]> git.proxmox.com Git - mirror_qemu.git/blob - target/loongarch/vec_helper.c
projects / mirror_qemu.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
target/loongarch: Implement xvsrln xvsran
[mirror_qemu.git] / target / loongarch / vec_helper.c
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 /*
3 * QEMU LoongArch vector helper functions.
4 *
5 * Copyright (c) 2022-2023 Loongson Technology Corporation Limited
6 */
7
8 #include "qemu/osdep.h"
9 #include "cpu.h"
10 #include "exec/exec-all.h"
11 #include "exec/helper-proto.h"
12 #include "fpu/softfloat.h"
13 #include "internals.h"
14 #include "tcg/tcg.h"
15 #include "vec.h"
16 #include "tcg/tcg-gvec-desc.h"
17
18 #define DO_ADD(a, b) (a + b)
19 #define DO_SUB(a, b) (a - b)
20
21 #define DO_ODD_EVEN(NAME, BIT, E1, E2, DO_OP) \
22 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
23 { \
24 int i; \
25 VReg *Vd = (VReg *)vd; \
26 VReg *Vj = (VReg *)vj; \
27 VReg *Vk = (VReg *)vk; \
28 typedef __typeof(Vd->E1(0)) TD; \
29 int oprsz = simd_oprsz(desc); \
30 \
31 for (i = 0; i < oprsz / (BIT / 8); i++) { \
32 Vd->E1(i) = DO_OP((TD)Vj->E2(2 * i + 1), (TD)Vk->E2(2 * i)); \
33 } \
34 }
35
36 DO_ODD_EVEN(vhaddw_h_b, 16, H, B, DO_ADD)
37 DO_ODD_EVEN(vhaddw_w_h, 32, W, H, DO_ADD)
38 DO_ODD_EVEN(vhaddw_d_w, 64, D, W, DO_ADD)
39
40 void HELPER(vhaddw_q_d)(void *vd, void *vj, void *vk, uint32_t desc)
41 {
42 int i;
43 VReg *Vd = (VReg *)vd;
44 VReg *Vj = (VReg *)vj;
45 VReg *Vk = (VReg *)vk;
46 int oprsz = simd_oprsz(desc);
47
48 for (i = 0; i < oprsz / 16 ; i++) {
49 Vd->Q(i) = int128_add(int128_makes64(Vj->D(2 * i + 1)),
50 int128_makes64(Vk->D(2 * i)));
51 }
52 }
53
54 DO_ODD_EVEN(vhsubw_h_b, 16, H, B, DO_SUB)
55 DO_ODD_EVEN(vhsubw_w_h, 32, W, H, DO_SUB)
56 DO_ODD_EVEN(vhsubw_d_w, 64, D, W, DO_SUB)
57
58 void HELPER(vhsubw_q_d)(void *vd, void *vj, void *vk, uint32_t desc)
59 {
60 int i;
61 VReg *Vd = (VReg *)vd;
62 VReg *Vj = (VReg *)vj;
63 VReg *Vk = (VReg *)vk;
64 int oprsz = simd_oprsz(desc);
65
66 for (i = 0; i < oprsz / 16; i++) {
67 Vd->Q(i) = int128_sub(int128_makes64(Vj->D(2 * i + 1)),
68 int128_makes64(Vk->D(2 * i)));
69 }
70 }
71
72 DO_ODD_EVEN(vhaddw_hu_bu, 16, UH, UB, DO_ADD)
73 DO_ODD_EVEN(vhaddw_wu_hu, 32, UW, UH, DO_ADD)
74 DO_ODD_EVEN(vhaddw_du_wu, 64, UD, UW, DO_ADD)
75
76 void HELPER(vhaddw_qu_du)(void *vd, void *vj, void *vk, uint32_t desc)
77 {
78 int i;
79 VReg *Vd = (VReg *)vd;
80 VReg *Vj = (VReg *)vj;
81 VReg *Vk = (VReg *)vk;
82 int oprsz = simd_oprsz(desc);
83
84 for (i = 0; i < oprsz / 16; i ++) {
85 Vd->Q(i) = int128_add(int128_make64(Vj->UD(2 * i + 1)),
86 int128_make64(Vk->UD(2 * i)));
87 }
88 }
89
90 DO_ODD_EVEN(vhsubw_hu_bu, 16, UH, UB, DO_SUB)
91 DO_ODD_EVEN(vhsubw_wu_hu, 32, UW, UH, DO_SUB)
92 DO_ODD_EVEN(vhsubw_du_wu, 64, UD, UW, DO_SUB)
93
94 void HELPER(vhsubw_qu_du)(void *vd, void *vj, void *vk, uint32_t desc)
95 {
96 int i;
97 VReg *Vd = (VReg *)vd;
98 VReg *Vj = (VReg *)vj;
99 VReg *Vk = (VReg *)vk;
100 int oprsz = simd_oprsz(desc);
101
102 for (i = 0; i < oprsz / 16; i++) {
103 Vd->Q(i) = int128_sub(int128_make64(Vj->UD(2 * i + 1)),
104 int128_make64(Vk->UD(2 * i)));
105 }
106 }
107
108 #define DO_EVEN(NAME, BIT, E1, E2, DO_OP) \
109 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
110 { \
111 int i; \
112 VReg *Vd = (VReg *)vd; \
113 VReg *Vj = (VReg *)vj; \
114 VReg *Vk = (VReg *)vk; \
115 typedef __typeof(Vd->E1(0)) TD; \
116 int oprsz = simd_oprsz(desc); \
117 \
118 for (i = 0; i < oprsz / (BIT / 8); i++) { \
119 Vd->E1(i) = DO_OP((TD)Vj->E2(2 * i) ,(TD)Vk->E2(2 * i)); \
120 } \
121 }
122
123 #define DO_ODD(NAME, BIT, E1, E2, DO_OP) \
124 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
125 { \
126 int i; \
127 VReg *Vd = (VReg *)vd; \
128 VReg *Vj = (VReg *)vj; \
129 VReg *Vk = (VReg *)vk; \
130 typedef __typeof(Vd->E1(0)) TD; \
131 int oprsz = simd_oprsz(desc); \
132 \
133 for (i = 0; i < oprsz / (BIT / 8); i++) { \
134 Vd->E1(i) = DO_OP((TD)Vj->E2(2 * i + 1), (TD)Vk->E2(2 * i + 1)); \
135 } \
136 }
137
138 void HELPER(vaddwev_q_d)(void *vd, void *vj, void *vk, uint32_t desc)
139 {
140 int i;
141 VReg *Vd = (VReg *)vd;
142 VReg *Vj = (VReg *)vj;
143 VReg *Vk = (VReg *)vk;
144 int oprsz = simd_oprsz(desc);
145
146 for (i = 0; i < oprsz / 16; i++) {
147 Vd->Q(i) = int128_add(int128_makes64(Vj->D(2 * i)),
148 int128_makes64(Vk->D(2 * i)));
149 }
150 }
151
152 DO_EVEN(vaddwev_h_b, 16, H, B, DO_ADD)
153 DO_EVEN(vaddwev_w_h, 32, W, H, DO_ADD)
154 DO_EVEN(vaddwev_d_w, 64, D, W, DO_ADD)
155
156 void HELPER(vaddwod_q_d)(void *vd, void *vj, void *vk, uint32_t desc)
157 {
158 int i;
159 VReg *Vd = (VReg *)vd;
160 VReg *Vj = (VReg *)vj;
161 VReg *Vk = (VReg *)vk;
162 int oprsz = simd_oprsz(desc);
163
164 for (i = 0; i < oprsz / 16; i++) {
165 Vd->Q(i) = int128_add(int128_makes64(Vj->D(2 * i +1)),
166 int128_makes64(Vk->D(2 * i +1)));
167 }
168 }
169
170 DO_ODD(vaddwod_h_b, 16, H, B, DO_ADD)
171 DO_ODD(vaddwod_w_h, 32, W, H, DO_ADD)
172 DO_ODD(vaddwod_d_w, 64, D, W, DO_ADD)
173
174 void HELPER(vsubwev_q_d)(void *vd, void *vj, void *vk, uint32_t desc)
175 {
176 int i;
177 VReg *Vd = (VReg *)vd;
178 VReg *Vj = (VReg *)vj;
179 VReg *Vk = (VReg *)vk;
180 int oprsz = simd_oprsz(desc);
181
182 for (i = 0; i < oprsz / 16; i++) {
183 Vd->Q(i) = int128_sub(int128_makes64(Vj->D(2 * i)),
184 int128_makes64(Vk->D(2 * i)));
185 }
186 }
187
188 DO_EVEN(vsubwev_h_b, 16, H, B, DO_SUB)
189 DO_EVEN(vsubwev_w_h, 32, W, H, DO_SUB)
190 DO_EVEN(vsubwev_d_w, 64, D, W, DO_SUB)
191
192 void HELPER(vsubwod_q_d)(void *vd, void *vj, void *vk, uint32_t desc)
193 {
194 int i;
195 VReg *Vd = (VReg *)vd;
196 VReg *Vj = (VReg *)vj;
197 VReg *Vk = (VReg *)vk;
198 int oprsz = simd_oprsz(desc);
199
200 for (i = 0; i < oprsz / 16; i++) {
201 Vd->Q(i) = int128_sub(int128_makes64(Vj->D(2 * i + 1)),
202 int128_makes64(Vk->D(2 * i + 1)));
203 }
204 }
205
206 DO_ODD(vsubwod_h_b, 16, H, B, DO_SUB)
207 DO_ODD(vsubwod_w_h, 32, W, H, DO_SUB)
208 DO_ODD(vsubwod_d_w, 64, D, W, DO_SUB)
209
210 void HELPER(vaddwev_q_du)(void *vd, void *vj, void *vk, uint32_t desc)
211 {
212 int i;
213 VReg *Vd = (VReg *)vd;
214 VReg *Vj = (VReg *)vj;
215 VReg *Vk = (VReg *)vk;
216 int oprsz = simd_oprsz(desc);
217
218 for (i = 0; i < oprsz / 16; i++) {
219 Vd->Q(i) = int128_add(int128_make64(Vj->UD(2 * i)),
220 int128_make64(Vk->UD(2 * i)));
221 }
222 }
223
224 DO_EVEN(vaddwev_h_bu, 16, UH, UB, DO_ADD)
225 DO_EVEN(vaddwev_w_hu, 32, UW, UH, DO_ADD)
226 DO_EVEN(vaddwev_d_wu, 64, UD, UW, DO_ADD)
227
228 void HELPER(vaddwod_q_du)(void *vd, void *vj, void *vk, uint32_t desc)
229 {
230 int i;
231 VReg *Vd = (VReg *)vd;
232 VReg *Vj = (VReg *)vj;
233 VReg *Vk = (VReg *)vk;
234 int oprsz = simd_oprsz(desc);
235
236 for (i = 0; i < oprsz / 16; i++) {
237 Vd->Q(i) = int128_add(int128_make64(Vj->UD(2 * i + 1)),
238 int128_make64(Vk->UD(2 * i + 1)));
239 }
240 }
241
242 DO_ODD(vaddwod_h_bu, 16, UH, UB, DO_ADD)
243 DO_ODD(vaddwod_w_hu, 32, UW, UH, DO_ADD)
244 DO_ODD(vaddwod_d_wu, 64, UD, UW, DO_ADD)
245
246 void HELPER(vsubwev_q_du)(void *vd, void *vj, void *vk, uint32_t desc)
247 {
248 int i;
249 VReg *Vd = (VReg *)vd;
250 VReg *Vj = (VReg *)vj;
251 VReg *Vk = (VReg *)vk;
252 int oprsz = simd_oprsz(desc);
253
254 for (i = 0; i < oprsz / 16; i++) {
255 Vd->Q(i) = int128_sub(int128_make64(Vj->UD(2 * i)),
256 int128_make64(Vk->UD(2 * i)));
257 }
258 }
259
260 DO_EVEN(vsubwev_h_bu, 16, UH, UB, DO_SUB)
261 DO_EVEN(vsubwev_w_hu, 32, UW, UH, DO_SUB)
262 DO_EVEN(vsubwev_d_wu, 64, UD, UW, DO_SUB)
263
264 void HELPER(vsubwod_q_du)(void *vd, void *vj, void *vk, uint32_t desc)
265 {
266 int i;
267 VReg *Vd = (VReg *)vd;
268 VReg *Vj = (VReg *)vj;
269 VReg *Vk = (VReg *)vk;
270 int oprsz = simd_oprsz(desc);
271
272 for (i = 0; i < oprsz / 16; i++) {
273 Vd->Q(i) = int128_sub(int128_make64(Vj->UD(2 * i + 1)),
274 int128_make64(Vk->UD(2 * i + 1)));
275 }
276 }
277
278 DO_ODD(vsubwod_h_bu, 16, UH, UB, DO_SUB)
279 DO_ODD(vsubwod_w_hu, 32, UW, UH, DO_SUB)
280 DO_ODD(vsubwod_d_wu, 64, UD, UW, DO_SUB)
281
282 #define DO_EVEN_U_S(NAME, BIT, ES1, EU1, ES2, EU2, DO_OP) \
283 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
284 { \
285 int i; \
286 VReg *Vd = (VReg *)vd; \
287 VReg *Vj = (VReg *)vj; \
288 VReg *Vk = (VReg *)vk; \
289 typedef __typeof(Vd->ES1(0)) TDS; \
290 typedef __typeof(Vd->EU1(0)) TDU; \
291 int oprsz = simd_oprsz(desc); \
292 \
293 for (i = 0; i < oprsz / (BIT / 8); i++) { \
294 Vd->ES1(i) = DO_OP((TDU)Vj->EU2(2 * i) ,(TDS)Vk->ES2(2 * i)); \
295 } \
296 }
297
298 #define DO_ODD_U_S(NAME, BIT, ES1, EU1, ES2, EU2, DO_OP) \
299 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
300 { \
301 int i; \
302 VReg *Vd = (VReg *)vd; \
303 VReg *Vj = (VReg *)vj; \
304 VReg *Vk = (VReg *)vk; \
305 typedef __typeof(Vd->ES1(0)) TDS; \
306 typedef __typeof(Vd->EU1(0)) TDU; \
307 int oprsz = simd_oprsz(desc); \
308 \
309 for (i = 0; i < oprsz / (BIT / 8); i++) { \
310 Vd->ES1(i) = DO_OP((TDU)Vj->EU2(2 * i + 1), (TDS)Vk->ES2(2 * i + 1)); \
311 } \
312 }
313
314 void HELPER(vaddwev_q_du_d)(void *vd, void *vj, void *vk, uint32_t desc)
315 {
316 int i;
317 VReg *Vd = (VReg *)vd;
318 VReg *Vj = (VReg *)vj;
319 VReg *Vk = (VReg *)vk;
320 int oprsz = simd_oprsz(desc);
321
322 for (i = 0; i < oprsz / 16; i++) {
323 Vd->Q(i) = int128_add(int128_make64(Vj->UD(2 * i)),
324 int128_makes64(Vk->D(2 * i)));
325 }
326 }
327
328 DO_EVEN_U_S(vaddwev_h_bu_b, 16, H, UH, B, UB, DO_ADD)
329 DO_EVEN_U_S(vaddwev_w_hu_h, 32, W, UW, H, UH, DO_ADD)
330 DO_EVEN_U_S(vaddwev_d_wu_w, 64, D, UD, W, UW, DO_ADD)
331
332 void HELPER(vaddwod_q_du_d)(void *vd, void *vj, void *vk, uint32_t desc)
333 {
334 int i;
335 VReg *Vd = (VReg *)vd;
336 VReg *Vj = (VReg *)vj;
337 VReg *Vk = (VReg *)vk;
338 int oprsz = simd_oprsz(desc);
339
340 for (i = 0; i < oprsz / 16; i++) {
341 Vd->Q(i) = int128_add(int128_make64(Vj->UD(2 * i + 1)),
342 int128_makes64(Vk->D(2 * i + 1)));
343 }
344 }
345
346 DO_ODD_U_S(vaddwod_h_bu_b, 16, H, UH, B, UB, DO_ADD)
347 DO_ODD_U_S(vaddwod_w_hu_h, 32, W, UW, H, UH, DO_ADD)
348 DO_ODD_U_S(vaddwod_d_wu_w, 64, D, UD, W, UW, DO_ADD)
349
350 #define DO_VAVG(a, b) ((a >> 1) + (b >> 1) + (a & b & 1))
351 #define DO_VAVGR(a, b) ((a >> 1) + (b >> 1) + ((a | b) & 1))
352
353 #define DO_3OP(NAME, BIT, E, DO_OP) \
354 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
355 { \
356 int i; \
357 VReg *Vd = (VReg *)vd; \
358 VReg *Vj = (VReg *)vj; \
359 VReg *Vk = (VReg *)vk; \
360 int oprsz = simd_oprsz(desc); \
361 \
362 for (i = 0; i < oprsz / (BIT / 8); i++) { \
363 Vd->E(i) = DO_OP(Vj->E(i), Vk->E(i)); \
364 } \
365 }
366
367 DO_3OP(vavg_b, 8, B, DO_VAVG)
368 DO_3OP(vavg_h, 16, H, DO_VAVG)
369 DO_3OP(vavg_w, 32, W, DO_VAVG)
370 DO_3OP(vavg_d, 64, D, DO_VAVG)
371 DO_3OP(vavgr_b, 8, B, DO_VAVGR)
372 DO_3OP(vavgr_h, 16, H, DO_VAVGR)
373 DO_3OP(vavgr_w, 32, W, DO_VAVGR)
374 DO_3OP(vavgr_d, 64, D, DO_VAVGR)
375 DO_3OP(vavg_bu, 8, UB, DO_VAVG)
376 DO_3OP(vavg_hu, 16, UH, DO_VAVG)
377 DO_3OP(vavg_wu, 32, UW, DO_VAVG)
378 DO_3OP(vavg_du, 64, UD, DO_VAVG)
379 DO_3OP(vavgr_bu, 8, UB, DO_VAVGR)
380 DO_3OP(vavgr_hu, 16, UH, DO_VAVGR)
381 DO_3OP(vavgr_wu, 32, UW, DO_VAVGR)
382 DO_3OP(vavgr_du, 64, UD, DO_VAVGR)
383
384 #define DO_VABSD(a, b) ((a > b) ? (a -b) : (b-a))
385
386 DO_3OP(vabsd_b, 8, B, DO_VABSD)
387 DO_3OP(vabsd_h, 16, H, DO_VABSD)
