xref: /openbmc/qemu/target/riscv/vector_helper.c (revision 4fd71d19)
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/cpu_ldst.h"
26 #include "exec/page-protection.h"
27 #include "exec/helper-proto.h"
28 #include "fpu/softfloat.h"
29 #include "tcg/tcg-gvec-desc.h"
30 #include "internals.h"
31 #include "vector_internals.h"
32 #include <math.h>
33 
34 target_ulong HELPER(vsetvl)(CPURISCVState *env, target_ulong s1,
35                             target_ulong s2)
36 {
37     int vlmax, vl;
38     RISCVCPU *cpu = env_archcpu(env);
39     uint64_t vlmul = FIELD_EX64(s2, VTYPE, VLMUL);
40     uint8_t vsew = FIELD_EX64(s2, VTYPE, VSEW);
41     uint16_t sew = 8 << vsew;
42     uint8_t ediv = FIELD_EX64(s2, VTYPE, VEDIV);
43     int xlen = riscv_cpu_xlen(env);
44     bool vill = (s2 >> (xlen - 1)) & 0x1;
45     target_ulong reserved = s2 &
46                             MAKE_64BIT_MASK(R_VTYPE_RESERVED_SHIFT,
47                                             xlen - 1 - R_VTYPE_RESERVED_SHIFT);
48     uint16_t vlen = cpu->cfg.vlenb << 3;
49     int8_t lmul;
50 
51     if (vlmul & 4) {
52         /*
53          * Fractional LMUL, check:
54          *
55          * VLEN * LMUL >= SEW
56          * VLEN >> (8 - lmul) >= sew
57          * (vlenb << 3) >> (8 - lmul) >= sew
58          */
59         if (vlmul == 4 || (vlen >> (8 - vlmul)) < sew) {
60             vill = true;
61         }
62     }
63 
64     if ((sew > cpu->cfg.elen) || vill || (ediv != 0) || (reserved != 0)) {
65         /* only set vill bit. */
66         env->vill = 1;
67         env->vtype = 0;
68         env->vl = 0;
69         env->vstart = 0;
70         return 0;
71     }
72 
73     /* lmul encoded as in DisasContext::lmul */
74     lmul = sextract32(FIELD_EX64(s2, VTYPE, VLMUL), 0, 3);
75     vlmax = vext_get_vlmax(cpu->cfg.vlenb, vsew, lmul);
76     if (s1 <= vlmax) {
77         vl = s1;
78     } else {
79         vl = vlmax;
80     }
81     env->vl = vl;
82     env->vtype = s2;
83     env->vstart = 0;
84     env->vill = 0;
85     return vl;
86 }
87 
88 /*
89  * Get the maximum number of elements can be operated.
90  *
91  * log2_esz: log2 of element size in bytes.
92  */
93 static inline uint32_t vext_max_elems(uint32_t desc, uint32_t log2_esz)
94 {
95     /*
96      * As simd_desc support at most 2048 bytes, the max vlen is 1024 bits.
97      * so vlen in bytes (vlenb) is encoded as maxsz.
98      */
99     uint32_t vlenb = simd_maxsz(desc);
100 
101     /* Return VLMAX */
102     int scale = vext_lmul(desc) - log2_esz;
103     return scale < 0 ? vlenb >> -scale : vlenb << scale;
104 }
105 
106 static inline target_ulong adjust_addr(CPURISCVState *env, target_ulong addr)
107 {
108     return (addr & ~env->cur_pmmask) | env->cur_pmbase;
109 }
110 
111 /*
112  * This function checks watchpoint before real load operation.
113  *
114  * In system mode, the TLB API probe_access is enough for watchpoint check.
115  * In user mode, there is no watchpoint support now.
116  *
117  * It will trigger an exception if there is no mapping in TLB
118  * and page table walk can't fill the TLB entry. Then the guest
119  * software can return here after process the exception or never return.
120  */
121 static void probe_pages(CPURISCVState *env, target_ulong addr,
122                         target_ulong len, uintptr_t ra,
123                         MMUAccessType access_type)
124 {
125     target_ulong pagelen = -(addr | TARGET_PAGE_MASK);
126     target_ulong curlen = MIN(pagelen, len);
127     int mmu_index = riscv_env_mmu_index(env, false);
128 
129     probe_access(env, adjust_addr(env, addr), curlen, access_type,
130                  mmu_index, ra);
131     if (len > curlen) {
132         addr += curlen;
133         curlen = len - curlen;
134         probe_access(env, adjust_addr(env, addr), curlen, access_type,
135                      mmu_index, ra);
136     }
137 }
138 
139 static inline void vext_set_elem_mask(void *v0, int index,
140                                       uint8_t value)
141 {
142     int idx = index / 64;
143     int pos = index % 64;
144     uint64_t old = ((uint64_t *)v0)[idx];
145     ((uint64_t *)v0)[idx] = deposit64(old, pos, 1, value);
146 }
147 
148 /* elements operations for load and store */
149 typedef void vext_ldst_elem_fn(CPURISCVState *env, abi_ptr addr,
150                                uint32_t idx, void *vd, uintptr_t retaddr);
151 
152 #define GEN_VEXT_LD_ELEM(NAME, ETYPE, H, LDSUF)            \
153 static void NAME(CPURISCVState *env, abi_ptr addr,         \
154                  uint32_t idx, void *vd, uintptr_t retaddr)\
155 {                                                          \
156     ETYPE *cur = ((ETYPE *)vd + H(idx));                   \
157     *cur = cpu_##LDSUF##_data_ra(env, addr, retaddr);      \
158 }                                                          \
159 
160 GEN_VEXT_LD_ELEM(lde_b, int8_t,  H1, ldsb)
161 GEN_VEXT_LD_ELEM(lde_h, int16_t, H2, ldsw)
162 GEN_VEXT_LD_ELEM(lde_w, int32_t, H4, ldl)
163 GEN_VEXT_LD_ELEM(lde_d, int64_t, H8, ldq)
164 
165 #define GEN_VEXT_ST_ELEM(NAME, ETYPE, H, STSUF)            \
166 static void NAME(CPURISCVState *env, abi_ptr addr,         \
167                  uint32_t idx, void *vd, uintptr_t retaddr)\
168 {                                                          \
169     ETYPE data = *((ETYPE *)vd + H(idx));                  \
170     cpu_##STSUF##_data_ra(env, addr, data, retaddr);       \
171 }
172 
173 GEN_VEXT_ST_ELEM(ste_b, int8_t,  H1, stb)
174 GEN_VEXT_ST_ELEM(ste_h, int16_t, H2, stw)
175 GEN_VEXT_ST_ELEM(ste_w, int32_t, H4, stl)
176 GEN_VEXT_ST_ELEM(ste_d, int64_t, H8, stq)
177 
178 static void vext_set_tail_elems_1s(target_ulong vl, void *vd,
179                                    uint32_t desc, uint32_t nf,
180                                    uint32_t esz, uint32_t max_elems)
181 {
182     uint32_t vta = vext_vta(desc);
183     int k;
184 
185     if (vta == 0) {
186         return;
187     }
188 
189     for (k = 0; k < nf; ++k) {
190         vext_set_elems_1s(vd, vta, (k * max_elems + vl) * esz,
191                           (k * max_elems + max_elems) * esz);
192     }
193 }
194 
195 /*
196  * stride: access vector element from strided memory
197  */
198 static void
199 vext_ldst_stride(void *vd, void *v0, target_ulong base,
200                  target_ulong stride, CPURISCVState *env,
201                  uint32_t desc, uint32_t vm,
202                  vext_ldst_elem_fn *ldst_elem,
203                  uint32_t log2_esz, uintptr_t ra)
204 {
205     uint32_t i, k;
206     uint32_t nf = vext_nf(desc);
207     uint32_t max_elems = vext_max_elems(desc, log2_esz);
208     uint32_t esz = 1 << log2_esz;
209     uint32_t vma = vext_vma(desc);
210 
211     VSTART_CHECK_EARLY_EXIT(env);
212 
213     for (i = env->vstart; i < env->vl; env->vstart = ++i) {
214         k = 0;
215         while (k < nf) {
216             if (!vm && !vext_elem_mask(v0, i)) {
217                 /* set masked-off elements to 1s */
218                 vext_set_elems_1s(vd, vma, (i + k * max_elems) * esz,
219                                   (i + k * max_elems + 1) * esz);
220                 k++;
221                 continue;
222             }
223             target_ulong addr = base + stride * i + (k << log2_esz);
224             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
225             k++;
226         }
227     }
228     env->vstart = 0;
229 
230     vext_set_tail_elems_1s(env->vl, vd, desc, nf, esz, max_elems);
231 }
232 
233 #define GEN_VEXT_LD_STRIDE(NAME, ETYPE, LOAD_FN)                        \
234 void HELPER(NAME)(void *vd, void * v0, target_ulong base,               \
235                   target_ulong stride, CPURISCVState *env,              \
236                   uint32_t desc)                                        \
237 {                                                                       \
238     uint32_t vm = vext_vm(desc);                                        \
239     vext_ldst_stride(vd, v0, base, stride, env, desc, vm, LOAD_FN,      \
240                      ctzl(sizeof(ETYPE)), GETPC());                     \
241 }
242 
243 GEN_VEXT_LD_STRIDE(vlse8_v,  int8_t,  lde_b)
244 GEN_VEXT_LD_STRIDE(vlse16_v, int16_t, lde_h)
245 GEN_VEXT_LD_STRIDE(vlse32_v, int32_t, lde_w)
246 GEN_VEXT_LD_STRIDE(vlse64_v, int64_t, lde_d)
247 
248 #define GEN_VEXT_ST_STRIDE(NAME, ETYPE, STORE_FN)                       \
249 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
250                   target_ulong stride, CPURISCVState *env,              \
251                   uint32_t desc)                                        \
252 {                                                                       \
253     uint32_t vm = vext_vm(desc);                                        \
254     vext_ldst_stride(vd, v0, base, stride, env, desc, vm, STORE_FN,     \
255                      ctzl(sizeof(ETYPE)), GETPC());                     \
256 }
257 
258 GEN_VEXT_ST_STRIDE(vsse8_v,  int8_t,  ste_b)
259 GEN_VEXT_ST_STRIDE(vsse16_v, int16_t, ste_h)
260 GEN_VEXT_ST_STRIDE(vsse32_v, int32_t, ste_w)
261 GEN_VEXT_ST_STRIDE(vsse64_v, int64_t, ste_d)
262 
263 /*
264  * unit-stride: access elements stored contiguously in memory
265  */
266 
267 /* unmasked unit-stride load and store operation */
268 static void
269 vext_ldst_us(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc,
270              vext_ldst_elem_fn *ldst_elem, uint32_t log2_esz, uint32_t evl,
271              uintptr_t ra)
272 {
273     uint32_t i, k;
274     uint32_t nf = vext_nf(desc);
275     uint32_t max_elems = vext_max_elems(desc, log2_esz);
276     uint32_t esz = 1 << log2_esz;
277 
278     VSTART_CHECK_EARLY_EXIT(env);
279 
280     /* load bytes from guest memory */
281     for (i = env->vstart; i < evl; env->vstart = ++i) {
282         k = 0;
283         while (k < nf) {
284             target_ulong addr = base + ((i * nf + k) << log2_esz);
285             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
286             k++;
287         }
288     }
289     env->vstart = 0;
290 
291     vext_set_tail_elems_1s(evl, vd, desc, nf, esz, max_elems);
292 }
293 
294 /*
295  * masked unit-stride load and store operation will be a special case of
296  * stride, stride = NF * sizeof (ETYPE)
297  */
298 
299 #define GEN_VEXT_LD_US(NAME, ETYPE, LOAD_FN)                            \
300 void HELPER(NAME##_mask)(void *vd, void *v0, target_ulong base,         \
301                          CPURISCVState *env, uint32_t desc)             \
302 {                                                                       \
303     uint32_t stride = vext_nf(desc) << ctzl(sizeof(ETYPE));             \
304     vext_ldst_stride(vd, v0, base, stride, env, desc, false, LOAD_FN,   \
305                      ctzl(sizeof(ETYPE)), GETPC());                     \
306 }                                                                       \
307                                                                         \
308 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
309                   CPURISCVState *env, uint32_t desc)                    \
310 {                                                                       \
311     vext_ldst_us(vd, base, env, desc, LOAD_FN,                          \
312                  ctzl(sizeof(ETYPE)), env->vl, GETPC());                \
313 }
314 
315 GEN_VEXT_LD_US(vle8_v,  int8_t,  lde_b)
316 GEN_VEXT_LD_US(vle16_v, int16_t, lde_h)
317 GEN_VEXT_LD_US(vle32_v, int32_t, lde_w)
318 GEN_VEXT_LD_US(vle64_v, int64_t, lde_d)
319 
320 #define GEN_VEXT_ST_US(NAME, ETYPE, STORE_FN)                            \
321 void HELPER(NAME##_mask)(void *vd, void *v0, target_ulong base,          \
322                          CPURISCVState *env, uint32_t desc)              \
323 {                                                                        \
324     uint32_t stride = vext_nf(desc) << ctzl(sizeof(ETYPE));              \
325     vext_ldst_stride(vd, v0, base, stride, env, desc, false, STORE_FN,   \
326                      ctzl(sizeof(ETYPE)), GETPC());                      \
327 }                                                                        \
328                                                                          \
329 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                 \
330                   CPURISCVState *env, uint32_t desc)                     \
331 {                                                                        \
332     vext_ldst_us(vd, base, env, desc, STORE_FN,                          \
333                  ctzl(sizeof(ETYPE)), env->vl, GETPC());                 \
334 }
335 
336 GEN_VEXT_ST_US(vse8_v,  int8_t,  ste_b)
337 GEN_VEXT_ST_US(vse16_v, int16_t, ste_h)
338 GEN_VEXT_ST_US(vse32_v, int32_t, ste_w)
339 GEN_VEXT_ST_US(vse64_v, int64_t, ste_d)
340 
341 /*
342  * unit stride mask load and store, EEW = 1
343  */
344 void HELPER(vlm_v)(void *vd, void *v0, target_ulong base,
345                     CPURISCVState *env, uint32_t desc)
346 {
347     /* evl = ceil(vl/8) */
348     uint8_t evl = (env->vl + 7) >> 3;
349     vext_ldst_us(vd, base, env, desc, lde_b,
350                  0, evl, GETPC());
351 }
352 
353 void HELPER(vsm_v)(void *vd, void *v0, target_ulong base,
354                     CPURISCVState *env, uint32_t desc)
355 {
356     /* evl = ceil(vl/8) */
357     uint8_t evl = (env->vl + 7) >> 3;
358     vext_ldst_us(vd, base, env, desc, ste_b,
359                  0, evl, GETPC());
360 }
361 
362 /*
363  * index: access vector element from indexed memory
364  */
365 typedef target_ulong vext_get_index_addr(target_ulong base,
366         uint32_t idx, void *vs2);
367 
368 #define GEN_VEXT_GET_INDEX_ADDR(NAME, ETYPE, H)        \
369 static target_ulong NAME(target_ulong base,            \
370                          uint32_t idx, void *vs2)      \
371 {                                                      \
372     return (base + *((ETYPE *)vs2 + H(idx)));          \
373 }
374 
375 GEN_VEXT_GET_INDEX_ADDR(idx_b, uint8_t,  H1)
376 GEN_VEXT_GET_INDEX_ADDR(idx_h, uint16_t, H2)
377 GEN_VEXT_GET_INDEX_ADDR(idx_w, uint32_t, H4)
378 GEN_VEXT_GET_INDEX_ADDR(idx_d, uint64_t, H8)
379 
380 static inline void
381 vext_ldst_index(void *vd, void *v0, target_ulong base,
382                 void *vs2, CPURISCVState *env, uint32_t desc,
383                 vext_get_index_addr get_index_addr,
384                 vext_ldst_elem_fn *ldst_elem,
385                 uint32_t log2_esz, uintptr_t ra)
386 {
387     uint32_t i, k;
388     uint32_t nf = vext_nf(desc);
389     uint32_t vm = vext_vm(desc);
390     uint32_t max_elems = vext_max_elems(desc, log2_esz);
391     uint32_t esz = 1 << log2_esz;
392     uint32_t vma = vext_vma(desc);
393 
394     VSTART_CHECK_EARLY_EXIT(env);
395 
396     /* load bytes from guest memory */
397     for (i = env->vstart; i < env->vl; env->vstart = ++i) {
398         k = 0;
399         while (k < nf) {
400             if (!vm && !vext_elem_mask(v0, i)) {
401                 /* set masked-off elements to 1s */
402                 vext_set_elems_1s(vd, vma, (i + k * max_elems) * esz,
403                                   (i + k * max_elems + 1) * esz);
404                 k++;
405                 continue;
406             }
407             abi_ptr addr = get_index_addr(base, i, vs2) + (k << log2_esz);
408             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
409             k++;
410         }
411     }
412     env->vstart = 0;
413 
414     vext_set_tail_elems_1s(env->vl, vd, desc, nf, esz, max_elems);
415 }
416 
417 #define GEN_VEXT_LD_INDEX(NAME, ETYPE, INDEX_FN, LOAD_FN)                  \
418 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                   \
419                   void *vs2, CPURISCVState *env, uint32_t desc)            \
420 {                                                                          \
421     vext_ldst_index(vd, v0, base, vs2, env, desc, INDEX_FN,                \
422                     LOAD_FN, ctzl(sizeof(ETYPE)), GETPC());                \
423 }
424 
425 GEN_VEXT_LD_INDEX(vlxei8_8_v,   int8_t,  idx_b, lde_b)
426 GEN_VEXT_LD_INDEX(vlxei8_16_v,  int16_t, idx_b, lde_h)
427 GEN_VEXT_LD_INDEX(vlxei8_32_v,  int32_t, idx_b, lde_w)
428 GEN_VEXT_LD_INDEX(vlxei8_64_v,  int64_t, idx_b, lde_d)
429 GEN_VEXT_LD_INDEX(vlxei16_8_v,  int8_t,  idx_h, lde_b)
430 GEN_VEXT_LD_INDEX(vlxei16_16_v, int16_t, idx_h, lde_h)
431 GEN_VEXT_LD_INDEX(vlxei16_32_v, int32_t, idx_h, lde_w)
432 GEN_VEXT_LD_INDEX(vlxei16_64_v, int64_t, idx_h, lde_d)
433 GEN_VEXT_LD_INDEX(vlxei32_8_v,  int8_t,  idx_w, lde_b)
434 GEN_VEXT_LD_INDEX(vlxei32_16_v, int16_t, idx_w, lde_h)
435 GEN_VEXT_LD_INDEX(vlxei32_32_v, int32_t, idx_w, lde_w)
436 GEN_VEXT_LD_INDEX(vlxei32_64_v, int64_t, idx_w, lde_d)
437 GEN_VEXT_LD_INDEX(vlxei64_8_v,  int8_t,  idx_d, lde_b)
438 GEN_VEXT_LD_INDEX(vlxei64_16_v, int16_t, idx_d, lde_h)
439 GEN_VEXT_LD_INDEX(vlxei64_32_v, int32_t, idx_d, lde_w)
440 GEN_VEXT_LD_INDEX(vlxei64_64_v, int64_t, idx_d, lde_d)
441 
442 #define GEN_VEXT_ST_INDEX(NAME, ETYPE, INDEX_FN, STORE_FN)       \
443 void HELPER(NAME)(void *vd, void *v0, target_ulong base,         \
444                   void *vs2, CPURISCVState *env, uint32_t desc)  \
445 {                                                                \
446     vext_ldst_index(vd, v0, base, vs2, env, desc, INDEX_FN,      \
447                     STORE_FN, ctzl(sizeof(ETYPE)),               \
448                     GETPC());                                    \
449 }
450 
451 GEN_VEXT_ST_INDEX(vsxei8_8_v,   int8_t,  idx_b, ste_b)
452 GEN_VEXT_ST_INDEX(vsxei8_16_v,  int16_t, idx_b, ste_h)
453 GEN_VEXT_ST_INDEX(vsxei8_32_v,  int32_t, idx_b, ste_w)
454 GEN_VEXT_ST_INDEX(vsxei8_64_v,  int64_t, idx_b, ste_d)
455 GEN_VEXT_ST_INDEX(vsxei16_8_v,  int8_t,  idx_h, ste_b)
456 GEN_VEXT_ST_INDEX(vsxei16_16_v, int16_t, idx_h, ste_h)
457 GEN_VEXT_ST_INDEX(vsxei16_32_v, int32_t, idx_h, ste_w)
458 GEN_VEXT_ST_INDEX(vsxei16_64_v, int64_t, idx_h, ste_d)
459 GEN_VEXT_ST_INDEX(vsxei32_8_v,  int8_t,  idx_w, ste_b)
460 GEN_VEXT_ST_INDEX(vsxei32_16_v, int16_t, idx_w, ste_h)
461 GEN_VEXT_ST_INDEX(vsxei32_32_v, int32_t, idx_w, ste_w)
462 GEN_VEXT_ST_INDEX(vsxei32_64_v, int64_t, idx_w, ste_d)
463 GEN_VEXT_ST_INDEX(vsxei64_8_v,  int8_t,  idx_d, ste_b)
464 GEN_VEXT_ST_INDEX(vsxei64_16_v, int16_t, idx_d, ste_h)
465 GEN_VEXT_ST_INDEX(vsxei64_32_v, int32_t, idx_d, ste_w)
466 GEN_VEXT_ST_INDEX(vsxei64_64_v, int64_t, idx_d, ste_d)
467 
468 /*
469  * unit-stride fault-only-fisrt load instructions
470  */
471 static inline void
472 vext_ldff(void *vd, void *v0, target_ulong base,
473           CPURISCVState *env, uint32_t desc,
474           vext_ldst_elem_fn *ldst_elem,
475           uint32_t log2_esz, uintptr_t ra)
476 {
477     void *host;
478     uint32_t i, k, vl = 0;
479     uint32_t nf = vext_nf(desc);
480     uint32_t vm = vext_vm(desc);
481     uint32_t max_elems = vext_max_elems(desc, log2_esz);
482     uint32_t esz = 1 << log2_esz;
483     uint32_t vma = vext_vma(desc);
484     target_ulong addr, offset, remain;
485     int mmu_index = riscv_env_mmu_index(env, false);
486 
487     VSTART_CHECK_EARLY_EXIT(env);
488 
489     /* probe every access */
490     for (i = env->vstart; i < env->vl; i++) {
491         if (!vm && !vext_elem_mask(v0, i)) {
492             continue;
493         }
494         addr = adjust_addr(env, base + i * (nf << log2_esz));
495         if (i == 0) {
496             probe_pages(env, addr, nf << log2_esz, ra, MMU_DATA_LOAD);
497         } else {
498             /* if it triggers an exception, no need to check watchpoint */
499             remain = nf << log2_esz;
500             while (remain > 0) {
501                 offset = -(addr | TARGET_PAGE_MASK);
502                 host = tlb_vaddr_to_host(env, addr, MMU_DATA_LOAD, mmu_index);
503                 if (host) {
504 #ifdef CONFIG_USER_ONLY
505                     if (!page_check_range(addr, offset, PAGE_READ)) {
506                         vl = i;
507                         goto ProbeSuccess;
508                     }
509 #else
510                     probe_pages(env, addr, offset, ra, MMU_DATA_LOAD);
511 #endif
512                 } else {
513                     vl = i;
514                     goto ProbeSuccess;
515                 }
516                 if (remain <=  offset) {
517                     break;
518                 }
519                 remain -= offset;
520                 addr = adjust_addr(env, addr + offset);
521             }
522         }
523     }
524 ProbeSuccess:
525     /* load bytes from guest memory */
526     if (vl != 0) {
527         env->vl = vl;
528     }
529     for (i = env->vstart; i < env->vl; i++) {
530         k = 0;
531         while (k < nf) {
532             if (!vm && !vext_elem_mask(v0, i)) {
533                 /* set masked-off elements to 1s */
534                 vext_set_elems_1s(vd, vma, (i + k * max_elems) * esz,
535                                   (i + k * max_elems + 1) * esz);
536                 k++;
537                 continue;
538             }
539             addr = base + ((i * nf + k) << log2_esz);
540             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
541             k++;
542         }
543     }
544     env->vstart = 0;
545 
546     vext_set_tail_elems_1s(env->vl, vd, desc, nf, esz, max_elems);
547 }
548 
549 #define GEN_VEXT_LDFF(NAME, ETYPE, LOAD_FN)               \
550 void HELPER(NAME)(void *vd, void *v0, target_ulong base,  \
551                   CPURISCVState *env, uint32_t desc)      \
552 {                                                         \
553     vext_ldff(vd, v0, base, env, desc, LOAD_FN,           \
554               ctzl(sizeof(ETYPE)), GETPC());              \
555 }
556 
557 GEN_VEXT_LDFF(vle8ff_v,  int8_t,  lde_b)
558 GEN_VEXT_LDFF(vle16ff_v, int16_t, lde_h)
559 GEN_VEXT_LDFF(vle32ff_v, int32_t, lde_w)
560 GEN_VEXT_LDFF(vle64ff_v, int64_t, lde_d)
561 
562 #define DO_SWAP(N, M) (M)
563 #define DO_AND(N, M)  (N & M)
564 #define DO_XOR(N, M)  (N ^ M)
565 #define DO_OR(N, M)   (N | M)
566 #define DO_ADD(N, M)  (N + M)
567 
568 /* Signed min/max */
569 #define DO_MAX(N, M)  ((N) >= (M) ? (N) : (M))
570 #define DO_MIN(N, M)  ((N) >= (M) ? (M) : (N))
571 
572 /*
573  * load and store whole register instructions
574  */
575 static void
576 vext_ldst_whole(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc,
577                 vext_ldst_elem_fn *ldst_elem, uint32_t log2_esz, uintptr_t ra)
578 {
579     uint32_t i, k, off, pos;
580     uint32_t nf = vext_nf(desc);
581     uint32_t vlenb = riscv_cpu_cfg(env)->vlenb;
582     uint32_t max_elems = vlenb >> log2_esz;
583 
584     if (env->vstart >= ((vlenb * nf) >> log2_esz)) {
585         env->vstart = 0;
586         return;
587     }
588 
589     k = env->vstart / max_elems;
590     off = env->vstart % max_elems;
591 
592     if (off) {
593         /* load/store rest of elements of current segment pointed by vstart */
594         for (pos = off; pos < max_elems; pos++, env->vstart++) {
595             target_ulong addr = base + ((pos + k * max_elems) << log2_esz);
596             ldst_elem(env, adjust_addr(env, addr), pos + k * max_elems, vd,
597                       ra);
598         }
599         k++;
600     }
601 
602     /* load/store elements for rest of segments */
603     for (; k < nf; k++) {
604         for (i = 0; i < max_elems; i++, env->vstart++) {
605             target_ulong addr = base + ((i + k * max_elems) << log2_esz);
606             ldst_elem(env, adjust_addr(env, addr), i + k * max_elems, vd, ra);
607         }
608     }
609 
610     env->vstart = 0;
611 }
612 
613 #define GEN_VEXT_LD_WHOLE(NAME, ETYPE, LOAD_FN)      \
614 void HELPER(NAME)(void *vd, target_ulong base,       \
615                   CPURISCVState *env, uint32_t desc) \
616 {                                                    \
617     vext_ldst_whole(vd, base, env, desc, LOAD_FN,    \
618                     ctzl(sizeof(ETYPE)), GETPC());   \
619 }
620 
621 GEN_VEXT_LD_WHOLE(vl1re8_v,  int8_t,  lde_b)
622 GEN_VEXT_LD_WHOLE(vl1re16_v, int16_t, lde_h)
623 GEN_VEXT_LD_WHOLE(vl1re32_v, int32_t, lde_w)
624 GEN_VEXT_LD_WHOLE(vl1re64_v, int64_t, lde_d)
625 GEN_VEXT_LD_WHOLE(vl2re8_v,  int8_t,  lde_b)
626 GEN_VEXT_LD_WHOLE(vl2re16_v, int16_t, lde_h)
627 GEN_VEXT_LD_WHOLE(vl2re32_v, int32_t, lde_w)
628 GEN_VEXT_LD_WHOLE(vl2re64_v, int64_t, lde_d)
629 GEN_VEXT_LD_WHOLE(vl4re8_v,  int8_t,  lde_b)
630 GEN_VEXT_LD_WHOLE(vl4re16_v, int16_t, lde_h)
631 GEN_VEXT_LD_WHOLE(vl4re32_v, int32_t, lde_w)
632 GEN_VEXT_LD_WHOLE(vl4re64_v, int64_t, lde_d)
