xref: /openbmc/qemu/target/riscv/vector_helper.c (revision ffe98631)
1 /*
2  * RISC-V Vector Extension Helpers for QEMU.
3  *
4  * Copyright (c) 2020 T-Head Semiconductor Co., Ltd. All rights reserved.
5  *
6  * This program is free software; you can redistribute it and/or modify it
7  * under the terms and conditions of the GNU General Public License,
8  * version 2 or later, as published by the Free Software Foundation.
9  *
10  * This program is distributed in the hope it will be useful, but WITHOUT
11  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
13  * more details.
14  *
15  * You should have received a copy of the GNU General Public License along with
16  * this program.  If not, see <http://www.gnu.org/licenses/>.
17  */
18 
19 #include "qemu/osdep.h"
20 #include "qemu/host-utils.h"
21 #include "qemu/bitops.h"
22 #include "cpu.h"
23 #include "exec/memop.h"
24 #include "exec/exec-all.h"
25 #include "exec/helper-proto.h"
26 #include "fpu/softfloat.h"
27 #include "tcg/tcg-gvec-desc.h"
28 #include "internals.h"
29 #include <math.h>
30 
31 target_ulong HELPER(vsetvl)(CPURISCVState *env, target_ulong s1,
32                             target_ulong s2)
33 {
34     int vlmax, vl;
35     RISCVCPU *cpu = env_archcpu(env);
36     uint64_t lmul = FIELD_EX64(s2, VTYPE, VLMUL);
37     uint16_t sew = 8 << FIELD_EX64(s2, VTYPE, VSEW);
38     uint8_t ediv = FIELD_EX64(s2, VTYPE, VEDIV);
39     int xlen = riscv_cpu_xlen(env);
40     bool vill = (s2 >> (xlen - 1)) & 0x1;
41     target_ulong reserved = s2 &
42                             MAKE_64BIT_MASK(R_VTYPE_RESERVED_SHIFT,
43                                             xlen - 1 - R_VTYPE_RESERVED_SHIFT);
44 
45     if (lmul & 4) {
46         /* Fractional LMUL. */
47         if (lmul == 4 ||
48             cpu->cfg.elen >> (8 - lmul) < sew) {
49             vill = true;
50         }
51     }
52 
53     if ((sew > cpu->cfg.elen)
54         || vill
55         || (ediv != 0)
56         || (reserved != 0)) {
57         /* only set vill bit. */
58         env->vill = 1;
59         env->vtype = 0;
60         env->vl = 0;
61         env->vstart = 0;
62         return 0;
63     }
64 
65     vlmax = vext_get_vlmax(cpu, s2);
66     if (s1 <= vlmax) {
67         vl = s1;
68     } else {
69         vl = vlmax;
70     }
71     env->vl = vl;
72     env->vtype = s2;
73     env->vstart = 0;
74     env->vill = 0;
75     return vl;
76 }
77 
78 /*
79  * Note that vector data is stored in host-endian 64-bit chunks,
80  * so addressing units smaller than that needs a host-endian fixup.
81  */
82 #if HOST_BIG_ENDIAN
83 #define H1(x)   ((x) ^ 7)
84 #define H1_2(x) ((x) ^ 6)
85 #define H1_4(x) ((x) ^ 4)
86 #define H2(x)   ((x) ^ 3)
87 #define H4(x)   ((x) ^ 1)
88 #define H8(x)   ((x))
89 #else
90 #define H1(x)   (x)
91 #define H1_2(x) (x)
92 #define H1_4(x) (x)
93 #define H2(x)   (x)
94 #define H4(x)   (x)
95 #define H8(x)   (x)
96 #endif
97 
98 static inline uint32_t vext_nf(uint32_t desc)
99 {
100     return FIELD_EX32(simd_data(desc), VDATA, NF);
101 }
102 
103 static inline uint32_t vext_vm(uint32_t desc)
104 {
105     return FIELD_EX32(simd_data(desc), VDATA, VM);
106 }
107 
108 /*
109  * Encode LMUL to lmul as following:
110  *     LMUL    vlmul    lmul
111  *      1       000       0
112  *      2       001       1
113  *      4       010       2
114  *      8       011       3
115  *      -       100       -
116  *     1/8      101      -3
117  *     1/4      110      -2
118  *     1/2      111      -1
119  */
120 static inline int32_t vext_lmul(uint32_t desc)
121 {
122     return sextract32(FIELD_EX32(simd_data(desc), VDATA, LMUL), 0, 3);
123 }
124 
125 static inline uint32_t vext_vta(uint32_t desc)
126 {
127     return FIELD_EX32(simd_data(desc), VDATA, VTA);
128 }
129 
130 static inline uint32_t vext_vma(uint32_t desc)
131 {
132     return FIELD_EX32(simd_data(desc), VDATA, VMA);
133 }
134 
135 static inline uint32_t vext_vta_all_1s(uint32_t desc)
136 {
137     return FIELD_EX32(simd_data(desc), VDATA, VTA_ALL_1S);
138 }
139 
140 /*
141  * Get the maximum number of elements can be operated.
142  *
143  * log2_esz: log2 of element size in bytes.
144  */
145 static inline uint32_t vext_max_elems(uint32_t desc, uint32_t log2_esz)
146 {
147     /*
148      * As simd_desc support at most 2048 bytes, the max vlen is 1024 bits.
149      * so vlen in bytes (vlenb) is encoded as maxsz.
150      */
151     uint32_t vlenb = simd_maxsz(desc);
152 
153     /* Return VLMAX */
154     int scale = vext_lmul(desc) - log2_esz;
155     return scale < 0 ? vlenb >> -scale : vlenb << scale;
156 }
157 
158 /*
159  * Get number of total elements, including prestart, body and tail elements.
160  * Note that when LMUL < 1, the tail includes the elements past VLMAX that
161  * are held in the same vector register.
162  */
163 static inline uint32_t vext_get_total_elems(CPURISCVState *env, uint32_t desc,
164                                             uint32_t esz)
165 {
166     uint32_t vlenb = simd_maxsz(desc);
167     uint32_t sew = 1 << FIELD_EX64(env->vtype, VTYPE, VSEW);
168     int8_t emul = ctzl(esz) - ctzl(sew) + vext_lmul(desc) < 0 ? 0 :
169                   ctzl(esz) - ctzl(sew) + vext_lmul(desc);
170     return (vlenb << emul) / esz;
171 }
172 
173 static inline target_ulong adjust_addr(CPURISCVState *env, target_ulong addr)
174 {
175     return (addr & env->cur_pmmask) | env->cur_pmbase;
176 }
177 
178 /*
179  * This function checks watchpoint before real load operation.
180  *
181  * In softmmu mode, the TLB API probe_access is enough for watchpoint check.
182  * In user mode, there is no watchpoint support now.
183  *
184  * It will trigger an exception if there is no mapping in TLB
185  * and page table walk can't fill the TLB entry. Then the guest
186  * software can return here after process the exception or never return.
187  */
188 static void probe_pages(CPURISCVState *env, target_ulong addr,
189                         target_ulong len, uintptr_t ra,
190                         MMUAccessType access_type)
191 {
192     target_ulong pagelen = -(addr | TARGET_PAGE_MASK);
193     target_ulong curlen = MIN(pagelen, len);
194 
195     probe_access(env, adjust_addr(env, addr), curlen, access_type,
196                  cpu_mmu_index(env, false), ra);
197     if (len > curlen) {
198         addr += curlen;
199         curlen = len - curlen;
200         probe_access(env, adjust_addr(env, addr), curlen, access_type,
201                      cpu_mmu_index(env, false), ra);
202     }
203 }
204 
205 /* set agnostic elements to 1s */
206 static void vext_set_elems_1s(void *base, uint32_t is_agnostic, uint32_t cnt,
207                               uint32_t tot)
208 {
209     if (is_agnostic == 0) {
210         /* policy undisturbed */
211         return;
212     }
213     if (tot - cnt == 0) {
214         return;
215     }
216     memset(base + cnt, -1, tot - cnt);
217 }
218 
219 static inline void vext_set_elem_mask(void *v0, int index,
220                                       uint8_t value)
221 {
222     int idx = index / 64;
223     int pos = index % 64;
224     uint64_t old = ((uint64_t *)v0)[idx];
225     ((uint64_t *)v0)[idx] = deposit64(old, pos, 1, value);
226 }
227 
228 /*
229  * Earlier designs (pre-0.9) had a varying number of bits
230  * per mask value (MLEN). In the 0.9 design, MLEN=1.
231  * (Section 4.5)
232  */
233 static inline int vext_elem_mask(void *v0, int index)
234 {
235     int idx = index / 64;
236     int pos = index  % 64;
237     return (((uint64_t *)v0)[idx] >> pos) & 1;
238 }
239 
240 /* elements operations for load and store */
241 typedef void vext_ldst_elem_fn(CPURISCVState *env, target_ulong addr,
242                                uint32_t idx, void *vd, uintptr_t retaddr);
243 
244 #define GEN_VEXT_LD_ELEM(NAME, ETYPE, H, LDSUF)            \
245 static void NAME(CPURISCVState *env, abi_ptr addr,         \
246                  uint32_t idx, void *vd, uintptr_t retaddr)\
247 {                                                          \
248     ETYPE *cur = ((ETYPE *)vd + H(idx));                   \
249     *cur = cpu_##LDSUF##_data_ra(env, addr, retaddr);      \
250 }                                                          \
251 
252 GEN_VEXT_LD_ELEM(lde_b, int8_t,  H1, ldsb)
253 GEN_VEXT_LD_ELEM(lde_h, int16_t, H2, ldsw)
254 GEN_VEXT_LD_ELEM(lde_w, int32_t, H4, ldl)
255 GEN_VEXT_LD_ELEM(lde_d, int64_t, H8, ldq)
256 
257 #define GEN_VEXT_ST_ELEM(NAME, ETYPE, H, STSUF)            \
258 static void NAME(CPURISCVState *env, abi_ptr addr,         \
259                  uint32_t idx, void *vd, uintptr_t retaddr)\
260 {                                                          \
261     ETYPE data = *((ETYPE *)vd + H(idx));                  \
262     cpu_##STSUF##_data_ra(env, addr, data, retaddr);       \
263 }
264 
265 GEN_VEXT_ST_ELEM(ste_b, int8_t,  H1, stb)
266 GEN_VEXT_ST_ELEM(ste_h, int16_t, H2, stw)
267 GEN_VEXT_ST_ELEM(ste_w, int32_t, H4, stl)
268 GEN_VEXT_ST_ELEM(ste_d, int64_t, H8, stq)
269 
270 /*
271  *** stride: access vector element from strided memory
272  */
273 static void
274 vext_ldst_stride(void *vd, void *v0, target_ulong base,
275                  target_ulong stride, CPURISCVState *env,
276                  uint32_t desc, uint32_t vm,
277                  vext_ldst_elem_fn *ldst_elem,
278                  uint32_t log2_esz, uintptr_t ra)
279 {
280     uint32_t i, k;
281     uint32_t nf = vext_nf(desc);
282     uint32_t max_elems = vext_max_elems(desc, log2_esz);
283     uint32_t esz = 1 << log2_esz;
284     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
285     uint32_t vta = vext_vta(desc);
286     uint32_t vma = vext_vma(desc);
287 
288     for (i = env->vstart; i < env->vl; i++, env->vstart++) {
289         k = 0;
290         while (k < nf) {
291             if (!vm && !vext_elem_mask(v0, i)) {
292                 /* set masked-off elements to 1s */
293                 vext_set_elems_1s(vd, vma, (i + k * max_elems) * esz,
294                                   (i + k * max_elems + 1) * esz);
295                 k++;
296                 continue;
297             }
298             target_ulong addr = base + stride * i + (k << log2_esz);
299             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
300             k++;
301         }
302     }
303     env->vstart = 0;
304     /* set tail elements to 1s */
305     for (k = 0; k < nf; ++k) {
306         vext_set_elems_1s(vd, vta, (k * max_elems + env->vl) * esz,
307                           (k * max_elems + max_elems) * esz);
308     }
309     if (nf * max_elems % total_elems != 0) {
310         uint32_t vlenb = env_archcpu(env)->cfg.vlen >> 3;
311         uint32_t registers_used =
312             ((nf * max_elems) * esz + (vlenb - 1)) / vlenb;
313         vext_set_elems_1s(vd, vta, (nf * max_elems) * esz,
314                           registers_used * vlenb);
315     }
316 }
317 
318 #define GEN_VEXT_LD_STRIDE(NAME, ETYPE, LOAD_FN)                        \
319 void HELPER(NAME)(void *vd, void * v0, target_ulong base,               \
320                   target_ulong stride, CPURISCVState *env,              \
321                   uint32_t desc)                                        \
322 {                                                                       \
323     uint32_t vm = vext_vm(desc);                                        \
324     vext_ldst_stride(vd, v0, base, stride, env, desc, vm, LOAD_FN,      \
325                      ctzl(sizeof(ETYPE)), GETPC());                     \
326 }
327 
328 GEN_VEXT_LD_STRIDE(vlse8_v,  int8_t,  lde_b)
329 GEN_VEXT_LD_STRIDE(vlse16_v, int16_t, lde_h)
330 GEN_VEXT_LD_STRIDE(vlse32_v, int32_t, lde_w)
331 GEN_VEXT_LD_STRIDE(vlse64_v, int64_t, lde_d)
332 
333 #define GEN_VEXT_ST_STRIDE(NAME, ETYPE, STORE_FN)                       \
334 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
335                   target_ulong stride, CPURISCVState *env,              \
336                   uint32_t desc)                                        \
337 {                                                                       \
338     uint32_t vm = vext_vm(desc);                                        \
339     vext_ldst_stride(vd, v0, base, stride, env, desc, vm, STORE_FN,     \
340                      ctzl(sizeof(ETYPE)), GETPC());                     \
341 }
342 
343 GEN_VEXT_ST_STRIDE(vsse8_v,  int8_t,  ste_b)
344 GEN_VEXT_ST_STRIDE(vsse16_v, int16_t, ste_h)
345 GEN_VEXT_ST_STRIDE(vsse32_v, int32_t, ste_w)
346 GEN_VEXT_ST_STRIDE(vsse64_v, int64_t, ste_d)
347 
348 /*
349  *** unit-stride: access elements stored contiguously in memory
350  */
351 
352 /* unmasked unit-stride load and store operation*/
353 static void
354 vext_ldst_us(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc,
355              vext_ldst_elem_fn *ldst_elem, uint32_t log2_esz, uint32_t evl,
356              uintptr_t ra)
357 {
358     uint32_t i, k;
359     uint32_t nf = vext_nf(desc);
360     uint32_t max_elems = vext_max_elems(desc, log2_esz);
361     uint32_t esz = 1 << log2_esz;
362     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
363     uint32_t vta = vext_vta(desc);
364 
365     /* load bytes from guest memory */
366     for (i = env->vstart; i < evl; i++, env->vstart++) {
367         k = 0;
368         while (k < nf) {
369             target_ulong addr = base + ((i * nf + k) << log2_esz);
370             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
371             k++;
372         }
373     }
374     env->vstart = 0;
375     /* set tail elements to 1s */
376     for (k = 0; k < nf; ++k) {
377         vext_set_elems_1s(vd, vta, (k * max_elems + evl) * esz,
378                           (k * max_elems + max_elems) * esz);
379     }
380     if (nf * max_elems % total_elems != 0) {
381         uint32_t vlenb = env_archcpu(env)->cfg.vlen >> 3;
382         uint32_t registers_used =
383             ((nf * max_elems) * esz + (vlenb - 1)) / vlenb;
384         vext_set_elems_1s(vd, vta, (nf * max_elems) * esz,
385                           registers_used * vlenb);
386     }
387 }
388 
389 /*
390  * masked unit-stride load and store operation will be a special case of stride,
391  * stride = NF * sizeof (MTYPE)
392  */
393 
394 #define GEN_VEXT_LD_US(NAME, ETYPE, LOAD_FN)                            \
395 void HELPER(NAME##_mask)(void *vd, void *v0, target_ulong base,         \
396                          CPURISCVState *env, uint32_t desc)             \
397 {                                                                       \
398     uint32_t stride = vext_nf(desc) << ctzl(sizeof(ETYPE));             \
399     vext_ldst_stride(vd, v0, base, stride, env, desc, false, LOAD_FN,   \
400                      ctzl(sizeof(ETYPE)), GETPC());                     \
401 }                                                                       \
402                                                                         \
403 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
404                   CPURISCVState *env, uint32_t desc)                    \
405 {                                                                       \
406     vext_ldst_us(vd, base, env, desc, LOAD_FN,                          \
407                  ctzl(sizeof(ETYPE)), env->vl, GETPC());                \
408 }
409 
410 GEN_VEXT_LD_US(vle8_v,  int8_t,  lde_b)
411 GEN_VEXT_LD_US(vle16_v, int16_t, lde_h)
412 GEN_VEXT_LD_US(vle32_v, int32_t, lde_w)
413 GEN_VEXT_LD_US(vle64_v, int64_t, lde_d)
414 
415 #define GEN_VEXT_ST_US(NAME, ETYPE, STORE_FN)                            \
416 void HELPER(NAME##_mask)(void *vd, void *v0, target_ulong base,          \
417                          CPURISCVState *env, uint32_t desc)              \
418 {                                                                        \
419     uint32_t stride = vext_nf(desc) << ctzl(sizeof(ETYPE));              \
420     vext_ldst_stride(vd, v0, base, stride, env, desc, false, STORE_FN,   \
421                      ctzl(sizeof(ETYPE)), GETPC());                      \
422 }                                                                        \
423                                                                          \
424 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                 \
425                   CPURISCVState *env, uint32_t desc)                     \
426 {                                                                        \
427     vext_ldst_us(vd, base, env, desc, STORE_FN,                          \
428                  ctzl(sizeof(ETYPE)), env->vl, GETPC());                 \
429 }
430 
431 GEN_VEXT_ST_US(vse8_v,  int8_t,  ste_b)
432 GEN_VEXT_ST_US(vse16_v, int16_t, ste_h)
433 GEN_VEXT_ST_US(vse32_v, int32_t, ste_w)
434 GEN_VEXT_ST_US(vse64_v, int64_t, ste_d)
435 
436 /*
437  *** unit stride mask load and store, EEW = 1
438  */
439 void HELPER(vlm_v)(void *vd, void *v0, target_ulong base,
440                     CPURISCVState *env, uint32_t desc)
441 {
442     /* evl = ceil(vl/8) */
443     uint8_t evl = (env->vl + 7) >> 3;
444     vext_ldst_us(vd, base, env, desc, lde_b,
445                  0, evl, GETPC());
446 }
447 
448 void HELPER(vsm_v)(void *vd, void *v0, target_ulong base,
449                     CPURISCVState *env, uint32_t desc)
450 {
451     /* evl = ceil(vl/8) */
452     uint8_t evl = (env->vl + 7) >> 3;
453     vext_ldst_us(vd, base, env, desc, ste_b,
454                  0, evl, GETPC());
455 }
456 
457 /*
458  *** index: access vector element from indexed memory
459  */
460 typedef target_ulong vext_get_index_addr(target_ulong base,
461         uint32_t idx, void *vs2);
462 
463 #define GEN_VEXT_GET_INDEX_ADDR(NAME, ETYPE, H)        \
464 static target_ulong NAME(target_ulong base,            \
465                          uint32_t idx, void *vs2)      \
466 {                                                      \
467     return (base + *((ETYPE *)vs2 + H(idx)));          \
468 }
469 
470 GEN_VEXT_GET_INDEX_ADDR(idx_b, uint8_t,  H1)
471 GEN_VEXT_GET_INDEX_ADDR(idx_h, uint16_t, H2)
472 GEN_VEXT_GET_INDEX_ADDR(idx_w, uint32_t, H4)
473 GEN_VEXT_GET_INDEX_ADDR(idx_d, uint64_t, H8)
474 
475 static inline void
476 vext_ldst_index(void *vd, void *v0, target_ulong base,
477                 void *vs2, CPURISCVState *env, uint32_t desc,
478                 vext_get_index_addr get_index_addr,
479                 vext_ldst_elem_fn *ldst_elem,
480                 uint32_t log2_esz, uintptr_t ra)
481 {
482     uint32_t i, k;
483     uint32_t nf = vext_nf(desc);
484     uint32_t vm = vext_vm(desc);
485     uint32_t max_elems = vext_max_elems(desc, log2_esz);
486     uint32_t esz = 1 << log2_esz;
487     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
488     uint32_t vta = vext_vta(desc);
489     uint32_t vma = vext_vma(desc);
490 
491     /* load bytes from guest memory */
492     for (i = env->vstart; i < env->vl; i++, env->vstart++) {
493         k = 0;
494         while (k < nf) {
495             if (!vm && !vext_elem_mask(v0, i)) {
496                 /* set masked-off elements to 1s */
497                 vext_set_elems_1s(vd, vma, (i + k * max_elems) * esz,
498                                   (i + k * max_elems + 1) * esz);
499                 k++;
500                 continue;
501             }
502             abi_ptr addr = get_index_addr(base, i, vs2) + (k << log2_esz);
503             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
504             k++;
505         }
506     }
507     env->vstart = 0;
508     /* set tail elements to 1s */
509     for (k = 0; k < nf; ++k) {
510         vext_set_elems_1s(vd, vta, (k * max_elems + env->vl) * esz,
511                           (k * max_elems + max_elems) * esz);
512     }
513     if (nf * max_elems % total_elems != 0) {
514         uint32_t vlenb = env_archcpu(env)->cfg.vlen >> 3;
515         uint32_t registers_used =
516             ((nf * max_elems) * esz + (vlenb - 1)) / vlenb;
517         vext_set_elems_1s(vd, vta, (nf * max_elems) * esz,
518                           registers_used * vlenb);
519     }
520 }
521 
522 #define GEN_VEXT_LD_INDEX(NAME, ETYPE, INDEX_FN, LOAD_FN)                  \
523 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                   \
524                   void *vs2, CPURISCVState *env, uint32_t desc)            \
525 {                                                                          \
526     vext_ldst_index(vd, v0, base, vs2, env, desc, INDEX_FN,                \
527                     LOAD_FN, ctzl(sizeof(ETYPE)), GETPC());                \
528 }
529 
530 GEN_VEXT_LD_INDEX(vlxei8_8_v,   int8_t,  idx_b, lde_b)
531 GEN_VEXT_LD_INDEX(vlxei8_16_v,  int16_t, idx_b, lde_h)
532 GEN_VEXT_LD_INDEX(vlxei8_32_v,  int32_t, idx_b, lde_w)
533 GEN_VEXT_LD_INDEX(vlxei8_64_v,  int64_t, idx_b, lde_d)
534 GEN_VEXT_LD_INDEX(vlxei16_8_v,  int8_t,  idx_h, lde_b)
535 GEN_VEXT_LD_INDEX(vlxei16_16_v, int16_t, idx_h, lde_h)
536 GEN_VEXT_LD_INDEX(vlxei16_32_v, int32_t, idx_h, lde_w)
537 GEN_VEXT_LD_INDEX(vlxei16_64_v, int64_t, idx_h, lde_d)
538 GEN_VEXT_LD_INDEX(vlxei32_8_v,  int8_t,  idx_w, lde_b)
539 GEN_VEXT_LD_INDEX(vlxei32_16_v, int16_t, idx_w, lde_h)
540 GEN_VEXT_LD_INDEX(vlxei32_32_v, int32_t, idx_w, lde_w)
541 GEN_VEXT_LD_INDEX(vlxei32_64_v, int64_t, idx_w, lde_d)
542 GEN_VEXT_LD_INDEX(vlxei64_8_v,  int8_t,  idx_d, lde_b)
543 GEN_VEXT_LD_INDEX(vlxei64_16_v, int16_t, idx_d, lde_h)
544 GEN_VEXT_LD_INDEX(vlxei64_32_v, int32_t, idx_d, lde_w)
545 GEN_VEXT_LD_INDEX(vlxei64_64_v, int64_t, idx_d, lde_d)
546 
547 #define GEN_VEXT_ST_INDEX(NAME, ETYPE, INDEX_FN, STORE_FN)       \
548 void HELPER(NAME)(void *vd, void *v0, target_ulong base,         \
549                   void *vs2, CPURISCVState *env, uint32_t desc)  \
550 {                                                                \
551     vext_ldst_index(vd, v0, base, vs2, env, desc, INDEX_FN,      \
552                     STORE_FN, ctzl(sizeof(ETYPE)),               \
553                     GETPC());                                    \
554 }
555 
556 GEN_VEXT_ST_INDEX(vsxei8_8_v,   int8_t,  idx_b, ste_b)
557 GEN_VEXT_ST_INDEX(vsxei8_16_v,  int16_t, idx_b, ste_h)
558 GEN_VEXT_ST_INDEX(vsxei8_32_v,  int32_t, idx_b, ste_w)
559 GEN_VEXT_ST_INDEX(vsxei8_64_v,  int64_t, idx_b, ste_d)
560 GEN_VEXT_ST_INDEX(vsxei16_8_v,  int8_t,  idx_h, ste_b)
561 GEN_VEXT_ST_INDEX(vsxei16_16_v, int16_t, idx_h, ste_h)
562 GEN_VEXT_ST_INDEX(vsxei16_32_v, int32_t, idx_h, ste_w)
563 GEN_VEXT_ST_INDEX(vsxei16_64_v, int64_t, idx_h, ste_d)
564 GEN_VEXT_ST_INDEX(vsxei32_8_v,  int8_t,  idx_w, ste_b)
565 GEN_VEXT_ST_INDEX(vsxei32_16_v, int16_t, idx_w, ste_h)
566 GEN_VEXT_ST_INDEX(vsxei32_32_v, int32_t, idx_w, ste_w)
567 GEN_VEXT_ST_INDEX(vsxei32_64_v, int64_t, idx_w, ste_d)
568 GEN_VEXT_ST_INDEX(vsxei64_8_v,  int8_t,  idx_d, ste_b)
569 GEN_VEXT_ST_INDEX(vsxei64_16_v, int16_t, idx_d, ste_h)
570 GEN_VEXT_ST_INDEX(vsxei64_32_v, int32_t, idx_d, ste_w)
571 GEN_VEXT_ST_INDEX(vsxei64_64_v, int64_t, idx_d, ste_d)
572 
573 /*
574  *** unit-stride fault-only-fisrt load instructions
575  */
576 static inline void
577 vext_ldff(void *vd, void *v0, target_ulong base,
578           CPURISCVState *env, uint32_t desc,
579           vext_ldst_elem_fn *ldst_elem,
580           uint32_t log2_esz, uintptr_t ra)
581 {
582     void *host;
583     uint32_t i, k, vl = 0;
584     uint32_t nf = vext_nf(desc);
585     uint32_t vm = vext_vm(desc);
586     uint32_t max_elems = vext_max_elems(desc, log2_esz);
587     uint32_t esz = 1 << log2_esz;
588     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
589     uint32_t vta = vext_vta(desc);
590     uint32_t vma = vext_vma(desc);
591     target_ulong addr, offset, remain;
592 
593     /* probe every access*/
594     for (i = env->vstart; i < env->vl; i++) {
595         if (!vm && !vext_elem_mask(v0, i)) {
596             continue;
597         }
598         addr = adjust_addr(env, base + i * (nf << log2_esz));
599         if (i == 0) {
600             probe_pages(env, addr, nf << log2_esz, ra, MMU_DATA_LOAD);
601         } else {
602             /* if it triggers an exception, no need to check watchpoint */
603             remain = nf << log2_esz;
604             while (remain > 0) {
605                 offset = -(addr | TARGET_PAGE_MASK);
606                 host = tlb_vaddr_to_host(env, addr, MMU_DATA_LOAD,
607                                          cpu_mmu_index(env, false));
608                 if (host) {
609 #ifdef CONFIG_USER_ONLY
610                     if (page_check_range(addr, offset, PAGE_READ) < 0) {
611                         vl = i;
612                         goto ProbeSuccess;
613                     }
614 #else
615                     probe_pages(env, addr, offset, ra, MMU_DATA_LOAD);
616 #endif
617                 } else {
618                     vl = i;
619                     goto ProbeSuccess;
620                 }
621                 if (remain <=  offset) {
622                     break;
623                 }
624                 remain -= offset;
625                 addr = adjust_addr(env, addr + offset);
626             }
627         }
628     }
629 ProbeSuccess:
630     /* load bytes from guest memory */
631     if (vl != 0) {
632         env->vl = vl;
633     }
634     for (i = env->vstart; i < env->vl; i++) {
635         k = 0;
636         while (k < nf) {
637             if (!vm && !vext_elem_mask(v0, i)) {
638                 /* set masked-off elements to 1s */
639                 vext_set_elems_1s(vd, vma, (i + k * max_elems) * esz,
640                                   (i + k * max_elems + 1) * esz);
641                 k++;
642                 continue;
643             }
644             target_ulong addr = base + ((i * nf + k) << log2_esz);
645             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
646             k++;
647         }
648     }
649     env->vstart = 0;
650     /* set tail elements to 1s */
651     for (k = 0; k < nf; ++k) {
652         vext_set_elems_1s(vd, vta, (k * max_elems + env->vl) * esz,
653                           (k * max_elems + max_elems) * esz);
654     }
655     if (nf * max_elems % total_elems != 0) {
656         uint32_t vlenb = env_archcpu(env)->cfg.vlen >> 3;
657         uint32_t registers_used =
658             ((nf * max_elems) * esz + (vlenb - 1)) / vlenb;
659         vext_set_elems_1s(vd, vta, (nf * max_elems) * esz,
660                           registers_used * vlenb);
661     }
662 }
663 
664 #define GEN_VEXT_LDFF(NAME, ETYPE, LOAD_FN)               \
665 void HELPER(NAME)(void *vd, void *v0, target_ulong base,  \
666                   CPURISCVState *env, uint32_t desc)      \
667 {                                                         \
668     vext_ldff(vd, v0, base, env, desc, LOAD_FN,           \
669               ctzl(sizeof(ETYPE)), GETPC());              \
670 }
671 
672 GEN_VEXT_LDFF(vle8ff_v,  int8_t,  lde_b)
673 GEN_VEXT_LDFF(vle16ff_v, int16_t, lde_h)
674 GEN_VEXT_LDFF(vle32ff_v, int32_t, lde_w)
675 GEN_VEXT_LDFF(vle64ff_v, int64_t, lde_d)
676 
677 #define DO_SWAP(N, M) (M)
678 #define DO_AND(N, M)  (N & M)
679 #define DO_XOR(N, M)  (N ^ M)
680 #define DO_OR(N, M)   (N | M)
681 #define DO_ADD(N, M)  (N + M)
682 
683 /* Signed min/max */
684 #define DO_MAX(N, M)  ((N) >= (M) ? (N) : (M))
685 #define DO_MIN(N, M)  ((N) >= (M) ? (M) : (N))
686 
687 /* Unsigned min/max */
688 #define DO_MAXU(N, M) DO_MAX((UMTYPE)N, (UMTYPE)M)
689 #define DO_MINU(N, M) DO_MIN((UMTYPE)N, (UMTYPE)M)
690 
691 /*
692  *** load and store whole register instructions
693  */
694 static void
695 vext_ldst_whole(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc,
696                 vext_ldst_elem_fn *ldst_elem, uint32_t log2_esz, uintptr_t ra)
697 {
698     uint32_t i, k, off, pos;
699     uint32_t nf = vext_nf(desc);
700     uint32_t vlenb = env_archcpu(env)->cfg.vlen >> 3;
701     uint32_t max_elems = vlenb >> log2_esz;
702 
703     k = env->vstart / max_elems;
704     off = env->vstart % max_elems;
705 
706     if (off) {
707         /* load/store rest of elements of current segment pointed by vstart */
708         for (pos = off; pos < max_elems; pos++, env->vstart++) {
709             target_ulong addr = base + ((pos + k * max_elems) << log2_esz);
710             ldst_elem(env, adjust_addr(env, addr), pos + k * max_elems, vd, ra);
711         }
712         k++;
713     }
714 
715     /* load/store elements for rest of segments */
716     for (; k < nf; k++) {
717         for (i = 0; i < max_elems; i++, env->vstart++) {
718             target_ulong addr = base + ((i + k * max_elems) << log2_esz);
719             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
720         }
721     }
722 
723     env->vstart = 0;
724 }
725 
726 #define GEN_VEXT_LD_WHOLE(NAME, ETYPE, LOAD_FN)      \
727 void HELPER(NAME)(void *vd, target_ulong base,       \
728                   CPURISCVState *env, uint32_t desc) \
729 {                                                    \
730     vext_ldst_whole(vd, base, env, desc, LOAD_FN,    \
731                     ctzl(sizeof(ETYPE)), GETPC());   \
732 }
733 
734 GEN_VEXT_LD_WHOLE(vl1re8_v,  int8_t,  lde_b)
735 GEN_VEXT_LD_WHOLE(vl1re16_v, int16_t, lde_h)
736 GEN_VEXT_LD_WHOLE(vl1re32_v, int32_t, lde_w)
737 GEN_VEXT_LD_WHOLE(vl1re64_v, int64_t, lde_d)
738 GEN_VEXT_LD_WHOLE(vl2re8_v,  int8_t,  lde_b)
739 GEN_VEXT_LD_WHOLE(vl2re16_v, int16_t, lde_h)
740 GEN_VEXT_LD_WHOLE(vl2re32_v, int32_t, lde_w)
741 GEN_VEXT_LD_WHOLE(vl2re64_v, int64_t, lde_d)
742 GEN_VEXT_LD_WHOLE(vl4re8_v,  int8_t,  lde_b)
743 GEN_VEXT_LD_WHOLE(vl4re16_v, int16_t, lde_h)
744 GEN_VEXT_LD_WHOLE(vl4re32_v, int32_t, lde_w)
745 GEN_VEXT_LD_WHOLE(vl4re64_v, int64_t, lde_d)
746 GEN_VEXT_LD_WHOLE(vl8re8_v,  int8_t,  lde_b)
747 GEN_VEXT_LD_WHOLE(vl8re16_v, int16_t, lde_h)
748 GEN_VEXT_LD_WHOLE(vl8re32_v, int32_t, lde_w)
749 GEN_VEXT_LD_WHOLE(vl8re64_v, int64_t, lde_d)
750 
751 #define GEN_VEXT_ST_WHOLE(NAME, ETYPE, STORE_FN)     \
752 void HELPER(NAME)(void *vd, target_ulong base,       \
753                   CPURISCVState *env, uint32_t desc) \
754 {                                                    \
755     vext_ldst_whole(vd, base, env, desc, STORE_FN,   \
756                     ctzl(sizeof(ETYPE)), GETPC());   \
757 }
758 
759 GEN_VEXT_ST_WHOLE(vs1r_v, int8_t, ste_b)
760 GEN_VEXT_ST_WHOLE(vs2r_v, int8_t, ste_b)
761 GEN_VEXT_ST_WHOLE(vs4r_v, int8_t, ste_b)
762 GEN_VEXT_ST_WHOLE(vs8r_v, int8_t, ste_b)
763 
764 /*
765  *** Vector Integer Arithmetic Instructions
766  */
767 
768 /* expand macro args before macro */
769 #define RVVCALL(macro, ...)  macro(__VA_ARGS__)
770 
771 /* (TD, T1, T2, TX1, TX2) */
772 #define OP_SSS_B int8_t, int8_t, int8_t, int8_t, int8_t
773 #define OP_SSS_H int16_t, int16_t, int16_t, int16_t, int16_t
774 #define OP_SSS_W int32_t, int32_t, int32_t, int32_t, int32_t
775 #define OP_SSS_D int64_t, int64_t, int64_t, int64_t, int64_t
776 #define OP_UUU_B uint8_t, uint8_t, uint8_t, uint8_t, uint8_t
777 #define OP_UUU_H uint16_t, uint16_t, uint16_t, uint16_t, uint16_t
778 #define OP_UUU_W uint32_t, uint32_t, uint32_t, uint32_t, uint32_t
779 #define OP_UUU_D uint64_t, uint64_t, uint64_t, uint64_t, uint64_t
780 #define OP_SUS_B int8_t, uint8_t, int8_t, uint8_t, int8_t
781 #define OP_SUS_H int16_t, uint16_t, int16_t, uint16_t, int16_t
782 #define OP_SUS_W int32_t, uint32_t, int32_t, uint32_t, int32_t
783 #define OP_SUS_D int64_t, uint64_t, int64_t, uint64_t, int64_t
784 #define WOP_UUU_B uint16_t, uint8_t, uint8_t, uint16_t, uint16_t
785 #define WOP_UUU_H uint32_t, uint16_t, uint16_t, uint32_t, uint32_t
786 #define WOP_UUU_W uint64_t, uint32_t, uint32_t, uint64_t, uint64_t
787 #define WOP_SSS_B int16_t, int8_t, int8_t, int16_t, int16_t
788 #define WOP_SSS_H int32_t, int16_t, int16_t, int32_t, int32_t
789 #define WOP_SSS_W int64_t, int32_t, int32_t, int64_t, int64_t
790 #define WOP_SUS_B int16_t, uint8_t, int8_t, uint16_t, int16_t
791 #define WOP_SUS_H int32_t, uint16_t, int16_t, uint32_t, int32_t
792 #define WOP_SUS_W int64_t, uint32_t, int32_t, uint64_t, int64_t
793 #define WOP_SSU_B int16_t, int8_t, uint8_t, int16_t, uint16_t
794 #define WOP_SSU_H int32_t, int16_t, uint16_t, int32_t, uint32_t
795 #define WOP_SSU_W int64_t, int32_t, uint32_t, int64_t, uint64_t
796 #define NOP_SSS_B int8_t, int8_t, int16_t, int8_t, int16_t
797 #define NOP_SSS_H int16_t, int16_t, int32_t, int16_t, int32_t
798 #define NOP_SSS_W int32_t, int32_t, int64_t, int32_t, int64_t
799 #define NOP_UUU_B uint8_t, uint8_t, uint16_t, uint8_t, uint16_t
800 #define NOP_UUU_H uint16_t, uint16_t, uint32_t, uint16_t, uint32_t
801 #define NOP_UUU_W uint32_t, uint32_t, uint64_t, uint32_t, uint64_t
802 
803 /* operation of two vector elements */
804 typedef void opivv2_fn(void *vd, void *vs1, void *vs2, int i);
805 
806 #define OPIVV2(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)    \
807 static void do_##NAME(void *vd, void *vs1, void *vs2, int i)    \
808 {                                                               \
809     TX1 s1 = *((T1 *)vs1 + HS1(i));                             \
810     TX2 s2 = *((T2 *)vs2 + HS2(i));                             \
811     *((TD *)vd + HD(i)) = OP(s2, s1);                           \
812 }
813 #define DO_SUB(N, M) (N - M)
814 #define DO_RSUB(N, M) (M - N)
815 
816 RVVCALL(OPIVV2, vadd_vv_b, OP_SSS_B, H1, H1, H1, DO_ADD)
817 RVVCALL(OPIVV2, vadd_vv_h, OP_SSS_H, H2, H2, H2, DO_ADD)
818 RVVCALL(OPIVV2, vadd_vv_w, OP_SSS_W, H4, H4, H4, DO_ADD)
819 RVVCALL(OPIVV2, vadd_vv_d, OP_SSS_D, H8, H8, H8, DO_ADD)
820 RVVCALL(OPIVV2, vsub_vv_b, OP_SSS_B, H1, H1, H1, DO_SUB)
821 RVVCALL(OPIVV2, vsub_vv_h, OP_SSS_H, H2, H2, H2, DO_SUB)
822 RVVCALL(OPIVV2, vsub_vv_w, OP_SSS_W, H4, H4, H4, DO_SUB)
823 RVVCALL(OPIVV2, vsub_vv_d, OP_SSS_D, H8, H8, H8, DO_SUB)
824 
825 static void do_vext_vv(void *vd, void *v0, void *vs1, void *vs2,
826                        CPURISCVState *env, uint32_t desc,
827                        opivv2_fn *fn, uint32_t esz)
828 {
829     uint32_t vm = vext_vm(desc);
830     uint32_t vl = env->vl;
831     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
832     uint32_t vta = vext_vta(desc);
833     uint32_t vma = vext_vma(desc);
834     uint32_t i;
835 
836     for (i = env->vstart; i < vl; i++) {
837         if (!vm && !vext_elem_mask(v0, i)) {
838             /* set masked-off elements to 1s */
839             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);
840             continue;
841         }
842         fn(vd, vs1, vs2, i);
843     }
844     env->vstart = 0;
845     /* set tail elements to 1s */
846     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);
847 }
848 
849 /* generate the helpers for OPIVV */
850 #define GEN_VEXT_VV(NAME, ESZ)                            \
851 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
852                   void *vs2, CPURISCVState *env,          \
853                   uint32_t desc)                          \
854 {                                                         \
855     do_vext_vv(vd, v0, vs1, vs2, env, desc,               \
856                do_##NAME, ESZ);                           \
857 }
858 
859 GEN_VEXT_VV(vadd_vv_b, 1)
860 GEN_VEXT_VV(vadd_vv_h, 2)
861 GEN_VEXT_VV(vadd_vv_w, 4)
862 GEN_VEXT_VV(vadd_vv_d, 8)
863 GEN_VEXT_VV(vsub_vv_b, 1)
864 GEN_VEXT_VV(vsub_vv_h, 2)
865 GEN_VEXT_VV(vsub_vv_w, 4)
866 GEN_VEXT_VV(vsub_vv_d, 8)
867 
868 typedef void opivx2_fn(void *vd, target_long s1, void *vs2, int i);
869 
870 /*
871  * (T1)s1 gives the real operator type.