388 DO_3OP(vabsd_w, 32, W, DO_VABSD)
389 DO_3OP(vabsd_d, 64, D, DO_VABSD)
390 DO_3OP(vabsd_bu, 8, UB, DO_VABSD)
391 DO_3OP(vabsd_hu, 16, UH, DO_VABSD)
392 DO_3OP(vabsd_wu, 32, UW, DO_VABSD)
393 DO_3OP(vabsd_du, 64, UD, DO_VABSD)
394
395 #define DO_VABS(a) ((a < 0) ? (-a) : (a))
396
397 #define DO_VADDA(NAME, BIT, E) \
398 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
399 { \
400 int i; \
401 VReg *Vd = (VReg *)vd; \
402 VReg *Vj = (VReg *)vj; \
403 VReg *Vk = (VReg *)vk; \
404 int oprsz = simd_oprsz(desc); \
405 \
406 for (i = 0; i < oprsz / (BIT / 8); i++) { \
407 Vd->E(i) = DO_VABS(Vj->E(i)) + DO_VABS(Vk->E(i)); \
408 } \
409 }
410
411 DO_VADDA(vadda_b, 8, B)
412 DO_VADDA(vadda_h, 16, H)
413 DO_VADDA(vadda_w, 32, W)
414 DO_VADDA(vadda_d, 64, D)
415
416 #define DO_MIN(a, b) (a < b ? a : b)
417 #define DO_MAX(a, b) (a > b ? a : b)
418
419 #define VMINMAXI(NAME, BIT, E, DO_OP) \
420 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
421 { \
422 int i; \
423 VReg *Vd = (VReg *)vd; \
424 VReg *Vj = (VReg *)vj; \
425 typedef __typeof(Vd->E(0)) TD; \
426 int oprsz = simd_oprsz(desc); \
427 \
428 for (i = 0; i < oprsz / (BIT / 8); i++) { \
429 Vd->E(i) = DO_OP(Vj->E(i), (TD)imm); \
430 } \
431 }
432
433 VMINMAXI(vmini_b, 8, B, DO_MIN)
434 VMINMAXI(vmini_h, 16, H, DO_MIN)
435 VMINMAXI(vmini_w, 32, W, DO_MIN)
436 VMINMAXI(vmini_d, 64, D, DO_MIN)
437 VMINMAXI(vmaxi_b, 8, B, DO_MAX)
438 VMINMAXI(vmaxi_h, 16, H, DO_MAX)
439 VMINMAXI(vmaxi_w, 32, W, DO_MAX)
440 VMINMAXI(vmaxi_d, 64, D, DO_MAX)
441 VMINMAXI(vmini_bu, 8, UB, DO_MIN)
442 VMINMAXI(vmini_hu, 16, UH, DO_MIN)
443 VMINMAXI(vmini_wu, 32, UW, DO_MIN)
444 VMINMAXI(vmini_du, 64, UD, DO_MIN)
445 VMINMAXI(vmaxi_bu, 8, UB, DO_MAX)
446 VMINMAXI(vmaxi_hu, 16, UH, DO_MAX)
447 VMINMAXI(vmaxi_wu, 32, UW, DO_MAX)
448 VMINMAXI(vmaxi_du, 64, UD, DO_MAX)
449
450 #define DO_VMUH(NAME, BIT, E1, E2, DO_OP) \
451 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
452 { \
453 int i; \
454 VReg *Vd = (VReg *)vd; \
455 VReg *Vj = (VReg *)vj; \
456 VReg *Vk = (VReg *)vk; \
457 typedef __typeof(Vd->E1(0)) T; \
458 int oprsz = simd_oprsz(desc); \
459 \
460 for (i = 0; i < oprsz / (BIT / 8); i++) { \
461 Vd->E2(i) = ((T)Vj->E2(i)) * ((T)Vk->E2(i)) >> BIT; \
462 } \
463 }
464
465 void HELPER(vmuh_d)(void *vd, void *vj, void *vk, uint32_t desc)
466 {
467 int i;
468 uint64_t l, h;
469 VReg *Vd = (VReg *)vd;
470 VReg *Vj = (VReg *)vj;
471 VReg *Vk = (VReg *)vk;
472 int oprsz = simd_oprsz(desc);
473
474 for (i = 0; i < oprsz / 8; i++) {
475 muls64(&l, &h, Vj->D(i), Vk->D(i));
476 Vd->D(i) = h;
477 }
478 }
479
480 DO_VMUH(vmuh_b, 8, H, B, DO_MUH)
481 DO_VMUH(vmuh_h, 16, W, H, DO_MUH)
482 DO_VMUH(vmuh_w, 32, D, W, DO_MUH)
483
484 void HELPER(vmuh_du)(void *vd, void *vj, void *vk, uint32_t desc)
485 {
486 int i;
487 uint64_t l, h;
488 VReg *Vd = (VReg *)vd;
489 VReg *Vj = (VReg *)vj;
490 VReg *Vk = (VReg *)vk;
491 int oprsz = simd_oprsz(desc);
492
493 for (i = 0; i < oprsz / 8; i++) {
494 mulu64(&l, &h, Vj->D(i), Vk->D(i));
495 Vd->D(i) = h;
496 }
497 }
498
499 DO_VMUH(vmuh_bu, 8, UH, UB, DO_MUH)
500 DO_VMUH(vmuh_hu, 16, UW, UH, DO_MUH)
501 DO_VMUH(vmuh_wu, 32, UD, UW, DO_MUH)
502
503 #define DO_MUL(a, b) (a * b)
504
505 DO_EVEN(vmulwev_h_b, 16, H, B, DO_MUL)
506 DO_EVEN(vmulwev_w_h, 32, W, H, DO_MUL)
507 DO_EVEN(vmulwev_d_w, 64, D, W, DO_MUL)
508
509 DO_ODD(vmulwod_h_b, 16, H, B, DO_MUL)
510 DO_ODD(vmulwod_w_h, 32, W, H, DO_MUL)
511 DO_ODD(vmulwod_d_w, 64, D, W, DO_MUL)
512
513 DO_EVEN(vmulwev_h_bu, 16, UH, UB, DO_MUL)
514 DO_EVEN(vmulwev_w_hu, 32, UW, UH, DO_MUL)
515 DO_EVEN(vmulwev_d_wu, 64, UD, UW, DO_MUL)
516
517 DO_ODD(vmulwod_h_bu, 16, UH, UB, DO_MUL)
518 DO_ODD(vmulwod_w_hu, 32, UW, UH, DO_MUL)
519 DO_ODD(vmulwod_d_wu, 64, UD, UW, DO_MUL)
520
521 DO_EVEN_U_S(vmulwev_h_bu_b, 16, H, UH, B, UB, DO_MUL)
522 DO_EVEN_U_S(vmulwev_w_hu_h, 32, W, UW, H, UH, DO_MUL)
523 DO_EVEN_U_S(vmulwev_d_wu_w, 64, D, UD, W, UW, DO_MUL)
524
525 DO_ODD_U_S(vmulwod_h_bu_b, 16, H, UH, B, UB, DO_MUL)
526 DO_ODD_U_S(vmulwod_w_hu_h, 32, W, UW, H, UH, DO_MUL)
527 DO_ODD_U_S(vmulwod_d_wu_w, 64, D, UD, W, UW, DO_MUL)
528
529 #define DO_MADD(a, b, c) (a + b * c)
530 #define DO_MSUB(a, b, c) (a - b * c)
531
532 #define VMADDSUB(NAME, BIT, E, DO_OP) \
533 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
534 { \
535 int i; \
536 VReg *Vd = (VReg *)vd; \
537 VReg *Vj = (VReg *)vj; \
538 VReg *Vk = (VReg *)vk; \
539 int oprsz = simd_oprsz(desc); \
540 \
541 for (i = 0; i < oprsz / (BIT / 8); i++) { \
542 Vd->E(i) = DO_OP(Vd->E(i), Vj->E(i) ,Vk->E(i)); \
543 } \
544 }
545
546 VMADDSUB(vmadd_b, 8, B, DO_MADD)
547 VMADDSUB(vmadd_h, 16, H, DO_MADD)
548 VMADDSUB(vmadd_w, 32, W, DO_MADD)
549 VMADDSUB(vmadd_d, 64, D, DO_MADD)
550 VMADDSUB(vmsub_b, 8, B, DO_MSUB)
551 VMADDSUB(vmsub_h, 16, H, DO_MSUB)
552 VMADDSUB(vmsub_w, 32, W, DO_MSUB)
553 VMADDSUB(vmsub_d, 64, D, DO_MSUB)
554
555 #define VMADDWEV(NAME, BIT, E1, E2, DO_OP) \
556 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
557 { \
558 int i; \
559 VReg *Vd = (VReg *)vd; \
560 VReg *Vj = (VReg *)vj; \
561 VReg *Vk = (VReg *)vk; \
562 typedef __typeof(Vd->E1(0)) TD; \
563 int oprsz = simd_oprsz(desc); \
564 \
565 for (i = 0; i < oprsz / (BIT / 8); i++) { \
566 Vd->E1(i) += DO_OP((TD)Vj->E2(2 * i), (TD)Vk->E2(2 * i)); \
567 } \
568 }
569
570 VMADDWEV(vmaddwev_h_b, 16, H, B, DO_MUL)
571 VMADDWEV(vmaddwev_w_h, 32, W, H, DO_MUL)
572 VMADDWEV(vmaddwev_d_w, 64, D, W, DO_MUL)
573 VMADDWEV(vmaddwev_h_bu, 16, UH, UB, DO_MUL)
574 VMADDWEV(vmaddwev_w_hu, 32, UW, UH, DO_MUL)
575 VMADDWEV(vmaddwev_d_wu, 64, UD, UW, DO_MUL)
576
577 #define VMADDWOD(NAME, BIT, E1, E2, DO_OP) \
578 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
579 { \
580 int i; \
581 VReg *Vd = (VReg *)vd; \
582 VReg *Vj = (VReg *)vj; \
583 VReg *Vk = (VReg *)vk; \
584 typedef __typeof(Vd->E1(0)) TD; \
585 int oprsz = simd_oprsz(desc); \
586 \
587 for (i = 0; i < oprsz / (BIT / 8); i++) { \
588 Vd->E1(i) += DO_OP((TD)Vj->E2(2 * i + 1), \
589 (TD)Vk->E2(2 * i + 1)); \
590 } \
591 }
592
593 VMADDWOD(vmaddwod_h_b, 16, H, B, DO_MUL)
594 VMADDWOD(vmaddwod_w_h, 32, W, H, DO_MUL)
595 VMADDWOD(vmaddwod_d_w, 64, D, W, DO_MUL)
596 VMADDWOD(vmaddwod_h_bu, 16, UH, UB, DO_MUL)
597 VMADDWOD(vmaddwod_w_hu, 32, UW, UH, DO_MUL)
598 VMADDWOD(vmaddwod_d_wu, 64, UD, UW, DO_MUL)
599
600 #define VMADDWEV_U_S(NAME, BIT, ES1, EU1, ES2, EU2, DO_OP) \
601 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
602 { \
603 int i; \
604 VReg *Vd = (VReg *)vd; \
605 VReg *Vj = (VReg *)vj; \
606 VReg *Vk = (VReg *)vk; \
607 typedef __typeof(Vd->ES1(0)) TS1; \
608 typedef __typeof(Vd->EU1(0)) TU1; \
609 int oprsz = simd_oprsz(desc); \
610 \
611 for (i = 0; i < oprsz / (BIT / 8); i++) { \
612 Vd->ES1(i) += DO_OP((TU1)Vj->EU2(2 * i), \
613 (TS1)Vk->ES2(2 * i)); \
614 } \
615 }
616
617 VMADDWEV_U_S(vmaddwev_h_bu_b, 16, H, UH, B, UB, DO_MUL)
618 VMADDWEV_U_S(vmaddwev_w_hu_h, 32, W, UW, H, UH, DO_MUL)
619 VMADDWEV_U_S(vmaddwev_d_wu_w, 64, D, UD, W, UW, DO_MUL)
620
621 #define VMADDWOD_U_S(NAME, BIT, ES1, EU1, ES2, EU2, DO_OP) \
622 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
623 { \
624 int i; \
625 VReg *Vd = (VReg *)vd; \
626 VReg *Vj = (VReg *)vj; \
627 VReg *Vk = (VReg *)vk; \
628 typedef __typeof(Vd->ES1(0)) TS1; \
629 typedef __typeof(Vd->EU1(0)) TU1; \
630 int oprsz = simd_oprsz(desc); \
631 \
632 for (i = 0; i < oprsz / (BIT / 8); i++) { \
633 Vd->ES1(i) += DO_OP((TU1)Vj->EU2(2 * i + 1), \
634 (TS1)Vk->ES2(2 * i + 1)); \
635 } \
636 }
637
638 VMADDWOD_U_S(vmaddwod_h_bu_b, 16, H, UH, B, UB, DO_MUL)
639 VMADDWOD_U_S(vmaddwod_w_hu_h, 32, W, UW, H, UH, DO_MUL)
640 VMADDWOD_U_S(vmaddwod_d_wu_w, 64, D, UD, W, UW, DO_MUL)
641
642 #define DO_DIVU(N, M) (unlikely(M == 0) ? 0 : N / M)
643 #define DO_REMU(N, M) (unlikely(M == 0) ? 0 : N % M)
644 #define DO_DIV(N, M) (unlikely(M == 0) ? 0 :\
645 unlikely((N == -N) && (M == (__typeof(N))(-1))) ? N : N / M)
646 #define DO_REM(N, M) (unlikely(M == 0) ? 0 :\
647 unlikely((N == -N) && (M == (__typeof(N))(-1))) ? 0 : N % M)
648
649 #define VDIV(NAME, BIT, E, DO_OP) \
650 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
651 { \
652 int i; \
653 VReg *Vd = (VReg *)vd; \
654 VReg *Vj = (VReg *)vj; \
655 VReg *Vk = (VReg *)vk; \
656 int oprsz = simd_oprsz(desc); \
657 \
658 for (i = 0; i < oprsz / (BIT / 8); i++) { \
659 Vd->E(i) = DO_OP(Vj->E(i), Vk->E(i)); \
660 } \
661 }
662
663 VDIV(vdiv_b, 8, B, DO_DIV)
664 VDIV(vdiv_h, 16, H, DO_DIV)
665 VDIV(vdiv_w, 32, W, DO_DIV)
666 VDIV(vdiv_d, 64, D, DO_DIV)
667 VDIV(vdiv_bu, 8, UB, DO_DIVU)
668 VDIV(vdiv_hu, 16, UH, DO_DIVU)
669 VDIV(vdiv_wu, 32, UW, DO_DIVU)
670 VDIV(vdiv_du, 64, UD, DO_DIVU)
671 VDIV(vmod_b, 8, B, DO_REM)
672 VDIV(vmod_h, 16, H, DO_REM)
673 VDIV(vmod_w, 32, W, DO_REM)
674 VDIV(vmod_d, 64, D, DO_REM)
675 VDIV(vmod_bu, 8, UB, DO_REMU)
676 VDIV(vmod_hu, 16, UH, DO_REMU)
677 VDIV(vmod_wu, 32, UW, DO_REMU)
678 VDIV(vmod_du, 64, UD, DO_REMU)
679
680 #define VSAT_S(NAME, BIT, E) \
681 void HELPER(NAME)(void *vd, void *vj, uint64_t max, uint32_t desc) \
682 { \
683 int i; \
684 VReg *Vd = (VReg *)vd; \
685 VReg *Vj = (VReg *)vj; \
686 typedef __typeof(Vd->E(0)) TD; \
687 int oprsz = simd_oprsz(desc); \
688 \
689 for (i = 0; i < oprsz / (BIT / 8); i++) { \
690 Vd->E(i) = Vj->E(i) > (TD)max ? (TD)max : \
691 Vj->E(i) < (TD)~max ? (TD)~max: Vj->E(i); \
692 } \
693 }
694
695 VSAT_S(vsat_b, 8, B)
696 VSAT_S(vsat_h, 16, H)
697 VSAT_S(vsat_w, 32, W)
698 VSAT_S(vsat_d, 64, D)
699
700 #define VSAT_U(NAME, BIT, E) \
701 void HELPER(NAME)(void *vd, void *vj, uint64_t max, uint32_t desc) \
702 { \
703 int i; \
704 VReg *Vd = (VReg *)vd; \
705 VReg *Vj = (VReg *)vj; \
706 typedef __typeof(Vd->E(0)) TD; \
707 int oprsz = simd_oprsz(desc); \
708 \
709 for (i = 0; i < oprsz / (BIT / 8); i++) { \
710 Vd->E(i) = Vj->E(i) > (TD)max ? (TD)max : Vj->E(i); \
711 } \
712 }
713
714 VSAT_U(vsat_bu, 8, UB)
715 VSAT_U(vsat_hu, 16, UH)
716 VSAT_U(vsat_wu, 32, UW)
717 VSAT_U(vsat_du, 64, UD)
718
719 #define VEXTH(NAME, BIT, E1, E2) \
720 void HELPER(NAME)(void *vd, void *vj, uint32_t desc) \
721 { \
722 int i, j, ofs; \
723 VReg *Vd = (VReg *)vd; \
724 VReg *Vj = (VReg *)vj; \
725 int oprsz = simd_oprsz(desc); \
726 \
727 ofs = LSX_LEN / BIT; \
728 for (i = 0; i < oprsz / 16; i++) { \
729 for (j = 0; j < ofs; j++) { \
730 Vd->E1(j + i * ofs) = Vj->E2(j + ofs + ofs * 2 * i); \
731 } \
732 } \
733 }
734
735 void HELPER(vexth_q_d)(void *vd, void *vj, uint32_t desc)
736 {
737 int i;
738 VReg *Vd = (VReg *)vd;
739 VReg *Vj = (VReg *)vj;
740 int oprsz = simd_oprsz(desc);
741
742 for (i = 0; i < oprsz / 16; i++) {
743 Vd->Q(i) = int128_makes64(Vj->D(2 * i + 1));
744 }
745 }
746
747 void HELPER(vexth_qu_du)(void *vd, void *vj, uint32_t desc)
748 {
749 int i;
750 VReg *Vd = (VReg *)vd;
751 VReg *Vj = (VReg *)vj;
752 int oprsz = simd_oprsz(desc);
753
754 for (i = 0; i < oprsz / 16; i++) {
755 Vd->Q(i) = int128_make64(Vj->UD(2 * i + 1));
756 }
757 }
758
759 VEXTH(vexth_h_b, 16, H, B)
760 VEXTH(vexth_w_h, 32, W, H)
761 VEXTH(vexth_d_w, 64, D, W)
762 VEXTH(vexth_hu_bu, 16, UH, UB)
763 VEXTH(vexth_wu_hu, 32, UW, UH)
764 VEXTH(vexth_du_wu, 64, UD, UW)
765
766 #define VEXT2XV(NAME, BIT, E1, E2) \
767 void HELPER(NAME)(void *vd, void *vj, uint32_t desc) \
768 { \