633 GEN_VEXT_LD_WHOLE(vl8re8_v,  int8_t,  lde_b)
634 GEN_VEXT_LD_WHOLE(vl8re16_v, int16_t, lde_h)
635 GEN_VEXT_LD_WHOLE(vl8re32_v, int32_t, lde_w)
636 GEN_VEXT_LD_WHOLE(vl8re64_v, int64_t, lde_d)
637 
638 #define GEN_VEXT_ST_WHOLE(NAME, ETYPE, STORE_FN)     \
639 void HELPER(NAME)(void *vd, target_ulong base,       \
640                   CPURISCVState *env, uint32_t desc) \
641 {                                                    \
642     vext_ldst_whole(vd, base, env, desc, STORE_FN,   \
643                     ctzl(sizeof(ETYPE)), GETPC());   \
644 }
645 
646 GEN_VEXT_ST_WHOLE(vs1r_v, int8_t, ste_b)
647 GEN_VEXT_ST_WHOLE(vs2r_v, int8_t, ste_b)
648 GEN_VEXT_ST_WHOLE(vs4r_v, int8_t, ste_b)
649 GEN_VEXT_ST_WHOLE(vs8r_v, int8_t, ste_b)
650 
651 /*
652  * Vector Integer Arithmetic Instructions
653  */
654 
655 /* (TD, T1, T2, TX1, TX2) */
656 #define OP_SSS_B int8_t, int8_t, int8_t, int8_t, int8_t
657 #define OP_SSS_H int16_t, int16_t, int16_t, int16_t, int16_t
658 #define OP_SSS_W int32_t, int32_t, int32_t, int32_t, int32_t
659 #define OP_SSS_D int64_t, int64_t, int64_t, int64_t, int64_t
660 #define OP_SUS_B int8_t, uint8_t, int8_t, uint8_t, int8_t
661 #define OP_SUS_H int16_t, uint16_t, int16_t, uint16_t, int16_t
662 #define OP_SUS_W int32_t, uint32_t, int32_t, uint32_t, int32_t
663 #define OP_SUS_D int64_t, uint64_t, int64_t, uint64_t, int64_t
664 #define WOP_SSS_B int16_t, int8_t, int8_t, int16_t, int16_t
665 #define WOP_SSS_H int32_t, int16_t, int16_t, int32_t, int32_t
666 #define WOP_SSS_W int64_t, int32_t, int32_t, int64_t, int64_t
667 #define WOP_SUS_B int16_t, uint8_t, int8_t, uint16_t, int16_t
668 #define WOP_SUS_H int32_t, uint16_t, int16_t, uint32_t, int32_t
669 #define WOP_SUS_W int64_t, uint32_t, int32_t, uint64_t, int64_t
670 #define WOP_SSU_B int16_t, int8_t, uint8_t, int16_t, uint16_t
671 #define WOP_SSU_H int32_t, int16_t, uint16_t, int32_t, uint32_t
672 #define WOP_SSU_W int64_t, int32_t, uint32_t, int64_t, uint64_t
673 #define NOP_SSS_B int8_t, int8_t, int16_t, int8_t, int16_t
674 #define NOP_SSS_H int16_t, int16_t, int32_t, int16_t, int32_t
675 #define NOP_SSS_W int32_t, int32_t, int64_t, int32_t, int64_t
676 #define NOP_UUU_B uint8_t, uint8_t, uint16_t, uint8_t, uint16_t
677 #define NOP_UUU_H uint16_t, uint16_t, uint32_t, uint16_t, uint32_t
678 #define NOP_UUU_W uint32_t, uint32_t, uint64_t, uint32_t, uint64_t
679 
680 #define DO_SUB(N, M) (N - M)
681 #define DO_RSUB(N, M) (M - N)
682 
683 RVVCALL(OPIVV2, vadd_vv_b, OP_SSS_B, H1, H1, H1, DO_ADD)
684 RVVCALL(OPIVV2, vadd_vv_h, OP_SSS_H, H2, H2, H2, DO_ADD)
685 RVVCALL(OPIVV2, vadd_vv_w, OP_SSS_W, H4, H4, H4, DO_ADD)
686 RVVCALL(OPIVV2, vadd_vv_d, OP_SSS_D, H8, H8, H8, DO_ADD)
687 RVVCALL(OPIVV2, vsub_vv_b, OP_SSS_B, H1, H1, H1, DO_SUB)
688 RVVCALL(OPIVV2, vsub_vv_h, OP_SSS_H, H2, H2, H2, DO_SUB)
689 RVVCALL(OPIVV2, vsub_vv_w, OP_SSS_W, H4, H4, H4, DO_SUB)
690 RVVCALL(OPIVV2, vsub_vv_d, OP_SSS_D, H8, H8, H8, DO_SUB)
691 
692 GEN_VEXT_VV(vadd_vv_b, 1)
693 GEN_VEXT_VV(vadd_vv_h, 2)
694 GEN_VEXT_VV(vadd_vv_w, 4)
695 GEN_VEXT_VV(vadd_vv_d, 8)
696 GEN_VEXT_VV(vsub_vv_b, 1)
697 GEN_VEXT_VV(vsub_vv_h, 2)
698 GEN_VEXT_VV(vsub_vv_w, 4)
699 GEN_VEXT_VV(vsub_vv_d, 8)
700 
701 
702 RVVCALL(OPIVX2, vadd_vx_b, OP_SSS_B, H1, H1, DO_ADD)
703 RVVCALL(OPIVX2, vadd_vx_h, OP_SSS_H, H2, H2, DO_ADD)
704 RVVCALL(OPIVX2, vadd_vx_w, OP_SSS_W, H4, H4, DO_ADD)
705 RVVCALL(OPIVX2, vadd_vx_d, OP_SSS_D, H8, H8, DO_ADD)
706 RVVCALL(OPIVX2, vsub_vx_b, OP_SSS_B, H1, H1, DO_SUB)
707 RVVCALL(OPIVX2, vsub_vx_h, OP_SSS_H, H2, H2, DO_SUB)
708 RVVCALL(OPIVX2, vsub_vx_w, OP_SSS_W, H4, H4, DO_SUB)
709 RVVCALL(OPIVX2, vsub_vx_d, OP_SSS_D, H8, H8, DO_SUB)
710 RVVCALL(OPIVX2, vrsub_vx_b, OP_SSS_B, H1, H1, DO_RSUB)
711 RVVCALL(OPIVX2, vrsub_vx_h, OP_SSS_H, H2, H2, DO_RSUB)
712 RVVCALL(OPIVX2, vrsub_vx_w, OP_SSS_W, H4, H4, DO_RSUB)
713 RVVCALL(OPIVX2, vrsub_vx_d, OP_SSS_D, H8, H8, DO_RSUB)
714 
715 GEN_VEXT_VX(vadd_vx_b, 1)
716 GEN_VEXT_VX(vadd_vx_h, 2)
717 GEN_VEXT_VX(vadd_vx_w, 4)
718 GEN_VEXT_VX(vadd_vx_d, 8)
719 GEN_VEXT_VX(vsub_vx_b, 1)
720 GEN_VEXT_VX(vsub_vx_h, 2)
721 GEN_VEXT_VX(vsub_vx_w, 4)
722 GEN_VEXT_VX(vsub_vx_d, 8)
723 GEN_VEXT_VX(vrsub_vx_b, 1)
724 GEN_VEXT_VX(vrsub_vx_h, 2)
725 GEN_VEXT_VX(vrsub_vx_w, 4)
726 GEN_VEXT_VX(vrsub_vx_d, 8)
727 
728 void HELPER(vec_rsubs8)(void *d, void *a, uint64_t b, uint32_t desc)
729 {
730     intptr_t oprsz = simd_oprsz(desc);
731     intptr_t i;
732 
733     for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
734         *(uint8_t *)(d + i) = (uint8_t)b - *(uint8_t *)(a + i);
735     }
736 }
737 
738 void HELPER(vec_rsubs16)(void *d, void *a, uint64_t b, uint32_t desc)
739 {
740     intptr_t oprsz = simd_oprsz(desc);
741     intptr_t i;
742 
743     for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
744         *(uint16_t *)(d + i) = (uint16_t)b - *(uint16_t *)(a + i);
745     }
746 }
747 
748 void HELPER(vec_rsubs32)(void *d, void *a, uint64_t b, uint32_t desc)
749 {
750     intptr_t oprsz = simd_oprsz(desc);
751     intptr_t i;
752 
753     for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
754         *(uint32_t *)(d + i) = (uint32_t)b - *(uint32_t *)(a + i);
755     }
756 }
757 
758 void HELPER(vec_rsubs64)(void *d, void *a, uint64_t b, uint32_t desc)
759 {
760     intptr_t oprsz = simd_oprsz(desc);
761     intptr_t i;
762 
763     for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
764         *(uint64_t *)(d + i) = b - *(uint64_t *)(a + i);
765     }
766 }
767 
768 /* Vector Widening Integer Add/Subtract */
769 #define WOP_UUU_B uint16_t, uint8_t, uint8_t, uint16_t, uint16_t
770 #define WOP_UUU_H uint32_t, uint16_t, uint16_t, uint32_t, uint32_t
771 #define WOP_UUU_W uint64_t, uint32_t, uint32_t, uint64_t, uint64_t
772 #define WOP_SSS_B int16_t, int8_t, int8_t, int16_t, int16_t
773 #define WOP_SSS_H int32_t, int16_t, int16_t, int32_t, int32_t
774 #define WOP_SSS_W int64_t, int32_t, int32_t, int64_t, int64_t
775 #define WOP_WUUU_B  uint16_t, uint8_t, uint16_t, uint16_t, uint16_t
776 #define WOP_WUUU_H  uint32_t, uint16_t, uint32_t, uint32_t, uint32_t
777 #define WOP_WUUU_W  uint64_t, uint32_t, uint64_t, uint64_t, uint64_t
778 #define WOP_WSSS_B  int16_t, int8_t, int16_t, int16_t, int16_t
779 #define WOP_WSSS_H  int32_t, int16_t, int32_t, int32_t, int32_t
780 #define WOP_WSSS_W  int64_t, int32_t, int64_t, int64_t, int64_t
781 RVVCALL(OPIVV2, vwaddu_vv_b, WOP_UUU_B, H2, H1, H1, DO_ADD)
782 RVVCALL(OPIVV2, vwaddu_vv_h, WOP_UUU_H, H4, H2, H2, DO_ADD)
783 RVVCALL(OPIVV2, vwaddu_vv_w, WOP_UUU_W, H8, H4, H4, DO_ADD)
784 RVVCALL(OPIVV2, vwsubu_vv_b, WOP_UUU_B, H2, H1, H1, DO_SUB)
785 RVVCALL(OPIVV2, vwsubu_vv_h, WOP_UUU_H, H4, H2, H2, DO_SUB)
786 RVVCALL(OPIVV2, vwsubu_vv_w, WOP_UUU_W, H8, H4, H4, DO_SUB)
787 RVVCALL(OPIVV2, vwadd_vv_b, WOP_SSS_B, H2, H1, H1, DO_ADD)
788 RVVCALL(OPIVV2, vwadd_vv_h, WOP_SSS_H, H4, H2, H2, DO_ADD)
789 RVVCALL(OPIVV2, vwadd_vv_w, WOP_SSS_W, H8, H4, H4, DO_ADD)
790 RVVCALL(OPIVV2, vwsub_vv_b, WOP_SSS_B, H2, H1, H1, DO_SUB)
791 RVVCALL(OPIVV2, vwsub_vv_h, WOP_SSS_H, H4, H2, H2, DO_SUB)
792 RVVCALL(OPIVV2, vwsub_vv_w, WOP_SSS_W, H8, H4, H4, DO_SUB)
793 RVVCALL(OPIVV2, vwaddu_wv_b, WOP_WUUU_B, H2, H1, H1, DO_ADD)
794 RVVCALL(OPIVV2, vwaddu_wv_h, WOP_WUUU_H, H4, H2, H2, DO_ADD)
795 RVVCALL(OPIVV2, vwaddu_wv_w, WOP_WUUU_W, H8, H4, H4, DO_ADD)
796 RVVCALL(OPIVV2, vwsubu_wv_b, WOP_WUUU_B, H2, H1, H1, DO_SUB)
797 RVVCALL(OPIVV2, vwsubu_wv_h, WOP_WUUU_H, H4, H2, H2, DO_SUB)
798 RVVCALL(OPIVV2, vwsubu_wv_w, WOP_WUUU_W, H8, H4, H4, DO_SUB)
799 RVVCALL(OPIVV2, vwadd_wv_b, WOP_WSSS_B, H2, H1, H1, DO_ADD)
800 RVVCALL(OPIVV2, vwadd_wv_h, WOP_WSSS_H, H4, H2, H2, DO_ADD)
801 RVVCALL(OPIVV2, vwadd_wv_w, WOP_WSSS_W, H8, H4, H4, DO_ADD)
802 RVVCALL(OPIVV2, vwsub_wv_b, WOP_WSSS_B, H2, H1, H1, DO_SUB)
803 RVVCALL(OPIVV2, vwsub_wv_h, WOP_WSSS_H, H4, H2, H2, DO_SUB)
804 RVVCALL(OPIVV2, vwsub_wv_w, WOP_WSSS_W, H8, H4, H4, DO_SUB)
805 GEN_VEXT_VV(vwaddu_vv_b, 2)
806 GEN_VEXT_VV(vwaddu_vv_h, 4)
807 GEN_VEXT_VV(vwaddu_vv_w, 8)
808 GEN_VEXT_VV(vwsubu_vv_b, 2)
809 GEN_VEXT_VV(vwsubu_vv_h, 4)
810 GEN_VEXT_VV(vwsubu_vv_w, 8)
811 GEN_VEXT_VV(vwadd_vv_b, 2)
812 GEN_VEXT_VV(vwadd_vv_h, 4)
813 GEN_VEXT_VV(vwadd_vv_w, 8)
814 GEN_VEXT_VV(vwsub_vv_b, 2)
815 GEN_VEXT_VV(vwsub_vv_h, 4)
816 GEN_VEXT_VV(vwsub_vv_w, 8)
817 GEN_VEXT_VV(vwaddu_wv_b, 2)
818 GEN_VEXT_VV(vwaddu_wv_h, 4)
819 GEN_VEXT_VV(vwaddu_wv_w, 8)
820 GEN_VEXT_VV(vwsubu_wv_b, 2)
821 GEN_VEXT_VV(vwsubu_wv_h, 4)
822 GEN_VEXT_VV(vwsubu_wv_w, 8)
823 GEN_VEXT_VV(vwadd_wv_b, 2)
824 GEN_VEXT_VV(vwadd_wv_h, 4)
825 GEN_VEXT_VV(vwadd_wv_w, 8)
826 GEN_VEXT_VV(vwsub_wv_b, 2)
827 GEN_VEXT_VV(vwsub_wv_h, 4)
828 GEN_VEXT_VV(vwsub_wv_w, 8)
829 
830 RVVCALL(OPIVX2, vwaddu_vx_b, WOP_UUU_B, H2, H1, DO_ADD)
831 RVVCALL(OPIVX2, vwaddu_vx_h, WOP_UUU_H, H4, H2, DO_ADD)
832 RVVCALL(OPIVX2, vwaddu_vx_w, WOP_UUU_W, H8, H4, DO_ADD)
833 RVVCALL(OPIVX2, vwsubu_vx_b, WOP_UUU_B, H2, H1, DO_SUB)
834 RVVCALL(OPIVX2, vwsubu_vx_h, WOP_UUU_H, H4, H2, DO_SUB)
835 RVVCALL(OPIVX2, vwsubu_vx_w, WOP_UUU_W, H8, H4, DO_SUB)
836 RVVCALL(OPIVX2, vwadd_vx_b, WOP_SSS_B, H2, H1, DO_ADD)
837 RVVCALL(OPIVX2, vwadd_vx_h, WOP_SSS_H, H4, H2, DO_ADD)
838 RVVCALL(OPIVX2, vwadd_vx_w, WOP_SSS_W, H8, H4, DO_ADD)
839 RVVCALL(OPIVX2, vwsub_vx_b, WOP_SSS_B, H2, H1, DO_SUB)
840 RVVCALL(OPIVX2, vwsub_vx_h, WOP_SSS_H, H4, H2, DO_SUB)
841 RVVCALL(OPIVX2, vwsub_vx_w, WOP_SSS_W, H8, H4, DO_SUB)
842 RVVCALL(OPIVX2, vwaddu_wx_b, WOP_WUUU_B, H2, H1, DO_ADD)
843 RVVCALL(OPIVX2, vwaddu_wx_h, WOP_WUUU_H, H4, H2, DO_ADD)
844 RVVCALL(OPIVX2, vwaddu_wx_w, WOP_WUUU_W, H8, H4, DO_ADD)
845 RVVCALL(OPIVX2, vwsubu_wx_b, WOP_WUUU_B, H2, H1, DO_SUB)
846 RVVCALL(OPIVX2, vwsubu_wx_h, WOP_WUUU_H, H4, H2, DO_SUB)
847 RVVCALL(OPIVX2, vwsubu_wx_w, WOP_WUUU_W, H8, H4, DO_SUB)
848 RVVCALL(OPIVX2, vwadd_wx_b, WOP_WSSS_B, H2, H1, DO_ADD)
849 RVVCALL(OPIVX2, vwadd_wx_h, WOP_WSSS_H, H4, H2, DO_ADD)
850 RVVCALL(OPIVX2, vwadd_wx_w, WOP_WSSS_W, H8, H4, DO_ADD)
851 RVVCALL(OPIVX2, vwsub_wx_b, WOP_WSSS_B, H2, H1, DO_SUB)
852 RVVCALL(OPIVX2, vwsub_wx_h, WOP_WSSS_H, H4, H2, DO_SUB)
853 RVVCALL(OPIVX2, vwsub_wx_w, WOP_WSSS_W, H8, H4, DO_SUB)
854 GEN_VEXT_VX(vwaddu_vx_b, 2)
855 GEN_VEXT_VX(vwaddu_vx_h, 4)
856 GEN_VEXT_VX(vwaddu_vx_w, 8)
857 GEN_VEXT_VX(vwsubu_vx_b, 2)
858 GEN_VEXT_VX(vwsubu_vx_h, 4)
859 GEN_VEXT_VX(vwsubu_vx_w, 8)
860 GEN_VEXT_VX(vwadd_vx_b, 2)
861 GEN_VEXT_VX(vwadd_vx_h, 4)
862 GEN_VEXT_VX(vwadd_vx_w, 8)
863 GEN_VEXT_VX(vwsub_vx_b, 2)
864 GEN_VEXT_VX(vwsub_vx_h, 4)
865 GEN_VEXT_VX(vwsub_vx_w, 8)
866 GEN_VEXT_VX(vwaddu_wx_b, 2)
867 GEN_VEXT_VX(vwaddu_wx_h, 4)
868 GEN_VEXT_VX(vwaddu_wx_w, 8)
869 GEN_VEXT_VX(vwsubu_wx_b, 2)
870 GEN_VEXT_VX(vwsubu_wx_h, 4)
871 GEN_VEXT_VX(vwsubu_wx_w, 8)
872 GEN_VEXT_VX(vwadd_wx_b, 2)
873 GEN_VEXT_VX(vwadd_wx_h, 4)
874 GEN_VEXT_VX(vwadd_wx_w, 8)
875 GEN_VEXT_VX(vwsub_wx_b, 2)
876 GEN_VEXT_VX(vwsub_wx_h, 4)
877 GEN_VEXT_VX(vwsub_wx_w, 8)
878 
879 /* Vector Integer Add-with-Carry / Subtract-with-Borrow Instructions */
880 #define DO_VADC(N, M, C) (N + M + C)
881 #define DO_VSBC(N, M, C) (N - M - C)
882 
883 #define GEN_VEXT_VADC_VVM(NAME, ETYPE, H, DO_OP)              \
884 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
885                   CPURISCVState *env, uint32_t desc)          \
886 {                                                             \
887     uint32_t vl = env->vl;                                    \
888     uint32_t esz = sizeof(ETYPE);                             \
889     uint32_t total_elems =                                    \
890         vext_get_total_elems(env, desc, esz);                 \
891     uint32_t vta = vext_vta(desc);                            \
892     uint32_t i;                                               \
893                                                               \
894     VSTART_CHECK_EARLY_EXIT(env);                             \
895                                                               \
896     for (i = env->vstart; i < vl; i++) {                      \
897         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
898         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
899         ETYPE carry = vext_elem_mask(v0, i);                  \
900                                                               \
901         *((ETYPE *)vd + H(i)) = DO_OP(s2, s1, carry);         \
902     }                                                         \
903     env->vstart = 0;                                          \
904     /* set tail elements to 1s */                             \
905     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);  \
906 }
907 
908 GEN_VEXT_VADC_VVM(vadc_vvm_b, uint8_t,  H1, DO_VADC)
909 GEN_VEXT_VADC_VVM(vadc_vvm_h, uint16_t, H2, DO_VADC)
910 GEN_VEXT_VADC_VVM(vadc_vvm_w, uint32_t, H4, DO_VADC)
911 GEN_VEXT_VADC_VVM(vadc_vvm_d, uint64_t, H8, DO_VADC)
912 
913 GEN_VEXT_VADC_VVM(vsbc_vvm_b, uint8_t,  H1, DO_VSBC)
914 GEN_VEXT_VADC_VVM(vsbc_vvm_h, uint16_t, H2, DO_VSBC)
915 GEN_VEXT_VADC_VVM(vsbc_vvm_w, uint32_t, H4, DO_VSBC)
916 GEN_VEXT_VADC_VVM(vsbc_vvm_d, uint64_t, H8, DO_VSBC)
917 
918 #define GEN_VEXT_VADC_VXM(NAME, ETYPE, H, DO_OP)                         \
919 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,        \
920                   CPURISCVState *env, uint32_t desc)                     \
921 {                                                                        \
922     uint32_t vl = env->vl;                                               \
923     uint32_t esz = sizeof(ETYPE);                                        \
924     uint32_t total_elems = vext_get_total_elems(env, desc, esz);         \
925     uint32_t vta = vext_vta(desc);                                       \
926     uint32_t i;                                                          \
927                                                                          \
928     VSTART_CHECK_EARLY_EXIT(env);                                        \
929                                                                          \
930     for (i = env->vstart; i < vl; i++) {                                 \
931         ETYPE s2 = *((ETYPE *)vs2 + H(i));                               \
932         ETYPE carry = vext_elem_mask(v0, i);                             \
933                                                                          \
934         *((ETYPE *)vd + H(i)) = DO_OP(s2, (ETYPE)(target_long)s1, carry);\
935     }                                                                    \
936     env->vstart = 0;                                                     \
937     /* set tail elements to 1s */                                        \
938     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);             \
939 }
940 
941 GEN_VEXT_VADC_VXM(vadc_vxm_b, uint8_t,  H1, DO_VADC)
942 GEN_VEXT_VADC_VXM(vadc_vxm_h, uint16_t, H2, DO_VADC)
943 GEN_VEXT_VADC_VXM(vadc_vxm_w, uint32_t, H4, DO_VADC)
944 GEN_VEXT_VADC_VXM(vadc_vxm_d, uint64_t, H8, DO_VADC)
945 
946 GEN_VEXT_VADC_VXM(vsbc_vxm_b, uint8_t,  H1, DO_VSBC)
947 GEN_VEXT_VADC_VXM(vsbc_vxm_h, uint16_t, H2, DO_VSBC)
948 GEN_VEXT_VADC_VXM(vsbc_vxm_w, uint32_t, H4, DO_VSBC)
949 GEN_VEXT_VADC_VXM(vsbc_vxm_d, uint64_t, H8, DO_VSBC)
950 
951 #define DO_MADC(N, M, C) (C ? (__typeof(N))(N + M + 1) <= N :           \
952                           (__typeof(N))(N + M) < N)
953 #define DO_MSBC(N, M, C) (C ? N <= M : N < M)
954 
955 #define GEN_VEXT_VMADC_VVM(NAME, ETYPE, H, DO_OP)             \
956 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
957                   CPURISCVState *env, uint32_t desc)          \
958 {                                                             \
959     uint32_t vl = env->vl;                                    \
960     uint32_t vm = vext_vm(desc);                              \
961     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;    \
962     uint32_t vta_all_1s = vext_vta_all_1s(desc);              \
963     uint32_t i;                                               \
964                                                               \
965     VSTART_CHECK_EARLY_EXIT(env);                             \
966                                                               \
967     for (i = env->vstart; i < vl; i++) {                      \
968         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
969         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
970         ETYPE carry = !vm && vext_elem_mask(v0, i);           \
971         vext_set_elem_mask(vd, i, DO_OP(s2, s1, carry));      \
972     }                                                         \
973     env->vstart = 0;                                          \
974     /*
975      * mask destination register are always tail-agnostic
976      * set tail elements to 1s
977      */                                                       \
978     if (vta_all_1s) {                                         \
979         for (; i < total_elems; i++) {                        \
980             vext_set_elem_mask(vd, i, 1);                     \
981         }                                                     \
982     }                                                         \
983 }
984 
985 GEN_VEXT_VMADC_VVM(vmadc_vvm_b, uint8_t,  H1, DO_MADC)
986 GEN_VEXT_VMADC_VVM(vmadc_vvm_h, uint16_t, H2, DO_MADC)
987 GEN_VEXT_VMADC_VVM(vmadc_vvm_w, uint32_t, H4, DO_MADC)
988 GEN_VEXT_VMADC_VVM(vmadc_vvm_d, uint64_t, H8, DO_MADC)
989 
990 GEN_VEXT_VMADC_VVM(vmsbc_vvm_b, uint8_t,  H1, DO_MSBC)
991 GEN_VEXT_VMADC_VVM(vmsbc_vvm_h, uint16_t, H2, DO_MSBC)
992 GEN_VEXT_VMADC_VVM(vmsbc_vvm_w, uint32_t, H4, DO_MSBC)
993 GEN_VEXT_VMADC_VVM(vmsbc_vvm_d, uint64_t, H8, DO_MSBC)
994 
995 #define GEN_VEXT_VMADC_VXM(NAME, ETYPE, H, DO_OP)               \
996 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,          \
997                   void *vs2, CPURISCVState *env, uint32_t desc) \
998 {                                                               \
999     uint32_t vl = env->vl;                                      \
1000     uint32_t vm = vext_vm(desc);                                \
1001     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;      \
1002     uint32_t vta_all_1s = vext_vta_all_1s(desc);                \
1003     uint32_t i;                                                 \
1004                                                                 \
1005     VSTART_CHECK_EARLY_EXIT(env);                               \
1006                                                                 \
1007     for (i = env->vstart; i < vl; i++) {                        \
1008         ETYPE s2 = *((ETYPE *)vs2 + H(i));                      \
1009         ETYPE carry = !vm && vext_elem_mask(v0, i);             \
1010         vext_set_elem_mask(vd, i,                               \
1011                 DO_OP(s2, (ETYPE)(target_long)s1, carry));      \
1012     }                                                           \
1013     env->vstart = 0;                                            \
1014     /*
1015      * mask destination register are always tail-agnostic
1016      * set tail elements to 1s
1017      */                                                         \
1018     if (vta_all_1s) {                                           \
1019         for (; i < total_elems; i++) {                          \
1020             vext_set_elem_mask(vd, i, 1);                       \
1021         }                                                       \
1022     }                                                           \
1023 }
1024 
1025 GEN_VEXT_VMADC_VXM(vmadc_vxm_b, uint8_t,  H1, DO_MADC)
1026 GEN_VEXT_VMADC_VXM(vmadc_vxm_h, uint16_t, H2, DO_MADC)
1027 GEN_VEXT_VMADC_VXM(vmadc_vxm_w, uint32_t, H4, DO_MADC)
1028 GEN_VEXT_VMADC_VXM(vmadc_vxm_d, uint64_t, H8, DO_MADC)
1029 
1030 GEN_VEXT_VMADC_VXM(vmsbc_vxm_b, uint8_t,  H1, DO_MSBC)
1031 GEN_VEXT_VMADC_VXM(vmsbc_vxm_h, uint16_t, H2, DO_MSBC)
1032 GEN_VEXT_VMADC_VXM(vmsbc_vxm_w, uint32_t, H4, DO_MSBC)
1033 GEN_VEXT_VMADC_VXM(vmsbc_vxm_d, uint64_t, H8, DO_MSBC)
1034 
1035 /* Vector Bitwise Logical Instructions */
1036 RVVCALL(OPIVV2, vand_vv_b, OP_SSS_B, H1, H1, H1, DO_AND)
1037 RVVCALL(OPIVV2, vand_vv_h, OP_SSS_H, H2, H2, H2, DO_AND)
1038 RVVCALL(OPIVV2, vand_vv_w, OP_SSS_W, H4, H4, H4, DO_AND)
1039 RVVCALL(OPIVV2, vand_vv_d, OP_SSS_D, H8, H8, H8, DO_AND)
1040 RVVCALL(OPIVV2, vor_vv_b, OP_SSS_B, H1, H1, H1, DO_OR)
1041 RVVCALL(OPIVV2, vor_vv_h, OP_SSS_H, H2, H2, H2, DO_OR)
1042 RVVCALL(OPIVV2, vor_vv_w, OP_SSS_W, H4, H4, H4, DO_OR)
1043 RVVCALL(OPIVV2, vor_vv_d, OP_SSS_D, H8, H8, H8, DO_OR)
1044 RVVCALL(OPIVV2, vxor_vv_b, OP_SSS_B, H1, H1, H1, DO_XOR)
1045 RVVCALL(OPIVV2, vxor_vv_h, OP_SSS_H, H2, H2, H2, DO_XOR)
1046 RVVCALL(OPIVV2, vxor_vv_w, OP_SSS_W, H4, H4, H4, DO_XOR)
1047 RVVCALL(OPIVV2, vxor_vv_d, OP_SSS_D, H8, H8, H8, DO_XOR)
1048 GEN_VEXT_VV(vand_vv_b, 1)
1049 GEN_VEXT_VV(vand_vv_h, 2)
1050 GEN_VEXT_VV(vand_vv_w, 4)
1051 GEN_VEXT_VV(vand_vv_d, 8)
1052 GEN_VEXT_VV(vor_vv_b, 1)
1053 GEN_VEXT_VV(vor_vv_h, 2)
1054 GEN_VEXT_VV(vor_vv_w, 4)
1055 GEN_VEXT_VV(vor_vv_d, 8)
1056 GEN_VEXT_VV(vxor_vv_b, 1)
1057 GEN_VEXT_VV(vxor_vv_h, 2)
1058 GEN_VEXT_VV(vxor_vv_w, 4)
1059 GEN_VEXT_VV(vxor_vv_d, 8)
1060 
1061 RVVCALL(OPIVX2, vand_vx_b, OP_SSS_B, H1, H1, DO_AND)
1062 RVVCALL(OPIVX2, vand_vx_h, OP_SSS_H, H2, H2, DO_AND)
1063 RVVCALL(OPIVX2, vand_vx_w, OP_SSS_W, H4, H4, DO_AND)
1064 RVVCALL(OPIVX2, vand_vx_d, OP_SSS_D, H8, H8, DO_AND)
1065 RVVCALL(OPIVX2, vor_vx_b, OP_SSS_B, H1, H1, DO_OR)
1066 RVVCALL(OPIVX2, vor_vx_h, OP_SSS_H, H2, H2, DO_OR)
1067 RVVCALL(OPIVX2, vor_vx_w, OP_SSS_W, H4, H4, DO_OR)
1068 RVVCALL(OPIVX2, vor_vx_d, OP_SSS_D, H8, H8, DO_OR)
1069 RVVCALL(OPIVX2, vxor_vx_b, OP_SSS_B, H1, H1, DO_XOR)
1070 RVVCALL(OPIVX2, vxor_vx_h, OP_SSS_H, H2, H2, DO_XOR)
1071 RVVCALL(OPIVX2, vxor_vx_w, OP_SSS_W, H4, H4, DO_XOR)
1072 RVVCALL(OPIVX2, vxor_vx_d, OP_SSS_D, H8, H8, DO_XOR)
1073 GEN_VEXT_VX(vand_vx_b, 1)
1074 GEN_VEXT_VX(vand_vx_h, 2)
1075 GEN_VEXT_VX(vand_vx_w, 4)
1076 GEN_VEXT_VX(vand_vx_d, 8)
1077 GEN_VEXT_VX(vor_vx_b, 1)
1078 GEN_VEXT_VX(vor_vx_h, 2)
1079 GEN_VEXT_VX(vor_vx_w, 4)
1080 GEN_VEXT_VX(vor_vx_d, 8)
1081 GEN_VEXT_VX(vxor_vx_b, 1)
1082 GEN_VEXT_VX(vxor_vx_h, 2)
1083 GEN_VEXT_VX(vxor_vx_w, 4)
1084 GEN_VEXT_VX(vxor_vx_d, 8)
1085 
1086 /* Vector Single-Width Bit Shift Instructions */
1087 #define DO_SLL(N, M)  (N << (M))
1088 #define DO_SRL(N, M)  (N >> (M))
1089 
1090 /* generate the helpers for shift instructions with two vector operators */
1091 #define GEN_VEXT_SHIFT_VV(NAME, TS1, TS2, HS1, HS2, OP, MASK)             \
1092 void HELPER(NAME)(void *vd, void *v0, void *vs1,                          \
1093                   void *vs2, CPURISCVState *env, uint32_t desc)           \
1094 {                                                                         \
1095     uint32_t vm = vext_vm(desc);                                          \
1096     uint32_t vl = env->vl;                                                \
1097     uint32_t esz = sizeof(TS1);                                           \
1098     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
1099     uint32_t vta = vext_vta(desc);                                        \
1100     uint32_t vma = vext_vma(desc);                                        \
1101     uint32_t i;                                                           \
1102                                                                           \
1103     VSTART_CHECK_EARLY_EXIT(env);                                         \
1104                                                                           \
1105     for (i = env->vstart; i < vl; i++) {                                  \
1106         if (!vm && !vext_elem_mask(v0, i)) {                              \
1107             /* set masked-off elements to 1s */                           \
1108             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
1109             continue;                                                     \
1110         }                                                                 \
1111         TS1 s1 = *((TS1 *)vs1 + HS1(i));                                  \
1112         TS2 s2 = *((TS2 *)vs2 + HS2(i));                                  \
1113         *((TS1 *)vd + HS1(i)) = OP(s2, s1 & MASK);                        \
1114     }                                                                     \
1115     env->vstart = 0;                                                      \
1116     /* set tail elements to 1s */                                         \
1117     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
1118 }
1119 