872  * (TX1)(T1)s1 expands the operator type of widen or narrow operations.
873  */
874 #define OPIVX2(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)             \
875 static void do_##NAME(void *vd, target_long s1, void *vs2, int i)   \
876 {                                                                   \
877     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
878     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1);                      \
879 }
880 
881 RVVCALL(OPIVX2, vadd_vx_b, OP_SSS_B, H1, H1, DO_ADD)
882 RVVCALL(OPIVX2, vadd_vx_h, OP_SSS_H, H2, H2, DO_ADD)
883 RVVCALL(OPIVX2, vadd_vx_w, OP_SSS_W, H4, H4, DO_ADD)
884 RVVCALL(OPIVX2, vadd_vx_d, OP_SSS_D, H8, H8, DO_ADD)
885 RVVCALL(OPIVX2, vsub_vx_b, OP_SSS_B, H1, H1, DO_SUB)
886 RVVCALL(OPIVX2, vsub_vx_h, OP_SSS_H, H2, H2, DO_SUB)
887 RVVCALL(OPIVX2, vsub_vx_w, OP_SSS_W, H4, H4, DO_SUB)
888 RVVCALL(OPIVX2, vsub_vx_d, OP_SSS_D, H8, H8, DO_SUB)
889 RVVCALL(OPIVX2, vrsub_vx_b, OP_SSS_B, H1, H1, DO_RSUB)
890 RVVCALL(OPIVX2, vrsub_vx_h, OP_SSS_H, H2, H2, DO_RSUB)
891 RVVCALL(OPIVX2, vrsub_vx_w, OP_SSS_W, H4, H4, DO_RSUB)
892 RVVCALL(OPIVX2, vrsub_vx_d, OP_SSS_D, H8, H8, DO_RSUB)
893 
894 static void do_vext_vx(void *vd, void *v0, target_long s1, void *vs2,
895                        CPURISCVState *env, uint32_t desc,
896                        opivx2_fn fn, uint32_t esz)
897 {
898     uint32_t vm = vext_vm(desc);
899     uint32_t vl = env->vl;
900     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
901     uint32_t vta = vext_vta(desc);
902     uint32_t vma = vext_vma(desc);
903     uint32_t i;
904 
905     for (i = env->vstart; i < vl; i++) {
906         if (!vm && !vext_elem_mask(v0, i)) {
907             /* set masked-off elements to 1s */
908             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);
909             continue;
910         }
911         fn(vd, s1, vs2, i);
912     }
913     env->vstart = 0;
914     /* set tail elements to 1s */
915     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);
916 }
917 
918 /* generate the helpers for OPIVX */
919 #define GEN_VEXT_VX(NAME, ESZ)                            \
920 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,    \
921                   void *vs2, CPURISCVState *env,          \
922                   uint32_t desc)                          \
923 {                                                         \
924     do_vext_vx(vd, v0, s1, vs2, env, desc,                \
925                do_##NAME, ESZ);                           \
926 }
927 
928 GEN_VEXT_VX(vadd_vx_b, 1)
929 GEN_VEXT_VX(vadd_vx_h, 2)
930 GEN_VEXT_VX(vadd_vx_w, 4)
931 GEN_VEXT_VX(vadd_vx_d, 8)
932 GEN_VEXT_VX(vsub_vx_b, 1)
933 GEN_VEXT_VX(vsub_vx_h, 2)
934 GEN_VEXT_VX(vsub_vx_w, 4)
935 GEN_VEXT_VX(vsub_vx_d, 8)
936 GEN_VEXT_VX(vrsub_vx_b, 1)
937 GEN_VEXT_VX(vrsub_vx_h, 2)
938 GEN_VEXT_VX(vrsub_vx_w, 4)
939 GEN_VEXT_VX(vrsub_vx_d, 8)
940 
941 void HELPER(vec_rsubs8)(void *d, void *a, uint64_t b, uint32_t desc)
942 {
943     intptr_t oprsz = simd_oprsz(desc);
944     intptr_t i;
945 
946     for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
947         *(uint8_t *)(d + i) = (uint8_t)b - *(uint8_t *)(a + i);
948     }
949 }
950 
951 void HELPER(vec_rsubs16)(void *d, void *a, uint64_t b, uint32_t desc)
952 {
953     intptr_t oprsz = simd_oprsz(desc);
954     intptr_t i;
955 
956     for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
957         *(uint16_t *)(d + i) = (uint16_t)b - *(uint16_t *)(a + i);
958     }
959 }
960 
961 void HELPER(vec_rsubs32)(void *d, void *a, uint64_t b, uint32_t desc)
962 {
963     intptr_t oprsz = simd_oprsz(desc);
964     intptr_t i;
965 
966     for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
967         *(uint32_t *)(d + i) = (uint32_t)b - *(uint32_t *)(a + i);
968     }
969 }
970 
971 void HELPER(vec_rsubs64)(void *d, void *a, uint64_t b, uint32_t desc)
972 {
973     intptr_t oprsz = simd_oprsz(desc);
974     intptr_t i;
975 
976     for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
977         *(uint64_t *)(d + i) = b - *(uint64_t *)(a + i);
978     }
979 }
980 
981 /* Vector Widening Integer Add/Subtract */
982 #define WOP_UUU_B uint16_t, uint8_t, uint8_t, uint16_t, uint16_t
983 #define WOP_UUU_H uint32_t, uint16_t, uint16_t, uint32_t, uint32_t
984 #define WOP_UUU_W uint64_t, uint32_t, uint32_t, uint64_t, uint64_t
985 #define WOP_SSS_B int16_t, int8_t, int8_t, int16_t, int16_t
986 #define WOP_SSS_H int32_t, int16_t, int16_t, int32_t, int32_t
987 #define WOP_SSS_W int64_t, int32_t, int32_t, int64_t, int64_t
988 #define WOP_WUUU_B  uint16_t, uint8_t, uint16_t, uint16_t, uint16_t
989 #define WOP_WUUU_H  uint32_t, uint16_t, uint32_t, uint32_t, uint32_t
990 #define WOP_WUUU_W  uint64_t, uint32_t, uint64_t, uint64_t, uint64_t
991 #define WOP_WSSS_B  int16_t, int8_t, int16_t, int16_t, int16_t
992 #define WOP_WSSS_H  int32_t, int16_t, int32_t, int32_t, int32_t
993 #define WOP_WSSS_W  int64_t, int32_t, int64_t, int64_t, int64_t
994 RVVCALL(OPIVV2, vwaddu_vv_b, WOP_UUU_B, H2, H1, H1, DO_ADD)
995 RVVCALL(OPIVV2, vwaddu_vv_h, WOP_UUU_H, H4, H2, H2, DO_ADD)
996 RVVCALL(OPIVV2, vwaddu_vv_w, WOP_UUU_W, H8, H4, H4, DO_ADD)
997 RVVCALL(OPIVV2, vwsubu_vv_b, WOP_UUU_B, H2, H1, H1, DO_SUB)
998 RVVCALL(OPIVV2, vwsubu_vv_h, WOP_UUU_H, H4, H2, H2, DO_SUB)
999 RVVCALL(OPIVV2, vwsubu_vv_w, WOP_UUU_W, H8, H4, H4, DO_SUB)
1000 RVVCALL(OPIVV2, vwadd_vv_b, WOP_SSS_B, H2, H1, H1, DO_ADD)
1001 RVVCALL(OPIVV2, vwadd_vv_h, WOP_SSS_H, H4, H2, H2, DO_ADD)
1002 RVVCALL(OPIVV2, vwadd_vv_w, WOP_SSS_W, H8, H4, H4, DO_ADD)
1003 RVVCALL(OPIVV2, vwsub_vv_b, WOP_SSS_B, H2, H1, H1, DO_SUB)
1004 RVVCALL(OPIVV2, vwsub_vv_h, WOP_SSS_H, H4, H2, H2, DO_SUB)
1005 RVVCALL(OPIVV2, vwsub_vv_w, WOP_SSS_W, H8, H4, H4, DO_SUB)
1006 RVVCALL(OPIVV2, vwaddu_wv_b, WOP_WUUU_B, H2, H1, H1, DO_ADD)
1007 RVVCALL(OPIVV2, vwaddu_wv_h, WOP_WUUU_H, H4, H2, H2, DO_ADD)
1008 RVVCALL(OPIVV2, vwaddu_wv_w, WOP_WUUU_W, H8, H4, H4, DO_ADD)
1009 RVVCALL(OPIVV2, vwsubu_wv_b, WOP_WUUU_B, H2, H1, H1, DO_SUB)
1010 RVVCALL(OPIVV2, vwsubu_wv_h, WOP_WUUU_H, H4, H2, H2, DO_SUB)
1011 RVVCALL(OPIVV2, vwsubu_wv_w, WOP_WUUU_W, H8, H4, H4, DO_SUB)
1012 RVVCALL(OPIVV2, vwadd_wv_b, WOP_WSSS_B, H2, H1, H1, DO_ADD)
1013 RVVCALL(OPIVV2, vwadd_wv_h, WOP_WSSS_H, H4, H2, H2, DO_ADD)
1014 RVVCALL(OPIVV2, vwadd_wv_w, WOP_WSSS_W, H8, H4, H4, DO_ADD)
1015 RVVCALL(OPIVV2, vwsub_wv_b, WOP_WSSS_B, H2, H1, H1, DO_SUB)
1016 RVVCALL(OPIVV2, vwsub_wv_h, WOP_WSSS_H, H4, H2, H2, DO_SUB)
1017 RVVCALL(OPIVV2, vwsub_wv_w, WOP_WSSS_W, H8, H4, H4, DO_SUB)
1018 GEN_VEXT_VV(vwaddu_vv_b, 2)
1019 GEN_VEXT_VV(vwaddu_vv_h, 4)
1020 GEN_VEXT_VV(vwaddu_vv_w, 8)
1021 GEN_VEXT_VV(vwsubu_vv_b, 2)
1022 GEN_VEXT_VV(vwsubu_vv_h, 4)
1023 GEN_VEXT_VV(vwsubu_vv_w, 8)
1024 GEN_VEXT_VV(vwadd_vv_b, 2)
1025 GEN_VEXT_VV(vwadd_vv_h, 4)
1026 GEN_VEXT_VV(vwadd_vv_w, 8)
1027 GEN_VEXT_VV(vwsub_vv_b, 2)
1028 GEN_VEXT_VV(vwsub_vv_h, 4)
1029 GEN_VEXT_VV(vwsub_vv_w, 8)
1030 GEN_VEXT_VV(vwaddu_wv_b, 2)
1031 GEN_VEXT_VV(vwaddu_wv_h, 4)
1032 GEN_VEXT_VV(vwaddu_wv_w, 8)
1033 GEN_VEXT_VV(vwsubu_wv_b, 2)
1034 GEN_VEXT_VV(vwsubu_wv_h, 4)
1035 GEN_VEXT_VV(vwsubu_wv_w, 8)
1036 GEN_VEXT_VV(vwadd_wv_b, 2)
1037 GEN_VEXT_VV(vwadd_wv_h, 4)
1038 GEN_VEXT_VV(vwadd_wv_w, 8)
1039 GEN_VEXT_VV(vwsub_wv_b, 2)
1040 GEN_VEXT_VV(vwsub_wv_h, 4)
1041 GEN_VEXT_VV(vwsub_wv_w, 8)
1042 
1043 RVVCALL(OPIVX2, vwaddu_vx_b, WOP_UUU_B, H2, H1, DO_ADD)
1044 RVVCALL(OPIVX2, vwaddu_vx_h, WOP_UUU_H, H4, H2, DO_ADD)
1045 RVVCALL(OPIVX2, vwaddu_vx_w, WOP_UUU_W, H8, H4, DO_ADD)
1046 RVVCALL(OPIVX2, vwsubu_vx_b, WOP_UUU_B, H2, H1, DO_SUB)
1047 RVVCALL(OPIVX2, vwsubu_vx_h, WOP_UUU_H, H4, H2, DO_SUB)
1048 RVVCALL(OPIVX2, vwsubu_vx_w, WOP_UUU_W, H8, H4, DO_SUB)
1049 RVVCALL(OPIVX2, vwadd_vx_b, WOP_SSS_B, H2, H1, DO_ADD)
1050 RVVCALL(OPIVX2, vwadd_vx_h, WOP_SSS_H, H4, H2, DO_ADD)
1051 RVVCALL(OPIVX2, vwadd_vx_w, WOP_SSS_W, H8, H4, DO_ADD)
1052 RVVCALL(OPIVX2, vwsub_vx_b, WOP_SSS_B, H2, H1, DO_SUB)
1053 RVVCALL(OPIVX2, vwsub_vx_h, WOP_SSS_H, H4, H2, DO_SUB)
1054 RVVCALL(OPIVX2, vwsub_vx_w, WOP_SSS_W, H8, H4, DO_SUB)
1055 RVVCALL(OPIVX2, vwaddu_wx_b, WOP_WUUU_B, H2, H1, DO_ADD)
1056 RVVCALL(OPIVX2, vwaddu_wx_h, WOP_WUUU_H, H4, H2, DO_ADD)
1057 RVVCALL(OPIVX2, vwaddu_wx_w, WOP_WUUU_W, H8, H4, DO_ADD)
1058 RVVCALL(OPIVX2, vwsubu_wx_b, WOP_WUUU_B, H2, H1, DO_SUB)
1059 RVVCALL(OPIVX2, vwsubu_wx_h, WOP_WUUU_H, H4, H2, DO_SUB)
1060 RVVCALL(OPIVX2, vwsubu_wx_w, WOP_WUUU_W, H8, H4, DO_SUB)
1061 RVVCALL(OPIVX2, vwadd_wx_b, WOP_WSSS_B, H2, H1, DO_ADD)
1062 RVVCALL(OPIVX2, vwadd_wx_h, WOP_WSSS_H, H4, H2, DO_ADD)
1063 RVVCALL(OPIVX2, vwadd_wx_w, WOP_WSSS_W, H8, H4, DO_ADD)
1064 RVVCALL(OPIVX2, vwsub_wx_b, WOP_WSSS_B, H2, H1, DO_SUB)
1065 RVVCALL(OPIVX2, vwsub_wx_h, WOP_WSSS_H, H4, H2, DO_SUB)
1066 RVVCALL(OPIVX2, vwsub_wx_w, WOP_WSSS_W, H8, H4, DO_SUB)
1067 GEN_VEXT_VX(vwaddu_vx_b, 2)
1068 GEN_VEXT_VX(vwaddu_vx_h, 4)
1069 GEN_VEXT_VX(vwaddu_vx_w, 8)
1070 GEN_VEXT_VX(vwsubu_vx_b, 2)
1071 GEN_VEXT_VX(vwsubu_vx_h, 4)
1072 GEN_VEXT_VX(vwsubu_vx_w, 8)
1073 GEN_VEXT_VX(vwadd_vx_b, 2)
1074 GEN_VEXT_VX(vwadd_vx_h, 4)
1075 GEN_VEXT_VX(vwadd_vx_w, 8)
1076 GEN_VEXT_VX(vwsub_vx_b, 2)
1077 GEN_VEXT_VX(vwsub_vx_h, 4)
1078 GEN_VEXT_VX(vwsub_vx_w, 8)
1079 GEN_VEXT_VX(vwaddu_wx_b, 2)
1080 GEN_VEXT_VX(vwaddu_wx_h, 4)
1081 GEN_VEXT_VX(vwaddu_wx_w, 8)
1082 GEN_VEXT_VX(vwsubu_wx_b, 2)
1083 GEN_VEXT_VX(vwsubu_wx_h, 4)
1084 GEN_VEXT_VX(vwsubu_wx_w, 8)
1085 GEN_VEXT_VX(vwadd_wx_b, 2)
1086 GEN_VEXT_VX(vwadd_wx_h, 4)
1087 GEN_VEXT_VX(vwadd_wx_w, 8)
1088 GEN_VEXT_VX(vwsub_wx_b, 2)
1089 GEN_VEXT_VX(vwsub_wx_h, 4)
1090 GEN_VEXT_VX(vwsub_wx_w, 8)
1091 
1092 /* Vector Integer Add-with-Carry / Subtract-with-Borrow Instructions */
1093 #define DO_VADC(N, M, C) (N + M + C)
1094 #define DO_VSBC(N, M, C) (N - M - C)
1095 
1096 #define GEN_VEXT_VADC_VVM(NAME, ETYPE, H, DO_OP)              \
1097 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
1098                   CPURISCVState *env, uint32_t desc)          \
1099 {                                                             \
1100     uint32_t vl = env->vl;                                    \
1101     uint32_t esz = sizeof(ETYPE);                             \
1102     uint32_t total_elems =                                    \
1103         vext_get_total_elems(env, desc, esz);                 \
1104     uint32_t vta = vext_vta(desc);                            \
1105     uint32_t i;                                               \
1106                                                               \
1107     for (i = env->vstart; i < vl; i++) {                      \
1108         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
1109         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
1110         ETYPE carry = vext_elem_mask(v0, i);                  \
1111                                                               \
1112         *((ETYPE *)vd + H(i)) = DO_OP(s2, s1, carry);         \
1113     }                                                         \
1114     env->vstart = 0;                                          \
1115     /* set tail elements to 1s */                             \
1116     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);  \
1117 }
1118 
1119 GEN_VEXT_VADC_VVM(vadc_vvm_b, uint8_t,  H1, DO_VADC)
1120 GEN_VEXT_VADC_VVM(vadc_vvm_h, uint16_t, H2, DO_VADC)
1121 GEN_VEXT_VADC_VVM(vadc_vvm_w, uint32_t, H4, DO_VADC)
1122 GEN_VEXT_VADC_VVM(vadc_vvm_d, uint64_t, H8, DO_VADC)
1123 
1124 GEN_VEXT_VADC_VVM(vsbc_vvm_b, uint8_t,  H1, DO_VSBC)
1125 GEN_VEXT_VADC_VVM(vsbc_vvm_h, uint16_t, H2, DO_VSBC)
1126 GEN_VEXT_VADC_VVM(vsbc_vvm_w, uint32_t, H4, DO_VSBC)
1127 GEN_VEXT_VADC_VVM(vsbc_vvm_d, uint64_t, H8, DO_VSBC)
1128 
1129 #define GEN_VEXT_VADC_VXM(NAME, ETYPE, H, DO_OP)                         \
1130 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,        \
1131                   CPURISCVState *env, uint32_t desc)                     \
1132 {                                                                        \
1133     uint32_t vl = env->vl;                                               \
1134     uint32_t esz = sizeof(ETYPE);                                        \
1135     uint32_t total_elems = vext_get_total_elems(env, desc, esz);         \
1136     uint32_t vta = vext_vta(desc);                                       \
1137     uint32_t i;                                                          \
1138                                                                          \
1139     for (i = env->vstart; i < vl; i++) {                                 \
1140         ETYPE s2 = *((ETYPE *)vs2 + H(i));                               \
1141         ETYPE carry = vext_elem_mask(v0, i);                             \
1142                                                                          \
1143         *((ETYPE *)vd + H(i)) = DO_OP(s2, (ETYPE)(target_long)s1, carry);\
1144     }                                                                    \
1145     env->vstart = 0;                                          \
1146     /* set tail elements to 1s */                                        \
1147     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);             \
1148 }
1149 
1150 GEN_VEXT_VADC_VXM(vadc_vxm_b, uint8_t,  H1, DO_VADC)
1151 GEN_VEXT_VADC_VXM(vadc_vxm_h, uint16_t, H2, DO_VADC)
1152 GEN_VEXT_VADC_VXM(vadc_vxm_w, uint32_t, H4, DO_VADC)
1153 GEN_VEXT_VADC_VXM(vadc_vxm_d, uint64_t, H8, DO_VADC)
1154 
1155 GEN_VEXT_VADC_VXM(vsbc_vxm_b, uint8_t,  H1, DO_VSBC)
1156 GEN_VEXT_VADC_VXM(vsbc_vxm_h, uint16_t, H2, DO_VSBC)
1157 GEN_VEXT_VADC_VXM(vsbc_vxm_w, uint32_t, H4, DO_VSBC)
1158 GEN_VEXT_VADC_VXM(vsbc_vxm_d, uint64_t, H8, DO_VSBC)
1159 
1160 #define DO_MADC(N, M, C) (C ? (__typeof(N))(N + M + 1) <= N :           \
1161                           (__typeof(N))(N + M) < N)
1162 #define DO_MSBC(N, M, C) (C ? N <= M : N < M)
1163 
1164 #define GEN_VEXT_VMADC_VVM(NAME, ETYPE, H, DO_OP)             \
1165 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
1166                   CPURISCVState *env, uint32_t desc)          \
1167 {                                                             \
1168     uint32_t vl = env->vl;                                    \
1169     uint32_t vm = vext_vm(desc);                              \
1170     uint32_t total_elems = env_archcpu(env)->cfg.vlen;        \
1171     uint32_t vta_all_1s = vext_vta_all_1s(desc);              \
1172     uint32_t i;                                               \
1173                                                               \
1174     for (i = env->vstart; i < vl; i++) {                      \
1175         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
1176         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
1177         ETYPE carry = !vm && vext_elem_mask(v0, i);           \
1178         vext_set_elem_mask(vd, i, DO_OP(s2, s1, carry));      \
1179     }                                                         \
1180     env->vstart = 0;                                          \
1181     /* mask destination register are always tail-agnostic */  \
1182     /* set tail elements to 1s */                             \
1183     if (vta_all_1s) {                                         \
1184         for (; i < total_elems; i++) {                        \
1185             vext_set_elem_mask(vd, i, 1);                     \
1186         }                                                     \
1187     }                                                         \
1188 }
1189 
1190 GEN_VEXT_VMADC_VVM(vmadc_vvm_b, uint8_t,  H1, DO_MADC)
1191 GEN_VEXT_VMADC_VVM(vmadc_vvm_h, uint16_t, H2, DO_MADC)
1192 GEN_VEXT_VMADC_VVM(vmadc_vvm_w, uint32_t, H4, DO_MADC)
1193 GEN_VEXT_VMADC_VVM(vmadc_vvm_d, uint64_t, H8, DO_MADC)
1194 
1195 GEN_VEXT_VMADC_VVM(vmsbc_vvm_b, uint8_t,  H1, DO_MSBC)
1196 GEN_VEXT_VMADC_VVM(vmsbc_vvm_h, uint16_t, H2, DO_MSBC)
1197 GEN_VEXT_VMADC_VVM(vmsbc_vvm_w, uint32_t, H4, DO_MSBC)
1198 GEN_VEXT_VMADC_VVM(vmsbc_vvm_d, uint64_t, H8, DO_MSBC)
1199 
1200 #define GEN_VEXT_VMADC_VXM(NAME, ETYPE, H, DO_OP)               \
1201 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,          \
1202                   void *vs2, CPURISCVState *env, uint32_t desc) \
1203 {                                                               \
1204     uint32_t vl = env->vl;                                      \
1205     uint32_t vm = vext_vm(desc);                                \
1206     uint32_t total_elems = env_archcpu(env)->cfg.vlen;          \
1207     uint32_t vta_all_1s = vext_vta_all_1s(desc);                \
1208     uint32_t i;                                                 \
1209                                                                 \
1210     for (i = env->vstart; i < vl; i++) {                        \
1211         ETYPE s2 = *((ETYPE *)vs2 + H(i));                      \
1212         ETYPE carry = !vm && vext_elem_mask(v0, i);             \
1213         vext_set_elem_mask(vd, i,                               \
1214                 DO_OP(s2, (ETYPE)(target_long)s1, carry));      \
1215     }                                                           \
1216     env->vstart = 0;                                            \
1217     /* mask destination register are always tail-agnostic */    \
1218     /* set tail elements to 1s */                               \
1219     if (vta_all_1s) {                                           \
1220         for (; i < total_elems; i++) {                          \
1221             vext_set_elem_mask(vd, i, 1);                       \
1222         }                                                       \
1223     }                                                           \
1224 }
1225 
1226 GEN_VEXT_VMADC_VXM(vmadc_vxm_b, uint8_t,  H1, DO_MADC)
1227 GEN_VEXT_VMADC_VXM(vmadc_vxm_h, uint16_t, H2, DO_MADC)
1228 GEN_VEXT_VMADC_VXM(vmadc_vxm_w, uint32_t, H4, DO_MADC)
1229 GEN_VEXT_VMADC_VXM(vmadc_vxm_d, uint64_t, H8, DO_MADC)
1230 
1231 GEN_VEXT_VMADC_VXM(vmsbc_vxm_b, uint8_t,  H1, DO_MSBC)
1232 GEN_VEXT_VMADC_VXM(vmsbc_vxm_h, uint16_t, H2, DO_MSBC)
1233 GEN_VEXT_VMADC_VXM(vmsbc_vxm_w, uint32_t, H4, DO_MSBC)
1234 GEN_VEXT_VMADC_VXM(vmsbc_vxm_d, uint64_t, H8, DO_MSBC)
1235 
1236 /* Vector Bitwise Logical Instructions */
1237 RVVCALL(OPIVV2, vand_vv_b, OP_SSS_B, H1, H1, H1, DO_AND)
1238 RVVCALL(OPIVV2, vand_vv_h, OP_SSS_H, H2, H2, H2, DO_AND)
1239 RVVCALL(OPIVV2, vand_vv_w, OP_SSS_W, H4, H4, H4, DO_AND)
1240 RVVCALL(OPIVV2, vand_vv_d, OP_SSS_D, H8, H8, H8, DO_AND)
1241 RVVCALL(OPIVV2, vor_vv_b, OP_SSS_B, H1, H1, H1, DO_OR)
1242 RVVCALL(OPIVV2, vor_vv_h, OP_SSS_H, H2, H2, H2, DO_OR)
1243 RVVCALL(OPIVV2, vor_vv_w, OP_SSS_W, H4, H4, H4, DO_OR)
1244 RVVCALL(OPIVV2, vor_vv_d, OP_SSS_D, H8, H8, H8, DO_OR)
1245 RVVCALL(OPIVV2, vxor_vv_b, OP_SSS_B, H1, H1, H1, DO_XOR)
1246 RVVCALL(OPIVV2, vxor_vv_h, OP_SSS_H, H2, H2, H2, DO_XOR)
1247 RVVCALL(OPIVV2, vxor_vv_w, OP_SSS_W, H4, H4, H4, DO_XOR)
1248 RVVCALL(OPIVV2, vxor_vv_d, OP_SSS_D, H8, H8, H8, DO_XOR)
1249 GEN_VEXT_VV(vand_vv_b, 1)
1250 GEN_VEXT_VV(vand_vv_h, 2)
1251 GEN_VEXT_VV(vand_vv_w, 4)
1252 GEN_VEXT_VV(vand_vv_d, 8)
1253 GEN_VEXT_VV(vor_vv_b, 1)
1254 GEN_VEXT_VV(vor_vv_h, 2)
1255 GEN_VEXT_VV(vor_vv_w, 4)
1256 GEN_VEXT_VV(vor_vv_d, 8)
1257 GEN_VEXT_VV(vxor_vv_b, 1)
1258 GEN_VEXT_VV(vxor_vv_h, 2)
1259 GEN_VEXT_VV(vxor_vv_w, 4)
1260 GEN_VEXT_VV(vxor_vv_d, 8)
1261 
1262 RVVCALL(OPIVX2, vand_vx_b, OP_SSS_B, H1, H1, DO_AND)
1263 RVVCALL(OPIVX2, vand_vx_h, OP_SSS_H, H2, H2, DO_AND)
1264 RVVCALL(OPIVX2, vand_vx_w, OP_SSS_W, H4, H4, DO_AND)
1265 RVVCALL(OPIVX2, vand_vx_d, OP_SSS_D, H8, H8, DO_AND)
1266 RVVCALL(OPIVX2, vor_vx_b, OP_SSS_B, H1, H1, DO_OR)
1267 RVVCALL(OPIVX2, vor_vx_h, OP_SSS_H, H2, H2, DO_OR)
1268 RVVCALL(OPIVX2, vor_vx_w, OP_SSS_W, H4, H4, DO_OR)
1269 RVVCALL(OPIVX2, vor_vx_d, OP_SSS_D, H8, H8, DO_OR)
1270 RVVCALL(OPIVX2, vxor_vx_b, OP_SSS_B, H1, H1, DO_XOR)
1271 RVVCALL(OPIVX2, vxor_vx_h, OP_SSS_H, H2, H2, DO_XOR)
1272 RVVCALL(OPIVX2, vxor_vx_w, OP_SSS_W, H4, H4, DO_XOR)
1273 RVVCALL(OPIVX2, vxor_vx_d, OP_SSS_D, H8, H8, DO_XOR)
1274 GEN_VEXT_VX(vand_vx_b, 1)
1275 GEN_VEXT_VX(vand_vx_h, 2)
1276 GEN_VEXT_VX(vand_vx_w, 4)
1277 GEN_VEXT_VX(vand_vx_d, 8)
1278 GEN_VEXT_VX(vor_vx_b, 1)
1279 GEN_VEXT_VX(vor_vx_h, 2)
1280 GEN_VEXT_VX(vor_vx_w, 4)
1281 GEN_VEXT_VX(vor_vx_d, 8)
1282 GEN_VEXT_VX(vxor_vx_b, 1)
1283 GEN_VEXT_VX(vxor_vx_h, 2)
1284 GEN_VEXT_VX(vxor_vx_w, 4)
1285 GEN_VEXT_VX(vxor_vx_d, 8)
1286 
1287 /* Vector Single-Width Bit Shift Instructions */
1288 #define DO_SLL(N, M)  (N << (M))
1289 #define DO_SRL(N, M)  (N >> (M))
1290 
1291 /* generate the helpers for shift instructions with two vector operators */
1292 #define GEN_VEXT_SHIFT_VV(NAME, TS1, TS2, HS1, HS2, OP, MASK)             \
1293 void HELPER(NAME)(void *vd, void *v0, void *vs1,                          \
1294                   void *vs2, CPURISCVState *env, uint32_t desc)           \
1295 {                                                                         \
1296     uint32_t vm = vext_vm(desc);                                          \
1297     uint32_t vl = env->vl;                                                \
1298     uint32_t esz = sizeof(TS1);                                           \
1299     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
1300     uint32_t vta = vext_vta(desc);                                        \
1301     uint32_t vma = vext_vma(desc);                                        \
1302     uint32_t i;                                                           \
1303                                                                           \
1304     for (i = env->vstart; i < vl; i++) {                                  \
1305         if (!vm && !vext_elem_mask(v0, i)) {                              \
1306             /* set masked-off elements to 1s */                           \
1307             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
1308             continue;                                                     \
1309         }                                                                 \
1310         TS1 s1 = *((TS1 *)vs1 + HS1(i));                                  \
1311         TS2 s2 = *((TS2 *)vs2 + HS2(i));                                  \
1312         *((TS1 *)vd + HS1(i)) = OP(s2, s1 & MASK);                        \
1313     }                                                                     \
1314     env->vstart = 0;                                                      \
1315     /* set tail elements to 1s */                                         \
1316     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
1317 }
1318 
1319 GEN_VEXT_SHIFT_VV(vsll_vv_b, uint8_t,  uint8_t, H1, H1, DO_SLL, 0x7)
1320 GEN_VEXT_SHIFT_VV(vsll_vv_h, uint16_t, uint16_t, H2, H2, DO_SLL, 0xf)
1321 GEN_VEXT_SHIFT_VV(vsll_vv_w, uint32_t, uint32_t, H4, H4, DO_SLL, 0x1f)
1322 GEN_VEXT_SHIFT_VV(vsll_vv_d, uint64_t, uint64_t, H8, H8, DO_SLL, 0x3f)
1323 
1324 GEN_VEXT_SHIFT_VV(vsrl_vv_b, uint8_t, uint8_t, H1, H1, DO_SRL, 0x7)
1325 GEN_VEXT_SHIFT_VV(vsrl_vv_h, uint16_t, uint16_t, H2, H2, DO_SRL, 0xf)
1326 GEN_VEXT_SHIFT_VV(vsrl_vv_w, uint32_t, uint32_t, H4, H4, DO_SRL, 0x1f)
1327 GEN_VEXT_SHIFT_VV(vsrl_vv_d, uint64_t, uint64_t, H8, H8, DO_SRL, 0x3f)
1328 
1329 GEN_VEXT_SHIFT_VV(vsra_vv_b, uint8_t,  int8_t, H1, H1, DO_SRL, 0x7)
1330 GEN_VEXT_SHIFT_VV(vsra_vv_h, uint16_t, int16_t, H2, H2, DO_SRL, 0xf)
1331 GEN_VEXT_SHIFT_VV(vsra_vv_w, uint32_t, int32_t, H4, H4, DO_SRL, 0x1f)
1332 GEN_VEXT_SHIFT_VV(vsra_vv_d, uint64_t, int64_t, H8, H8, DO_SRL, 0x3f)
1333 
1334 /* generate the helpers for shift instructions with one vector and one scalar */
1335 #define GEN_VEXT_SHIFT_VX(NAME, TD, TS2, HD, HS2, OP, MASK) \
1336 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,      \
1337         void *vs2, CPURISCVState *env, uint32_t desc)       \
1338 {                                                           \
1339     uint32_t vm = vext_vm(desc);                            \
1340     uint32_t vl = env->vl;                                  \
1341     uint32_t esz = sizeof(TD);                              \
1342     uint32_t total_elems =                                  \
1343         vext_get_total_elems(env, desc, esz);               \
1344     uint32_t vta = vext_vta(desc);                          \
1345     uint32_t vma = vext_vma(desc);                          \
1346     uint32_t i;                                             \
1347                                                             \
1348     for (i = env->vstart; i < vl; i++) {                    \
1349         if (!vm && !vext_elem_mask(v0, i)) {                \
1350             /* set masked-off elements to 1s */             \
1351             vext_set_elems_1s(vd, vma, i * esz,             \
1352                               (i + 1) * esz);               \
1353             continue;                                       \
1354         }                                                   \
1355         TS2 s2 = *((TS2 *)vs2 + HS2(i));                    \
1356         *((TD *)vd + HD(i)) = OP(s2, s1 & MASK);            \
1357     }                                                       \
1358     env->vstart = 0;                                        \
1359     /* set tail elements to 1s */                           \
1360     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);\
1361 }
1362 
1363 GEN_VEXT_SHIFT_VX(vsll_vx_b, uint8_t, int8_t, H1, H1, DO_SLL, 0x7)
1364 GEN_VEXT_SHIFT_VX(vsll_vx_h, uint16_t, int16_t, H2, H2, DO_SLL, 0xf)
1365 GEN_VEXT_SHIFT_VX(vsll_vx_w, uint32_t, int32_t, H4, H4, DO_SLL, 0x1f)
1366 GEN_VEXT_SHIFT_VX(vsll_vx_d, uint64_t, int64_t, H8, H8, DO_SLL, 0x3f)
1367 
1368 GEN_VEXT_SHIFT_VX(vsrl_vx_b, uint8_t, uint8_t, H1, H1, DO_SRL, 0x7)
1369 GEN_VEXT_SHIFT_VX(vsrl_vx_h, uint16_t, uint16_t, H2, H2, DO_SRL, 0xf)
1370 GEN_VEXT_SHIFT_VX(vsrl_vx_w, uint32_t, uint32_t, H4, H4, DO_SRL, 0x1f)
1371 GEN_VEXT_SHIFT_VX(vsrl_vx_d, uint64_t, uint64_t, H8, H8, DO_SRL, 0x3f)
1372 
1373 GEN_VEXT_SHIFT_VX(vsra_vx_b, int8_t, int8_t, H1, H1, DO_SRL, 0x7)
1374 GEN_VEXT_SHIFT_VX(vsra_vx_h, int16_t, int16_t, H2, H2, DO_SRL, 0xf)
1375 GEN_VEXT_SHIFT_VX(vsra_vx_w, int32_t, int32_t, H4, H4, DO_SRL, 0x1f)
1376 GEN_VEXT_SHIFT_VX(vsra_vx_d, int64_t, int64_t, H8, H8, DO_SRL, 0x3f)
1377 
1378 /* Vector Narrowing Integer Right Shift Instructions */
1379 GEN_VEXT_SHIFT_VV(vnsrl_wv_b, uint8_t,  uint16_t, H1, H2, DO_SRL, 0xf)
1380 GEN_VEXT_SHIFT_VV(vnsrl_wv_h, uint16_t, uint32_t, H2, H4, DO_SRL, 0x1f)
1381 GEN_VEXT_SHIFT_VV(vnsrl_wv_w, uint32_t, uint64_t, H4, H8, DO_SRL, 0x3f)
1382 GEN_VEXT_SHIFT_VV(vnsra_wv_b, uint8_t,  int16_t, H1, H2, DO_SRL, 0xf)
1383 GEN_VEXT_SHIFT_VV(vnsra_wv_h, uint16_t, int32_t, H2, H4, DO_SRL, 0x1f)
1384 GEN_VEXT_SHIFT_VV(vnsra_wv_w, uint32_t, int64_t, H4, H8, DO_SRL, 0x3f)
1385 GEN_VEXT_SHIFT_VX(vnsrl_wx_b, uint8_t, uint16_t, H1, H2, DO_SRL, 0xf)
1386 GEN_VEXT_SHIFT_VX(vnsrl_wx_h, uint16_t, uint32_t, H2, H4, DO_SRL, 0x1f)
1387 GEN_VEXT_SHIFT_VX(vnsrl_wx_w, uint32_t, uint64_t, H4, H8, DO_SRL, 0x3f)
1388 GEN_VEXT_SHIFT_VX(vnsra_wx_b, int8_t, int16_t, H1, H2, DO_SRL, 0xf)
1389 GEN_VEXT_SHIFT_VX(vnsra_wx_h, int16_t, int32_t, H2, H4, DO_SRL, 0x1f)
1390 GEN_VEXT_SHIFT_VX(vnsra_wx_w, int32_t, int64_t, H4, H8, DO_SRL, 0x3f)
1391 
1392 /* Vector Integer Comparison Instructions */
1393 #define DO_MSEQ(N, M) (N == M)
1394 #define DO_MSNE(N, M) (N != M)
1395 #define DO_MSLT(N, M) (N < M)
1396 #define DO_MSLE(N, M) (N <= M)
1397 #define DO_MSGT(N, M) (N > M)
1398 
1399 #define GEN_VEXT_CMP_VV(NAME, ETYPE, H, DO_OP)                \
1400 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
1401                   CPURISCVState *env, uint32_t desc)          \
1402 {                                                             \
1403     uint32_t vm = vext_vm(desc);                              \
1404     uint32_t vl = env->vl;                                    \
1405     uint32_t total_elems = env_archcpu(env)->cfg.vlen;        \
1406     uint32_t vta_all_1s = vext_vta_all_1s(desc);              \
1407     uint32_t vma = vext_vma(desc);                            \
