)
467
468#define DEF_VMO(BITS, TBITS) \
469void HELPER(gvec_vmo##BITS)(void *v1, const void *v2, const void *v3, \
470 uint32_t desc) \
471{ \
472 int i, j; \
473 \
474 for (i = 0, j = 1; i < (128 / TBITS); i++, j += 2) { \
475 int##TBITS##_t a = (int##BITS##_t)s390_vec_read_element##BITS(v2, j); \
476 int##TBITS##_t b = (int##BITS##_t)s390_vec_read_element##BITS(v3, j); \
477 \
478 s390_vec_write_element##TBITS(v1, i, a * b); \
479 } \
480}
481DEF_VMO(8, 16)
482DEF_VMO(16, 32)
483DEF_VMO(32, 64)
484
485#define DEF_VMLO(BITS, TBITS) \
486void HELPER(gvec_vmlo##BITS)(void *v1, const void *v2, const void *v3, \
487 uint32_t desc) \
488{ \
489 int i, j; \
490 \
491 for (i = 0, j = 0; i < (128 / TBITS); i++, j += 2) { \
492 const uint##TBITS##_t a = s390_vec_read_element##BITS(v2, j); \
493 const uint##TBITS##_t b = s390_vec_read_element##BITS(v3, j); \
494 \
495 s390_vec_write_element##TBITS(v1, i, a * b); \
496 } \
497}
498DEF_VMLO(8, 16)
499DEF_VMLO(16, 32)
500DEF_VMLO(32, 64)
501
502#define DEF_VPOPCT(BITS) \
503void HELPER(gvec_vpopct##BITS)(void *v1, const void *v2, uint32_t desc) \
504{ \
505 int i; \
506 \
507 for (i = 0; i < (128 / BITS); i++) { \
508 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
509 \
510 s390_vec_write_element##BITS(v1, i, ctpop32(a)); \
511 } \
512}
513DEF_VPOPCT(8)
514DEF_VPOPCT(16)
515
516#define DEF_VERLLV(BITS) \
517void HELPER(gvec_verllv##BITS)(void *v1, const void *v2, const void *v3, \
518 uint32_t desc) \
519{ \
520 int i; \
521 \
522 for (i = 0; i < (128 / BITS); i++) { \
523 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
524 const uint##BITS##_t b = s390_vec_read_element##BITS(v3, i); \
525 \
526 s390_vec_write_element##BITS(v1, i, rol##BITS(a, b)); \
527 } \
528}
529DEF_VERLLV(8)
530DEF_VERLLV(16)
531
532#define DEF_VERLL(BITS) \
533void HELPER(gvec_verll##BITS)(void *v1, const void *v2, uint64_t count, \
534 uint32_t desc) \
535{ \
536 int i; \
537 \
538 for (i = 0; i < (128 / BITS); i++) { \
539 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
540 \
541 s390_vec_write_element##BITS(v1, i, rol##BITS(a, count)); \
542 } \
543}
544DEF_VERLL(8)
545DEF_VERLL(16)
546
547#define DEF_VERIM(BITS) \
548void HELPER(gvec_verim##BITS)(void *v1, const void *v2, const void *v3, \
549 uint32_t desc) \
550{ \
551 const uint8_t count = simd_data(desc); \
552 int i; \
553 \
554 for (i = 0; i < (128 / BITS); i++) { \
555 const uint##BITS##_t a = s390_vec_read_element##BITS(v1, i); \
556 const uint##BITS##_t b = s390_vec_read_element##BITS(v2, i); \
557 const uint##BITS##_t mask = s390_vec_read_element##BITS(v3, i); \
558 const uint##BITS##_t d = (a & ~mask) | (rol##BITS(b, count) & mask); \
559 \
560 s390_vec_write_element##BITS(v1, i, d); \
561 } \
562}
563DEF_VERIM(8)
564DEF_VERIM(16)
565
566void HELPER(gvec_vsl)(void *v1, const void *v2, uint64_t count,
567 uint32_t desc)
568{
569 s390_vec_shl(v1, v2, count);
570}
571
572void HELPER(gvec_vsra)(void *v1, const void *v2, uint64_t count,
573 uint32_t desc)
574{
575 s390_vec_sar(v1, v2, count);
576}
577
578void HELPER(gvec_vsrl)(void *v1, const void *v2, uint64_t count,
579 uint32_t desc)
580{
581 s390_vec_shr(v1, v2, count);
582}
583
584#define DEF_VSCBI(BITS) \
585void HELPER(gvec_vscbi##BITS)(void *v1, const void *v2, const void *v3, \
586 uint32_t desc) \
587{ \
588 int i; \
589 \
590 for (i = 0; i < (128 / BITS); i++) { \
591 const uint##BITS##_t a = s390_vec_read_element##BITS(v2, i); \
592 const uint##BITS##_t b = s390_vec_read_element##BITS(v3, i); \
593 \
594 s390_vec_write_element##BITS(v1, i, a < b); \
595 } \
596}
597DEF_VSCBI(8)
598DEF_VSCBI(16)
599
600void HELPER(gvec_vtm)(void *v1, const void *v2, CPUS390XState *env,
601 uint32_t desc)
602{
603 S390Vector tmp;
604
605 s390_vec_and(&tmp, v1, v2);
606 if (s390_vec_is_zero(&tmp)) {
607 /* Selected bits all zeros; or all mask bits zero */
608 env->cc_op = 0;
609 } else if (s390_vec_equal(&tmp, v2)) {
610 /* Selected bits all ones */
611 env->cc_op = 3;
612 } else {
613 /* Selected bits a mix of zeros and ones */
614 env->cc_op = 1;
615 }
616}
617