1120 GEN_VEXT_SHIFT_VV(vsll_vv_b, uint8_t,  uint8_t, H1, H1, DO_SLL, 0x7)
1121 GEN_VEXT_SHIFT_VV(vsll_vv_h, uint16_t, uint16_t, H2, H2, DO_SLL, 0xf)
1122 GEN_VEXT_SHIFT_VV(vsll_vv_w, uint32_t, uint32_t, H4, H4, DO_SLL, 0x1f)
1123 GEN_VEXT_SHIFT_VV(vsll_vv_d, uint64_t, uint64_t, H8, H8, DO_SLL, 0x3f)
1124 
1125 GEN_VEXT_SHIFT_VV(vsrl_vv_b, uint8_t, uint8_t, H1, H1, DO_SRL, 0x7)
1126 GEN_VEXT_SHIFT_VV(vsrl_vv_h, uint16_t, uint16_t, H2, H2, DO_SRL, 0xf)
1127 GEN_VEXT_SHIFT_VV(vsrl_vv_w, uint32_t, uint32_t, H4, H4, DO_SRL, 0x1f)
1128 GEN_VEXT_SHIFT_VV(vsrl_vv_d, uint64_t, uint64_t, H8, H8, DO_SRL, 0x3f)
1129 
1130 GEN_VEXT_SHIFT_VV(vsra_vv_b, uint8_t,  int8_t, H1, H1, DO_SRL, 0x7)
1131 GEN_VEXT_SHIFT_VV(vsra_vv_h, uint16_t, int16_t, H2, H2, DO_SRL, 0xf)
1132 GEN_VEXT_SHIFT_VV(vsra_vv_w, uint32_t, int32_t, H4, H4, DO_SRL, 0x1f)
1133 GEN_VEXT_SHIFT_VV(vsra_vv_d, uint64_t, int64_t, H8, H8, DO_SRL, 0x3f)
1134 
1135 /*
1136  * generate the helpers for shift instructions with one vector and one scalar
1137  */
1138 #define GEN_VEXT_SHIFT_VX(NAME, TD, TS2, HD, HS2, OP, MASK) \
1139 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,      \
1140                   void *vs2, CPURISCVState *env,            \
1141                   uint32_t desc)                            \
1142 {                                                           \
1143     uint32_t vm = vext_vm(desc);                            \
1144     uint32_t vl = env->vl;                                  \
1145     uint32_t esz = sizeof(TD);                              \
1146     uint32_t total_elems =                                  \
1147         vext_get_total_elems(env, desc, esz);               \
1148     uint32_t vta = vext_vta(desc);                          \
1149     uint32_t vma = vext_vma(desc);                          \
1150     uint32_t i;                                             \
1151                                                             \
1152     VSTART_CHECK_EARLY_EXIT(env);                           \
1153                                                             \
1154     for (i = env->vstart; i < vl; i++) {                    \
1155         if (!vm && !vext_elem_mask(v0, i)) {                \
1156             /* set masked-off elements to 1s */             \
1157             vext_set_elems_1s(vd, vma, i * esz,             \
1158                               (i + 1) * esz);               \
1159             continue;                                       \
1160         }                                                   \
1161         TS2 s2 = *((TS2 *)vs2 + HS2(i));                    \
1162         *((TD *)vd + HD(i)) = OP(s2, s1 & MASK);            \
1163     }                                                       \
1164     env->vstart = 0;                                        \
1165     /* set tail elements to 1s */                           \
1166     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);\
1167 }
1168 
1169 GEN_VEXT_SHIFT_VX(vsll_vx_b, uint8_t, int8_t, H1, H1, DO_SLL, 0x7)
1170 GEN_VEXT_SHIFT_VX(vsll_vx_h, uint16_t, int16_t, H2, H2, DO_SLL, 0xf)
1171 GEN_VEXT_SHIFT_VX(vsll_vx_w, uint32_t, int32_t, H4, H4, DO_SLL, 0x1f)
1172 GEN_VEXT_SHIFT_VX(vsll_vx_d, uint64_t, int64_t, H8, H8, DO_SLL, 0x3f)
1173 
1174 GEN_VEXT_SHIFT_VX(vsrl_vx_b, uint8_t, uint8_t, H1, H1, DO_SRL, 0x7)
1175 GEN_VEXT_SHIFT_VX(vsrl_vx_h, uint16_t, uint16_t, H2, H2, DO_SRL, 0xf)
1176 GEN_VEXT_SHIFT_VX(vsrl_vx_w, uint32_t, uint32_t, H4, H4, DO_SRL, 0x1f)
1177 GEN_VEXT_SHIFT_VX(vsrl_vx_d, uint64_t, uint64_t, H8, H8, DO_SRL, 0x3f)
1178 
1179 GEN_VEXT_SHIFT_VX(vsra_vx_b, int8_t, int8_t, H1, H1, DO_SRL, 0x7)
1180 GEN_VEXT_SHIFT_VX(vsra_vx_h, int16_t, int16_t, H2, H2, DO_SRL, 0xf)
1181 GEN_VEXT_SHIFT_VX(vsra_vx_w, int32_t, int32_t, H4, H4, DO_SRL, 0x1f)
1182 GEN_VEXT_SHIFT_VX(vsra_vx_d, int64_t, int64_t, H8, H8, DO_SRL, 0x3f)
1183 
1184 /* Vector Narrowing Integer Right Shift Instructions */
1185 GEN_VEXT_SHIFT_VV(vnsrl_wv_b, uint8_t,  uint16_t, H1, H2, DO_SRL, 0xf)
1186 GEN_VEXT_SHIFT_VV(vnsrl_wv_h, uint16_t, uint32_t, H2, H4, DO_SRL, 0x1f)
1187 GEN_VEXT_SHIFT_VV(vnsrl_wv_w, uint32_t, uint64_t, H4, H8, DO_SRL, 0x3f)
1188 GEN_VEXT_SHIFT_VV(vnsra_wv_b, uint8_t,  int16_t, H1, H2, DO_SRL, 0xf)
1189 GEN_VEXT_SHIFT_VV(vnsra_wv_h, uint16_t, int32_t, H2, H4, DO_SRL, 0x1f)
1190 GEN_VEXT_SHIFT_VV(vnsra_wv_w, uint32_t, int64_t, H4, H8, DO_SRL, 0x3f)
1191 GEN_VEXT_SHIFT_VX(vnsrl_wx_b, uint8_t, uint16_t, H1, H2, DO_SRL, 0xf)
1192 GEN_VEXT_SHIFT_VX(vnsrl_wx_h, uint16_t, uint32_t, H2, H4, DO_SRL, 0x1f)
1193 GEN_VEXT_SHIFT_VX(vnsrl_wx_w, uint32_t, uint64_t, H4, H8, DO_SRL, 0x3f)
1194 GEN_VEXT_SHIFT_VX(vnsra_wx_b, int8_t, int16_t, H1, H2, DO_SRL, 0xf)
1195 GEN_VEXT_SHIFT_VX(vnsra_wx_h, int16_t, int32_t, H2, H4, DO_SRL, 0x1f)
1196 GEN_VEXT_SHIFT_VX(vnsra_wx_w, int32_t, int64_t, H4, H8, DO_SRL, 0x3f)
1197 
1198 /* Vector Integer Comparison Instructions */
1199 #define DO_MSEQ(N, M) (N == M)
1200 #define DO_MSNE(N, M) (N != M)
1201 #define DO_MSLT(N, M) (N < M)
1202 #define DO_MSLE(N, M) (N <= M)
1203 #define DO_MSGT(N, M) (N > M)
1204 
1205 #define GEN_VEXT_CMP_VV(NAME, ETYPE, H, DO_OP)                \
1206 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
1207                   CPURISCVState *env, uint32_t desc)          \
1208 {                                                             \
1209     uint32_t vm = vext_vm(desc);                              \
1210     uint32_t vl = env->vl;                                    \
1211     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;    \
1212     uint32_t vta_all_1s = vext_vta_all_1s(desc);              \
1213     uint32_t vma = vext_vma(desc);                            \
1214     uint32_t i;                                               \
1215                                                               \
1216     VSTART_CHECK_EARLY_EXIT(env);                             \
1217                                                               \
1218     for (i = env->vstart; i < vl; i++) {                      \
1219         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
1220         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
1221         if (!vm && !vext_elem_mask(v0, i)) {                  \
1222             /* set masked-off elements to 1s */               \
1223             if (vma) {                                        \
1224                 vext_set_elem_mask(vd, i, 1);                 \
1225             }                                                 \
1226             continue;                                         \
1227         }                                                     \
1228         vext_set_elem_mask(vd, i, DO_OP(s2, s1));             \
1229     }                                                         \
1230     env->vstart = 0;                                          \
1231     /*
1232      * mask destination register are always tail-agnostic
1233      * set tail elements to 1s
1234      */                                                       \
1235     if (vta_all_1s) {                                         \
1236         for (; i < total_elems; i++) {                        \
1237             vext_set_elem_mask(vd, i, 1);                     \
1238         }                                                     \
1239     }                                                         \
1240 }
1241 
1242 GEN_VEXT_CMP_VV(vmseq_vv_b, uint8_t,  H1, DO_MSEQ)
1243 GEN_VEXT_CMP_VV(vmseq_vv_h, uint16_t, H2, DO_MSEQ)
1244 GEN_VEXT_CMP_VV(vmseq_vv_w, uint32_t, H4, DO_MSEQ)
1245 GEN_VEXT_CMP_VV(vmseq_vv_d, uint64_t, H8, DO_MSEQ)
1246 
1247 GEN_VEXT_CMP_VV(vmsne_vv_b, uint8_t,  H1, DO_MSNE)
1248 GEN_VEXT_CMP_VV(vmsne_vv_h, uint16_t, H2, DO_MSNE)
1249 GEN_VEXT_CMP_VV(vmsne_vv_w, uint32_t, H4, DO_MSNE)
1250 GEN_VEXT_CMP_VV(vmsne_vv_d, uint64_t, H8, DO_MSNE)
1251 
1252 GEN_VEXT_CMP_VV(vmsltu_vv_b, uint8_t,  H1, DO_MSLT)
1253 GEN_VEXT_CMP_VV(vmsltu_vv_h, uint16_t, H2, DO_MSLT)
1254 GEN_VEXT_CMP_VV(vmsltu_vv_w, uint32_t, H4, DO_MSLT)
1255 GEN_VEXT_CMP_VV(vmsltu_vv_d, uint64_t, H8, DO_MSLT)
1256 
1257 GEN_VEXT_CMP_VV(vmslt_vv_b, int8_t,  H1, DO_MSLT)
1258 GEN_VEXT_CMP_VV(vmslt_vv_h, int16_t, H2, DO_MSLT)
1259 GEN_VEXT_CMP_VV(vmslt_vv_w, int32_t, H4, DO_MSLT)
1260 GEN_VEXT_CMP_VV(vmslt_vv_d, int64_t, H8, DO_MSLT)
1261 
1262 GEN_VEXT_CMP_VV(vmsleu_vv_b, uint8_t,  H1, DO_MSLE)
1263 GEN_VEXT_CMP_VV(vmsleu_vv_h, uint16_t, H2, DO_MSLE)
1264 GEN_VEXT_CMP_VV(vmsleu_vv_w, uint32_t, H4, DO_MSLE)
1265 GEN_VEXT_CMP_VV(vmsleu_vv_d, uint64_t, H8, DO_MSLE)
1266 
1267 GEN_VEXT_CMP_VV(vmsle_vv_b, int8_t,  H1, DO_MSLE)
1268 GEN_VEXT_CMP_VV(vmsle_vv_h, int16_t, H2, DO_MSLE)
1269 GEN_VEXT_CMP_VV(vmsle_vv_w, int32_t, H4, DO_MSLE)
1270 GEN_VEXT_CMP_VV(vmsle_vv_d, int64_t, H8, DO_MSLE)
1271 
1272 #define GEN_VEXT_CMP_VX(NAME, ETYPE, H, DO_OP)                      \
1273 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,   \
1274                   CPURISCVState *env, uint32_t desc)                \
1275 {                                                                   \
1276     uint32_t vm = vext_vm(desc);                                    \
1277     uint32_t vl = env->vl;                                          \
1278     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;          \
1279     uint32_t vta_all_1s = vext_vta_all_1s(desc);                    \
1280     uint32_t vma = vext_vma(desc);                                  \
1281     uint32_t i;                                                     \
1282                                                                     \
1283     VSTART_CHECK_EARLY_EXIT(env);                                   \
1284                                                                     \
1285     for (i = env->vstart; i < vl; i++) {                            \
1286         ETYPE s2 = *((ETYPE *)vs2 + H(i));                          \
1287         if (!vm && !vext_elem_mask(v0, i)) {                        \
1288             /* set masked-off elements to 1s */                     \
1289             if (vma) {                                              \
1290                 vext_set_elem_mask(vd, i, 1);                       \
1291             }                                                       \
1292             continue;                                               \
1293         }                                                           \
1294         vext_set_elem_mask(vd, i,                                   \
1295                 DO_OP(s2, (ETYPE)(target_long)s1));                 \
1296     }                                                               \
1297     env->vstart = 0;                                                \
1298     /*
1299      * mask destination register are always tail-agnostic
1300      * set tail elements to 1s
1301      */                                                             \
1302     if (vta_all_1s) {                                               \
1303         for (; i < total_elems; i++) {                              \
1304             vext_set_elem_mask(vd, i, 1);                           \
1305         }                                                           \
1306     }                                                               \
1307 }
1308 
1309 GEN_VEXT_CMP_VX(vmseq_vx_b, uint8_t,  H1, DO_MSEQ)
1310 GEN_VEXT_CMP_VX(vmseq_vx_h, uint16_t, H2, DO_MSEQ)
1311 GEN_VEXT_CMP_VX(vmseq_vx_w, uint32_t, H4, DO_MSEQ)
1312 GEN_VEXT_CMP_VX(vmseq_vx_d, uint64_t, H8, DO_MSEQ)
1313 
1314 GEN_VEXT_CMP_VX(vmsne_vx_b, uint8_t,  H1, DO_MSNE)
1315 GEN_VEXT_CMP_VX(vmsne_vx_h, uint16_t, H2, DO_MSNE)
1316 GEN_VEXT_CMP_VX(vmsne_vx_w, uint32_t, H4, DO_MSNE)
1317 GEN_VEXT_CMP_VX(vmsne_vx_d, uint64_t, H8, DO_MSNE)
1318 
1319 GEN_VEXT_CMP_VX(vmsltu_vx_b, uint8_t,  H1, DO_MSLT)
1320 GEN_VEXT_CMP_VX(vmsltu_vx_h, uint16_t, H2, DO_MSLT)
1321 GEN_VEXT_CMP_VX(vmsltu_vx_w, uint32_t, H4, DO_MSLT)
1322 GEN_VEXT_CMP_VX(vmsltu_vx_d, uint64_t, H8, DO_MSLT)
1323 
1324 GEN_VEXT_CMP_VX(vmslt_vx_b, int8_t,  H1, DO_MSLT)
1325 GEN_VEXT_CMP_VX(vmslt_vx_h, int16_t, H2, DO_MSLT)
1326 GEN_VEXT_CMP_VX(vmslt_vx_w, int32_t, H4, DO_MSLT)
1327 GEN_VEXT_CMP_VX(vmslt_vx_d, int64_t, H8, DO_MSLT)
1328 
1329 GEN_VEXT_CMP_VX(vmsleu_vx_b, uint8_t,  H1, DO_MSLE)
1330 GEN_VEXT_CMP_VX(vmsleu_vx_h, uint16_t, H2, DO_MSLE)
1331 GEN_VEXT_CMP_VX(vmsleu_vx_w, uint32_t, H4, DO_MSLE)
1332 GEN_VEXT_CMP_VX(vmsleu_vx_d, uint64_t, H8, DO_MSLE)
1333 
1334 GEN_VEXT_CMP_VX(vmsle_vx_b, int8_t,  H1, DO_MSLE)
1335 GEN_VEXT_CMP_VX(vmsle_vx_h, int16_t, H2, DO_MSLE)
1336 GEN_VEXT_CMP_VX(vmsle_vx_w, int32_t, H4, DO_MSLE)
1337 GEN_VEXT_CMP_VX(vmsle_vx_d, int64_t, H8, DO_MSLE)
1338 
1339 GEN_VEXT_CMP_VX(vmsgtu_vx_b, uint8_t,  H1, DO_MSGT)
1340 GEN_VEXT_CMP_VX(vmsgtu_vx_h, uint16_t, H2, DO_MSGT)
1341 GEN_VEXT_CMP_VX(vmsgtu_vx_w, uint32_t, H4, DO_MSGT)
1342 GEN_VEXT_CMP_VX(vmsgtu_vx_d, uint64_t, H8, DO_MSGT)
1343 
1344 GEN_VEXT_CMP_VX(vmsgt_vx_b, int8_t,  H1, DO_MSGT)
1345 GEN_VEXT_CMP_VX(vmsgt_vx_h, int16_t, H2, DO_MSGT)
1346 GEN_VEXT_CMP_VX(vmsgt_vx_w, int32_t, H4, DO_MSGT)
1347 GEN_VEXT_CMP_VX(vmsgt_vx_d, int64_t, H8, DO_MSGT)
1348 
1349 /* Vector Integer Min/Max Instructions */
1350 RVVCALL(OPIVV2, vminu_vv_b, OP_UUU_B, H1, H1, H1, DO_MIN)
1351 RVVCALL(OPIVV2, vminu_vv_h, OP_UUU_H, H2, H2, H2, DO_MIN)
1352 RVVCALL(OPIVV2, vminu_vv_w, OP_UUU_W, H4, H4, H4, DO_MIN)
1353 RVVCALL(OPIVV2, vminu_vv_d, OP_UUU_D, H8, H8, H8, DO_MIN)
1354 RVVCALL(OPIVV2, vmin_vv_b, OP_SSS_B, H1, H1, H1, DO_MIN)
1355 RVVCALL(OPIVV2, vmin_vv_h, OP_SSS_H, H2, H2, H2, DO_MIN)
1356 RVVCALL(OPIVV2, vmin_vv_w, OP_SSS_W, H4, H4, H4, DO_MIN)
1357 RVVCALL(OPIVV2, vmin_vv_d, OP_SSS_D, H8, H8, H8, DO_MIN)
1358 RVVCALL(OPIVV2, vmaxu_vv_b, OP_UUU_B, H1, H1, H1, DO_MAX)
1359 RVVCALL(OPIVV2, vmaxu_vv_h, OP_UUU_H, H2, H2, H2, DO_MAX)
1360 RVVCALL(OPIVV2, vmaxu_vv_w, OP_UUU_W, H4, H4, H4, DO_MAX)
1361 RVVCALL(OPIVV2, vmaxu_vv_d, OP_UUU_D, H8, H8, H8, DO_MAX)
1362 RVVCALL(OPIVV2, vmax_vv_b, OP_SSS_B, H1, H1, H1, DO_MAX)
1363 RVVCALL(OPIVV2, vmax_vv_h, OP_SSS_H, H2, H2, H2, DO_MAX)
1364 RVVCALL(OPIVV2, vmax_vv_w, OP_SSS_W, H4, H4, H4, DO_MAX)
1365 RVVCALL(OPIVV2, vmax_vv_d, OP_SSS_D, H8, H8, H8, DO_MAX)
1366 GEN_VEXT_VV(vminu_vv_b, 1)
1367 GEN_VEXT_VV(vminu_vv_h, 2)
1368 GEN_VEXT_VV(vminu_vv_w, 4)
1369 GEN_VEXT_VV(vminu_vv_d, 8)
1370 GEN_VEXT_VV(vmin_vv_b, 1)
1371 GEN_VEXT_VV(vmin_vv_h, 2)
1372 GEN_VEXT_VV(vmin_vv_w, 4)
1373 GEN_VEXT_VV(vmin_vv_d, 8)
1374 GEN_VEXT_VV(vmaxu_vv_b, 1)
1375 GEN_VEXT_VV(vmaxu_vv_h, 2)
1376 GEN_VEXT_VV(vmaxu_vv_w, 4)
1377 GEN_VEXT_VV(vmaxu_vv_d, 8)
1378 GEN_VEXT_VV(vmax_vv_b, 1)
1379 GEN_VEXT_VV(vmax_vv_h, 2)
1380 GEN_VEXT_VV(vmax_vv_w, 4)
1381 GEN_VEXT_VV(vmax_vv_d, 8)
1382 
1383 RVVCALL(OPIVX2, vminu_vx_b, OP_UUU_B, H1, H1, DO_MIN)
1384 RVVCALL(OPIVX2, vminu_vx_h, OP_UUU_H, H2, H2, DO_MIN)
1385 RVVCALL(OPIVX2, vminu_vx_w, OP_UUU_W, H4, H4, DO_MIN)
1386 RVVCALL(OPIVX2, vminu_vx_d, OP_UUU_D, H8, H8, DO_MIN)
1387 RVVCALL(OPIVX2, vmin_vx_b, OP_SSS_B, H1, H1, DO_MIN)
1388 RVVCALL(OPIVX2, vmin_vx_h, OP_SSS_H, H2, H2, DO_MIN)
1389 RVVCALL(OPIVX2, vmin_vx_w, OP_SSS_W, H4, H4, DO_MIN)
1390 RVVCALL(OPIVX2, vmin_vx_d, OP_SSS_D, H8, H8, DO_MIN)
1391 RVVCALL(OPIVX2, vmaxu_vx_b, OP_UUU_B, H1, H1, DO_MAX)
1392 RVVCALL(OPIVX2, vmaxu_vx_h, OP_UUU_H, H2, H2, DO_MAX)
1393 RVVCALL(OPIVX2, vmaxu_vx_w, OP_UUU_W, H4, H4, DO_MAX)
1394 RVVCALL(OPIVX2, vmaxu_vx_d, OP_UUU_D, H8, H8, DO_MAX)
1395 RVVCALL(OPIVX2, vmax_vx_b, OP_SSS_B, H1, H1, DO_MAX)
1396 RVVCALL(OPIVX2, vmax_vx_h, OP_SSS_H, H2, H2, DO_MAX)
1397 RVVCALL(OPIVX2, vmax_vx_w, OP_SSS_W, H4, H4, DO_MAX)
1398 RVVCALL(OPIVX2, vmax_vx_d, OP_SSS_D, H8, H8, DO_MAX)
1399 GEN_VEXT_VX(vminu_vx_b, 1)
1400 GEN_VEXT_VX(vminu_vx_h, 2)
1401 GEN_VEXT_VX(vminu_vx_w, 4)
1402 GEN_VEXT_VX(vminu_vx_d, 8)
1403 GEN_VEXT_VX(vmin_vx_b, 1)
1404 GEN_VEXT_VX(vmin_vx_h, 2)
1405 GEN_VEXT_VX(vmin_vx_w, 4)
1406 GEN_VEXT_VX(vmin_vx_d, 8)
1407 GEN_VEXT_VX(vmaxu_vx_b, 1)
1408 GEN_VEXT_VX(vmaxu_vx_h, 2)
1409 GEN_VEXT_VX(vmaxu_vx_w, 4)
1410 GEN_VEXT_VX(vmaxu_vx_d, 8)
1411 GEN_VEXT_VX(vmax_vx_b, 1)
1412 GEN_VEXT_VX(vmax_vx_h, 2)
1413 GEN_VEXT_VX(vmax_vx_w, 4)
1414 GEN_VEXT_VX(vmax_vx_d, 8)
1415 
1416 /* Vector Single-Width Integer Multiply Instructions */
1417 #define DO_MUL(N, M) (N * M)
1418 RVVCALL(OPIVV2, vmul_vv_b, OP_SSS_B, H1, H1, H1, DO_MUL)
1419 RVVCALL(OPIVV2, vmul_vv_h, OP_SSS_H, H2, H2, H2, DO_MUL)
1420 RVVCALL(OPIVV2, vmul_vv_w, OP_SSS_W, H4, H4, H4, DO_MUL)
1421 RVVCALL(OPIVV2, vmul_vv_d, OP_SSS_D, H8, H8, H8, DO_MUL)
1422 GEN_VEXT_VV(vmul_vv_b, 1)
1423 GEN_VEXT_VV(vmul_vv_h, 2)
1424 GEN_VEXT_VV(vmul_vv_w, 4)
1425 GEN_VEXT_VV(vmul_vv_d, 8)
1426 
1427 static int8_t do_mulh_b(int8_t s2, int8_t s1)
1428 {
1429     return (int16_t)s2 * (int16_t)s1 >> 8;
1430 }
1431 
1432 static int16_t do_mulh_h(int16_t s2, int16_t s1)
1433 {
1434     return (int32_t)s2 * (int32_t)s1 >> 16;
1435 }
1436 
1437 static int32_t do_mulh_w(int32_t s2, int32_t s1)
1438 {
1439     return (int64_t)s2 * (int64_t)s1 >> 32;
1440 }
1441 
1442 static int64_t do_mulh_d(int64_t s2, int64_t s1)
1443 {
1444     uint64_t hi_64, lo_64;
1445 
1446     muls64(&lo_64, &hi_64, s1, s2);
1447     return hi_64;
1448 }
1449 
1450 static uint8_t do_mulhu_b(uint8_t s2, uint8_t s1)
1451 {
1452     return (uint16_t)s2 * (uint16_t)s1 >> 8;
1453 }
1454 
1455 static uint16_t do_mulhu_h(uint16_t s2, uint16_t s1)
1456 {
1457     return (uint32_t)s2 * (uint32_t)s1 >> 16;
1458 }
1459 
1460 static uint32_t do_mulhu_w(uint32_t s2, uint32_t s1)
1461 {
1462     return (uint64_t)s2 * (uint64_t)s1 >> 32;
1463 }
1464 
1465 static uint64_t do_mulhu_d(uint64_t s2, uint64_t s1)
1466 {
1467     uint64_t hi_64, lo_64;
1468 
1469     mulu64(&lo_64, &hi_64, s2, s1);
1470     return hi_64;
1471 }
1472 
1473 static int8_t do_mulhsu_b(int8_t s2, uint8_t s1)
1474 {
1475     return (int16_t)s2 * (uint16_t)s1 >> 8;
1476 }
1477 
1478 static int16_t do_mulhsu_h(int16_t s2, uint16_t s1)
1479 {
1480     return (int32_t)s2 * (uint32_t)s1 >> 16;
1481 }
1482 
1483 static int32_t do_mulhsu_w(int32_t s2, uint32_t s1)
1484 {
1485     return (int64_t)s2 * (uint64_t)s1 >> 32;
1486 }
1487 
1488 /*
1489  * Let  A = signed operand,
1490  *      B = unsigned operand
1491  *      P = mulu64(A, B), unsigned product
1492  *
1493  * LET  X = 2 ** 64  - A, 2's complement of A
1494  *      SP = signed product
1495  * THEN
1496  *      IF A < 0
1497  *          SP = -X * B
1498  *             = -(2 ** 64 - A) * B
1499  *             = A * B - 2 ** 64 * B
1500  *             = P - 2 ** 64 * B
1501  *      ELSE
1502  *          SP = P
1503  * THEN
1504  *      HI_P -= (A < 0 ? B : 0)
1505  */
1506 
1507 static int64_t do_mulhsu_d(int64_t s2, uint64_t s1)
1508 {
1509     uint64_t hi_64, lo_64;
1510 
1511     mulu64(&lo_64, &hi_64, s2, s1);
1512 
1513     hi_64 -= s2 < 0 ? s1 : 0;
1514     return hi_64;
1515 }
1516 
1517 RVVCALL(OPIVV2, vmulh_vv_b, OP_SSS_B, H1, H1, H1, do_mulh_b)
1518 RVVCALL(OPIVV2, vmulh_vv_h, OP_SSS_H, H2, H2, H2, do_mulh_h)
1519 RVVCALL(OPIVV2, vmulh_vv_w, OP_SSS_W, H4, H4, H4, do_mulh_w)
1520 RVVCALL(OPIVV2, vmulh_vv_d, OP_SSS_D, H8, H8, H8, do_mulh_d)
1521 RVVCALL(OPIVV2, vmulhu_vv_b, OP_UUU_B, H1, H1, H1, do_mulhu_b)
1522 RVVCALL(OPIVV2, vmulhu_vv_h, OP_UUU_H, H2, H2, H2, do_mulhu_h)
1523 RVVCALL(OPIVV2, vmulhu_vv_w, OP_UUU_W, H4, H4, H4, do_mulhu_w)
1524 RVVCALL(OPIVV2, vmulhu_vv_d, OP_UUU_D, H8, H8, H8, do_mulhu_d)
1525 RVVCALL(OPIVV2, vmulhsu_vv_b, OP_SUS_B, H1, H1, H1, do_mulhsu_b)
1526 RVVCALL(OPIVV2, vmulhsu_vv_h, OP_SUS_H, H2, H2, H2, do_mulhsu_h)
1527 RVVCALL(OPIVV2, vmulhsu_vv_w, OP_SUS_W, H4, H4, H4, do_mulhsu_w)
1528 RVVCALL(OPIVV2, vmulhsu_vv_d, OP_SUS_D, H8, H8, H8, do_mulhsu_d)
1529 GEN_VEXT_VV(vmulh_vv_b, 1)
1530 GEN_VEXT_VV(vmulh_vv_h, 2)
1531 GEN_VEXT_VV(vmulh_vv_w, 4)
1532 GEN_VEXT_VV(vmulh_vv_d, 8)
1533 GEN_VEXT_VV(vmulhu_vv_b, 1)
1534 GEN_VEXT_VV(vmulhu_vv_h, 2)
1535 GEN_VEXT_VV(vmulhu_vv_w, 4)
1536 GEN_VEXT_VV(vmulhu_vv_d, 8)
1537 GEN_VEXT_VV(vmulhsu_vv_b, 1)
1538 GEN_VEXT_VV(vmulhsu_vv_h, 2)
1539 GEN_VEXT_VV(vmulhsu_vv_w, 4)
1540 GEN_VEXT_VV(vmulhsu_vv_d, 8)
1541 
1542 RVVCALL(OPIVX2, vmul_vx_b, OP_SSS_B, H1, H1, DO_MUL)
1543 RVVCALL(OPIVX2, vmul_vx_h, OP_SSS_H, H2, H2, DO_MUL)
1544 RVVCALL(OPIVX2, vmul_vx_w, OP_SSS_W, H4, H4, DO_MUL)
1545 RVVCALL(OPIVX2, vmul_vx_d, OP_SSS_D, H8, H8, DO_MUL)
1546 RVVCALL(OPIVX2, vmulh_vx_b, OP_SSS_B, H1, H1, do_mulh_b)
1547 RVVCALL(OPIVX2, vmulh_vx_h, OP_SSS_H, H2, H2, do_mulh_h)
1548 RVVCALL(OPIVX2, vmulh_vx_w, OP_SSS_W, H4, H4, do_mulh_w)
1549 RVVCALL(OPIVX2, vmulh_vx_d, OP_SSS_D, H8, H8, do_mulh_d)
1550 RVVCALL(OPIVX2, vmulhu_vx_b, OP_UUU_B, H1, H1, do_mulhu_b)
1551 RVVCALL(OPIVX2, vmulhu_vx_h, OP_UUU_H, H2, H2, do_mulhu_h)
1552 RVVCALL(OPIVX2, vmulhu_vx_w, OP_UUU_W, H4, H4, do_mulhu_w)
1553 RVVCALL(OPIVX2, vmulhu_vx_d, OP_UUU_D, H8, H8, do_mulhu_d)
1554 RVVCALL(OPIVX2, vmulhsu_vx_b, OP_SUS_B, H1, H1, do_mulhsu_b)
1555 RVVCALL(OPIVX2, vmulhsu_vx_h, OP_SUS_H, H2, H2, do_mulhsu_h)
1556 RVVCALL(OPIVX2, vmulhsu_vx_w, OP_SUS_W, H4, H4, do_mulhsu_w)
1557 RVVCALL(OPIVX2, vmulhsu_vx_d, OP_SUS_D, H8, H8, do_mulhsu_d)
1558 GEN_VEXT_VX(vmul_vx_b, 1)
1559 GEN_VEXT_VX(vmul_vx_h, 2)
1560 GEN_VEXT_VX(vmul_vx_w, 4)
1561 GEN_VEXT_VX(vmul_vx_d, 8)
1562 GEN_VEXT_VX(vmulh_vx_b, 1)
1563 GEN_VEXT_VX(vmulh_vx_h, 2)
1564 GEN_VEXT_VX(vmulh_vx_w, 4)
1565 GEN_VEXT_VX(vmulh_vx_d, 8)
1566 GEN_VEXT_VX(vmulhu_vx_b, 1)
1567 GEN_VEXT_VX(vmulhu_vx_h, 2)
1568 GEN_VEXT_VX(vmulhu_vx_w, 4)
1569 GEN_VEXT_VX(vmulhu_vx_d, 8)
1570 GEN_VEXT_VX(vmulhsu_vx_b, 1)
1571 GEN_VEXT_VX(vmulhsu_vx_h, 2)
1572 GEN_VEXT_VX(vmulhsu_vx_w, 4)
1573 GEN_VEXT_VX(vmulhsu_vx_d, 8)
1574 
1575 /* Vector Integer Divide Instructions */
1576 #define DO_DIVU(N, M) (unlikely(M == 0) ? (__typeof(N))(-1) : N / M)
1577 #define DO_REMU(N, M) (unlikely(M == 0) ? N : N % M)
1578 #define DO_DIV(N, M)  (unlikely(M == 0) ? (__typeof(N))(-1) : \
1579         unlikely((N == -N) && (M == (__typeof(N))(-1))) ? N : N / M)
1580 #define DO_REM(N, M)  (unlikely(M == 0) ? N : \
1581         unlikely((N == -N) && (M == (__typeof(N))(-1))) ? 0 : N % M)
1582 
1583 RVVCALL(OPIVV2, vdivu_vv_b, OP_UUU_B, H1, H1, H1, DO_DIVU)
1584 RVVCALL(OPIVV2, vdivu_vv_h, OP_UUU_H, H2, H2, H2, DO_DIVU)
1585 RVVCALL(OPIVV2, vdivu_vv_w, OP_UUU_W, H4, H4, H4, DO_DIVU)
1586 RVVCALL(OPIVV2, vdivu_vv_d, OP_UUU_D, H8, H8, H8, DO_DIVU)
1587 RVVCALL(OPIVV2, vdiv_vv_b, OP_SSS_B, H1, H1, H1, DO_DIV)
1588 RVVCALL(OPIVV2, vdiv_vv_h, OP_SSS_H, H2, H2, H2, DO_DIV)
1589 RVVCALL(OPIVV2, vdiv_vv_w, OP_SSS_W, H4, H4, H4, DO_DIV)
1590 RVVCALL(OPIVV2, vdiv_vv_d, OP_SSS_D, H8, H8, H8, DO_DIV)
1591 RVVCALL(OPIVV2, vremu_vv_b, OP_UUU_B, H1, H1, H1, DO_REMU)
1592 RVVCALL(OPIVV2, vremu_vv_h, OP_UUU_H, H2, H2, H2, DO_REMU)
1593 RVVCALL(OPIVV2, vremu_vv_w, OP_UUU_W, H4, H4, H4, DO_REMU)
1594 RVVCALL(OPIVV2, vremu_vv_d, OP_UUU_D, H8, H8, H8, DO_REMU)
1595 RVVCALL(OPIVV2, vrem_vv_b, OP_SSS_B, H1, H1, H1, DO_REM)
1596 RVVCALL(OPIVV2, vrem_vv_h, OP_SSS_H, H2, H2, H2, DO_REM)
1597 RVVCALL(OPIVV2, vrem_vv_w, OP_SSS_W, H4, H4, H4, DO_REM)
1598 RVVCALL(OPIVV2, vrem_vv_d, OP_SSS_D, H8, H8, H8, DO_REM)
1599 GEN_VEXT_VV(vdivu_vv_b, 1)
1600 GEN_VEXT_VV(vdivu_vv_h, 2)
1601 GEN_VEXT_VV(vdivu_vv_w, 4)
1602 GEN_VEXT_VV(vdivu_vv_d, 8)
1603 GEN_VEXT_VV(vdiv_vv_b, 1)
1604 GEN_VEXT_VV(vdiv_vv_h, 2)
1605 GEN_VEXT_VV(vdiv_vv_w, 4)
1606 GEN_VEXT_VV(vdiv_vv_d, 8)
1607 GEN_VEXT_VV(vremu_vv_b, 1)
1608 GEN_VEXT_VV(vremu_vv_h, 2)
1609 GEN_VEXT_VV(vremu_vv_w, 4)
1610 GEN_VEXT_VV(vremu_vv_d, 8)
1611 GEN_VEXT_VV(vrem_vv_b, 1)
1612 GEN_VEXT_VV(vrem_vv_h, 2)
1613 GEN_VEXT_VV(vrem_vv_w, 4)
1614 GEN_VEXT_VV(vrem_vv_d, 8)
1615 
1616 RVVCALL(OPIVX2, vdivu_vx_b, OP_UUU_B, H1, H1, DO_DIVU)
1617 RVVCALL(OPIVX2, vdivu_vx_h, OP_UUU_H, H2, H2, DO_DIVU)
1618 RVVCALL(OPIVX2, vdivu_vx_w, OP_UUU_W, H4, H4, DO_DIVU)
1619 RVVCALL(OPIVX2, vdivu_vx_d, OP_UUU_D, H8, H8, DO_DIVU)
1620 RVVCALL(OPIVX2, vdiv_vx_b, OP_SSS_B, H1, H1, DO_DIV)
1621 RVVCALL(OPIVX2, vdiv_vx_h, OP_SSS_H, H2, H2, DO_DIV)
1622 RVVCALL(OPIVX2, vdiv_vx_w, OP_SSS_W, H4, H4, DO_DIV)
1623 RVVCALL(OPIVX2, vdiv_vx_d, OP_SSS_D, H8, H8, DO_DIV)
1624 RVVCALL(OPIVX2, vremu_vx_b, OP_UUU_B, H1, H1, DO_REMU)
1625 RVVCALL(OPIVX2, vremu_vx_h, OP_UUU_H, H2, H2, DO_REMU)
1626 RVVCALL(OPIVX2, vremu_vx_w, OP_UUU_W, H4, H4, DO_REMU)
1627 RVVCALL(OPIVX2, vremu_vx_d, OP_UUU_D, H8, H8, DO_REMU)
1628 RVVCALL(OPIVX2, vrem_vx_b, OP_SSS_B, H1, H1, DO_REM)
1629 RVVCALL(OPIVX2, vrem_vx_h, OP_SSS_H, H2, H2, DO_REM)
1630 RVVCALL(OPIVX2, vrem_vx_w, OP_SSS_W, H4, H4, DO_REM)
1631 RVVCALL(OPIVX2, vrem_vx_d, OP_SSS_D, H8, H8, DO_REM)
1632 GEN_VEXT_VX(vdivu_vx_b, 1)
1633 GEN_VEXT_VX(vdivu_vx_h, 2)
1634 GEN_VEXT_VX(vdivu_vx_w, 4)
1635 GEN_VEXT_VX(vdivu_vx_d, 8)
1636 GEN_VEXT_VX(vdiv_vx_b, 1)
1637 GEN_VEXT_VX(vdiv_vx_h, 2)