1408     uint32_t i;                                               \
1409                                                               \
1410     for (i = env->vstart; i < vl; i++) {                      \
1411         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
1412         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
1413         if (!vm && !vext_elem_mask(v0, i)) {                  \
1414             /* set masked-off elements to 1s */               \
1415             if (vma) {                                        \
1416                 vext_set_elem_mask(vd, i, 1);                 \
1417             }                                                 \
1418             continue;                                         \
1419         }                                                     \
1420         vext_set_elem_mask(vd, i, DO_OP(s2, s1));             \
1421     }                                                         \
1422     env->vstart = 0;                                          \
1423     /* mask destination register are always tail-agnostic */  \
1424     /* set tail elements to 1s */                             \
1425     if (vta_all_1s) {                                         \
1426         for (; i < total_elems; i++) {                        \
1427             vext_set_elem_mask(vd, i, 1);                     \
1428         }                                                     \
1429     }                                                         \
1430 }
1431 
1432 GEN_VEXT_CMP_VV(vmseq_vv_b, uint8_t,  H1, DO_MSEQ)
1433 GEN_VEXT_CMP_VV(vmseq_vv_h, uint16_t, H2, DO_MSEQ)
1434 GEN_VEXT_CMP_VV(vmseq_vv_w, uint32_t, H4, DO_MSEQ)
1435 GEN_VEXT_CMP_VV(vmseq_vv_d, uint64_t, H8, DO_MSEQ)
1436 
1437 GEN_VEXT_CMP_VV(vmsne_vv_b, uint8_t,  H1, DO_MSNE)
1438 GEN_VEXT_CMP_VV(vmsne_vv_h, uint16_t, H2, DO_MSNE)
1439 GEN_VEXT_CMP_VV(vmsne_vv_w, uint32_t, H4, DO_MSNE)
1440 GEN_VEXT_CMP_VV(vmsne_vv_d, uint64_t, H8, DO_MSNE)
1441 
1442 GEN_VEXT_CMP_VV(vmsltu_vv_b, uint8_t,  H1, DO_MSLT)
1443 GEN_VEXT_CMP_VV(vmsltu_vv_h, uint16_t, H2, DO_MSLT)
1444 GEN_VEXT_CMP_VV(vmsltu_vv_w, uint32_t, H4, DO_MSLT)
1445 GEN_VEXT_CMP_VV(vmsltu_vv_d, uint64_t, H8, DO_MSLT)
1446 
1447 GEN_VEXT_CMP_VV(vmslt_vv_b, int8_t,  H1, DO_MSLT)
1448 GEN_VEXT_CMP_VV(vmslt_vv_h, int16_t, H2, DO_MSLT)
1449 GEN_VEXT_CMP_VV(vmslt_vv_w, int32_t, H4, DO_MSLT)
1450 GEN_VEXT_CMP_VV(vmslt_vv_d, int64_t, H8, DO_MSLT)
1451 
1452 GEN_VEXT_CMP_VV(vmsleu_vv_b, uint8_t,  H1, DO_MSLE)
1453 GEN_VEXT_CMP_VV(vmsleu_vv_h, uint16_t, H2, DO_MSLE)
1454 GEN_VEXT_CMP_VV(vmsleu_vv_w, uint32_t, H4, DO_MSLE)
1455 GEN_VEXT_CMP_VV(vmsleu_vv_d, uint64_t, H8, DO_MSLE)
1456 
1457 GEN_VEXT_CMP_VV(vmsle_vv_b, int8_t,  H1, DO_MSLE)
1458 GEN_VEXT_CMP_VV(vmsle_vv_h, int16_t, H2, DO_MSLE)
1459 GEN_VEXT_CMP_VV(vmsle_vv_w, int32_t, H4, DO_MSLE)
1460 GEN_VEXT_CMP_VV(vmsle_vv_d, int64_t, H8, DO_MSLE)
1461 
1462 #define GEN_VEXT_CMP_VX(NAME, ETYPE, H, DO_OP)                      \
1463 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,   \
1464                   CPURISCVState *env, uint32_t desc)                \
1465 {                                                                   \
1466     uint32_t vm = vext_vm(desc);                                    \
1467     uint32_t vl = env->vl;                                          \
1468     uint32_t total_elems = env_archcpu(env)->cfg.vlen;              \
1469     uint32_t vta_all_1s = vext_vta_all_1s(desc);                    \
1470     uint32_t vma = vext_vma(desc);                                  \
1471     uint32_t i;                                                     \
1472                                                                     \
1473     for (i = env->vstart; i < vl; i++) {                            \
1474         ETYPE s2 = *((ETYPE *)vs2 + H(i));                          \
1475         if (!vm && !vext_elem_mask(v0, i)) {                        \
1476             /* set masked-off elements to 1s */                     \
1477             if (vma) {                                              \
1478                 vext_set_elem_mask(vd, i, 1);                       \
1479             }                                                       \
1480             continue;                                               \
1481         }                                                           \
1482         vext_set_elem_mask(vd, i,                                   \
1483                 DO_OP(s2, (ETYPE)(target_long)s1));                 \
1484     }                                                               \
1485     env->vstart = 0;                                                \
1486     /* mask destination register are always tail-agnostic */        \
1487     /* set tail elements to 1s */                                   \
1488     if (vta_all_1s) {                                               \
1489         for (; i < total_elems; i++) {                              \
1490             vext_set_elem_mask(vd, i, 1);                           \
1491         }                                                           \
1492     }                                                               \
1493 }
1494 
1495 GEN_VEXT_CMP_VX(vmseq_vx_b, uint8_t,  H1, DO_MSEQ)
1496 GEN_VEXT_CMP_VX(vmseq_vx_h, uint16_t, H2, DO_MSEQ)
1497 GEN_VEXT_CMP_VX(vmseq_vx_w, uint32_t, H4, DO_MSEQ)
1498 GEN_VEXT_CMP_VX(vmseq_vx_d, uint64_t, H8, DO_MSEQ)
1499 
1500 GEN_VEXT_CMP_VX(vmsne_vx_b, uint8_t,  H1, DO_MSNE)
1501 GEN_VEXT_CMP_VX(vmsne_vx_h, uint16_t, H2, DO_MSNE)
1502 GEN_VEXT_CMP_VX(vmsne_vx_w, uint32_t, H4, DO_MSNE)
1503 GEN_VEXT_CMP_VX(vmsne_vx_d, uint64_t, H8, DO_MSNE)
1504 
1505 GEN_VEXT_CMP_VX(vmsltu_vx_b, uint8_t,  H1, DO_MSLT)
1506 GEN_VEXT_CMP_VX(vmsltu_vx_h, uint16_t, H2, DO_MSLT)
1507 GEN_VEXT_CMP_VX(vmsltu_vx_w, uint32_t, H4, DO_MSLT)
1508 GEN_VEXT_CMP_VX(vmsltu_vx_d, uint64_t, H8, DO_MSLT)
1509 
1510 GEN_VEXT_CMP_VX(vmslt_vx_b, int8_t,  H1, DO_MSLT)
1511 GEN_VEXT_CMP_VX(vmslt_vx_h, int16_t, H2, DO_MSLT)
1512 GEN_VEXT_CMP_VX(vmslt_vx_w, int32_t, H4, DO_MSLT)
1513 GEN_VEXT_CMP_VX(vmslt_vx_d, int64_t, H8, DO_MSLT)
1514 
1515 GEN_VEXT_CMP_VX(vmsleu_vx_b, uint8_t,  H1, DO_MSLE)
1516 GEN_VEXT_CMP_VX(vmsleu_vx_h, uint16_t, H2, DO_MSLE)
1517 GEN_VEXT_CMP_VX(vmsleu_vx_w, uint32_t, H4, DO_MSLE)
1518 GEN_VEXT_CMP_VX(vmsleu_vx_d, uint64_t, H8, DO_MSLE)
1519 
1520 GEN_VEXT_CMP_VX(vmsle_vx_b, int8_t,  H1, DO_MSLE)
1521 GEN_VEXT_CMP_VX(vmsle_vx_h, int16_t, H2, DO_MSLE)
1522 GEN_VEXT_CMP_VX(vmsle_vx_w, int32_t, H4, DO_MSLE)
1523 GEN_VEXT_CMP_VX(vmsle_vx_d, int64_t, H8, DO_MSLE)
1524 
1525 GEN_VEXT_CMP_VX(vmsgtu_vx_b, uint8_t,  H1, DO_MSGT)
1526 GEN_VEXT_CMP_VX(vmsgtu_vx_h, uint16_t, H2, DO_MSGT)
1527 GEN_VEXT_CMP_VX(vmsgtu_vx_w, uint32_t, H4, DO_MSGT)
1528 GEN_VEXT_CMP_VX(vmsgtu_vx_d, uint64_t, H8, DO_MSGT)
1529 
1530 GEN_VEXT_CMP_VX(vmsgt_vx_b, int8_t,  H1, DO_MSGT)
1531 GEN_VEXT_CMP_VX(vmsgt_vx_h, int16_t, H2, DO_MSGT)
1532 GEN_VEXT_CMP_VX(vmsgt_vx_w, int32_t, H4, DO_MSGT)
1533 GEN_VEXT_CMP_VX(vmsgt_vx_d, int64_t, H8, DO_MSGT)
1534 
1535 /* Vector Integer Min/Max Instructions */
1536 RVVCALL(OPIVV2, vminu_vv_b, OP_UUU_B, H1, H1, H1, DO_MIN)
1537 RVVCALL(OPIVV2, vminu_vv_h, OP_UUU_H, H2, H2, H2, DO_MIN)
1538 RVVCALL(OPIVV2, vminu_vv_w, OP_UUU_W, H4, H4, H4, DO_MIN)
1539 RVVCALL(OPIVV2, vminu_vv_d, OP_UUU_D, H8, H8, H8, DO_MIN)
1540 RVVCALL(OPIVV2, vmin_vv_b, OP_SSS_B, H1, H1, H1, DO_MIN)
1541 RVVCALL(OPIVV2, vmin_vv_h, OP_SSS_H, H2, H2, H2, DO_MIN)
1542 RVVCALL(OPIVV2, vmin_vv_w, OP_SSS_W, H4, H4, H4, DO_MIN)
1543 RVVCALL(OPIVV2, vmin_vv_d, OP_SSS_D, H8, H8, H8, DO_MIN)
1544 RVVCALL(OPIVV2, vmaxu_vv_b, OP_UUU_B, H1, H1, H1, DO_MAX)
1545 RVVCALL(OPIVV2, vmaxu_vv_h, OP_UUU_H, H2, H2, H2, DO_MAX)
1546 RVVCALL(OPIVV2, vmaxu_vv_w, OP_UUU_W, H4, H4, H4, DO_MAX)
1547 RVVCALL(OPIVV2, vmaxu_vv_d, OP_UUU_D, H8, H8, H8, DO_MAX)
1548 RVVCALL(OPIVV2, vmax_vv_b, OP_SSS_B, H1, H1, H1, DO_MAX)
1549 RVVCALL(OPIVV2, vmax_vv_h, OP_SSS_H, H2, H2, H2, DO_MAX)
1550 RVVCALL(OPIVV2, vmax_vv_w, OP_SSS_W, H4, H4, H4, DO_MAX)
1551 RVVCALL(OPIVV2, vmax_vv_d, OP_SSS_D, H8, H8, H8, DO_MAX)
1552 GEN_VEXT_VV(vminu_vv_b, 1)
1553 GEN_VEXT_VV(vminu_vv_h, 2)
1554 GEN_VEXT_VV(vminu_vv_w, 4)
1555 GEN_VEXT_VV(vminu_vv_d, 8)
1556 GEN_VEXT_VV(vmin_vv_b, 1)
1557 GEN_VEXT_VV(vmin_vv_h, 2)
1558 GEN_VEXT_VV(vmin_vv_w, 4)
1559 GEN_VEXT_VV(vmin_vv_d, 8)
1560 GEN_VEXT_VV(vmaxu_vv_b, 1)
1561 GEN_VEXT_VV(vmaxu_vv_h, 2)
1562 GEN_VEXT_VV(vmaxu_vv_w, 4)
1563 GEN_VEXT_VV(vmaxu_vv_d, 8)
1564 GEN_VEXT_VV(vmax_vv_b, 1)
1565 GEN_VEXT_VV(vmax_vv_h, 2)
1566 GEN_VEXT_VV(vmax_vv_w, 4)
1567 GEN_VEXT_VV(vmax_vv_d, 8)
1568 
1569 RVVCALL(OPIVX2, vminu_vx_b, OP_UUU_B, H1, H1, DO_MIN)
1570 RVVCALL(OPIVX2, vminu_vx_h, OP_UUU_H, H2, H2, DO_MIN)
1571 RVVCALL(OPIVX2, vminu_vx_w, OP_UUU_W, H4, H4, DO_MIN)
1572 RVVCALL(OPIVX2, vminu_vx_d, OP_UUU_D, H8, H8, DO_MIN)
1573 RVVCALL(OPIVX2, vmin_vx_b, OP_SSS_B, H1, H1, DO_MIN)
1574 RVVCALL(OPIVX2, vmin_vx_h, OP_SSS_H, H2, H2, DO_MIN)
1575 RVVCALL(OPIVX2, vmin_vx_w, OP_SSS_W, H4, H4, DO_MIN)
1576 RVVCALL(OPIVX2, vmin_vx_d, OP_SSS_D, H8, H8, DO_MIN)
1577 RVVCALL(OPIVX2, vmaxu_vx_b, OP_UUU_B, H1, H1, DO_MAX)
1578 RVVCALL(OPIVX2, vmaxu_vx_h, OP_UUU_H, H2, H2, DO_MAX)
1579 RVVCALL(OPIVX2, vmaxu_vx_w, OP_UUU_W, H4, H4, DO_MAX)
1580 RVVCALL(OPIVX2, vmaxu_vx_d, OP_UUU_D, H8, H8, DO_MAX)
1581 RVVCALL(OPIVX2, vmax_vx_b, OP_SSS_B, H1, H1, DO_MAX)
1582 RVVCALL(OPIVX2, vmax_vx_h, OP_SSS_H, H2, H2, DO_MAX)
1583 RVVCALL(OPIVX2, vmax_vx_w, OP_SSS_W, H4, H4, DO_MAX)
1584 RVVCALL(OPIVX2, vmax_vx_d, OP_SSS_D, H8, H8, DO_MAX)
1585 GEN_VEXT_VX(vminu_vx_b, 1)
1586 GEN_VEXT_VX(vminu_vx_h, 2)
1587 GEN_VEXT_VX(vminu_vx_w, 4)
1588 GEN_VEXT_VX(vminu_vx_d, 8)
1589 GEN_VEXT_VX(vmin_vx_b, 1)
1590 GEN_VEXT_VX(vmin_vx_h, 2)
1591 GEN_VEXT_VX(vmin_vx_w, 4)
1592 GEN_VEXT_VX(vmin_vx_d, 8)
1593 GEN_VEXT_VX(vmaxu_vx_b, 1)
1594 GEN_VEXT_VX(vmaxu_vx_h, 2)
1595 GEN_VEXT_VX(vmaxu_vx_w, 4)
1596 GEN_VEXT_VX(vmaxu_vx_d, 8)
1597 GEN_VEXT_VX(vmax_vx_b, 1)
1598 GEN_VEXT_VX(vmax_vx_h, 2)
1599 GEN_VEXT_VX(vmax_vx_w, 4)
1600 GEN_VEXT_VX(vmax_vx_d, 8)
1601 
1602 /* Vector Single-Width Integer Multiply Instructions */
1603 #define DO_MUL(N, M) (N * M)
1604 RVVCALL(OPIVV2, vmul_vv_b, OP_SSS_B, H1, H1, H1, DO_MUL)
1605 RVVCALL(OPIVV2, vmul_vv_h, OP_SSS_H, H2, H2, H2, DO_MUL)
1606 RVVCALL(OPIVV2, vmul_vv_w, OP_SSS_W, H4, H4, H4, DO_MUL)
1607 RVVCALL(OPIVV2, vmul_vv_d, OP_SSS_D, H8, H8, H8, DO_MUL)
1608 GEN_VEXT_VV(vmul_vv_b, 1)
1609 GEN_VEXT_VV(vmul_vv_h, 2)
1610 GEN_VEXT_VV(vmul_vv_w, 4)
1611 GEN_VEXT_VV(vmul_vv_d, 8)
1612 
1613 static int8_t do_mulh_b(int8_t s2, int8_t s1)
1614 {
1615     return (int16_t)s2 * (int16_t)s1 >> 8;
1616 }
1617 
1618 static int16_t do_mulh_h(int16_t s2, int16_t s1)
1619 {
1620     return (int32_t)s2 * (int32_t)s1 >> 16;
1621 }
1622 
1623 static int32_t do_mulh_w(int32_t s2, int32_t s1)
1624 {
1625     return (int64_t)s2 * (int64_t)s1 >> 32;
1626 }
1627 
1628 static int64_t do_mulh_d(int64_t s2, int64_t s1)
1629 {
1630     uint64_t hi_64, lo_64;
1631 
1632     muls64(&lo_64, &hi_64, s1, s2);
1633     return hi_64;
1634 }
1635 
1636 static uint8_t do_mulhu_b(uint8_t s2, uint8_t s1)
1637 {
1638     return (uint16_t)s2 * (uint16_t)s1 >> 8;
1639 }
1640 
1641 static uint16_t do_mulhu_h(uint16_t s2, uint16_t s1)
1642 {
1643     return (uint32_t)s2 * (uint32_t)s1 >> 16;
1644 }
1645 
1646 static uint32_t do_mulhu_w(uint32_t s2, uint32_t s1)
1647 {
1648     return (uint64_t)s2 * (uint64_t)s1 >> 32;
1649 }
1650 
1651 static uint64_t do_mulhu_d(uint64_t s2, uint64_t s1)
1652 {
1653     uint64_t hi_64, lo_64;
1654 
1655     mulu64(&lo_64, &hi_64, s2, s1);
1656     return hi_64;
1657 }
1658 
1659 static int8_t do_mulhsu_b(int8_t s2, uint8_t s1)
1660 {
1661     return (int16_t)s2 * (uint16_t)s1 >> 8;
1662 }
1663 
1664 static int16_t do_mulhsu_h(int16_t s2, uint16_t s1)
1665 {
1666     return (int32_t)s2 * (uint32_t)s1 >> 16;
1667 }
1668 
1669 static int32_t do_mulhsu_w(int32_t s2, uint32_t s1)
1670 {
1671     return (int64_t)s2 * (uint64_t)s1 >> 32;
1672 }
1673 
1674 /*
1675  * Let  A = signed operand,
1676  *      B = unsigned operand
1677  *      P = mulu64(A, B), unsigned product
1678  *
1679  * LET  X = 2 ** 64  - A, 2's complement of A
1680  *      SP = signed product
1681  * THEN
1682  *      IF A < 0
1683  *          SP = -X * B
1684  *             = -(2 ** 64 - A) * B
1685  *             = A * B - 2 ** 64 * B
1686  *             = P - 2 ** 64 * B
1687  *      ELSE
1688  *          SP = P
1689  * THEN
1690  *      HI_P -= (A < 0 ? B : 0)
1691  */
1692 
1693 static int64_t do_mulhsu_d(int64_t s2, uint64_t s1)
1694 {
1695     uint64_t hi_64, lo_64;
1696 
1697     mulu64(&lo_64, &hi_64, s2, s1);
1698 
1699     hi_64 -= s2 < 0 ? s1 : 0;
1700     return hi_64;
1701 }
1702 
1703 RVVCALL(OPIVV2, vmulh_vv_b, OP_SSS_B, H1, H1, H1, do_mulh_b)
1704 RVVCALL(OPIVV2, vmulh_vv_h, OP_SSS_H, H2, H2, H2, do_mulh_h)
1705 RVVCALL(OPIVV2, vmulh_vv_w, OP_SSS_W, H4, H4, H4, do_mulh_w)
1706 RVVCALL(OPIVV2, vmulh_vv_d, OP_SSS_D, H8, H8, H8, do_mulh_d)
1707 RVVCALL(OPIVV2, vmulhu_vv_b, OP_UUU_B, H1, H1, H1, do_mulhu_b)
1708 RVVCALL(OPIVV2, vmulhu_vv_h, OP_UUU_H, H2, H2, H2, do_mulhu_h)
1709 RVVCALL(OPIVV2, vmulhu_vv_w, OP_UUU_W, H4, H4, H4, do_mulhu_w)
1710 RVVCALL(OPIVV2, vmulhu_vv_d, OP_UUU_D, H8, H8, H8, do_mulhu_d)
1711 RVVCALL(OPIVV2, vmulhsu_vv_b, OP_SUS_B, H1, H1, H1, do_mulhsu_b)
1712 RVVCALL(OPIVV2, vmulhsu_vv_h, OP_SUS_H, H2, H2, H2, do_mulhsu_h)
1713 RVVCALL(OPIVV2, vmulhsu_vv_w, OP_SUS_W, H4, H4, H4, do_mulhsu_w)
1714 RVVCALL(OPIVV2, vmulhsu_vv_d, OP_SUS_D, H8, H8, H8, do_mulhsu_d)
1715 GEN_VEXT_VV(vmulh_vv_b, 1)
1716 GEN_VEXT_VV(vmulh_vv_h, 2)
1717 GEN_VEXT_VV(vmulh_vv_w, 4)
1718 GEN_VEXT_VV(vmulh_vv_d, 8)
1719 GEN_VEXT_VV(vmulhu_vv_b, 1)
1720 GEN_VEXT_VV(vmulhu_vv_h, 2)
1721 GEN_VEXT_VV(vmulhu_vv_w, 4)
1722 GEN_VEXT_VV(vmulhu_vv_d, 8)
1723 GEN_VEXT_VV(vmulhsu_vv_b, 1)
1724 GEN_VEXT_VV(vmulhsu_vv_h, 2)
1725 GEN_VEXT_VV(vmulhsu_vv_w, 4)
1726 GEN_VEXT_VV(vmulhsu_vv_d, 8)
1727 
1728 RVVCALL(OPIVX2, vmul_vx_b, OP_SSS_B, H1, H1, DO_MUL)
1729 RVVCALL(OPIVX2, vmul_vx_h, OP_SSS_H, H2, H2, DO_MUL)
1730 RVVCALL(OPIVX2, vmul_vx_w, OP_SSS_W, H4, H4, DO_MUL)
1731 RVVCALL(OPIVX2, vmul_vx_d, OP_SSS_D, H8, H8, DO_MUL)
1732 RVVCALL(OPIVX2, vmulh_vx_b, OP_SSS_B, H1, H1, do_mulh_b)
1733 RVVCALL(OPIVX2, vmulh_vx_h, OP_SSS_H, H2, H2, do_mulh_h)
1734 RVVCALL(OPIVX2, vmulh_vx_w, OP_SSS_W, H4, H4, do_mulh_w)
1735 RVVCALL(OPIVX2, vmulh_vx_d, OP_SSS_D, H8, H8, do_mulh_d)
1736 RVVCALL(OPIVX2, vmulhu_vx_b, OP_UUU_B, H1, H1, do_mulhu_b)
1737 RVVCALL(OPIVX2, vmulhu_vx_h, OP_UUU_H, H2, H2, do_mulhu_h)
1738 RVVCALL(OPIVX2, vmulhu_vx_w, OP_UUU_W, H4, H4, do_mulhu_w)
1739 RVVCALL(OPIVX2, vmulhu_vx_d, OP_UUU_D, H8, H8, do_mulhu_d)
1740 RVVCALL(OPIVX2, vmulhsu_vx_b, OP_SUS_B, H1, H1, do_mulhsu_b)
1741 RVVCALL(OPIVX2, vmulhsu_vx_h, OP_SUS_H, H2, H2, do_mulhsu_h)
1742 RVVCALL(OPIVX2, vmulhsu_vx_w, OP_SUS_W, H4, H4, do_mulhsu_w)
1743 RVVCALL(OPIVX2, vmulhsu_vx_d, OP_SUS_D, H8, H8, do_mulhsu_d)
1744 GEN_VEXT_VX(vmul_vx_b, 1)
1745 GEN_VEXT_VX(vmul_vx_h, 2)
1746 GEN_VEXT_VX(vmul_vx_w, 4)
1747 GEN_VEXT_VX(vmul_vx_d, 8)
1748 GEN_VEXT_VX(vmulh_vx_b, 1)
1749 GEN_VEXT_VX(vmulh_vx_h, 2)
1750 GEN_VEXT_VX(vmulh_vx_w, 4)
1751 GEN_VEXT_VX(vmulh_vx_d, 8)
1752 GEN_VEXT_VX(vmulhu_vx_b, 1)
1753 GEN_VEXT_VX(vmulhu_vx_h, 2)
1754 GEN_VEXT_VX(vmulhu_vx_w, 4)
1755 GEN_VEXT_VX(vmulhu_vx_d, 8)
1756 GEN_VEXT_VX(vmulhsu_vx_b, 1)
1757 GEN_VEXT_VX(vmulhsu_vx_h, 2)
1758 GEN_VEXT_VX(vmulhsu_vx_w, 4)
1759 GEN_VEXT_VX(vmulhsu_vx_d, 8)
1760 
1761 /* Vector Integer Divide Instructions */
1762 #define DO_DIVU(N, M) (unlikely(M == 0) ? (__typeof(N))(-1) : N / M)
1763 #define DO_REMU(N, M) (unlikely(M == 0) ? N : N % M)
1764 #define DO_DIV(N, M)  (unlikely(M == 0) ? (__typeof(N))(-1) :\
1765         unlikely((N == -N) && (M == (__typeof(N))(-1))) ? N : N / M)
1766 #define DO_REM(N, M)  (unlikely(M == 0) ? N :\
1767         unlikely((N == -N) && (M == (__typeof(N))(-1))) ? 0 : N % M)
1768 
1769 RVVCALL(OPIVV2, vdivu_vv_b, OP_UUU_B, H1, H1, H1, DO_DIVU)
1770 RVVCALL(OPIVV2, vdivu_vv_h, OP_UUU_H, H2, H2, H2, DO_DIVU)
1771 RVVCALL(OPIVV2, vdivu_vv_w, OP_UUU_W, H4, H4, H4, DO_DIVU)
1772 RVVCALL(OPIVV2, vdivu_vv_d, OP_UUU_D, H8, H8, H8, DO_DIVU)
1773 RVVCALL(OPIVV2, vdiv_vv_b, OP_SSS_B, H1, H1, H1, DO_DIV)
1774 RVVCALL(OPIVV2, vdiv_vv_h, OP_SSS_H, H2, H2, H2, DO_DIV)
1775 RVVCALL(OPIVV2, vdiv_vv_w, OP_SSS_W, H4, H4, H4, DO_DIV)
1776 RVVCALL(OPIVV2, vdiv_vv_d, OP_SSS_D, H8, H8, H8, DO_DIV)
1777 RVVCALL(OPIVV2, vremu_vv_b, OP_UUU_B, H1, H1, H1, DO_REMU)
1778 RVVCALL(OPIVV2, vremu_vv_h, OP_UUU_H, H2, H2, H2, DO_REMU)
1779 RVVCALL(OPIVV2, vremu_vv_w, OP_UUU_W, H4, H4, H4, DO_REMU)
1780 RVVCALL(OPIVV2, vremu_vv_d, OP_UUU_D, H8, H8, H8, DO_REMU)
1781 RVVCALL(OPIVV2, vrem_vv_b, OP_SSS_B, H1, H1, H1, DO_REM)
1782 RVVCALL(OPIVV2, vrem_vv_h, OP_SSS_H, H2, H2, H2, DO_REM)
1783 RVVCALL(OPIVV2, vrem_vv_w, OP_SSS_W, H4, H4, H4, DO_REM)
1784 RVVCALL(OPIVV2, vrem_vv_d, OP_SSS_D, H8, H8, H8, DO_REM)
1785 GEN_VEXT_VV(vdivu_vv_b, 1)
1786 GEN_VEXT_VV(vdivu_vv_h, 2)
1787 GEN_VEXT_VV(vdivu_vv_w, 4)
1788 GEN_VEXT_VV(vdivu_vv_d, 8)
1789 GEN_VEXT_VV(vdiv_vv_b, 1)
1790 GEN_VEXT_VV(vdiv_vv_h, 2)
1791 GEN_VEXT_VV(vdiv_vv_w, 4)
1792 GEN_VEXT_VV(vdiv_vv_d, 8)
1793 GEN_VEXT_VV(vremu_vv_b, 1)
1794 GEN_VEXT_VV(vremu_vv_h, 2)
1795 GEN_VEXT_VV(vremu_vv_w, 4)
1796 GEN_VEXT_VV(vremu_vv_d, 8)
1797 GEN_VEXT_VV(vrem_vv_b, 1)
1798 GEN_VEXT_VV(vrem_vv_h, 2)
1799 GEN_VEXT_VV(vrem_vv_w, 4)
1800 GEN_VEXT_VV(vrem_vv_d, 8)
1801 
1802 RVVCALL(OPIVX2, vdivu_vx_b, OP_UUU_B, H1, H1, DO_DIVU)
1803 RVVCALL(OPIVX2, vdivu_vx_h, OP_UUU_H, H2, H2, DO_DIVU)
1804 RVVCALL(OPIVX2, vdivu_vx_w, OP_UUU_W, H4, H4, DO_DIVU)
1805 RVVCALL(OPIVX2, vdivu_vx_d, OP_UUU_D, H8, H8, DO_DIVU)
1806 RVVCALL(OPIVX2, vdiv_vx_b, OP_SSS_B, H1, H1, DO_DIV)
1807 RVVCALL(OPIVX2, vdiv_vx_h, OP_SSS_H, H2, H2, DO_DIV)
1808 RVVCALL(OPIVX2, vdiv_vx_w, OP_SSS_W, H4, H4, DO_DIV)
1809 RVVCALL(OPIVX2, vdiv_vx_d, OP_SSS_D, H8, H8, DO_DIV)
1810 RVVCALL(OPIVX2, vremu_vx_b, OP_UUU_B, H1, H1, DO_REMU)
1811 RVVCALL(OPIVX2, vremu_vx_h, OP_UUU_H, H2, H2, DO_REMU)
1812 RVVCALL(OPIVX2, vremu_vx_w, OP_UUU_W, H4, H4, DO_REMU)
1813 RVVCALL(OPIVX2, vremu_vx_d, OP_UUU_D, H8, H8, DO_REMU)
1814 RVVCALL(OPIVX2, vrem_vx_b, OP_SSS_B, H1, H1, DO_REM)
1815 RVVCALL(OPIVX2, vrem_vx_h, OP_SSS_H, H2, H2, DO_REM)
1816 RVVCALL(OPIVX2, vrem_vx_w, OP_SSS_W, H4, H4, DO_REM)
1817 RVVCALL(OPIVX2, vrem_vx_d, OP_SSS_D, H8, H8, DO_REM)
1818 GEN_VEXT_VX(vdivu_vx_b, 1)
1819 GEN_VEXT_VX(vdivu_vx_h, 2)
1820 GEN_VEXT_VX(vdivu_vx_w, 4)
1821 GEN_VEXT_VX(vdivu_vx_d, 8)
1822 GEN_VEXT_VX(vdiv_vx_b, 1)
1823 GEN_VEXT_VX(vdiv_vx_h, 2)
1824 GEN_VEXT_VX(vdiv_vx_w, 4)
1825 GEN_VEXT_VX(vdiv_vx_d, 8)
1826 GEN_VEXT_VX(vremu_vx_b, 1)
1827 GEN_VEXT_VX(vremu_vx_h, 2)
1828 GEN_VEXT_VX(vremu_vx_w, 4)
1829 GEN_VEXT_VX(vremu_vx_d, 8)
1830 GEN_VEXT_VX(vrem_vx_b, 1)
1831 GEN_VEXT_VX(vrem_vx_h, 2)
1832 GEN_VEXT_VX(vrem_vx_w, 4)
1833 GEN_VEXT_VX(vrem_vx_d, 8)
1834 
1835 /* Vector Widening Integer Multiply Instructions */
1836 RVVCALL(OPIVV2, vwmul_vv_b, WOP_SSS_B, H2, H1, H1, DO_MUL)
1837 RVVCALL(OPIVV2, vwmul_vv_h, WOP_SSS_H, H4, H2, H2, DO_MUL)
1838 RVVCALL(OPIVV2, vwmul_vv_w, WOP_SSS_W, H8, H4, H4, DO_MUL)
1839 RVVCALL(OPIVV2, vwmulu_vv_b, WOP_UUU_B, H2, H1, H1, DO_MUL)
1840 RVVCALL(OPIVV2, vwmulu_vv_h, WOP_UUU_H, H4, H2, H2, DO_MUL)
1841 RVVCALL(OPIVV2, vwmulu_vv_w, WOP_UUU_W, H8, H4, H4, DO_MUL)
1842 RVVCALL(OPIVV2, vwmulsu_vv_b, WOP_SUS_B, H2, H1, H1, DO_MUL)
1843 RVVCALL(OPIVV2, vwmulsu_vv_h, WOP_SUS_H, H4, H2, H2, DO_MUL)
1844 RVVCALL(OPIVV2, vwmulsu_vv_w, WOP_SUS_W, H8, H4, H4, DO_MUL)
1845 GEN_VEXT_VV(vwmul_vv_b, 2)
1846 GEN_VEXT_VV(vwmul_vv_h, 4)
1847 GEN_VEXT_VV(vwmul_vv_w, 8)
1848 GEN_VEXT_VV(vwmulu_vv_b, 2)
1849 GEN_VEXT_VV(vwmulu_vv_h, 4)
1850 GEN_VEXT_VV(vwmulu_vv_w, 8)
1851 GEN_VEXT_VV(vwmulsu_vv_b, 2)
1852 GEN_VEXT_VV(vwmulsu_vv_h, 4)
1853 GEN_VEXT_VV(vwmulsu_vv_w, 8)
1854 
1855 RVVCALL(OPIVX2, vwmul_vx_b, WOP_SSS_B, H2, H1, DO_MUL)
1856 RVVCALL(OPIVX2, vwmul_vx_h, WOP_SSS_H, H4, H2, DO_MUL)
1857 RVVCALL(OPIVX2, vwmul_vx_w, WOP_SSS_W, H8, H4, DO_MUL)
1858 RVVCALL(OPIVX2, vwmulu_vx_b, WOP_UUU_B, H2, H1, DO_MUL)
1859 RVVCALL(OPIVX2, vwmulu_vx_h, WOP_UUU_H, H4, H2, DO_MUL)
1860 RVVCALL(OPIVX2, vwmulu_vx_w, WOP_UUU_W, H8, H4, DO_MUL)
1861 RVVCALL(OPIVX2, vwmulsu_vx_b, WOP_SUS_B, H2, H1, DO_MUL)
1862 RVVCALL(OPIVX2, vwmulsu_vx_h, WOP_SUS_H, H4, H2, DO_MUL)
1863 RVVCALL(OPIVX2, vwmulsu_vx_w, WOP_SUS_W, H8, H4, DO_MUL)
1864 GEN_VEXT_VX(vwmul_vx_b, 2)
1865 GEN_VEXT_VX(vwmul_vx_h, 4)
1866 GEN_VEXT_VX(vwmul_vx_w, 8)
1867 GEN_VEXT_VX(vwmulu_vx_b, 2)
1868 GEN_VEXT_VX(vwmulu_vx_h, 4)
1869 GEN_VEXT_VX(vwmulu_vx_w, 8)
1870 GEN_VEXT_VX(vwmulsu_vx_b, 2)
1871 GEN_VEXT_VX(vwmulsu_vx_h, 4)
1872 GEN_VEXT_VX(vwmulsu_vx_w, 8)
1873 
1874 /* Vector Single-Width Integer Multiply-Add Instructions */
1875 #define OPIVV3(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)   \
1876 static void do_##NAME(void *vd, void *vs1, void *vs2, int i)       \
1877 {                                                                  \
1878     TX1 s1 = *((T1 *)vs1 + HS1(i));                                \
1879     TX2 s2 = *((T2 *)vs2 + HS2(i));                                \
1880     TD d = *((TD *)vd + HD(i));                                    \
1881     *((TD *)vd + HD(i)) = OP(s2, s1, d);                           \
1882 }
1883 
1884 #define DO_MACC(N, M, D) (M * N + D)
1885 #define DO_NMSAC(N, M, D) (-(M * N) + D)
1886 #define DO_MADD(N, M, D) (M * D + N)
1887 #define DO_NMSUB(N, M, D) (-(M * D) + N)
1888 RVVCALL(OPIVV3, vmacc_vv_b, OP_SSS_B, H1, H1, H1, DO_MACC)
1889 RVVCALL(OPIVV3, vmacc_vv_h, OP_SSS_H, H2, H2, H2, DO_MACC)
1890 RVVCALL(OPIVV3, vmacc_vv_w, OP_SSS_W, H4, H4, H4, DO_MACC)
1891 RVVCALL(OPIVV3, vmacc_vv_d, OP_SSS_D, H8, H8, H8, DO_MACC)
1892 RVVCALL(OPIVV3, vnmsac_vv_b, OP_SSS_B, H1, H1, H1, DO_NMSAC)
1893 RVVCALL(OPIVV3, vnmsac_vv_h, OP_SSS_H, H2, H2, H2, DO_NMSAC)
1894 RVVCALL(OPIVV3, vnmsac_vv_w, OP_SSS_W, H4, H4, H4, DO_NMSAC)
1895 RVVCALL(OPIVV3, vnmsac_vv_d, OP_SSS_D, H8, H8, H8, DO_NMSAC)
1896 RVVCALL(OPIVV3, vmadd_vv_b, OP_SSS_B, H1, H1, H1, DO_MADD)
1897 RVVCALL(OPIVV3, vmadd_vv_h, OP_SSS_H, H2, H2, H2, DO_MADD)
1898 RVVCALL(OPIVV3, vmadd_vv_w, OP_SSS_W, H4, H4, H4, DO_MADD)
1899 RVVCALL(OPIVV3, vmadd_vv_d, OP_SSS_D, H8, H8, H8, DO_MADD)
1900 RVVCALL(OPIVV3, vnmsub_vv_b, OP_SSS_B, H1, H1, H1, DO_NMSUB)
1901 RVVCALL(OPIVV3, vnmsub_vv_h, OP_SSS_H, H2, H2, H2, DO_NMSUB)
1902 RVVCALL(OPIVV3, vnmsub_vv_w, OP_SSS_W, H4, H4, H4, DO_NMSUB)
1903 RVVCALL(OPIVV3, vnmsub_vv_d, OP_SSS_D, H8, H8, H8, DO_NMSUB)
1904 GEN_VEXT_VV(vmacc_vv_b, 1)
1905 GEN_VEXT_VV(vmacc_vv_h, 2)
1906 GEN_VEXT_VV(vmacc_vv_w, 4)
1907 GEN_VEXT_VV(vmacc_vv_d, 8)
1908 GEN_VEXT_VV(vnmsac_vv_b, 1)
1909 GEN_VEXT_VV(vnmsac_vv_h, 2)
1910 GEN_VEXT_VV(vnmsac_vv_w, 4)
1911 GEN_VEXT_VV(vnmsac_vv_d, 8)
1912 GEN_VEXT_VV(vmadd_vv_b, 1)
1913 GEN_VEXT_VV(vmadd_vv_h, 2)
1914 GEN_VEXT_VV(vmadd_vv_w, 4)
1915 GEN_VEXT_VV(vmadd_vv_d, 8)
1916 GEN_VEXT_VV(vnmsub_vv_b, 1)
1917 GEN_VEXT_VV(vnmsub_vv_h, 2)
1918 GEN_VEXT_VV(vnmsub_vv_w, 4)
1919 GEN_VEXT_VV(vnmsub_vv_d, 8)
1920 
1921 #define OPIVX3(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)             \
1922 static void do_##NAME(void *vd, target_long s1, void *vs2, int i)   \
1923 {                                                                   \
1924     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
1925     TD d = *((TD *)vd + HD(i));                                     \
1926     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1, d);                   \
1927 }
1928 
1929 RVVCALL(OPIVX3, vmacc_vx_b, OP_SSS_B, H1, H1, DO_MACC)
1930 RVVCALL(OPIVX3, vmacc_vx_h, OP_SSS_H, H2, H2, DO_MACC)
1931 RVVCALL(OPIVX3, vmacc_vx_w, OP_SSS_W, H4, H4, DO_MACC)
1932 RVVCALL(OPIVX3, vmacc_vx_d, OP_SSS_D, H8, H8, DO_MACC)
1933 RVVCALL(OPIVX3, vnmsac_vx_b, OP_SSS_B, H1, H1, DO_NMSAC)
1934 RVVCALL(OPIVX3, vnmsac_vx_h, OP_SSS_H, H2, H2, DO_NMSAC)
1935 RVVCALL(OPIVX3, vnmsac_vx_w, OP_SSS_W, H4, H4, DO_NMSAC)
1936 RVVCALL(OPIVX3, vnmsac_vx_d, OP_SSS_D, H8, H8, DO_NMSAC)
1937 RVVCALL(OPIVX3, vmadd_vx_b, OP_SSS_B, H1, H1, DO_MADD)
1938 RVVCALL(OPIVX3, vmadd_vx_h, OP_SSS_H, H2, H2, DO_MADD)
1939 RVVCALL(OPIVX3, vmadd_vx_w, OP_SSS_W, H4, H4, DO_MADD)
1940 RVVCALL(OPIVX3, vmadd_vx_d, OP_SSS_D, H8, H8, DO_MADD)
1941 RVVCALL(OPIVX3, vnmsub_vx_b, OP_SSS_B, H1, H1, DO_NMSUB)
1942 RVVCALL(OPIVX3, vnmsub_vx_h, OP_SSS_H, H2, H2, DO_NMSUB)
1943 RVVCALL(OPIVX3, vnmsub_vx_w, OP_SSS_W, H4, H4, DO_NMSUB)
1944 RVVCALL(OPIVX3, vnmsub_vx_d, OP_SSS_D, H8, H8, DO_NMSUB)
1945 GEN_VEXT_VX(vmacc_vx_b, 1)
1946 GEN_VEXT_VX(vmacc_vx_h, 2)
1947 GEN_VEXT_VX(vmacc_vx_w, 4)
1948 GEN_VEXT_VX(vmacc_vx_d, 8)
1949 GEN_VEXT_VX(vnmsac_vx_b, 1)
1950 GEN_VEXT_VX(vnmsac_vx_h, 2)
1951 GEN_VEXT_VX(vnmsac_vx_w, 4)
1952 GEN_VEXT_VX(vnmsac_vx_d, 8)
1953 GEN_VEXT_VX(vmadd_vx_b, 1)
1954 GEN_VEXT_VX(vmadd_vx_h, 2)
1955 GEN_VEXT_VX(vmadd_vx_w, 4)
1956 GEN_VEXT_VX(vmadd_vx_d, 8)
1957 GEN_VEXT_VX(vnmsub_vx_b, 1)
1958 GEN_VEXT_VX(vnmsub_vx_h, 2)
1959 GEN_VEXT_VX(vnmsub_vx_w, 4)
1960 GEN_VEXT_VX(vnmsub_vx_d, 8)
1961 
1962 /* Vector Widening Integer Multiply-Add Instructions */
1963 RVVCALL(OPIVV3, vwmaccu_vv_b, WOP_UUU_B, H2, H1, H1, DO_MACC)
1964 RVVCALL(OPIVV3, vwmaccu_vv_h, WOP_UUU_H, H4, H2, H2, DO_MACC)
1965 RVVCALL(OPIVV3, vwmaccu_vv_w, WOP_UUU_W, H8, H4, H4, DO_MACC)
1966 RVVCALL(OPIVV3, vwmacc_vv_b, WOP_SSS_B, H2, H1, H1, DO_MACC)
1967 RVVCALL(OPIVV3, vwmacc_vv_h, WOP_SSS_H, H4, H2, H2, DO_MACC)
1968 RVVCALL(OPIVV3, vwmacc_vv_w, WOP_SSS_W, H8, H4, H4, DO_MACC)
1969 RVVCALL(OPIVV3, vwmaccsu_vv_b, WOP_SSU_B, H2, H1, H1, DO_MACC)
1970 RVVCALL(OPIVV3, vwmaccsu_vv_h, WOP_SSU_H, H4, H2, H2, DO_MACC)
1971 RVVCALL(OPIVV3, vwmaccsu_vv_w, WOP_SSU_W, H8, H4, H4, DO_MACC)
1972 GEN_VEXT_VV(vwmaccu_vv_b, 2)
1973 GEN_VEXT_VV(vwmaccu_vv_h, 4)
1974 GEN_VEXT_VV(vwmaccu_vv_w, 8)
1975 GEN_VEXT_VV(vwmacc_vv_b, 2)
1976 GEN_VEXT_VV(vwmacc_vv_h, 4)
1977 GEN_VEXT_VV(vwmacc_vv_w, 8)
1978 GEN_VEXT_VV(vwmaccsu_vv_b, 2)
1979 GEN_VEXT_VV(vwmaccsu_vv_h, 4)
1980 GEN_VEXT_VV(vwmaccsu_vv_w, 8)
1981 
1982 RVVCALL(OPIVX3, vwmaccu_vx_b, WOP_UUU_B, H2, H1, DO_MACC)
1983 RVVCALL(OPIVX3, vwmaccu_vx_h, WOP_UUU_H, H4, H2, DO_MACC)
1984 RVVCALL(OPIVX3, vwmaccu_vx_w, WOP_UUU_W, H8, H4, DO_MACC)
1985 RVVCALL(OPIVX3, vwmacc_vx_b, WOP_SSS_B, H2, H1, DO_MACC)
1986 RVVCALL(OPIVX3, vwmacc_vx_h, WOP_SSS_H, H4, H2, DO_MACC)
1987 RVVCALL(OPIVX3, vwmacc_vx_w, WOP_SSS_W, H8, H4, DO_MACC)
1988 RVVCALL(OPIVX3, vwmaccsu_vx_b, WOP_SSU_B, H2, H1, DO_MACC)
1989 RVVCALL(OPIVX3, vwmaccsu_vx_h, WOP_SSU_H, H4, H2, DO_MACC)
1990 RVVCALL(OPIVX3, vwmaccsu_vx_w, WOP_SSU_W, H8, H4, DO_MACC)
1991 RVVCALL(OPIVX3, vwmaccus_vx_b, WOP_SUS_B, H2, H1, DO_MACC)
1992 RVVCALL(OPIVX3, vwmaccus_vx_h, WOP_SUS_H, H4, H2, DO_MACC)
1993 RVVCALL(OPIVX3, vwmaccus_vx_w, WOP_SUS_W, H8, H4, DO_MACC)
1994 GEN_VEXT_VX(vwmaccu_vx_b, 2)
1995 GEN_VEXT_VX(vwmaccu_vx_h, 4)
1996 GEN_VEXT_VX(vwmaccu_vx_w, 8)
1997 GEN_VEXT_VX(vwmacc_vx_b, 2)
1998 GEN_VEXT_VX(vwmacc_vx_h, 4)
1999 GEN_VEXT_VX(vwmacc_vx_w, 8)
2000 GEN_VEXT_VX(vwmaccsu_vx_b, 2)
2001 GEN_VEXT_VX(vwmaccsu_vx_h, 4)
2002 GEN_VEXT_VX(vwmaccsu_vx_w, 8)
2003 GEN_VEXT_VX(vwmaccus_vx_b, 2)
2004 GEN_VEXT_VX(vwmaccus_vx_h, 4)
2005 GEN_VEXT_VX(vwmaccus_vx_w, 8)
2006 
2007 /* Vector Integer Merge and Move Instructions */
2008 #define GEN_VEXT_VMV_VV(NAME, ETYPE, H)                              \
2009 void HELPER(NAME)(void *vd, void *vs1, CPURISCVState *env,           \
2010                   uint32_t desc)                                     \
2011 {                                                                    \
2012     uint32_t vl = env->vl;                                           \
2013     uint32_t esz = sizeof(ETYPE);                                    \
2014     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
2015     uint32_t vta = vext_vta(desc);                                   \
2016     uint32_t i;                                                      \
2017                                                                      \
2018     for (i = env->vstart; i < vl; i++) {                             \