769 int i; \
770 VReg temp = {}; \
771 VReg *Vd = (VReg *)vd; \
772 VReg *Vj = (VReg *)vj; \
773 int oprsz = simd_oprsz(desc); \
774 \
775 for (i = 0; i < oprsz / (BIT / 8); i++) { \
776 temp.E1(i) = Vj->E2(i); \
777 } \
778 *Vd = temp; \
779 }
780
781 VEXT2XV(vext2xv_h_b, 16, H, B)
782 VEXT2XV(vext2xv_w_b, 32, W, B)
783 VEXT2XV(vext2xv_d_b, 64, D, B)
784 VEXT2XV(vext2xv_w_h, 32, W, H)
785 VEXT2XV(vext2xv_d_h, 64, D, H)
786 VEXT2XV(vext2xv_d_w, 64, D, W)
787 VEXT2XV(vext2xv_hu_bu, 16, UH, UB)
788 VEXT2XV(vext2xv_wu_bu, 32, UW, UB)
789 VEXT2XV(vext2xv_du_bu, 64, UD, UB)
790 VEXT2XV(vext2xv_wu_hu, 32, UW, UH)
791 VEXT2XV(vext2xv_du_hu, 64, UD, UH)
792 VEXT2XV(vext2xv_du_wu, 64, UD, UW)
793
794 #define DO_SIGNCOV(a, b) (a == 0 ? 0 : a < 0 ? -b : b)
795
796 DO_3OP(vsigncov_b, 8, B, DO_SIGNCOV)
797 DO_3OP(vsigncov_h, 16, H, DO_SIGNCOV)
798 DO_3OP(vsigncov_w, 32, W, DO_SIGNCOV)
799 DO_3OP(vsigncov_d, 64, D, DO_SIGNCOV)
800
801 static uint64_t do_vmskltz_b(int64_t val)
802 {
803 uint64_t m = 0x8080808080808080ULL;
804 uint64_t c = val & m;
805 c |= c << 7;
806 c |= c << 14;
807 c |= c << 28;
808 return c >> 56;
809 }
810
811 void HELPER(vmskltz_b)(void *vd, void *vj, uint32_t desc)
812 {
813 int i;
814 uint16_t temp = 0;
815 VReg *Vd = (VReg *)vd;
816 VReg *Vj = (VReg *)vj;
817 int oprsz = simd_oprsz(desc);
818
819 for (i = 0; i < oprsz / 16; i++) {
820 temp = 0;
821 temp = do_vmskltz_b(Vj->D(2 * i));
822 temp |= (do_vmskltz_b(Vj->D(2 * i + 1)) << 8);
823 Vd->D(2 * i) = temp;
824 Vd->D(2 * i + 1) = 0;
825 }
826 }
827
828 static uint64_t do_vmskltz_h(int64_t val)
829 {
830 uint64_t m = 0x8000800080008000ULL;
831 uint64_t c = val & m;
832 c |= c << 15;
833 c |= c << 30;
834 return c >> 60;
835 }
836
837 void HELPER(vmskltz_h)(void *vd, void *vj, uint32_t desc)
838 {
839 int i;
840 uint16_t temp = 0;
841 VReg *Vd = (VReg *)vd;
842 VReg *Vj = (VReg *)vj;
843 int oprsz = simd_oprsz(desc);
844
845 for (i = 0; i < oprsz / 16; i++) {
846 temp = 0;
847 temp = do_vmskltz_h(Vj->D(2 * i));
848 temp |= (do_vmskltz_h(Vj->D(2 * i + 1)) << 4);
849 Vd->D(2 * i) = temp;
850 Vd->D(2 * i + 1) = 0;
851 }
852 }
853
854 static uint64_t do_vmskltz_w(int64_t val)
855 {
856 uint64_t m = 0x8000000080000000ULL;
857 uint64_t c = val & m;
858 c |= c << 31;
859 return c >> 62;
860 }
861
862 void HELPER(vmskltz_w)(void *vd, void *vj, uint32_t desc)
863 {
864 int i;
865 uint16_t temp = 0;
866 VReg *Vd = (VReg *)vd;
867 VReg *Vj = (VReg *)vj;
868 int oprsz = simd_oprsz(desc);
869
870 for (i = 0; i < oprsz / 16; i++) {
871 temp = 0;
872 temp = do_vmskltz_w(Vj->D(2 * i));
873 temp |= (do_vmskltz_w(Vj->D(2 * i + 1)) << 2);
874 Vd->D(2 * i) = temp;
875 Vd->D(2 * i + 1) = 0;
876 }
877 }
878
879 static uint64_t do_vmskltz_d(int64_t val)
880 {
881 return (uint64_t)val >> 63;
882 }
883 void HELPER(vmskltz_d)(void *vd, void *vj, uint32_t desc)
884 {
885 int i;
886 uint16_t temp = 0;
887 VReg *Vd = (VReg *)vd;
888 VReg *Vj = (VReg *)vj;
889 int oprsz = simd_oprsz(desc);
890
891 for (i = 0; i < oprsz / 16; i++) {
892 temp = 0;
893 temp = do_vmskltz_d(Vj->D(2 * i));
894 temp |= (do_vmskltz_d(Vj->D(2 * i + 1)) << 1);
895 Vd->D(2 * i) = temp;
896 Vd->D(2 * i + 1) = 0;
897 }
898 }
899
900 void HELPER(vmskgez_b)(void *vd, void *vj, uint32_t desc)
901 {
902 int i;
903 uint16_t temp = 0;
904 VReg *Vd = (VReg *)vd;
905 VReg *Vj = (VReg *)vj;
906 int oprsz = simd_oprsz(desc);
907
908 for (i = 0; i < oprsz / 16; i++) {
909 temp = 0;
910 temp = do_vmskltz_b(Vj->D(2 * i));
911 temp |= (do_vmskltz_b(Vj->D(2 * i + 1)) << 8);
912 Vd->D(2 * i) = (uint16_t)(~temp);
913 Vd->D(2 * i + 1) = 0;
914 }
915 }
916
917 static uint64_t do_vmskez_b(uint64_t a)
918 {
919 uint64_t m = 0x7f7f7f7f7f7f7f7fULL;
920 uint64_t c = ~(((a & m) + m) | a | m);
921 c |= c << 7;
922 c |= c << 14;
923 c |= c << 28;
924 return c >> 56;
925 }
926
927 void HELPER(vmsknz_b)(void *vd, void *vj, uint32_t desc)
928 {
929 int i;
930 uint16_t temp = 0;
931 VReg *Vd = (VReg *)vd;
932 VReg *Vj = (VReg *)vj;
933 int oprsz = simd_oprsz(desc);
934
935 for (i = 0; i < oprsz / 16; i++) {
936 temp = 0;
937 temp = do_vmskez_b(Vj->D(2 * i));
938 temp |= (do_vmskez_b(Vj->D(2 * i + 1)) << 8);
939 Vd->D(2 * i) = (uint16_t)(~temp);
940 Vd->D(2 * i + 1) = 0;
941 }
942 }
943
944 void HELPER(vnori_b)(void *vd, void *vj, uint64_t imm, uint32_t desc)
945 {
946 int i;
947 VReg *Vd = (VReg *)vd;
948 VReg *Vj = (VReg *)vj;
949
950 for (i = 0; i < simd_oprsz(desc); i++) {
951 Vd->B(i) = ~(Vj->B(i) | (uint8_t)imm);
952 }
953 }
954
955 #define VSLLWIL(NAME, BIT, E1, E2) \
956 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
957 { \
958 int i, j, ofs; \
959 VReg temp = {}; \
960 VReg *Vd = (VReg *)vd; \
961 VReg *Vj = (VReg *)vj; \
962 int oprsz = simd_oprsz(desc); \
963 typedef __typeof(temp.E1(0)) TD; \
964 \
965 ofs = LSX_LEN / BIT; \
966 for (i = 0; i < oprsz / 16; i++) { \
967 for (j = 0; j < ofs; j++) { \
968 temp.E1(j + ofs * i) = (TD)Vj->E2(j + ofs * 2 * i) << (imm % BIT); \
969 } \
970 } \
971 *Vd = temp; \
972 }
973
974
975 void HELPER(vextl_q_d)(void *vd, void *vj, uint32_t desc)
976 {
977 int i;
978 VReg *Vd = (VReg *)vd;
979 VReg *Vj = (VReg *)vj;
980 int oprsz = simd_oprsz(desc);
981
982 for (i = 0; i < oprsz / 16; i++) {
983 Vd->Q(i) = int128_makes64(Vj->D(2 * i));
984 }
985 }
986
987 void HELPER(vextl_qu_du)(void *vd, void *vj, uint32_t desc)
988 {
989 int i;
990 VReg *Vd = (VReg *)vd;
991 VReg *Vj = (VReg *)vj;
992 int oprsz = simd_oprsz(desc);
993
994 for (i = 0; i < oprsz / 16; i++) {
995 Vd->Q(i) = int128_make64(Vj->UD(2 * i));
996 }
997 }
998
999 VSLLWIL(vsllwil_h_b, 16, H, B)
1000 VSLLWIL(vsllwil_w_h, 32, W, H)
1001 VSLLWIL(vsllwil_d_w, 64, D, W)
1002 VSLLWIL(vsllwil_hu_bu, 16, UH, UB)
1003 VSLLWIL(vsllwil_wu_hu, 32, UW, UH)
1004 VSLLWIL(vsllwil_du_wu, 64, UD, UW)
1005
1006 #define do_vsrlr(E, T) \
1007 static T do_vsrlr_ ##E(T s1, int sh) \
1008 { \
1009 if (sh == 0) { \
1010 return s1; \
1011 } else { \
1012 return (s1 >> sh) + ((s1 >> (sh - 1)) & 0x1); \
1013 } \
1014 }
1015
1016 do_vsrlr(B, uint8_t)
1017 do_vsrlr(H, uint16_t)
1018 do_vsrlr(W, uint32_t)
1019 do_vsrlr(D, uint64_t)
1020
1021 #define VSRLR(NAME, BIT, T, E) \
1022 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1023 { \
1024 int i; \
1025 VReg *Vd = (VReg *)vd; \
1026 VReg *Vj = (VReg *)vj; \
1027 VReg *Vk = (VReg *)vk; \
1028 int oprsz = simd_oprsz(desc); \
1029 \
1030 for (i = 0; i < oprsz / (BIT / 8); i++) { \
1031 Vd->E(i) = do_vsrlr_ ## E(Vj->E(i), ((T)Vk->E(i))%BIT); \
1032 } \
1033 }
1034
1035 VSRLR(vsrlr_b, 8, uint8_t, B)
1036 VSRLR(vsrlr_h, 16, uint16_t, H)
1037 VSRLR(vsrlr_w, 32, uint32_t, W)
1038 VSRLR(vsrlr_d, 64, uint64_t, D)
1039
1040 #define VSRLRI(NAME, BIT, E) \
1041 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1042 { \
1043 int i; \
1044 VReg *Vd = (VReg *)vd; \
1045 VReg *Vj = (VReg *)vj; \
1046 int oprsz = simd_oprsz(desc); \
1047 \
1048 for (i = 0; i < oprsz / (BIT / 8); i++) { \
1049 Vd->E(i) = do_vsrlr_ ## E(Vj->E(i), imm); \
1050 } \
1051 }
1052
1053 VSRLRI(vsrlri_b, 8, B)
1054 VSRLRI(vsrlri_h, 16, H)
1055 VSRLRI(vsrlri_w, 32, W)
1056 VSRLRI(vsrlri_d, 64, D)
1057
1058 #define do_vsrar(E, T) \
1059 static T do_vsrar_ ##E(T s1, int sh) \
1060 { \
1061 if (sh == 0) { \
1062 return s1; \
1063 } else { \
1064 return (s1 >> sh) + ((s1 >> (sh - 1)) & 0x1); \
1065 } \
1066 }
1067
1068 do_vsrar(B, int8_t)
1069 do_vsrar(H, int16_t)
1070 do_vsrar(W, int32_t)
1071 do_vsrar(D, int64_t)
1072
1073 #define VSRAR(NAME, BIT, T, E) \
1074 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1075 { \
1076 int i; \
1077 VReg *Vd = (VReg *)vd; \
1078 VReg *Vj = (VReg *)vj; \
1079 VReg *Vk = (VReg *)vk; \
1080 int oprsz = simd_oprsz(desc); \
1081 \
1082 for (i = 0; i < oprsz / (BIT / 8); i++) { \
1083 Vd->E(i) = do_vsrar_ ## E(Vj->E(i), ((T)Vk->E(i))%BIT); \
1084 } \
1085 }
1086
1087 VSRAR(vsrar_b, 8, uint8_t, B)
1088 VSRAR(vsrar_h, 16, uint16_t, H)
1089 VSRAR(vsrar_w, 32, uint32_t, W)
1090 VSRAR(vsrar_d, 64, uint64_t, D)
1091
1092 #define VSRARI(NAME, BIT, E) \
1093 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1094 { \
1095 int i; \
1096 VReg *Vd = (VReg *)vd; \
1097 VReg *Vj = (VReg *)vj; \
1098 int oprsz = simd_oprsz(desc); \
1099 \
1100 for (i = 0; i < oprsz / (BIT / 8); i++) { \
1101 Vd->E(i) = do_vsrar_ ## E(Vj->E(i), imm); \
1102 } \
1103 }
1104
1105 VSRARI(vsrari_b, 8, B)
1106 VSRARI(vsrari_h, 16, H)
1107 VSRARI(vsrari_w, 32, W)
1108 VSRARI(vsrari_d, 64, D)
1109
1110 #define R_SHIFT(a, b) (a >> b)
1111
1112 #define VSRLN(NAME, BIT, E1, E2) \
1113 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1114 { \
1115 int i, j, ofs; \
1116 VReg *Vd = (VReg *)vd; \
1117 VReg *Vj = (VReg *)vj; \
1118 VReg *Vk = (VReg *)vk; \
1119 int oprsz = simd_oprsz(desc); \
1120 \
1121 ofs = LSX_LEN / BIT; \
1122 for (i = 0; i < oprsz / 16; i++) { \
1123 for (j = 0; j < ofs; j++) { \
1124 Vd->E1(j + ofs * 2 * i) = R_SHIFT(Vj->E2(j + ofs * i), \
1125 Vk->E2(j + ofs * i) % BIT); \
1126 } \
1127 Vd->D(2 * i + 1) = 0; \
1128 } \
1129 }
1130
1131 VSRLN(vsrln_b_h, 16, B, UH)
1132 VSRLN(vsrln_h_w, 32, H, UW)
1133 VSRLN(vsrln_w_d, 64, W, UD)
1134
1135 #define VSRAN(NAME, BIT, E1, E2, E3) \
1136 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1137 { \
1138 int i, j, ofs; \
1139 VReg *Vd = (VReg *)vd; \
1140 VReg *Vj = (VReg *)vj; \
1141 VReg *Vk = (VReg *)vk; \
1142 int oprsz = simd_oprsz(desc); \
1143 \
1144 ofs = LSX_LEN / BIT; \
1145 for (i = 0; i < oprsz / 16; i++) { \
1146 for (j = 0; j < ofs; j++) { \
1147 Vd->E1(j + ofs * 2 * i) = R_SHIFT(Vj->E2(j + ofs * i), \
1148 Vk->E3(j + ofs * i) % BIT); \
1149 } \
1150 Vd->D(2 * i + 1) = 0; \
1151 } \
1152 }
1153
1154 VSRAN(vsran_b_h, 16, B, H, UH)
1155 VSRAN(vsran_h_w, 32, H, W, UW)
1156 VSRAN(vsran_w_d, 64, W, D, UD)
1157
1158 #define VSRLNI(NAME, BIT, E1, E2) \
1159 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1160 { \
1161 int i, j, ofs; \
1162 VReg temp = {}; \
1163 VReg *Vd = (VReg *)vd; \
1164 VReg *Vj = (VReg *)vj; \
1165 int oprsz = simd_oprsz(desc); \
1166 \
1167 ofs = LSX_LEN / BIT; \
1168 for (i = 0; i < oprsz / 16; i++) { \
1169 for (j = 0; j < ofs; j++) { \
1170 temp.E1(j + ofs * 2 * i) = R_SHIFT(Vj->E2(j + ofs * i), imm); \
1171 temp.E1(j + ofs * (2 * i + 1)) = R_SHIFT(Vd->E2(j + ofs * i), \
1172 imm); \
1173 } \
1174 } \
1175 *Vd = temp; \
1176 }
1177
1178 void HELPER(vsrlni_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1179 {
1180 int i;
1181 VReg temp = {};
1182 VReg *Vd = (VReg *)vd;
1183 VReg *Vj = (VReg *)vj;
1184
1185 for (i = 0; i < 2; i++) {
1186 temp.D(2 * i) = int128_getlo(int128_urshift(Vj->Q(i), imm % 128));
1187 temp.D(2 * i +1) = int128_getlo(int128_urshift(Vd->Q(i), imm % 128));
1188 }
1189 *Vd = temp;
1190 }
1191
1192 VSRLNI(vsrlni_b_h, 16, B, UH)
1193 VSRLNI(vsrlni_h_w, 32, H, UW)
1194 VSRLNI(vsrlni_w_d, 64, W, UD)
1195
1196 #define VSRANI(NAME, BIT, E1, E2) \
1197 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1198 { \
1199 int i, j, ofs; \
1200 VReg temp = {}; \
1201 VReg *Vd = (VReg *)vd; \
1202 VReg *Vj = (VReg *)vj; \
1203 int oprsz = simd_oprsz(desc); \
1204 \
1205 ofs = LSX_LEN / BIT; \
1206 for (i = 0; i < oprsz / 16; i++) { \
1207 for (j = 0; j < ofs; j++) { \
1208 temp.E1(j + ofs * 2 * i) = R_SHIFT(Vj->E2(j + ofs * i), imm); \