1638 GEN_VEXT_VX(vdiv_vx_w, 4)
1639 GEN_VEXT_VX(vdiv_vx_d, 8)
1640 GEN_VEXT_VX(vremu_vx_b, 1)
1641 GEN_VEXT_VX(vremu_vx_h, 2)
1642 GEN_VEXT_VX(vremu_vx_w, 4)
1643 GEN_VEXT_VX(vremu_vx_d, 8)
1644 GEN_VEXT_VX(vrem_vx_b, 1)
1645 GEN_VEXT_VX(vrem_vx_h, 2)
1646 GEN_VEXT_VX(vrem_vx_w, 4)
1647 GEN_VEXT_VX(vrem_vx_d, 8)
1648 
1649 /* Vector Widening Integer Multiply Instructions */
1650 RVVCALL(OPIVV2, vwmul_vv_b, WOP_SSS_B, H2, H1, H1, DO_MUL)
1651 RVVCALL(OPIVV2, vwmul_vv_h, WOP_SSS_H, H4, H2, H2, DO_MUL)
1652 RVVCALL(OPIVV2, vwmul_vv_w, WOP_SSS_W, H8, H4, H4, DO_MUL)
1653 RVVCALL(OPIVV2, vwmulu_vv_b, WOP_UUU_B, H2, H1, H1, DO_MUL)
1654 RVVCALL(OPIVV2, vwmulu_vv_h, WOP_UUU_H, H4, H2, H2, DO_MUL)
1655 RVVCALL(OPIVV2, vwmulu_vv_w, WOP_UUU_W, H8, H4, H4, DO_MUL)
1656 RVVCALL(OPIVV2, vwmulsu_vv_b, WOP_SUS_B, H2, H1, H1, DO_MUL)
1657 RVVCALL(OPIVV2, vwmulsu_vv_h, WOP_SUS_H, H4, H2, H2, DO_MUL)
1658 RVVCALL(OPIVV2, vwmulsu_vv_w, WOP_SUS_W, H8, H4, H4, DO_MUL)
1659 GEN_VEXT_VV(vwmul_vv_b, 2)
1660 GEN_VEXT_VV(vwmul_vv_h, 4)
1661 GEN_VEXT_VV(vwmul_vv_w, 8)
1662 GEN_VEXT_VV(vwmulu_vv_b, 2)
1663 GEN_VEXT_VV(vwmulu_vv_h, 4)
1664 GEN_VEXT_VV(vwmulu_vv_w, 8)
1665 GEN_VEXT_VV(vwmulsu_vv_b, 2)
1666 GEN_VEXT_VV(vwmulsu_vv_h, 4)
1667 GEN_VEXT_VV(vwmulsu_vv_w, 8)
1668 
1669 RVVCALL(OPIVX2, vwmul_vx_b, WOP_SSS_B, H2, H1, DO_MUL)
1670 RVVCALL(OPIVX2, vwmul_vx_h, WOP_SSS_H, H4, H2, DO_MUL)
1671 RVVCALL(OPIVX2, vwmul_vx_w, WOP_SSS_W, H8, H4, DO_MUL)
1672 RVVCALL(OPIVX2, vwmulu_vx_b, WOP_UUU_B, H2, H1, DO_MUL)
1673 RVVCALL(OPIVX2, vwmulu_vx_h, WOP_UUU_H, H4, H2, DO_MUL)
1674 RVVCALL(OPIVX2, vwmulu_vx_w, WOP_UUU_W, H8, H4, DO_MUL)
1675 RVVCALL(OPIVX2, vwmulsu_vx_b, WOP_SUS_B, H2, H1, DO_MUL)
1676 RVVCALL(OPIVX2, vwmulsu_vx_h, WOP_SUS_H, H4, H2, DO_MUL)
1677 RVVCALL(OPIVX2, vwmulsu_vx_w, WOP_SUS_W, H8, H4, DO_MUL)
1678 GEN_VEXT_VX(vwmul_vx_b, 2)
1679 GEN_VEXT_VX(vwmul_vx_h, 4)
1680 GEN_VEXT_VX(vwmul_vx_w, 8)
1681 GEN_VEXT_VX(vwmulu_vx_b, 2)
1682 GEN_VEXT_VX(vwmulu_vx_h, 4)
1683 GEN_VEXT_VX(vwmulu_vx_w, 8)
1684 GEN_VEXT_VX(vwmulsu_vx_b, 2)
1685 GEN_VEXT_VX(vwmulsu_vx_h, 4)
1686 GEN_VEXT_VX(vwmulsu_vx_w, 8)
1687 
1688 /* Vector Single-Width Integer Multiply-Add Instructions */
1689 #define OPIVV3(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)       \
1690 static void do_##NAME(void *vd, void *vs1, void *vs2, int i)       \
1691 {                                                                  \
1692     TX1 s1 = *((T1 *)vs1 + HS1(i));                                \
1693     TX2 s2 = *((T2 *)vs2 + HS2(i));                                \
1694     TD d = *((TD *)vd + HD(i));                                    \
1695     *((TD *)vd + HD(i)) = OP(s2, s1, d);                           \
1696 }
1697 
1698 #define DO_MACC(N, M, D) (M * N + D)
1699 #define DO_NMSAC(N, M, D) (-(M * N) + D)
1700 #define DO_MADD(N, M, D) (M * D + N)
1701 #define DO_NMSUB(N, M, D) (-(M * D) + N)
1702 RVVCALL(OPIVV3, vmacc_vv_b, OP_SSS_B, H1, H1, H1, DO_MACC)
1703 RVVCALL(OPIVV3, vmacc_vv_h, OP_SSS_H, H2, H2, H2, DO_MACC)
1704 RVVCALL(OPIVV3, vmacc_vv_w, OP_SSS_W, H4, H4, H4, DO_MACC)
1705 RVVCALL(OPIVV3, vmacc_vv_d, OP_SSS_D, H8, H8, H8, DO_MACC)
1706 RVVCALL(OPIVV3, vnmsac_vv_b, OP_SSS_B, H1, H1, H1, DO_NMSAC)
1707 RVVCALL(OPIVV3, vnmsac_vv_h, OP_SSS_H, H2, H2, H2, DO_NMSAC)
1708 RVVCALL(OPIVV3, vnmsac_vv_w, OP_SSS_W, H4, H4, H4, DO_NMSAC)
1709 RVVCALL(OPIVV3, vnmsac_vv_d, OP_SSS_D, H8, H8, H8, DO_NMSAC)
1710 RVVCALL(OPIVV3, vmadd_vv_b, OP_SSS_B, H1, H1, H1, DO_MADD)
1711 RVVCALL(OPIVV3, vmadd_vv_h, OP_SSS_H, H2, H2, H2, DO_MADD)
1712 RVVCALL(OPIVV3, vmadd_vv_w, OP_SSS_W, H4, H4, H4, DO_MADD)
1713 RVVCALL(OPIVV3, vmadd_vv_d, OP_SSS_D, H8, H8, H8, DO_MADD)
1714 RVVCALL(OPIVV3, vnmsub_vv_b, OP_SSS_B, H1, H1, H1, DO_NMSUB)
1715 RVVCALL(OPIVV3, vnmsub_vv_h, OP_SSS_H, H2, H2, H2, DO_NMSUB)
1716 RVVCALL(OPIVV3, vnmsub_vv_w, OP_SSS_W, H4, H4, H4, DO_NMSUB)
1717 RVVCALL(OPIVV3, vnmsub_vv_d, OP_SSS_D, H8, H8, H8, DO_NMSUB)
1718 GEN_VEXT_VV(vmacc_vv_b, 1)
1719 GEN_VEXT_VV(vmacc_vv_h, 2)
1720 GEN_VEXT_VV(vmacc_vv_w, 4)
1721 GEN_VEXT_VV(vmacc_vv_d, 8)
1722 GEN_VEXT_VV(vnmsac_vv_b, 1)
1723 GEN_VEXT_VV(vnmsac_vv_h, 2)
1724 GEN_VEXT_VV(vnmsac_vv_w, 4)
1725 GEN_VEXT_VV(vnmsac_vv_d, 8)
1726 GEN_VEXT_VV(vmadd_vv_b, 1)
1727 GEN_VEXT_VV(vmadd_vv_h, 2)
1728 GEN_VEXT_VV(vmadd_vv_w, 4)
1729 GEN_VEXT_VV(vmadd_vv_d, 8)
1730 GEN_VEXT_VV(vnmsub_vv_b, 1)
1731 GEN_VEXT_VV(vnmsub_vv_h, 2)
1732 GEN_VEXT_VV(vnmsub_vv_w, 4)
1733 GEN_VEXT_VV(vnmsub_vv_d, 8)
1734 
1735 #define OPIVX3(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)             \
1736 static void do_##NAME(void *vd, target_long s1, void *vs2, int i)   \
1737 {                                                                   \
1738     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
1739     TD d = *((TD *)vd + HD(i));                                     \
1740     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1, d);                   \
1741 }
1742 
1743 RVVCALL(OPIVX3, vmacc_vx_b, OP_SSS_B, H1, H1, DO_MACC)
1744 RVVCALL(OPIVX3, vmacc_vx_h, OP_SSS_H, H2, H2, DO_MACC)
1745 RVVCALL(OPIVX3, vmacc_vx_w, OP_SSS_W, H4, H4, DO_MACC)
1746 RVVCALL(OPIVX3, vmacc_vx_d, OP_SSS_D, H8, H8, DO_MACC)
1747 RVVCALL(OPIVX3, vnmsac_vx_b, OP_SSS_B, H1, H1, DO_NMSAC)
1748 RVVCALL(OPIVX3, vnmsac_vx_h, OP_SSS_H, H2, H2, DO_NMSAC)
1749 RVVCALL(OPIVX3, vnmsac_vx_w, OP_SSS_W, H4, H4, DO_NMSAC)
1750 RVVCALL(OPIVX3, vnmsac_vx_d, OP_SSS_D, H8, H8, DO_NMSAC)
1751 RVVCALL(OPIVX3, vmadd_vx_b, OP_SSS_B, H1, H1, DO_MADD)
1752 RVVCALL(OPIVX3, vmadd_vx_h, OP_SSS_H, H2, H2, DO_MADD)
1753 RVVCALL(OPIVX3, vmadd_vx_w, OP_SSS_W, H4, H4, DO_MADD)
1754 RVVCALL(OPIVX3, vmadd_vx_d, OP_SSS_D, H8, H8, DO_MADD)
1755 RVVCALL(OPIVX3, vnmsub_vx_b, OP_SSS_B, H1, H1, DO_NMSUB)
1756 RVVCALL(OPIVX3, vnmsub_vx_h, OP_SSS_H, H2, H2, DO_NMSUB)
1757 RVVCALL(OPIVX3, vnmsub_vx_w, OP_SSS_W, H4, H4, DO_NMSUB)
1758 RVVCALL(OPIVX3, vnmsub_vx_d, OP_SSS_D, H8, H8, DO_NMSUB)
1759 GEN_VEXT_VX(vmacc_vx_b, 1)
1760 GEN_VEXT_VX(vmacc_vx_h, 2)
1761 GEN_VEXT_VX(vmacc_vx_w, 4)
1762 GEN_VEXT_VX(vmacc_vx_d, 8)
1763 GEN_VEXT_VX(vnmsac_vx_b, 1)
1764 GEN_VEXT_VX(vnmsac_vx_h, 2)
1765 GEN_VEXT_VX(vnmsac_vx_w, 4)
1766 GEN_VEXT_VX(vnmsac_vx_d, 8)
1767 GEN_VEXT_VX(vmadd_vx_b, 1)
1768 GEN_VEXT_VX(vmadd_vx_h, 2)
1769 GEN_VEXT_VX(vmadd_vx_w, 4)
1770 GEN_VEXT_VX(vmadd_vx_d, 8)
1771 GEN_VEXT_VX(vnmsub_vx_b, 1)
1772 GEN_VEXT_VX(vnmsub_vx_h, 2)
1773 GEN_VEXT_VX(vnmsub_vx_w, 4)
1774 GEN_VEXT_VX(vnmsub_vx_d, 8)
1775 
1776 /* Vector Widening Integer Multiply-Add Instructions */
1777 RVVCALL(OPIVV3, vwmaccu_vv_b, WOP_UUU_B, H2, H1, H1, DO_MACC)
1778 RVVCALL(OPIVV3, vwmaccu_vv_h, WOP_UUU_H, H4, H2, H2, DO_MACC)
1779 RVVCALL(OPIVV3, vwmaccu_vv_w, WOP_UUU_W, H8, H4, H4, DO_MACC)
1780 RVVCALL(OPIVV3, vwmacc_vv_b, WOP_SSS_B, H2, H1, H1, DO_MACC)
1781 RVVCALL(OPIVV3, vwmacc_vv_h, WOP_SSS_H, H4, H2, H2, DO_MACC)
1782 RVVCALL(OPIVV3, vwmacc_vv_w, WOP_SSS_W, H8, H4, H4, DO_MACC)
1783 RVVCALL(OPIVV3, vwmaccsu_vv_b, WOP_SSU_B, H2, H1, H1, DO_MACC)
1784 RVVCALL(OPIVV3, vwmaccsu_vv_h, WOP_SSU_H, H4, H2, H2, DO_MACC)
1785 RVVCALL(OPIVV3, vwmaccsu_vv_w, WOP_SSU_W, H8, H4, H4, DO_MACC)
1786 GEN_VEXT_VV(vwmaccu_vv_b, 2)
1787 GEN_VEXT_VV(vwmaccu_vv_h, 4)
1788 GEN_VEXT_VV(vwmaccu_vv_w, 8)
1789 GEN_VEXT_VV(vwmacc_vv_b, 2)
1790 GEN_VEXT_VV(vwmacc_vv_h, 4)
1791 GEN_VEXT_VV(vwmacc_vv_w, 8)
1792 GEN_VEXT_VV(vwmaccsu_vv_b, 2)
1793 GEN_VEXT_VV(vwmaccsu_vv_h, 4)
1794 GEN_VEXT_VV(vwmaccsu_vv_w, 8)
1795 
1796 RVVCALL(OPIVX3, vwmaccu_vx_b, WOP_UUU_B, H2, H1, DO_MACC)
1797 RVVCALL(OPIVX3, vwmaccu_vx_h, WOP_UUU_H, H4, H2, DO_MACC)
1798 RVVCALL(OPIVX3, vwmaccu_vx_w, WOP_UUU_W, H8, H4, DO_MACC)
1799 RVVCALL(OPIVX3, vwmacc_vx_b, WOP_SSS_B, H2, H1, DO_MACC)
1800 RVVCALL(OPIVX3, vwmacc_vx_h, WOP_SSS_H, H4, H2, DO_MACC)
1801 RVVCALL(OPIVX3, vwmacc_vx_w, WOP_SSS_W, H8, H4, DO_MACC)
1802 RVVCALL(OPIVX3, vwmaccsu_vx_b, WOP_SSU_B, H2, H1, DO_MACC)
1803 RVVCALL(OPIVX3, vwmaccsu_vx_h, WOP_SSU_H, H4, H2, DO_MACC)
1804 RVVCALL(OPIVX3, vwmaccsu_vx_w, WOP_SSU_W, H8, H4, DO_MACC)
1805 RVVCALL(OPIVX3, vwmaccus_vx_b, WOP_SUS_B, H2, H1, DO_MACC)
1806 RVVCALL(OPIVX3, vwmaccus_vx_h, WOP_SUS_H, H4, H2, DO_MACC)
1807 RVVCALL(OPIVX3, vwmaccus_vx_w, WOP_SUS_W, H8, H4, DO_MACC)
1808 GEN_VEXT_VX(vwmaccu_vx_b, 2)
1809 GEN_VEXT_VX(vwmaccu_vx_h, 4)
1810 GEN_VEXT_VX(vwmaccu_vx_w, 8)
1811 GEN_VEXT_VX(vwmacc_vx_b, 2)
1812 GEN_VEXT_VX(vwmacc_vx_h, 4)
1813 GEN_VEXT_VX(vwmacc_vx_w, 8)
1814 GEN_VEXT_VX(vwmaccsu_vx_b, 2)
1815 GEN_VEXT_VX(vwmaccsu_vx_h, 4)
1816 GEN_VEXT_VX(vwmaccsu_vx_w, 8)
1817 GEN_VEXT_VX(vwmaccus_vx_b, 2)
1818 GEN_VEXT_VX(vwmaccus_vx_h, 4)
1819 GEN_VEXT_VX(vwmaccus_vx_w, 8)
1820 
1821 /* Vector Integer Merge and Move Instructions */
1822 #define GEN_VEXT_VMV_VV(NAME, ETYPE, H)                              \
1823 void HELPER(NAME)(void *vd, void *vs1, CPURISCVState *env,           \
1824                   uint32_t desc)                                     \
1825 {                                                                    \
1826     uint32_t vl = env->vl;                                           \
1827     uint32_t esz = sizeof(ETYPE);                                    \
1828     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
1829     uint32_t vta = vext_vta(desc);                                   \
1830     uint32_t i;                                                      \
1831                                                                      \
1832     VSTART_CHECK_EARLY_EXIT(env);                                    \
1833                                                                      \
1834     for (i = env->vstart; i < vl; i++) {                             \
1835         ETYPE s1 = *((ETYPE *)vs1 + H(i));                           \
1836         *((ETYPE *)vd + H(i)) = s1;                                  \
1837     }                                                                \
1838     env->vstart = 0;                                                 \
1839     /* set tail elements to 1s */                                    \
1840     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
1841 }
1842 
1843 GEN_VEXT_VMV_VV(vmv_v_v_b, int8_t,  H1)
1844 GEN_VEXT_VMV_VV(vmv_v_v_h, int16_t, H2)
1845 GEN_VEXT_VMV_VV(vmv_v_v_w, int32_t, H4)
1846 GEN_VEXT_VMV_VV(vmv_v_v_d, int64_t, H8)
1847 
1848 #define GEN_VEXT_VMV_VX(NAME, ETYPE, H)                              \
1849 void HELPER(NAME)(void *vd, uint64_t s1, CPURISCVState *env,         \
1850                   uint32_t desc)                                     \
1851 {                                                                    \
1852     uint32_t vl = env->vl;                                           \
1853     uint32_t esz = sizeof(ETYPE);                                    \
1854     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
1855     uint32_t vta = vext_vta(desc);                                   \
1856     uint32_t i;                                                      \
1857                                                                      \
1858     VSTART_CHECK_EARLY_EXIT(env);                                    \
1859                                                                      \
1860     for (i = env->vstart; i < vl; i++) {                             \
1861         *((ETYPE *)vd + H(i)) = (ETYPE)s1;                           \
1862     }                                                                \
1863     env->vstart = 0;                                                 \
1864     /* set tail elements to 1s */                                    \
1865     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
1866 }
1867 
1868 GEN_VEXT_VMV_VX(vmv_v_x_b, int8_t,  H1)
1869 GEN_VEXT_VMV_VX(vmv_v_x_h, int16_t, H2)
1870 GEN_VEXT_VMV_VX(vmv_v_x_w, int32_t, H4)
1871 GEN_VEXT_VMV_VX(vmv_v_x_d, int64_t, H8)
1872 
1873 #define GEN_VEXT_VMERGE_VV(NAME, ETYPE, H)                           \
1874 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,          \
1875                   CPURISCVState *env, uint32_t desc)                 \
1876 {                                                                    \
1877     uint32_t vl = env->vl;                                           \
1878     uint32_t esz = sizeof(ETYPE);                                    \
1879     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
1880     uint32_t vta = vext_vta(desc);                                   \
1881     uint32_t i;                                                      \
1882                                                                      \
1883     VSTART_CHECK_EARLY_EXIT(env);                                    \
1884                                                                      \
1885     for (i = env->vstart; i < vl; i++) {                             \
1886         ETYPE *vt = (!vext_elem_mask(v0, i) ? vs2 : vs1);            \
1887         *((ETYPE *)vd + H(i)) = *(vt + H(i));                        \
1888     }                                                                \
1889     env->vstart = 0;                                                 \
1890     /* set tail elements to 1s */                                    \
1891     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
1892 }
1893 
1894 GEN_VEXT_VMERGE_VV(vmerge_vvm_b, int8_t,  H1)
1895 GEN_VEXT_VMERGE_VV(vmerge_vvm_h, int16_t, H2)
1896 GEN_VEXT_VMERGE_VV(vmerge_vvm_w, int32_t, H4)
1897 GEN_VEXT_VMERGE_VV(vmerge_vvm_d, int64_t, H8)
1898 
1899 #define GEN_VEXT_VMERGE_VX(NAME, ETYPE, H)                           \
1900 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,               \
1901                   void *vs2, CPURISCVState *env, uint32_t desc)      \
1902 {                                                                    \
1903     uint32_t vl = env->vl;                                           \
1904     uint32_t esz = sizeof(ETYPE);                                    \
1905     uint32_t total_elems = vext_get_total_elems(env, desc, esz);     \
1906     uint32_t vta = vext_vta(desc);                                   \
1907     uint32_t i;                                                      \
1908                                                                      \
1909     VSTART_CHECK_EARLY_EXIT(env);                                    \
1910                                                                      \
1911     for (i = env->vstart; i < vl; i++) {                             \
1912         ETYPE s2 = *((ETYPE *)vs2 + H(i));                           \
1913         ETYPE d = (!vext_elem_mask(v0, i) ? s2 :                     \
1914                    (ETYPE)(target_long)s1);                          \
1915         *((ETYPE *)vd + H(i)) = d;                                   \
1916     }                                                                \
1917     env->vstart = 0;                                                 \
1918     /* set tail elements to 1s */                                    \
1919     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);         \
1920 }
1921 
1922 GEN_VEXT_VMERGE_VX(vmerge_vxm_b, int8_t,  H1)
1923 GEN_VEXT_VMERGE_VX(vmerge_vxm_h, int16_t, H2)
1924 GEN_VEXT_VMERGE_VX(vmerge_vxm_w, int32_t, H4)
1925 GEN_VEXT_VMERGE_VX(vmerge_vxm_d, int64_t, H8)
1926 
1927 /*
1928  * Vector Fixed-Point Arithmetic Instructions
1929  */
1930 
1931 /* Vector Single-Width Saturating Add and Subtract */
1932 
1933 /*
1934  * As fixed point instructions probably have round mode and saturation,
1935  * define common macros for fixed point here.
1936  */
1937 typedef void opivv2_rm_fn(void *vd, void *vs1, void *vs2, int i,
1938                           CPURISCVState *env, int vxrm);
1939 
1940 #define OPIVV2_RM(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)     \
1941 static inline void                                                  \
1942 do_##NAME(void *vd, void *vs1, void *vs2, int i,                    \
1943           CPURISCVState *env, int vxrm)                             \
1944 {                                                                   \
1945     TX1 s1 = *((T1 *)vs1 + HS1(i));                                 \
1946     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
1947     *((TD *)vd + HD(i)) = OP(env, vxrm, s2, s1);                    \
1948 }
1949 
1950 static inline void
1951 vext_vv_rm_1(void *vd, void *v0, void *vs1, void *vs2,
1952              CPURISCVState *env,
1953              uint32_t vl, uint32_t vm, int vxrm,
1954              opivv2_rm_fn *fn, uint32_t vma, uint32_t esz)
1955 {
1956     VSTART_CHECK_EARLY_EXIT(env);
1957 
1958     for (uint32_t i = env->vstart; i < vl; i++) {
1959         if (!vm && !vext_elem_mask(v0, i)) {
1960             /* set masked-off elements to 1s */
1961             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);
1962             continue;
1963         }
1964         fn(vd, vs1, vs2, i, env, vxrm);
1965     }
1966     env->vstart = 0;
1967 }
1968 
1969 static inline void
1970 vext_vv_rm_2(void *vd, void *v0, void *vs1, void *vs2,
1971              CPURISCVState *env,
1972              uint32_t desc,
1973              opivv2_rm_fn *fn, uint32_t esz)
1974 {
1975     uint32_t vm = vext_vm(desc);
1976     uint32_t vl = env->vl;
1977     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
1978     uint32_t vta = vext_vta(desc);
1979     uint32_t vma = vext_vma(desc);
1980 
1981     switch (env->vxrm) {
1982     case 0: /* rnu */
1983         vext_vv_rm_1(vd, v0, vs1, vs2,
1984                      env, vl, vm, 0, fn, vma, esz);
1985         break;
1986     case 1: /* rne */
1987         vext_vv_rm_1(vd, v0, vs1, vs2,
1988                      env, vl, vm, 1, fn, vma, esz);
1989         break;
1990     case 2: /* rdn */
1991         vext_vv_rm_1(vd, v0, vs1, vs2,
1992                      env, vl, vm, 2, fn, vma, esz);
1993         break;
1994     default: /* rod */
1995         vext_vv_rm_1(vd, v0, vs1, vs2,
1996                      env, vl, vm, 3, fn, vma, esz);
1997         break;
1998     }
1999     /* set tail elements to 1s */
2000     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);
2001 }
2002 
2003 /* generate helpers for fixed point instructions with OPIVV format */
2004 #define GEN_VEXT_VV_RM(NAME, ESZ)                               \
2005 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,     \
2006                   CPURISCVState *env, uint32_t desc)            \
2007 {                                                               \
2008     vext_vv_rm_2(vd, v0, vs1, vs2, env, desc,                   \
2009                  do_##NAME, ESZ);                               \
2010 }
2011 
2012 static inline uint8_t saddu8(CPURISCVState *env, int vxrm, uint8_t a,
2013                              uint8_t b)
2014 {
2015     uint8_t res = a + b;
2016     if (res < a) {
2017         res = UINT8_MAX;
2018         env->vxsat = 0x1;
2019     }
2020     return res;
2021 }
2022 
2023 static inline uint16_t saddu16(CPURISCVState *env, int vxrm, uint16_t a,
2024                                uint16_t b)
2025 {
2026     uint16_t res = a + b;
2027     if (res < a) {
2028         res = UINT16_MAX;
2029         env->vxsat = 0x1;
2030     }
2031     return res;
2032 }
2033 
2034 static inline uint32_t saddu32(CPURISCVState *env, int vxrm, uint32_t a,
2035                                uint32_t b)
2036 {
2037     uint32_t res = a + b;
2038     if (res < a) {
2039         res = UINT32_MAX;
2040         env->vxsat = 0x1;
2041     }
2042     return res;
2043 }
2044 
2045 static inline uint64_t saddu64(CPURISCVState *env, int vxrm, uint64_t a,
2046                                uint64_t b)
2047 {
2048     uint64_t res = a + b;
2049     if (res < a) {
2050         res = UINT64_MAX;
2051         env->vxsat = 0x1;
2052     }
2053     return res;
2054 }
2055 
2056 RVVCALL(OPIVV2_RM, vsaddu_vv_b, OP_UUU_B, H1, H1, H1, saddu8)
2057 RVVCALL(OPIVV2_RM, vsaddu_vv_h, OP_UUU_H, H2, H2, H2, saddu16)
2058 RVVCALL(OPIVV2_RM, vsaddu_vv_w, OP_UUU_W, H4, H4, H4, saddu32)
2059 RVVCALL(OPIVV2_RM, vsaddu_vv_d, OP_UUU_D, H8, H8, H8, saddu64)
2060 GEN_VEXT_VV_RM(vsaddu_vv_b, 1)
2061 GEN_VEXT_VV_RM(vsaddu_vv_h, 2)
2062 GEN_VEXT_VV_RM(vsaddu_vv_w, 4)
2063 GEN_VEXT_VV_RM(vsaddu_vv_d, 8)
2064 
2065 typedef void opivx2_rm_fn(void *vd, target_long s1, void *vs2, int i,
2066                           CPURISCVState *env, int vxrm);
2067 
2068 #define OPIVX2_RM(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)          \
2069 static inline void                                                  \
2070 do_##NAME(void *vd, target_long s1, void *vs2, int i,               \
2071           CPURISCVState *env, int vxrm)                             \
2072 {                                                                   \
2073     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
2074     *((TD *)vd + HD(i)) = OP(env, vxrm, s2, (TX1)(T1)s1);           \
2075 }
2076 
2077 static inline void
2078 vext_vx_rm_1(void *vd, void *v0, target_long s1, void *vs2,
2079              CPURISCVState *env,
2080              uint32_t vl, uint32_t vm, int vxrm,
2081              opivx2_rm_fn *fn, uint32_t vma, uint32_t esz)
2082 {
2083     VSTART_CHECK_EARLY_EXIT(env);
2084 
2085     for (uint32_t i = env->vstart; i < vl; i++) {
2086         if (!vm && !vext_elem_mask(v0, i)) {
2087             /* set masked-off elements to 1s */
2088             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);
2089             continue;
2090         }
2091         fn(vd, s1, vs2, i, env, vxrm);
2092     }
2093     env->vstart = 0;
2094 }
2095 
2096 static inline void
2097 vext_vx_rm_2(void *vd, void *v0, target_long s1, void *vs2,
2098              CPURISCVState *env,
2099              uint32_t desc,
2100              opivx2_rm_fn *fn, uint32_t esz)
2101 {
2102     uint32_t vm = vext_vm(desc);
2103     uint32_t vl = env->vl;
2104     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
2105     uint32_t vta = vext_vta(desc);
2106     uint32_t vma = vext_vma(desc);
2107 
2108     switch (env->vxrm) {
2109     case 0: /* rnu */
2110         vext_vx_rm_1(vd, v0, s1, vs2,
2111                      env, vl, vm, 0, fn, vma, esz);
2112         break;
2113     case 1: /* rne */
2114         vext_vx_rm_1(vd, v0, s1, vs2,
2115                      env, vl, vm, 1, fn, vma, esz);
2116         break;
2117     case 2: /* rdn */
2118         vext_vx_rm_1(vd, v0, s1, vs2,
2119                      env, vl, vm, 2, fn, vma, esz);
2120         break;
2121     default: /* rod */
2122         vext_vx_rm_1(vd, v0, s1, vs2,
2123                      env, vl, vm, 3, fn, vma, esz);
2124         break;
2125     }
2126     /* set tail elements to 1s */
2127     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);
2128 }
2129 
2130 /* generate helpers for fixed point instructions with OPIVX format */
2131 #define GEN_VEXT_VX_RM(NAME, ESZ)                         \
2132 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,    \
2133                   void *vs2, CPURISCVState *env,          \
2134                   uint32_t desc)                          \
2135 {                                                         \
2136     vext_vx_rm_2(vd, v0, s1, vs2, env, desc,              \
2137                  do_##NAME, ESZ);                         \
2138 }
2139 
2140 RVVCALL(OPIVX2_RM, vsaddu_vx_b, OP_UUU_B, H1, H1, saddu8)
2141 RVVCALL(OPIVX2_RM, vsaddu_vx_h, OP_UUU_H, H2, H2, saddu16)
2142 RVVCALL(OPIVX2_RM, vsaddu_vx_w, OP_UUU_W, H4, H4, saddu32)
2143 RVVCALL(OPIVX2_RM, vsaddu_vx_d, OP_UUU_D, H8, H8, saddu64)
2144 GEN_VEXT_VX_RM(vsaddu_vx_b, 1)
2145 GEN_VEXT_VX_RM(vsaddu_vx_h, 2)
2146 GEN_VEXT_VX_RM(vsaddu_vx_w, 4)
2147 GEN_VEXT_VX_RM(vsaddu_vx_d, 8)
2148 
2149 static inline int8_t sadd8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2150 {
2151     int8_t res = a + b;
2152     if ((res ^ a) & (res ^ b) & INT8_MIN) {
2153         res = a > 0 ? INT8_MAX : INT8_MIN;
2154         env->vxsat = 0x1;
2155     }
2156     return res;
2157 }
2158 
2159 static inline int16_t sadd16(CPURISCVState *env, int vxrm, int16_t a,
2160                              int16_t b)
2161 {
2162     int16_t res = a + b;
2163     if ((res ^ a) & (res ^ b) & INT16_MIN) {
2164         res = a > 0 ? INT16_MAX : INT16_MIN;
2165         env->vxsat = 0x1;
2166     }
2167     return res;
2168 }
2169 
2170 static inline int32_t sadd32(CPURISCVState *env, int vxrm, int32_t a,
2171                              int32_t b)
2172 {
2173     int32_t res = a + b;
2174     if ((res ^ a) & (res ^ b) & INT32_MIN) {
2175         res = a > 0 ? INT32_MAX : INT32_MIN;
2176         env->vxsat = 0x1;