2019         ETYPE s1 = *((ETYPE *)vs1 + H(i));                           \
2020         *((ETYPE *)vd + H(i)) = s1;                                  \
2021     }                                                                \
2022     env->vstart = 0;                                                 \
2023     /* set tail elements to 1s */                                    \
2024     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
2025 }
2026 
2027 GEN_VEXT_VMV_VV(vmv_v_v_b, int8_t,  H1)
2028 GEN_VEXT_VMV_VV(vmv_v_v_h, int16_t, H2)
2029 GEN_VEXT_VMV_VV(vmv_v_v_w, int32_t, H4)
2030 GEN_VEXT_VMV_VV(vmv_v_v_d, int64_t, H8)
2031 
2032 #define GEN_VEXT_VMV_VX(NAME, ETYPE, H)                              \
2033 void HELPER(NAME)(void *vd, uint64_t s1, CPURISCVState *env,         \
2034                   uint32_t desc)                                     \
2035 {                                                                    \
2036     uint32_t vl = env->vl;                                           \
2037     uint32_t esz = sizeof(ETYPE);                                    \
2038     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
2039     uint32_t vta = vext_vta(desc);                                   \
2040     uint32_t i;                                                      \
2041                                                                      \
2042     for (i = env->vstart; i < vl; i++) {                             \
2043         *((ETYPE *)vd + H(i)) = (ETYPE)s1;                           \
2044     }                                                                \
2045     env->vstart = 0;                                                 \
2046     /* set tail elements to 1s */                                    \
2047     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
2048 }
2049 
2050 GEN_VEXT_VMV_VX(vmv_v_x_b, int8_t,  H1)
2051 GEN_VEXT_VMV_VX(vmv_v_x_h, int16_t, H2)
2052 GEN_VEXT_VMV_VX(vmv_v_x_w, int32_t, H4)
2053 GEN_VEXT_VMV_VX(vmv_v_x_d, int64_t, H8)
2054 
2055 #define GEN_VEXT_VMERGE_VV(NAME, ETYPE, H)                           \
2056 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,          \
2057                   CPURISCVState *env, uint32_t desc)                 \
2058 {                                                                    \
2059     uint32_t vl = env->vl;                                           \
2060     uint32_t esz = sizeof(ETYPE);                                    \
2061     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
2062     uint32_t vta = vext_vta(desc);                                   \
2063     uint32_t i;                                                      \
2064                                                                      \
2065     for (i = env->vstart; i < vl; i++) {                             \
2066         ETYPE *vt = (!vext_elem_mask(v0, i) ? vs2 : vs1);            \
2067         *((ETYPE *)vd + H(i)) = *(vt + H(i));                        \
2068     }                                                                \
2069     env->vstart = 0;                                                 \
2070     /* set tail elements to 1s */                                    \
2071     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
2072 }
2073 
2074 GEN_VEXT_VMERGE_VV(vmerge_vvm_b, int8_t,  H1)
2075 GEN_VEXT_VMERGE_VV(vmerge_vvm_h, int16_t, H2)
2076 GEN_VEXT_VMERGE_VV(vmerge_vvm_w, int32_t, H4)
2077 GEN_VEXT_VMERGE_VV(vmerge_vvm_d, int64_t, H8)
2078 
2079 #define GEN_VEXT_VMERGE_VX(NAME, ETYPE, H)                           \
2080 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,               \
2081                   void *vs2, CPURISCVState *env, uint32_t desc)      \
2082 {                                                                    \
2083     uint32_t vl = env->vl;                                           \
2084     uint32_t esz = sizeof(ETYPE);                                    \
2085     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
2086     uint32_t vta = vext_vta(desc);                                   \
2087     uint32_t i;                                                      \
2088                                                                      \
2089     for (i = env->vstart; i < vl; i++) {                             \
2090         ETYPE s2 = *((ETYPE *)vs2 + H(i));                           \
2091         ETYPE d = (!vext_elem_mask(v0, i) ? s2 :                     \
2092                    (ETYPE)(target_long)s1);                          \
2093         *((ETYPE *)vd + H(i)) = d;                                   \
2094     }                                                                \
2095     env->vstart = 0;                                                 \
2096     /* set tail elements to 1s */                                    \
2097     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
2098 }
2099 
2100 GEN_VEXT_VMERGE_VX(vmerge_vxm_b, int8_t,  H1)
2101 GEN_VEXT_VMERGE_VX(vmerge_vxm_h, int16_t, H2)
2102 GEN_VEXT_VMERGE_VX(vmerge_vxm_w, int32_t, H4)
2103 GEN_VEXT_VMERGE_VX(vmerge_vxm_d, int64_t, H8)
2104 
2105 /*
2106  *** Vector Fixed-Point Arithmetic Instructions
2107  */
2108 
2109 /* Vector Single-Width Saturating Add and Subtract */
2110 
2111 /*
2112  * As fixed point instructions probably have round mode and saturation,
2113  * define common macros for fixed point here.
2114  */
2115 typedef void opivv2_rm_fn(void *vd, void *vs1, void *vs2, int i,
2116                           CPURISCVState *env, int vxrm);
2117 
2118 #define OPIVV2_RM(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)     \
2119 static inline void                                                  \
2120 do_##NAME(void *vd, void *vs1, void *vs2, int i,                    \
2121           CPURISCVState *env, int vxrm)                             \
2122 {                                                                   \
2123     TX1 s1 = *((T1 *)vs1 + HS1(i));                                 \
2124     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
2125     *((TD *)vd + HD(i)) = OP(env, vxrm, s2, s1);                    \
2126 }
2127 
2128 static inline void
2129 vext_vv_rm_1(void *vd, void *v0, void *vs1, void *vs2,
2130              CPURISCVState *env,
2131              uint32_t vl, uint32_t vm, int vxrm,
2132              opivv2_rm_fn *fn, uint32_t vma, uint32_t esz)
2133 {
2134     for (uint32_t i = env->vstart; i < vl; i++) {
2135         if (!vm && !vext_elem_mask(v0, i)) {
2136             /* set masked-off elements to 1s */
2137             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);
2138             continue;
2139         }
2140         fn(vd, vs1, vs2, i, env, vxrm);
2141     }
2142     env->vstart = 0;
2143 }
2144 
2145 static inline void
2146 vext_vv_rm_2(void *vd, void *v0, void *vs1, void *vs2,
2147              CPURISCVState *env,
2148              uint32_t desc,
2149              opivv2_rm_fn *fn, uint32_t esz)
2150 {
2151     uint32_t vm = vext_vm(desc);
2152     uint32_t vl = env->vl;
2153     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
2154     uint32_t vta = vext_vta(desc);
2155     uint32_t vma = vext_vma(desc);
2156 
2157     switch (env->vxrm) {
2158     case 0: /* rnu */
2159         vext_vv_rm_1(vd, v0, vs1, vs2,
2160                      env, vl, vm, 0, fn, vma, esz);
2161         break;
2162     case 1: /* rne */
2163         vext_vv_rm_1(vd, v0, vs1, vs2,
2164                      env, vl, vm, 1, fn, vma, esz);
2165         break;
2166     case 2: /* rdn */
2167         vext_vv_rm_1(vd, v0, vs1, vs2,
2168                      env, vl, vm, 2, fn, vma, esz);
2169         break;
2170     default: /* rod */
2171         vext_vv_rm_1(vd, v0, vs1, vs2,
2172                      env, vl, vm, 3, fn, vma, esz);
2173         break;
2174     }
2175     /* set tail elements to 1s */
2176     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);
2177 }
2178 
2179 /* generate helpers for fixed point instructions with OPIVV format */
2180 #define GEN_VEXT_VV_RM(NAME, ESZ)                               \
2181 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,     \
2182                   CPURISCVState *env, uint32_t desc)            \
2183 {                                                               \
2184     vext_vv_rm_2(vd, v0, vs1, vs2, env, desc,                   \
2185                  do_##NAME, ESZ);                               \
2186 }
2187 
2188 static inline uint8_t saddu8(CPURISCVState *env, int vxrm, uint8_t a, uint8_t b)
2189 {
2190     uint8_t res = a + b;
2191     if (res < a) {
2192         res = UINT8_MAX;
2193         env->vxsat = 0x1;
2194     }
2195     return res;
2196 }
2197 
2198 static inline uint16_t saddu16(CPURISCVState *env, int vxrm, uint16_t a,
2199                                uint16_t b)
2200 {
2201     uint16_t res = a + b;
2202     if (res < a) {
2203         res = UINT16_MAX;
2204         env->vxsat = 0x1;
2205     }
2206     return res;
2207 }
2208 
2209 static inline uint32_t saddu32(CPURISCVState *env, int vxrm, uint32_t a,
2210                                uint32_t b)
2211 {
2212     uint32_t res = a + b;
2213     if (res < a) {
2214         res = UINT32_MAX;
2215         env->vxsat = 0x1;
2216     }
2217     return res;
2218 }
2219 
2220 static inline uint64_t saddu64(CPURISCVState *env, int vxrm, uint64_t a,
2221                                uint64_t b)
2222 {
2223     uint64_t res = a + b;
2224     if (res < a) {
2225         res = UINT64_MAX;
2226         env->vxsat = 0x1;
2227     }
2228     return res;
2229 }
2230 
2231 RVVCALL(OPIVV2_RM, vsaddu_vv_b, OP_UUU_B, H1, H1, H1, saddu8)
2232 RVVCALL(OPIVV2_RM, vsaddu_vv_h, OP_UUU_H, H2, H2, H2, saddu16)
2233 RVVCALL(OPIVV2_RM, vsaddu_vv_w, OP_UUU_W, H4, H4, H4, saddu32)
2234 RVVCALL(OPIVV2_RM, vsaddu_vv_d, OP_UUU_D, H8, H8, H8, saddu64)
2235 GEN_VEXT_VV_RM(vsaddu_vv_b, 1)
2236 GEN_VEXT_VV_RM(vsaddu_vv_h, 2)
2237 GEN_VEXT_VV_RM(vsaddu_vv_w, 4)
2238 GEN_VEXT_VV_RM(vsaddu_vv_d, 8)
2239 
2240 typedef void opivx2_rm_fn(void *vd, target_long s1, void *vs2, int i,
2241                           CPURISCVState *env, int vxrm);
2242 
2243 #define OPIVX2_RM(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)          \
2244 static inline void                                                  \
2245 do_##NAME(void *vd, target_long s1, void *vs2, int i,               \
2246           CPURISCVState *env, int vxrm)                             \
2247 {                                                                   \
2248     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
2249     *((TD *)vd + HD(i)) = OP(env, vxrm, s2, (TX1)(T1)s1);           \
2250 }
2251 
2252 static inline void
2253 vext_vx_rm_1(void *vd, void *v0, target_long s1, void *vs2,
2254              CPURISCVState *env,
2255              uint32_t vl, uint32_t vm, int vxrm,
2256              opivx2_rm_fn *fn, uint32_t vma, uint32_t esz)
2257 {
2258     for (uint32_t i = env->vstart; i < vl; i++) {
2259         if (!vm && !vext_elem_mask(v0, i)) {
2260             /* set masked-off elements to 1s */
2261             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);
2262             continue;
2263         }
2264         fn(vd, s1, vs2, i, env, vxrm);
2265     }
2266     env->vstart = 0;
2267 }
2268 
2269 static inline void
2270 vext_vx_rm_2(void *vd, void *v0, target_long s1, void *vs2,
2271              CPURISCVState *env,
2272              uint32_t desc,
2273              opivx2_rm_fn *fn, uint32_t esz)
2274 {
2275     uint32_t vm = vext_vm(desc);
2276     uint32_t vl = env->vl;
2277     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
2278     uint32_t vta = vext_vta(desc);
2279     uint32_t vma = vext_vma(desc);
2280 
2281     switch (env->vxrm) {
2282     case 0: /* rnu */
2283         vext_vx_rm_1(vd, v0, s1, vs2,
2284                      env, vl, vm, 0, fn, vma, esz);
2285         break;
2286     case 1: /* rne */
2287         vext_vx_rm_1(vd, v0, s1, vs2,
2288                      env, vl, vm, 1, fn, vma, esz);
2289         break;
2290     case 2: /* rdn */
2291         vext_vx_rm_1(vd, v0, s1, vs2,
2292                      env, vl, vm, 2, fn, vma, esz);
2293         break;
2294     default: /* rod */
2295         vext_vx_rm_1(vd, v0, s1, vs2,
2296                      env, vl, vm, 3, fn, vma, esz);
2297         break;
2298     }
2299     /* set tail elements to 1s */
2300     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);
2301 }
2302 
2303 /* generate helpers for fixed point instructions with OPIVX format */
2304 #define GEN_VEXT_VX_RM(NAME, ESZ)                         \
2305 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,    \
2306         void *vs2, CPURISCVState *env, uint32_t desc)     \
2307 {                                                         \
2308     vext_vx_rm_2(vd, v0, s1, vs2, env, desc,              \
2309                  do_##NAME, ESZ);                         \
2310 }
2311 
2312 RVVCALL(OPIVX2_RM, vsaddu_vx_b, OP_UUU_B, H1, H1, saddu8)
2313 RVVCALL(OPIVX2_RM, vsaddu_vx_h, OP_UUU_H, H2, H2, saddu16)
2314 RVVCALL(OPIVX2_RM, vsaddu_vx_w, OP_UUU_W, H4, H4, saddu32)
2315 RVVCALL(OPIVX2_RM, vsaddu_vx_d, OP_UUU_D, H8, H8, saddu64)
2316 GEN_VEXT_VX_RM(vsaddu_vx_b, 1)
2317 GEN_VEXT_VX_RM(vsaddu_vx_h, 2)
2318 GEN_VEXT_VX_RM(vsaddu_vx_w, 4)
2319 GEN_VEXT_VX_RM(vsaddu_vx_d, 8)
2320 
2321 static inline int8_t sadd8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2322 {
2323     int8_t res = a + b;
2324     if ((res ^ a) & (res ^ b) & INT8_MIN) {
2325         res = a > 0 ? INT8_MAX : INT8_MIN;
2326         env->vxsat = 0x1;
2327     }
2328     return res;
2329 }
2330 
2331 static inline int16_t sadd16(CPURISCVState *env, int vxrm, int16_t a, int16_t b)
2332 {
2333     int16_t res = a + b;
2334     if ((res ^ a) & (res ^ b) & INT16_MIN) {
2335         res = a > 0 ? INT16_MAX : INT16_MIN;
2336         env->vxsat = 0x1;
2337     }
2338     return res;
2339 }
2340 
2341 static inline int32_t sadd32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2342 {
2343     int32_t res = a + b;
2344     if ((res ^ a) & (res ^ b) & INT32_MIN) {
2345         res = a > 0 ? INT32_MAX : INT32_MIN;
2346         env->vxsat = 0x1;
2347     }
2348     return res;
2349 }
2350 
2351 static inline int64_t sadd64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2352 {
2353     int64_t res = a + b;
2354     if ((res ^ a) & (res ^ b) & INT64_MIN) {
2355         res = a > 0 ? INT64_MAX : INT64_MIN;
2356         env->vxsat = 0x1;
2357     }
2358     return res;
2359 }
2360 
2361 RVVCALL(OPIVV2_RM, vsadd_vv_b, OP_SSS_B, H1, H1, H1, sadd8)
2362 RVVCALL(OPIVV2_RM, vsadd_vv_h, OP_SSS_H, H2, H2, H2, sadd16)
2363 RVVCALL(OPIVV2_RM, vsadd_vv_w, OP_SSS_W, H4, H4, H4, sadd32)
2364 RVVCALL(OPIVV2_RM, vsadd_vv_d, OP_SSS_D, H8, H8, H8, sadd64)
2365 GEN_VEXT_VV_RM(vsadd_vv_b, 1)
2366 GEN_VEXT_VV_RM(vsadd_vv_h, 2)
2367 GEN_VEXT_VV_RM(vsadd_vv_w, 4)
2368 GEN_VEXT_VV_RM(vsadd_vv_d, 8)
2369 
2370 RVVCALL(OPIVX2_RM, vsadd_vx_b, OP_SSS_B, H1, H1, sadd8)
2371 RVVCALL(OPIVX2_RM, vsadd_vx_h, OP_SSS_H, H2, H2, sadd16)
2372 RVVCALL(OPIVX2_RM, vsadd_vx_w, OP_SSS_W, H4, H4, sadd32)
2373 RVVCALL(OPIVX2_RM, vsadd_vx_d, OP_SSS_D, H8, H8, sadd64)
2374 GEN_VEXT_VX_RM(vsadd_vx_b, 1)
2375 GEN_VEXT_VX_RM(vsadd_vx_h, 2)
2376 GEN_VEXT_VX_RM(vsadd_vx_w, 4)
2377 GEN_VEXT_VX_RM(vsadd_vx_d, 8)
2378 
2379 static inline uint8_t ssubu8(CPURISCVState *env, int vxrm, uint8_t a, uint8_t b)
2380 {
2381     uint8_t res = a - b;
2382     if (res > a) {
2383         res = 0;
2384         env->vxsat = 0x1;
2385     }
2386     return res;
2387 }
2388 
2389 static inline uint16_t ssubu16(CPURISCVState *env, int vxrm, uint16_t a,
2390                                uint16_t b)
2391 {
2392     uint16_t res = a - b;
2393     if (res > a) {
2394         res = 0;
2395         env->vxsat = 0x1;
2396     }
2397     return res;
2398 }
2399 
2400 static inline uint32_t ssubu32(CPURISCVState *env, int vxrm, uint32_t a,
2401                                uint32_t b)
2402 {
2403     uint32_t res = a - b;
2404     if (res > a) {
2405         res = 0;
2406         env->vxsat = 0x1;
2407     }
2408     return res;
2409 }
2410 
2411 static inline uint64_t ssubu64(CPURISCVState *env, int vxrm, uint64_t a,
2412                                uint64_t b)
2413 {
2414     uint64_t res = a - b;
2415     if (res > a) {
2416         res = 0;
2417         env->vxsat = 0x1;
2418     }
2419     return res;
2420 }
2421 
2422 RVVCALL(OPIVV2_RM, vssubu_vv_b, OP_UUU_B, H1, H1, H1, ssubu8)
2423 RVVCALL(OPIVV2_RM, vssubu_vv_h, OP_UUU_H, H2, H2, H2, ssubu16)
2424 RVVCALL(OPIVV2_RM, vssubu_vv_w, OP_UUU_W, H4, H4, H4, ssubu32)
2425 RVVCALL(OPIVV2_RM, vssubu_vv_d, OP_UUU_D, H8, H8, H8, ssubu64)
2426 GEN_VEXT_VV_RM(vssubu_vv_b, 1)
2427 GEN_VEXT_VV_RM(vssubu_vv_h, 2)
2428 GEN_VEXT_VV_RM(vssubu_vv_w, 4)
2429 GEN_VEXT_VV_RM(vssubu_vv_d, 8)
2430 
2431 RVVCALL(OPIVX2_RM, vssubu_vx_b, OP_UUU_B, H1, H1, ssubu8)
2432 RVVCALL(OPIVX2_RM, vssubu_vx_h, OP_UUU_H, H2, H2, ssubu16)
2433 RVVCALL(OPIVX2_RM, vssubu_vx_w, OP_UUU_W, H4, H4, ssubu32)
2434 RVVCALL(OPIVX2_RM, vssubu_vx_d, OP_UUU_D, H8, H8, ssubu64)
2435 GEN_VEXT_VX_RM(vssubu_vx_b, 1)
2436 GEN_VEXT_VX_RM(vssubu_vx_h, 2)
2437 GEN_VEXT_VX_RM(vssubu_vx_w, 4)
2438 GEN_VEXT_VX_RM(vssubu_vx_d, 8)
2439 
2440 static inline int8_t ssub8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2441 {
2442     int8_t res = a - b;
2443     if ((res ^ a) & (a ^ b) & INT8_MIN) {
2444         res = a >= 0 ? INT8_MAX : INT8_MIN;
2445         env->vxsat = 0x1;
2446     }
2447     return res;
2448 }
2449 
2450 static inline int16_t ssub16(CPURISCVState *env, int vxrm, int16_t a, int16_t b)
2451 {
2452     int16_t res = a - b;
2453     if ((res ^ a) & (a ^ b) & INT16_MIN) {
2454         res = a >= 0 ? INT16_MAX : INT16_MIN;
2455         env->vxsat = 0x1;
2456     }
2457     return res;
2458 }
2459 
2460 static inline int32_t ssub32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2461 {
2462     int32_t res = a - b;
2463     if ((res ^ a) & (a ^ b) & INT32_MIN) {
2464         res = a >= 0 ? INT32_MAX : INT32_MIN;
2465         env->vxsat = 0x1;
2466     }
2467     return res;
2468 }
2469 
2470 static inline int64_t ssub64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2471 {
2472     int64_t res = a - b;
2473     if ((res ^ a) & (a ^ b) & INT64_MIN) {
2474         res = a >= 0 ? INT64_MAX : INT64_MIN;
2475         env->vxsat = 0x1;
2476     }
2477     return res;
2478 }
2479 
2480 RVVCALL(OPIVV2_RM, vssub_vv_b, OP_SSS_B, H1, H1, H1, ssub8)
2481 RVVCALL(OPIVV2_RM, vssub_vv_h, OP_SSS_H, H2, H2, H2, ssub16)
2482 RVVCALL(OPIVV2_RM, vssub_vv_w, OP_SSS_W, H4, H4, H4, ssub32)
2483 RVVCALL(OPIVV2_RM, vssub_vv_d, OP_SSS_D, H8, H8, H8, ssub64)
2484 GEN_VEXT_VV_RM(vssub_vv_b, 1)
2485 GEN_VEXT_VV_RM(vssub_vv_h, 2)
2486 GEN_VEXT_VV_RM(vssub_vv_w, 4)
2487 GEN_VEXT_VV_RM(vssub_vv_d, 8)
2488 
2489 RVVCALL(OPIVX2_RM, vssub_vx_b, OP_SSS_B, H1, H1, ssub8)
2490 RVVCALL(OPIVX2_RM, vssub_vx_h, OP_SSS_H, H2, H2, ssub16)
2491 RVVCALL(OPIVX2_RM, vssub_vx_w, OP_SSS_W, H4, H4, ssub32)
2492 RVVCALL(OPIVX2_RM, vssub_vx_d, OP_SSS_D, H8, H8, ssub64)
2493 GEN_VEXT_VX_RM(vssub_vx_b, 1)
2494 GEN_VEXT_VX_RM(vssub_vx_h, 2)
2495 GEN_VEXT_VX_RM(vssub_vx_w, 4)
2496 GEN_VEXT_VX_RM(vssub_vx_d, 8)
2497 
2498 /* Vector Single-Width Averaging Add and Subtract */
2499 static inline uint8_t get_round(int vxrm, uint64_t v, uint8_t shift)
2500 {
2501     uint8_t d = extract64(v, shift, 1);
2502     uint8_t d1;
2503     uint64_t D1, D2;
2504 
2505     if (shift == 0 || shift > 64) {
2506         return 0;
2507     }
2508 
2509     d1 = extract64(v, shift - 1, 1);
2510     D1 = extract64(v, 0, shift);
2511     if (vxrm == 0) { /* round-to-nearest-up (add +0.5 LSB) */
2512         return d1;
2513     } else if (vxrm == 1) { /* round-to-nearest-even */
2514         if (shift > 1) {
2515             D2 = extract64(v, 0, shift - 1);
2516             return d1 & ((D2 != 0) | d);
2517         } else {
2518             return d1 & d;
2519         }
2520     } else if (vxrm == 3) { /* round-to-odd (OR bits into LSB, aka "jam") */
2521         return !d & (D1 != 0);
2522     }
2523     return 0; /* round-down (truncate) */
2524 }
2525 
2526 static inline int32_t aadd32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2527 {
2528     int64_t res = (int64_t)a + b;
2529     uint8_t round = get_round(vxrm, res, 1);
2530 
2531     return (res >> 1) + round;
2532 }
2533 
2534 static inline int64_t aadd64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2535 {
2536     int64_t res = a + b;
2537     uint8_t round = get_round(vxrm, res, 1);
2538     int64_t over = (res ^ a) & (res ^ b) & INT64_MIN;
2539 
2540     /* With signed overflow, bit 64 is inverse of bit 63. */
2541     return ((res >> 1) ^ over) + round;
2542 }
2543 
2544 RVVCALL(OPIVV2_RM, vaadd_vv_b, OP_SSS_B, H1, H1, H1, aadd32)
2545 RVVCALL(OPIVV2_RM, vaadd_vv_h, OP_SSS_H, H2, H2, H2, aadd32)
2546 RVVCALL(OPIVV2_RM, vaadd_vv_w, OP_SSS_W, H4, H4, H4, aadd32)
2547 RVVCALL(OPIVV2_RM, vaadd_vv_d, OP_SSS_D, H8, H8, H8, aadd64)
2548 GEN_VEXT_VV_RM(vaadd_vv_b, 1)
2549 GEN_VEXT_VV_RM(vaadd_vv_h, 2)
2550 GEN_VEXT_VV_RM(vaadd_vv_w, 4)
2551 GEN_VEXT_VV_RM(vaadd_vv_d, 8)
2552 
2553 RVVCALL(OPIVX2_RM, vaadd_vx_b, OP_SSS_B, H1, H1, aadd32)
2554 RVVCALL(OPIVX2_RM, vaadd_vx_h, OP_SSS_H, H2, H2, aadd32)
2555 RVVCALL(OPIVX2_RM, vaadd_vx_w, OP_SSS_W, H4, H4, aadd32)
2556 RVVCALL(OPIVX2_RM, vaadd_vx_d, OP_SSS_D, H8, H8, aadd64)
2557 GEN_VEXT_VX_RM(vaadd_vx_b, 1)
2558 GEN_VEXT_VX_RM(vaadd_vx_h, 2)
2559 GEN_VEXT_VX_RM(vaadd_vx_w, 4)
2560 GEN_VEXT_VX_RM(vaadd_vx_d, 8)
2561 
2562 static inline uint32_t aaddu32(CPURISCVState *env, int vxrm,
2563                                uint32_t a, uint32_t b)
2564 {
2565     uint64_t res = (uint64_t)a + b;
2566     uint8_t round = get_round(vxrm, res, 1);
2567 
2568     return (res >> 1) + round;
2569 }
2570 
2571 static inline uint64_t aaddu64(CPURISCVState *env, int vxrm,
2572                                uint64_t a, uint64_t b)
2573 {
2574     uint64_t res = a + b;
2575     uint8_t round = get_round(vxrm, res, 1);
2576     uint64_t over = (uint64_t)(res < a) << 63;
2577 
2578     return ((res >> 1) | over) + round;
2579 }
2580 
2581 RVVCALL(OPIVV2_RM, vaaddu_vv_b, OP_UUU_B, H1, H1, H1, aaddu32)
2582 RVVCALL(OPIVV2_RM, vaaddu_vv_h, OP_UUU_H, H2, H2, H2, aaddu32)
2583 RVVCALL(OPIVV2_RM, vaaddu_vv_w, OP_UUU_W, H4, H4, H4, aaddu32)
2584 RVVCALL(OPIVV2_RM, vaaddu_vv_d, OP_UUU_D, H8, H8, H8, aaddu64)
2585 GEN_VEXT_VV_RM(vaaddu_vv_b, 1)
2586 GEN_VEXT_VV_RM(vaaddu_vv_h, 2)
2587 GEN_VEXT_VV_RM(vaaddu_vv_w, 4)
2588 GEN_VEXT_VV_RM(vaaddu_vv_d, 8)
2589 
2590 RVVCALL(OPIVX2_RM, vaaddu_vx_b, OP_UUU_B, H1, H1, aaddu32)
2591 RVVCALL(OPIVX2_RM, vaaddu_vx_h, OP_UUU_H, H2, H2, aaddu32)
2592 RVVCALL(OPIVX2_RM, vaaddu_vx_w, OP_UUU_W, H4, H4, aaddu32)
2593 RVVCALL(OPIVX2_RM, vaaddu_vx_d, OP_UUU_D, H8, H8, aaddu64)
2594 GEN_VEXT_VX_RM(vaaddu_vx_b, 1)
2595 GEN_VEXT_VX_RM(vaaddu_vx_h, 2)
2596 GEN_VEXT_VX_RM(vaaddu_vx_w, 4)
2597 GEN_VEXT_VX_RM(vaaddu_vx_d, 8)
2598 
2599 static inline int32_t asub32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2600 {
2601     int64_t res = (int64_t)a - b;
2602     uint8_t round = get_round(vxrm, res, 1);
2603 
2604     return (res >> 1) + round;
2605 }
2606 
2607 static inline int64_t asub64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2608 {
2609     int64_t res = (int64_t)a - b;
2610     uint8_t round = get_round(vxrm, res, 1);
2611     int64_t over = (res ^ a) & (a ^ b) & INT64_MIN;
2612 
2613     /* With signed overflow, bit 64 is inverse of bit 63. */
2614     return ((res >> 1) ^ over) + round;
2615 }
2616 
2617 RVVCALL(OPIVV2_RM, vasub_vv_b, OP_SSS_B, H1, H1, H1, asub32)
2618 RVVCALL(OPIVV2_RM, vasub_vv_h, OP_SSS_H, H2, H2, H2, asub32)
2619 RVVCALL(OPIVV2_RM, vasub_vv_w, OP_SSS_W, H4, H4, H4, asub32)
2620 RVVCALL(OPIVV2_RM, vasub_vv_d, OP_SSS_D, H8, H8, H8, asub64)
2621 GEN_VEXT_VV_RM(vasub_vv_b, 1)
2622 GEN_VEXT_VV_RM(vasub_vv_h, 2)
2623 GEN_VEXT_VV_RM(vasub_vv_w, 4)
2624 GEN_VEXT_VV_RM(vasub_vv_d, 8)
2625 
2626 RVVCALL(OPIVX2_RM, vasub_vx_b, OP_SSS_B, H1, H1, asub32)
2627 RVVCALL(OPIVX2_RM, vasub_vx_h, OP_SSS_H, H2, H2, asub32)
2628 RVVCALL(OPIVX2_RM, vasub_vx_w, OP_SSS_W, H4, H4, asub32)
2629 RVVCALL(OPIVX2_RM, vasub_vx_d, OP_SSS_D, H8, H8, asub64)
2630 GEN_VEXT_VX_RM(vasub_vx_b, 1)
2631 GEN_VEXT_VX_RM(vasub_vx_h, 2)
2632 GEN_VEXT_VX_RM(vasub_vx_w, 4)
2633 GEN_VEXT_VX_RM(vasub_vx_d, 8)
2634 
2635 static inline uint32_t asubu32(CPURISCVState *env, int vxrm,
2636                                uint32_t a, uint32_t b)
2637 {
2638     int64_t res = (int64_t)a - b;
2639     uint8_t round = get_round(vxrm, res, 1);
2640 
2641     return (res >> 1) + round;
2642 }
2643 
2644 static inline uint64_t asubu64(CPURISCVState *env, int vxrm,
2645                                uint64_t a, uint64_t b)
2646 {
2647     uint64_t res = (uint64_t)a - b;
2648     uint8_t round = get_round(vxrm, res, 1);
2649     uint64_t over = (uint64_t)(res > a) << 63;
2650 
2651     return ((res >> 1) | over) + round;
2652 }
2653 
2654 RVVCALL(OPIVV2_RM, vasubu_vv_b, OP_UUU_B, H1, H1, H1, asubu32)
2655 RVVCALL(OPIVV2_RM, vasubu_vv_h, OP_UUU_H, H2, H2, H2, asubu32)
2656 RVVCALL(OPIVV2_RM, vasubu_vv_w, OP_UUU_W, H4, H4, H4, asubu32)
2657 RVVCALL(OPIVV2_RM, vasubu_vv_d, OP_UUU_D, H8, H8, H8, asubu64)
2658 GEN_VEXT_VV_RM(vasubu_vv_b, 1)
2659 GEN_VEXT_VV_RM(vasubu_vv_h, 2)
2660 GEN_VEXT_VV_RM(vasubu_vv_w, 4)
2661 GEN_VEXT_VV_RM(vasubu_vv_d, 8)
2662 
2663 RVVCALL(OPIVX2_RM, vasubu_vx_b, OP_UUU_B, H1, H1, asubu32)
2664 RVVCALL(OPIVX2_RM, vasubu_vx_h, OP_UUU_H, H2, H2, asubu32)
2665 RVVCALL(OPIVX2_RM, vasubu_vx_w, OP_UUU_W, H4, H4, asubu32)
2666 RVVCALL(OPIVX2_RM, vasubu_vx_d, OP_UUU_D, H8, H8, asubu64)
2667 GEN_VEXT_VX_RM(vasubu_vx_b, 1)
2668 GEN_VEXT_VX_RM(vasubu_vx_h, 2)
2669 GEN_VEXT_VX_RM(vasubu_vx_w, 4)
2670 GEN_VEXT_VX_RM(vasubu_vx_d, 8)
2671 
2672 /* Vector Single-Width Fractional Multiply with Rounding and Saturation */
2673 static inline int8_t vsmul8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2674 {
2675     uint8_t round;
2676     int16_t res;
2677 
2678     res = (int16_t)a * (int16_t)b;
2679     round = get_round(vxrm, res, 7);
2680     res   = (res >> 7) + round;
2681 
2682     if (res > INT8_MAX) {
2683         env->vxsat = 0x1;
2684         return INT8_MAX;
2685     } else if (res < INT8_MIN) {
2686         env->vxsat = 0x1;
2687         return INT8_MIN;
2688     } else {
2689         return res;
2690     }
2691 }
2692 
2693 static int16_t vsmul16(CPURISCVState *env, int vxrm, int16_t a, int16_t b)
2694 {
2695     uint8_t round;
2696     int32_t res;
2697 
2698     res = (int32_t)a * (int32_t)b;
2699     round = get_round(vxrm, res, 15);
2700     res   = (res >> 15) + round;
2701 
2702     if (res > INT16_MAX) {
2703         env->vxsat = 0x1;
2704         return INT16_MAX;
2705     } else if (res < INT16_MIN) {
2706         env->vxsat = 0x1;
2707         return INT16_MIN;
2708     } else {
2709         return res;
2710     }
2711 }
2712 
2713 static int32_t vsmul32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2714 {
2715     uint8_t round;
2716     int64_t res;
2717 
2718     res = (int64_t)a * (int64_t)b;
2719     round = get_round(vxrm, res, 31);
2720     res   = (res >> 31) + round;
2721 
2722     if (res > INT32_MAX) {
2723         env->vxsat = 0x1;
2724         return INT32_MAX;
2725     } else if (res < INT32_MIN) {
2726         env->vxsat = 0x1;
2727         return INT32_MIN;
2728     } else {
2729         return res;
2730     }
2731 }
2732 
2733 static int64_t vsmul64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2734 {
2735     uint8_t round;
2736     uint64_t hi_64, lo_64;
2737     int64_t res;
2738 
2739     if (a == INT64_MIN && b == INT64_MIN) {
2740         env->vxsat = 1;
2741         return INT64_MAX;
2742     }
2743 
2744     muls64(&lo_64, &hi_64, a, b);
2745     round = get_round(vxrm, lo_64, 63);
2746     /*
2747      * Cannot overflow, as there are always
2748      * 2 sign bits after multiply.