1209 temp.E1(j + ofs * (2 * i + 1)) = R_SHIFT(Vd->E2(j + ofs * i), \
1210 imm); \
1211 } \
1212 } \
1213 *Vd = temp; \
1214 }
1215
1216 void HELPER(vsrani_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1217 {
1218 int i;
1219 VReg temp = {};
1220 VReg *Vd = (VReg *)vd;
1221 VReg *Vj = (VReg *)vj;
1222
1223 for (i = 0; i < 2; i++) {
1224 temp.D(2 * i) = int128_getlo(int128_rshift(Vj->Q(i), imm % 128));
1225 temp.D(2 * i + 1) = int128_getlo(int128_rshift(Vd->Q(i), imm % 128));
1226 }
1227 *Vd = temp;
1228 }
1229
1230 VSRANI(vsrani_b_h, 16, B, H)
1231 VSRANI(vsrani_h_w, 32, H, W)
1232 VSRANI(vsrani_w_d, 64, W, D)
1233
1234 #define VSRLRN(NAME, BIT, T, E1, E2) \
1235 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1236 { \
1237 int i; \
1238 VReg *Vd = (VReg *)vd; \
1239 VReg *Vj = (VReg *)vj; \
1240 VReg *Vk = (VReg *)vk; \
1241 \
1242 for (i = 0; i < LSX_LEN/BIT; i++) { \
1243 Vd->E1(i) = do_vsrlr_ ## E2(Vj->E2(i), ((T)Vk->E2(i))%BIT); \
1244 } \
1245 Vd->D(1) = 0; \
1246 }
1247
1248 VSRLRN(vsrlrn_b_h, 16, uint16_t, B, H)
1249 VSRLRN(vsrlrn_h_w, 32, uint32_t, H, W)
1250 VSRLRN(vsrlrn_w_d, 64, uint64_t, W, D)
1251
1252 #define VSRARN(NAME, BIT, T, E1, E2) \
1253 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1254 { \
1255 int i; \
1256 VReg *Vd = (VReg *)vd; \
1257 VReg *Vj = (VReg *)vj; \
1258 VReg *Vk = (VReg *)vk; \
1259 \
1260 for (i = 0; i < LSX_LEN/BIT; i++) { \
1261 Vd->E1(i) = do_vsrar_ ## E2(Vj->E2(i), ((T)Vk->E2(i))%BIT); \
1262 } \
1263 Vd->D(1) = 0; \
1264 }
1265
1266 VSRARN(vsrarn_b_h, 16, uint8_t, B, H)
1267 VSRARN(vsrarn_h_w, 32, uint16_t, H, W)
1268 VSRARN(vsrarn_w_d, 64, uint32_t, W, D)
1269
1270 #define VSRLRNI(NAME, BIT, E1, E2) \
1271 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1272 { \
1273 int i, max; \
1274 VReg temp; \
1275 VReg *Vd = (VReg *)vd; \
1276 VReg *Vj = (VReg *)vj; \
1277 \
1278 temp.D(0) = 0; \
1279 temp.D(1) = 0; \
1280 max = LSX_LEN/BIT; \
1281 for (i = 0; i < max; i++) { \
1282 temp.E1(i) = do_vsrlr_ ## E2(Vj->E2(i), imm); \
1283 temp.E1(i + max) = do_vsrlr_ ## E2(Vd->E2(i), imm); \
1284 } \
1285 *Vd = temp; \
1286 }
1287
1288 void HELPER(vsrlrni_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1289 {
1290 VReg temp;
1291 VReg *Vd = (VReg *)vd;
1292 VReg *Vj = (VReg *)vj;
1293 Int128 r1, r2;
1294
1295 if (imm == 0) {
1296 temp.D(0) = int128_getlo(Vj->Q(0));
1297 temp.D(1) = int128_getlo(Vd->Q(0));
1298 } else {
1299 r1 = int128_and(int128_urshift(Vj->Q(0), (imm -1)), int128_one());
1300 r2 = int128_and(int128_urshift(Vd->Q(0), (imm -1)), int128_one());
1301
1302 temp.D(0) = int128_getlo(int128_add(int128_urshift(Vj->Q(0), imm), r1));
1303 temp.D(1) = int128_getlo(int128_add(int128_urshift(Vd->Q(0), imm), r2));
1304 }
1305 *Vd = temp;
1306 }
1307
1308 VSRLRNI(vsrlrni_b_h, 16, B, H)
1309 VSRLRNI(vsrlrni_h_w, 32, H, W)
1310 VSRLRNI(vsrlrni_w_d, 64, W, D)
1311
1312 #define VSRARNI(NAME, BIT, E1, E2) \
1313 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1314 { \
1315 int i, max; \
1316 VReg temp; \
1317 VReg *Vd = (VReg *)vd; \
1318 VReg *Vj = (VReg *)vj; \
1319 \
1320 temp.D(0) = 0; \
1321 temp.D(1) = 0; \
1322 max = LSX_LEN/BIT; \
1323 for (i = 0; i < max; i++) { \
1324 temp.E1(i) = do_vsrar_ ## E2(Vj->E2(i), imm); \
1325 temp.E1(i + max) = do_vsrar_ ## E2(Vd->E2(i), imm); \
1326 } \
1327 *Vd = temp; \
1328 }
1329
1330 void HELPER(vsrarni_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1331 {
1332 VReg temp;
1333 VReg *Vd = (VReg *)vd;
1334 VReg *Vj = (VReg *)vj;
1335 Int128 r1, r2;
1336
1337 if (imm == 0) {
1338 temp.D(0) = int128_getlo(Vj->Q(0));
1339 temp.D(1) = int128_getlo(Vd->Q(0));
1340 } else {
1341 r1 = int128_and(int128_rshift(Vj->Q(0), (imm -1)), int128_one());
1342 r2 = int128_and(int128_rshift(Vd->Q(0), (imm -1)), int128_one());
1343
1344 temp.D(0) = int128_getlo(int128_add(int128_rshift(Vj->Q(0), imm), r1));
1345 temp.D(1) = int128_getlo(int128_add(int128_rshift(Vd->Q(0), imm), r2));
1346 }
1347 *Vd = temp;
1348 }
1349
1350 VSRARNI(vsrarni_b_h, 16, B, H)
1351 VSRARNI(vsrarni_h_w, 32, H, W)
1352 VSRARNI(vsrarni_w_d, 64, W, D)
1353
1354 #define SSRLNS(NAME, T1, T2, T3) \
1355 static T1 do_ssrlns_ ## NAME(T2 e2, int sa, int sh) \
1356 { \
1357 T1 shft_res; \
1358 if (sa == 0) { \
1359 shft_res = e2; \
1360 } else { \
1361 shft_res = (((T1)e2) >> sa); \
1362 } \
1363 T3 mask; \
1364 mask = (1ull << sh) -1; \
1365 if (shft_res > mask) { \
1366 return mask; \
1367 } else { \
1368 return shft_res; \
1369 } \
1370 }
1371
1372 SSRLNS(B, uint16_t, int16_t, uint8_t)
1373 SSRLNS(H, uint32_t, int32_t, uint16_t)
1374 SSRLNS(W, uint64_t, int64_t, uint32_t)
1375
1376 #define VSSRLN(NAME, BIT, T, E1, E2) \
1377 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1378 { \
1379 int i; \
1380 VReg *Vd = (VReg *)vd; \
1381 VReg *Vj = (VReg *)vj; \
1382 VReg *Vk = (VReg *)vk; \
1383 \
1384 for (i = 0; i < LSX_LEN/BIT; i++) { \
1385 Vd->E1(i) = do_ssrlns_ ## E1(Vj->E2(i), (T)Vk->E2(i)% BIT, BIT/2 -1); \
1386 } \
1387 Vd->D(1) = 0; \
1388 }
1389
1390 VSSRLN(vssrln_b_h, 16, uint16_t, B, H)
1391 VSSRLN(vssrln_h_w, 32, uint32_t, H, W)
1392 VSSRLN(vssrln_w_d, 64, uint64_t, W, D)
1393
1394 #define SSRANS(E, T1, T2) \
1395 static T1 do_ssrans_ ## E(T1 e2, int sa, int sh) \
1396 { \
1397 T1 shft_res; \
1398 if (sa == 0) { \
1399 shft_res = e2; \
1400 } else { \
1401 shft_res = e2 >> sa; \
1402 } \
1403 T2 mask; \
1404 mask = (1ll << sh) -1; \
1405 if (shft_res > mask) { \
1406 return mask; \
1407 } else if (shft_res < -(mask +1)) { \
1408 return ~mask; \
1409 } else { \
1410 return shft_res; \
1411 } \
1412 }
1413
1414 SSRANS(B, int16_t, int8_t)
1415 SSRANS(H, int32_t, int16_t)
1416 SSRANS(W, int64_t, int32_t)
1417
1418 #define VSSRAN(NAME, BIT, T, E1, E2) \
1419 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1420 { \
1421 int i; \
1422 VReg *Vd = (VReg *)vd; \
1423 VReg *Vj = (VReg *)vj; \
1424 VReg *Vk = (VReg *)vk; \
1425 \
1426 for (i = 0; i < LSX_LEN/BIT; i++) { \
1427 Vd->E1(i) = do_ssrans_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2 -1); \
1428 } \
1429 Vd->D(1) = 0; \
1430 }
1431
1432 VSSRAN(vssran_b_h, 16, uint16_t, B, H)
1433 VSSRAN(vssran_h_w, 32, uint32_t, H, W)
1434 VSSRAN(vssran_w_d, 64, uint64_t, W, D)
1435
1436 #define SSRLNU(E, T1, T2, T3) \
1437 static T1 do_ssrlnu_ ## E(T3 e2, int sa, int sh) \
1438 { \
1439 T1 shft_res; \
1440 if (sa == 0) { \
1441 shft_res = e2; \
1442 } else { \
1443 shft_res = (((T1)e2) >> sa); \
1444 } \
1445 T2 mask; \
1446 mask = (1ull << sh) -1; \
1447 if (shft_res > mask) { \
1448 return mask; \
1449 } else { \
1450 return shft_res; \
1451 } \
1452 }
1453
1454 SSRLNU(B, uint16_t, uint8_t, int16_t)
1455 SSRLNU(H, uint32_t, uint16_t, int32_t)
1456 SSRLNU(W, uint64_t, uint32_t, int64_t)
1457
1458 #define VSSRLNU(NAME, BIT, T, E1, E2) \
1459 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1460 { \
1461 int i; \
1462 VReg *Vd = (VReg *)vd; \
1463 VReg *Vj = (VReg *)vj; \
1464 VReg *Vk = (VReg *)vk; \
1465 \
1466 for (i = 0; i < LSX_LEN/BIT; i++) { \
1467 Vd->E1(i) = do_ssrlnu_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2); \
1468 } \
1469 Vd->D(1) = 0; \
1470 }
1471
1472 VSSRLNU(vssrln_bu_h, 16, uint16_t, B, H)
1473 VSSRLNU(vssrln_hu_w, 32, uint32_t, H, W)
1474 VSSRLNU(vssrln_wu_d, 64, uint64_t, W, D)
1475
1476 #define SSRANU(E, T1, T2, T3) \
1477 static T1 do_ssranu_ ## E(T3 e2, int sa, int sh) \
1478 { \
1479 T1 shft_res; \
1480 if (sa == 0) { \
1481 shft_res = e2; \
1482 } else { \
1483 shft_res = e2 >> sa; \
1484 } \
1485 if (e2 < 0) { \
1486 shft_res = 0; \
1487 } \
1488 T2 mask; \
1489 mask = (1ull << sh) -1; \
1490 if (shft_res > mask) { \
1491 return mask; \
1492 } else { \
1493 return shft_res; \
1494 } \
1495 }
1496
1497 SSRANU(B, uint16_t, uint8_t, int16_t)
1498 SSRANU(H, uint32_t, uint16_t, int32_t)
1499 SSRANU(W, uint64_t, uint32_t, int64_t)
1500
1501 #define VSSRANU(NAME, BIT, T, E1, E2) \
1502 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1503 { \
1504 int i; \
1505 VReg *Vd = (VReg *)vd; \
1506 VReg *Vj = (VReg *)vj; \
1507 VReg *Vk = (VReg *)vk; \
1508 \
1509 for (i = 0; i < LSX_LEN/BIT; i++) { \
1510 Vd->E1(i) = do_ssranu_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2); \
1511 } \
1512 Vd->D(1) = 0; \
1513 }
1514
1515 VSSRANU(vssran_bu_h, 16, uint16_t, B, H)
1516 VSSRANU(vssran_hu_w, 32, uint32_t, H, W)
1517 VSSRANU(vssran_wu_d, 64, uint64_t, W, D)
1518
1519 #define VSSRLNI(NAME, BIT, E1, E2) \
1520 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1521 { \
1522 int i; \
1523 VReg temp; \
1524 VReg *Vd = (VReg *)vd; \
1525 VReg *Vj = (VReg *)vj; \
1526 \
1527 for (i = 0; i < LSX_LEN/BIT; i++) { \
1528 temp.E1(i) = do_ssrlns_ ## E1(Vj->E2(i), imm, BIT/2 -1); \
1529 temp.E1(i + LSX_LEN/BIT) = do_ssrlns_ ## E1(Vd->E2(i), imm, BIT/2 -1);\
1530 } \
1531 *Vd = temp; \
1532 }
1533
1534 void HELPER(vssrlni_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1535 {
1536 Int128 shft_res1, shft_res2, mask;
1537 VReg *Vd = (VReg *)vd;
1538 VReg *Vj = (VReg *)vj;
1539
1540 if (imm == 0) {
1541 shft_res1 = Vj->Q(0);
1542 shft_res2 = Vd->Q(0);
1543 } else {
1544 shft_res1 = int128_urshift(Vj->Q(0), imm);
1545 shft_res2 = int128_urshift(Vd->Q(0), imm);
1546 }
1547 mask = int128_sub(int128_lshift(int128_one(), 63), int128_one());
1548
1549 if (int128_ult(mask, shft_res1)) {
1550 Vd->D(0) = int128_getlo(mask);
1551 }else {
1552 Vd->D(0) = int128_getlo(shft_res1);
1553 }
1554
1555 if (int128_ult(mask, shft_res2)) {
1556 Vd->D(1) = int128_getlo(mask);
1557 }else {
1558 Vd->D(1) = int128_getlo(shft_res2);
1559 }
1560 }
1561
1562 VSSRLNI(vssrlni_b_h, 16, B, H)
1563 VSSRLNI(vssrlni_h_w, 32, H, W)
1564 VSSRLNI(vssrlni_w_d, 64, W, D)
1565
1566 #define VSSRANI(NAME, BIT, E1, E2) \
1567 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1568 { \
1569 int i; \
1570 VReg temp; \
1571 VReg *Vd = (VReg *)vd; \
1572 VReg *Vj = (VReg *)vj; \
1573 \
1574 for (i = 0; i < LSX_LEN/BIT; i++) { \
1575 temp.E1(i) = do_ssrans_ ## E1(Vj->E2(i), imm, BIT/2 -1); \
1576 temp.E1(i + LSX_LEN/BIT) = do_ssrans_ ## E1(Vd->E2(i), imm, BIT/2 -1); \
1577 } \
1578 *Vd = temp; \
1579 }
1580
1581 void HELPER(vssrani_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1582 {
1583 Int128 shft_res1, shft_res2, mask, min;
1584 VReg *Vd = (VReg *)vd;
1585 VReg *Vj = (VReg *)vj;
1586
1587 if (imm == 0) {
1588 shft_res1 = Vj->Q(0);
1589 shft_res2 = Vd->Q(0);
1590 } else {
1591 shft_res1 = int128_rshift(Vj->Q(0), imm);
1592 shft_res2 = int128_rshift(Vd->Q(0), imm);
1593 }
1594 mask = int128_sub(int128_lshift(int128_one(), 63), int128_one());
1595 min = int128_lshift(int128_one(), 63);
1596
1597 if (int128_gt(shft_res1, mask)) {
1598 Vd->D(0) = int128_getlo(mask);
1599 } else if (int128_lt(shft_res1, int128_neg(min))) {
1600 Vd->D(0) = int128_getlo(min);
1601 } else {
1602 Vd->D(0) = int128_getlo(shft_res1);
1603 }
1604
1605 if (int128_gt(shft_res2, mask)) {
1606 Vd->D(1) = int128_getlo(mask);
1607 } else if (int128_lt(shft_res2, int128_neg(min))) {
1608 Vd->D(1) = int128_getlo(min);
1609 } else {
1610 Vd->D(1) = int128_getlo(shft_res2);
1611 }
1612 }
1613
1614 VSSRANI(vssrani_b_h, 16, B, H)
1615 VSSRANI(vssrani_h_w, 32, H, W)
1616 VSSRANI(vssrani_w_d, 64, W, D)
1617
1618 #define VSSRLNUI(NAME, BIT, E1, E2) \