2177     }
2178     return res;
2179 }
2180 
2181 static inline int64_t sadd64(CPURISCVState *env, int vxrm, int64_t a,
2182                              int64_t b)
2183 {
2184     int64_t res = a + b;
2185     if ((res ^ a) & (res ^ b) & INT64_MIN) {
2186         res = a > 0 ? INT64_MAX : INT64_MIN;
2187         env->vxsat = 0x1;
2188     }
2189     return res;
2190 }
2191 
2192 RVVCALL(OPIVV2_RM, vsadd_vv_b, OP_SSS_B, H1, H1, H1, sadd8)
2193 RVVCALL(OPIVV2_RM, vsadd_vv_h, OP_SSS_H, H2, H2, H2, sadd16)
2194 RVVCALL(OPIVV2_RM, vsadd_vv_w, OP_SSS_W, H4, H4, H4, sadd32)
2195 RVVCALL(OPIVV2_RM, vsadd_vv_d, OP_SSS_D, H8, H8, H8, sadd64)
2196 GEN_VEXT_VV_RM(vsadd_vv_b, 1)
2197 GEN_VEXT_VV_RM(vsadd_vv_h, 2)
2198 GEN_VEXT_VV_RM(vsadd_vv_w, 4)
2199 GEN_VEXT_VV_RM(vsadd_vv_d, 8)
2200 
2201 RVVCALL(OPIVX2_RM, vsadd_vx_b, OP_SSS_B, H1, H1, sadd8)
2202 RVVCALL(OPIVX2_RM, vsadd_vx_h, OP_SSS_H, H2, H2, sadd16)
2203 RVVCALL(OPIVX2_RM, vsadd_vx_w, OP_SSS_W, H4, H4, sadd32)
2204 RVVCALL(OPIVX2_RM, vsadd_vx_d, OP_SSS_D, H8, H8, sadd64)
2205 GEN_VEXT_VX_RM(vsadd_vx_b, 1)
2206 GEN_VEXT_VX_RM(vsadd_vx_h, 2)
2207 GEN_VEXT_VX_RM(vsadd_vx_w, 4)
2208 GEN_VEXT_VX_RM(vsadd_vx_d, 8)
2209 
2210 static inline uint8_t ssubu8(CPURISCVState *env, int vxrm, uint8_t a,
2211                              uint8_t b)
2212 {
2213     uint8_t res = a - b;
2214     if (res > a) {
2215         res = 0;
2216         env->vxsat = 0x1;
2217     }
2218     return res;
2219 }
2220 
2221 static inline uint16_t ssubu16(CPURISCVState *env, int vxrm, uint16_t a,
2222                                uint16_t b)
2223 {
2224     uint16_t res = a - b;
2225     if (res > a) {
2226         res = 0;
2227         env->vxsat = 0x1;
2228     }
2229     return res;
2230 }
2231 
2232 static inline uint32_t ssubu32(CPURISCVState *env, int vxrm, uint32_t a,
2233                                uint32_t b)
2234 {
2235     uint32_t res = a - b;
2236     if (res > a) {
2237         res = 0;
2238         env->vxsat = 0x1;
2239     }
2240     return res;
2241 }
2242 
2243 static inline uint64_t ssubu64(CPURISCVState *env, int vxrm, uint64_t a,
2244                                uint64_t b)
2245 {
2246     uint64_t res = a - b;
2247     if (res > a) {
2248         res = 0;
2249         env->vxsat = 0x1;
2250     }
2251     return res;
2252 }
2253 
2254 RVVCALL(OPIVV2_RM, vssubu_vv_b, OP_UUU_B, H1, H1, H1, ssubu8)
2255 RVVCALL(OPIVV2_RM, vssubu_vv_h, OP_UUU_H, H2, H2, H2, ssubu16)
2256 RVVCALL(OPIVV2_RM, vssubu_vv_w, OP_UUU_W, H4, H4, H4, ssubu32)
2257 RVVCALL(OPIVV2_RM, vssubu_vv_d, OP_UUU_D, H8, H8, H8, ssubu64)
2258 GEN_VEXT_VV_RM(vssubu_vv_b, 1)
2259 GEN_VEXT_VV_RM(vssubu_vv_h, 2)
2260 GEN_VEXT_VV_RM(vssubu_vv_w, 4)
2261 GEN_VEXT_VV_RM(vssubu_vv_d, 8)
2262 
2263 RVVCALL(OPIVX2_RM, vssubu_vx_b, OP_UUU_B, H1, H1, ssubu8)
2264 RVVCALL(OPIVX2_RM, vssubu_vx_h, OP_UUU_H, H2, H2, ssubu16)
2265 RVVCALL(OPIVX2_RM, vssubu_vx_w, OP_UUU_W, H4, H4, ssubu32)
2266 RVVCALL(OPIVX2_RM, vssubu_vx_d, OP_UUU_D, H8, H8, ssubu64)
2267 GEN_VEXT_VX_RM(vssubu_vx_b, 1)
2268 GEN_VEXT_VX_RM(vssubu_vx_h, 2)
2269 GEN_VEXT_VX_RM(vssubu_vx_w, 4)
2270 GEN_VEXT_VX_RM(vssubu_vx_d, 8)
2271 
2272 static inline int8_t ssub8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2273 {
2274     int8_t res = a - b;
2275     if ((res ^ a) & (a ^ b) & INT8_MIN) {
2276         res = a >= 0 ? INT8_MAX : INT8_MIN;
2277         env->vxsat = 0x1;
2278     }
2279     return res;
2280 }
2281 
2282 static inline int16_t ssub16(CPURISCVState *env, int vxrm, int16_t a,
2283                              int16_t b)
2284 {
2285     int16_t res = a - b;
2286     if ((res ^ a) & (a ^ b) & INT16_MIN) {
2287         res = a >= 0 ? INT16_MAX : INT16_MIN;
2288         env->vxsat = 0x1;
2289     }
2290     return res;
2291 }
2292 
2293 static inline int32_t ssub32(CPURISCVState *env, int vxrm, int32_t a,
2294                              int32_t b)
2295 {
2296     int32_t res = a - b;
2297     if ((res ^ a) & (a ^ b) & INT32_MIN) {
2298         res = a >= 0 ? INT32_MAX : INT32_MIN;
2299         env->vxsat = 0x1;
2300     }
2301     return res;
2302 }
2303 
2304 static inline int64_t ssub64(CPURISCVState *env, int vxrm, int64_t a,
2305                              int64_t b)
2306 {
2307     int64_t res = a - b;
2308     if ((res ^ a) & (a ^ b) & INT64_MIN) {
2309         res = a >= 0 ? INT64_MAX : INT64_MIN;
2310         env->vxsat = 0x1;
2311     }
2312     return res;
2313 }
2314 
2315 RVVCALL(OPIVV2_RM, vssub_vv_b, OP_SSS_B, H1, H1, H1, ssub8)
2316 RVVCALL(OPIVV2_RM, vssub_vv_h, OP_SSS_H, H2, H2, H2, ssub16)
2317 RVVCALL(OPIVV2_RM, vssub_vv_w, OP_SSS_W, H4, H4, H4, ssub32)
2318 RVVCALL(OPIVV2_RM, vssub_vv_d, OP_SSS_D, H8, H8, H8, ssub64)
2319 GEN_VEXT_VV_RM(vssub_vv_b, 1)
2320 GEN_VEXT_VV_RM(vssub_vv_h, 2)
2321 GEN_VEXT_VV_RM(vssub_vv_w, 4)
2322 GEN_VEXT_VV_RM(vssub_vv_d, 8)
2323 
2324 RVVCALL(OPIVX2_RM, vssub_vx_b, OP_SSS_B, H1, H1, ssub8)
2325 RVVCALL(OPIVX2_RM, vssub_vx_h, OP_SSS_H, H2, H2, ssub16)
2326 RVVCALL(OPIVX2_RM, vssub_vx_w, OP_SSS_W, H4, H4, ssub32)
2327 RVVCALL(OPIVX2_RM, vssub_vx_d, OP_SSS_D, H8, H8, ssub64)
2328 GEN_VEXT_VX_RM(vssub_vx_b, 1)
2329 GEN_VEXT_VX_RM(vssub_vx_h, 2)
2330 GEN_VEXT_VX_RM(vssub_vx_w, 4)
2331 GEN_VEXT_VX_RM(vssub_vx_d, 8)
2332 
2333 /* Vector Single-Width Averaging Add and Subtract */
2334 static inline uint8_t get_round(int vxrm, uint64_t v, uint8_t shift)
2335 {
2336     uint8_t d = extract64(v, shift, 1);
2337     uint8_t d1;
2338     uint64_t D1, D2;
2339 
2340     if (shift == 0 || shift > 64) {
2341         return 0;
2342     }
2343 
2344     d1 = extract64(v, shift - 1, 1);
2345     D1 = extract64(v, 0, shift);
2346     if (vxrm == 0) { /* round-to-nearest-up (add +0.5 LSB) */
2347         return d1;
2348     } else if (vxrm == 1) { /* round-to-nearest-even */
2349         if (shift > 1) {
2350             D2 = extract64(v, 0, shift - 1);
2351             return d1 & ((D2 != 0) | d);
2352         } else {
2353             return d1 & d;
2354         }
2355     } else if (vxrm == 3) { /* round-to-odd (OR bits into LSB, aka "jam") */
2356         return !d & (D1 != 0);
2357     }
2358     return 0; /* round-down (truncate) */
2359 }
2360 
2361 static inline int32_t aadd32(CPURISCVState *env, int vxrm, int32_t a,
2362                              int32_t b)
2363 {
2364     int64_t res = (int64_t)a + b;
2365     uint8_t round = get_round(vxrm, res, 1);
2366 
2367     return (res >> 1) + round;
2368 }
2369 
2370 static inline int64_t aadd64(CPURISCVState *env, int vxrm, int64_t a,
2371                              int64_t b)
2372 {
2373     int64_t res = a + b;
2374     uint8_t round = get_round(vxrm, res, 1);
2375     int64_t over = (res ^ a) & (res ^ b) & INT64_MIN;
2376 
2377     /* With signed overflow, bit 64 is inverse of bit 63. */
2378     return ((res >> 1) ^ over) + round;
2379 }
2380 
2381 RVVCALL(OPIVV2_RM, vaadd_vv_b, OP_SSS_B, H1, H1, H1, aadd32)
2382 RVVCALL(OPIVV2_RM, vaadd_vv_h, OP_SSS_H, H2, H2, H2, aadd32)
2383 RVVCALL(OPIVV2_RM, vaadd_vv_w, OP_SSS_W, H4, H4, H4, aadd32)
2384 RVVCALL(OPIVV2_RM, vaadd_vv_d, OP_SSS_D, H8, H8, H8, aadd64)
2385 GEN_VEXT_VV_RM(vaadd_vv_b, 1)
2386 GEN_VEXT_VV_RM(vaadd_vv_h, 2)
2387 GEN_VEXT_VV_RM(vaadd_vv_w, 4)
2388 GEN_VEXT_VV_RM(vaadd_vv_d, 8)
2389 
2390 RVVCALL(OPIVX2_RM, vaadd_vx_b, OP_SSS_B, H1, H1, aadd32)
2391 RVVCALL(OPIVX2_RM, vaadd_vx_h, OP_SSS_H, H2, H2, aadd32)
2392 RVVCALL(OPIVX2_RM, vaadd_vx_w, OP_SSS_W, H4, H4, aadd32)
2393 RVVCALL(OPIVX2_RM, vaadd_vx_d, OP_SSS_D, H8, H8, aadd64)
2394 GEN_VEXT_VX_RM(vaadd_vx_b, 1)
2395 GEN_VEXT_VX_RM(vaadd_vx_h, 2)
2396 GEN_VEXT_VX_RM(vaadd_vx_w, 4)
2397 GEN_VEXT_VX_RM(vaadd_vx_d, 8)
2398 
2399 static inline uint32_t aaddu32(CPURISCVState *env, int vxrm,
2400                                uint32_t a, uint32_t b)
2401 {
2402     uint64_t res = (uint64_t)a + b;
2403     uint8_t round = get_round(vxrm, res, 1);
2404 
2405     return (res >> 1) + round;
2406 }
2407 
2408 static inline uint64_t aaddu64(CPURISCVState *env, int vxrm,
2409                                uint64_t a, uint64_t b)
2410 {
2411     uint64_t res = a + b;
2412     uint8_t round = get_round(vxrm, res, 1);
2413     uint64_t over = (uint64_t)(res < a) << 63;
2414 
2415     return ((res >> 1) | over) + round;
2416 }
2417 
2418 RVVCALL(OPIVV2_RM, vaaddu_vv_b, OP_UUU_B, H1, H1, H1, aaddu32)
2419 RVVCALL(OPIVV2_RM, vaaddu_vv_h, OP_UUU_H, H2, H2, H2, aaddu32)
2420 RVVCALL(OPIVV2_RM, vaaddu_vv_w, OP_UUU_W, H4, H4, H4, aaddu32)
2421 RVVCALL(OPIVV2_RM, vaaddu_vv_d, OP_UUU_D, H8, H8, H8, aaddu64)
2422 GEN_VEXT_VV_RM(vaaddu_vv_b, 1)
2423 GEN_VEXT_VV_RM(vaaddu_vv_h, 2)
2424 GEN_VEXT_VV_RM(vaaddu_vv_w, 4)
2425 GEN_VEXT_VV_RM(vaaddu_vv_d, 8)
2426 
2427 RVVCALL(OPIVX2_RM, vaaddu_vx_b, OP_UUU_B, H1, H1, aaddu32)
2428 RVVCALL(OPIVX2_RM, vaaddu_vx_h, OP_UUU_H, H2, H2, aaddu32)
2429 RVVCALL(OPIVX2_RM, vaaddu_vx_w, OP_UUU_W, H4, H4, aaddu32)
2430 RVVCALL(OPIVX2_RM, vaaddu_vx_d, OP_UUU_D, H8, H8, aaddu64)
2431 GEN_VEXT_VX_RM(vaaddu_vx_b, 1)
2432 GEN_VEXT_VX_RM(vaaddu_vx_h, 2)
2433 GEN_VEXT_VX_RM(vaaddu_vx_w, 4)
2434 GEN_VEXT_VX_RM(vaaddu_vx_d, 8)
2435 
2436 static inline int32_t asub32(CPURISCVState *env, int vxrm, int32_t a,
2437                              int32_t b)
2438 {
2439     int64_t res = (int64_t)a - b;
2440     uint8_t round = get_round(vxrm, res, 1);
2441 
2442     return (res >> 1) + round;
2443 }
2444 
2445 static inline int64_t asub64(CPURISCVState *env, int vxrm, int64_t a,
2446                              int64_t b)
2447 {
2448     int64_t res = (int64_t)a - b;
2449     uint8_t round = get_round(vxrm, res, 1);
2450     int64_t over = (res ^ a) & (a ^ b) & INT64_MIN;
2451 
2452     /* With signed overflow, bit 64 is inverse of bit 63. */
2453     return ((res >> 1) ^ over) + round;
2454 }
2455 
2456 RVVCALL(OPIVV2_RM, vasub_vv_b, OP_SSS_B, H1, H1, H1, asub32)
2457 RVVCALL(OPIVV2_RM, vasub_vv_h, OP_SSS_H, H2, H2, H2, asub32)
2458 RVVCALL(OPIVV2_RM, vasub_vv_w, OP_SSS_W, H4, H4, H4, asub32)
2459 RVVCALL(OPIVV2_RM, vasub_vv_d, OP_SSS_D, H8, H8, H8, asub64)
2460 GEN_VEXT_VV_RM(vasub_vv_b, 1)
2461 GEN_VEXT_VV_RM(vasub_vv_h, 2)
2462 GEN_VEXT_VV_RM(vasub_vv_w, 4)
2463 GEN_VEXT_VV_RM(vasub_vv_d, 8)
2464 
2465 RVVCALL(OPIVX2_RM, vasub_vx_b, OP_SSS_B, H1, H1, asub32)
2466 RVVCALL(OPIVX2_RM, vasub_vx_h, OP_SSS_H, H2, H2, asub32)
2467 RVVCALL(OPIVX2_RM, vasub_vx_w, OP_SSS_W, H4, H4, asub32)
2468 RVVCALL(OPIVX2_RM, vasub_vx_d, OP_SSS_D, H8, H8, asub64)
2469 GEN_VEXT_VX_RM(vasub_vx_b, 1)
2470 GEN_VEXT_VX_RM(vasub_vx_h, 2)
2471 GEN_VEXT_VX_RM(vasub_vx_w, 4)
2472 GEN_VEXT_VX_RM(vasub_vx_d, 8)
2473 
2474 static inline uint32_t asubu32(CPURISCVState *env, int vxrm,
2475                                uint32_t a, uint32_t b)
2476 {
2477     int64_t res = (int64_t)a - b;
2478     uint8_t round = get_round(vxrm, res, 1);
2479 
2480     return (res >> 1) + round;
2481 }
2482 
2483 static inline uint64_t asubu64(CPURISCVState *env, int vxrm,
2484                                uint64_t a, uint64_t b)
2485 {
2486     uint64_t res = (uint64_t)a - b;
2487     uint8_t round = get_round(vxrm, res, 1);
2488     uint64_t over = (uint64_t)(res > a) << 63;
2489 
2490     return ((res >> 1) | over) + round;
2491 }
2492 
2493 RVVCALL(OPIVV2_RM, vasubu_vv_b, OP_UUU_B, H1, H1, H1, asubu32)
2494 RVVCALL(OPIVV2_RM, vasubu_vv_h, OP_UUU_H, H2, H2, H2, asubu32)
2495 RVVCALL(OPIVV2_RM, vasubu_vv_w, OP_UUU_W, H4, H4, H4, asubu32)
2496 RVVCALL(OPIVV2_RM, vasubu_vv_d, OP_UUU_D, H8, H8, H8, asubu64)
2497 GEN_VEXT_VV_RM(vasubu_vv_b, 1)
2498 GEN_VEXT_VV_RM(vasubu_vv_h, 2)
2499 GEN_VEXT_VV_RM(vasubu_vv_w, 4)
2500 GEN_VEXT_VV_RM(vasubu_vv_d, 8)
2501 
2502 RVVCALL(OPIVX2_RM, vasubu_vx_b, OP_UUU_B, H1, H1, asubu32)
2503 RVVCALL(OPIVX2_RM, vasubu_vx_h, OP_UUU_H, H2, H2, asubu32)
2504 RVVCALL(OPIVX2_RM, vasubu_vx_w, OP_UUU_W, H4, H4, asubu32)
2505 RVVCALL(OPIVX2_RM, vasubu_vx_d, OP_UUU_D, H8, H8, asubu64)
2506 GEN_VEXT_VX_RM(vasubu_vx_b, 1)
2507 GEN_VEXT_VX_RM(vasubu_vx_h, 2)
2508 GEN_VEXT_VX_RM(vasubu_vx_w, 4)
2509 GEN_VEXT_VX_RM(vasubu_vx_d, 8)
2510 
2511 /* Vector Single-Width Fractional Multiply with Rounding and Saturation */
2512 static inline int8_t vsmul8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2513 {
2514     uint8_t round;
2515     int16_t res;
2516 
2517     res = (int16_t)a * (int16_t)b;
2518     round = get_round(vxrm, res, 7);
2519     res = (res >> 7) + round;
2520 
2521     if (res > INT8_MAX) {
2522         env->vxsat = 0x1;
2523         return INT8_MAX;
2524     } else if (res < INT8_MIN) {
2525         env->vxsat = 0x1;
2526         return INT8_MIN;
2527     } else {
2528         return res;
2529     }
2530 }
2531 
2532 static int16_t vsmul16(CPURISCVState *env, int vxrm, int16_t a, int16_t b)
2533 {
2534     uint8_t round;
2535     int32_t res;
2536 
2537     res = (int32_t)a * (int32_t)b;
2538     round = get_round(vxrm, res, 15);
2539     res = (res >> 15) + round;
2540 
2541     if (res > INT16_MAX) {
2542         env->vxsat = 0x1;
2543         return INT16_MAX;
2544     } else if (res < INT16_MIN) {
2545         env->vxsat = 0x1;
2546         return INT16_MIN;
2547     } else {
2548         return res;
2549     }
2550 }
2551 
2552 static int32_t vsmul32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2553 {
2554     uint8_t round;
2555     int64_t res;
2556 
2557     res = (int64_t)a * (int64_t)b;
2558     round = get_round(vxrm, res, 31);
2559     res = (res >> 31) + round;
2560 
2561     if (res > INT32_MAX) {
2562         env->vxsat = 0x1;
2563         return INT32_MAX;
2564     } else if (res < INT32_MIN) {
2565         env->vxsat = 0x1;
2566         return INT32_MIN;
2567     } else {
2568         return res;
2569     }
2570 }
2571 
2572 static int64_t vsmul64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2573 {
2574     uint8_t round;
2575     uint64_t hi_64, lo_64;
2576     int64_t res;
2577 
2578     if (a == INT64_MIN && b == INT64_MIN) {
2579         env->vxsat = 1;
2580         return INT64_MAX;
2581     }
2582 
2583     muls64(&lo_64, &hi_64, a, b);
2584     round = get_round(vxrm, lo_64, 63);
2585     /*
2586      * Cannot overflow, as there are always
2587      * 2 sign bits after multiply.
2588      */
2589     res = (hi_64 << 1) | (lo_64 >> 63);
2590     if (round) {
2591         if (res == INT64_MAX) {
2592             env->vxsat = 1;
2593         } else {
2594             res += 1;
2595         }
2596     }
2597     return res;
2598 }
2599 
2600 RVVCALL(OPIVV2_RM, vsmul_vv_b, OP_SSS_B, H1, H1, H1, vsmul8)
2601 RVVCALL(OPIVV2_RM, vsmul_vv_h, OP_SSS_H, H2, H2, H2, vsmul16)
2602 RVVCALL(OPIVV2_RM, vsmul_vv_w, OP_SSS_W, H4, H4, H4, vsmul32)
2603 RVVCALL(OPIVV2_RM, vsmul_vv_d, OP_SSS_D, H8, H8, H8, vsmul64)
2604 GEN_VEXT_VV_RM(vsmul_vv_b, 1)
2605 GEN_VEXT_VV_RM(vsmul_vv_h, 2)
2606 GEN_VEXT_VV_RM(vsmul_vv_w, 4)
2607 GEN_VEXT_VV_RM(vsmul_vv_d, 8)
2608 
2609 RVVCALL(OPIVX2_RM, vsmul_vx_b, OP_SSS_B, H1, H1, vsmul8)
2610 RVVCALL(OPIVX2_RM, vsmul_vx_h, OP_SSS_H, H2, H2, vsmul16)
2611 RVVCALL(OPIVX2_RM, vsmul_vx_w, OP_SSS_W, H4, H4, vsmul32)
2612 RVVCALL(OPIVX2_RM, vsmul_vx_d, OP_SSS_D, H8, H8, vsmul64)
2613 GEN_VEXT_VX_RM(vsmul_vx_b, 1)
2614 GEN_VEXT_VX_RM(vsmul_vx_h, 2)
2615 GEN_VEXT_VX_RM(vsmul_vx_w, 4)
2616 GEN_VEXT_VX_RM(vsmul_vx_d, 8)
2617 
2618 /* Vector Single-Width Scaling Shift Instructions */
2619 static inline uint8_t
2620 vssrl8(CPURISCVState *env, int vxrm, uint8_t a, uint8_t b)
2621 {
2622     uint8_t round, shift = b & 0x7;
2623     uint8_t res;
2624 
2625     round = get_round(vxrm, a, shift);
2626     res = (a >> shift) + round;
2627     return res;
2628 }
2629 static inline uint16_t
2630 vssrl16(CPURISCVState *env, int vxrm, uint16_t a, uint16_t b)
2631 {
2632     uint8_t round, shift = b & 0xf;
2633 
2634     round = get_round(vxrm, a, shift);
2635     return (a >> shift) + round;
2636 }
2637 static inline uint32_t
2638 vssrl32(CPURISCVState *env, int vxrm, uint32_t a, uint32_t b)
2639 {
2640     uint8_t round, shift = b & 0x1f;
2641 
2642     round = get_round(vxrm, a, shift);
2643     return (a >> shift) + round;
2644 }
2645 static inline uint64_t
2646 vssrl64(CPURISCVState *env, int vxrm, uint64_t a, uint64_t b)
2647 {
2648     uint8_t round, shift = b & 0x3f;
2649 
2650     round = get_round(vxrm, a, shift);
2651     return (a >> shift) + round;
2652 }
2653 RVVCALL(OPIVV2_RM, vssrl_vv_b, OP_UUU_B, H1, H1, H1, vssrl8)
2654 RVVCALL(OPIVV2_RM, vssrl_vv_h, OP_UUU_H, H2, H2, H2, vssrl16)
2655 RVVCALL(OPIVV2_RM, vssrl_vv_w, OP_UUU_W, H4, H4, H4, vssrl32)
2656 RVVCALL(OPIVV2_RM, vssrl_vv_d, OP_UUU_D, H8, H8, H8, vssrl64)
2657 GEN_VEXT_VV_RM(vssrl_vv_b, 1)
2658 GEN_VEXT_VV_RM(vssrl_vv_h, 2)
2659 GEN_VEXT_VV_RM(vssrl_vv_w, 4)
2660 GEN_VEXT_VV_RM(vssrl_vv_d, 8)
2661 
2662 RVVCALL(OPIVX2_RM, vssrl_vx_b, OP_UUU_B, H1, H1, vssrl8)
2663 RVVCALL(OPIVX2_RM, vssrl_vx_h, OP_UUU_H, H2, H2, vssrl16)
2664 RVVCALL(OPIVX2_RM, vssrl_vx_w, OP_UUU_W, H4, H4, vssrl32)
2665 RVVCALL(OPIVX2_RM, vssrl_vx_d, OP_UUU_D, H8, H8, vssrl64)
2666 GEN_VEXT_VX_RM(vssrl_vx_b, 1)
2667 GEN_VEXT_VX_RM(vssrl_vx_h, 2)
2668 GEN_VEXT_VX_RM(vssrl_vx_w, 4)
2669 GEN_VEXT_VX_RM(vssrl_vx_d, 8)
2670 
2671 static inline int8_t
2672 vssra8(CPURISCVState *env, int vxrm, int8_t a, int8_t b)
2673 {
2674     uint8_t round, shift = b & 0x7;
2675 
2676     round = get_round(vxrm, a, shift);
2677     return (a >> shift) + round;
2678 }
2679 static inline int16_t
2680 vssra16(CPURISCVState *env, int vxrm, int16_t a, int16_t b)
2681 {
2682     uint8_t round, shift = b & 0xf;
2683 
2684     round = get_round(vxrm, a, shift);
2685     return (a >> shift) + round;
2686 }
2687 static inline int32_t
2688 vssra32(CPURISCVState *env, int vxrm, int32_t a, int32_t b)
2689 {
2690     uint8_t round, shift = b & 0x1f;
2691 
2692     round = get_round(vxrm, a, shift);
2693     return (a >> shift) + round;
2694 }
2695 static inline int64_t
2696 vssra64(CPURISCVState *env, int vxrm, int64_t a, int64_t b)
2697 {
2698     uint8_t round, shift = b & 0x3f;
2699 
2700     round = get_round(vxrm, a, shift);
2701     return (a >> shift) + round;
2702 }
2703 
2704 RVVCALL(OPIVV2_RM, vssra_vv_b, OP_SSS_B, H1, H1, H1, vssra8)
2705 RVVCALL(OPIVV2_RM, vssra_vv_h, OP_SSS_H, H2, H2, H2, vssra16)
2706 RVVCALL(OPIVV2_RM, vssra_vv_w, OP_SSS_W, H4, H4, H4, vssra32)
2707 RVVCALL(OPIVV2_RM, vssra_vv_d, OP_SSS_D, H8, H8, H8, vssra64)
2708 GEN_VEXT_VV_RM(vssra_vv_b, 1)
2709 GEN_VEXT_VV_RM(vssra_vv_h, 2)
2710 GEN_VEXT_VV_RM(vssra_vv_w, 4)
2711 GEN_VEXT_VV_RM(vssra_vv_d, 8)
2712 
2713 RVVCALL(OPIVX2_RM, vssra_vx_b, OP_SSS_B, H1, H1, vssra8)
2714 RVVCALL(OPIVX2_RM, vssra_vx_h, OP_SSS_H, H2, H2, vssra16)
2715 RVVCALL(OPIVX2_RM, vssra_vx_w, OP_SSS_W, H4, H4, vssra32)
2716 RVVCALL(OPIVX2_RM, vssra_vx_d, OP_SSS_D, H8, H8, vssra64)
2717 GEN_VEXT_VX_RM(vssra_vx_b, 1)
2718 GEN_VEXT_VX_RM(vssra_vx_h, 2)
2719 GEN_VEXT_VX_RM(vssra_vx_w, 4)
2720 GEN_VEXT_VX_RM(vssra_vx_d, 8)
2721 
2722 /* Vector Narrowing Fixed-Point Clip Instructions */
2723 static inline int8_t
2724 vnclip8(CPURISCVState *env, int vxrm, int16_t a, int8_t b)
2725 {
2726     uint8_t round, shift = b & 0xf;
2727     int16_t res;
2728 
2729     round = get_round(vxrm, a, shift);
2730     res = (a >> shift) + round;
2731     if (res > INT8_MAX) {
2732         env->vxsat = 0x1;
2733         return INT8_MAX;
2734     } else if (res < INT8_MIN) {
2735         env->vxsat = 0x1;
2736         return INT8_MIN;
2737     } else {
2738         return res;
2739     }
2740 }
2741 
2742 static inline int16_t
2743 vnclip16(CPURISCVState *env, int vxrm, int32_t a, int16_t b)
2744 {
2745     uint8_t round, shift = b & 0x1f;
2746     int32_t res;
2747 
2748     round = get_round(vxrm, a, shift);
2749     res = (a >> shift) + round;
2750     if (res > INT16_MAX) {
2751         env->vxsat = 0x1;
2752         return INT16_MAX;
2753     } else if (res < INT16_MIN) {
2754         env->vxsat = 0x1;
2755         return INT16_MIN;
2756     } else {
2757         return res;
2758     }
2759 }
2760 
2761 static inline int32_t
2762 vnclip32(CPURISCVState *env, int vxrm, int64_t a, int32_t b)
2763 {
2764     uint8_t round, shift = b & 0x3f;
2765     int64_t res;
2766 
2767     round = get_round(vxrm, a, shift);
2768     res = (a >> shift) + round;
2769     if (res > INT32_MAX) {
2770         env->vxsat = 0x1;
2771         return INT32_MAX;
2772     } else if (res < INT32_MIN) {
2773         env->vxsat = 0x1;
2774         return INT32_MIN;
2775     } else {
2776         return res;
2777     }
2778 }
2779 
2780 RVVCALL(OPIVV2_RM, vnclip_wv_b, NOP_SSS_B, H1, H2, H1, vnclip8)
2781 RVVCALL(OPIVV2_RM, vnclip_wv_h, NOP_SSS_H, H2, H4, H2, vnclip16)
2782 RVVCALL(OPIVV2_RM, vnclip_wv_w, NOP_SSS_W, H4, H8, H4, vnclip32)
2783 GEN_VEXT_VV_RM(vnclip_wv_b, 1)
2784 GEN_VEXT_VV_RM(vnclip_wv_h, 2)
2785 GEN_VEXT_VV_RM(vnclip_wv_w, 4)
2786 
2787 RVVCALL(OPIVX2_RM, vnclip_wx_b, NOP_SSS_B, H1, H2, vnclip8)
2788 RVVCALL(OPIVX2_RM, vnclip_wx_h, NOP_SSS_H, H2, H4, vnclip16)
2789 RVVCALL(OPIVX2_RM, vnclip_wx_w, NOP_SSS_W, H4, H8, vnclip32)
2790 GEN_VEXT_VX_RM(vnclip_wx_b, 1)
2791 GEN_VEXT_VX_RM(vnclip_wx_h, 2)
2792 GEN_VEXT_VX_RM(vnclip_wx_w, 4)
2793 
2794 static inline uint8_t
2795 vnclipu8(CPURISCVState *env, int vxrm, uint16_t a, uint8_t b)
2796 {
2797     uint8_t round, shift = b & 0xf;
2798     uint16_t res;
2799 
2800     round = get_round(vxrm, a, shift);
2801     res = (a >> shift) + round;
2802     if (res > UINT8_MAX) {
2803         env->vxsat = 0x1;
2804         return UINT8_MAX;
2805     } else {
2806         return res;
2807     }
2808 }
2809 
2810 static inline uint16_t
2811 vnclipu16(CPURISCVState *env, int vxrm, uint32_t a, uint16_t b)
2812 {
2813     uint8_t round, shift = b & 0x1f;
2814     uint32_t res;
2815 
2816     round = get_round(vxrm, a, shift);
2817     res = (a >> shift) + round;
2818     if (res > UINT16_MAX) {
2819         env->vxsat = 0x1;
2820         return UINT16_MAX;
2821     } else {
2822         return res;
2823     }
2824 }
2825 
2826 static inline uint32_t
2827 vnclipu32(CPURISCVState *env, int vxrm, uint64_t a, uint32_t b)
2828 {
2829     uint8_t round, shift = b & 0x3f;
2830     uint64_t res;
2831 
2832     round = get_round(vxrm, a, shift);
2833     res = (a >> shift) + round;
2834     if (res > UINT32_MAX) {
2835         env->vxsat = 0x1;
2836         return UINT32_MAX;
2837     } else {
2838         return res;
2839     }
2840 }
2841 
2842 RVVCALL(OPIVV2_RM, vnclipu_wv_b, NOP_UUU_B, H1, H2, H1, vnclipu8)
2843 RVVCALL(OPIVV2_RM, vnclipu_wv_h, NOP_UUU_H, H2, H4, H2, vnclipu16)
2844 RVVCALL(OPIVV2_RM, vnclipu_wv_w, NOP_UUU_W, H4, H8, H4, vnclipu32)
2845 GEN_VEXT_VV_RM(vnclipu_wv_b, 1)
2846 GEN_VEXT_VV_RM(vnclipu_wv_h, 2)
2847 GEN_VEXT_VV_RM(vnclipu_wv_w, 4)
2848 
2849 RVVCALL(OPIVX2_RM, vnclipu_wx_b, NOP_UUU_B, H1, H2, vnclipu8)
2850 RVVCALL(OPIVX2_RM, vnclipu_wx_h, NOP_UUU_H, H2, H4, vnclipu16)
2851 RVVCALL(OPIVX2_RM, vnclipu_wx_w, NOP_UUU_W, H4, H8, vnclipu32)
2852 GEN_VEXT_VX_RM(vnclipu_wx_b, 1)
2853 GEN_VEXT_VX_RM(vnclipu_wx_h, 2)
2854 GEN_VEXT_VX_RM(vnclipu_wx_w, 4)
2855 
2856 /*
2857  * Vector Float Point Arithmetic Instructions
2858  */
2859 /* Vector Single-Width Floating-Point Add/Subtract Instructions */
2860 #define OPFVV2(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)   \
2861 static void do_##NAME(void *vd, void *vs1, void *vs2, int i,   \
2862                       CPURISCVState *env)                      \
2863 {                                                              \
2864     TX1 s1 = *((T1 *)vs1 + HS1(i));                            \
2865     TX2 s2 = *((T2 *)vs2 + HS2(i));                            \
2866     *((TD *)vd + HD(i)) = OP(s2, s1, &env->fp_status);         \
2867 }
2868 
2869 #define GEN_VEXT_VV_ENV(NAME, ESZ)                        \
2870 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
2871                   void *vs2, CPURISCVState *env,          \
2872                   uint32_t desc)                          \
2873 {                                                         \
2874     uint32_t vm = vext_vm(desc);                          \
2875     uint32_t vl = env->vl;                                \
2876     uint32_t total_elems =                                \
2877         vext_get_total_elems(env, desc, ESZ);             \
2878     uint32_t vta = vext_vta(desc);                        \
2879     uint32_t vma = vext_vma(desc);                        \
2880     uint32_t i;                                           \
2881                                                           \
2882     VSTART_CHECK_EARLY_EXIT(env);                         \
2883                                                           \
2884     for (i = env->vstart; i < vl; i++) {                  \