2749      */
2750     res = (hi_64 << 1) | (lo_64 >> 63);
2751     if (round) {
2752         if (res == INT64_MAX) {
2753             env->vxsat = 1;
2754         } else {
2755             res += 1;
2756         }
2757     }
2758     return res;
2759 }
2760 
2761 RVVCALL(OPIVV2_RM, vsmul_vv_b, OP_SSS_B, H1, H1, H1, vsmul8)
2762 RVVCALL(OPIVV2_RM, vsmul_vv_h, OP_SSS_H, H2, H2, H2, vsmul16)
2763 RVVCALL(OPIVV2_RM, vsmul_vv_w, OP_SSS_W, H4, H4, H4, vsmul32)
2764 RVVCALL(OPIVV2_RM, vsmul_vv_d, OP_SSS_D, H8, H8, H8, vsmul64)
2765 GEN_VEXT_VV_RM(vsmul_vv_b, 1)
2766 GEN_VEXT_VV_RM(vsmul_vv_h, 2)
2767 GEN_VEXT_VV_RM(vsmul_vv_w, 4)
2768 GEN_VEXT_VV_RM(vsmul_vv_d, 8)
2769 
2770 RVVCALL(OPIVX2_RM, vsmul_vx_b, OP_SSS_B, H1, H1, vsmul8)
2771 RVVCALL(OPIVX2_RM, vsmul_vx_h, OP_SSS_H, H2, H2, vsmul16)
2772 RVVCALL(OPIVX2_RM, vsmul_vx_w, OP_SSS_W, H4, H4, vsmul32)
2773 RVVCALL(OPIVX2_RM, vsmul_vx_d, OP_SSS_D, H8, H8, vsmul64)
2774 GEN_VEXT_VX_RM(vsmul_vx_b, 1)
2775 GEN_VEXT_VX_RM(vsmul_vx_h, 2)
2776 GEN_VEXT_VX_RM(vsmul_vx_w, 4)
2777 GEN_VEXT_VX_RM(vsmul_vx_d, 8)
2778 
2779 /* Vector Single-Width Scaling Shift Instructions */
2780 static inline uint8_t
2781 vssrl8(CPURISCVState *env, int vxrm, uint8_t a, uint8_t b)
2782 {
2783     uint8_t round, shift = b & 0x7;
2784     uint8_t res;
2785 
2786     round = get_round(vxrm, a, shift);
2787     res   = (a >> shift)  + round;
2788     return res;
2789 }
2790 static inline uint16_t
2791 vssrl16(CPURISCVState *env, int vxrm, uint16_t a, uint16_t b)
2792 {
2793     uint8_t round, shift = b & 0xf;
2794 
2795     round = get_round(vxrm, a, shift);
2796     return (a >> shift) + round;
2797 }
2798 static inline uint32_t
2799 vssrl32(CPURISCVState *env, int vxrm, uint32_t a, uint32_t b)
2800 {
2801     uint8_t round, shift = b & 0x1f;
2802 
2803     round = get_round(vxrm, a, shift);
2804     return (a >> shift) + round;
2805 }
2806 static inline uint64_t
2807 vssrl64(CPURISCVState *env, int vxrm, uint64_t a, uint64_t b)
2808 {
2809     uint8_t round, shift = b & 0x3f;
2810 
2811     round = get_round(vxrm, a, shift);
2812     return (a >> shift) + round;
2813 }
2814 RVVCALL(OPIVV2_RM, vssrl_vv_b, OP_UUU_B, H1, H1, H1, vssrl8)
2815 RVVCALL(OPIVV2_RM, vssrl_vv_h, OP_UUU_H, H2, H2, H2, vssrl16)
2816 RVVCALL(OPIVV2_RM, vssrl_vv_w, OP_UUU_W, H4, H4, H4, vssrl32)
2817 RVVCALL(OPIVV2_RM, vssrl_vv_d, OP_UUU_D, H8, H8, H8, vssrl64)
2818 GEN_VEXT_VV_RM(vssrl_vv_b, 1)
2819 GEN_VEXT_VV_RM(vssrl_vv_h, 2)
2820 GEN_VEXT_VV_RM(vssrl_vv_w, 4)
2821 GEN_VEXT_VV_RM(vssrl_vv_d, 8)
2822 
2823 RVVCALL(OPIVX2_RM, vssrl_vx_b, OP_UUU_B, H1, H1, vssrl8)
2824 RVVCALL(OPIVX2_RM, vssrl_vx_h, OP_UUU_H, H2, H2, vssrl16)
2825 RVVCALL(OPIVX2_RM, vssrl_vx_w, OP_UUU_W, H4, H4, vssrl32)
2826 RVVCALL(OPIVX2_RM, vssrl_vx_d, OP_UUU_D, H8, H8, vssrl64)
2827 GEN_VEXT_VX_RM(vssrl_vx_b, 1)
2828 GEN_VEXT_VX_RM(vssrl_vx_h, 2)
2829 GEN_VEXT_VX_RM(vssrl_vx_w, 4)
2830 GEN_VEXT_VX_RM(vssrl_vx_d, 8)
2831 
2832 static inline int8_t
2833 vssra8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2834 {
2835     uint8_t round, shift = b & 0x7;
2836 
2837     round = get_round(vxrm, a, shift);
2838     return (a >> shift) + round;
2839 }
2840 static inline int16_t
2841 vssra16(CPURISCVState *env, int vxrm, int16_t a, int16_t b)
2842 {
2843     uint8_t round, shift = b & 0xf;
2844 
2845     round = get_round(vxrm, a, shift);
2846     return (a >> shift) + round;
2847 }
2848 static inline int32_t
2849 vssra32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2850 {
2851     uint8_t round, shift = b & 0x1f;
2852 
2853     round = get_round(vxrm, a, shift);
2854     return (a >> shift) + round;
2855 }
2856 static inline int64_t
2857 vssra64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2858 {
2859     uint8_t round, shift = b & 0x3f;
2860 
2861     round = get_round(vxrm, a, shift);
2862     return (a >> shift) + round;
2863 }
2864 
2865 RVVCALL(OPIVV2_RM, vssra_vv_b, OP_SSS_B, H1, H1, H1, vssra8)
2866 RVVCALL(OPIVV2_RM, vssra_vv_h, OP_SSS_H, H2, H2, H2, vssra16)
2867 RVVCALL(OPIVV2_RM, vssra_vv_w, OP_SSS_W, H4, H4, H4, vssra32)
2868 RVVCALL(OPIVV2_RM, vssra_vv_d, OP_SSS_D, H8, H8, H8, vssra64)
2869 GEN_VEXT_VV_RM(vssra_vv_b, 1)
2870 GEN_VEXT_VV_RM(vssra_vv_h, 2)
2871 GEN_VEXT_VV_RM(vssra_vv_w, 4)
2872 GEN_VEXT_VV_RM(vssra_vv_d, 8)
2873 
2874 RVVCALL(OPIVX2_RM, vssra_vx_b, OP_SSS_B, H1, H1, vssra8)
2875 RVVCALL(OPIVX2_RM, vssra_vx_h, OP_SSS_H, H2, H2, vssra16)
2876 RVVCALL(OPIVX2_RM, vssra_vx_w, OP_SSS_W, H4, H4, vssra32)
2877 RVVCALL(OPIVX2_RM, vssra_vx_d, OP_SSS_D, H8, H8, vssra64)
2878 GEN_VEXT_VX_RM(vssra_vx_b, 1)
2879 GEN_VEXT_VX_RM(vssra_vx_h, 2)
2880 GEN_VEXT_VX_RM(vssra_vx_w, 4)
2881 GEN_VEXT_VX_RM(vssra_vx_d, 8)
2882 
2883 /* Vector Narrowing Fixed-Point Clip Instructions */
2884 static inline int8_t
2885 vnclip8(CPURISCVState *env, int vxrm, int16_t a, int8_t b)
2886 {
2887     uint8_t round, shift = b & 0xf;
2888     int16_t res;
2889 
2890     round = get_round(vxrm, a, shift);
2891     res   = (a >> shift)  + round;
2892     if (res > INT8_MAX) {
2893         env->vxsat = 0x1;
2894         return INT8_MAX;
2895     } else if (res < INT8_MIN) {
2896         env->vxsat = 0x1;
2897         return INT8_MIN;
2898     } else {
2899         return res;
2900     }
2901 }
2902 
2903 static inline int16_t
2904 vnclip16(CPURISCVState *env, int vxrm, int32_t a, int16_t b)
2905 {
2906     uint8_t round, shift = b & 0x1f;
2907     int32_t res;
2908 
2909     round = get_round(vxrm, a, shift);
2910     res   = (a >> shift)  + round;
2911     if (res > INT16_MAX) {
2912         env->vxsat = 0x1;
2913         return INT16_MAX;
2914     } else if (res < INT16_MIN) {
2915         env->vxsat = 0x1;
2916         return INT16_MIN;
2917     } else {
2918         return res;
2919     }
2920 }
2921 
2922 static inline int32_t
2923 vnclip32(CPURISCVState *env, int vxrm, int64_t a, int32_t b)
2924 {
2925     uint8_t round, shift = b & 0x3f;
2926     int64_t res;
2927 
2928     round = get_round(vxrm, a, shift);
2929     res   = (a >> shift)  + round;
2930     if (res > INT32_MAX) {
2931         env->vxsat = 0x1;
2932         return INT32_MAX;
2933     } else if (res < INT32_MIN) {
2934         env->vxsat = 0x1;
2935         return INT32_MIN;
2936     } else {
2937         return res;
2938     }
2939 }
2940 
2941 RVVCALL(OPIVV2_RM, vnclip_wv_b, NOP_SSS_B, H1, H2, H1, vnclip8)
2942 RVVCALL(OPIVV2_RM, vnclip_wv_h, NOP_SSS_H, H2, H4, H2, vnclip16)
2943 RVVCALL(OPIVV2_RM, vnclip_wv_w, NOP_SSS_W, H4, H8, H4, vnclip32)
2944 GEN_VEXT_VV_RM(vnclip_wv_b, 1)
2945 GEN_VEXT_VV_RM(vnclip_wv_h, 2)
2946 GEN_VEXT_VV_RM(vnclip_wv_w, 4)
2947 
2948 RVVCALL(OPIVX2_RM, vnclip_wx_b, NOP_SSS_B, H1, H2, vnclip8)
2949 RVVCALL(OPIVX2_RM, vnclip_wx_h, NOP_SSS_H, H2, H4, vnclip16)
2950 RVVCALL(OPIVX2_RM, vnclip_wx_w, NOP_SSS_W, H4, H8, vnclip32)
2951 GEN_VEXT_VX_RM(vnclip_wx_b, 1)
2952 GEN_VEXT_VX_RM(vnclip_wx_h, 2)
2953 GEN_VEXT_VX_RM(vnclip_wx_w, 4)
2954 
2955 static inline uint8_t
2956 vnclipu8(CPURISCVState *env, int vxrm, uint16_t a, uint8_t b)
2957 {
2958     uint8_t round, shift = b & 0xf;
2959     uint16_t res;
2960 
2961     round = get_round(vxrm, a, shift);
2962     res   = (a >> shift)  + round;
2963     if (res > UINT8_MAX) {
2964         env->vxsat = 0x1;
2965         return UINT8_MAX;
2966     } else {
2967         return res;
2968     }
2969 }
2970 
2971 static inline uint16_t
2972 vnclipu16(CPURISCVState *env, int vxrm, uint32_t a, uint16_t b)
2973 {
2974     uint8_t round, shift = b & 0x1f;
2975     uint32_t res;
2976 
2977     round = get_round(vxrm, a, shift);
2978     res   = (a >> shift)  + round;
2979     if (res > UINT16_MAX) {
2980         env->vxsat = 0x1;
2981         return UINT16_MAX;
2982     } else {
2983         return res;
2984     }
2985 }
2986 
2987 static inline uint32_t
2988 vnclipu32(CPURISCVState *env, int vxrm, uint64_t a, uint32_t b)
2989 {
2990     uint8_t round, shift = b & 0x3f;
2991     uint64_t res;
2992 
2993     round = get_round(vxrm, a, shift);
2994     res   = (a >> shift)  + round;
2995     if (res > UINT32_MAX) {
2996         env->vxsat = 0x1;
2997         return UINT32_MAX;
2998     } else {
2999         return res;
3000     }
3001 }
3002 
3003 RVVCALL(OPIVV2_RM, vnclipu_wv_b, NOP_UUU_B, H1, H2, H1, vnclipu8)
3004 RVVCALL(OPIVV2_RM, vnclipu_wv_h, NOP_UUU_H, H2, H4, H2, vnclipu16)
3005 RVVCALL(OPIVV2_RM, vnclipu_wv_w, NOP_UUU_W, H4, H8, H4, vnclipu32)
3006 GEN_VEXT_VV_RM(vnclipu_wv_b, 1)
3007 GEN_VEXT_VV_RM(vnclipu_wv_h, 2)
3008 GEN_VEXT_VV_RM(vnclipu_wv_w, 4)
3009 
3010 RVVCALL(OPIVX2_RM, vnclipu_wx_b, NOP_UUU_B, H1, H2, vnclipu8)
3011 RVVCALL(OPIVX2_RM, vnclipu_wx_h, NOP_UUU_H, H2, H4, vnclipu16)
3012 RVVCALL(OPIVX2_RM, vnclipu_wx_w, NOP_UUU_W, H4, H8, vnclipu32)
3013 GEN_VEXT_VX_RM(vnclipu_wx_b, 1)
3014 GEN_VEXT_VX_RM(vnclipu_wx_h, 2)
3015 GEN_VEXT_VX_RM(vnclipu_wx_w, 4)
3016 
3017 /*
3018  *** Vector Float Point Arithmetic Instructions
3019  */
3020 /* Vector Single-Width Floating-Point Add/Subtract Instructions */
3021 #define OPFVV2(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)   \
3022 static void do_##NAME(void *vd, void *vs1, void *vs2, int i,   \
3023                       CPURISCVState *env)                      \
3024 {                                                              \
3025     TX1 s1 = *((T1 *)vs1 + HS1(i));                            \
3026     TX2 s2 = *((T2 *)vs2 + HS2(i));                            \
3027     *((TD *)vd + HD(i)) = OP(s2, s1, &env->fp_status);         \
3028 }
3029 
3030 #define GEN_VEXT_VV_ENV(NAME, ESZ)                        \
3031 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
3032                   void *vs2, CPURISCVState *env,          \
3033                   uint32_t desc)                          \
3034 {                                                         \
3035     uint32_t vm = vext_vm(desc);                          \
3036     uint32_t vl = env->vl;                                \
3037     uint32_t total_elems =                                \
3038         vext_get_total_elems(env, desc, ESZ);             \
3039     uint32_t vta = vext_vta(desc);                        \
3040     uint32_t vma = vext_vma(desc);                        \
3041     uint32_t i;                                           \
3042                                                           \
3043     for (i = env->vstart; i < vl; i++) {                  \
3044         if (!vm && !vext_elem_mask(v0, i)) {              \
3045             /* set masked-off elements to 1s */           \
3046             vext_set_elems_1s(vd, vma, i * ESZ,           \
3047                               (i + 1) * ESZ);             \
3048             continue;                                     \
3049         }                                                 \
3050         do_##NAME(vd, vs1, vs2, i, env);                  \
3051     }                                                     \
3052     env->vstart = 0;                                      \
3053     /* set tail elements to 1s */                         \
3054     vext_set_elems_1s(vd, vta, vl * ESZ,                  \
3055                       total_elems * ESZ);                 \
3056 }
3057 
3058 RVVCALL(OPFVV2, vfadd_vv_h, OP_UUU_H, H2, H2, H2, float16_add)
3059 RVVCALL(OPFVV2, vfadd_vv_w, OP_UUU_W, H4, H4, H4, float32_add)
3060 RVVCALL(OPFVV2, vfadd_vv_d, OP_UUU_D, H8, H8, H8, float64_add)
3061 GEN_VEXT_VV_ENV(vfadd_vv_h, 2)
3062 GEN_VEXT_VV_ENV(vfadd_vv_w, 4)
3063 GEN_VEXT_VV_ENV(vfadd_vv_d, 8)
3064 
3065 #define OPFVF2(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)        \
3066 static void do_##NAME(void *vd, uint64_t s1, void *vs2, int i, \
3067                       CPURISCVState *env)                      \
3068 {                                                              \
3069     TX2 s2 = *((T2 *)vs2 + HS2(i));                            \
3070     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1, &env->fp_status);\
3071 }
3072 
3073 #define GEN_VEXT_VF(NAME, ESZ)                            \
3074 void HELPER(NAME)(void *vd, void *v0, uint64_t s1,        \
3075                   void *vs2, CPURISCVState *env,          \
3076                   uint32_t desc)                          \
3077 {                                                         \
3078     uint32_t vm = vext_vm(desc);                          \
3079     uint32_t vl = env->vl;                                \
3080     uint32_t total_elems =                                \
3081         vext_get_total_elems(env, desc, ESZ);              \
3082     uint32_t vta = vext_vta(desc);                        \
3083     uint32_t vma = vext_vma(desc);                        \
3084     uint32_t i;                                           \
3085                                                           \
3086     for (i = env->vstart; i < vl; i++) {                  \
3087         if (!vm && !vext_elem_mask(v0, i)) {              \
3088             /* set masked-off elements to 1s */           \
3089             vext_set_elems_1s(vd, vma, i * ESZ,           \
3090                               (i + 1) * ESZ);             \
3091             continue;                                     \
3092         }                                                 \
3093         do_##NAME(vd, s1, vs2, i, env);                   \
3094     }                                                     \
3095     env->vstart = 0;                                      \
3096     /* set tail elements to 1s */                         \
3097     vext_set_elems_1s(vd, vta, vl * ESZ,                  \
3098                       total_elems * ESZ);                 \
3099 }
3100 
3101 RVVCALL(OPFVF2, vfadd_vf_h, OP_UUU_H, H2, H2, float16_add)
3102 RVVCALL(OPFVF2, vfadd_vf_w, OP_UUU_W, H4, H4, float32_add)
3103 RVVCALL(OPFVF2, vfadd_vf_d, OP_UUU_D, H8, H8, float64_add)
3104 GEN_VEXT_VF(vfadd_vf_h, 2)
3105 GEN_VEXT_VF(vfadd_vf_w, 4)
3106 GEN_VEXT_VF(vfadd_vf_d, 8)
3107 
3108 RVVCALL(OPFVV2, vfsub_vv_h, OP_UUU_H, H2, H2, H2, float16_sub)
3109 RVVCALL(OPFVV2, vfsub_vv_w, OP_UUU_W, H4, H4, H4, float32_sub)
3110 RVVCALL(OPFVV2, vfsub_vv_d, OP_UUU_D, H8, H8, H8, float64_sub)
3111 GEN_VEXT_VV_ENV(vfsub_vv_h, 2)
3112 GEN_VEXT_VV_ENV(vfsub_vv_w, 4)
3113 GEN_VEXT_VV_ENV(vfsub_vv_d, 8)
3114 RVVCALL(OPFVF2, vfsub_vf_h, OP_UUU_H, H2, H2, float16_sub)
3115 RVVCALL(OPFVF2, vfsub_vf_w, OP_UUU_W, H4, H4, float32_sub)
3116 RVVCALL(OPFVF2, vfsub_vf_d, OP_UUU_D, H8, H8, float64_sub)
3117 GEN_VEXT_VF(vfsub_vf_h, 2)
3118 GEN_VEXT_VF(vfsub_vf_w, 4)
3119 GEN_VEXT_VF(vfsub_vf_d, 8)
3120 
3121 static uint16_t float16_rsub(uint16_t a, uint16_t b, float_status *s)
3122 {
3123     return float16_sub(b, a, s);
3124 }
3125 
3126 static uint32_t float32_rsub(uint32_t a, uint32_t b, float_status *s)
3127 {
3128     return float32_sub(b, a, s);
3129 }
3130 
3131 static uint64_t float64_rsub(uint64_t a, uint64_t b, float_status *s)
3132 {
3133     return float64_sub(b, a, s);
3134 }
3135 
3136 RVVCALL(OPFVF2, vfrsub_vf_h, OP_UUU_H, H2, H2, float16_rsub)
3137 RVVCALL(OPFVF2, vfrsub_vf_w, OP_UUU_W, H4, H4, float32_rsub)
3138 RVVCALL(OPFVF2, vfrsub_vf_d, OP_UUU_D, H8, H8, float64_rsub)
3139 GEN_VEXT_VF(vfrsub_vf_h, 2)
3140 GEN_VEXT_VF(vfrsub_vf_w, 4)
3141 GEN_VEXT_VF(vfrsub_vf_d, 8)
3142 
3143 /* Vector Widening Floating-Point Add/Subtract Instructions */
3144 static uint32_t vfwadd16(uint16_t a, uint16_t b, float_status *s)
3145 {
3146     return float32_add(float16_to_float32(a, true, s),
3147             float16_to_float32(b, true, s), s);
3148 }
3149 
3150 static uint64_t vfwadd32(uint32_t a, uint32_t b, float_status *s)
3151 {
3152     return float64_add(float32_to_float64(a, s),
3153             float32_to_float64(b, s), s);
3154 
3155 }
3156 
3157 RVVCALL(OPFVV2, vfwadd_vv_h, WOP_UUU_H, H4, H2, H2, vfwadd16)
3158 RVVCALL(OPFVV2, vfwadd_vv_w, WOP_UUU_W, H8, H4, H4, vfwadd32)
3159 GEN_VEXT_VV_ENV(vfwadd_vv_h, 4)
3160 GEN_VEXT_VV_ENV(vfwadd_vv_w, 8)
3161 RVVCALL(OPFVF2, vfwadd_vf_h, WOP_UUU_H, H4, H2, vfwadd16)
3162 RVVCALL(OPFVF2, vfwadd_vf_w, WOP_UUU_W, H8, H4, vfwadd32)
3163 GEN_VEXT_VF(vfwadd_vf_h, 4)
3164 GEN_VEXT_VF(vfwadd_vf_w, 8)
3165 
3166 static uint32_t vfwsub16(uint16_t a, uint16_t b, float_status *s)
3167 {
3168     return float32_sub(float16_to_float32(a, true, s),
3169             float16_to_float32(b, true, s), s);
3170 }
3171 
3172 static uint64_t vfwsub32(uint32_t a, uint32_t b, float_status *s)
3173 {
3174     return float64_sub(float32_to_float64(a, s),
3175             float32_to_float64(b, s), s);
3176 
3177 }
3178 
3179 RVVCALL(OPFVV2, vfwsub_vv_h, WOP_UUU_H, H4, H2, H2, vfwsub16)
3180 RVVCALL(OPFVV2, vfwsub_vv_w, WOP_UUU_W, H8, H4, H4, vfwsub32)
3181 GEN_VEXT_VV_ENV(vfwsub_vv_h, 4)
3182 GEN_VEXT_VV_ENV(vfwsub_vv_w, 8)
3183 RVVCALL(OPFVF2, vfwsub_vf_h, WOP_UUU_H, H4, H2, vfwsub16)
3184 RVVCALL(OPFVF2, vfwsub_vf_w, WOP_UUU_W, H8, H4, vfwsub32)
3185 GEN_VEXT_VF(vfwsub_vf_h, 4)
3186 GEN_VEXT_VF(vfwsub_vf_w, 8)
3187 
3188 static uint32_t vfwaddw16(uint32_t a, uint16_t b, float_status *s)
3189 {
3190     return float32_add(a, float16_to_float32(b, true, s), s);
3191 }
3192 
3193 static uint64_t vfwaddw32(uint64_t a, uint32_t b, float_status *s)
3194 {
3195     return float64_add(a, float32_to_float64(b, s), s);
3196 }
3197 
3198 RVVCALL(OPFVV2, vfwadd_wv_h, WOP_WUUU_H, H4, H2, H2, vfwaddw16)
3199 RVVCALL(OPFVV2, vfwadd_wv_w, WOP_WUUU_W, H8, H4, H4, vfwaddw32)
3200 GEN_VEXT_VV_ENV(vfwadd_wv_h, 4)
3201 GEN_VEXT_VV_ENV(vfwadd_wv_w, 8)
3202 RVVCALL(OPFVF2, vfwadd_wf_h, WOP_WUUU_H, H4, H2, vfwaddw16)
3203 RVVCALL(OPFVF2, vfwadd_wf_w, WOP_WUUU_W, H8, H4, vfwaddw32)
3204 GEN_VEXT_VF(vfwadd_wf_h, 4)
3205 GEN_VEXT_VF(vfwadd_wf_w, 8)
3206 
3207 static uint32_t vfwsubw16(uint32_t a, uint16_t b, float_status *s)
3208 {
3209     return float32_sub(a, float16_to_float32(b, true, s), s);
3210 }
3211 
3212 static uint64_t vfwsubw32(uint64_t a, uint32_t b, float_status *s)
3213 {
3214     return float64_sub(a, float32_to_float64(b, s), s);
3215 }
3216 
3217 RVVCALL(OPFVV2, vfwsub_wv_h, WOP_WUUU_H, H4, H2, H2, vfwsubw16)
3218 RVVCALL(OPFVV2, vfwsub_wv_w, WOP_WUUU_W, H8, H4, H4, vfwsubw32)
3219 GEN_VEXT_VV_ENV(vfwsub_wv_h, 4)
3220 GEN_VEXT_VV_ENV(vfwsub_wv_w, 8)
3221 RVVCALL(OPFVF2, vfwsub_wf_h, WOP_WUUU_H, H4, H2, vfwsubw16)
3222 RVVCALL(OPFVF2, vfwsub_wf_w, WOP_WUUU_W, H8, H4, vfwsubw32)
3223 GEN_VEXT_VF(vfwsub_wf_h, 4)
3224 GEN_VEXT_VF(vfwsub_wf_w, 8)
3225 
3226 /* Vector Single-Width Floating-Point Multiply/Divide Instructions */
3227 RVVCALL(OPFVV2, vfmul_vv_h, OP_UUU_H, H2, H2, H2, float16_mul)
3228 RVVCALL(OPFVV2, vfmul_vv_w, OP_UUU_W, H4, H4, H4, float32_mul)
3229 RVVCALL(OPFVV2, vfmul_vv_d, OP_UUU_D, H8, H8, H8, float64_mul)
3230 GEN_VEXT_VV_ENV(vfmul_vv_h, 2)
3231 GEN_VEXT_VV_ENV(vfmul_vv_w, 4)
3232 GEN_VEXT_VV_ENV(vfmul_vv_d, 8)
3233 RVVCALL(OPFVF2, vfmul_vf_h, OP_UUU_H, H2, H2, float16_mul)
3234 RVVCALL(OPFVF2, vfmul_vf_w, OP_UUU_W, H4, H4, float32_mul)
3235 RVVCALL(OPFVF2, vfmul_vf_d, OP_UUU_D, H8, H8, float64_mul)
3236 GEN_VEXT_VF(vfmul_vf_h, 2)
3237 GEN_VEXT_VF(vfmul_vf_w, 4)
3238 GEN_VEXT_VF(vfmul_vf_d, 8)
3239 
3240 RVVCALL(OPFVV2, vfdiv_vv_h, OP_UUU_H, H2, H2, H2, float16_div)
3241 RVVCALL(OPFVV2, vfdiv_vv_w, OP_UUU_W, H4, H4, H4, float32_div)
3242 RVVCALL(OPFVV2, vfdiv_vv_d, OP_UUU_D, H8, H8, H8, float64_div)
3243 GEN_VEXT_VV_ENV(vfdiv_vv_h, 2)
3244 GEN_VEXT_VV_ENV(vfdiv_vv_w, 4)
3245 GEN_VEXT_VV_ENV(vfdiv_vv_d, 8)
3246 RVVCALL(OPFVF2, vfdiv_vf_h, OP_UUU_H, H2, H2, float16_div)
3247 RVVCALL(OPFVF2, vfdiv_vf_w, OP_UUU_W, H4, H4, float32_div)
3248 RVVCALL(OPFVF2, vfdiv_vf_d, OP_UUU_D, H8, H8, float64_div)
3249 GEN_VEXT_VF(vfdiv_vf_h, 2)
3250 GEN_VEXT_VF(vfdiv_vf_w, 4)
3251 GEN_VEXT_VF(vfdiv_vf_d, 8)
3252 
3253 static uint16_t float16_rdiv(uint16_t a, uint16_t b, float_status *s)
3254 {
3255     return float16_div(b, a, s);
3256 }
3257 
3258 static uint32_t float32_rdiv(uint32_t a, uint32_t b, float_status *s)
3259 {
3260     return float32_div(b, a, s);
3261 }
3262 
3263 static uint64_t float64_rdiv(uint64_t a, uint64_t b, float_status *s)
3264 {
3265     return float64_div(b, a, s);
3266 }
3267 
3268 RVVCALL(OPFVF2, vfrdiv_vf_h, OP_UUU_H, H2, H2, float16_rdiv)
3269 RVVCALL(OPFVF2, vfrdiv_vf_w, OP_UUU_W, H4, H4, float32_rdiv)
3270 RVVCALL(OPFVF2, vfrdiv_vf_d, OP_UUU_D, H8, H8, float64_rdiv)
3271 GEN_VEXT_VF(vfrdiv_vf_h, 2)
3272 GEN_VEXT_VF(vfrdiv_vf_w, 4)
3273 GEN_VEXT_VF(vfrdiv_vf_d, 8)
3274 
3275 /* Vector Widening Floating-Point Multiply */
3276 static uint32_t vfwmul16(uint16_t a, uint16_t b, float_status *s)
3277 {
3278     return float32_mul(float16_to_float32(a, true, s),
3279             float16_to_float32(b, true, s), s);
3280 }
3281 
3282 static uint64_t vfwmul32(uint32_t a, uint32_t b, float_status *s)
3283 {
3284     return float64_mul(float32_to_float64(a, s),
3285             float32_to_float64(b, s), s);
3286 
3287 }
3288 RVVCALL(OPFVV2, vfwmul_vv_h, WOP_UUU_H, H4, H2, H2, vfwmul16)
3289 RVVCALL(OPFVV2, vfwmul_vv_w, WOP_UUU_W, H8, H4, H4, vfwmul32)
3290 GEN_VEXT_VV_ENV(vfwmul_vv_h, 4)
3291 GEN_VEXT_VV_ENV(vfwmul_vv_w, 8)
3292 RVVCALL(OPFVF2, vfwmul_vf_h, WOP_UUU_H, H4, H2, vfwmul16)
3293 RVVCALL(OPFVF2, vfwmul_vf_w, WOP_UUU_W, H8, H4, vfwmul32)
3294 GEN_VEXT_VF(vfwmul_vf_h, 4)
3295 GEN_VEXT_VF(vfwmul_vf_w, 8)
3296 
3297 /* Vector Single-Width Floating-Point Fused Multiply-Add Instructions */
3298 #define OPFVV3(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)       \
3299 static void do_##NAME(void *vd, void *vs1, void *vs2, int i,       \
3300         CPURISCVState *env)                                        \
3301 {                                                                  \
3302     TX1 s1 = *((T1 *)vs1 + HS1(i));                                \
3303     TX2 s2 = *((T2 *)vs2 + HS2(i));                                \
3304     TD d = *((TD *)vd + HD(i));                                    \
3305     *((TD *)vd + HD(i)) = OP(s2, s1, d, &env->fp_status);          \
3306 }
3307 
3308 static uint16_t fmacc16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3309 {
3310     return float16_muladd(a, b, d, 0, s);
3311 }
3312 
3313 static uint32_t fmacc32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3314 {
3315     return float32_muladd(a, b, d, 0, s);
3316 }
3317 
3318 static uint64_t fmacc64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3319 {
3320     return float64_muladd(a, b, d, 0, s);
3321 }
3322 
3323 RVVCALL(OPFVV3, vfmacc_vv_h, OP_UUU_H, H2, H2, H2, fmacc16)
3324 RVVCALL(OPFVV3, vfmacc_vv_w, OP_UUU_W, H4, H4, H4, fmacc32)
3325 RVVCALL(OPFVV3, vfmacc_vv_d, OP_UUU_D, H8, H8, H8, fmacc64)
3326 GEN_VEXT_VV_ENV(vfmacc_vv_h, 2)
3327 GEN_VEXT_VV_ENV(vfmacc_vv_w, 4)
3328 GEN_VEXT_VV_ENV(vfmacc_vv_d, 8)
3329 
3330 #define OPFVF3(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)           \
3331 static void do_##NAME(void *vd, uint64_t s1, void *vs2, int i,    \
3332         CPURISCVState *env)                                       \
3333 {                                                                 \
3334     TX2 s2 = *((T2 *)vs2 + HS2(i));                               \
3335     TD d = *((TD *)vd + HD(i));                                   \
3336     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1, d, &env->fp_status);\
3337 }
3338 
3339 RVVCALL(OPFVF3, vfmacc_vf_h, OP_UUU_H, H2, H2, fmacc16)
3340 RVVCALL(OPFVF3, vfmacc_vf_w, OP_UUU_W, H4, H4, fmacc32)
3341 RVVCALL(OPFVF3, vfmacc_vf_d, OP_UUU_D, H8, H8, fmacc64)
3342 GEN_VEXT_VF(vfmacc_vf_h, 2)
3343 GEN_VEXT_VF(vfmacc_vf_w, 4)
3344 GEN_VEXT_VF(vfmacc_vf_d, 8)
3345 
3346 static uint16_t fnmacc16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3347 {
3348     return float16_muladd(a, b, d,
3349             float_muladd_negate_c | float_muladd_negate_product, s);
3350 }
3351 
3352 static uint32_t fnmacc32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3353 {
3354     return float32_muladd(a, b, d,