1619 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1620 { \
1621 int i; \
1622 VReg temp; \
1623 VReg *Vd = (VReg *)vd; \
1624 VReg *Vj = (VReg *)vj; \
1625 \
1626 for (i = 0; i < LSX_LEN/BIT; i++) { \
1627 temp.E1(i) = do_ssrlnu_ ## E1(Vj->E2(i), imm, BIT/2); \
1628 temp.E1(i + LSX_LEN/BIT) = do_ssrlnu_ ## E1(Vd->E2(i), imm, BIT/2); \
1629 } \
1630 *Vd = temp; \
1631 }
1632
1633 void HELPER(vssrlni_du_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1634 {
1635 Int128 shft_res1, shft_res2, mask;
1636 VReg *Vd = (VReg *)vd;
1637 VReg *Vj = (VReg *)vj;
1638
1639 if (imm == 0) {
1640 shft_res1 = Vj->Q(0);
1641 shft_res2 = Vd->Q(0);
1642 } else {
1643 shft_res1 = int128_urshift(Vj->Q(0), imm);
1644 shft_res2 = int128_urshift(Vd->Q(0), imm);
1645 }
1646 mask = int128_sub(int128_lshift(int128_one(), 64), int128_one());
1647
1648 if (int128_ult(mask, shft_res1)) {
1649 Vd->D(0) = int128_getlo(mask);
1650 }else {
1651 Vd->D(0) = int128_getlo(shft_res1);
1652 }
1653
1654 if (int128_ult(mask, shft_res2)) {
1655 Vd->D(1) = int128_getlo(mask);
1656 }else {
1657 Vd->D(1) = int128_getlo(shft_res2);
1658 }
1659 }
1660
1661 VSSRLNUI(vssrlni_bu_h, 16, B, H)
1662 VSSRLNUI(vssrlni_hu_w, 32, H, W)
1663 VSSRLNUI(vssrlni_wu_d, 64, W, D)
1664
1665 #define VSSRANUI(NAME, BIT, E1, E2) \
1666 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1667 { \
1668 int i; \
1669 VReg temp; \
1670 VReg *Vd = (VReg *)vd; \
1671 VReg *Vj = (VReg *)vj; \
1672 \
1673 for (i = 0; i < LSX_LEN/BIT; i++) { \
1674 temp.E1(i) = do_ssranu_ ## E1(Vj->E2(i), imm, BIT/2); \
1675 temp.E1(i + LSX_LEN/BIT) = do_ssranu_ ## E1(Vd->E2(i), imm, BIT/2); \
1676 } \
1677 *Vd = temp; \
1678 }
1679
1680 void HELPER(vssrani_du_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1681 {
1682 Int128 shft_res1, shft_res2, mask;
1683 VReg *Vd = (VReg *)vd;
1684 VReg *Vj = (VReg *)vj;
1685
1686 if (imm == 0) {
1687 shft_res1 = Vj->Q(0);
1688 shft_res2 = Vd->Q(0);
1689 } else {
1690 shft_res1 = int128_rshift(Vj->Q(0), imm);
1691 shft_res2 = int128_rshift(Vd->Q(0), imm);
1692 }
1693
1694 if (int128_lt(Vj->Q(0), int128_zero())) {
1695 shft_res1 = int128_zero();
1696 }
1697
1698 if (int128_lt(Vd->Q(0), int128_zero())) {
1699 shft_res2 = int128_zero();
1700 }
1701
1702 mask = int128_sub(int128_lshift(int128_one(), 64), int128_one());
1703
1704 if (int128_ult(mask, shft_res1)) {
1705 Vd->D(0) = int128_getlo(mask);
1706 }else {
1707 Vd->D(0) = int128_getlo(shft_res1);
1708 }
1709
1710 if (int128_ult(mask, shft_res2)) {
1711 Vd->D(1) = int128_getlo(mask);
1712 }else {
1713 Vd->D(1) = int128_getlo(shft_res2);
1714 }
1715 }
1716
1717 VSSRANUI(vssrani_bu_h, 16, B, H)
1718 VSSRANUI(vssrani_hu_w, 32, H, W)
1719 VSSRANUI(vssrani_wu_d, 64, W, D)
1720
1721 #define SSRLRNS(E1, E2, T1, T2, T3) \
1722 static T1 do_ssrlrns_ ## E1(T2 e2, int sa, int sh) \
1723 { \
1724 T1 shft_res; \
1725 \
1726 shft_res = do_vsrlr_ ## E2(e2, sa); \
1727 T1 mask; \
1728 mask = (1ull << sh) -1; \
1729 if (shft_res > mask) { \
1730 return mask; \
1731 } else { \
1732 return shft_res; \
1733 } \
1734 }
1735
1736 SSRLRNS(B, H, uint16_t, int16_t, uint8_t)
1737 SSRLRNS(H, W, uint32_t, int32_t, uint16_t)
1738 SSRLRNS(W, D, uint64_t, int64_t, uint32_t)
1739
1740 #define VSSRLRN(NAME, BIT, T, E1, E2) \
1741 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1742 { \
1743 int i; \
1744 VReg *Vd = (VReg *)vd; \
1745 VReg *Vj = (VReg *)vj; \
1746 VReg *Vk = (VReg *)vk; \
1747 \
1748 for (i = 0; i < LSX_LEN/BIT; i++) { \
1749 Vd->E1(i) = do_ssrlrns_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2 -1); \
1750 } \
1751 Vd->D(1) = 0; \
1752 }
1753
1754 VSSRLRN(vssrlrn_b_h, 16, uint16_t, B, H)
1755 VSSRLRN(vssrlrn_h_w, 32, uint32_t, H, W)
1756 VSSRLRN(vssrlrn_w_d, 64, uint64_t, W, D)
1757
1758 #define SSRARNS(E1, E2, T1, T2) \
1759 static T1 do_ssrarns_ ## E1(T1 e2, int sa, int sh) \
1760 { \
1761 T1 shft_res; \
1762 \
1763 shft_res = do_vsrar_ ## E2(e2, sa); \
1764 T2 mask; \
1765 mask = (1ll << sh) -1; \
1766 if (shft_res > mask) { \
1767 return mask; \
1768 } else if (shft_res < -(mask +1)) { \
1769 return ~mask; \
1770 } else { \
1771 return shft_res; \
1772 } \
1773 }
1774
1775 SSRARNS(B, H, int16_t, int8_t)
1776 SSRARNS(H, W, int32_t, int16_t)
1777 SSRARNS(W, D, int64_t, int32_t)
1778
1779 #define VSSRARN(NAME, BIT, T, E1, E2) \
1780 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1781 { \
1782 int i; \
1783 VReg *Vd = (VReg *)vd; \
1784 VReg *Vj = (VReg *)vj; \
1785 VReg *Vk = (VReg *)vk; \
1786 \
1787 for (i = 0; i < LSX_LEN/BIT; i++) { \
1788 Vd->E1(i) = do_ssrarns_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2 -1); \
1789 } \
1790 Vd->D(1) = 0; \
1791 }
1792
1793 VSSRARN(vssrarn_b_h, 16, uint16_t, B, H)
1794 VSSRARN(vssrarn_h_w, 32, uint32_t, H, W)
1795 VSSRARN(vssrarn_w_d, 64, uint64_t, W, D)
1796
1797 #define SSRLRNU(E1, E2, T1, T2, T3) \
1798 static T1 do_ssrlrnu_ ## E1(T3 e2, int sa, int sh) \
1799 { \
1800 T1 shft_res; \
1801 \
1802 shft_res = do_vsrlr_ ## E2(e2, sa); \
1803 \
1804 T2 mask; \
1805 mask = (1ull << sh) -1; \
1806 if (shft_res > mask) { \
1807 return mask; \
1808 } else { \
1809 return shft_res; \
1810 } \
1811 }
1812
1813 SSRLRNU(B, H, uint16_t, uint8_t, int16_t)
1814 SSRLRNU(H, W, uint32_t, uint16_t, int32_t)
1815 SSRLRNU(W, D, uint64_t, uint32_t, int64_t)
1816
1817 #define VSSRLRNU(NAME, BIT, T, E1, E2) \
1818 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1819 { \
1820 int i; \
1821 VReg *Vd = (VReg *)vd; \
1822 VReg *Vj = (VReg *)vj; \
1823 VReg *Vk = (VReg *)vk; \
1824 \
1825 for (i = 0; i < LSX_LEN/BIT; i++) { \
1826 Vd->E1(i) = do_ssrlrnu_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2); \
1827 } \
1828 Vd->D(1) = 0; \
1829 }
1830
1831 VSSRLRNU(vssrlrn_bu_h, 16, uint16_t, B, H)
1832 VSSRLRNU(vssrlrn_hu_w, 32, uint32_t, H, W)
1833 VSSRLRNU(vssrlrn_wu_d, 64, uint64_t, W, D)
1834
1835 #define SSRARNU(E1, E2, T1, T2, T3) \
1836 static T1 do_ssrarnu_ ## E1(T3 e2, int sa, int sh) \
1837 { \
1838 T1 shft_res; \
1839 \
1840 if (e2 < 0) { \
1841 shft_res = 0; \
1842 } else { \
1843 shft_res = do_vsrar_ ## E2(e2, sa); \
1844 } \
1845 T2 mask; \
1846 mask = (1ull << sh) -1; \
1847 if (shft_res > mask) { \
1848 return mask; \
1849 } else { \
1850 return shft_res; \
1851 } \
1852 }
1853
1854 SSRARNU(B, H, uint16_t, uint8_t, int16_t)
1855 SSRARNU(H, W, uint32_t, uint16_t, int32_t)
1856 SSRARNU(W, D, uint64_t, uint32_t, int64_t)
1857
1858 #define VSSRARNU(NAME, BIT, T, E1, E2) \
1859 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
1860 { \
1861 int i; \
1862 VReg *Vd = (VReg *)vd; \
1863 VReg *Vj = (VReg *)vj; \
1864 VReg *Vk = (VReg *)vk; \
1865 \
1866 for (i = 0; i < LSX_LEN/BIT; i++) { \
1867 Vd->E1(i) = do_ssrarnu_ ## E1(Vj->E2(i), (T)Vk->E2(i)%BIT, BIT/2); \
1868 } \
1869 Vd->D(1) = 0; \
1870 }
1871
1872 VSSRARNU(vssrarn_bu_h, 16, uint16_t, B, H)
1873 VSSRARNU(vssrarn_hu_w, 32, uint32_t, H, W)
1874 VSSRARNU(vssrarn_wu_d, 64, uint64_t, W, D)
1875
1876 #define VSSRLRNI(NAME, BIT, E1, E2) \
1877 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1878 { \
1879 int i; \
1880 VReg temp; \
1881 VReg *Vd = (VReg *)vd; \
1882 VReg *Vj = (VReg *)vj; \
1883 \
1884 for (i = 0; i < LSX_LEN/BIT; i++) { \
1885 temp.E1(i) = do_ssrlrns_ ## E1(Vj->E2(i), imm, BIT/2 -1); \
1886 temp.E1(i + LSX_LEN/BIT) = do_ssrlrns_ ## E1(Vd->E2(i), imm, BIT/2 -1);\
1887 } \
1888 *Vd = temp; \
1889 }
1890
1891 #define VSSRLRNI_Q(NAME, sh) \
1892 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1893 { \
1894 Int128 shft_res1, shft_res2, mask, r1, r2; \
1895 VReg *Vd = (VReg *)vd; \
1896 VReg *Vj = (VReg *)vj; \
1897 \
1898 if (imm == 0) { \
1899 shft_res1 = Vj->Q(0); \
1900 shft_res2 = Vd->Q(0); \
1901 } else { \
1902 r1 = int128_and(int128_urshift(Vj->Q(0), (imm -1)), int128_one()); \
1903 r2 = int128_and(int128_urshift(Vd->Q(0), (imm -1)), int128_one()); \
1904 \
1905 shft_res1 = (int128_add(int128_urshift(Vj->Q(0), imm), r1)); \
1906 shft_res2 = (int128_add(int128_urshift(Vd->Q(0), imm), r2)); \
1907 } \
1908 \
1909 mask = int128_sub(int128_lshift(int128_one(), sh), int128_one()); \
1910 \
1911 if (int128_ult(mask, shft_res1)) { \
1912 Vd->D(0) = int128_getlo(mask); \
1913 }else { \
1914 Vd->D(0) = int128_getlo(shft_res1); \
1915 } \
1916 \
1917 if (int128_ult(mask, shft_res2)) { \
1918 Vd->D(1) = int128_getlo(mask); \
1919 }else { \
1920 Vd->D(1) = int128_getlo(shft_res2); \
1921 } \
1922 }
1923
1924 VSSRLRNI(vssrlrni_b_h, 16, B, H)
1925 VSSRLRNI(vssrlrni_h_w, 32, H, W)
1926 VSSRLRNI(vssrlrni_w_d, 64, W, D)
1927 VSSRLRNI_Q(vssrlrni_d_q, 63)
1928
1929 #define VSSRARNI(NAME, BIT, E1, E2) \
1930 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1931 { \
1932 int i; \
1933 VReg temp; \
1934 VReg *Vd = (VReg *)vd; \
1935 VReg *Vj = (VReg *)vj; \
1936 \
1937 for (i = 0; i < LSX_LEN/BIT; i++) { \
1938 temp.E1(i) = do_ssrarns_ ## E1(Vj->E2(i), imm, BIT/2 -1); \
1939 temp.E1(i + LSX_LEN/BIT) = do_ssrarns_ ## E1(Vd->E2(i), imm, BIT/2 -1); \
1940 } \
1941 *Vd = temp; \
1942 }
1943
1944 void HELPER(vssrarni_d_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
1945 {
1946 Int128 shft_res1, shft_res2, mask1, mask2, r1, r2;
1947 VReg *Vd = (VReg *)vd;
1948 VReg *Vj = (VReg *)vj;
1949
1950 if (imm == 0) {
1951 shft_res1 = Vj->Q(0);
1952 shft_res2 = Vd->Q(0);
1953 } else {
1954 r1 = int128_and(int128_rshift(Vj->Q(0), (imm -1)), int128_one());
1955 r2 = int128_and(int128_rshift(Vd->Q(0), (imm -1)), int128_one());
1956
1957 shft_res1 = int128_add(int128_rshift(Vj->Q(0), imm), r1);
1958 shft_res2 = int128_add(int128_rshift(Vd->Q(0), imm), r2);
1959 }
1960
1961 mask1 = int128_sub(int128_lshift(int128_one(), 63), int128_one());
1962 mask2 = int128_lshift(int128_one(), 63);
1963
1964 if (int128_gt(shft_res1, mask1)) {
1965 Vd->D(0) = int128_getlo(mask1);
1966 } else if (int128_lt(shft_res1, int128_neg(mask2))) {
1967 Vd->D(0) = int128_getlo(mask2);
1968 } else {
1969 Vd->D(0) = int128_getlo(shft_res1);
1970 }
1971
1972 if (int128_gt(shft_res2, mask1)) {
1973 Vd->D(1) = int128_getlo(mask1);
1974 } else if (int128_lt(shft_res2, int128_neg(mask2))) {
1975 Vd->D(1) = int128_getlo(mask2);
1976 } else {
1977 Vd->D(1) = int128_getlo(shft_res2);
1978 }
1979 }
1980
1981 VSSRARNI(vssrarni_b_h, 16, B, H)
1982 VSSRARNI(vssrarni_h_w, 32, H, W)
1983 VSSRARNI(vssrarni_w_d, 64, W, D)
1984
1985 #define VSSRLRNUI(NAME, BIT, E1, E2) \
1986 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
1987 { \
1988 int i; \
1989 VReg temp; \
1990 VReg *Vd = (VReg *)vd; \
1991 VReg *Vj = (VReg *)vj; \
1992 \
1993 for (i = 0; i < LSX_LEN/BIT; i++) { \
1994 temp.E1(i) = do_ssrlrnu_ ## E1(Vj->E2(i), imm, BIT/2); \
1995 temp.E1(i + LSX_LEN/BIT) = do_ssrlrnu_ ## E1(Vd->E2(i), imm, BIT/2); \
1996 } \
1997 *Vd = temp; \
1998 }
1999
2000 VSSRLRNUI(vssrlrni_bu_h, 16, B, H)
2001 VSSRLRNUI(vssrlrni_hu_w, 32, H, W)
2002 VSSRLRNUI(vssrlrni_wu_d, 64, W, D)
2003 VSSRLRNI_Q(vssrlrni_du_q, 64)
2004
2005 #define VSSRARNUI(NAME, BIT, E1, E2) \
2006 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
2007 { \
2008 int i; \
2009 VReg temp; \
2010 VReg *Vd = (VReg *)vd; \
2011 VReg *Vj = (VReg *)vj; \
2012 \
2013 for (i = 0; i < LSX_LEN/BIT; i++) { \
2014 temp.E1(i) = do_ssrarnu_ ## E1(Vj->E2(i), imm, BIT/2); \