2885         if (!vm && !vext_elem_mask(v0, i)) {              \
2886             /* set masked-off elements to 1s */           \
2887             vext_set_elems_1s(vd, vma, i * ESZ,           \
2888                               (i + 1) * ESZ);             \
2889             continue;                                     \
2890         }                                                 \
2891         do_##NAME(vd, vs1, vs2, i, env);                  \
2892     }                                                     \
2893     env->vstart = 0;                                      \
2894     /* set tail elements to 1s */                         \
2895     vext_set_elems_1s(vd, vta, vl * ESZ,                  \
2896                       total_elems * ESZ);                 \
2897 }
2898 
2899 RVVCALL(OPFVV2, vfadd_vv_h, OP_UUU_H, H2, H2, H2, float16_add)
2900 RVVCALL(OPFVV2, vfadd_vv_w, OP_UUU_W, H4, H4, H4, float32_add)
2901 RVVCALL(OPFVV2, vfadd_vv_d, OP_UUU_D, H8, H8, H8, float64_add)
2902 GEN_VEXT_VV_ENV(vfadd_vv_h, 2)
2903 GEN_VEXT_VV_ENV(vfadd_vv_w, 4)
2904 GEN_VEXT_VV_ENV(vfadd_vv_d, 8)
2905 
2906 #define OPFVF2(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)        \
2907 static void do_##NAME(void *vd, uint64_t s1, void *vs2, int i, \
2908                       CPURISCVState *env)                      \
2909 {                                                              \
2910     TX2 s2 = *((T2 *)vs2 + HS2(i));                            \
2911     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1, &env->fp_status);\
2912 }
2913 
2914 #define GEN_VEXT_VF(NAME, ESZ)                            \
2915 void HELPER(NAME)(void *vd, void *v0, uint64_t s1,        \
2916                   void *vs2, CPURISCVState *env,          \
2917                   uint32_t desc)                          \
2918 {                                                         \
2919     uint32_t vm = vext_vm(desc);                          \
2920     uint32_t vl = env->vl;                                \
2921     uint32_t total_elems =                                \
2922         vext_get_total_elems(env, desc, ESZ);             \
2923     uint32_t vta = vext_vta(desc);                        \
2924     uint32_t vma = vext_vma(desc);                        \
2925     uint32_t i;                                           \
2926                                                           \
2927     VSTART_CHECK_EARLY_EXIT(env);                         \
2928                                                           \
2929     for (i = env->vstart; i < vl; i++) {                  \
2930         if (!vm && !vext_elem_mask(v0, i)) {              \
2931             /* set masked-off elements to 1s */           \
2932             vext_set_elems_1s(vd, vma, i * ESZ,           \
2933                               (i + 1) * ESZ);             \
2934             continue;                                     \
2935         }                                                 \
2936         do_##NAME(vd, s1, vs2, i, env);                   \
2937     }                                                     \
2938     env->vstart = 0;                                      \
2939     /* set tail elements to 1s */                         \
2940     vext_set_elems_1s(vd, vta, vl * ESZ,                  \
2941                       total_elems * ESZ);                 \
2942 }
2943 
2944 RVVCALL(OPFVF2, vfadd_vf_h, OP_UUU_H, H2, H2, float16_add)
2945 RVVCALL(OPFVF2, vfadd_vf_w, OP_UUU_W, H4, H4, float32_add)
2946 RVVCALL(OPFVF2, vfadd_vf_d, OP_UUU_D, H8, H8, float64_add)
2947 GEN_VEXT_VF(vfadd_vf_h, 2)
2948 GEN_VEXT_VF(vfadd_vf_w, 4)
2949 GEN_VEXT_VF(vfadd_vf_d, 8)
2950 
2951 RVVCALL(OPFVV2, vfsub_vv_h, OP_UUU_H, H2, H2, H2, float16_sub)
2952 RVVCALL(OPFVV2, vfsub_vv_w, OP_UUU_W, H4, H4, H4, float32_sub)
2953 RVVCALL(OPFVV2, vfsub_vv_d, OP_UUU_D, H8, H8, H8, float64_sub)
2954 GEN_VEXT_VV_ENV(vfsub_vv_h, 2)
2955 GEN_VEXT_VV_ENV(vfsub_vv_w, 4)
2956 GEN_VEXT_VV_ENV(vfsub_vv_d, 8)
2957 RVVCALL(OPFVF2, vfsub_vf_h, OP_UUU_H, H2, H2, float16_sub)
2958 RVVCALL(OPFVF2, vfsub_vf_w, OP_UUU_W, H4, H4, float32_sub)
2959 RVVCALL(OPFVF2, vfsub_vf_d, OP_UUU_D, H8, H8, float64_sub)
2960 GEN_VEXT_VF(vfsub_vf_h, 2)
2961 GEN_VEXT_VF(vfsub_vf_w, 4)
2962 GEN_VEXT_VF(vfsub_vf_d, 8)
2963 
2964 static uint16_t float16_rsub(uint16_t a, uint16_t b, float_status *s)
2965 {
2966     return float16_sub(b, a, s);
2967 }
2968 
2969 static uint32_t float32_rsub(uint32_t a, uint32_t b, float_status *s)
2970 {
2971     return float32_sub(b, a, s);
2972 }
2973 
2974 static uint64_t float64_rsub(uint64_t a, uint64_t b, float_status *s)
2975 {
2976     return float64_sub(b, a, s);
2977 }
2978 
2979 RVVCALL(OPFVF2, vfrsub_vf_h, OP_UUU_H, H2, H2, float16_rsub)
2980 RVVCALL(OPFVF2, vfrsub_vf_w, OP_UUU_W, H4, H4, float32_rsub)
2981 RVVCALL(OPFVF2, vfrsub_vf_d, OP_UUU_D, H8, H8, float64_rsub)
2982 GEN_VEXT_VF(vfrsub_vf_h, 2)
2983 GEN_VEXT_VF(vfrsub_vf_w, 4)
2984 GEN_VEXT_VF(vfrsub_vf_d, 8)
2985 
2986 /* Vector Widening Floating-Point Add/Subtract Instructions */
2987 static uint32_t vfwadd16(uint16_t a, uint16_t b, float_status *s)
2988 {
2989     return float32_add(float16_to_float32(a, true, s),
2990                        float16_to_float32(b, true, s), s);
2991 }
2992 
2993 static uint64_t vfwadd32(uint32_t a, uint32_t b, float_status *s)
2994 {
2995     return float64_add(float32_to_float64(a, s),
2996                        float32_to_float64(b, s), s);
2997 
2998 }
2999 
3000 RVVCALL(OPFVV2, vfwadd_vv_h, WOP_UUU_H, H4, H2, H2, vfwadd16)
3001 RVVCALL(OPFVV2, vfwadd_vv_w, WOP_UUU_W, H8, H4, H4, vfwadd32)
3002 GEN_VEXT_VV_ENV(vfwadd_vv_h, 4)
3003 GEN_VEXT_VV_ENV(vfwadd_vv_w, 8)
3004 RVVCALL(OPFVF2, vfwadd_vf_h, WOP_UUU_H, H4, H2, vfwadd16)
3005 RVVCALL(OPFVF2, vfwadd_vf_w, WOP_UUU_W, H8, H4, vfwadd32)
3006 GEN_VEXT_VF(vfwadd_vf_h, 4)
3007 GEN_VEXT_VF(vfwadd_vf_w, 8)
3008 
3009 static uint32_t vfwsub16(uint16_t a, uint16_t b, float_status *s)
3010 {
3011     return float32_sub(float16_to_float32(a, true, s),
3012                        float16_to_float32(b, true, s), s);
3013 }
3014 
3015 static uint64_t vfwsub32(uint32_t a, uint32_t b, float_status *s)
3016 {
3017     return float64_sub(float32_to_float64(a, s),
3018                        float32_to_float64(b, s), s);
3019 
3020 }
3021 
3022 RVVCALL(OPFVV2, vfwsub_vv_h, WOP_UUU_H, H4, H2, H2, vfwsub16)
3023 RVVCALL(OPFVV2, vfwsub_vv_w, WOP_UUU_W, H8, H4, H4, vfwsub32)
3024 GEN_VEXT_VV_ENV(vfwsub_vv_h, 4)
3025 GEN_VEXT_VV_ENV(vfwsub_vv_w, 8)
3026 RVVCALL(OPFVF2, vfwsub_vf_h, WOP_UUU_H, H4, H2, vfwsub16)
3027 RVVCALL(OPFVF2, vfwsub_vf_w, WOP_UUU_W, H8, H4, vfwsub32)
3028 GEN_VEXT_VF(vfwsub_vf_h, 4)
3029 GEN_VEXT_VF(vfwsub_vf_w, 8)
3030 
3031 static uint32_t vfwaddw16(uint32_t a, uint16_t b, float_status *s)
3032 {
3033     return float32_add(a, float16_to_float32(b, true, s), s);
3034 }
3035 
3036 static uint64_t vfwaddw32(uint64_t a, uint32_t b, float_status *s)
3037 {
3038     return float64_add(a, float32_to_float64(b, s), s);
3039 }
3040 
3041 RVVCALL(OPFVV2, vfwadd_wv_h, WOP_WUUU_H, H4, H2, H2, vfwaddw16)
3042 RVVCALL(OPFVV2, vfwadd_wv_w, WOP_WUUU_W, H8, H4, H4, vfwaddw32)
3043 GEN_VEXT_VV_ENV(vfwadd_wv_h, 4)
3044 GEN_VEXT_VV_ENV(vfwadd_wv_w, 8)
3045 RVVCALL(OPFVF2, vfwadd_wf_h, WOP_WUUU_H, H4, H2, vfwaddw16)
3046 RVVCALL(OPFVF2, vfwadd_wf_w, WOP_WUUU_W, H8, H4, vfwaddw32)
3047 GEN_VEXT_VF(vfwadd_wf_h, 4)
3048 GEN_VEXT_VF(vfwadd_wf_w, 8)
3049 
3050 static uint32_t vfwsubw16(uint32_t a, uint16_t b, float_status *s)
3051 {
3052     return float32_sub(a, float16_to_float32(b, true, s), s);
3053 }
3054 
3055 static uint64_t vfwsubw32(uint64_t a, uint32_t b, float_status *s)
3056 {
3057     return float64_sub(a, float32_to_float64(b, s), s);
3058 }
3059 
3060 RVVCALL(OPFVV2, vfwsub_wv_h, WOP_WUUU_H, H4, H2, H2, vfwsubw16)
3061 RVVCALL(OPFVV2, vfwsub_wv_w, WOP_WUUU_W, H8, H4, H4, vfwsubw32)
3062 GEN_VEXT_VV_ENV(vfwsub_wv_h, 4)
3063 GEN_VEXT_VV_ENV(vfwsub_wv_w, 8)
3064 RVVCALL(OPFVF2, vfwsub_wf_h, WOP_WUUU_H, H4, H2, vfwsubw16)
3065 RVVCALL(OPFVF2, vfwsub_wf_w, WOP_WUUU_W, H8, H4, vfwsubw32)
3066 GEN_VEXT_VF(vfwsub_wf_h, 4)
3067 GEN_VEXT_VF(vfwsub_wf_w, 8)
3068 
3069 /* Vector Single-Width Floating-Point Multiply/Divide Instructions */
3070 RVVCALL(OPFVV2, vfmul_vv_h, OP_UUU_H, H2, H2, H2, float16_mul)
3071 RVVCALL(OPFVV2, vfmul_vv_w, OP_UUU_W, H4, H4, H4, float32_mul)
3072 RVVCALL(OPFVV2, vfmul_vv_d, OP_UUU_D, H8, H8, H8, float64_mul)
3073 GEN_VEXT_VV_ENV(vfmul_vv_h, 2)
3074 GEN_VEXT_VV_ENV(vfmul_vv_w, 4)
3075 GEN_VEXT_VV_ENV(vfmul_vv_d, 8)
3076 RVVCALL(OPFVF2, vfmul_vf_h, OP_UUU_H, H2, H2, float16_mul)
3077 RVVCALL(OPFVF2, vfmul_vf_w, OP_UUU_W, H4, H4, float32_mul)
3078 RVVCALL(OPFVF2, vfmul_vf_d, OP_UUU_D, H8, H8, float64_mul)
3079 GEN_VEXT_VF(vfmul_vf_h, 2)
3080 GEN_VEXT_VF(vfmul_vf_w, 4)
3081 GEN_VEXT_VF(vfmul_vf_d, 8)
3082 
3083 RVVCALL(OPFVV2, vfdiv_vv_h, OP_UUU_H, H2, H2, H2, float16_div)
3084 RVVCALL(OPFVV2, vfdiv_vv_w, OP_UUU_W, H4, H4, H4, float32_div)
3085 RVVCALL(OPFVV2, vfdiv_vv_d, OP_UUU_D, H8, H8, H8, float64_div)
3086 GEN_VEXT_VV_ENV(vfdiv_vv_h, 2)
3087 GEN_VEXT_VV_ENV(vfdiv_vv_w, 4)
3088 GEN_VEXT_VV_ENV(vfdiv_vv_d, 8)
3089 RVVCALL(OPFVF2, vfdiv_vf_h, OP_UUU_H, H2, H2, float16_div)
3090 RVVCALL(OPFVF2, vfdiv_vf_w, OP_UUU_W, H4, H4, float32_div)
3091 RVVCALL(OPFVF2, vfdiv_vf_d, OP_UUU_D, H8, H8, float64_div)
3092 GEN_VEXT_VF(vfdiv_vf_h, 2)
3093 GEN_VEXT_VF(vfdiv_vf_w, 4)
3094 GEN_VEXT_VF(vfdiv_vf_d, 8)
3095 
3096 static uint16_t float16_rdiv(uint16_t a, uint16_t b, float_status *s)
3097 {
3098     return float16_div(b, a, s);
3099 }
3100 
3101 static uint32_t float32_rdiv(uint32_t a, uint32_t b, float_status *s)
3102 {
3103     return float32_div(b, a, s);
3104 }
3105 
3106 static uint64_t float64_rdiv(uint64_t a, uint64_t b, float_status *s)
3107 {
3108     return float64_div(b, a, s);
3109 }
3110 
3111 RVVCALL(OPFVF2, vfrdiv_vf_h, OP_UUU_H, H2, H2, float16_rdiv)
3112 RVVCALL(OPFVF2, vfrdiv_vf_w, OP_UUU_W, H4, H4, float32_rdiv)
3113 RVVCALL(OPFVF2, vfrdiv_vf_d, OP_UUU_D, H8, H8, float64_rdiv)
3114 GEN_VEXT_VF(vfrdiv_vf_h, 2)
3115 GEN_VEXT_VF(vfrdiv_vf_w, 4)
3116 GEN_VEXT_VF(vfrdiv_vf_d, 8)
3117 
3118 /* Vector Widening Floating-Point Multiply */
3119 static uint32_t vfwmul16(uint16_t a, uint16_t b, float_status *s)
3120 {
3121     return float32_mul(float16_to_float32(a, true, s),
3122                        float16_to_float32(b, true, s), s);
3123 }
3124 
3125 static uint64_t vfwmul32(uint32_t a, uint32_t b, float_status *s)
3126 {
3127     return float64_mul(float32_to_float64(a, s),
3128                        float32_to_float64(b, s), s);
3129 
3130 }
3131 RVVCALL(OPFVV2, vfwmul_vv_h, WOP_UUU_H, H4, H2, H2, vfwmul16)
3132 RVVCALL(OPFVV2, vfwmul_vv_w, WOP_UUU_W, H8, H4, H4, vfwmul32)
3133 GEN_VEXT_VV_ENV(vfwmul_vv_h, 4)
3134 GEN_VEXT_VV_ENV(vfwmul_vv_w, 8)
3135 RVVCALL(OPFVF2, vfwmul_vf_h, WOP_UUU_H, H4, H2, vfwmul16)
3136 RVVCALL(OPFVF2, vfwmul_vf_w, WOP_UUU_W, H8, H4, vfwmul32)
3137 GEN_VEXT_VF(vfwmul_vf_h, 4)
3138 GEN_VEXT_VF(vfwmul_vf_w, 8)
3139 
3140 /* Vector Single-Width Floating-Point Fused Multiply-Add Instructions */
3141 #define OPFVV3(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)       \
3142 static void do_##NAME(void *vd, void *vs1, void *vs2, int i,       \
3143                       CPURISCVState *env)                          \
3144 {                                                                  \
3145     TX1 s1 = *((T1 *)vs1 + HS1(i));                                \
3146     TX2 s2 = *((T2 *)vs2 + HS2(i));                                \
3147     TD d = *((TD *)vd + HD(i));                                    \
3148     *((TD *)vd + HD(i)) = OP(s2, s1, d, &env->fp_status);          \
3149 }
3150 
3151 static uint16_t fmacc16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3152 {
3153     return float16_muladd(a, b, d, 0, s);
3154 }
3155 
3156 static uint32_t fmacc32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3157 {
3158     return float32_muladd(a, b, d, 0, s);
3159 }
3160 
3161 static uint64_t fmacc64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3162 {
3163     return float64_muladd(a, b, d, 0, s);
3164 }
3165 
3166 RVVCALL(OPFVV3, vfmacc_vv_h, OP_UUU_H, H2, H2, H2, fmacc16)
3167 RVVCALL(OPFVV3, vfmacc_vv_w, OP_UUU_W, H4, H4, H4, fmacc32)
3168 RVVCALL(OPFVV3, vfmacc_vv_d, OP_UUU_D, H8, H8, H8, fmacc64)
3169 GEN_VEXT_VV_ENV(vfmacc_vv_h, 2)
3170 GEN_VEXT_VV_ENV(vfmacc_vv_w, 4)
3171 GEN_VEXT_VV_ENV(vfmacc_vv_d, 8)
3172 
3173 #define OPFVF3(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)           \
3174 static void do_##NAME(void *vd, uint64_t s1, void *vs2, int i,    \
3175                       CPURISCVState *env)                         \
3176 {                                                                 \
3177     TX2 s2 = *((T2 *)vs2 + HS2(i));                               \
3178     TD d = *((TD *)vd + HD(i));                                   \
3179     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1, d, &env->fp_status);\
3180 }
3181 
3182 RVVCALL(OPFVF3, vfmacc_vf_h, OP_UUU_H, H2, H2, fmacc16)
3183 RVVCALL(OPFVF3, vfmacc_vf_w, OP_UUU_W, H4, H4, fmacc32)
3184 RVVCALL(OPFVF3, vfmacc_vf_d, OP_UUU_D, H8, H8, fmacc64)
3185 GEN_VEXT_VF(vfmacc_vf_h, 2)
3186 GEN_VEXT_VF(vfmacc_vf_w, 4)
3187 GEN_VEXT_VF(vfmacc_vf_d, 8)
3188 
3189 static uint16_t fnmacc16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3190 {
3191     return float16_muladd(a, b, d, float_muladd_negate_c |
3192                                    float_muladd_negate_product, s);
3193 }
3194 
3195 static uint32_t fnmacc32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3196 {
3197     return float32_muladd(a, b, d, float_muladd_negate_c |
3198                                    float_muladd_negate_product, s);
3199 }
3200 
3201 static uint64_t fnmacc64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3202 {
3203     return float64_muladd(a, b, d, float_muladd_negate_c |
3204                                    float_muladd_negate_product, s);
3205 }
3206 
3207 RVVCALL(OPFVV3, vfnmacc_vv_h, OP_UUU_H, H2, H2, H2, fnmacc16)
3208 RVVCALL(OPFVV3, vfnmacc_vv_w, OP_UUU_W, H4, H4, H4, fnmacc32)
3209 RVVCALL(OPFVV3, vfnmacc_vv_d, OP_UUU_D, H8, H8, H8, fnmacc64)
3210 GEN_VEXT_VV_ENV(vfnmacc_vv_h, 2)
3211 GEN_VEXT_VV_ENV(vfnmacc_vv_w, 4)
3212 GEN_VEXT_VV_ENV(vfnmacc_vv_d, 8)
3213 RVVCALL(OPFVF3, vfnmacc_vf_h, OP_UUU_H, H2, H2, fnmacc16)
3214 RVVCALL(OPFVF3, vfnmacc_vf_w, OP_UUU_W, H4, H4, fnmacc32)
3215 RVVCALL(OPFVF3, vfnmacc_vf_d, OP_UUU_D, H8, H8, fnmacc64)
3216 GEN_VEXT_VF(vfnmacc_vf_h, 2)
3217 GEN_VEXT_VF(vfnmacc_vf_w, 4)
3218 GEN_VEXT_VF(vfnmacc_vf_d, 8)
3219 
3220 static uint16_t fmsac16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3221 {
3222     return float16_muladd(a, b, d, float_muladd_negate_c, s);
3223 }
3224 
3225 static uint32_t fmsac32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3226 {
3227     return float32_muladd(a, b, d, float_muladd_negate_c, s);
3228 }
3229 
3230 static uint64_t fmsac64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3231 {
3232     return float64_muladd(a, b, d, float_muladd_negate_c, s);
3233 }
3234 
3235 RVVCALL(OPFVV3, vfmsac_vv_h, OP_UUU_H, H2, H2, H2, fmsac16)
3236 RVVCALL(OPFVV3, vfmsac_vv_w, OP_UUU_W, H4, H4, H4, fmsac32)
3237 RVVCALL(OPFVV3, vfmsac_vv_d, OP_UUU_D, H8, H8, H8, fmsac64)
3238 GEN_VEXT_VV_ENV(vfmsac_vv_h, 2)
3239 GEN_VEXT_VV_ENV(vfmsac_vv_w, 4)
3240 GEN_VEXT_VV_ENV(vfmsac_vv_d, 8)
3241 RVVCALL(OPFVF3, vfmsac_vf_h, OP_UUU_H, H2, H2, fmsac16)
3242 RVVCALL(OPFVF3, vfmsac_vf_w, OP_UUU_W, H4, H4, fmsac32)
3243 RVVCALL(OPFVF3, vfmsac_vf_d, OP_UUU_D, H8, H8, fmsac64)
3244 GEN_VEXT_VF(vfmsac_vf_h, 2)
3245 GEN_VEXT_VF(vfmsac_vf_w, 4)
3246 GEN_VEXT_VF(vfmsac_vf_d, 8)
3247 
3248 static uint16_t fnmsac16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3249 {
3250     return float16_muladd(a, b, d, float_muladd_negate_product, s);
3251 }
3252 
3253 static uint32_t fnmsac32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3254 {
3255     return float32_muladd(a, b, d, float_muladd_negate_product, s);
3256 }
3257 
3258 static uint64_t fnmsac64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3259 {
3260     return float64_muladd(a, b, d, float_muladd_negate_product, s);
3261 }
3262 
3263 RVVCALL(OPFVV3, vfnmsac_vv_h, OP_UUU_H, H2, H2, H2, fnmsac16)
3264 RVVCALL(OPFVV3, vfnmsac_vv_w, OP_UUU_W, H4, H4, H4, fnmsac32)
3265 RVVCALL(OPFVV3, vfnmsac_vv_d, OP_UUU_D, H8, H8, H8, fnmsac64)
3266 GEN_VEXT_VV_ENV(vfnmsac_vv_h, 2)
3267 GEN_VEXT_VV_ENV(vfnmsac_vv_w, 4)
3268 GEN_VEXT_VV_ENV(vfnmsac_vv_d, 8)
3269 RVVCALL(OPFVF3, vfnmsac_vf_h, OP_UUU_H, H2, H2, fnmsac16)
3270 RVVCALL(OPFVF3, vfnmsac_vf_w, OP_UUU_W, H4, H4, fnmsac32)
3271 RVVCALL(OPFVF3, vfnmsac_vf_d, OP_UUU_D, H8, H8, fnmsac64)
3272 GEN_VEXT_VF(vfnmsac_vf_h, 2)
3273 GEN_VEXT_VF(vfnmsac_vf_w, 4)
3274 GEN_VEXT_VF(vfnmsac_vf_d, 8)
3275 
3276 static uint16_t fmadd16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3277 {
3278     return float16_muladd(d, b, a, 0, s);
3279 }
3280 
3281 static uint32_t fmadd32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3282 {
3283     return float32_muladd(d, b, a, 0, s);
3284 }
3285 
3286 static uint64_t fmadd64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3287 {
3288     return float64_muladd(d, b, a, 0, s);
3289 }
3290 
3291 RVVCALL(OPFVV3, vfmadd_vv_h, OP_UUU_H, H2, H2, H2, fmadd16)
3292 RVVCALL(OPFVV3, vfmadd_vv_w, OP_UUU_W, H4, H4, H4, fmadd32)
3293 RVVCALL(OPFVV3, vfmadd_vv_d, OP_UUU_D, H8, H8, H8, fmadd64)
3294 GEN_VEXT_VV_ENV(vfmadd_vv_h, 2)
3295 GEN_VEXT_VV_ENV(vfmadd_vv_w, 4)
3296 GEN_VEXT_VV_ENV(vfmadd_vv_d, 8)
3297 RVVCALL(OPFVF3, vfmadd_vf_h, OP_UUU_H, H2, H2, fmadd16)
3298 RVVCALL(OPFVF3, vfmadd_vf_w, OP_UUU_W, H4, H4, fmadd32)
3299 RVVCALL(OPFVF3, vfmadd_vf_d, OP_UUU_D, H8, H8, fmadd64)
3300 GEN_VEXT_VF(vfmadd_vf_h, 2)
3301 GEN_VEXT_VF(vfmadd_vf_w, 4)
3302 GEN_VEXT_VF(vfmadd_vf_d, 8)
3303 
3304 static uint16_t fnmadd16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3305 {
3306     return float16_muladd(d, b, a, float_muladd_negate_c |
3307                                    float_muladd_negate_product, s);
3308 }
3309 
3310 static uint32_t fnmadd32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3311 {
3312     return float32_muladd(d, b, a, float_muladd_negate_c |
3313                                    float_muladd_negate_product, s);
3314 }
3315 
3316 static uint64_t fnmadd64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3317 {
3318     return float64_muladd(d, b, a, float_muladd_negate_c |
3319                                    float_muladd_negate_product, s);
3320 }
3321 
3322 RVVCALL(OPFVV3, vfnmadd_vv_h, OP_UUU_H, H2, H2, H2, fnmadd16)
3323 RVVCALL(OPFVV3, vfnmadd_vv_w, OP_UUU_W, H4, H4, H4, fnmadd32)
3324 RVVCALL(OPFVV3, vfnmadd_vv_d, OP_UUU_D, H8, H8, H8, fnmadd64)
3325 GEN_VEXT_VV_ENV(vfnmadd_vv_h, 2)
3326 GEN_VEXT_VV_ENV(vfnmadd_vv_w, 4)
3327 GEN_VEXT_VV_ENV(vfnmadd_vv_d, 8)
3328 RVVCALL(OPFVF3, vfnmadd_vf_h, OP_UUU_H, H2, H2, fnmadd16)
3329 RVVCALL(OPFVF3, vfnmadd_vf_w, OP_UUU_W, H4, H4, fnmadd32)
3330 RVVCALL(OPFVF3, vfnmadd_vf_d, OP_UUU_D, H8, H8, fnmadd64)
3331 GEN_VEXT_VF(vfnmadd_vf_h, 2)
3332 GEN_VEXT_VF(vfnmadd_vf_w, 4)
3333 GEN_VEXT_VF(vfnmadd_vf_d, 8)
3334 
3335 static uint16_t fmsub16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3336 {
3337     return float16_muladd(d, b, a, float_muladd_negate_c, s);
3338 }
3339 
3340 static uint32_t fmsub32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3341 {
3342     return float32_muladd(d, b, a, float_muladd_negate_c, s);
3343 }
3344 
3345 static uint64_t fmsub64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3346 {
3347     return float64_muladd(d, b, a, float_muladd_negate_c, s);
3348 }
3349 
3350 RVVCALL(OPFVV3, vfmsub_vv_h, OP_UUU_H, H2, H2, H2, fmsub16)
3351 RVVCALL(OPFVV3, vfmsub_vv_w, OP_UUU_W, H4, H4, H4, fmsub32)
3352 RVVCALL(OPFVV3, vfmsub_vv_d, OP_UUU_D, H8, H8, H8, fmsub64)
3353 GEN_VEXT_VV_ENV(vfmsub_vv_h, 2)
3354 GEN_VEXT_VV_ENV(vfmsub_vv_w, 4)
3355 GEN_VEXT_VV_ENV(vfmsub_vv_d, 8)
3356 RVVCALL(OPFVF3, vfmsub_vf_h, OP_UUU_H, H2, H2, fmsub16)
3357 RVVCALL(OPFVF3, vfmsub_vf_w, OP_UUU_W, H4, H4, fmsub32)
3358 RVVCALL(OPFVF3, vfmsub_vf_d, OP_UUU_D, H8, H8, fmsub64)
3359 GEN_VEXT_VF(vfmsub_vf_h, 2)
3360 GEN_VEXT_VF(vfmsub_vf_w, 4)
3361 GEN_VEXT_VF(vfmsub_vf_d, 8)
3362 
3363 static uint16_t fnmsub16(uint16_t a, uint16_t b, uint16_t d, float_status *s)
3364 {
3365     return float16_muladd(d, b, a, float_muladd_negate_product, s);
3366 }
3367 
3368 static uint32_t fnmsub32(uint32_t a, uint32_t b, uint32_t d, float_status *s)
3369 {
3370     return float32_muladd(d, b, a, float_muladd_negate_product, s);
3371 }
3372 
3373 static uint64_t fnmsub64(uint64_t a, uint64_t b, uint64_t d, float_status *s)
3374 {
3375     return float64_muladd(d, b, a, float_muladd_negate_product, s);
3376 }
3377 
3378 RVVCALL(OPFVV3, vfnmsub_vv_h, OP_UUU_H, H2, H2, H2, fnmsub16)
3379 RVVCALL(OPFVV3, vfnmsub_vv_w, OP_UUU_W, H4, H4, H4, fnmsub32)
3380 RVVCALL(OPFVV3, vfnmsub_vv_d, OP_UUU_D, H8, H8, H8, fnmsub64)
3381 GEN_VEXT_VV_ENV(vfnmsub_vv_h, 2)
3382 GEN_VEXT_VV_ENV(vfnmsub_vv_w, 4)
3383 GEN_VEXT_VV_ENV(vfnmsub_vv_d, 8)
3384 RVVCALL(OPFVF3, vfnmsub_vf_h, OP_UUU_H, H2, H2, fnmsub16)
3385 RVVCALL(OPFVF3, vfnmsub_vf_w, OP_UUU_W, H4, H4, fnmsub32)
3386 RVVCALL(OPFVF3, vfnmsub_vf_d, OP_UUU_D, H8, H8, fnmsub64)
3387 GEN_VEXT_VF(vfnmsub_vf_h, 2)
3388 GEN_VEXT_VF(vfnmsub_vf_w, 4)
3389 GEN_VEXT_VF(vfnmsub_vf_d, 8)
3390 
3391 /* Vector Widening Floating-Point Fused Multiply-Add Instructions */
3392 static uint32_t fwmacc16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3393 {
3394     return float32_muladd(float16_to_float32(a, true, s),
3395                           float16_to_float32(b, true, s), d, 0, s);
3396 }
3397 
3398 static uint64_t fwmacc32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3399 {
3400     return float64_muladd(float32_to_float64(a, s),
3401                           float32_to_float64(b, s), d, 0, s);
3402 }
3403 
3404 RVVCALL(OPFVV3, vfwmacc_vv_h, WOP_UUU_H, H4, H2, H2, fwmacc16)
3405 RVVCALL(OPFVV3, vfwmacc_vv_w, WOP_UUU_W, H8, H4, H4, fwmacc32)
3406 GEN_VEXT_VV_ENV(vfwmacc_vv_h, 4)
3407 GEN_VEXT_VV_ENV(vfwmacc_vv_w, 8)
3408 RVVCALL(OPFVF3, vfwmacc_vf_h, WOP_UUU_H, H4, H2, fwmacc16)
3409 RVVCALL(OPFVF3, vfwmacc_vf_w, WOP_UUU_W, H8, H4, fwmacc32)
3410 GEN_VEXT_VF(vfwmacc_vf_h, 4)
3411 GEN_VEXT_VF(vfwmacc_vf_w, 8)
3412 
3413 static uint32_t fwmaccbf16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3414 {
3415     return float32_muladd(bfloat16_to_float32(a, s),
3416                           bfloat16_to_float32(b, s), d, 0, s);
3417 }
3418 
3419 RVVCALL(OPFVV3, vfwmaccbf16_vv, WOP_UUU_H, H4, H2, H2, fwmaccbf16)
3420 GEN_VEXT_VV_ENV(vfwmaccbf16_vv, 4)
3421 RVVCALL(OPFVF3, vfwmaccbf16_vf, WOP_UUU_H, H4, H2, fwmaccbf16)
3422 GEN_VEXT_VF(vfwmaccbf16_vf, 4)
3423 
3424 static uint32_t fwnmacc16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3425 {
3426     return float32_muladd(float16_to_float32(a, true, s),
3427                           float16_to_float32(b, true, s), d,
3428                           float_muladd_negate_c | float_muladd_negate_product,
3429                           s);
3430 }
3431 
3432 static uint64_t fwnmacc32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3433 {
3434     return float64_muladd(float32_to_float64(a, s), float32_to_float64(b, s),
3435                           d, float_muladd_negate_c |
3436                              float_muladd_negate_product, s);
3437 }
3438 
3439 RVVCALL(OPFVV3, vfwnmacc_vv_h, WOP_UUU_H, H4, H2, H2, fwnmacc16)
3440 RVVCALL(OPFVV3, vfwnmacc_vv_w, WOP_UUU_W, H8, H4, H4, fwnmacc32)
3441 GEN_VEXT_VV_ENV(vfwnmacc_vv_h, 4)
3442 GEN_VEXT_VV_ENV(vfwnmacc_vv_w, 8)
3443 RVVCALL(OPFVF3, vfwnmacc_vf_h, WOP_UUU_H, H4, H2, fwnmacc16)
3444 RVVCALL(OPFVF3, vfwnmacc_vf_w, WOP_UUU_W, H8, H4, fwnmacc32)
3445 GEN_VEXT_VF(vfwnmacc_vf_h, 4)
3446 GEN_VEXT_VF(vfwnmacc_vf_w, 8)
3447 
3448 static uint32_t fwmsac16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3449 {
3450     return float32_muladd(float16_to_float32(a, true, s),
3451                           float16_to_float32(b, true, s), d,
3452                           float_muladd_negate_c, s);
3453 }
3454 
3455 static uint64_t fwmsac32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3456 {
3457     return float64_muladd(float32_to_float64(a, s),
3458                           float32_to_float64(b, s), d,
3459                           float_muladd_negate_c, s);
3460 }
3461 
3462 RVVCALL(OPFVV3, vfwmsac_vv_h, WOP_UUU_H, H4, H2, H2, fwmsac16)