3355             float_muladd_negate_c | float_muladd_negate_product, s);
3356 }
3357 
3358 static uint64_t fnmacc64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3359 {
3360     return float64_muladd(a, b, d,
3361             float_muladd_negate_c | float_muladd_negate_product, s);
3362 }
3363 
3364 RVVCALL(OPFVV3, vfnmacc_vv_h, OP_UUU_H, H2, H2, H2, fnmacc16)
3365 RVVCALL(OPFVV3, vfnmacc_vv_w, OP_UUU_W, H4, H4, H4, fnmacc32)
3366 RVVCALL(OPFVV3, vfnmacc_vv_d, OP_UUU_D, H8, H8, H8, fnmacc64)
3367 GEN_VEXT_VV_ENV(vfnmacc_vv_h, 2)
3368 GEN_VEXT_VV_ENV(vfnmacc_vv_w, 4)
3369 GEN_VEXT_VV_ENV(vfnmacc_vv_d, 8)
3370 RVVCALL(OPFVF3, vfnmacc_vf_h, OP_UUU_H, H2, H2, fnmacc16)
3371 RVVCALL(OPFVF3, vfnmacc_vf_w, OP_UUU_W, H4, H4, fnmacc32)
3372 RVVCALL(OPFVF3, vfnmacc_vf_d, OP_UUU_D, H8, H8, fnmacc64)
3373 GEN_VEXT_VF(vfnmacc_vf_h, 2)
3374 GEN_VEXT_VF(vfnmacc_vf_w, 4)
3375 GEN_VEXT_VF(vfnmacc_vf_d, 8)
3376 
3377 static uint16_t fmsac16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3378 {
3379     return float16_muladd(a, b, d, float_muladd_negate_c, s);
3380 }
3381 
3382 static uint32_t fmsac32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3383 {
3384     return float32_muladd(a, b, d, float_muladd_negate_c, s);
3385 }
3386 
3387 static uint64_t fmsac64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3388 {
3389     return float64_muladd(a, b, d, float_muladd_negate_c, s);
3390 }
3391 
3392 RVVCALL(OPFVV3, vfmsac_vv_h, OP_UUU_H, H2, H2, H2, fmsac16)
3393 RVVCALL(OPFVV3, vfmsac_vv_w, OP_UUU_W, H4, H4, H4, fmsac32)
3394 RVVCALL(OPFVV3, vfmsac_vv_d, OP_UUU_D, H8, H8, H8, fmsac64)
3395 GEN_VEXT_VV_ENV(vfmsac_vv_h, 2)
3396 GEN_VEXT_VV_ENV(vfmsac_vv_w, 4)
3397 GEN_VEXT_VV_ENV(vfmsac_vv_d, 8)
3398 RVVCALL(OPFVF3, vfmsac_vf_h, OP_UUU_H, H2, H2, fmsac16)
3399 RVVCALL(OPFVF3, vfmsac_vf_w, OP_UUU_W, H4, H4, fmsac32)
3400 RVVCALL(OPFVF3, vfmsac_vf_d, OP_UUU_D, H8, H8, fmsac64)
3401 GEN_VEXT_VF(vfmsac_vf_h, 2)
3402 GEN_VEXT_VF(vfmsac_vf_w, 4)
3403 GEN_VEXT_VF(vfmsac_vf_d, 8)
3404 
3405 static uint16_t fnmsac16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3406 {
3407     return float16_muladd(a, b, d, float_muladd_negate_product, s);
3408 }
3409 
3410 static uint32_t fnmsac32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3411 {
3412     return float32_muladd(a, b, d, float_muladd_negate_product, s);
3413 }
3414 
3415 static uint64_t fnmsac64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3416 {
3417     return float64_muladd(a, b, d, float_muladd_negate_product, s);
3418 }
3419 
3420 RVVCALL(OPFVV3, vfnmsac_vv_h, OP_UUU_H, H2, H2, H2, fnmsac16)
3421 RVVCALL(OPFVV3, vfnmsac_vv_w, OP_UUU_W, H4, H4, H4, fnmsac32)
3422 RVVCALL(OPFVV3, vfnmsac_vv_d, OP_UUU_D, H8, H8, H8, fnmsac64)
3423 GEN_VEXT_VV_ENV(vfnmsac_vv_h, 2)
3424 GEN_VEXT_VV_ENV(vfnmsac_vv_w, 4)
3425 GEN_VEXT_VV_ENV(vfnmsac_vv_d, 8)
3426 RVVCALL(OPFVF3, vfnmsac_vf_h, OP_UUU_H, H2, H2, fnmsac16)
3427 RVVCALL(OPFVF3, vfnmsac_vf_w, OP_UUU_W, H4, H4, fnmsac32)
3428 RVVCALL(OPFVF3, vfnmsac_vf_d, OP_UUU_D, H8, H8, fnmsac64)
3429 GEN_VEXT_VF(vfnmsac_vf_h, 2)
3430 GEN_VEXT_VF(vfnmsac_vf_w, 4)
3431 GEN_VEXT_VF(vfnmsac_vf_d, 8)
3432 
3433 static uint16_t fmadd16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3434 {
3435     return float16_muladd(d, b, a, 0, s);
3436 }
3437 
3438 static uint32_t fmadd32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3439 {
3440     return float32_muladd(d, b, a, 0, s);
3441 }
3442 
3443 static uint64_t fmadd64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3444 {
3445     return float64_muladd(d, b, a, 0, s);
3446 }
3447 
3448 RVVCALL(OPFVV3, vfmadd_vv_h, OP_UUU_H, H2, H2, H2, fmadd16)
3449 RVVCALL(OPFVV3, vfmadd_vv_w, OP_UUU_W, H4, H4, H4, fmadd32)
3450 RVVCALL(OPFVV3, vfmadd_vv_d, OP_UUU_D, H8, H8, H8, fmadd64)
3451 GEN_VEXT_VV_ENV(vfmadd_vv_h, 2)
3452 GEN_VEXT_VV_ENV(vfmadd_vv_w, 4)
3453 GEN_VEXT_VV_ENV(vfmadd_vv_d, 8)
3454 RVVCALL(OPFVF3, vfmadd_vf_h, OP_UUU_H, H2, H2, fmadd16)
3455 RVVCALL(OPFVF3, vfmadd_vf_w, OP_UUU_W, H4, H4, fmadd32)
3456 RVVCALL(OPFVF3, vfmadd_vf_d, OP_UUU_D, H8, H8, fmadd64)
3457 GEN_VEXT_VF(vfmadd_vf_h, 2)
3458 GEN_VEXT_VF(vfmadd_vf_w, 4)
3459 GEN_VEXT_VF(vfmadd_vf_d, 8)
3460 
3461 static uint16_t fnmadd16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3462 {
3463     return float16_muladd(d, b, a,
3464             float_muladd_negate_c | float_muladd_negate_product, s);
3465 }
3466 
3467 static uint32_t fnmadd32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3468 {
3469     return float32_muladd(d, b, a,
3470             float_muladd_negate_c | float_muladd_negate_product, s);
3471 }
3472 
3473 static uint64_t fnmadd64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3474 {
3475     return float64_muladd(d, b, a,
3476             float_muladd_negate_c | float_muladd_negate_product, s);
3477 }
3478 
3479 RVVCALL(OPFVV3, vfnmadd_vv_h, OP_UUU_H, H2, H2, H2, fnmadd16)
3480 RVVCALL(OPFVV3, vfnmadd_vv_w, OP_UUU_W, H4, H4, H4, fnmadd32)
3481 RVVCALL(OPFVV3, vfnmadd_vv_d, OP_UUU_D, H8, H8, H8, fnmadd64)
3482 GEN_VEXT_VV_ENV(vfnmadd_vv_h, 2)
3483 GEN_VEXT_VV_ENV(vfnmadd_vv_w, 4)
3484 GEN_VEXT_VV_ENV(vfnmadd_vv_d, 8)
3485 RVVCALL(OPFVF3, vfnmadd_vf_h, OP_UUU_H, H2, H2, fnmadd16)
3486 RVVCALL(OPFVF3, vfnmadd_vf_w, OP_UUU_W, H4, H4, fnmadd32)
3487 RVVCALL(OPFVF3, vfnmadd_vf_d, OP_UUU_D, H8, H8, fnmadd64)
3488 GEN_VEXT_VF(vfnmadd_vf_h, 2)
3489 GEN_VEXT_VF(vfnmadd_vf_w, 4)
3490 GEN_VEXT_VF(vfnmadd_vf_d, 8)
3491 
3492 static uint16_t fmsub16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3493 {
3494     return float16_muladd(d, b, a, float_muladd_negate_c, s);
3495 }
3496 
3497 static uint32_t fmsub32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3498 {
3499     return float32_muladd(d, b, a, float_muladd_negate_c, s);
3500 }
3501 
3502 static uint64_t fmsub64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3503 {
3504     return float64_muladd(d, b, a, float_muladd_negate_c, s);
3505 }
3506 
3507 RVVCALL(OPFVV3, vfmsub_vv_h, OP_UUU_H, H2, H2, H2, fmsub16)
3508 RVVCALL(OPFVV3, vfmsub_vv_w, OP_UUU_W, H4, H4, H4, fmsub32)
3509 RVVCALL(OPFVV3, vfmsub_vv_d, OP_UUU_D, H8, H8, H8, fmsub64)
3510 GEN_VEXT_VV_ENV(vfmsub_vv_h, 2)
3511 GEN_VEXT_VV_ENV(vfmsub_vv_w, 4)
3512 GEN_VEXT_VV_ENV(vfmsub_vv_d, 8)
3513 RVVCALL(OPFVF3, vfmsub_vf_h, OP_UUU_H, H2, H2, fmsub16)
3514 RVVCALL(OPFVF3, vfmsub_vf_w, OP_UUU_W, H4, H4, fmsub32)
3515 RVVCALL(OPFVF3, vfmsub_vf_d, OP_UUU_D, H8, H8, fmsub64)
3516 GEN_VEXT_VF(vfmsub_vf_h, 2)
3517 GEN_VEXT_VF(vfmsub_vf_w, 4)
3518 GEN_VEXT_VF(vfmsub_vf_d, 8)
3519 
3520 static uint16_t fnmsub16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3521 {
3522     return float16_muladd(d, b, a, float_muladd_negate_product, s);
3523 }
3524 
3525 static uint32_t fnmsub32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3526 {
3527     return float32_muladd(d, b, a, float_muladd_negate_product, s);
3528 }
3529 
3530 static uint64_t fnmsub64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3531 {
3532     return float64_muladd(d, b, a, float_muladd_negate_product, s);
3533 }
3534 
3535 RVVCALL(OPFVV3, vfnmsub_vv_h, OP_UUU_H, H2, H2, H2, fnmsub16)
3536 RVVCALL(OPFVV3, vfnmsub_vv_w, OP_UUU_W, H4, H4, H4, fnmsub32)
3537 RVVCALL(OPFVV3, vfnmsub_vv_d, OP_UUU_D, H8, H8, H8, fnmsub64)
3538 GEN_VEXT_VV_ENV(vfnmsub_vv_h, 2)
3539 GEN_VEXT_VV_ENV(vfnmsub_vv_w, 4)
3540 GEN_VEXT_VV_ENV(vfnmsub_vv_d, 8)
3541 RVVCALL(OPFVF3, vfnmsub_vf_h, OP_UUU_H, H2, H2, fnmsub16)
3542 RVVCALL(OPFVF3, vfnmsub_vf_w, OP_UUU_W, H4, H4, fnmsub32)
3543 RVVCALL(OPFVF3, vfnmsub_vf_d, OP_UUU_D, H8, H8, fnmsub64)
3544 GEN_VEXT_VF(vfnmsub_vf_h, 2)
3545 GEN_VEXT_VF(vfnmsub_vf_w, 4)
3546 GEN_VEXT_VF(vfnmsub_vf_d, 8)
3547 
3548 /* Vector Widening Floating-Point Fused Multiply-Add Instructions */
3549 static uint32_t fwmacc16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3550 {
3551     return float32_muladd(float16_to_float32(a, true, s),
3552                         float16_to_float32(b, true, s), d, 0, s);
3553 }
3554 
3555 static uint64_t fwmacc32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3556 {
3557     return float64_muladd(float32_to_float64(a, s),
3558                         float32_to_float64(b, s), d, 0, s);
3559 }
3560 
3561 RVVCALL(OPFVV3, vfwmacc_vv_h, WOP_UUU_H, H4, H2, H2, fwmacc16)
3562 RVVCALL(OPFVV3, vfwmacc_vv_w, WOP_UUU_W, H8, H4, H4, fwmacc32)
3563 GEN_VEXT_VV_ENV(vfwmacc_vv_h, 4)
3564 GEN_VEXT_VV_ENV(vfwmacc_vv_w, 8)
3565 RVVCALL(OPFVF3, vfwmacc_vf_h, WOP_UUU_H, H4, H2, fwmacc16)
3566 RVVCALL(OPFVF3, vfwmacc_vf_w, WOP_UUU_W, H8, H4, fwmacc32)
3567 GEN_VEXT_VF(vfwmacc_vf_h, 4)
3568 GEN_VEXT_VF(vfwmacc_vf_w, 8)
3569 
3570 static uint32_t fwnmacc16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3571 {
3572     return float32_muladd(float16_to_float32(a, true, s),
3573                         float16_to_float32(b, true, s), d,
3574                         float_muladd_negate_c | float_muladd_negate_product, s);
3575 }
3576 
3577 static uint64_t fwnmacc32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3578 {
3579     return float64_muladd(float32_to_float64(a, s),
3580                         float32_to_float64(b, s), d,
3581                         float_muladd_negate_c | float_muladd_negate_product, s);
3582 }
3583 
3584 RVVCALL(OPFVV3, vfwnmacc_vv_h, WOP_UUU_H, H4, H2, H2, fwnmacc16)
3585 RVVCALL(OPFVV3, vfwnmacc_vv_w, WOP_UUU_W, H8, H4, H4, fwnmacc32)
3586 GEN_VEXT_VV_ENV(vfwnmacc_vv_h, 4)
3587 GEN_VEXT_VV_ENV(vfwnmacc_vv_w, 8)
3588 RVVCALL(OPFVF3, vfwnmacc_vf_h, WOP_UUU_H, H4, H2, fwnmacc16)
3589 RVVCALL(OPFVF3, vfwnmacc_vf_w, WOP_UUU_W, H8, H4, fwnmacc32)
3590 GEN_VEXT_VF(vfwnmacc_vf_h, 4)
3591 GEN_VEXT_VF(vfwnmacc_vf_w, 8)
3592 
3593 static uint32_t fwmsac16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3594 {
3595     return float32_muladd(float16_to_float32(a, true, s),
3596                         float16_to_float32(b, true, s), d,
3597                         float_muladd_negate_c, s);
3598 }
3599 
3600 static uint64_t fwmsac32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3601 {
3602     return float64_muladd(float32_to_float64(a, s),
3603                         float32_to_float64(b, s), d,
3604                         float_muladd_negate_c, s);
3605 }
3606 
3607 RVVCALL(OPFVV3, vfwmsac_vv_h, WOP_UUU_H, H4, H2, H2, fwmsac16)
3608 RVVCALL(OPFVV3, vfwmsac_vv_w, WOP_UUU_W, H8, H4, H4, fwmsac32)
3609 GEN_VEXT_VV_ENV(vfwmsac_vv_h, 4)
3610 GEN_VEXT_VV_ENV(vfwmsac_vv_w, 8)
3611 RVVCALL(OPFVF3, vfwmsac_vf_h, WOP_UUU_H, H4, H2, fwmsac16)
3612 RVVCALL(OPFVF3, vfwmsac_vf_w, WOP_UUU_W, H8, H4, fwmsac32)
3613 GEN_VEXT_VF(vfwmsac_vf_h, 4)
3614 GEN_VEXT_VF(vfwmsac_vf_w, 8)
3615 
3616 static uint32_t fwnmsac16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3617 {
3618     return float32_muladd(float16_to_float32(a, true, s),
3619                         float16_to_float32(b, true, s), d,
3620                         float_muladd_negate_product, s);
3621 }
3622 
3623 static uint64_t fwnmsac32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3624 {
3625     return float64_muladd(float32_to_float64(a, s),
3626                         float32_to_float64(b, s), d,
3627                         float_muladd_negate_product, s);
3628 }
3629 
3630 RVVCALL(OPFVV3, vfwnmsac_vv_h, WOP_UUU_H, H4, H2, H2, fwnmsac16)
3631 RVVCALL(OPFVV3, vfwnmsac_vv_w, WOP_UUU_W, H8, H4, H4, fwnmsac32)
3632 GEN_VEXT_VV_ENV(vfwnmsac_vv_h, 4)
3633 GEN_VEXT_VV_ENV(vfwnmsac_vv_w, 8)
3634 RVVCALL(OPFVF3, vfwnmsac_vf_h, WOP_UUU_H, H4, H2, fwnmsac16)
3635 RVVCALL(OPFVF3, vfwnmsac_vf_w, WOP_UUU_W, H8, H4, fwnmsac32)
3636 GEN_VEXT_VF(vfwnmsac_vf_h, 4)
3637 GEN_VEXT_VF(vfwnmsac_vf_w, 8)
3638 
3639 /* Vector Floating-Point Square-Root Instruction */
3640 /* (TD, T2, TX2) */
3641 #define OP_UU_H uint16_t, uint16_t, uint16_t
3642 #define OP_UU_W uint32_t, uint32_t, uint32_t
3643 #define OP_UU_D uint64_t, uint64_t, uint64_t
3644 
3645 #define OPFVV1(NAME, TD, T2, TX2, HD, HS2, OP)        \
3646 static void do_##NAME(void *vd, void *vs2, int i,      \
3647         CPURISCVState *env)                            \
3648 {                                                      \
3649     TX2 s2 = *((T2 *)vs2 + HS2(i));                    \
3650     *((TD *)vd + HD(i)) = OP(s2, &env->fp_status);     \
3651 }
3652 
3653 #define GEN_VEXT_V_ENV(NAME, ESZ)                      \
3654 void HELPER(NAME)(void *vd, void *v0, void *vs2,       \
3655         CPURISCVState *env, uint32_t desc)             \
3656 {                                                      \
3657     uint32_t vm = vext_vm(desc);                       \
3658     uint32_t vl = env->vl;                             \
3659     uint32_t total_elems =                             \
3660         vext_get_total_elems(env, desc, ESZ);          \
3661     uint32_t vta = vext_vta(desc);                     \
3662     uint32_t vma = vext_vma(desc);                     \
3663     uint32_t i;                                        \
3664                                                        \
3665     if (vl == 0) {                                     \
3666         return;                                        \
3667     }                                                  \
3668     for (i = env->vstart; i < vl; i++) {               \
3669         if (!vm && !vext_elem_mask(v0, i)) {           \
3670             /* set masked-off elements to 1s */        \
3671             vext_set_elems_1s(vd, vma, i * ESZ,        \
3672                               (i + 1) * ESZ);          \
3673             continue;                                  \
3674         }                                              \
3675         do_##NAME(vd, vs2, i, env);                    \
3676     }                                                  \
3677     env->vstart = 0;                                   \
3678     vext_set_elems_1s(vd, vta, vl * ESZ,               \
3679                       total_elems * ESZ);              \
3680 }
3681 
3682 RVVCALL(OPFVV1, vfsqrt_v_h, OP_UU_H, H2, H2, float16_sqrt)
3683 RVVCALL(OPFVV1, vfsqrt_v_w, OP_UU_W, H4, H4, float32_sqrt)
3684 RVVCALL(OPFVV1, vfsqrt_v_d, OP_UU_D, H8, H8, float64_sqrt)
3685 GEN_VEXT_V_ENV(vfsqrt_v_h, 2)
3686 GEN_VEXT_V_ENV(vfsqrt_v_w, 4)
3687 GEN_VEXT_V_ENV(vfsqrt_v_d, 8)
3688 
3689 /*
3690  * Vector Floating-Point Reciprocal Square-Root Estimate Instruction
3691  *
3692  * Adapted from riscv-v-spec recip.c:
3693  * https://github.com/riscv/riscv-v-spec/blob/master/recip.c
3694  */
3695 static uint64_t frsqrt7(uint64_t f, int exp_size, int frac_size)
3696 {
3697     uint64_t sign = extract64(f, frac_size + exp_size, 1);
3698     uint64_t exp = extract64(f, frac_size, exp_size);
3699     uint64_t frac = extract64(f, 0, frac_size);
3700 
3701     const uint8_t lookup_table[] = {
3702         52, 51, 50, 48, 47, 46, 44, 43,
3703         42, 41, 40, 39, 38, 36, 35, 34,
3704         33, 32, 31, 30, 30, 29, 28, 27,
3705         26, 25, 24, 23, 23, 22, 21, 20,
3706         19, 19, 18, 17, 16, 16, 15, 14,
3707         14, 13, 12, 12, 11, 10, 10, 9,
3708         9, 8, 7, 7, 6, 6, 5, 4,
3709         4, 3, 3, 2, 2, 1, 1, 0,
3710         127, 125, 123, 121, 119, 118, 116, 114,
3711         113, 111, 109, 108, 106, 105, 103, 102,
3712         100, 99, 97, 96, 95, 93, 92, 91,
3713         90, 88, 87, 86, 85, 84, 83, 82,
3714         80, 79, 78, 77, 76, 75, 74, 73,
3715         72, 71, 70, 70, 69, 68, 67, 66,
3716         65, 64, 63, 63, 62, 61, 60, 59,
3717         59, 58, 57, 56, 56, 55, 54, 53
3718     };
3719     const int precision = 7;
3720 
3721     if (exp == 0 && frac != 0) { /* subnormal */
3722         /* Normalize the subnormal. */
3723         while (extract64(frac, frac_size - 1, 1) == 0) {
3724             exp--;
3725             frac <<= 1;
3726         }
3727 
3728         frac = (frac << 1) & MAKE_64BIT_MASK(0, frac_size);
3729     }
3730 
3731     int idx = ((exp & 1) << (precision - 1)) |
3732                 (frac >> (frac_size - precision + 1));
3733     uint64_t out_frac = (uint64_t)(lookup_table[idx]) <<
3734                             (frac_size - precision);
3735     uint64_t out_exp = (3 * MAKE_64BIT_MASK(0, exp_size - 1) + ~exp) / 2;
3736 
3737     uint64_t val = 0;
3738     val = deposit64(val, 0, frac_size, out_frac);
3739     val = deposit64(val, frac_size, exp_size, out_exp);
3740     val = deposit64(val, frac_size + exp_size, 1, sign);
3741     return val;
3742 }
3743 
3744 static float16 frsqrt7_h(float16 f, float_status *s)
3745 {
3746     int exp_size = 5, frac_size = 10;
3747     bool sign = float16_is_neg(f);
3748 
3749     /*
3750      * frsqrt7(sNaN) = canonical NaN
3751      * frsqrt7(-inf) = canonical NaN
3752      * frsqrt7(-normal) = canonical NaN
3753      * frsqrt7(-subnormal) = canonical NaN
3754      */
3755     if (float16_is_signaling_nan(f, s) ||
3756             (float16_is_infinity(f) && sign) ||
3757             (float16_is_normal(f) && sign) ||
3758             (float16_is_zero_or_denormal(f) && !float16_is_zero(f) && sign)) {
3759         s->float_exception_flags |= float_flag_invalid;
3760         return float16_default_nan(s);
3761     }
3762 
3763     /* frsqrt7(qNaN) = canonical NaN */
3764     if (float16_is_quiet_nan(f, s)) {
3765         return float16_default_nan(s);
3766     }
3767 
3768     /* frsqrt7(+-0) = +-inf */
3769     if (float16_is_zero(f)) {
3770         s->float_exception_flags |= float_flag_divbyzero;
3771         return float16_set_sign(float16_infinity, sign);
3772     }
3773 
3774     /* frsqrt7(+inf) = +0 */
3775     if (float16_is_infinity(f) && !sign) {
3776         return float16_set_sign(float16_zero, sign);
3777     }
3778 
3779     /* +normal, +subnormal */
3780     uint64_t val = frsqrt7(f, exp_size, frac_size);
3781     return make_float16(val);
3782 }
3783 
3784 static float32 frsqrt7_s(float32 f, float_status *s)
3785 {
3786     int exp_size = 8, frac_size = 23;
3787     bool sign = float32_is_neg(f);
3788 
3789     /*
3790      * frsqrt7(sNaN) = canonical NaN
3791      * frsqrt7(-inf) = canonical NaN
3792      * frsqrt7(-normal) = canonical NaN
3793      * frsqrt7(-subnormal) = canonical NaN
3794      */
3795     if (float32_is_signaling_nan(f, s) ||
3796             (float32_is_infinity(f) && sign) ||
3797             (float32_is_normal(f) && sign) ||
3798             (float32_is_zero_or_denormal(f) && !float32_is_zero(f) && sign)) {
3799         s->float_exception_flags |= float_flag_invalid;
3800         return float32_default_nan(s);
3801     }
3802 
3803     /* frsqrt7(qNaN) = canonical NaN */
3804     if (float32_is_quiet_nan(f, s)) {
3805         return float32_default_nan(s);
3806     }
3807 
3808     /* frsqrt7(+-0) = +-inf */
3809     if (float32_is_zero(f)) {
3810         s->float_exception_flags |= float_flag_divbyzero;
3811         return float32_set_sign(float32_infinity, sign);
3812     }
3813 
3814     /* frsqrt7(+inf) = +0 */
3815     if (float32_is_infinity(f) && !sign) {
3816         return float32_set_sign(float32_zero, sign);
3817     }
3818 
3819     /* +normal, +subnormal */
3820     uint64_t val = frsqrt7(f, exp_size, frac_size);
3821     return make_float32(val);
3822 }
3823 
3824 static float64 frsqrt7_d(float64 f, float_status *s)
3825 {
3826     int exp_size = 11, frac_size = 52;
3827     bool sign = float64_is_neg(f);
3828 
3829     /*
3830      * frsqrt7(sNaN) = canonical NaN
3831      * frsqrt7(-inf) = canonical NaN
3832      * frsqrt7(-normal) = canonical NaN
3833      * frsqrt7(-subnormal) = canonical NaN
3834      */
3835     if (float64_is_signaling_nan(f, s) ||
3836             (float64_is_infinity(f) && sign) ||
3837             (float64_is_normal(f) && sign) ||
3838             (float64_is_zero_or_denormal(f) && !float64_is_zero(f) && sign)) {
3839         s->float_exception_flags |= float_flag_invalid;
3840         return float64_default_nan(s);
3841     }
3842 
3843     /* frsqrt7(qNaN) = canonical NaN */
3844     if (float64_is_quiet_nan(f, s)) {
3845         return float64_default_nan(s);
3846     }
3847 
3848     /* frsqrt7(+-0) = +-inf */
3849     if (float64_is_zero(f)) {
3850         s->float_exception_flags |= float_flag_divbyzero;
3851         return float64_set_sign(float64_infinity, sign);
3852     }
3853 
3854     /* frsqrt7(+inf) = +0 */
3855     if (float64_is_infinity(f) && !sign) {
3856         return float64_set_sign(float64_zero, sign);
3857     }
3858 
3859     /* +normal, +subnormal */
3860     uint64_t val = frsqrt7(f, exp_size, frac_size);
3861     return make_float64(val);
3862 }
3863 
3864 RVVCALL(OPFVV1, vfrsqrt7_v_h, OP_UU_H, H2, H2, frsqrt7_h)
3865 RVVCALL(OPFVV1, vfrsqrt7_v_w, OP_UU_W, H4, H4, frsqrt7_s)
3866 RVVCALL(OPFVV1, vfrsqrt7_v_d, OP_UU_D, H8, H8, frsqrt7_d)
3867 GEN_VEXT_V_ENV(vfrsqrt7_v_h, 2)
3868 GEN_VEXT_V_ENV(vfrsqrt7_v_w, 4)
3869 GEN_VEXT_V_ENV(vfrsqrt7_v_d, 8)
3870 
3871 /*
3872  * Vector Floating-Point Reciprocal Estimate Instruction
3873  *
3874  * Adapted from riscv-v-spec recip.c:
3875  * https://github.com/riscv/riscv-v-spec/blob/master/recip.c
3876  */
3877 static uint64_t frec7(uint64_t f, int exp_size, int frac_size,
3878                       float_status *s)
3879 {
3880     uint64_t sign = extract64(f, frac_size + exp_size, 1);
3881     uint64_t exp = extract64(f, frac_size, exp_size);
3882     uint64_t frac = extract64(f, 0, frac_size);
3883 
3884     const uint8_t lookup_table[] = {
3885         127, 125, 123, 121, 119, 117, 116, 114,
3886         112, 110, 109, 107, 105, 104, 102, 100,
3887         99, 97, 96, 94, 93, 91, 90, 88,
3888         87, 85, 84, 83, 81, 80, 79, 77,
3889         76, 75, 74, 72, 71, 70, 69, 68,
3890         66, 65, 64, 63, 62, 61, 60, 59,
3891         58, 57, 56, 55, 54, 53, 52, 51,
3892         50, 49, 48, 47, 46, 45, 44, 43,
3893         42, 41, 40, 40, 39, 38, 37, 36,
3894         35, 35, 34, 33, 32, 31, 31, 30,
3895         29, 28, 28, 27, 26, 25, 25, 24,
3896         23, 23, 22, 21, 21, 20, 19, 19,
3897         18, 17, 17, 16, 15, 15, 14, 14,
3898         13, 12, 12, 11, 11, 10, 9, 9,
3899         8, 8, 7, 7, 6, 5, 5, 4,
3900         4, 3, 3, 2, 2, 1, 1, 0
3901     };
3902     const int precision = 7;
3903 
3904     if (exp == 0 && frac != 0) { /* subnormal */
3905         /* Normalize the subnormal. */
3906         while (extract64(frac, frac_size - 1, 1) == 0) {
3907             exp--;
3908             frac <<= 1;
3909         }
3910 
3911         frac = (frac << 1) & MAKE_64BIT_MASK(0, frac_size);
3912 
3913         if (exp != 0 && exp != UINT64_MAX) {
3914             /*
3915              * Overflow to inf or max value of same sign,
3916              * depending on sign and rounding mode.
3917              */
3918             s->float_exception_flags |= (float_flag_inexact |
3919                                          float_flag_overflow);
3920 
3921             if ((s->float_rounding_mode == float_round_to_zero) ||
3922                 ((s->float_rounding_mode == float_round_down) && !sign) ||
3923                 ((s->float_rounding_mode == float_round_up) && sign)) {
3924                 /* Return greatest/negative finite value. */
3925                 return (sign << (exp_size + frac_size)) |
3926                     (MAKE_64BIT_MASK(frac_size, exp_size) - 1);
3927             } else {
3928                 /* Return +-inf. */
3929                 return (sign << (exp_size + frac_size)) |
3930                     MAKE_64BIT_MASK(frac_size, exp_size);
3931             }
3932         }
3933     }
3934 
3935     int idx = frac >> (frac_size - precision);
3936     uint64_t out_frac = (uint64_t)(lookup_table[idx]) <<
3937                             (frac_size - precision);
3938     uint64_t out_exp = 2 * MAKE_64BIT_MASK(0, exp_size - 1) + ~exp;
3939 
3940     if (out_exp == 0 || out_exp == UINT64_MAX) {
3941         /*
3942          * The result is subnormal, but don't raise the underflow exception,
3943          * because there's no additional loss of precision.