2015 temp.E1(i + LSX_LEN/BIT) = do_ssrarnu_ ## E1(Vd->E2(i), imm, BIT/2); \
2016 } \
2017 *Vd = temp; \
2018 }
2019
2020 void HELPER(vssrarni_du_q)(void *vd, void *vj, uint64_t imm, uint32_t desc)
2021 {
2022 Int128 shft_res1, shft_res2, mask1, mask2, r1, r2;
2023 VReg *Vd = (VReg *)vd;
2024 VReg *Vj = (VReg *)vj;
2025
2026 if (imm == 0) {
2027 shft_res1 = Vj->Q(0);
2028 shft_res2 = Vd->Q(0);
2029 } else {
2030 r1 = int128_and(int128_rshift(Vj->Q(0), (imm -1)), int128_one());
2031 r2 = int128_and(int128_rshift(Vd->Q(0), (imm -1)), int128_one());
2032
2033 shft_res1 = int128_add(int128_rshift(Vj->Q(0), imm), r1);
2034 shft_res2 = int128_add(int128_rshift(Vd->Q(0), imm), r2);
2035 }
2036
2037 if (int128_lt(Vj->Q(0), int128_zero())) {
2038 shft_res1 = int128_zero();
2039 }
2040 if (int128_lt(Vd->Q(0), int128_zero())) {
2041 shft_res2 = int128_zero();
2042 }
2043
2044 mask1 = int128_sub(int128_lshift(int128_one(), 64), int128_one());
2045 mask2 = int128_lshift(int128_one(), 64);
2046
2047 if (int128_gt(shft_res1, mask1)) {
2048 Vd->D(0) = int128_getlo(mask1);
2049 } else if (int128_lt(shft_res1, int128_neg(mask2))) {
2050 Vd->D(0) = int128_getlo(mask2);
2051 } else {
2052 Vd->D(0) = int128_getlo(shft_res1);
2053 }
2054
2055 if (int128_gt(shft_res2, mask1)) {
2056 Vd->D(1) = int128_getlo(mask1);
2057 } else if (int128_lt(shft_res2, int128_neg(mask2))) {
2058 Vd->D(1) = int128_getlo(mask2);
2059 } else {
2060 Vd->D(1) = int128_getlo(shft_res2);
2061 }
2062 }
2063
2064 VSSRARNUI(vssrarni_bu_h, 16, B, H)
2065 VSSRARNUI(vssrarni_hu_w, 32, H, W)
2066 VSSRARNUI(vssrarni_wu_d, 64, W, D)
2067
2068 #define DO_2OP(NAME, BIT, E, DO_OP) \
2069 void HELPER(NAME)(void *vd, void *vj, uint32_t desc) \
2070 { \
2071 int i; \
2072 VReg *Vd = (VReg *)vd; \
2073 VReg *Vj = (VReg *)vj; \
2074 \
2075 for (i = 0; i < LSX_LEN/BIT; i++) \
2076 { \
2077 Vd->E(i) = DO_OP(Vj->E(i)); \
2078 } \
2079 }
2080
2081 #define DO_CLO_B(N) (clz32(~N & 0xff) - 24)
2082 #define DO_CLO_H(N) (clz32(~N & 0xffff) - 16)
2083 #define DO_CLO_W(N) (clz32(~N))
2084 #define DO_CLO_D(N) (clz64(~N))
2085 #define DO_CLZ_B(N) (clz32(N) - 24)
2086 #define DO_CLZ_H(N) (clz32(N) - 16)
2087 #define DO_CLZ_W(N) (clz32(N))
2088 #define DO_CLZ_D(N) (clz64(N))
2089
2090 DO_2OP(vclo_b, 8, UB, DO_CLO_B)
2091 DO_2OP(vclo_h, 16, UH, DO_CLO_H)
2092 DO_2OP(vclo_w, 32, UW, DO_CLO_W)
2093 DO_2OP(vclo_d, 64, UD, DO_CLO_D)
2094 DO_2OP(vclz_b, 8, UB, DO_CLZ_B)
2095 DO_2OP(vclz_h, 16, UH, DO_CLZ_H)
2096 DO_2OP(vclz_w, 32, UW, DO_CLZ_W)
2097 DO_2OP(vclz_d, 64, UD, DO_CLZ_D)
2098
2099 #define VPCNT(NAME, BIT, E, FN) \
2100 void HELPER(NAME)(void *vd, void *vj, uint32_t desc) \
2101 { \
2102 int i; \
2103 VReg *Vd = (VReg *)vd; \
2104 VReg *Vj = (VReg *)vj; \
2105 \
2106 for (i = 0; i < LSX_LEN/BIT; i++) \
2107 { \
2108 Vd->E(i) = FN(Vj->E(i)); \
2109 } \
2110 }
2111
2112 VPCNT(vpcnt_b, 8, UB, ctpop8)
2113 VPCNT(vpcnt_h, 16, UH, ctpop16)
2114 VPCNT(vpcnt_w, 32, UW, ctpop32)
2115 VPCNT(vpcnt_d, 64, UD, ctpop64)
2116
2117 #define DO_BITCLR(a, bit) (a & ~(1ull << bit))
2118 #define DO_BITSET(a, bit) (a | 1ull << bit)
2119 #define DO_BITREV(a, bit) (a ^ (1ull << bit))
2120
2121 #define DO_BIT(NAME, BIT, E, DO_OP) \
2122 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t v) \
2123 { \
2124 int i; \
2125 VReg *Vd = (VReg *)vd; \
2126 VReg *Vj = (VReg *)vj; \
2127 VReg *Vk = (VReg *)vk; \
2128 \
2129 for (i = 0; i < LSX_LEN/BIT; i++) { \
2130 Vd->E(i) = DO_OP(Vj->E(i), Vk->E(i)%BIT); \
2131 } \
2132 }
2133
2134 DO_BIT(vbitclr_b, 8, UB, DO_BITCLR)
2135 DO_BIT(vbitclr_h, 16, UH, DO_BITCLR)
2136 DO_BIT(vbitclr_w, 32, UW, DO_BITCLR)
2137 DO_BIT(vbitclr_d, 64, UD, DO_BITCLR)
2138 DO_BIT(vbitset_b, 8, UB, DO_BITSET)
2139 DO_BIT(vbitset_h, 16, UH, DO_BITSET)
2140 DO_BIT(vbitset_w, 32, UW, DO_BITSET)
2141 DO_BIT(vbitset_d, 64, UD, DO_BITSET)
2142 DO_BIT(vbitrev_b, 8, UB, DO_BITREV)
2143 DO_BIT(vbitrev_h, 16, UH, DO_BITREV)
2144 DO_BIT(vbitrev_w, 32, UW, DO_BITREV)
2145 DO_BIT(vbitrev_d, 64, UD, DO_BITREV)
2146
2147 #define DO_BITI(NAME, BIT, E, DO_OP) \
2148 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t v) \
2149 { \
2150 int i; \
2151 VReg *Vd = (VReg *)vd; \
2152 VReg *Vj = (VReg *)vj; \
2153 \
2154 for (i = 0; i < LSX_LEN/BIT; i++) { \
2155 Vd->E(i) = DO_OP(Vj->E(i), imm); \
2156 } \
2157 }
2158
2159 DO_BITI(vbitclri_b, 8, UB, DO_BITCLR)
2160 DO_BITI(vbitclri_h, 16, UH, DO_BITCLR)
2161 DO_BITI(vbitclri_w, 32, UW, DO_BITCLR)
2162 DO_BITI(vbitclri_d, 64, UD, DO_BITCLR)
2163 DO_BITI(vbitseti_b, 8, UB, DO_BITSET)
2164 DO_BITI(vbitseti_h, 16, UH, DO_BITSET)
2165 DO_BITI(vbitseti_w, 32, UW, DO_BITSET)
2166 DO_BITI(vbitseti_d, 64, UD, DO_BITSET)
2167 DO_BITI(vbitrevi_b, 8, UB, DO_BITREV)
2168 DO_BITI(vbitrevi_h, 16, UH, DO_BITREV)
2169 DO_BITI(vbitrevi_w, 32, UW, DO_BITREV)
2170 DO_BITI(vbitrevi_d, 64, UD, DO_BITREV)
2171
2172 #define VFRSTP(NAME, BIT, MASK, E) \
2173 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
2174 { \
2175 int i, m; \
2176 VReg *Vd = (VReg *)vd; \
2177 VReg *Vj = (VReg *)vj; \
2178 VReg *Vk = (VReg *)vk; \
2179 \
2180 for (i = 0; i < LSX_LEN/BIT; i++) { \
2181 if (Vj->E(i) < 0) { \
2182 break; \
2183 } \
2184 } \
2185 m = Vk->E(0) & MASK; \
2186 Vd->E(m) = i; \
2187 }
2188
2189 VFRSTP(vfrstp_b, 8, 0xf, B)
2190 VFRSTP(vfrstp_h, 16, 0x7, H)
2191
2192 #define VFRSTPI(NAME, BIT, E) \
2193 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
2194 { \
2195 int i, m; \
2196 VReg *Vd = (VReg *)vd; \
2197 VReg *Vj = (VReg *)vj; \
2198 \
2199 for (i = 0; i < LSX_LEN/BIT; i++) { \
2200 if (Vj->E(i) < 0) { \
2201 break; \
2202 } \
2203 } \
2204 m = imm % (LSX_LEN/BIT); \
2205 Vd->E(m) = i; \
2206 }
2207
2208 VFRSTPI(vfrstpi_b, 8, B)
2209 VFRSTPI(vfrstpi_h, 16, H)
2210
2211 static void vec_update_fcsr0_mask(CPULoongArchState *env,
2212 uintptr_t pc, int mask)
2213 {
2214 int flags = get_float_exception_flags(&env->fp_status);
2215
2216 set_float_exception_flags(0, &env->fp_status);
2217
2218 flags &= ~mask;
2219
2220 if (flags) {
2221 flags = ieee_ex_to_loongarch(flags);
2222 UPDATE_FP_CAUSE(env->fcsr0, flags);
2223 }
2224
2225 if (GET_FP_ENABLES(env->fcsr0) & flags) {
2226 do_raise_exception(env, EXCCODE_FPE, pc);
2227 } else {
2228 UPDATE_FP_FLAGS(env->fcsr0, flags);
2229 }
2230 }
2231
2232 static void vec_update_fcsr0(CPULoongArchState *env, uintptr_t pc)
2233 {
2234 vec_update_fcsr0_mask(env, pc, 0);
2235 }
2236
2237 static inline void vec_clear_cause(CPULoongArchState *env)
2238 {
2239 SET_FP_CAUSE(env->fcsr0, 0);
2240 }
2241
2242 #define DO_3OP_F(NAME, BIT, E, FN) \
2243 void HELPER(NAME)(void *vd, void *vj, void *vk, \
2244 CPULoongArchState *env, uint32_t desc) \
2245 { \
2246 int i; \
2247 VReg *Vd = (VReg *)vd; \
2248 VReg *Vj = (VReg *)vj; \
2249 VReg *Vk = (VReg *)vk; \
2250 \
2251 vec_clear_cause(env); \
2252 for (i = 0; i < LSX_LEN/BIT; i++) { \
2253 Vd->E(i) = FN(Vj->E(i), Vk->E(i), &env->fp_status); \
2254 vec_update_fcsr0(env, GETPC()); \
2255 } \
2256 }
2257
2258 DO_3OP_F(vfadd_s, 32, UW, float32_add)
2259 DO_3OP_F(vfadd_d, 64, UD, float64_add)
2260 DO_3OP_F(vfsub_s, 32, UW, float32_sub)
2261 DO_3OP_F(vfsub_d, 64, UD, float64_sub)
2262 DO_3OP_F(vfmul_s, 32, UW, float32_mul)
2263 DO_3OP_F(vfmul_d, 64, UD, float64_mul)
2264 DO_3OP_F(vfdiv_s, 32, UW, float32_div)
2265 DO_3OP_F(vfdiv_d, 64, UD, float64_div)
2266 DO_3OP_F(vfmax_s, 32, UW, float32_maxnum)
2267 DO_3OP_F(vfmax_d, 64, UD, float64_maxnum)
2268 DO_3OP_F(vfmin_s, 32, UW, float32_minnum)
2269 DO_3OP_F(vfmin_d, 64, UD, float64_minnum)
2270 DO_3OP_F(vfmaxa_s, 32, UW, float32_maxnummag)
2271 DO_3OP_F(vfmaxa_d, 64, UD, float64_maxnummag)
2272 DO_3OP_F(vfmina_s, 32, UW, float32_minnummag)
2273 DO_3OP_F(vfmina_d, 64, UD, float64_minnummag)
2274
2275 #define DO_4OP_F(NAME, BIT, E, FN, flags) \
2276 void HELPER(NAME)(void *vd, void *vj, void *vk, void *va, \
2277 CPULoongArchState *env, uint32_t desc) \
2278 { \
2279 int i; \
2280 VReg *Vd = (VReg *)vd; \
2281 VReg *Vj = (VReg *)vj; \
2282 VReg *Vk = (VReg *)vk; \
2283 VReg *Va = (VReg *)va; \
2284 \
2285 vec_clear_cause(env); \
2286 for (i = 0; i < LSX_LEN/BIT; i++) { \
2287 Vd->E(i) = FN(Vj->E(i), Vk->E(i), Va->E(i), flags, &env->fp_status); \
2288 vec_update_fcsr0(env, GETPC()); \
2289 } \
2290 }
2291
2292 DO_4OP_F(vfmadd_s, 32, UW, float32_muladd, 0)
2293 DO_4OP_F(vfmadd_d, 64, UD, float64_muladd, 0)
2294 DO_4OP_F(vfmsub_s, 32, UW, float32_muladd, float_muladd_negate_c)
2295 DO_4OP_F(vfmsub_d, 64, UD, float64_muladd, float_muladd_negate_c)
2296 DO_4OP_F(vfnmadd_s, 32, UW, float32_muladd, float_muladd_negate_result)
2297 DO_4OP_F(vfnmadd_d, 64, UD, float64_muladd, float_muladd_negate_result)
2298 DO_4OP_F(vfnmsub_s, 32, UW, float32_muladd,
2299 float_muladd_negate_c | float_muladd_negate_result)
2300 DO_4OP_F(vfnmsub_d, 64, UD, float64_muladd,
2301 float_muladd_negate_c | float_muladd_negate_result)
2302
2303 #define DO_2OP_F(NAME, BIT, E, FN) \
2304 void HELPER(NAME)(void *vd, void *vj, \
2305 CPULoongArchState *env, uint32_t desc) \
2306 { \
2307 int i; \
2308 VReg *Vd = (VReg *)vd; \
2309 VReg *Vj = (VReg *)vj; \
2310 \
2311 vec_clear_cause(env); \
2312 for (i = 0; i < LSX_LEN/BIT; i++) { \
2313 Vd->E(i) = FN(env, Vj->E(i)); \
2314 } \
2315 }
2316
2317 #define FLOGB(BIT, T) \
2318 static T do_flogb_## BIT(CPULoongArchState *env, T fj) \
2319 { \
2320 T fp, fd; \
2321 float_status *status = &env->fp_status; \
2322 FloatRoundMode old_mode = get_float_rounding_mode(status); \
2323 \
2324 set_float_rounding_mode(float_round_down, status); \
2325 fp = float ## BIT ##_log2(fj, status); \
2326 fd = float ## BIT ##_round_to_int(fp, status); \
2327 set_float_rounding_mode(old_mode, status); \
2328 vec_update_fcsr0_mask(env, GETPC(), float_flag_inexact); \
2329 return fd; \
2330 }
2331
2332 FLOGB(32, uint32_t)
2333 FLOGB(64, uint64_t)
2334
2335 #define FCLASS(NAME, BIT, E, FN) \
2336 void HELPER(NAME)(void *vd, void *vj, \
2337 CPULoongArchState *env, uint32_t desc) \
2338 { \
2339 int i; \
2340 VReg *Vd = (VReg *)vd; \
2341 VReg *Vj = (VReg *)vj; \
2342 \
2343 for (i = 0; i < LSX_LEN/BIT; i++) { \
2344 Vd->E(i) = FN(env, Vj->E(i)); \
2345 } \
2346 }
2347
2348 FCLASS(vfclass_s, 32, UW, helper_fclass_s)
2349 FCLASS(vfclass_d, 64, UD, helper_fclass_d)
2350
2351 #define FSQRT(BIT, T) \
2352 static T do_fsqrt_## BIT(CPULoongArchState *env, T fj) \
2353 { \
2354 T fd; \
2355 fd = float ## BIT ##_sqrt(fj, &env->fp_status); \
2356 vec_update_fcsr0(env, GETPC()); \
2357 return fd; \
2358 }
2359
2360 FSQRT(32, uint32_t)
2361 FSQRT(64, uint64_t)
2362
2363 #define FRECIP(BIT, T) \
2364 static T do_frecip_## BIT(CPULoongArchState *env, T fj) \
2365 { \
2366 T fd; \
2367 fd = float ## BIT ##_div(float ## BIT ##_one, fj, &env->fp_status); \
2368 vec_update_fcsr0(env, GETPC()); \
2369 return fd; \
2370 }
2371
2372 FRECIP(32, uint32_t)
2373 FRECIP(64, uint64_t)
2374
2375 #define FRSQRT(BIT, T) \
2376 static T do_frsqrt_## BIT(CPULoongArchState *env, T fj) \
2377 { \
2378 T fd, fp; \
2379 fp = float ## BIT ##_sqrt(fj, &env->fp_status); \
2380 fd = float ## BIT ##_div(float ## BIT ##_one, fp, &env->fp_status); \
2381 vec_update_fcsr0(env, GETPC()); \
2382 return fd; \
2383 }
2384
2385 FRSQRT(32, uint32_t)
2386 FRSQRT(64, uint64_t)
2387
2388 DO_2OP_F(vflogb_s, 32, UW, do_flogb_32)