3463 RVVCALL(OPFVV3, vfwmsac_vv_w, WOP_UUU_W, H8, H4, H4, fwmsac32)
3464 GEN_VEXT_VV_ENV(vfwmsac_vv_h, 4)
3465 GEN_VEXT_VV_ENV(vfwmsac_vv_w, 8)
3466 RVVCALL(OPFVF3, vfwmsac_vf_h, WOP_UUU_H, H4, H2, fwmsac16)
3467 RVVCALL(OPFVF3, vfwmsac_vf_w, WOP_UUU_W, H8, H4, fwmsac32)
3468 GEN_VEXT_VF(vfwmsac_vf_h, 4)
3469 GEN_VEXT_VF(vfwmsac_vf_w, 8)
3470 
3471 static uint32_t fwnmsac16(uint16_t a, uint16_t b, uint32_t d, float_status *s)
3472 {
3473     return float32_muladd(float16_to_float32(a, true, s),
3474                           float16_to_float32(b, true, s), d,
3475                           float_muladd_negate_product, s);
3476 }
3477 
3478 static uint64_t fwnmsac32(uint32_t a, uint32_t b, uint64_t d, float_status *s)
3479 {
3480     return float64_muladd(float32_to_float64(a, s),
3481                           float32_to_float64(b, s), d,
3482                           float_muladd_negate_product, s);
3483 }
3484 
3485 RVVCALL(OPFVV3, vfwnmsac_vv_h, WOP_UUU_H, H4, H2, H2, fwnmsac16)
3486 RVVCALL(OPFVV3, vfwnmsac_vv_w, WOP_UUU_W, H8, H4, H4, fwnmsac32)
3487 GEN_VEXT_VV_ENV(vfwnmsac_vv_h, 4)
3488 GEN_VEXT_VV_ENV(vfwnmsac_vv_w, 8)
3489 RVVCALL(OPFVF3, vfwnmsac_vf_h, WOP_UUU_H, H4, H2, fwnmsac16)
3490 RVVCALL(OPFVF3, vfwnmsac_vf_w, WOP_UUU_W, H8, H4, fwnmsac32)
3491 GEN_VEXT_VF(vfwnmsac_vf_h, 4)
3492 GEN_VEXT_VF(vfwnmsac_vf_w, 8)
3493 
3494 /* Vector Floating-Point Square-Root Instruction */
3495 #define OPFVV1(NAME, TD, T2, TX2, HD, HS2, OP)         \
3496 static void do_##NAME(void *vd, void *vs2, int i,      \
3497                       CPURISCVState *env)              \
3498 {                                                      \
3499     TX2 s2 = *((T2 *)vs2 + HS2(i));                    \
3500     *((TD *)vd + HD(i)) = OP(s2, &env->fp_status);     \
3501 }
3502 
3503 #define GEN_VEXT_V_ENV(NAME, ESZ)                      \
3504 void HELPER(NAME)(void *vd, void *v0, void *vs2,       \
3505                   CPURISCVState *env, uint32_t desc)   \
3506 {                                                      \
3507     uint32_t vm = vext_vm(desc);                       \
3508     uint32_t vl = env->vl;                             \
3509     uint32_t total_elems =                             \
3510         vext_get_total_elems(env, desc, ESZ);          \
3511     uint32_t vta = vext_vta(desc);                     \
3512     uint32_t vma = vext_vma(desc);                     \
3513     uint32_t i;                                        \
3514                                                        \
3515     VSTART_CHECK_EARLY_EXIT(env);                      \
3516                                                        \
3517     if (vl == 0) {                                     \
3518         return;                                        \
3519     }                                                  \
3520     for (i = env->vstart; i < vl; i++) {               \
3521         if (!vm && !vext_elem_mask(v0, i)) {           \
3522             /* set masked-off elements to 1s */        \
3523             vext_set_elems_1s(vd, vma, i * ESZ,        \
3524                               (i + 1) * ESZ);          \
3525             continue;                                  \
3526         }                                              \
3527         do_##NAME(vd, vs2, i, env);                    \
3528     }                                                  \
3529     env->vstart = 0;                                   \
3530     vext_set_elems_1s(vd, vta, vl * ESZ,               \
3531                       total_elems * ESZ);              \
3532 }
3533 
3534 RVVCALL(OPFVV1, vfsqrt_v_h, OP_UU_H, H2, H2, float16_sqrt)
3535 RVVCALL(OPFVV1, vfsqrt_v_w, OP_UU_W, H4, H4, float32_sqrt)
3536 RVVCALL(OPFVV1, vfsqrt_v_d, OP_UU_D, H8, H8, float64_sqrt)
3537 GEN_VEXT_V_ENV(vfsqrt_v_h, 2)
3538 GEN_VEXT_V_ENV(vfsqrt_v_w, 4)
3539 GEN_VEXT_V_ENV(vfsqrt_v_d, 8)
3540 
3541 /*
3542  * Vector Floating-Point Reciprocal Square-Root Estimate Instruction
3543  *
3544  * Adapted from riscv-v-spec recip.c:
3545  * https://github.com/riscv/riscv-v-spec/blob/master/recip.c
3546  */
3547 static uint64_t frsqrt7(uint64_t f, int exp_size, int frac_size)
3548 {
3549     uint64_t sign = extract64(f, frac_size + exp_size, 1);
3550     uint64_t exp = extract64(f, frac_size, exp_size);
3551     uint64_t frac = extract64(f, 0, frac_size);
3552 
3553     const uint8_t lookup_table[] = {
3554         52, 51, 50, 48, 47, 46, 44, 43,
3555         42, 41, 40, 39, 38, 36, 35, 34,
3556         33, 32, 31, 30, 30, 29, 28, 27,
3557         26, 25, 24, 23, 23, 22, 21, 20,
3558         19, 19, 18, 17, 16, 16, 15, 14,
3559         14, 13, 12, 12, 11, 10, 10, 9,
3560         9, 8, 7, 7, 6, 6, 5, 4,
3561         4, 3, 3, 2, 2, 1, 1, 0,
3562         127, 125, 123, 121, 119, 118, 116, 114,
3563         113, 111, 109, 108, 106, 105, 103, 102,
3564         100, 99, 97, 96, 95, 93, 92, 91,
3565         90, 88, 87, 86, 85, 84, 83, 82,
3566         80, 79, 78, 77, 76, 75, 74, 73,
3567         72, 71, 70, 70, 69, 68, 67, 66,
3568         65, 64, 63, 63, 62, 61, 60, 59,
3569         59, 58, 57, 56, 56, 55, 54, 53
3570     };
3571     const int precision = 7;
3572 
3573     if (exp == 0 && frac != 0) { /* subnormal */
3574         /* Normalize the subnormal. */
3575         while (extract64(frac, frac_size - 1, 1) == 0) {
3576             exp--;
3577             frac <<= 1;
3578         }
3579 
3580         frac = (frac << 1) & MAKE_64BIT_MASK(0, frac_size);
3581     }
3582 
3583     int idx = ((exp & 1) << (precision - 1)) |
3584               (frac >> (frac_size - precision + 1));
3585     uint64_t out_frac = (uint64_t)(lookup_table[idx]) <<
3586                         (frac_size - precision);
3587     uint64_t out_exp = (3 * MAKE_64BIT_MASK(0, exp_size - 1) + ~exp) / 2;
3588 
3589     uint64_t val = 0;
3590     val = deposit64(val, 0, frac_size, out_frac);
3591     val = deposit64(val, frac_size, exp_size, out_exp);
3592     val = deposit64(val, frac_size + exp_size, 1, sign);
3593     return val;
3594 }
3595 
3596 static float16 frsqrt7_h(float16 f, float_status *s)
3597 {
3598     int exp_size = 5, frac_size = 10;
3599     bool sign = float16_is_neg(f);
3600 
3601     /*
3602      * frsqrt7(sNaN) = canonical NaN
3603      * frsqrt7(-inf) = canonical NaN
3604      * frsqrt7(-normal) = canonical NaN
3605      * frsqrt7(-subnormal) = canonical NaN
3606      */
3607     if (float16_is_signaling_nan(f, s) ||
3608         (float16_is_infinity(f) && sign) ||
3609         (float16_is_normal(f) && sign) ||
3610         (float16_is_zero_or_denormal(f) && !float16_is_zero(f) && sign)) {
3611         s->float_exception_flags |= float_flag_invalid;
3612         return float16_default_nan(s);
3613     }
3614 
3615     /* frsqrt7(qNaN) = canonical NaN */
3616     if (float16_is_quiet_nan(f, s)) {
3617         return float16_default_nan(s);
3618     }
3619 
3620     /* frsqrt7(+-0) = +-inf */
3621     if (float16_is_zero(f)) {
3622         s->float_exception_flags |= float_flag_divbyzero;
3623         return float16_set_sign(float16_infinity, sign);
3624     }
3625 
3626     /* frsqrt7(+inf) = +0 */
3627     if (float16_is_infinity(f) && !sign) {
3628         return float16_set_sign(float16_zero, sign);
3629     }
3630 
3631     /* +normal, +subnormal */
3632     uint64_t val = frsqrt7(f, exp_size, frac_size);
3633     return make_float16(val);
3634 }
3635 
3636 static float32 frsqrt7_s(float32 f, float_status *s)
3637 {
3638     int exp_size = 8, frac_size = 23;
3639     bool sign = float32_is_neg(f);
3640 
3641     /*
3642      * frsqrt7(sNaN) = canonical NaN
3643      * frsqrt7(-inf) = canonical NaN
3644      * frsqrt7(-normal) = canonical NaN
3645      * frsqrt7(-subnormal) = canonical NaN
3646      */
3647     if (float32_is_signaling_nan(f, s) ||
3648         (float32_is_infinity(f) && sign) ||
3649         (float32_is_normal(f) && sign) ||
3650         (float32_is_zero_or_denormal(f) && !float32_is_zero(f) && sign)) {
3651         s->float_exception_flags |= float_flag_invalid;
3652         return float32_default_nan(s);
3653     }
3654 
3655     /* frsqrt7(qNaN) = canonical NaN */
3656     if (float32_is_quiet_nan(f, s)) {
3657         return float32_default_nan(s);
3658     }
3659 
3660     /* frsqrt7(+-0) = +-inf */
3661     if (float32_is_zero(f)) {
3662         s->float_exception_flags |= float_flag_divbyzero;
3663         return float32_set_sign(float32_infinity, sign);
3664     }
3665 
3666     /* frsqrt7(+inf) = +0 */
3667     if (float32_is_infinity(f) && !sign) {
3668         return float32_set_sign(float32_zero, sign);
3669     }
3670 
3671     /* +normal, +subnormal */
3672     uint64_t val = frsqrt7(f, exp_size, frac_size);
3673     return make_float32(val);
3674 }
3675 
3676 static float64 frsqrt7_d(float64 f, float_status *s)
3677 {
3678     int exp_size = 11, frac_size = 52;
3679     bool sign = float64_is_neg(f);
3680 
3681     /*
3682      * frsqrt7(sNaN) = canonical NaN
3683      * frsqrt7(-inf) = canonical NaN
3684      * frsqrt7(-normal) = canonical NaN
3685      * frsqrt7(-subnormal) = canonical NaN
3686      */
3687     if (float64_is_signaling_nan(f, s) ||
3688         (float64_is_infinity(f) && sign) ||
3689         (float64_is_normal(f) && sign) ||
3690         (float64_is_zero_or_denormal(f) && !float64_is_zero(f) && sign)) {
3691         s->float_exception_flags |= float_flag_invalid;
3692         return float64_default_nan(s);
3693     }
3694 
3695     /* frsqrt7(qNaN) = canonical NaN */
3696     if (float64_is_quiet_nan(f, s)) {
3697         return float64_default_nan(s);
3698     }
3699 
3700     /* frsqrt7(+-0) = +-inf */
3701     if (float64_is_zero(f)) {
3702         s->float_exception_flags |= float_flag_divbyzero;
3703         return float64_set_sign(float64_infinity, sign);
3704     }
3705 
3706     /* frsqrt7(+inf) = +0 */
3707     if (float64_is_infinity(f) && !sign) {
3708         return float64_set_sign(float64_zero, sign);
3709     }
3710 
3711     /* +normal, +subnormal */
3712     uint64_t val = frsqrt7(f, exp_size, frac_size);
3713     return make_float64(val);
3714 }
3715 
3716 RVVCALL(OPFVV1, vfrsqrt7_v_h, OP_UU_H, H2, H2, frsqrt7_h)
3717 RVVCALL(OPFVV1, vfrsqrt7_v_w, OP_UU_W, H4, H4, frsqrt7_s)
3718 RVVCALL(OPFVV1, vfrsqrt7_v_d, OP_UU_D, H8, H8, frsqrt7_d)
3719 GEN_VEXT_V_ENV(vfrsqrt7_v_h, 2)
3720 GEN_VEXT_V_ENV(vfrsqrt7_v_w, 4)
3721 GEN_VEXT_V_ENV(vfrsqrt7_v_d, 8)
3722 
3723 /*
3724  * Vector Floating-Point Reciprocal Estimate Instruction
3725  *
3726  * Adapted from riscv-v-spec recip.c:
3727  * https://github.com/riscv/riscv-v-spec/blob/master/recip.c
3728  */
3729 static uint64_t frec7(uint64_t f, int exp_size, int frac_size,
3730                       float_status *s)
3731 {
3732     uint64_t sign = extract64(f, frac_size + exp_size, 1);
3733     uint64_t exp = extract64(f, frac_size, exp_size);
3734     uint64_t frac = extract64(f, 0, frac_size);
3735 
3736     const uint8_t lookup_table[] = {
3737         127, 125, 123, 121, 119, 117, 116, 114,
3738         112, 110, 109, 107, 105, 104, 102, 100,
3739         99, 97, 96, 94, 93, 91, 90, 88,
3740         87, 85, 84, 83, 81, 80, 79, 77,
3741         76, 75, 74, 72, 71, 70, 69, 68,
3742         66, 65, 64, 63, 62, 61, 60, 59,
3743         58, 57, 56, 55, 54, 53, 52, 51,
3744         50, 49, 48, 47, 46, 45, 44, 43,
3745         42, 41, 40, 40, 39, 38, 37, 36,
3746         35, 35, 34, 33, 32, 31, 31, 30,
3747         29, 28, 28, 27, 26, 25, 25, 24,
3748         23, 23, 22, 21, 21, 20, 19, 19,
3749         18, 17, 17, 16, 15, 15, 14, 14,
3750         13, 12, 12, 11, 11, 10, 9, 9,
3751         8, 8, 7, 7, 6, 5, 5, 4,
3752         4, 3, 3, 2, 2, 1, 1, 0
3753     };
3754     const int precision = 7;
3755 
3756     if (exp == 0 && frac != 0) { /* subnormal */
3757         /* Normalize the subnormal. */
3758         while (extract64(frac, frac_size - 1, 1) == 0) {
3759             exp--;
3760             frac <<= 1;
3761         }
3762 
3763         frac = (frac << 1) & MAKE_64BIT_MASK(0, frac_size);
3764 
3765         if (exp != 0 && exp != UINT64_MAX) {
3766             /*
3767              * Overflow to inf or max value of same sign,
3768              * depending on sign and rounding mode.
3769              */
3770             s->float_exception_flags |= (float_flag_inexact |
3771                                          float_flag_overflow);
3772 
3773             if ((s->float_rounding_mode == float_round_to_zero) ||
3774                 ((s->float_rounding_mode == float_round_down) && !sign) ||
3775                 ((s->float_rounding_mode == float_round_up) && sign)) {
3776                 /* Return greatest/negative finite value. */
3777                 return (sign << (exp_size + frac_size)) |
3778                        (MAKE_64BIT_MASK(frac_size, exp_size) - 1);
3779             } else {
3780                 /* Return +-inf. */
3781                 return (sign << (exp_size + frac_size)) |
3782                        MAKE_64BIT_MASK(frac_size, exp_size);
3783             }
3784         }
3785     }
3786 
3787     int idx = frac >> (frac_size - precision);
3788     uint64_t out_frac = (uint64_t)(lookup_table[idx]) <<
3789                         (frac_size - precision);
3790     uint64_t out_exp = 2 * MAKE_64BIT_MASK(0, exp_size - 1) + ~exp;
3791 
3792     if (out_exp == 0 || out_exp == UINT64_MAX) {
3793         /*
3794          * The result is subnormal, but don't raise the underflow exception,
3795          * because there's no additional loss of precision.
3796          */
3797         out_frac = (out_frac >> 1) | MAKE_64BIT_MASK(frac_size - 1, 1);
3798         if (out_exp == UINT64_MAX) {
3799             out_frac >>= 1;
3800             out_exp = 0;
3801         }
3802     }
3803 
3804     uint64_t val = 0;
3805     val = deposit64(val, 0, frac_size, out_frac);
3806     val = deposit64(val, frac_size, exp_size, out_exp);
3807     val = deposit64(val, frac_size + exp_size, 1, sign);
3808     return val;
3809 }
3810 
3811 static float16 frec7_h(float16 f, float_status *s)
3812 {
3813     int exp_size = 5, frac_size = 10;
3814     bool sign = float16_is_neg(f);
3815 
3816     /* frec7(+-inf) = +-0 */
3817     if (float16_is_infinity(f)) {
3818         return float16_set_sign(float16_zero, sign);
3819     }
3820 
3821     /* frec7(+-0) = +-inf */
3822     if (float16_is_zero(f)) {
3823         s->float_exception_flags |= float_flag_divbyzero;
3824         return float16_set_sign(float16_infinity, sign);
3825     }
3826 
3827     /* frec7(sNaN) = canonical NaN */
3828     if (float16_is_signaling_nan(f, s)) {
3829         s->float_exception_flags |= float_flag_invalid;
3830         return float16_default_nan(s);
3831     }
3832 
3833     /* frec7(qNaN) = canonical NaN */
3834     if (float16_is_quiet_nan(f, s)) {
3835         return float16_default_nan(s);
3836     }
3837 
3838     /* +-normal, +-subnormal */
3839     uint64_t val = frec7(f, exp_size, frac_size, s);
3840     return make_float16(val);
3841 }
3842 
3843 static float32 frec7_s(float32 f, float_status *s)
3844 {
3845     int exp_size = 8, frac_size = 23;
3846     bool sign = float32_is_neg(f);
3847 
3848     /* frec7(+-inf) = +-0 */
3849     if (float32_is_infinity(f)) {
3850         return float32_set_sign(float32_zero, sign);
3851     }
3852 
3853     /* frec7(+-0) = +-inf */
3854     if (float32_is_zero(f)) {
3855         s->float_exception_flags |= float_flag_divbyzero;
3856         return float32_set_sign(float32_infinity, sign);
3857     }
3858 
3859     /* frec7(sNaN) = canonical NaN */
3860     if (float32_is_signaling_nan(f, s)) {
3861         s->float_exception_flags |= float_flag_invalid;
3862         return float32_default_nan(s);
3863     }
3864 
3865     /* frec7(qNaN) = canonical NaN */
3866     if (float32_is_quiet_nan(f, s)) {
3867         return float32_default_nan(s);
3868     }
3869 
3870     /* +-normal, +-subnormal */
3871     uint64_t val = frec7(f, exp_size, frac_size, s);
3872     return make_float32(val);
3873 }
3874 
3875 static float64 frec7_d(float64 f, float_status *s)
3876 {
3877     int exp_size = 11, frac_size = 52;
3878     bool sign = float64_is_neg(f);
3879 
3880     /* frec7(+-inf) = +-0 */
3881     if (float64_is_infinity(f)) {
3882         return float64_set_sign(float64_zero, sign);
3883     }
3884 
3885     /* frec7(+-0) = +-inf */
3886     if (float64_is_zero(f)) {
3887         s->float_exception_flags |= float_flag_divbyzero;
3888         return float64_set_sign(float64_infinity, sign);
3889     }
3890 
3891     /* frec7(sNaN) = canonical NaN */
3892     if (float64_is_signaling_nan(f, s)) {
3893         s->float_exception_flags |= float_flag_invalid;
3894         return float64_default_nan(s);
3895     }
3896 
3897     /* frec7(qNaN) = canonical NaN */
3898     if (float64_is_quiet_nan(f, s)) {
3899         return float64_default_nan(s);
3900     }
3901 
3902     /* +-normal, +-subnormal */
3903     uint64_t val = frec7(f, exp_size, frac_size, s);
3904     return make_float64(val);
3905 }
3906 
3907 RVVCALL(OPFVV1, vfrec7_v_h, OP_UU_H, H2, H2, frec7_h)
3908 RVVCALL(OPFVV1, vfrec7_v_w, OP_UU_W, H4, H4, frec7_s)
3909 RVVCALL(OPFVV1, vfrec7_v_d, OP_UU_D, H8, H8, frec7_d)
3910 GEN_VEXT_V_ENV(vfrec7_v_h, 2)
3911 GEN_VEXT_V_ENV(vfrec7_v_w, 4)
3912 GEN_VEXT_V_ENV(vfrec7_v_d, 8)
3913 
3914 /* Vector Floating-Point MIN/MAX Instructions */
3915 RVVCALL(OPFVV2, vfmin_vv_h, OP_UUU_H, H2, H2, H2, float16_minimum_number)
3916 RVVCALL(OPFVV2, vfmin_vv_w, OP_UUU_W, H4, H4, H4, float32_minimum_number)
3917 RVVCALL(OPFVV2, vfmin_vv_d, OP_UUU_D, H8, H8, H8, float64_minimum_number)
3918 GEN_VEXT_VV_ENV(vfmin_vv_h, 2)
3919 GEN_VEXT_VV_ENV(vfmin_vv_w, 4)
3920 GEN_VEXT_VV_ENV(vfmin_vv_d, 8)
3921 RVVCALL(OPFVF2, vfmin_vf_h, OP_UUU_H, H2, H2, float16_minimum_number)
3922 RVVCALL(OPFVF2, vfmin_vf_w, OP_UUU_W, H4, H4, float32_minimum_number)
3923 RVVCALL(OPFVF2, vfmin_vf_d, OP_UUU_D, H8, H8, float64_minimum_number)
3924 GEN_VEXT_VF(vfmin_vf_h, 2)
3925 GEN_VEXT_VF(vfmin_vf_w, 4)
3926 GEN_VEXT_VF(vfmin_vf_d, 8)
3927 
3928 RVVCALL(OPFVV2, vfmax_vv_h, OP_UUU_H, H2, H2, H2, float16_maximum_number)
3929 RVVCALL(OPFVV2, vfmax_vv_w, OP_UUU_W, H4, H4, H4, float32_maximum_number)
3930 RVVCALL(OPFVV2, vfmax_vv_d, OP_UUU_D, H8, H8, H8, float64_maximum_number)
3931 GEN_VEXT_VV_ENV(vfmax_vv_h, 2)
3932 GEN_VEXT_VV_ENV(vfmax_vv_w, 4)
3933 GEN_VEXT_VV_ENV(vfmax_vv_d, 8)
3934 RVVCALL(OPFVF2, vfmax_vf_h, OP_UUU_H, H2, H2, float16_maximum_number)
3935 RVVCALL(OPFVF2, vfmax_vf_w, OP_UUU_W, H4, H4, float32_maximum_number)
3936 RVVCALL(OPFVF2, vfmax_vf_d, OP_UUU_D, H8, H8, float64_maximum_number)
3937 GEN_VEXT_VF(vfmax_vf_h, 2)
3938 GEN_VEXT_VF(vfmax_vf_w, 4)
3939 GEN_VEXT_VF(vfmax_vf_d, 8)
3940 
3941 /* Vector Floating-Point Sign-Injection Instructions */
3942 static uint16_t fsgnj16(uint16_t a, uint16_t b, float_status *s)
3943 {
3944     return deposit64(b, 0, 15, a);
3945 }
3946 
3947 static uint32_t fsgnj32(uint32_t a, uint32_t b, float_status *s)
3948 {
3949     return deposit64(b, 0, 31, a);
3950 }
3951 
3952 static uint64_t fsgnj64(uint64_t a, uint64_t b, float_status *s)
3953 {
3954     return deposit64(b, 0, 63, a);
3955 }
3956 
3957 RVVCALL(OPFVV2, vfsgnj_vv_h, OP_UUU_H, H2, H2, H2, fsgnj16)
3958 RVVCALL(OPFVV2, vfsgnj_vv_w, OP_UUU_W, H4, H4, H4, fsgnj32)
3959 RVVCALL(OPFVV2, vfsgnj_vv_d, OP_UUU_D, H8, H8, H8, fsgnj64)
3960 GEN_VEXT_VV_ENV(vfsgnj_vv_h, 2)
3961 GEN_VEXT_VV_ENV(vfsgnj_vv_w, 4)
3962 GEN_VEXT_VV_ENV(vfsgnj_vv_d, 8)
3963 RVVCALL(OPFVF2, vfsgnj_vf_h, OP_UUU_H, H2, H2, fsgnj16)
3964 RVVCALL(OPFVF2, vfsgnj_vf_w, OP_UUU_W, H4, H4, fsgnj32)
3965 RVVCALL(OPFVF2, vfsgnj_vf_d, OP_UUU_D, H8, H8, fsgnj64)
3966 GEN_VEXT_VF(vfsgnj_vf_h, 2)
3967 GEN_VEXT_VF(vfsgnj_vf_w, 4)
3968 GEN_VEXT_VF(vfsgnj_vf_d, 8)
3969 
3970 static uint16_t fsgnjn16(uint16_t a, uint16_t b, float_status *s)
3971 {
3972     return deposit64(~b, 0, 15, a);
3973 }
3974 
3975 static uint32_t fsgnjn32(uint32_t a, uint32_t b, float_status *s)
3976 {
3977     return deposit64(~b, 0, 31, a);
3978 }
3979 
3980 static uint64_t fsgnjn64(uint64_t a, uint64_t b, float_status *s)
3981 {
3982     return deposit64(~b, 0, 63, a);
3983 }
3984 
3985 RVVCALL(OPFVV2, vfsgnjn_vv_h, OP_UUU_H, H2, H2, H2, fsgnjn16)
3986 RVVCALL(OPFVV2, vfsgnjn_vv_w, OP_UUU_W, H4, H4, H4, fsgnjn32)
3987 RVVCALL(OPFVV2, vfsgnjn_vv_d, OP_UUU_D, H8, H8, H8, fsgnjn64)
3988 GEN_VEXT_VV_ENV(vfsgnjn_vv_h, 2)
3989 GEN_VEXT_VV_ENV(vfsgnjn_vv_w, 4)
3990 GEN_VEXT_VV_ENV(vfsgnjn_vv_d, 8)
3991 RVVCALL(OPFVF2, vfsgnjn_vf_h, OP_UUU_H, H2, H2, fsgnjn16)
3992 RVVCALL(OPFVF2, vfsgnjn_vf_w, OP_UUU_W, H4, H4, fsgnjn32)
3993 RVVCALL(OPFVF2, vfsgnjn_vf_d, OP_UUU_D, H8, H8, fsgnjn64)
3994 GEN_VEXT_VF(vfsgnjn_vf_h, 2)
3995 GEN_VEXT_VF(vfsgnjn_vf_w, 4)
3996 GEN_VEXT_VF(vfsgnjn_vf_d, 8)
3997 
3998 static uint16_t fsgnjx16(uint16_t a, uint16_t b, float_status *s)
3999 {
4000     return deposit64(b ^ a, 0, 15, a);
4001 }
4002 
4003 static uint32_t fsgnjx32(uint32_t a, uint32_t b, float_status *s)
4004 {
4005     return deposit64(b ^ a, 0, 31, a);
4006 }
4007 
4008 static uint64_t fsgnjx64(uint64_t a, uint64_t b, float_status *s)
4009 {
4010     return deposit64(b ^ a, 0, 63, a);
4011 }
4012 
4013 RVVCALL(OPFVV2, vfsgnjx_vv_h, OP_UUU_H, H2, H2, H2, fsgnjx16)
4014 RVVCALL(OPFVV2, vfsgnjx_vv_w, OP_UUU_W, H4, H4, H4, fsgnjx32)
4015 RVVCALL(OPFVV2, vfsgnjx_vv_d, OP_UUU_D, H8, H8, H8, fsgnjx64)
4016 GEN_VEXT_VV_ENV(vfsgnjx_vv_h, 2)
4017 GEN_VEXT_VV_ENV(vfsgnjx_vv_w, 4)
4018 GEN_VEXT_VV_ENV(vfsgnjx_vv_d, 8)
4019 RVVCALL(OPFVF2, vfsgnjx_vf_h, OP_UUU_H, H2, H2, fsgnjx16)
4020 RVVCALL(OPFVF2, vfsgnjx_vf_w, OP_UUU_W, H4, H4, fsgnjx32)
4021 RVVCALL(OPFVF2, vfsgnjx_vf_d, OP_UUU_D, H8, H8, fsgnjx64)
4022 GEN_VEXT_VF(vfsgnjx_vf_h, 2)
4023 GEN_VEXT_VF(vfsgnjx_vf_w, 4)
4024 GEN_VEXT_VF(vfsgnjx_vf_d, 8)
4025 
4026 /* Vector Floating-Point Compare Instructions */
4027 #define GEN_VEXT_CMP_VV_ENV(NAME, ETYPE, H, DO_OP)            \
4028 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,   \
4029                   CPURISCVState *env, uint32_t desc)          \
4030 {                                                             \
4031     uint32_t vm = vext_vm(desc);                              \
4032     uint32_t vl = env->vl;                                    \
4033     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;    \
4034     uint32_t vta_all_1s = vext_vta_all_1s(desc);              \
4035     uint32_t vma = vext_vma(desc);                            \
4036     uint32_t i;                                               \
4037                                                               \
4038     VSTART_CHECK_EARLY_EXIT(env);                             \
4039                                                               \
4040     for (i = env->vstart; i < vl; i++) {                      \
4041         ETYPE s1 = *((ETYPE *)vs1 + H(i));                    \
4042         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
4043         if (!vm && !vext_elem_mask(v0, i)) {                  \
4044             /* set masked-off elements to 1s */               \
4045             if (vma) {                                        \
4046                 vext_set_elem_mask(vd, i, 1);                 \
4047             }                                                 \
4048             continue;                                         \
4049         }                                                     \
4050         vext_set_elem_mask(vd, i,                             \
4051                            DO_OP(s2, s1, &env->fp_status));   \
4052     }                                                         \
4053     env->vstart = 0;                                          \
4054     /*
4055      * mask destination register are always tail-agnostic
4056      * set tail elements to 1s
4057      */                                                       \
4058     if (vta_all_1s) {                                         \
4059         for (; i < total_elems; i++) {                        \
4060             vext_set_elem_mask(vd, i, 1);                     \
4061         }                                                     \
4062     }                                                         \
4063 }
4064 
4065 GEN_VEXT_CMP_VV_ENV(vmfeq_vv_h, uint16_t, H2, float16_eq_quiet)
4066 GEN_VEXT_CMP_VV_ENV(vmfeq_vv_w, uint32_t, H4, float32_eq_quiet)
4067 GEN_VEXT_CMP_VV_ENV(vmfeq_vv_d, uint64_t, H8, float64_eq_quiet)
4068 
4069 #define GEN_VEXT_CMP_VF(NAME, ETYPE, H, DO_OP)                      \
4070 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2,       \