3944          */
3945         out_frac = (out_frac >> 1) | MAKE_64BIT_MASK(frac_size - 1, 1);
3946         if (out_exp == UINT64_MAX) {
3947             out_frac >>= 1;
3948             out_exp = 0;
3949         }
3950     }
3951 
3952     uint64_t val = 0;
3953     val = deposit64(val, 0, frac_size, out_frac);
3954     val = deposit64(val, frac_size, exp_size, out_exp);
3955     val = deposit64(val, frac_size + exp_size, 1, sign);
3956     return val;
3957 }
3958 
3959 static float16 frec7_h(float16 f, float_status *s)
3960 {
3961     int exp_size = 5, frac_size = 10;
3962     bool sign = float16_is_neg(f);
3963 
3964     /* frec7(+-inf) = +-0 */
3965     if (float16_is_infinity(f)) {
3966         return float16_set_sign(float16_zero, sign);
3967     }
3968 
3969     /* frec7(+-0) = +-inf */
3970     if (float16_is_zero(f)) {
3971         s->float_exception_flags |= float_flag_divbyzero;
3972         return float16_set_sign(float16_infinity, sign);
3973     }
3974 
3975     /* frec7(sNaN) = canonical NaN */
3976     if (float16_is_signaling_nan(f, s)) {
3977         s->float_exception_flags |= float_flag_invalid;
3978         return float16_default_nan(s);
3979     }
3980 
3981     /* frec7(qNaN) = canonical NaN */
3982     if (float16_is_quiet_nan(f, s)) {
3983         return float16_default_nan(s);
3984     }
3985 
3986     /* +-normal, +-subnormal */
3987     uint64_t val = frec7(f, exp_size, frac_size, s);
3988     return make_float16(val);
3989 }
3990 
3991 static float32 frec7_s(float32 f, float_status *s)
3992 {
3993     int exp_size = 8, frac_size = 23;
3994     bool sign = float32_is_neg(f);
3995 
3996     /* frec7(+-inf) = +-0 */
3997     if (float32_is_infinity(f)) {
3998         return float32_set_sign(float32_zero, sign);
3999     }
4000 
4001     /* frec7(+-0) = +-inf */
4002     if (float32_is_zero(f)) {
4003         s->float_exception_flags |= float_flag_divbyzero;
4004         return float32_set_sign(float32_infinity, sign);
4005     }
4006 
4007     /* frec7(sNaN) = canonical NaN */
4008     if (float32_is_signaling_nan(f, s)) {
4009         s->float_exception_flags |= float_flag_invalid;
4010         return float32_default_nan(s);
4011     }
4012 
4013     /* frec7(qNaN) = canonical NaN */
4014     if (float32_is_quiet_nan(f, s)) {
4015         return float32_default_nan(s);
4016     }
4017 
4018     /* +-normal, +-subnormal */
4019     uint64_t val = frec7(f, exp_size, frac_size, s);
4020     return make_float32(val);
4021 }
4022 
4023 static float64 frec7_d(float64 f, float_status *s)
4024 {
4025     int exp_size = 11, frac_size = 52;
4026     bool sign = float64_is_neg(f);
4027 
4028     /* frec7(+-inf) = +-0 */
4029     if (float64_is_infinity(f)) {
4030         return float64_set_sign(float64_zero, sign);
4031     }
4032 
4033     /* frec7(+-0) = +-inf */
4034     if (float64_is_zero(f)) {
4035         s->float_exception_flags |= float_flag_divbyzero;
4036         return float64_set_sign(float64_infinity, sign);
4037     }
4038 
4039     /* frec7(sNaN) = canonical NaN */
4040     if (float64_is_signaling_nan(f, s)) {
4041         s->float_exception_flags |= float_flag_invalid;
4042         return float64_default_nan(s);
4043     }
4044 
4045     /* frec7(qNaN) = canonical NaN */
4046     if (float64_is_quiet_nan(f, s)) {
4047         return float64_default_nan(s);
4048     }
4049 
4050     /* +-normal, +-subnormal */
4051     uint64_t val = frec7(f, exp_size, frac_size, s);
4052     return make_float64(val);
4053 }
4054 
4055 RVVCALL(OPFVV1, vfrec7_v_h, OP_UU_H, H2, H2, frec7_h)
4056 RVVCALL(OPFVV1, vfrec7_v_w, OP_UU_W, H4, H4, frec7_s)
4057 RVVCALL(OPFVV1, vfrec7_v_d, OP_UU_D, H8, H8, frec7_d)
4058 GEN_VEXT_V_ENV(vfrec7_v_h, 2)
4059 GEN_VEXT_V_ENV(vfrec7_v_w, 4)
4060 GEN_VEXT_V_ENV(vfrec7_v_d, 8)
4061 
4062 /* Vector Floating-Point MIN/MAX Instructions */
4063 RVVCALL(OPFVV2, vfmin_vv_h, OP_UUU_H, H2, H2, H2, float16_minimum_number)
4064 RVVCALL(OPFVV2, vfmin_vv_w, OP_UUU_W, H4, H4, H4, float32_minimum_number)
4065 RVVCALL(OPFVV2, vfmin_vv_d, OP_UUU_D, H8, H8, H8, float64_minimum_number)
4066 GEN_VEXT_VV_ENV(vfmin_vv_h, 2)
4067 GEN_VEXT_VV_ENV(vfmin_vv_w, 4)
4068 GEN_VEXT_VV_ENV(vfmin_vv_d, 8)
4069 RVVCALL(OPFVF2, vfmin_vf_h, OP_UUU_H, H2, H2, float16_minimum_number)
4070 RVVCALL(OPFVF2, vfmin_vf_w, OP_UUU_W, H4, H4, float32_minimum_number)
4071 RVVCALL(OPFVF2, vfmin_vf_d, OP_UUU_D, H8, H8, float64_minimum_number)
4072 GEN_VEXT_VF(vfmin_vf_h, 2)
4073 GEN_VEXT_VF(vfmin_vf_w, 4)
4074 GEN_VEXT_VF(vfmin_vf_d, 8)
4075 
4076 RVVCALL(OPFVV2, vfmax_vv_h, OP_UUU_H, H2, H2, H2, float16_maximum_number)
4077 RVVCALL(OPFVV2, vfmax_vv_w, OP_UUU_W, H4, H4, H4, float32_maximum_number)
4078 RVVCALL(OPFVV2, vfmax_vv_d, OP_UUU_D, H8, H8, H8, float64_maximum_number)
4079 GEN_VEXT_VV_ENV(vfmax_vv_h, 2)
4080 GEN_VEXT_VV_ENV(vfmax_vv_w, 4)
4081 GEN_VEXT_VV_ENV(vfmax_vv_d, 8)
4082 RVVCALL(OPFVF2, vfmax_vf_h, OP_UUU_H, H2, H2, float16_maximum_number)
4083 RVVCALL(OPFVF2, vfmax_vf_w, OP_UUU_W, H4, H4, float32_maximum_number)
4084 RVVCALL(OPFVF2, vfmax_vf_d, OP_UUU_D, H8, H8, float64_maximum_number)
4085 GEN_VEXT_VF(vfmax_vf_h, 2)
4086 GEN_VEXT_VF(vfmax_vf_w, 4)
4087 GEN_VEXT_VF(vfmax_vf_d, 8)
4088 
4089 /* Vector Floating-Point Sign-Injection Instructions */
4090 static uint16_t fsgnj16(uint16_t a, uint16_t b, float_status *s)
4091 {
4092     return deposit64(b, 0, 15, a);
4093 }
4094 
4095 static uint32_t fsgnj32(uint32_t a, uint32_t b, float_status *s)
4096 {
4097     return deposit64(b, 0, 31, a);
4098 }
4099 
4100 static uint64_t fsgnj64(uint64_t a, uint64_t b, float_status *s)
4101 {
4102     return deposit64(b, 0, 63, a);
4103 }
4104 
4105 RVVCALL(OPFVV2, vfsgnj_vv_h, OP_UUU_H, H2, H2, H2, fsgnj16)
4106 RVVCALL(OPFVV2, vfsgnj_vv_w, OP_UUU_W, H4, H4, H4, fsgnj32)
4107 RVVCALL(OPFVV2, vfsgnj_vv_d, OP_UUU_D, H8, H8, H8, fsgnj64)
4108 GEN_VEXT_VV_ENV(vfsgnj_vv_h, 2)
4109 GEN_VEXT_VV_ENV(vfsgnj_vv_w, 4)
4110 GEN_VEXT_VV_ENV(vfsgnj_vv_d, 8)
4111 RVVCALL(OPFVF2, vfsgnj_vf_h, OP_UUU_H, H2, H2, fsgnj16)
4112 RVVCALL(OPFVF2, vfsgnj_vf_w, OP_UUU_W, H4, H4, fsgnj32)
4113 RVVCALL(OPFVF2, vfsgnj_vf_d, OP_UUU_D, H8, H8, fsgnj64)
4114 GEN_VEXT_VF(vfsgnj_vf_h, 2)
4115 GEN_VEXT_VF(vfsgnj_vf_w, 4)
4116 GEN_VEXT_VF(vfsgnj_vf_d, 8)
4117 
4118 static uint16_t fsgnjn16(uint16_t a, uint16_t b, float_status *s)
4119 {
4120     return deposit64(~b, 0, 15, a);
4121 }
4122 
4123 static uint32_t fsgnjn32(uint32_t a, uint32_t b, float_status *s)
4124 {
4125     return deposit64(~b, 0, 31, a);
4126 }
4127 
4128 static uint64_t fsgnjn64(uint64_t a, uint64_t b, float_status *s)
4129 {
4130     return deposit64(~b, 0, 63, a);
4131 }
4132 
4133 RVVCALL(OPFVV2, vfsgnjn_vv_h, OP_UUU_H, H2, H2, H2, fsgnjn16)
4134 RVVCALL(OPFVV2, vfsgnjn_vv_w, OP_UUU_W, H4, H4, H4, fsgnjn32)
4135 RVVCALL(OPFVV2, vfsgnjn_vv_d, OP_UUU_D, H8, H8, H8, fsgnjn64)
4136 GEN_VEXT_VV_ENV(vfsgnjn_vv_h, 2)
4137 GEN_VEXT_VV_ENV(vfsgnjn_vv_w, 4)
4138 GEN_VEXT_VV_ENV(vfsgnjn_vv_d, 8)
4139 RVVCALL(OPFVF2, vfsgnjn_vf_h, OP_UUU_H, H2, H2, fsgnjn16)
4140 RVVCALL(OPFVF2, vfsgnjn_vf_w, OP_UUU_W, H4, H4, fsgnjn32)
4141 RVVCALL(OPFVF2, vfsgnjn_vf_d, OP_UUU_D, H8, H8, fsgnjn64)
4142 GEN_VEXT_VF(vfsgnjn_vf_h, 2)
4143 GEN_VEXT_VF(vfsgnjn_vf_w, 4)
4144 GEN_VEXT_VF(vfsgnjn_vf_d, 8)
4145 
4146 static uint16_t fsgnjx16(uint16_t a, uint16_t b, float_status *s)
4147 {
4148     return deposit64(b ^ a, 0, 15, a);
4149 }
4150 
4151 static uint32_t fsgnjx32(uint32_t a, uint32_t b, float_status *s)
4152 {
4153     return deposit64(b ^ a, 0, 31, a);
4154 }
4155 
4156 static uint64_t fsgnjx64(uint64_t a, uint64_t b, float_status *s)
4157 {
4158     return deposit64(b ^ a, 0, 63, a);
4159 }
4160 
4161 RVVCALL(OPFVV2, vfsgnjx_vv_h, OP_UUU_H, H2, H2, H2, fsgnjx16)
4162 RVVCALL(OPFVV2, vfsgnjx_vv_w, OP_UUU_W, H4, H4, H4, fsgnjx32)
4163 RVVCALL(OPFVV2, vfsgnjx_vv_d, OP_UUU_D, H8, H8, H8, fsgnjx64)
4164 GEN_VEXT_VV_ENV(vfsgnjx_vv_h, 2)
4165 GEN_VEXT_VV_ENV(vfsgnjx_vv_w, 4)
4166 GEN_VEXT_VV_ENV(vfsgnjx_vv_d, 8)
4167 RVVCALL(OPFVF2, vfsgnjx_vf_h, OP_UUU_H, H2, H2, fsgnjx16)
4168 RVVCALL(OPFVF2, vfsgnjx_vf_w, OP_UUU_W, H4, H4, fsgnjx32)
4169 RVVCALL(OPFVF2, vfsgnjx_vf_d, OP_UUU_D, H8, H8, fsgnjx64)
4170 GEN_VEXT_VF(vfsgnjx_vf_h, 2)
4171 GEN_VEXT_VF(vfsgnjx_vf_w, 4)
4172 GEN_VEXT_VF(vfsgnjx_vf_d, 8)
4173 
4174 /* Vector Floating-Point Compare Instructions */
4175 #define GEN_VEXT_CMP_VV_ENV(NAME, ETYPE, H, DO_OP)            \
4176 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
4177                   CPURISCVState *env, uint32_t desc)          \
4178 {                                                             \
4179     uint32_t vm = vext_vm(desc);                              \
4180     uint32_t vl = env->vl;                                    \
4181     uint32_t total_elems = env_archcpu(env)->cfg.vlen;        \
4182     uint32_t vta_all_1s = vext_vta_all_1s(desc);              \
4183     uint32_t vma = vext_vma(desc);                            \
4184     uint32_t i;                                               \
4185                                                               \
4186     for (i = env->vstart; i < vl; i++) {                      \
4187         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
4188         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
4189         if (!vm && !vext_elem_mask(v0, i)) {                  \
4190             /* set masked-off elements to 1s */               \
4191             if (vma) {                                        \
4192                 vext_set_elem_mask(vd, i, 1);                 \
4193             }                                                 \
4194             continue;                                         \
4195         }                                                     \
4196         vext_set_elem_mask(vd, i,                             \
4197                            DO_OP(s2, s1, &env->fp_status));   \
4198     }                                                         \
4199     env->vstart = 0;                                          \
4200     /* mask destination register are always tail-agnostic */  \
4201     /* set tail elements to 1s */                             \
4202     if (vta_all_1s) {                                         \
4203         for (; i < total_elems; i++) {                        \
4204             vext_set_elem_mask(vd, i, 1);                     \
4205         }                                                     \
4206     }                                                         \
4207 }
4208 
4209 GEN_VEXT_CMP_VV_ENV(vmfeq_vv_h, uint16_t, H2, float16_eq_quiet)
4210 GEN_VEXT_CMP_VV_ENV(vmfeq_vv_w, uint32_t, H4, float32_eq_quiet)
4211 GEN_VEXT_CMP_VV_ENV(vmfeq_vv_d, uint64_t, H8, float64_eq_quiet)
4212 
4213 #define GEN_VEXT_CMP_VF(NAME, ETYPE, H, DO_OP)                      \
4214 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2,       \
4215                   CPURISCVState *env, uint32_t desc)                \
4216 {                                                                   \
4217     uint32_t vm = vext_vm(desc);                                    \
4218     uint32_t vl = env->vl;                                          \
4219     uint32_t total_elems = env_archcpu(env)->cfg.vlen;              \
4220     uint32_t vta_all_1s = vext_vta_all_1s(desc);                    \
4221     uint32_t vma = vext_vma(desc);                                  \
4222     uint32_t i;                                                     \
4223                                                                     \
4224     for (i = env->vstart; i < vl; i++) {                            \
4225         ETYPE s2 = *((ETYPE *)vs2 + H(i));                          \
4226         if (!vm && !vext_elem_mask(v0, i)) {                        \
4227             /* set masked-off elements to 1s */                     \
4228             if (vma) {                                              \
4229                 vext_set_elem_mask(vd, i, 1);                       \
4230             }                                                       \
4231             continue;                                               \
4232         }                                                           \
4233         vext_set_elem_mask(vd, i,                                   \
4234                            DO_OP(s2, (ETYPE)s1, &env->fp_status));  \
4235     }                                                               \
4236     env->vstart = 0;                                                \
4237     /* mask destination register are always tail-agnostic */        \
4238     /* set tail elements to 1s */                                   \
4239     if (vta_all_1s) {                                               \
4240         for (; i < total_elems; i++) {                              \
4241             vext_set_elem_mask(vd, i, 1);                           \
4242         }                                                           \
4243     }                                                               \
4244 }
4245 
4246 GEN_VEXT_CMP_VF(vmfeq_vf_h, uint16_t, H2, float16_eq_quiet)
4247 GEN_VEXT_CMP_VF(vmfeq_vf_w, uint32_t, H4, float32_eq_quiet)
4248 GEN_VEXT_CMP_VF(vmfeq_vf_d, uint64_t, H8, float64_eq_quiet)
4249 
4250 static bool vmfne16(uint16_t a, uint16_t b, float_status *s)
4251 {
4252     FloatRelation compare = float16_compare_quiet(a, b, s);
4253     return compare != float_relation_equal;
4254 }
4255 
4256 static bool vmfne32(uint32_t a, uint32_t b, float_status *s)
4257 {
4258     FloatRelation compare = float32_compare_quiet(a, b, s);
4259     return compare != float_relation_equal;
4260 }
4261 
4262 static bool vmfne64(uint64_t a, uint64_t b, float_status *s)
4263 {
4264     FloatRelation compare = float64_compare_quiet(a, b, s);
4265     return compare != float_relation_equal;
4266 }
4267 
4268 GEN_VEXT_CMP_VV_ENV(vmfne_vv_h, uint16_t, H2, vmfne16)
4269 GEN_VEXT_CMP_VV_ENV(vmfne_vv_w, uint32_t, H4, vmfne32)
4270 GEN_VEXT_CMP_VV_ENV(vmfne_vv_d, uint64_t, H8, vmfne64)
4271 GEN_VEXT_CMP_VF(vmfne_vf_h, uint16_t, H2, vmfne16)
4272 GEN_VEXT_CMP_VF(vmfne_vf_w, uint32_t, H4, vmfne32)
4273 GEN_VEXT_CMP_VF(vmfne_vf_d, uint64_t, H8, vmfne64)
4274 
4275 GEN_VEXT_CMP_VV_ENV(vmflt_vv_h, uint16_t, H2, float16_lt)
4276 GEN_VEXT_CMP_VV_ENV(vmflt_vv_w, uint32_t, H4, float32_lt)
4277 GEN_VEXT_CMP_VV_ENV(vmflt_vv_d, uint64_t, H8, float64_lt)
4278 GEN_VEXT_CMP_VF(vmflt_vf_h, uint16_t, H2, float16_lt)
4279 GEN_VEXT_CMP_VF(vmflt_vf_w, uint32_t, H4, float32_lt)
4280 GEN_VEXT_CMP_VF(vmflt_vf_d, uint64_t, H8, float64_lt)
4281 
4282 GEN_VEXT_CMP_VV_ENV(vmfle_vv_h, uint16_t, H2, float16_le)
4283 GEN_VEXT_CMP_VV_ENV(vmfle_vv_w, uint32_t, H4, float32_le)
4284 GEN_VEXT_CMP_VV_ENV(vmfle_vv_d, uint64_t, H8, float64_le)
4285 GEN_VEXT_CMP_VF(vmfle_vf_h, uint16_t, H2, float16_le)
4286 GEN_VEXT_CMP_VF(vmfle_vf_w, uint32_t, H4, float32_le)
4287 GEN_VEXT_CMP_VF(vmfle_vf_d, uint64_t, H8, float64_le)
4288 
4289 static bool vmfgt16(uint16_t a, uint16_t b, float_status *s)
4290 {
4291     FloatRelation compare = float16_compare(a, b, s);
4292     return compare == float_relation_greater;
4293 }
4294 
4295 static bool vmfgt32(uint32_t a, uint32_t b, float_status *s)
4296 {
4297     FloatRelation compare = float32_compare(a, b, s);
4298     return compare == float_relation_greater;
4299 }
4300 
4301 static bool vmfgt64(uint64_t a, uint64_t b, float_status *s)
4302 {
4303     FloatRelation compare = float64_compare(a, b, s);
4304     return compare == float_relation_greater;
4305 }
4306 
4307 GEN_VEXT_CMP_VF(vmfgt_vf_h, uint16_t, H2, vmfgt16)
4308 GEN_VEXT_CMP_VF(vmfgt_vf_w, uint32_t, H4, vmfgt32)
4309 GEN_VEXT_CMP_VF(vmfgt_vf_d, uint64_t, H8, vmfgt64)
4310 
4311 static bool vmfge16(uint16_t a, uint16_t b, float_status *s)
4312 {
4313     FloatRelation compare = float16_compare(a, b, s);
4314     return compare == float_relation_greater ||
4315            compare == float_relation_equal;
4316 }
4317 
4318 static bool vmfge32(uint32_t a, uint32_t b, float_status *s)
4319 {
4320     FloatRelation compare = float32_compare(a, b, s);
4321     return compare == float_relation_greater ||
4322            compare == float_relation_equal;
4323 }
4324 
4325 static bool vmfge64(uint64_t a, uint64_t b, float_status *s)
4326 {
4327     FloatRelation compare = float64_compare(a, b, s);
4328     return compare == float_relation_greater ||
4329            compare == float_relation_equal;
4330 }
4331 
4332 GEN_VEXT_CMP_VF(vmfge_vf_h, uint16_t, H2, vmfge16)
4333 GEN_VEXT_CMP_VF(vmfge_vf_w, uint32_t, H4, vmfge32)
4334 GEN_VEXT_CMP_VF(vmfge_vf_d, uint64_t, H8, vmfge64)
4335 
4336 /* Vector Floating-Point Classify Instruction */
4337 #define OPIVV1(NAME, TD, T2, TX2, HD, HS2, OP)         \
4338 static void do_##NAME(void *vd, void *vs2, int i)      \
4339 {                                                      \
4340     TX2 s2 = *((T2 *)vs2 + HS2(i));                    \
4341     *((TD *)vd + HD(i)) = OP(s2);                      \
4342 }
4343 
4344 #define GEN_VEXT_V(NAME, ESZ)                          \
4345 void HELPER(NAME)(void *vd, void *v0, void *vs2,       \
4346                   CPURISCVState *env, uint32_t desc)   \
4347 {                                                      \
4348     uint32_t vm = vext_vm(desc);                       \
4349     uint32_t vl = env->vl;                             \
4350     uint32_t total_elems =                             \
4351         vext_get_total_elems(env, desc, ESZ);          \
4352     uint32_t vta = vext_vta(desc);                     \
4353     uint32_t vma = vext_vma(desc);                     \
4354     uint32_t i;                                        \
4355                                                        \
4356     for (i = env->vstart; i < vl; i++) {               \
4357         if (!vm && !vext_elem_mask(v0, i)) {           \
4358             /* set masked-off elements to 1s */        \
4359             vext_set_elems_1s(vd, vma, i * ESZ,        \
4360                               (i + 1) * ESZ);          \
4361             continue;                                  \
4362         }                                              \
4363         do_##NAME(vd, vs2, i);                         \
4364     }                                                  \
4365     env->vstart = 0;                                   \
4366     /* set tail elements to 1s */                      \
4367     vext_set_elems_1s(vd, vta, vl * ESZ,               \
4368                       total_elems * ESZ);              \
4369 }
4370 
4371 target_ulong fclass_h(uint64_t frs1)
4372 {
4373     float16 f = frs1;
4374     bool sign = float16_is_neg(f);
4375 
4376     if (float16_is_infinity(f)) {
4377         return sign ? 1 << 0 : 1 << 7;
4378     } else if (float16_is_zero(f)) {
4379         return sign ? 1 << 3 : 1 << 4;
4380     } else if (float16_is_zero_or_denormal(f)) {
4381         return sign ? 1 << 2 : 1 << 5;
4382     } else if (float16_is_any_nan(f)) {
4383         float_status s = { }; /* for snan_bit_is_one */
4384         return float16_is_quiet_nan(f, &s) ? 1 << 9 : 1 << 8;
4385     } else {
4386         return sign ? 1 << 1 : 1 << 6;
4387     }
4388 }
4389 
4390 target_ulong fclass_s(uint64_t frs1)
4391 {
4392     float32 f = frs1;
4393     bool sign = float32_is_neg(f);
4394 
4395     if (float32_is_infinity(f)) {
4396         return sign ? 1 << 0 : 1 << 7;
4397     } else if (float32_is_zero(f)) {
4398         return sign ? 1 << 3 : 1 << 4;
4399     } else if (float32_is_zero_or_denormal(f)) {
4400         return sign ? 1 << 2 : 1 << 5;
4401     } else if (float32_is_any_nan(f)) {
4402         float_status s = { }; /* for snan_bit_is_one */
4403         return float32_is_quiet_nan(f, &s) ? 1 << 9 : 1 << 8;
4404     } else {
4405         return sign ? 1 << 1 : 1 << 6;
4406     }
4407 }
4408 
4409 target_ulong fclass_d(uint64_t frs1)
4410 {
4411     float64 f = frs1;
4412     bool sign = float64_is_neg(f);
4413 
4414     if (float64_is_infinity(f)) {
4415         return sign ? 1 << 0 : 1 << 7;
4416     } else if (float64_is_zero(f)) {
4417         return sign ? 1 << 3 : 1 << 4;
4418     } else if (float64_is_zero_or_denormal(f)) {
4419         return sign ? 1 << 2 : 1 << 5;
4420     } else if (float64_is_any_nan(f)) {
4421         float_status s = { }; /* for snan_bit_is_one */
4422         return float64_is_quiet_nan(f, &s) ? 1 << 9 : 1 << 8;
4423     } else {
4424         return sign ? 1 << 1 : 1 << 6;
4425     }
4426 }
4427 
4428 RVVCALL(OPIVV1, vfclass_v_h, OP_UU_H, H2, H2, fclass_h)
4429 RVVCALL(OPIVV1, vfclass_v_w, OP_UU_W, H4, H4, fclass_s)
4430 RVVCALL(OPIVV1, vfclass_v_d, OP_UU_D, H8, H8, fclass_d)
4431 GEN_VEXT_V(vfclass_v_h, 2)
4432 GEN_VEXT_V(vfclass_v_w, 4)
4433 GEN_VEXT_V(vfclass_v_d, 8)
4434 
4435 /* Vector Floating-Point Merge Instruction */
4436 
4437 #define GEN_VFMERGE_VF(NAME, ETYPE, H)                        \
4438 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2, \
4439                   CPURISCVState *env, uint32_t desc)          \
4440 {                                                             \
4441     uint32_t vm = vext_vm(desc);                              \
4442     uint32_t vl = env->vl;                                    \
4443     uint32_t esz = sizeof(ETYPE);                             \
4444     uint32_t total_elems =                                    \
4445         vext_get_total_elems(env, desc, esz);                 \
4446     uint32_t vta = vext_vta(desc);                            \
4447     uint32_t i;                                               \
4448                                                               \
4449     for (i = env->vstart; i < vl; i++) {                      \
4450         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
4451         *((ETYPE *)vd + H(i))                                 \
4452           = (!vm && !vext_elem_mask(v0, i) ? s2 : s1);        \
4453     }                                                         \
4454     env->vstart = 0;                                          \
4455     /* set tail elements to 1s */                             \
4456     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);  \
4457 }
4458 
4459 GEN_VFMERGE_VF(vfmerge_vfm_h, int16_t, H2)
4460 GEN_VFMERGE_VF(vfmerge_vfm_w, int32_t, H4)
4461 GEN_VFMERGE_VF(vfmerge_vfm_d, int64_t, H8)
4462 
4463 /* Single-Width Floating-Point/Integer Type-Convert Instructions */
4464 /* vfcvt.xu.f.v vd, vs2, vm # Convert float to unsigned integer. */
4465 RVVCALL(OPFVV1, vfcvt_xu_f_v_h, OP_UU_H, H2, H2, float16_to_uint16)
4466 RVVCALL(OPFVV1, vfcvt_xu_f_v_w, OP_UU_W, H4, H4, float32_to_uint32)
4467 RVVCALL(OPFVV1, vfcvt_xu_f_v_d, OP_UU_D, H8, H8, float64_to_uint64)
4468 GEN_VEXT_V_ENV(vfcvt_xu_f_v_h, 2)
4469 GEN_VEXT_V_ENV(vfcvt_xu_f_v_w, 4)
4470 GEN_VEXT_V_ENV(vfcvt_xu_f_v_d, 8)
4471 
4472 /* vfcvt.x.f.v vd, vs2, vm # Convert float to signed integer. */
4473 RVVCALL(OPFVV1, vfcvt_x_f_v_h, OP_UU_H, H2, H2, float16_to_int16)
4474 RVVCALL(OPFVV1, vfcvt_x_f_v_w, OP_UU_W, H4, H4, float32_to_int32)
4475 RVVCALL(OPFVV1, vfcvt_x_f_v_d, OP_UU_D, H8, H8, float64_to_int64)
4476 GEN_VEXT_V_ENV(vfcvt_x_f_v_h, 2)
4477 GEN_VEXT_V_ENV(vfcvt_x_f_v_w, 4)
4478 GEN_VEXT_V_ENV(vfcvt_x_f_v_d, 8)
4479 
4480 /* vfcvt.f.xu.v vd, vs2, vm # Convert unsigned integer to float. */
4481 RVVCALL(OPFVV1, vfcvt_f_xu_v_h, OP_UU_H, H2, H2, uint16_to_float16)
4482 RVVCALL(OPFVV1, vfcvt_f_xu_v_w, OP_UU_W, H4, H4, uint32_to_float32)
4483 RVVCALL(OPFVV1, vfcvt_f_xu_v_d, OP_UU_D, H8, H8, uint64_to_float64)
4484 GEN_VEXT_V_ENV(vfcvt_f_xu_v_h, 2)
4485 GEN_VEXT_V_ENV(vfcvt_f_xu_v_w, 4)
4486 GEN_VEXT_V_ENV(vfcvt_f_xu_v_d, 8)
4487 
4488 /* vfcvt.f.x.v vd, vs2, vm # Convert integer to float. */
4489 RVVCALL(OPFVV1, vfcvt_f_x_v_h, OP_UU_H, H2, H2, int16_to_float16)
4490 RVVCALL(OPFVV1, vfcvt_f_x_v_w, OP_UU_W, H4, H4, int32_to_float32)
4491 RVVCALL(OPFVV1, vfcvt_f_x_v_d, OP_UU_D, H8, H8, int64_to_float64)
4492 GEN_VEXT_V_ENV(vfcvt_f_x_v_h, 2)
4493 GEN_VEXT_V_ENV(vfcvt_f_x_v_w, 4)
4494 GEN_VEXT_V_ENV(vfcvt_f_x_v_d, 8)
4495 
4496 /* Widening Floating-Point/Integer Type-Convert Instructions */
4497 /* (TD, T2, TX2) */
4498 #define WOP_UU_B uint16_t, uint8_t,  uint8_t
4499 #define WOP_UU_H uint32_t, uint16_t, uint16_t
4500 #define WOP_UU_W uint64_t, uint32_t, uint32_t
4501 /* vfwcvt.xu.f.v vd, vs2, vm # Convert float to double-width unsigned integer.*/
4502 RVVCALL(OPFVV1, vfwcvt_xu_f_v_h, WOP_UU_H, H4, H2, float16_to_uint32)
4503 RVVCALL(OPFVV1, vfwcvt_xu_f_v_w, WOP_UU_W, H8, H4, float32_to_uint64)
4504 GEN_VEXT_V_ENV(vfwcvt_xu_f_v_h, 4)
4505 GEN_VEXT_V_ENV(vfwcvt_xu_f_v_w, 8)
4506 
4507 /* vfwcvt.x.f.v vd, vs2, vm # Convert float to double-width signed integer. */
4508 RVVCALL(OPFVV1, vfwcvt_x_f_v_h, WOP_UU_H, H4, H2, float16_to_int32)
4509 RVVCALL(OPFVV1, vfwcvt_x_f_v_w, WOP_UU_W, H8, H4, float32_to_int64)
4510 GEN_VEXT_V_ENV(vfwcvt_x_f_v_h, 4)
4511 GEN_VEXT_V_ENV(vfwcvt_x_f_v_w, 8)
4512 
4513 /* vfwcvt.f.xu.v vd, vs2, vm # Convert unsigned integer to double-width float */
4514 RVVCALL(OPFVV1, vfwcvt_f_xu_v_b, WOP_UU_B, H2, H1, uint8_to_float16)
4515 RVVCALL(OPFVV1, vfwcvt_f_xu_v_h, WOP_UU_H, H4, H2, uint16_to_float32)
4516 RVVCALL(OPFVV1, vfwcvt_f_xu_v_w, WOP_UU_W, H8, H4, uint32_to_float64)
4517 GEN_VEXT_V_ENV(vfwcvt_f_xu_v_b, 2)
4518 GEN_VEXT_V_ENV(vfwcvt_f_xu_v_h, 4)
4519 GEN_VEXT_V_ENV(vfwcvt_f_xu_v_w, 8)
4520 
4521 /* vfwcvt.f.x.v vd, vs2, vm # Convert integer to double-width float. */
4522 RVVCALL(OPFVV1, vfwcvt_f_x_v_b, WOP_UU_B, H2, H1, int8_to_float16)
4523 RVVCALL(OPFVV1, vfwcvt_f_x_v_h, WOP_UU_H, H4, H2, int16_to_float32)
4524 RVVCALL(OPFVV1, vfwcvt_f_x_v_w, WOP_UU_W, H8, H4, int32_to_float64)
4525 GEN_VEXT_V_ENV(vfwcvt_f_x_v_b, 2)
4526 GEN_VEXT_V_ENV(vfwcvt_f_x_v_h, 4)
4527 GEN_VEXT_V_ENV(vfwcvt_f_x_v_w, 8)
4528 
4529 /*
4530  * vfwcvt.f.f.v vd, vs2, vm
4531  * Convert single-width float to double-width float.