2389 DO_2OP_F(vflogb_d, 64, UD, do_flogb_64)
2390 DO_2OP_F(vfsqrt_s, 32, UW, do_fsqrt_32)
2391 DO_2OP_F(vfsqrt_d, 64, UD, do_fsqrt_64)
2392 DO_2OP_F(vfrecip_s, 32, UW, do_frecip_32)
2393 DO_2OP_F(vfrecip_d, 64, UD, do_frecip_64)
2394 DO_2OP_F(vfrsqrt_s, 32, UW, do_frsqrt_32)
2395 DO_2OP_F(vfrsqrt_d, 64, UD, do_frsqrt_64)
2396
2397 static uint32_t float16_cvt_float32(uint16_t h, float_status *status)
2398 {
2399 return float16_to_float32(h, true, status);
2400 }
2401 static uint64_t float32_cvt_float64(uint32_t s, float_status *status)
2402 {
2403 return float32_to_float64(s, status);
2404 }
2405
2406 static uint16_t float32_cvt_float16(uint32_t s, float_status *status)
2407 {
2408 return float32_to_float16(s, true, status);
2409 }
2410 static uint32_t float64_cvt_float32(uint64_t d, float_status *status)
2411 {
2412 return float64_to_float32(d, status);
2413 }
2414
2415 void HELPER(vfcvtl_s_h)(void *vd, void *vj,
2416 CPULoongArchState *env, uint32_t desc)
2417 {
2418 int i;
2419 VReg temp;
2420 VReg *Vd = (VReg *)vd;
2421 VReg *Vj = (VReg *)vj;
2422
2423 vec_clear_cause(env);
2424 for (i = 0; i < LSX_LEN/32; i++) {
2425 temp.UW(i) = float16_cvt_float32(Vj->UH(i), &env->fp_status);
2426 vec_update_fcsr0(env, GETPC());
2427 }
2428 *Vd = temp;
2429 }
2430
2431 void HELPER(vfcvtl_d_s)(void *vd, void *vj,
2432 CPULoongArchState *env, uint32_t desc)
2433 {
2434 int i;
2435 VReg temp;
2436 VReg *Vd = (VReg *)vd;
2437 VReg *Vj = (VReg *)vj;
2438
2439 vec_clear_cause(env);
2440 for (i = 0; i < LSX_LEN/64; i++) {
2441 temp.UD(i) = float32_cvt_float64(Vj->UW(i), &env->fp_status);
2442 vec_update_fcsr0(env, GETPC());
2443 }
2444 *Vd = temp;
2445 }
2446
2447 void HELPER(vfcvth_s_h)(void *vd, void *vj,
2448 CPULoongArchState *env, uint32_t desc)
2449 {
2450 int i;
2451 VReg temp;
2452 VReg *Vd = (VReg *)vd;
2453 VReg *Vj = (VReg *)vj;
2454
2455 vec_clear_cause(env);
2456 for (i = 0; i < LSX_LEN/32; i++) {
2457 temp.UW(i) = float16_cvt_float32(Vj->UH(i + 4), &env->fp_status);
2458 vec_update_fcsr0(env, GETPC());
2459 }
2460 *Vd = temp;
2461 }
2462
2463 void HELPER(vfcvth_d_s)(void *vd, void *vj,
2464 CPULoongArchState *env, uint32_t desc)
2465 {
2466 int i;
2467 VReg temp;
2468 VReg *Vd = (VReg *)vd;
2469 VReg *Vj = (VReg *)vj;
2470
2471 vec_clear_cause(env);
2472 for (i = 0; i < LSX_LEN/64; i++) {
2473 temp.UD(i) = float32_cvt_float64(Vj->UW(i + 2), &env->fp_status);
2474 vec_update_fcsr0(env, GETPC());
2475 }
2476 *Vd = temp;
2477 }
2478
2479 void HELPER(vfcvt_h_s)(void *vd, void *vj, void *vk,
2480 CPULoongArchState *env, uint32_t desc)
2481 {
2482 int i;
2483 VReg temp;
2484 VReg *Vd = (VReg *)vd;
2485 VReg *Vj = (VReg *)vj;
2486 VReg *Vk = (VReg *)vk;
2487
2488 vec_clear_cause(env);
2489 for(i = 0; i < LSX_LEN/32; i++) {
2490 temp.UH(i + 4) = float32_cvt_float16(Vj->UW(i), &env->fp_status);
2491 temp.UH(i) = float32_cvt_float16(Vk->UW(i), &env->fp_status);
2492 vec_update_fcsr0(env, GETPC());
2493 }
2494 *Vd = temp;
2495 }
2496
2497 void HELPER(vfcvt_s_d)(void *vd, void *vj, void *vk,
2498 CPULoongArchState *env, uint32_t desc)
2499 {
2500 int i;
2501 VReg temp;
2502 VReg *Vd = (VReg *)vd;
2503 VReg *Vj = (VReg *)vj;
2504 VReg *Vk = (VReg *)vk;
2505
2506 vec_clear_cause(env);
2507 for(i = 0; i < LSX_LEN/64; i++) {
2508 temp.UW(i + 2) = float64_cvt_float32(Vj->UD(i), &env->fp_status);
2509 temp.UW(i) = float64_cvt_float32(Vk->UD(i), &env->fp_status);
2510 vec_update_fcsr0(env, GETPC());
2511 }
2512 *Vd = temp;
2513 }
2514
2515 void HELPER(vfrint_s)(void *vd, void *vj,
2516 CPULoongArchState *env, uint32_t desc)
2517 {
2518 int i;
2519 VReg *Vd = (VReg *)vd;
2520 VReg *Vj = (VReg *)vj;
2521
2522 vec_clear_cause(env);
2523 for (i = 0; i < 4; i++) {
2524 Vd->W(i) = float32_round_to_int(Vj->UW(i), &env->fp_status);
2525 vec_update_fcsr0(env, GETPC());
2526 }
2527 }
2528
2529 void HELPER(vfrint_d)(void *vd, void *vj,
2530 CPULoongArchState *env, uint32_t desc)
2531 {
2532 int i;
2533 VReg *Vd = (VReg *)vd;
2534 VReg *Vj = (VReg *)vj;
2535
2536 vec_clear_cause(env);
2537 for (i = 0; i < 2; i++) {
2538 Vd->D(i) = float64_round_to_int(Vj->UD(i), &env->fp_status);
2539 vec_update_fcsr0(env, GETPC());
2540 }
2541 }
2542
2543 #define FCVT_2OP(NAME, BIT, E, MODE) \
2544 void HELPER(NAME)(void *vd, void *vj, \
2545 CPULoongArchState *env, uint32_t desc) \
2546 { \
2547 int i; \
2548 VReg *Vd = (VReg *)vd; \
2549 VReg *Vj = (VReg *)vj; \
2550 \
2551 vec_clear_cause(env); \
2552 for (i = 0; i < LSX_LEN/BIT; i++) { \
2553 FloatRoundMode old_mode = get_float_rounding_mode(&env->fp_status); \
2554 set_float_rounding_mode(MODE, &env->fp_status); \
2555 Vd->E(i) = float## BIT ## _round_to_int(Vj->E(i), &env->fp_status); \
2556 set_float_rounding_mode(old_mode, &env->fp_status); \
2557 vec_update_fcsr0(env, GETPC()); \
2558 } \
2559 }
2560
2561 FCVT_2OP(vfrintrne_s, 32, UW, float_round_nearest_even)
2562 FCVT_2OP(vfrintrne_d, 64, UD, float_round_nearest_even)
2563 FCVT_2OP(vfrintrz_s, 32, UW, float_round_to_zero)
2564 FCVT_2OP(vfrintrz_d, 64, UD, float_round_to_zero)
2565 FCVT_2OP(vfrintrp_s, 32, UW, float_round_up)
2566 FCVT_2OP(vfrintrp_d, 64, UD, float_round_up)
2567 FCVT_2OP(vfrintrm_s, 32, UW, float_round_down)
2568 FCVT_2OP(vfrintrm_d, 64, UD, float_round_down)
2569
2570 #define FTINT(NAME, FMT1, FMT2, T1, T2, MODE) \
2571 static T2 do_ftint ## NAME(CPULoongArchState *env, T1 fj) \
2572 { \
2573 T2 fd; \
2574 FloatRoundMode old_mode = get_float_rounding_mode(&env->fp_status); \
2575 \
2576 set_float_rounding_mode(MODE, &env->fp_status); \
2577 fd = do_## FMT1 ##_to_## FMT2(env, fj); \
2578 set_float_rounding_mode(old_mode, &env->fp_status); \
2579 return fd; \
2580 }
2581
2582 #define DO_FTINT(FMT1, FMT2, T1, T2) \
2583 static T2 do_## FMT1 ##_to_## FMT2(CPULoongArchState *env, T1 fj) \
2584 { \
2585 T2 fd; \
2586 \
2587 fd = FMT1 ##_to_## FMT2(fj, &env->fp_status); \
2588 if (get_float_exception_flags(&env->fp_status) & (float_flag_invalid)) { \
2589 if (FMT1 ##_is_any_nan(fj)) { \
2590 fd = 0; \
2591 } \
2592 } \
2593 vec_update_fcsr0(env, GETPC()); \
2594 return fd; \
2595 }
2596
2597 DO_FTINT(float32, int32, uint32_t, uint32_t)
2598 DO_FTINT(float64, int64, uint64_t, uint64_t)
2599 DO_FTINT(float32, uint32, uint32_t, uint32_t)
2600 DO_FTINT(float64, uint64, uint64_t, uint64_t)
2601 DO_FTINT(float64, int32, uint64_t, uint32_t)
2602 DO_FTINT(float32, int64, uint32_t, uint64_t)
2603
2604 FTINT(rne_w_s, float32, int32, uint32_t, uint32_t, float_round_nearest_even)
2605 FTINT(rne_l_d, float64, int64, uint64_t, uint64_t, float_round_nearest_even)
2606 FTINT(rp_w_s, float32, int32, uint32_t, uint32_t, float_round_up)
2607 FTINT(rp_l_d, float64, int64, uint64_t, uint64_t, float_round_up)
2608 FTINT(rz_w_s, float32, int32, uint32_t, uint32_t, float_round_to_zero)
2609 FTINT(rz_l_d, float64, int64, uint64_t, uint64_t, float_round_to_zero)
2610 FTINT(rm_w_s, float32, int32, uint32_t, uint32_t, float_round_down)
2611 FTINT(rm_l_d, float64, int64, uint64_t, uint64_t, float_round_down)
2612
2613 DO_2OP_F(vftintrne_w_s, 32, UW, do_ftintrne_w_s)
2614 DO_2OP_F(vftintrne_l_d, 64, UD, do_ftintrne_l_d)
2615 DO_2OP_F(vftintrp_w_s, 32, UW, do_ftintrp_w_s)
2616 DO_2OP_F(vftintrp_l_d, 64, UD, do_ftintrp_l_d)
2617 DO_2OP_F(vftintrz_w_s, 32, UW, do_ftintrz_w_s)
2618 DO_2OP_F(vftintrz_l_d, 64, UD, do_ftintrz_l_d)
2619 DO_2OP_F(vftintrm_w_s, 32, UW, do_ftintrm_w_s)
2620 DO_2OP_F(vftintrm_l_d, 64, UD, do_ftintrm_l_d)
2621 DO_2OP_F(vftint_w_s, 32, UW, do_float32_to_int32)
2622 DO_2OP_F(vftint_l_d, 64, UD, do_float64_to_int64)
2623
2624 FTINT(rz_wu_s, float32, uint32, uint32_t, uint32_t, float_round_to_zero)
2625 FTINT(rz_lu_d, float64, uint64, uint64_t, uint64_t, float_round_to_zero)
2626
2627 DO_2OP_F(vftintrz_wu_s, 32, UW, do_ftintrz_wu_s)
2628 DO_2OP_F(vftintrz_lu_d, 64, UD, do_ftintrz_lu_d)
2629 DO_2OP_F(vftint_wu_s, 32, UW, do_float32_to_uint32)
2630 DO_2OP_F(vftint_lu_d, 64, UD, do_float64_to_uint64)
2631
2632 FTINT(rm_w_d, float64, int32, uint64_t, uint32_t, float_round_down)
2633 FTINT(rp_w_d, float64, int32, uint64_t, uint32_t, float_round_up)
2634 FTINT(rz_w_d, float64, int32, uint64_t, uint32_t, float_round_to_zero)
2635 FTINT(rne_w_d, float64, int32, uint64_t, uint32_t, float_round_nearest_even)
2636
2637 #define FTINT_W_D(NAME, FN) \
2638 void HELPER(NAME)(void *vd, void *vj, void *vk, \
2639 CPULoongArchState *env, uint32_t desc) \
2640 { \
2641 int i; \
2642 VReg temp; \
2643 VReg *Vd = (VReg *)vd; \
2644 VReg *Vj = (VReg *)vj; \
2645 VReg *Vk = (VReg *)vk; \
2646 \
2647 vec_clear_cause(env); \
2648 for (i = 0; i < 2; i++) { \
2649 temp.W(i + 2) = FN(env, Vj->UD(i)); \
2650 temp.W(i) = FN(env, Vk->UD(i)); \
2651 } \
2652 *Vd = temp; \
2653 }
2654
2655 FTINT_W_D(vftint_w_d, do_float64_to_int32)
2656 FTINT_W_D(vftintrm_w_d, do_ftintrm_w_d)
2657 FTINT_W_D(vftintrp_w_d, do_ftintrp_w_d)
2658 FTINT_W_D(vftintrz_w_d, do_ftintrz_w_d)
2659 FTINT_W_D(vftintrne_w_d, do_ftintrne_w_d)
2660
2661 FTINT(rml_l_s, float32, int64, uint32_t, uint64_t, float_round_down)
2662 FTINT(rpl_l_s, float32, int64, uint32_t, uint64_t, float_round_up)
2663 FTINT(rzl_l_s, float32, int64, uint32_t, uint64_t, float_round_to_zero)
2664 FTINT(rnel_l_s, float32, int64, uint32_t, uint64_t, float_round_nearest_even)
2665 FTINT(rmh_l_s, float32, int64, uint32_t, uint64_t, float_round_down)
2666 FTINT(rph_l_s, float32, int64, uint32_t, uint64_t, float_round_up)
2667 FTINT(rzh_l_s, float32, int64, uint32_t, uint64_t, float_round_to_zero)
2668 FTINT(rneh_l_s, float32, int64, uint32_t, uint64_t, float_round_nearest_even)
2669
2670 #define FTINTL_L_S(NAME, FN) \
2671 void HELPER(NAME)(void *vd, void *vj, \
2672 CPULoongArchState *env, uint32_t desc) \
2673 { \
2674 int i; \
2675 VReg temp; \
2676 VReg *Vd = (VReg *)vd; \
2677 VReg *Vj = (VReg *)vj; \
2678 \
2679 vec_clear_cause(env); \
2680 for (i = 0; i < 2; i++) { \
2681 temp.D(i) = FN(env, Vj->UW(i)); \
2682 } \
2683 *Vd = temp; \
2684 }
2685
2686 FTINTL_L_S(vftintl_l_s, do_float32_to_int64)
2687 FTINTL_L_S(vftintrml_l_s, do_ftintrml_l_s)
2688 FTINTL_L_S(vftintrpl_l_s, do_ftintrpl_l_s)
2689 FTINTL_L_S(vftintrzl_l_s, do_ftintrzl_l_s)
2690 FTINTL_L_S(vftintrnel_l_s, do_ftintrnel_l_s)
2691
2692 #define FTINTH_L_S(NAME, FN) \
2693 void HELPER(NAME)(void *vd, void *vj, \
2694 CPULoongArchState *env, uint32_t desc) \
2695 { \
2696 int i; \
2697 VReg temp; \
2698 VReg *Vd = (VReg *)vd; \
2699 VReg *Vj = (VReg *)vj; \
2700 \
2701 vec_clear_cause(env); \
2702 for (i = 0; i < 2; i++) { \
2703 temp.D(i) = FN(env, Vj->UW(i + 2)); \
2704 } \
2705 *Vd = temp; \
2706 }
2707
2708 FTINTH_L_S(vftinth_l_s, do_float32_to_int64)
2709 FTINTH_L_S(vftintrmh_l_s, do_ftintrmh_l_s)
2710 FTINTH_L_S(vftintrph_l_s, do_ftintrph_l_s)
2711 FTINTH_L_S(vftintrzh_l_s, do_ftintrzh_l_s)
2712 FTINTH_L_S(vftintrneh_l_s, do_ftintrneh_l_s)
2713
2714 #define FFINT(NAME, FMT1, FMT2, T1, T2) \
2715 static T2 do_ffint_ ## NAME(CPULoongArchState *env, T1 fj) \
2716 { \
2717 T2 fd; \
2718 \
2719 fd = FMT1 ##_to_## FMT2(fj, &env->fp_status); \
2720 vec_update_fcsr0(env, GETPC()); \
2721 return fd; \
2722 }
2723
2724 FFINT(s_w, int32, float32, int32_t, uint32_t)
2725 FFINT(d_l, int64, float64, int64_t, uint64_t)
2726 FFINT(s_wu, uint32, float32, uint32_t, uint32_t)
2727 FFINT(d_lu, uint64, float64, uint64_t, uint64_t)