4532  */
4533 static uint32_t vfwcvtffv16(uint16_t a, float_status *s)
4534 {
4535     return float16_to_float32(a, true, s);
4536 }
4537 
4538 RVVCALL(OPFVV1, vfwcvt_f_f_v_h, WOP_UU_H, H4, H2, vfwcvtffv16)
4539 RVVCALL(OPFVV1, vfwcvt_f_f_v_w, WOP_UU_W, H8, H4, float32_to_float64)
4540 GEN_VEXT_V_ENV(vfwcvt_f_f_v_h, 4)
4541 GEN_VEXT_V_ENV(vfwcvt_f_f_v_w, 8)
4542 
4543 /* Narrowing Floating-Point/Integer Type-Convert Instructions */
4544 /* (TD, T2, TX2) */
4545 #define NOP_UU_B uint8_t,  uint16_t, uint32_t
4546 #define NOP_UU_H uint16_t, uint32_t, uint32_t
4547 #define NOP_UU_W uint32_t, uint64_t, uint64_t
4548 /* vfncvt.xu.f.v vd, vs2, vm # Convert float to unsigned integer. */
4549 RVVCALL(OPFVV1, vfncvt_xu_f_w_b, NOP_UU_B, H1, H2, float16_to_uint8)
4550 RVVCALL(OPFVV1, vfncvt_xu_f_w_h, NOP_UU_H, H2, H4, float32_to_uint16)
4551 RVVCALL(OPFVV1, vfncvt_xu_f_w_w, NOP_UU_W, H4, H8, float64_to_uint32)
4552 GEN_VEXT_V_ENV(vfncvt_xu_f_w_b, 1)
4553 GEN_VEXT_V_ENV(vfncvt_xu_f_w_h, 2)
4554 GEN_VEXT_V_ENV(vfncvt_xu_f_w_w, 4)
4555 
4556 /* vfncvt.x.f.v vd, vs2, vm # Convert double-width float to signed integer. */
4557 RVVCALL(OPFVV1, vfncvt_x_f_w_b, NOP_UU_B, H1, H2, float16_to_int8)
4558 RVVCALL(OPFVV1, vfncvt_x_f_w_h, NOP_UU_H, H2, H4, float32_to_int16)
4559 RVVCALL(OPFVV1, vfncvt_x_f_w_w, NOP_UU_W, H4, H8, float64_to_int32)
4560 GEN_VEXT_V_ENV(vfncvt_x_f_w_b, 1)
4561 GEN_VEXT_V_ENV(vfncvt_x_f_w_h, 2)
4562 GEN_VEXT_V_ENV(vfncvt_x_f_w_w, 4)
4563 
4564 /* vfncvt.f.xu.v vd, vs2, vm # Convert double-width unsigned integer to float */
4565 RVVCALL(OPFVV1, vfncvt_f_xu_w_h, NOP_UU_H, H2, H4, uint32_to_float16)
4566 RVVCALL(OPFVV1, vfncvt_f_xu_w_w, NOP_UU_W, H4, H8, uint64_to_float32)
4567 GEN_VEXT_V_ENV(vfncvt_f_xu_w_h, 2)
4568 GEN_VEXT_V_ENV(vfncvt_f_xu_w_w, 4)
4569 
4570 /* vfncvt.f.x.v vd, vs2, vm # Convert double-width integer to float. */
4571 RVVCALL(OPFVV1, vfncvt_f_x_w_h, NOP_UU_H, H2, H4, int32_to_float16)
4572 RVVCALL(OPFVV1, vfncvt_f_x_w_w, NOP_UU_W, H4, H8, int64_to_float32)
4573 GEN_VEXT_V_ENV(vfncvt_f_x_w_h, 2)
4574 GEN_VEXT_V_ENV(vfncvt_f_x_w_w, 4)
4575 
4576 /* vfncvt.f.f.v vd, vs2, vm # Convert double float to single-width float. */
4577 static uint16_t vfncvtffv16(uint32_t a, float_status *s)
4578 {
4579     return float32_to_float16(a, true, s);
4580 }
4581 
4582 RVVCALL(OPFVV1, vfncvt_f_f_w_h, NOP_UU_H, H2, H4, vfncvtffv16)
4583 RVVCALL(OPFVV1, vfncvt_f_f_w_w, NOP_UU_W, H4, H8, float64_to_float32)
4584 GEN_VEXT_V_ENV(vfncvt_f_f_w_h, 2)
4585 GEN_VEXT_V_ENV(vfncvt_f_f_w_w, 4)
4586 
4587 /*
4588  *** Vector Reduction Operations
4589  */
4590 /* Vector Single-Width Integer Reduction Instructions */
4591 #define GEN_VEXT_RED(NAME, TD, TS2, HD, HS2, OP)          \
4592 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
4593         void *vs2, CPURISCVState *env, uint32_t desc)     \
4594 {                                                         \
4595     uint32_t vm = vext_vm(desc);                          \
4596     uint32_t vl = env->vl;                                \
4597     uint32_t esz = sizeof(TD);                            \
4598     uint32_t vlenb = simd_maxsz(desc);                    \
4599     uint32_t vta = vext_vta(desc);                        \
4600     uint32_t i;                                           \
4601     TD s1 =  *((TD *)vs1 + HD(0));                        \
4602                                                           \
4603     for (i = env->vstart; i < vl; i++) {                  \
4604         TS2 s2 = *((TS2 *)vs2 + HS2(i));                  \
4605         if (!vm && !vext_elem_mask(v0, i)) {              \
4606             continue;                                     \
4607         }                                                 \
4608         s1 = OP(s1, (TD)s2);                              \
4609     }                                                     \
4610     *((TD *)vd + HD(0)) = s1;                             \
4611     env->vstart = 0;                                      \
4612     /* set tail elements to 1s */                         \
4613     vext_set_elems_1s(vd, vta, esz, vlenb);               \
4614 }
4615 
4616 /* vd[0] = sum(vs1[0], vs2[*]) */
4617 GEN_VEXT_RED(vredsum_vs_b, int8_t,  int8_t,  H1, H1, DO_ADD)
4618 GEN_VEXT_RED(vredsum_vs_h, int16_t, int16_t, H2, H2, DO_ADD)
4619 GEN_VEXT_RED(vredsum_vs_w, int32_t, int32_t, H4, H4, DO_ADD)
4620 GEN_VEXT_RED(vredsum_vs_d, int64_t, int64_t, H8, H8, DO_ADD)
4621 
4622 /* vd[0] = maxu(vs1[0], vs2[*]) */
4623 GEN_VEXT_RED(vredmaxu_vs_b, uint8_t,  uint8_t,  H1, H1, DO_MAX)
4624 GEN_VEXT_RED(vredmaxu_vs_h, uint16_t, uint16_t, H2, H2, DO_MAX)
4625 GEN_VEXT_RED(vredmaxu_vs_w, uint32_t, uint32_t, H4, H4, DO_MAX)
4626 GEN_VEXT_RED(vredmaxu_vs_d, uint64_t, uint64_t, H8, H8, DO_MAX)
4627 
4628 /* vd[0] = max(vs1[0], vs2[*]) */
4629 GEN_VEXT_RED(vredmax_vs_b, int8_t,  int8_t,  H1, H1, DO_MAX)
4630 GEN_VEXT_RED(vredmax_vs_h, int16_t, int16_t, H2, H2, DO_MAX)
4631 GEN_VEXT_RED(vredmax_vs_w, int32_t, int32_t, H4, H4, DO_MAX)
4632 GEN_VEXT_RED(vredmax_vs_d, int64_t, int64_t, H8, H8, DO_MAX)
4633 
4634 /* vd[0] = minu(vs1[0], vs2[*]) */
4635 GEN_VEXT_RED(vredminu_vs_b, uint8_t,  uint8_t,  H1, H1, DO_MIN)
4636 GEN_VEXT_RED(vredminu_vs_h, uint16_t, uint16_t, H2, H2, DO_MIN)
4637 GEN_VEXT_RED(vredminu_vs_w, uint32_t, uint32_t, H4, H4, DO_MIN)
4638 GEN_VEXT_RED(vredminu_vs_d, uint64_t, uint64_t, H8, H8, DO_MIN)
4639 
4640 /* vd[0] = min(vs1[0], vs2[*]) */
4641 GEN_VEXT_RED(vredmin_vs_b, int8_t,  int8_t,  H1, H1, DO_MIN)
4642 GEN_VEXT_RED(vredmin_vs_h, int16_t, int16_t, H2, H2, DO_MIN)
4643 GEN_VEXT_RED(vredmin_vs_w, int32_t, int32_t, H4, H4, DO_MIN)
4644 GEN_VEXT_RED(vredmin_vs_d, int64_t, int64_t, H8, H8, DO_MIN)
4645 
4646 /* vd[0] = and(vs1[0], vs2[*]) */
4647 GEN_VEXT_RED(vredand_vs_b, int8_t,  int8_t,  H1, H1, DO_AND)
4648 GEN_VEXT_RED(vredand_vs_h, int16_t, int16_t, H2, H2, DO_AND)
4649 GEN_VEXT_RED(vredand_vs_w, int32_t, int32_t, H4, H4, DO_AND)
4650 GEN_VEXT_RED(vredand_vs_d, int64_t, int64_t, H8, H8, DO_AND)
4651 
4652 /* vd[0] = or(vs1[0], vs2[*]) */
4653 GEN_VEXT_RED(vredor_vs_b, int8_t,  int8_t,  H1, H1, DO_OR)
4654 GEN_VEXT_RED(vredor_vs_h, int16_t, int16_t, H2, H2, DO_OR)
4655 GEN_VEXT_RED(vredor_vs_w, int32_t, int32_t, H4, H4, DO_OR)
4656 GEN_VEXT_RED(vredor_vs_d, int64_t, int64_t, H8, H8, DO_OR)
4657 
4658 /* vd[0] = xor(vs1[0], vs2[*]) */
4659 GEN_VEXT_RED(vredxor_vs_b, int8_t,  int8_t,  H1, H1, DO_XOR)
4660 GEN_VEXT_RED(vredxor_vs_h, int16_t, int16_t, H2, H2, DO_XOR)
4661 GEN_VEXT_RED(vredxor_vs_w, int32_t, int32_t, H4, H4, DO_XOR)
4662 GEN_VEXT_RED(vredxor_vs_d, int64_t, int64_t, H8, H8, DO_XOR)
4663 
4664 /* Vector Widening Integer Reduction Instructions */
4665 /* signed sum reduction into double-width accumulator */
4666 GEN_VEXT_RED(vwredsum_vs_b, int16_t, int8_t,  H2, H1, DO_ADD)
4667 GEN_VEXT_RED(vwredsum_vs_h, int32_t, int16_t, H4, H2, DO_ADD)
4668 GEN_VEXT_RED(vwredsum_vs_w, int64_t, int32_t, H8, H4, DO_ADD)
4669 
4670 /* Unsigned sum reduction into double-width accumulator */
4671 GEN_VEXT_RED(vwredsumu_vs_b, uint16_t, uint8_t,  H2, H1, DO_ADD)
4672 GEN_VEXT_RED(vwredsumu_vs_h, uint32_t, uint16_t, H4, H2, DO_ADD)
4673 GEN_VEXT_RED(vwredsumu_vs_w, uint64_t, uint32_t, H8, H4, DO_ADD)
4674 
4675 /* Vector Single-Width Floating-Point Reduction Instructions */
4676 #define GEN_VEXT_FRED(NAME, TD, TS2, HD, HS2, OP)          \
4677 void HELPER(NAME)(void *vd, void *v0, void *vs1,           \
4678                   void *vs2, CPURISCVState *env,           \
4679                   uint32_t desc)                           \
4680 {                                                          \
4681     uint32_t vm = vext_vm(desc);                           \
4682     uint32_t vl = env->vl;                                 \
4683     uint32_t esz = sizeof(TD);                             \
4684     uint32_t vlenb = simd_maxsz(desc);                     \
4685     uint32_t vta = vext_vta(desc);                         \
4686     uint32_t i;                                            \
4687     TD s1 =  *((TD *)vs1 + HD(0));                         \
4688                                                            \
4689     for (i = env->vstart; i < vl; i++) {                   \
4690         TS2 s2 = *((TS2 *)vs2 + HS2(i));                   \
4691         if (!vm && !vext_elem_mask(v0, i)) {               \
4692             continue;                                      \
4693         }                                                  \
4694         s1 = OP(s1, (TD)s2, &env->fp_status);              \
4695     }                                                      \
4696     *((TD *)vd + HD(0)) = s1;                              \
4697     env->vstart = 0;                                       \
4698     /* set tail elements to 1s */                          \
4699     vext_set_elems_1s(vd, vta, esz, vlenb);                \
4700 }
4701 
4702 /* Unordered sum */
4703 GEN_VEXT_FRED(vfredusum_vs_h, uint16_t, uint16_t, H2, H2, float16_add)
4704 GEN_VEXT_FRED(vfredusum_vs_w, uint32_t, uint32_t, H4, H4, float32_add)
4705 GEN_VEXT_FRED(vfredusum_vs_d, uint64_t, uint64_t, H8, H8, float64_add)
4706 
4707 /* Ordered sum */
4708 GEN_VEXT_FRED(vfredosum_vs_h, uint16_t, uint16_t, H2, H2, float16_add)
4709 GEN_VEXT_FRED(vfredosum_vs_w, uint32_t, uint32_t, H4, H4, float32_add)
4710 GEN_VEXT_FRED(vfredosum_vs_d, uint64_t, uint64_t, H8, H8, float64_add)
4711 
4712 /* Maximum value */
4713 GEN_VEXT_FRED(vfredmax_vs_h, uint16_t, uint16_t, H2, H2, float16_maximum_number)
4714 GEN_VEXT_FRED(vfredmax_vs_w, uint32_t, uint32_t, H4, H4, float32_maximum_number)
4715 GEN_VEXT_FRED(vfredmax_vs_d, uint64_t, uint64_t, H8, H8, float64_maximum_number)
4716 
4717 /* Minimum value */
4718 GEN_VEXT_FRED(vfredmin_vs_h, uint16_t, uint16_t, H2, H2, float16_minimum_number)
4719 GEN_VEXT_FRED(vfredmin_vs_w, uint32_t, uint32_t, H4, H4, float32_minimum_number)
4720 GEN_VEXT_FRED(vfredmin_vs_d, uint64_t, uint64_t, H8, H8, float64_minimum_number)
4721 
4722 /* Vector Widening Floating-Point Add Instructions */
4723 static uint32_t fwadd16(uint32_t a, uint16_t b, float_status *s)
4724 {
4725     return float32_add(a, float16_to_float32(b, true, s), s);
4726 }
4727 
4728 static uint64_t fwadd32(uint64_t a, uint32_t b, float_status *s)
4729 {
4730     return float64_add(a, float32_to_float64(b, s), s);
4731 }
4732 
4733 /* Vector Widening Floating-Point Reduction Instructions */
4734 /* Ordered/unordered reduce 2*SEW = 2*SEW + sum(promote(SEW)) */
4735 GEN_VEXT_FRED(vfwredusum_vs_h, uint32_t, uint16_t, H4, H2, fwadd16)
4736 GEN_VEXT_FRED(vfwredusum_vs_w, uint64_t, uint32_t, H8, H4, fwadd32)
4737 GEN_VEXT_FRED(vfwredosum_vs_h, uint32_t, uint16_t, H4, H2, fwadd16)
4738 GEN_VEXT_FRED(vfwredosum_vs_w, uint64_t, uint32_t, H8, H4, fwadd32)
4739 
4740 /*
4741  *** Vector Mask Operations
4742  */
4743 /* Vector Mask-Register Logical Instructions */
4744 #define GEN_VEXT_MASK_VV(NAME, OP)                        \
4745 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
4746                   void *vs2, CPURISCVState *env,          \
4747                   uint32_t desc)                          \
4748 {                                                         \
4749     uint32_t vl = env->vl;                                \
4750     uint32_t total_elems = env_archcpu(env)->cfg.vlen;    \
4751     uint32_t vta_all_1s = vext_vta_all_1s(desc);          \
4752     uint32_t i;                                           \
4753     int a, b;                                             \
4754                                                           \
4755     for (i = env->vstart; i < vl; i++) {                  \
4756         a = vext_elem_mask(vs1, i);                       \
4757         b = vext_elem_mask(vs2, i);                       \
4758         vext_set_elem_mask(vd, i, OP(b, a));              \
4759     }                                                     \
4760     env->vstart = 0;                                      \
4761     /* mask destination register are always tail-         \
4762      * agnostic                                           \
4763      */                                                   \
4764     /* set tail elements to 1s */                         \
4765     if (vta_all_1s) {                                     \
4766         for (; i < total_elems; i++) {                    \
4767             vext_set_elem_mask(vd, i, 1);                 \
4768         }                                                 \
4769     }                                                     \
4770 }
4771 
4772 #define DO_NAND(N, M)  (!(N & M))
4773 #define DO_ANDNOT(N, M)  (N & !M)
4774 #define DO_NOR(N, M)  (!(N | M))
4775 #define DO_ORNOT(N, M)  (N | !M)
4776 #define DO_XNOR(N, M)  (!(N ^ M))
4777 
4778 GEN_VEXT_MASK_VV(vmand_mm, DO_AND)
4779 GEN_VEXT_MASK_VV(vmnand_mm, DO_NAND)
4780 GEN_VEXT_MASK_VV(vmandn_mm, DO_ANDNOT)
4781 GEN_VEXT_MASK_VV(vmxor_mm, DO_XOR)
4782 GEN_VEXT_MASK_VV(vmor_mm, DO_OR)
4783 GEN_VEXT_MASK_VV(vmnor_mm, DO_NOR)
4784 GEN_VEXT_MASK_VV(vmorn_mm, DO_ORNOT)
4785 GEN_VEXT_MASK_VV(vmxnor_mm, DO_XNOR)
4786 
4787 /* Vector count population in mask vcpop */
4788 target_ulong HELPER(vcpop_m)(void *v0, void *vs2, CPURISCVState *env,
4789                              uint32_t desc)
4790 {
4791     target_ulong cnt = 0;
4792     uint32_t vm = vext_vm(desc);
4793     uint32_t vl = env->vl;
4794     int i;
4795 
4796     for (i = env->vstart; i < vl; i++) {
4797         if (vm || vext_elem_mask(v0, i)) {
4798             if (vext_elem_mask(vs2, i)) {
4799                 cnt++;
4800             }
4801         }
4802     }
4803     env->vstart = 0;
4804     return cnt;
4805 }
4806 
4807 /* vfirst find-first-set mask bit*/
4808 target_ulong HELPER(vfirst_m)(void *v0, void *vs2, CPURISCVState *env,
4809                               uint32_t desc)
4810 {
4811     uint32_t vm = vext_vm(desc);
4812     uint32_t vl = env->vl;
4813     int i;
4814 
4815     for (i = env->vstart; i < vl; i++) {
4816         if (vm || vext_elem_mask(v0, i)) {
4817             if (vext_elem_mask(vs2, i)) {
4818                 return i;
4819             }
4820         }
4821     }
4822     env->vstart = 0;
4823     return -1LL;
4824 }
4825 
4826 enum set_mask_type {
4827     ONLY_FIRST = 1,
4828     INCLUDE_FIRST,
4829     BEFORE_FIRST,
4830 };
4831 
4832 static void vmsetm(void *vd, void *v0, void *vs2, CPURISCVState *env,
4833                    uint32_t desc, enum set_mask_type type)
4834 {
4835     uint32_t vm = vext_vm(desc);
4836     uint32_t vl = env->vl;
4837     uint32_t total_elems = env_archcpu(env)->cfg.vlen;
4838     uint32_t vta_all_1s = vext_vta_all_1s(desc);
4839     uint32_t vma = vext_vma(desc);
4840     int i;
4841     bool first_mask_bit = false;
4842 
4843     for (i = env->vstart; i < vl; i++) {
4844         if (!vm && !vext_elem_mask(v0, i)) {
4845             /* set masked-off elements to 1s */
4846             if (vma) {
4847                 vext_set_elem_mask(vd, i, 1);
4848             }
4849             continue;
4850         }
4851         /* write a zero to all following active elements */
4852         if (first_mask_bit) {
4853             vext_set_elem_mask(vd, i, 0);
4854             continue;
4855         }
4856         if (vext_elem_mask(vs2, i)) {
4857             first_mask_bit = true;
4858             if (type == BEFORE_FIRST) {
4859                 vext_set_elem_mask(vd, i, 0);
4860             } else {
4861                 vext_set_elem_mask(vd, i, 1);
4862             }
4863         } else {
4864             if (type == ONLY_FIRST) {
4865                 vext_set_elem_mask(vd, i, 0);
4866             } else {
4867                 vext_set_elem_mask(vd, i, 1);
4868             }
4869         }
4870     }
4871     env->vstart = 0;
4872     /* mask destination register are always tail-agnostic */
4873     /* set tail elements to 1s */
4874     if (vta_all_1s) {
4875         for (; i < total_elems; i++) {
4876             vext_set_elem_mask(vd, i, 1);
4877         }
4878     }
4879 }
4880 
4881 void HELPER(vmsbf_m)(void *vd, void *v0, void *vs2, CPURISCVState *env,
4882                      uint32_t desc)
4883 {
4884     vmsetm(vd, v0, vs2, env, desc, BEFORE_FIRST);
4885 }
4886 
4887 void HELPER(vmsif_m)(void *vd, void *v0, void *vs2, CPURISCVState *env,
4888                      uint32_t desc)
4889 {
4890     vmsetm(vd, v0, vs2, env, desc, INCLUDE_FIRST);
4891 }
4892 
4893 void HELPER(vmsof_m)(void *vd, void *v0, void *vs2, CPURISCVState *env,
4894                      uint32_t desc)
4895 {
4896     vmsetm(vd, v0, vs2, env, desc, ONLY_FIRST);
4897 }
4898 
4899 /* Vector Iota Instruction */
4900 #define GEN_VEXT_VIOTA_M(NAME, ETYPE, H)                                  \
4901 void HELPER(NAME)(void *vd, void *v0, void *vs2, CPURISCVState *env,      \
4902                   uint32_t desc)                                          \
4903 {                                                                         \
4904     uint32_t vm = vext_vm(desc);                                          \
4905     uint32_t vl = env->vl;                                                \
4906     uint32_t esz = sizeof(ETYPE);                                         \
4907     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4908     uint32_t vta = vext_vta(desc);                                        \
4909     uint32_t vma = vext_vma(desc);                                        \
4910     uint32_t sum = 0;                                                     \
4911     int i;                                                                \
4912                                                                           \
4913     for (i = env->vstart; i < vl; i++) {                                  \
4914         if (!vm && !vext_elem_mask(v0, i)) {                              \
4915             /* set masked-off elements to 1s */                           \
4916             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4917             continue;                                                     \
4918         }                                                                 \
4919         *((ETYPE *)vd + H(i)) = sum;                                      \
4920         if (vext_elem_mask(vs2, i)) {                                     \
4921             sum++;                                                        \
4922         }                                                                 \
4923     }                                                                     \
4924     env->vstart = 0;                                                      \
4925     /* set tail elements to 1s */                                         \
4926     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4927 }
4928 
4929 GEN_VEXT_VIOTA_M(viota_m_b, uint8_t,  H1)
4930 GEN_VEXT_VIOTA_M(viota_m_h, uint16_t, H2)
4931 GEN_VEXT_VIOTA_M(viota_m_w, uint32_t, H4)
4932 GEN_VEXT_VIOTA_M(viota_m_d, uint64_t, H8)
4933 
4934 /* Vector Element Index Instruction */
4935 #define GEN_VEXT_VID_V(NAME, ETYPE, H)                                    \
4936 void HELPER(NAME)(void *vd, void *v0, CPURISCVState *env, uint32_t desc)  \
4937 {                                                                         \
4938     uint32_t vm = vext_vm(desc);                                          \
4939     uint32_t vl = env->vl;                                                \
4940     uint32_t esz = sizeof(ETYPE);                                         \
4941     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4942     uint32_t vta = vext_vta(desc);                                        \
4943     uint32_t vma = vext_vma(desc);                                        \
4944     int i;                                                                \
4945                                                                           \
4946     for (i = env->vstart; i < vl; i++) {                                  \
4947         if (!vm && !vext_elem_mask(v0, i)) {                              \
4948             /* set masked-off elements to 1s */                           \
4949             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4950             continue;                                                     \
4951         }                                                                 \
4952         *((ETYPE *)vd + H(i)) = i;                                        \
4953     }                                                                     \
4954     env->vstart = 0;                                                      \
4955     /* set tail elements to 1s */                                         \
4956     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4957 }
4958 
4959 GEN_VEXT_VID_V(vid_v_b, uint8_t,  H1)
4960 GEN_VEXT_VID_V(vid_v_h, uint16_t, H2)
4961 GEN_VEXT_VID_V(vid_v_w, uint32_t, H4)
4962 GEN_VEXT_VID_V(vid_v_d, uint64_t, H8)
4963 
4964 /*
4965  *** Vector Permutation Instructions
4966  */
4967 
4968 /* Vector Slide Instructions */
4969 #define GEN_VEXT_VSLIDEUP_VX(NAME, ETYPE, H)                              \
4970 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,         \
4971                   CPURISCVState *env, uint32_t desc)                      \
4972 {                                                                         \
4973     uint32_t vm = vext_vm(desc);                                          \
4974     uint32_t vl = env->vl;                                                \
4975     uint32_t esz = sizeof(ETYPE);                                         \
4976     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4977     uint32_t vta = vext_vta(desc);                                        \
4978     uint32_t vma = vext_vma(desc);                                        \
4979     target_ulong offset = s1, i_min, i;                                   \
4980                                                                           \
4981     i_min = MAX(env->vstart, offset);                                     \
4982     for (i = i_min; i < vl; i++) {                                        \
4983         if (!vm && !vext_elem_mask(v0, i)) {                              \
4984             /* set masked-off elements to 1s */                           \
4985             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4986             continue;                                                     \
4987         }                                                                 \
4988         *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i - offset));          \
4989     }                                                                     \
4990     /* set tail elements to 1s */                                         \
4991     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4992 }
4993 
4994 /* vslideup.vx vd, vs2, rs1, vm # vd[i+rs1] = vs2[i] */
4995 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_b, uint8_t,  H1)
4996 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_h, uint16_t, H2)
4997 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_w, uint32_t, H4)
4998 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_d, uint64_t, H8)
4999 
5000 #define GEN_VEXT_VSLIDEDOWN_VX(NAME, ETYPE, H)                            \
5001 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,         \
5002                   CPURISCVState *env, uint32_t desc)                      \
5003 {                                                                         \
5004     uint32_t vlmax = vext_max_elems(desc, ctzl(sizeof(ETYPE)));           \
5005     uint32_t vm = vext_vm(desc);                                          \
5006     uint32_t vl = env->vl;                                                \
5007     uint32_t esz = sizeof(ETYPE);                                         \
5008     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5009     uint32_t vta = vext_vta(desc);                                        \
5010     uint32_t vma = vext_vma(desc);                                        \
5011     target_ulong i_max, i;                                                \
5012                                                                           \
5013     i_max = MAX(MIN(s1 < vlmax ? vlmax - s1 : 0, vl), env->vstart);       \
5014     for (i = env->vstart; i < i_max; ++i) {                               \
5015         if (!vm && !vext_elem_mask(v0, i)) {                              \
5016             /* set masked-off elements to 1s */                           \
5017             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
5018             continue;                                                     \
5019         }                                                                 \
5020         *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i + s1));              \
5021     }                                                                     \
5022                                                                           \
5023     for (i = i_max; i < vl; ++i) {                                        \
5024         if (vm || vext_elem_mask(v0, i)) {                                \
5025             *((ETYPE *)vd + H(i)) = 0;                                    \
5026         }                                                                 \
5027     }                                                                     \
5028                                                                           \
5029     env->vstart = 0;                                                      \
5030     /* set tail elements to 1s */                                         \
5031     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5032 }
5033 
5034 /* vslidedown.vx vd, vs2, rs1, vm # vd[i] = vs2[i+rs1] */
5035 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_b, uint8_t,  H1)
5036 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_h, uint16_t, H2)
5037 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_w, uint32_t, H4)
5038 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_d, uint64_t, H8)
5039 
5040 #define GEN_VEXT_VSLIE1UP(BITWIDTH, H)                                      \
5041 static void vslide1up_##BITWIDTH(void *vd, void *v0, target_ulong s1,       \
5042                      void *vs2, CPURISCVState *env, uint32_t desc)          \
5043 {                                                                           \
5044     typedef uint##BITWIDTH##_t ETYPE;                                       \
5045     uint32_t vm = vext_vm(desc);                                            \
5046     uint32_t vl = env->vl;                                                  \
5047     uint32_t esz = sizeof(ETYPE);                                           \
5048     uint32_t total_elems = vext_get_total_elems(env, desc, esz);            \
5049     uint32_t vta = vext_vta(desc);                                          \
5050     uint32_t vma = vext_vma(desc);                                          \
5051     uint32_t i;                                                             \
5052                                                                             \
5053     for (i = env->vstart; i < vl; i++) {                                    \
5054         if (!vm && !vext_elem_mask(v0, i)) {                                \
5055             /* set masked-off elements to 1s */                             \
5056             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);             \
5057             continue;                                                       \
5058         }                                                                   \
5059         if (i == 0) {                                                       \
5060             *((ETYPE *)vd + H(i)) = s1;                                     \
5061         } else {                                                            \
5062             *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i - 1));             \
5063         }                                                                   \
5064     }                                                                       \
5065     env->vstart = 0;                                                        \
5066     /* set tail elements to 1s */                                           \
5067     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);                \
5068 }
5069 
5070 GEN_VEXT_VSLIE1UP(8,  H1)
5071 GEN_VEXT_VSLIE1UP(16, H2)
5072 GEN_VEXT_VSLIE1UP(32, H4)
5073 GEN_VEXT_VSLIE1UP(64, H8)
5074 
5075 #define GEN_VEXT_VSLIDE1UP_VX(NAME, BITWIDTH)                     \
5076 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2, \
5077                   CPURISCVState *env, uint32_t desc)              \
5078 {                                                                 \
5079     vslide1up_##BITWIDTH(vd, v0, s1, vs2, env, desc);             \
5080 }
5081 
5082 /* vslide1up.vx vd, vs2, rs1, vm # vd[0]=x[rs1], vd[i+1] = vs2[i] */
5083 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_b, 8)
5084 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_h, 16)
5085 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_w, 32)
5086 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_d, 64)
5087 
5088 #define GEN_VEXT_VSLIDE1DOWN(BITWIDTH, H)                                     \
5089 static void vslide1down_##BITWIDTH(void *vd, void *v0, target_ulong s1,       \
5090                        void *vs2, CPURISCVState *env, uint32_t desc)          \
5091 {                                                                             \
5092     typedef uint##BITWIDTH##_t ETYPE;                                         \
5093     uint32_t vm = vext_vm(desc);                                              \
5094     uint32_t vl = env->vl;                                                    \
5095     uint32_t esz = sizeof(ETYPE);                                             \
5096     uint32_t total_elems = vext_get_total_elems(env, desc, esz);              \
5097     uint32_t vta = vext_vta(desc);                                            \
5098     uint32_t vma = vext_vma(desc);                                            \
5099     uint32_t i;                                                               \
5100                                                                               \
5101     for (i = env->vstart; i < vl; i++) {                                      \
5102         if (!vm && !vext_elem_mask(v0, i)) {                                  \
5103             /* set masked-off elements to 1s */                               \
5104             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);               \
5105             continue;                                                         \
5106         }                                                                     \
5107         if (i == vl - 1) {                                                    \
5108             *((ETYPE *)vd + H(i)) = s1;                                       \
5109         } else {                                                              \
5110             *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i + 1));               \
5111         }                                                                     \
5112     }                                                                         \
5113     env->vstart = 0;                                                          \
5114     /* set tail elements to 1s */                                             \
5115     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);                  \
5116 }
5117 
5118 GEN_VEXT_VSLIDE1DOWN(8,  H1)
5119 GEN_VEXT_VSLIDE1DOWN(16, H2)
5120 GEN_VEXT_VSLIDE1DOWN(32, H4)
5121 GEN_VEXT_VSLIDE1DOWN(64, H8)
5122 
5123 #define GEN_VEXT_VSLIDE1DOWN_VX(NAME, BITWIDTH)                   \
5124 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2, \
5125                   CPURISCVState *env, uint32_t desc)              \
5126 {                                                                 \
5127     vslide1down_##BITWIDTH(vd, v0, s1, vs2, env, desc);           \
5128 }
5129 
5130 /* vslide1down.vx vd, vs2, rs1, vm # vd[i] = vs2[i+1], vd[vl-1]=x[rs1] */
5131 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_b, 8)
5132 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_h, 16)
5133 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_w, 32)
5134 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_d, 64)
5135 
5136 /* Vector Floating-Point Slide Instructions */
5137 #define GEN_VEXT_VFSLIDE1UP_VF(NAME, BITWIDTH)                \
5138 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2, \
5139                   CPURISCVState *env, uint32_t desc)          \
5140 {                                                             \
5141     vslide1up_##BITWIDTH(vd, v0, s1, vs2, env, desc);         \
5142 }
5143 
5144 /* vfslide1up.vf vd, vs2, rs1, vm # vd[0]=f[rs1], vd[i+1] = vs2[i] */
5145 GEN_VEXT_VFSLIDE1UP_VF(vfslide1up_vf_h, 16)
5146 GEN_VEXT_VFSLIDE1UP_VF(vfslide1up_vf_w, 32)
5147 GEN_VEXT_VFSLIDE1UP_VF(vfslide1up_vf_d, 64)
5148 
5149 #define GEN_VEXT_VFSLIDE1DOWN_VF(NAME, BITWIDTH)              \
5150 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2, \
5151                   CPURISCVState *env, uint32_t desc)          \
5152 {                                                             \
5153     vslide1down_##BITWIDTH(vd, v0, s1, vs2, env, desc);       \
5154 }
5155 
5156 /* vfslide1down.vf vd, vs2, rs1, vm # vd[i] = vs2[i+1], vd[vl-1]=f[rs1] */
5157 GEN_VEXT_VFSLIDE1DOWN_VF(vfslide1down_vf_h, 16)
5158 GEN_VEXT_VFSLIDE1DOWN_VF(vfslide1down_vf_w, 32)
5159 GEN_VEXT_VFSLIDE1DOWN_VF(vfslide1down_vf_d, 64)
5160 
5161 /* Vector Register Gather Instruction */
5162 #define GEN_VEXT_VRGATHER_VV(NAME, TS1, TS2, HS1, HS2)                    \
5163 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,               \
5164                   CPURISCVState *env, uint32_t desc)                      \
5165 {                                                                         \
5166     uint32_t vlmax = vext_max_elems(desc, ctzl(sizeof(TS2)));             \
5167     uint32_t vm = vext_vm(desc);                                          \
5168     uint32_t vl = env->vl;                                                \
5169     uint32_t esz = sizeof(TS2);                                           \
5170     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5171     uint32_t vta = vext_vta(desc);                                        \
5172     uint32_t vma = vext_vma(desc);                                        \
5173     uint64_t index;                                                       \
5174     uint32_t i;                                                           \
5175                                                                           \
5176     for (i = env->vstart; i < vl; i++) {                                  \
5177         if (!vm && !vext_elem_mask(v0, i)) {                              \
5178             /* set masked-off elements to 1s */                           \
5179             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
5180             continue;                                                     \
5181         }                                                                 \
5182         index = *((TS1 *)vs1 + HS1(i));                                   \
5183         if (index >= vlmax) {                                             \
5184             *((TS2 *)vd + HS2(i)) = 0;                                    \
5185         } else {                                                          \
5186             *((TS2 *)vd + HS2(i)) = *((TS2 *)vs2 + HS2(index));           \
5187         }                                                                 \
5188     }                                                                     \
5189     env->vstart = 0;                                                      \
5190     /* set tail elements to 1s */                                         \
5191     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5192 }
5193 
5194 /* vd[i] = (vs1[i] >= VLMAX) ? 0 : vs2[vs1[i]]; */
5195 GEN_VEXT_VRGATHER_VV(vrgather_vv_b, uint8_t,  uint8_t,  H1, H1)
5196 GEN_VEXT_VRGATHER_VV(vrgather_vv_h, uint16_t, uint16_t, H2, H2)
5197 GEN_VEXT_VRGATHER_VV(vrgather_vv_w, uint32_t, uint32_t, H4, H4)
5198 GEN_VEXT_VRGATHER_VV(vrgather_vv_d, uint64_t, uint64_t, H8, H8)
5199 
5200 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_b, uint16_t, uint8_t,  H2, H1)
5201 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_h, uint16_t, uint16_t, H2, H2)
5202 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_w, uint16_t, uint32_t, H2, H4)
5203 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_d, uint16_t, uint64_t, H2, H8)
5204 
5205 #define GEN_VEXT_VRGATHER_VX(NAME, ETYPE, H)                              \
5206 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,         \
5207                   CPURISCVState *env, uint32_t desc)                      \
5208 {                                                                         \
5209     uint32_t vlmax = vext_max_elems(desc, ctzl(sizeof(ETYPE)));           \
5210     uint32_t vm = vext_vm(desc);                                          \
5211     uint32_t vl = env->vl;                                                \
5212     uint32_t esz = sizeof(ETYPE);                                         \
5213     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5214     uint32_t vta = vext_vta(desc);                                        \
5215     uint32_t vma = vext_vma(desc);                                        \
5216     uint64_t index = s1;                                                  \
5217     uint32_t i;                                                           \
5218                                                                           \
5219     for (i = env->vstart; i < vl; i++) {                                  \
5220         if (!vm && !vext_elem_mask(v0, i)) {                              \
5221             /* set masked-off elements to 1s */                           \
5222             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
5223             continue;                                                     \
5224         }                                                                 \
5225         if (index >= vlmax) {                                             \
5226             *((ETYPE *)vd + H(i)) = 0;                                    \
5227         } else {                                                          \
5228             *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(index));           \
5229         }                                                                 \
5230     }                                                                     \
5231     env->vstart = 0;                                                      \
5232     /* set tail elements to 1s */                                         \
5233     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5234 }
5235 
5236 /* vd[i] = (x[rs1] >= VLMAX) ? 0 : vs2[rs1] */
5237 GEN_VEXT_VRGATHER_VX(vrgather_vx_b, uint8_t,  H1)
5238 GEN_VEXT_VRGATHER_VX(vrgather_vx_h, uint16_t, H2)
5239 GEN_VEXT_VRGATHER_VX(vrgather_vx_w, uint32_t, H4)
5240 GEN_VEXT_VRGATHER_VX(vrgather_vx_d, uint64_t, H8)
5241 
5242 /* Vector Compress Instruction */
5243 #define GEN_VEXT_VCOMPRESS_VM(NAME, ETYPE, H)                             \
5244 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,               \
5245                   CPURISCVState *env, uint32_t desc)                      \
5246 {                                                                         \
5247     uint32_t vl = env->vl;                                                \
5248     uint32_t esz = sizeof(ETYPE);                                         \
5249     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5250     uint32_t vta = vext_vta(desc);                                        \
5251     uint32_t num = 0, i;                                                  \
5252                                                                           \
5253     for (i = env->vstart; i < vl; i++) {                                  \
5254         if (!vext_elem_mask(vs1, i)) {                                    \
5255             continue;                                                     \
5256         }                                                                 \
5257         *((ETYPE *)vd + H(num)) = *((ETYPE *)vs2 + H(i));                 \
5258         num++;                                                            \
5259     }                                                                     \
5260     env->vstart = 0;                                                      \
5261     /* set tail elements to 1s */                                         \
5262     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5263 }
5264 
5265 /* Compress into vd elements of vs2 where vs1 is enabled */
5266 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_b, uint8_t,  H1)
5267 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_h, uint16_t, H2)
5268 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_w, uint32_t, H4)
5269 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_d, uint64_t, H8)
5270 
5271 /* Vector Whole Register Move */
5272 void HELPER(vmvr_v)(void *vd, void *vs2, CPURISCVState *env, uint32_t desc)
5273 {
5274     /* EEW = SEW */
5275     uint32_t maxsz = simd_maxsz(desc);
5276     uint32_t sewb = 1 << FIELD_EX64(env->vtype, VTYPE, VSEW);
5277     uint32_t startb = env->vstart * sewb;
5278     uint32_t i = startb;
5279 
5280     memcpy((uint8_t *)vd + H1(i),
5281            (uint8_t *)vs2 + H1(i),
5282            maxsz - startb);
5283 
5284     env->vstart = 0;
5285 }
5286 
5287 /* Vector Integer Extension */
5288 #define GEN_VEXT_INT_EXT(NAME, ETYPE, DTYPE, HD, HS1)            \
5289 void HELPER(NAME)(void *vd, void *v0, void *vs2,                 \
5290                   CPURISCVState *env, uint32_t desc)             \
5291 {                                                                \
5292     uint32_t vl = env->vl;                                       \
5293     uint32_t vm = vext_vm(desc);                                 \
5294     uint32_t esz = sizeof(ETYPE);                                \
5295     uint32_t total_elems = vext_get_total_elems(env, desc, esz); \
5296     uint32_t vta = vext_vta(desc);                               \
5297     uint32_t vma = vext_vma(desc);                               \
5298     uint32_t i;                                                  \
5299                                                                  \
5300     for (i = env->vstart; i < vl; i++) {                         \
5301         if (!vm && !vext_elem_mask(v0, i)) {                     \
5302             /* set masked-off elements to 1s */                  \
5303             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);  \
5304             continue;                                            \
5305         }                                                        \
5306         *((ETYPE *)vd + HD(i)) = *((DTYPE *)vs2 + HS1(i));       \
5307     }                                                            \
5308     env->vstart = 0;                                             \
5309     /* set tail elements to 1s */                                \
5310     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);     \
5311 }
5312 
5313 GEN_VEXT_INT_EXT(vzext_vf2_h, uint16_t, uint8_t,  H2, H1)
5314 GEN_VEXT_INT_EXT(vzext_vf2_w, uint32_t, uint16_t, H4, H2)
5315 GEN_VEXT_INT_EXT(vzext_vf2_d, uint64_t, uint32_t, H8, H4)
5316 GEN_VEXT_INT_EXT(vzext_vf4_w, uint32_t, uint8_t,  H4, H1)
5317 GEN_VEXT_INT_EXT(vzext_vf4_d, uint64_t, uint16_t, H8, H2)
5318 GEN_VEXT_INT_EXT(vzext_vf8_d, uint64_t, uint8_t,  H8, H1)
5319 
5320 GEN_VEXT_INT_EXT(vsext_vf2_h, int16_t, int8_t,  H2, H1)
5321 GEN_VEXT_INT_EXT(vsext_vf2_w, int32_t, int16_t, H4, H2)
5322 GEN_VEXT_INT_EXT(vsext_vf2_d, int64_t, int32_t, H8, H4)
5323 GEN_VEXT_INT_EXT(vsext_vf4_w, int32_t, int8_t,  H4, H1)
5324 GEN_VEXT_INT_EXT(vsext_vf4_d, int64_t, int16_t, H8, H2)
5325 GEN_VEXT_INT_EXT(vsext_vf8_d, int64_t, int8_t,  H8, H1)
5326