2728
2729 DO_2OP_F(vffint_s_w, 32, W, do_ffint_s_w)
2730 DO_2OP_F(vffint_d_l, 64, D, do_ffint_d_l)
2731 DO_2OP_F(vffint_s_wu, 32, UW, do_ffint_s_wu)
2732 DO_2OP_F(vffint_d_lu, 64, UD, do_ffint_d_lu)
2733
2734 void HELPER(vffintl_d_w)(void *vd, void *vj,
2735 CPULoongArchState *env, uint32_t desc)
2736 {
2737 int i;
2738 VReg temp;
2739 VReg *Vd = (VReg *)vd;
2740 VReg *Vj = (VReg *)vj;
2741
2742 vec_clear_cause(env);
2743 for (i = 0; i < 2; i++) {
2744 temp.D(i) = int32_to_float64(Vj->W(i), &env->fp_status);
2745 vec_update_fcsr0(env, GETPC());
2746 }
2747 *Vd = temp;
2748 }
2749
2750 void HELPER(vffinth_d_w)(void *vd, void *vj,
2751 CPULoongArchState *env, uint32_t desc)
2752 {
2753 int i;
2754 VReg temp;
2755 VReg *Vd = (VReg *)vd;
2756 VReg *Vj = (VReg *)vj;
2757
2758 vec_clear_cause(env);
2759 for (i = 0; i < 2; i++) {
2760 temp.D(i) = int32_to_float64(Vj->W(i + 2), &env->fp_status);
2761 vec_update_fcsr0(env, GETPC());
2762 }
2763 *Vd = temp;
2764 }
2765
2766 void HELPER(vffint_s_l)(void *vd, void *vj, void *vk,
2767 CPULoongArchState *env, uint32_t desc)
2768 {
2769 int i;
2770 VReg temp;
2771 VReg *Vd = (VReg *)vd;
2772 VReg *Vj = (VReg *)vj;
2773 VReg *Vk = (VReg *)vk;
2774
2775 vec_clear_cause(env);
2776 for (i = 0; i < 2; i++) {
2777 temp.W(i + 2) = int64_to_float32(Vj->D(i), &env->fp_status);
2778 temp.W(i) = int64_to_float32(Vk->D(i), &env->fp_status);
2779 vec_update_fcsr0(env, GETPC());
2780 }
2781 *Vd = temp;
2782 }
2783
2784 #define VSEQ(a, b) (a == b ? -1 : 0)
2785 #define VSLE(a, b) (a <= b ? -1 : 0)
2786 #define VSLT(a, b) (a < b ? -1 : 0)
2787
2788 #define VCMPI(NAME, BIT, E, DO_OP) \
2789 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t v) \
2790 { \
2791 int i; \
2792 VReg *Vd = (VReg *)vd; \
2793 VReg *Vj = (VReg *)vj; \
2794 typedef __typeof(Vd->E(0)) TD; \
2795 \
2796 for (i = 0; i < LSX_LEN/BIT; i++) { \
2797 Vd->E(i) = DO_OP(Vj->E(i), (TD)imm); \
2798 } \
2799 }
2800
2801 VCMPI(vseqi_b, 8, B, VSEQ)
2802 VCMPI(vseqi_h, 16, H, VSEQ)
2803 VCMPI(vseqi_w, 32, W, VSEQ)
2804 VCMPI(vseqi_d, 64, D, VSEQ)
2805 VCMPI(vslei_b, 8, B, VSLE)
2806 VCMPI(vslei_h, 16, H, VSLE)
2807 VCMPI(vslei_w, 32, W, VSLE)
2808 VCMPI(vslei_d, 64, D, VSLE)
2809 VCMPI(vslei_bu, 8, UB, VSLE)
2810 VCMPI(vslei_hu, 16, UH, VSLE)
2811 VCMPI(vslei_wu, 32, UW, VSLE)
2812 VCMPI(vslei_du, 64, UD, VSLE)
2813 VCMPI(vslti_b, 8, B, VSLT)
2814 VCMPI(vslti_h, 16, H, VSLT)
2815 VCMPI(vslti_w, 32, W, VSLT)
2816 VCMPI(vslti_d, 64, D, VSLT)
2817 VCMPI(vslti_bu, 8, UB, VSLT)
2818 VCMPI(vslti_hu, 16, UH, VSLT)
2819 VCMPI(vslti_wu, 32, UW, VSLT)
2820 VCMPI(vslti_du, 64, UD, VSLT)
2821
2822 static uint64_t vfcmp_common(CPULoongArchState *env,
2823 FloatRelation cmp, uint32_t flags)
2824 {
2825 uint64_t ret = 0;
2826
2827 switch (cmp) {
2828 case float_relation_less:
2829 ret = (flags & FCMP_LT);
2830 break;
2831 case float_relation_equal:
2832 ret = (flags & FCMP_EQ);
2833 break;
2834 case float_relation_greater:
2835 ret = (flags & FCMP_GT);
2836 break;
2837 case float_relation_unordered:
2838 ret = (flags & FCMP_UN);
2839 break;
2840 default:
2841 g_assert_not_reached();
2842 }
2843
2844 if (ret) {
2845 ret = -1;
2846 }
2847
2848 return ret;
2849 }
2850
2851 #define VFCMP(NAME, BIT, E, FN) \
2852 void HELPER(NAME)(CPULoongArchState *env, \
2853 uint32_t vd, uint32_t vj, uint32_t vk, uint32_t flags) \
2854 { \
2855 int i; \
2856 VReg t; \
2857 VReg *Vd = &(env->fpr[vd].vreg); \
2858 VReg *Vj = &(env->fpr[vj].vreg); \
2859 VReg *Vk = &(env->fpr[vk].vreg); \
2860 \
2861 vec_clear_cause(env); \
2862 for (i = 0; i < LSX_LEN/BIT ; i++) { \
2863 FloatRelation cmp; \
2864 cmp = FN(Vj->E(i), Vk->E(i), &env->fp_status); \
2865 t.E(i) = vfcmp_common(env, cmp, flags); \
2866 vec_update_fcsr0(env, GETPC()); \
2867 } \
2868 *Vd = t; \
2869 }
2870
2871 VFCMP(vfcmp_c_s, 32, UW, float32_compare_quiet)
2872 VFCMP(vfcmp_s_s, 32, UW, float32_compare)
2873 VFCMP(vfcmp_c_d, 64, UD, float64_compare_quiet)
2874 VFCMP(vfcmp_s_d, 64, UD, float64_compare)
2875
2876 void HELPER(vbitseli_b)(void *vd, void *vj, uint64_t imm, uint32_t v)
2877 {
2878 int i;
2879 VReg *Vd = (VReg *)vd;
2880 VReg *Vj = (VReg *)vj;
2881
2882 for (i = 0; i < 16; i++) {
2883 Vd->B(i) = (~Vd->B(i) & Vj->B(i)) | (Vd->B(i) & imm);
2884 }
2885 }
2886
2887 /* Copy from target/arm/tcg/sve_helper.c */
2888 static inline bool do_match2(uint64_t n, uint64_t m0, uint64_t m1, int esz)
2889 {
2890 uint64_t bits = 8 << esz;
2891 uint64_t ones = dup_const(esz, 1);
2892 uint64_t signs = ones << (bits - 1);
2893 uint64_t cmp0, cmp1;
2894
2895 cmp1 = dup_const(esz, n);
2896 cmp0 = cmp1 ^ m0;
2897 cmp1 = cmp1 ^ m1;
2898 cmp0 = (cmp0 - ones) & ~cmp0;
2899 cmp1 = (cmp1 - ones) & ~cmp1;
2900 return (cmp0 | cmp1) & signs;
2901 }
2902
2903 #define SETANYEQZ(NAME, MO) \
2904 void HELPER(NAME)(CPULoongArchState *env, uint32_t cd, uint32_t vj) \
2905 { \
2906 VReg *Vj = &(env->fpr[vj].vreg); \
2907 \
2908 env->cf[cd & 0x7] = do_match2(0, Vj->D(0), Vj->D(1), MO); \
2909 }
2910 SETANYEQZ(vsetanyeqz_b, MO_8)
2911 SETANYEQZ(vsetanyeqz_h, MO_16)
2912 SETANYEQZ(vsetanyeqz_w, MO_32)
2913 SETANYEQZ(vsetanyeqz_d, MO_64)
2914
2915 #define SETALLNEZ(NAME, MO) \
2916 void HELPER(NAME)(CPULoongArchState *env, uint32_t cd, uint32_t vj) \
2917 { \
2918 VReg *Vj = &(env->fpr[vj].vreg); \
2919 \
2920 env->cf[cd & 0x7]= !do_match2(0, Vj->D(0), Vj->D(1), MO); \
2921 }
2922 SETALLNEZ(vsetallnez_b, MO_8)
2923 SETALLNEZ(vsetallnez_h, MO_16)
2924 SETALLNEZ(vsetallnez_w, MO_32)
2925 SETALLNEZ(vsetallnez_d, MO_64)
2926
2927 #define VPACKEV(NAME, BIT, E) \
2928 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
2929 { \
2930 int i; \
2931 VReg temp; \
2932 VReg *Vd = (VReg *)vd; \
2933 VReg *Vj = (VReg *)vj; \
2934 VReg *Vk = (VReg *)vk; \
2935 \
2936 for (i = 0; i < LSX_LEN/BIT; i++) { \
2937 temp.E(2 * i + 1) = Vj->E(2 * i); \
2938 temp.E(2 *i) = Vk->E(2 * i); \
2939 } \
2940 *Vd = temp; \
2941 }
2942
2943 VPACKEV(vpackev_b, 16, B)
2944 VPACKEV(vpackev_h, 32, H)
2945 VPACKEV(vpackev_w, 64, W)
2946 VPACKEV(vpackev_d, 128, D)
2947
2948 #define VPACKOD(NAME, BIT, E) \
2949 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
2950 { \
2951 int i; \
2952 VReg temp; \
2953 VReg *Vd = (VReg *)vd; \
2954 VReg *Vj = (VReg *)vj; \
2955 VReg *Vk = (VReg *)vk; \
2956 \
2957 for (i = 0; i < LSX_LEN/BIT; i++) { \
2958 temp.E(2 * i + 1) = Vj->E(2 * i + 1); \
2959 temp.E(2 * i) = Vk->E(2 * i + 1); \
2960 } \
2961 *Vd = temp; \
2962 }
2963
2964 VPACKOD(vpackod_b, 16, B)
2965 VPACKOD(vpackod_h, 32, H)
2966 VPACKOD(vpackod_w, 64, W)
2967 VPACKOD(vpackod_d, 128, D)
2968
2969 #define VPICKEV(NAME, BIT, E) \
2970 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
2971 { \
2972 int i; \
2973 VReg temp; \
2974 VReg *Vd = (VReg *)vd; \
2975 VReg *Vj = (VReg *)vj; \
2976 VReg *Vk = (VReg *)vk; \
2977 \
2978 for (i = 0; i < LSX_LEN/BIT; i++) { \
2979 temp.E(i + LSX_LEN/BIT) = Vj->E(2 * i); \
2980 temp.E(i) = Vk->E(2 * i); \
2981 } \
2982 *Vd = temp; \
2983 }
2984
2985 VPICKEV(vpickev_b, 16, B)
2986 VPICKEV(vpickev_h, 32, H)
2987 VPICKEV(vpickev_w, 64, W)
2988 VPICKEV(vpickev_d, 128, D)
2989
2990 #define VPICKOD(NAME, BIT, E) \
2991 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
2992 { \
2993 int i; \
2994 VReg temp; \
2995 VReg *Vd = (VReg *)vd; \
2996 VReg *Vj = (VReg *)vj; \
2997 VReg *Vk = (VReg *)vk; \
2998 \
2999 for (i = 0; i < LSX_LEN/BIT; i++) { \
3000 temp.E(i + LSX_LEN/BIT) = Vj->E(2 * i + 1); \
3001 temp.E(i) = Vk->E(2 * i + 1); \
3002 } \
3003 *Vd = temp; \
3004 }
3005
3006 VPICKOD(vpickod_b, 16, B)
3007 VPICKOD(vpickod_h, 32, H)
3008 VPICKOD(vpickod_w, 64, W)
3009 VPICKOD(vpickod_d, 128, D)
3010
3011 #define VILVL(NAME, BIT, E) \
3012 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
3013 { \
3014 int i; \
3015 VReg temp; \
3016 VReg *Vd = (VReg *)vd; \
3017 VReg *Vj = (VReg *)vj; \
3018 VReg *Vk = (VReg *)vk; \
3019 \
3020 for (i = 0; i < LSX_LEN/BIT; i++) { \
3021 temp.E(2 * i + 1) = Vj->E(i); \
3022 temp.E(2 * i) = Vk->E(i); \
3023 } \
3024 *Vd = temp; \
3025 }
3026
3027 VILVL(vilvl_b, 16, B)
3028 VILVL(vilvl_h, 32, H)
3029 VILVL(vilvl_w, 64, W)
3030 VILVL(vilvl_d, 128, D)
3031
3032 #define VILVH(NAME, BIT, E) \
3033 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
3034 { \
3035 int i; \
3036 VReg temp; \
3037 VReg *Vd = (VReg *)vd; \
3038 VReg *Vj = (VReg *)vj; \
3039 VReg *Vk = (VReg *)vk; \
3040 \
3041 for (i = 0; i < LSX_LEN/BIT; i++) { \
3042 temp.E(2 * i + 1) = Vj->E(i + LSX_LEN/BIT); \
3043 temp.E(2 * i) = Vk->E(i + LSX_LEN/BIT); \
3044 } \
3045 *Vd = temp; \
3046 }
3047
3048 VILVH(vilvh_b, 16, B)
3049 VILVH(vilvh_h, 32, H)
3050 VILVH(vilvh_w, 64, W)
3051 VILVH(vilvh_d, 128, D)
3052
3053 void HELPER(vshuf_b)(void *vd, void *vj, void *vk, void *va, uint32_t desc)
3054 {
3055 int i, m;
3056 VReg temp;
3057 VReg *Vd = (VReg *)vd;
3058 VReg *Vj = (VReg *)vj;
3059 VReg *Vk = (VReg *)vk;
3060 VReg *Va = (VReg *)va;
3061
3062 m = LSX_LEN/8;
3063 for (i = 0; i < m ; i++) {
3064 uint64_t k = (uint8_t)Va->B(i) % (2 * m);
3065 temp.B(i) = k < m ? Vk->B(k) : Vj->B(k - m);
3066 }
3067 *Vd = temp;
3068 }
3069
3070 #define VSHUF(NAME, BIT, E) \
3071 void HELPER(NAME)(void *vd, void *vj, void *vk, uint32_t desc) \
3072 { \
3073 int i, m; \
3074 VReg temp; \
3075 VReg *Vd = (VReg *)vd; \
3076 VReg *Vj = (VReg *)vj; \
3077 VReg *Vk = (VReg *)vk; \
3078 \
3079 m = LSX_LEN/BIT; \
3080 for (i = 0; i < m; i++) { \
3081 uint64_t k = ((uint8_t) Vd->E(i)) % (2 * m); \
3082 temp.E(i) = k < m ? Vk->E(k) : Vj->E(k - m); \
3083 } \
3084 *Vd = temp; \
3085 }
3086
3087 VSHUF(vshuf_h, 16, H)
3088 VSHUF(vshuf_w, 32, W)
3089 VSHUF(vshuf_d, 64, D)
3090
3091 #define VSHUF4I(NAME, BIT, E) \
3092 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
3093 { \
3094 int i; \
3095 VReg temp; \
3096 VReg *Vd = (VReg *)vd; \
3097 VReg *Vj = (VReg *)vj; \
3098 \
3099 for (i = 0; i < LSX_LEN/BIT; i++) { \
3100 temp.E(i) = Vj->E(((i) & 0xfc) + (((imm) >> \
3101 (2 * ((i) & 0x03))) & 0x03)); \
3102 } \
3103 *Vd = temp; \
3104 }
3105
3106 VSHUF4I(vshuf4i_b, 8, B)
3107 VSHUF4I(vshuf4i_h, 16, H)
3108 VSHUF4I(vshuf4i_w, 32, W)
3109
3110 void HELPER(vshuf4i_d)(void *vd, void *vj, uint64_t imm, uint32_t desc)
3111 {
3112 VReg *Vd = (VReg *)vd;
3113 VReg *Vj = (VReg *)vj;
3114
3115 VReg temp;
3116 temp.D(0) = (imm & 2 ? Vj : Vd)->D(imm & 1);
3117 temp.D(1) = (imm & 8 ? Vj : Vd)->D((imm >> 2) & 1);
3118 *Vd = temp;
3119 }
3120
3121 void HELPER(vpermi_w)(void *vd, void *vj, uint64_t imm, uint32_t desc)
3122 {
3123 VReg temp;
3124 VReg *Vd = (VReg *)vd;
3125 VReg *Vj = (VReg *)vj;
3126
3127 temp.W(0) = Vj->W(imm & 0x3);
3128 temp.W(1) = Vj->W((imm >> 2) & 0x3);
3129 temp.W(2) = Vd->W((imm >> 4) & 0x3);
3130 temp.W(3) = Vd->W((imm >> 6) & 0x3);
3131 *Vd = temp;
3132 }
3133
3134 #define VEXTRINS(NAME, BIT, E, MASK) \
3135 void HELPER(NAME)(void *vd, void *vj, uint64_t imm, uint32_t desc) \
3136 { \
3137 int ins, extr; \
3138 VReg *Vd = (VReg *)vd; \
3139 VReg *Vj = (VReg *)vj; \
3140 \
3141 ins = (imm >> 4) & MASK; \
3142 extr = imm & MASK; \
3143 Vd->E(ins) = Vj->E(extr); \
3144 }
3145
3146 VEXTRINS(vextrins_b, 8, B, 0xf)
3147 VEXTRINS(vextrins_h, 16, H, 0x7)
3148 VEXTRINS(vextrins_w, 32, W, 0x3)
3149 VEXTRINS(vextrins_d, 64, D, 0x1)
QEMU mirror