4071                   CPURISCVState *env, uint32_t desc)                \
4072 {                                                                   \
4073     uint32_t vm = vext_vm(desc);                                    \
4074     uint32_t vl = env->vl;                                          \
4075     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;          \
4076     uint32_t vta_all_1s = vext_vta_all_1s(desc);                    \
4077     uint32_t vma = vext_vma(desc);                                  \
4078     uint32_t i;                                                     \
4079                                                                     \
4080     VSTART_CHECK_EARLY_EXIT(env);                                   \
4081                                                                     \
4082     for (i = env->vstart; i < vl; i++) {                            \
4083         ETYPE s2 = *((ETYPE *)vs2 + H(i));                          \
4084         if (!vm && !vext_elem_mask(v0, i)) {                        \
4085             /* set masked-off elements to 1s */                     \
4086             if (vma) {                                              \
4087                 vext_set_elem_mask(vd, i, 1);                       \
4088             }                                                       \
4089             continue;                                               \
4090         }                                                           \
4091         vext_set_elem_mask(vd, i,                                   \
4092                            DO_OP(s2, (ETYPE)s1, &env->fp_status));  \
4093     }                                                               \
4094     env->vstart = 0;                                                \
4095     /*
4096      * mask destination register are always tail-agnostic
4097      * set tail elements to 1s
4098      */                                                             \
4099     if (vta_all_1s) {                                               \
4100         for (; i < total_elems; i++) {                              \
4101             vext_set_elem_mask(vd, i, 1);                           \
4102         }                                                           \
4103     }                                                               \
4104 }
4105 
4106 GEN_VEXT_CMP_VF(vmfeq_vf_h, uint16_t, H2, float16_eq_quiet)
4107 GEN_VEXT_CMP_VF(vmfeq_vf_w, uint32_t, H4, float32_eq_quiet)
4108 GEN_VEXT_CMP_VF(vmfeq_vf_d, uint64_t, H8, float64_eq_quiet)
4109 
4110 static bool vmfne16(uint16_t a, uint16_t b, float_status *s)
4111 {
4112     FloatRelation compare = float16_compare_quiet(a, b, s);
4113     return compare != float_relation_equal;
4114 }
4115 
4116 static bool vmfne32(uint32_t a, uint32_t b, float_status *s)
4117 {
4118     FloatRelation compare = float32_compare_quiet(a, b, s);
4119     return compare != float_relation_equal;
4120 }
4121 
4122 static bool vmfne64(uint64_t a, uint64_t b, float_status *s)
4123 {
4124     FloatRelation compare = float64_compare_quiet(a, b, s);
4125     return compare != float_relation_equal;
4126 }
4127 
4128 GEN_VEXT_CMP_VV_ENV(vmfne_vv_h, uint16_t, H2, vmfne16)
4129 GEN_VEXT_CMP_VV_ENV(vmfne_vv_w, uint32_t, H4, vmfne32)
4130 GEN_VEXT_CMP_VV_ENV(vmfne_vv_d, uint64_t, H8, vmfne64)
4131 GEN_VEXT_CMP_VF(vmfne_vf_h, uint16_t, H2, vmfne16)
4132 GEN_VEXT_CMP_VF(vmfne_vf_w, uint32_t, H4, vmfne32)
4133 GEN_VEXT_CMP_VF(vmfne_vf_d, uint64_t, H8, vmfne64)
4134 
4135 GEN_VEXT_CMP_VV_ENV(vmflt_vv_h, uint16_t, H2, float16_lt)
4136 GEN_VEXT_CMP_VV_ENV(vmflt_vv_w, uint32_t, H4, float32_lt)
4137 GEN_VEXT_CMP_VV_ENV(vmflt_vv_d, uint64_t, H8, float64_lt)
4138 GEN_VEXT_CMP_VF(vmflt_vf_h, uint16_t, H2, float16_lt)
4139 GEN_VEXT_CMP_VF(vmflt_vf_w, uint32_t, H4, float32_lt)
4140 GEN_VEXT_CMP_VF(vmflt_vf_d, uint64_t, H8, float64_lt)
4141 
4142 GEN_VEXT_CMP_VV_ENV(vmfle_vv_h, uint16_t, H2, float16_le)
4143 GEN_VEXT_CMP_VV_ENV(vmfle_vv_w, uint32_t, H4, float32_le)
4144 GEN_VEXT_CMP_VV_ENV(vmfle_vv_d, uint64_t, H8, float64_le)
4145 GEN_VEXT_CMP_VF(vmfle_vf_h, uint16_t, H2, float16_le)
4146 GEN_VEXT_CMP_VF(vmfle_vf_w, uint32_t, H4, float32_le)
4147 GEN_VEXT_CMP_VF(vmfle_vf_d, uint64_t, H8, float64_le)
4148 
4149 static bool vmfgt16(uint16_t a, uint16_t b, float_status *s)
4150 {
4151     FloatRelation compare = float16_compare(a, b, s);
4152     return compare == float_relation_greater;
4153 }
4154 
4155 static bool vmfgt32(uint32_t a, uint32_t b, float_status *s)
4156 {
4157     FloatRelation compare = float32_compare(a, b, s);
4158     return compare == float_relation_greater;
4159 }
4160 
4161 static bool vmfgt64(uint64_t a, uint64_t b, float_status *s)
4162 {
4163     FloatRelation compare = float64_compare(a, b, s);
4164     return compare == float_relation_greater;
4165 }
4166 
4167 GEN_VEXT_CMP_VF(vmfgt_vf_h, uint16_t, H2, vmfgt16)
4168 GEN_VEXT_CMP_VF(vmfgt_vf_w, uint32_t, H4, vmfgt32)
4169 GEN_VEXT_CMP_VF(vmfgt_vf_d, uint64_t, H8, vmfgt64)
4170 
4171 static bool vmfge16(uint16_t a, uint16_t b, float_status *s)
4172 {
4173     FloatRelation compare = float16_compare(a, b, s);
4174     return compare == float_relation_greater ||
4175            compare == float_relation_equal;
4176 }
4177 
4178 static bool vmfge32(uint32_t a, uint32_t b, float_status *s)
4179 {
4180     FloatRelation compare = float32_compare(a, b, s);
4181     return compare == float_relation_greater ||
4182            compare == float_relation_equal;
4183 }
4184 
4185 static bool vmfge64(uint64_t a, uint64_t b, float_status *s)
4186 {
4187     FloatRelation compare = float64_compare(a, b, s);
4188     return compare == float_relation_greater ||
4189            compare == float_relation_equal;
4190 }
4191 
4192 GEN_VEXT_CMP_VF(vmfge_vf_h, uint16_t, H2, vmfge16)
4193 GEN_VEXT_CMP_VF(vmfge_vf_w, uint32_t, H4, vmfge32)
4194 GEN_VEXT_CMP_VF(vmfge_vf_d, uint64_t, H8, vmfge64)
4195 
4196 /* Vector Floating-Point Classify Instruction */
4197 target_ulong fclass_h(uint64_t frs1)
4198 {
4199     float16 f = frs1;
4200     bool sign = float16_is_neg(f);
4201 
4202     if (float16_is_infinity(f)) {
4203         return sign ? 1 << 0 : 1 << 7;
4204     } else if (float16_is_zero(f)) {
4205         return sign ? 1 << 3 : 1 << 4;
4206     } else if (float16_is_zero_or_denormal(f)) {
4207         return sign ? 1 << 2 : 1 << 5;
4208     } else if (float16_is_any_nan(f)) {
4209         float_status s = { }; /* for snan_bit_is_one */
4210         return float16_is_quiet_nan(f, &s) ? 1 << 9 : 1 << 8;
4211     } else {
4212         return sign ? 1 << 1 : 1 << 6;
4213     }
4214 }
4215 
4216 target_ulong fclass_s(uint64_t frs1)
4217 {
4218     float32 f = frs1;
4219     bool sign = float32_is_neg(f);
4220 
4221     if (float32_is_infinity(f)) {
4222         return sign ? 1 << 0 : 1 << 7;
4223     } else if (float32_is_zero(f)) {
4224         return sign ? 1 << 3 : 1 << 4;
4225     } else if (float32_is_zero_or_denormal(f)) {
4226         return sign ? 1 << 2 : 1 << 5;
4227     } else if (float32_is_any_nan(f)) {
4228         float_status s = { }; /* for snan_bit_is_one */
4229         return float32_is_quiet_nan(f, &s) ? 1 << 9 : 1 << 8;
4230     } else {
4231         return sign ? 1 << 1 : 1 << 6;
4232     }
4233 }
4234 
4235 target_ulong fclass_d(uint64_t frs1)
4236 {
4237     float64 f = frs1;
4238     bool sign = float64_is_neg(f);
4239 
4240     if (float64_is_infinity(f)) {
4241         return sign ? 1 << 0 : 1 << 7;
4242     } else if (float64_is_zero(f)) {
4243         return sign ? 1 << 3 : 1 << 4;
4244     } else if (float64_is_zero_or_denormal(f)) {
4245         return sign ? 1 << 2 : 1 << 5;
4246     } else if (float64_is_any_nan(f)) {
4247         float_status s = { }; /* for snan_bit_is_one */
4248         return float64_is_quiet_nan(f, &s) ? 1 << 9 : 1 << 8;
4249     } else {
4250         return sign ? 1 << 1 : 1 << 6;
4251     }
4252 }
4253 
4254 RVVCALL(OPIVV1, vfclass_v_h, OP_UU_H, H2, H2, fclass_h)
4255 RVVCALL(OPIVV1, vfclass_v_w, OP_UU_W, H4, H4, fclass_s)
4256 RVVCALL(OPIVV1, vfclass_v_d, OP_UU_D, H8, H8, fclass_d)
4257 GEN_VEXT_V(vfclass_v_h, 2)
4258 GEN_VEXT_V(vfclass_v_w, 4)
4259 GEN_VEXT_V(vfclass_v_d, 8)
4260 
4261 /* Vector Floating-Point Merge Instruction */
4262 
4263 #define GEN_VFMERGE_VF(NAME, ETYPE, H)                        \
4264 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2, \
4265                   CPURISCVState *env, uint32_t desc)          \
4266 {                                                             \
4267     uint32_t vm = vext_vm(desc);                              \
4268     uint32_t vl = env->vl;                                    \
4269     uint32_t esz = sizeof(ETYPE);                             \
4270     uint32_t total_elems =                                    \
4271         vext_get_total_elems(env, desc, esz);                 \
4272     uint32_t vta = vext_vta(desc);                            \
4273     uint32_t i;                                               \
4274                                                               \
4275     VSTART_CHECK_EARLY_EXIT(env);                             \
4276                                                               \
4277     for (i = env->vstart; i < vl; i++) {                      \
4278         ETYPE s2 = *((ETYPE *)vs2 + H(i));                    \
4279         *((ETYPE *)vd + H(i)) =                               \
4280             (!vm && !vext_elem_mask(v0, i) ? s2 : s1);        \
4281     }                                                         \
4282     env->vstart = 0;                                          \
4283     /* set tail elements to 1s */                             \
4284     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);  \
4285 }
4286 
4287 GEN_VFMERGE_VF(vfmerge_vfm_h, int16_t, H2)
4288 GEN_VFMERGE_VF(vfmerge_vfm_w, int32_t, H4)
4289 GEN_VFMERGE_VF(vfmerge_vfm_d, int64_t, H8)
4290 
4291 /* Single-Width Floating-Point/Integer Type-Convert Instructions */
4292 /* vfcvt.xu.f.v vd, vs2, vm # Convert float to unsigned integer. */
4293 RVVCALL(OPFVV1, vfcvt_xu_f_v_h, OP_UU_H, H2, H2, float16_to_uint16)
4294 RVVCALL(OPFVV1, vfcvt_xu_f_v_w, OP_UU_W, H4, H4, float32_to_uint32)
4295 RVVCALL(OPFVV1, vfcvt_xu_f_v_d, OP_UU_D, H8, H8, float64_to_uint64)
4296 GEN_VEXT_V_ENV(vfcvt_xu_f_v_h, 2)
4297 GEN_VEXT_V_ENV(vfcvt_xu_f_v_w, 4)
4298 GEN_VEXT_V_ENV(vfcvt_xu_f_v_d, 8)
4299 
4300 /* vfcvt.x.f.v vd, vs2, vm # Convert float to signed integer. */
4301 RVVCALL(OPFVV1, vfcvt_x_f_v_h, OP_UU_H, H2, H2, float16_to_int16)
4302 RVVCALL(OPFVV1, vfcvt_x_f_v_w, OP_UU_W, H4, H4, float32_to_int32)
4303 RVVCALL(OPFVV1, vfcvt_x_f_v_d, OP_UU_D, H8, H8, float64_to_int64)
4304 GEN_VEXT_V_ENV(vfcvt_x_f_v_h, 2)
4305 GEN_VEXT_V_ENV(vfcvt_x_f_v_w, 4)
4306 GEN_VEXT_V_ENV(vfcvt_x_f_v_d, 8)
4307 
4308 /* vfcvt.f.xu.v vd, vs2, vm # Convert unsigned integer to float. */
4309 RVVCALL(OPFVV1, vfcvt_f_xu_v_h, OP_UU_H, H2, H2, uint16_to_float16)
4310 RVVCALL(OPFVV1, vfcvt_f_xu_v_w, OP_UU_W, H4, H4, uint32_to_float32)
4311 RVVCALL(OPFVV1, vfcvt_f_xu_v_d, OP_UU_D, H8, H8, uint64_to_float64)
4312 GEN_VEXT_V_ENV(vfcvt_f_xu_v_h, 2)
4313 GEN_VEXT_V_ENV(vfcvt_f_xu_v_w, 4)
4314 GEN_VEXT_V_ENV(vfcvt_f_xu_v_d, 8)
4315 
4316 /* vfcvt.f.x.v vd, vs2, vm # Convert integer to float. */
4317 RVVCALL(OPFVV1, vfcvt_f_x_v_h, OP_UU_H, H2, H2, int16_to_float16)
4318 RVVCALL(OPFVV1, vfcvt_f_x_v_w, OP_UU_W, H4, H4, int32_to_float32)
4319 RVVCALL(OPFVV1, vfcvt_f_x_v_d, OP_UU_D, H8, H8, int64_to_float64)
4320 GEN_VEXT_V_ENV(vfcvt_f_x_v_h, 2)
4321 GEN_VEXT_V_ENV(vfcvt_f_x_v_w, 4)
4322 GEN_VEXT_V_ENV(vfcvt_f_x_v_d, 8)
4323 
4324 /* Widening Floating-Point/Integer Type-Convert Instructions */
4325 /* (TD, T2, TX2) */
4326 #define WOP_UU_B uint16_t, uint8_t,  uint8_t
4327 #define WOP_UU_H uint32_t, uint16_t, uint16_t
4328 #define WOP_UU_W uint64_t, uint32_t, uint32_t
4329 /*
4330  * vfwcvt.xu.f.v vd, vs2, vm # Convert float to double-width unsigned integer.
4331  */
4332 RVVCALL(OPFVV1, vfwcvt_xu_f_v_h, WOP_UU_H, H4, H2, float16_to_uint32)
4333 RVVCALL(OPFVV1, vfwcvt_xu_f_v_w, WOP_UU_W, H8, H4, float32_to_uint64)
4334 GEN_VEXT_V_ENV(vfwcvt_xu_f_v_h, 4)
4335 GEN_VEXT_V_ENV(vfwcvt_xu_f_v_w, 8)
4336 
4337 /* vfwcvt.x.f.v vd, vs2, vm # Convert float to double-width signed integer. */
4338 RVVCALL(OPFVV1, vfwcvt_x_f_v_h, WOP_UU_H, H4, H2, float16_to_int32)
4339 RVVCALL(OPFVV1, vfwcvt_x_f_v_w, WOP_UU_W, H8, H4, float32_to_int64)
4340 GEN_VEXT_V_ENV(vfwcvt_x_f_v_h, 4)
4341 GEN_VEXT_V_ENV(vfwcvt_x_f_v_w, 8)
4342 
4343 /*
4344  * vfwcvt.f.xu.v vd, vs2, vm # Convert unsigned integer to double-width float.
4345  */
4346 RVVCALL(OPFVV1, vfwcvt_f_xu_v_b, WOP_UU_B, H2, H1, uint8_to_float16)
4347 RVVCALL(OPFVV1, vfwcvt_f_xu_v_h, WOP_UU_H, H4, H2, uint16_to_float32)
4348 RVVCALL(OPFVV1, vfwcvt_f_xu_v_w, WOP_UU_W, H8, H4, uint32_to_float64)
4349 GEN_VEXT_V_ENV(vfwcvt_f_xu_v_b, 2)
4350 GEN_VEXT_V_ENV(vfwcvt_f_xu_v_h, 4)
4351 GEN_VEXT_V_ENV(vfwcvt_f_xu_v_w, 8)
4352 
4353 /* vfwcvt.f.x.v vd, vs2, vm # Convert integer to double-width float. */
4354 RVVCALL(OPFVV1, vfwcvt_f_x_v_b, WOP_UU_B, H2, H1, int8_to_float16)
4355 RVVCALL(OPFVV1, vfwcvt_f_x_v_h, WOP_UU_H, H4, H2, int16_to_float32)
4356 RVVCALL(OPFVV1, vfwcvt_f_x_v_w, WOP_UU_W, H8, H4, int32_to_float64)
4357 GEN_VEXT_V_ENV(vfwcvt_f_x_v_b, 2)
4358 GEN_VEXT_V_ENV(vfwcvt_f_x_v_h, 4)
4359 GEN_VEXT_V_ENV(vfwcvt_f_x_v_w, 8)
4360 
4361 /*
4362  * vfwcvt.f.f.v vd, vs2, vm # Convert single-width float to double-width float.
4363  */
4364 static uint32_t vfwcvtffv16(uint16_t a, float_status *s)
4365 {
4366     return float16_to_float32(a, true, s);
4367 }
4368 
4369 RVVCALL(OPFVV1, vfwcvt_f_f_v_h, WOP_UU_H, H4, H2, vfwcvtffv16)
4370 RVVCALL(OPFVV1, vfwcvt_f_f_v_w, WOP_UU_W, H8, H4, float32_to_float64)
4371 GEN_VEXT_V_ENV(vfwcvt_f_f_v_h, 4)
4372 GEN_VEXT_V_ENV(vfwcvt_f_f_v_w, 8)
4373 
4374 RVVCALL(OPFVV1, vfwcvtbf16_f_f_v, WOP_UU_H, H4, H2, bfloat16_to_float32)
4375 GEN_VEXT_V_ENV(vfwcvtbf16_f_f_v, 4)
4376 
4377 /* Narrowing Floating-Point/Integer Type-Convert Instructions */
4378 /* (TD, T2, TX2) */
4379 #define NOP_UU_B uint8_t,  uint16_t, uint32_t
4380 #define NOP_UU_H uint16_t, uint32_t, uint32_t
4381 #define NOP_UU_W uint32_t, uint64_t, uint64_t
4382 /* vfncvt.xu.f.v vd, vs2, vm # Convert float to unsigned integer. */
4383 RVVCALL(OPFVV1, vfncvt_xu_f_w_b, NOP_UU_B, H1, H2, float16_to_uint8)
4384 RVVCALL(OPFVV1, vfncvt_xu_f_w_h, NOP_UU_H, H2, H4, float32_to_uint16)
4385 RVVCALL(OPFVV1, vfncvt_xu_f_w_w, NOP_UU_W, H4, H8, float64_to_uint32)
4386 GEN_VEXT_V_ENV(vfncvt_xu_f_w_b, 1)
4387 GEN_VEXT_V_ENV(vfncvt_xu_f_w_h, 2)
4388 GEN_VEXT_V_ENV(vfncvt_xu_f_w_w, 4)
4389 
4390 /* vfncvt.x.f.v vd, vs2, vm # Convert double-width float to signed integer. */
4391 RVVCALL(OPFVV1, vfncvt_x_f_w_b, NOP_UU_B, H1, H2, float16_to_int8)
4392 RVVCALL(OPFVV1, vfncvt_x_f_w_h, NOP_UU_H, H2, H4, float32_to_int16)
4393 RVVCALL(OPFVV1, vfncvt_x_f_w_w, NOP_UU_W, H4, H8, float64_to_int32)
4394 GEN_VEXT_V_ENV(vfncvt_x_f_w_b, 1)
4395 GEN_VEXT_V_ENV(vfncvt_x_f_w_h, 2)
4396 GEN_VEXT_V_ENV(vfncvt_x_f_w_w, 4)
4397 
4398 /*
4399  * vfncvt.f.xu.v vd, vs2, vm # Convert double-width unsigned integer to float.
4400  */
4401 RVVCALL(OPFVV1, vfncvt_f_xu_w_h, NOP_UU_H, H2, H4, uint32_to_float16)
4402 RVVCALL(OPFVV1, vfncvt_f_xu_w_w, NOP_UU_W, H4, H8, uint64_to_float32)
4403 GEN_VEXT_V_ENV(vfncvt_f_xu_w_h, 2)
4404 GEN_VEXT_V_ENV(vfncvt_f_xu_w_w, 4)
4405 
4406 /* vfncvt.f.x.v vd, vs2, vm # Convert double-width integer to float. */
4407 RVVCALL(OPFVV1, vfncvt_f_x_w_h, NOP_UU_H, H2, H4, int32_to_float16)
4408 RVVCALL(OPFVV1, vfncvt_f_x_w_w, NOP_UU_W, H4, H8, int64_to_float32)
4409 GEN_VEXT_V_ENV(vfncvt_f_x_w_h, 2)
4410 GEN_VEXT_V_ENV(vfncvt_f_x_w_w, 4)
4411 
4412 /* vfncvt.f.f.v vd, vs2, vm # Convert double float to single-width float. */
4413 static uint16_t vfncvtffv16(uint32_t a, float_status *s)
4414 {
4415     return float32_to_float16(a, true, s);
4416 }
4417 
4418 RVVCALL(OPFVV1, vfncvt_f_f_w_h, NOP_UU_H, H2, H4, vfncvtffv16)
4419 RVVCALL(OPFVV1, vfncvt_f_f_w_w, NOP_UU_W, H4, H8, float64_to_float32)
4420 GEN_VEXT_V_ENV(vfncvt_f_f_w_h, 2)
4421 GEN_VEXT_V_ENV(vfncvt_f_f_w_w, 4)
4422 
4423 RVVCALL(OPFVV1, vfncvtbf16_f_f_w, NOP_UU_H, H2, H4, float32_to_bfloat16)
4424 GEN_VEXT_V_ENV(vfncvtbf16_f_f_w, 2)
4425 
4426 /*
4427  * Vector Reduction Operations
4428  */
4429 /* Vector Single-Width Integer Reduction Instructions */
4430 #define GEN_VEXT_RED(NAME, TD, TS2, HD, HS2, OP)          \
4431 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
4432                   void *vs2, CPURISCVState *env,          \
4433                   uint32_t desc)                          \
4434 {                                                         \
4435     uint32_t vm = vext_vm(desc);                          \
4436     uint32_t vl = env->vl;                                \
4437     uint32_t esz = sizeof(TD);                            \
4438     uint32_t vlenb = simd_maxsz(desc);                    \
4439     uint32_t vta = vext_vta(desc);                        \
4440     uint32_t i;                                           \
4441     TD s1 =  *((TD *)vs1 + HD(0));                        \
4442                                                           \
4443     for (i = env->vstart; i < vl; i++) {                  \
4444         TS2 s2 = *((TS2 *)vs2 + HS2(i));                  \
4445         if (!vm && !vext_elem_mask(v0, i)) {              \
4446             continue;                                     \
4447         }                                                 \
4448         s1 = OP(s1, (TD)s2);                              \
4449     }                                                     \
4450     *((TD *)vd + HD(0)) = s1;                             \
4451     env->vstart = 0;                                      \
4452     /* set tail elements to 1s */                         \
4453     vext_set_elems_1s(vd, vta, esz, vlenb);               \
4454 }
4455 
4456 /* vd[0] = sum(vs1[0], vs2[*]) */
4457 GEN_VEXT_RED(vredsum_vs_b, int8_t,  int8_t,  H1, H1, DO_ADD)
4458 GEN_VEXT_RED(vredsum_vs_h, int16_t, int16_t, H2, H2, DO_ADD)
4459 GEN_VEXT_RED(vredsum_vs_w, int32_t, int32_t, H4, H4, DO_ADD)
4460 GEN_VEXT_RED(vredsum_vs_d, int64_t, int64_t, H8, H8, DO_ADD)
4461 
4462 /* vd[0] = maxu(vs1[0], vs2[*]) */
4463 GEN_VEXT_RED(vredmaxu_vs_b, uint8_t,  uint8_t,  H1, H1, DO_MAX)
4464 GEN_VEXT_RED(vredmaxu_vs_h, uint16_t, uint16_t, H2, H2, DO_MAX)
4465 GEN_VEXT_RED(vredmaxu_vs_w, uint32_t, uint32_t, H4, H4, DO_MAX)
4466 GEN_VEXT_RED(vredmaxu_vs_d, uint64_t, uint64_t, H8, H8, DO_MAX)
4467 
4468 /* vd[0] = max(vs1[0], vs2[*]) */
4469 GEN_VEXT_RED(vredmax_vs_b, int8_t,  int8_t,  H1, H1, DO_MAX)
4470 GEN_VEXT_RED(vredmax_vs_h, int16_t, int16_t, H2, H2, DO_MAX)
4471 GEN_VEXT_RED(vredmax_vs_w, int32_t, int32_t, H4, H4, DO_MAX)
4472 GEN_VEXT_RED(vredmax_vs_d, int64_t, int64_t, H8, H8, DO_MAX)
4473 
4474 /* vd[0] = minu(vs1[0], vs2[*]) */
4475 GEN_VEXT_RED(vredminu_vs_b, uint8_t,  uint8_t,  H1, H1, DO_MIN)
4476 GEN_VEXT_RED(vredminu_vs_h, uint16_t, uint16_t, H2, H2, DO_MIN)
4477 GEN_VEXT_RED(vredminu_vs_w, uint32_t, uint32_t, H4, H4, DO_MIN)
4478 GEN_VEXT_RED(vredminu_vs_d, uint64_t, uint64_t, H8, H8, DO_MIN)
4479 
4480 /* vd[0] = min(vs1[0], vs2[*]) */
4481 GEN_VEXT_RED(vredmin_vs_b, int8_t,  int8_t,  H1, H1, DO_MIN)
4482 GEN_VEXT_RED(vredmin_vs_h, int16_t, int16_t, H2, H2, DO_MIN)
4483 GEN_VEXT_RED(vredmin_vs_w, int32_t, int32_t, H4, H4, DO_MIN)
4484 GEN_VEXT_RED(vredmin_vs_d, int64_t, int64_t, H8, H8, DO_MIN)
4485 
4486 /* vd[0] = and(vs1[0], vs2[*]) */
4487 GEN_VEXT_RED(vredand_vs_b, int8_t,  int8_t,  H1, H1, DO_AND)
4488 GEN_VEXT_RED(vredand_vs_h, int16_t, int16_t, H2, H2, DO_AND)
4489 GEN_VEXT_RED(vredand_vs_w, int32_t, int32_t, H4, H4, DO_AND)
4490 GEN_VEXT_RED(vredand_vs_d, int64_t, int64_t, H8, H8, DO_AND)
4491 
4492 /* vd[0] = or(vs1[0], vs2[*]) */
4493 GEN_VEXT_RED(vredor_vs_b, int8_t,  int8_t,  H1, H1, DO_OR)
4494 GEN_VEXT_RED(vredor_vs_h, int16_t, int16_t, H2, H2, DO_OR)
4495 GEN_VEXT_RED(vredor_vs_w, int32_t, int32_t, H4, H4, DO_OR)
4496 GEN_VEXT_RED(vredor_vs_d, int64_t, int64_t, H8, H8, DO_OR)
4497 
4498 /* vd[0] = xor(vs1[0], vs2[*]) */
4499 GEN_VEXT_RED(vredxor_vs_b, int8_t,  int8_t,  H1, H1, DO_XOR)
4500 GEN_VEXT_RED(vredxor_vs_h, int16_t, int16_t, H2, H2, DO_XOR)
4501 GEN_VEXT_RED(vredxor_vs_w, int32_t, int32_t, H4, H4, DO_XOR)
4502 GEN_VEXT_RED(vredxor_vs_d, int64_t, int64_t, H8, H8, DO_XOR)
4503 
4504 /* Vector Widening Integer Reduction Instructions */
4505 /* signed sum reduction into double-width accumulator */
4506 GEN_VEXT_RED(vwredsum_vs_b, int16_t, int8_t,  H2, H1, DO_ADD)
4507 GEN_VEXT_RED(vwredsum_vs_h, int32_t, int16_t, H4, H2, DO_ADD)
4508 GEN_VEXT_RED(vwredsum_vs_w, int64_t, int32_t, H8, H4, DO_ADD)
4509 
4510 /* Unsigned sum reduction into double-width accumulator */
4511 GEN_VEXT_RED(vwredsumu_vs_b, uint16_t, uint8_t,  H2, H1, DO_ADD)
4512 GEN_VEXT_RED(vwredsumu_vs_h, uint32_t, uint16_t, H4, H2, DO_ADD)
4513 GEN_VEXT_RED(vwredsumu_vs_w, uint64_t, uint32_t, H8, H4, DO_ADD)
4514 
4515 /* Vector Single-Width Floating-Point Reduction Instructions */
4516 #define GEN_VEXT_FRED(NAME, TD, TS2, HD, HS2, OP)          \
4517 void HELPER(NAME)(void *vd, void *v0, void *vs1,           \
4518                   void *vs2, CPURISCVState *env,           \
4519                   uint32_t desc)                           \
4520 {                                                          \
4521     uint32_t vm = vext_vm(desc);                           \
4522     uint32_t vl = env->vl;                                 \
4523     uint32_t esz = sizeof(TD);                             \
4524     uint32_t vlenb = simd_maxsz(desc);                     \
4525     uint32_t vta = vext_vta(desc);                         \
4526     uint32_t i;                                            \
4527     TD s1 =  *((TD *)vs1 + HD(0));                         \
4528                                                            \
4529     for (i = env->vstart; i < vl; i++) {                   \
4530         TS2 s2 = *((TS2 *)vs2 + HS2(i));                   \
4531         if (!vm && !vext_elem_mask(v0, i)) {               \
4532             continue;                                      \
4533         }                                                  \
4534         s1 = OP(s1, (TD)s2, &env->fp_status);              \
4535     }                                                      \
4536     *((TD *)vd + HD(0)) = s1;                              \
4537     env->vstart = 0;                                       \
4538     /* set tail elements to 1s */                          \
4539     vext_set_elems_1s(vd, vta, esz, vlenb);                \
4540 }
4541 
4542 /* Unordered sum */
4543 GEN_VEXT_FRED(vfredusum_vs_h, uint16_t, uint16_t, H2, H2, float16_add)
4544 GEN_VEXT_FRED(vfredusum_vs_w, uint32_t, uint32_t, H4, H4, float32_add)
4545 GEN_VEXT_FRED(vfredusum_vs_d, uint64_t, uint64_t, H8, H8, float64_add)
4546 
4547 /* Ordered sum */
4548 GEN_VEXT_FRED(vfredosum_vs_h, uint16_t, uint16_t, H2, H2, float16_add)
4549 GEN_VEXT_FRED(vfredosum_vs_w, uint32_t, uint32_t, H4, H4, float32_add)
4550 GEN_VEXT_FRED(vfredosum_vs_d, uint64_t, uint64_t, H8, H8, float64_add)
4551 
4552 /* Maximum value */
4553 GEN_VEXT_FRED(vfredmax_vs_h, uint16_t, uint16_t, H2, H2,
4554               float16_maximum_number)
4555 GEN_VEXT_FRED(vfredmax_vs_w, uint32_t, uint32_t, H4, H4,
4556               float32_maximum_number)
4557 GEN_VEXT_FRED(vfredmax_vs_d, uint64_t, uint64_t, H8, H8,
4558               float64_maximum_number)
4559 
4560 /* Minimum value */
4561 GEN_VEXT_FRED(vfredmin_vs_h, uint16_t, uint16_t, H2, H2,
4562               float16_minimum_number)
4563 GEN_VEXT_FRED(vfredmin_vs_w, uint32_t, uint32_t, H4, H4,
4564               float32_minimum_number)
4565 GEN_VEXT_FRED(vfredmin_vs_d, uint64_t, uint64_t, H8, H8,
4566               float64_minimum_number)
4567 
4568 /* Vector Widening Floating-Point Add Instructions */
4569 static uint32_t fwadd16(uint32_t a, uint16_t b, float_status *s)
4570 {
4571     return float32_add(a, float16_to_float32(b, true, s), s);
4572 }
4573 
4574 static uint64_t fwadd32(uint64_t a, uint32_t b, float_status *s)
4575 {
4576     return float64_add(a, float32_to_float64(b, s), s);
4577 }
4578 
4579 /* Vector Widening Floating-Point Reduction Instructions */
4580 /* Ordered/unordered reduce 2*SEW = 2*SEW + sum(promote(SEW)) */
4581 GEN_VEXT_FRED(vfwredusum_vs_h, uint32_t, uint16_t, H4, H2, fwadd16)
4582 GEN_VEXT_FRED(vfwredusum_vs_w, uint64_t, uint32_t, H8, H4, fwadd32)
4583 GEN_VEXT_FRED(vfwredosum_vs_h, uint32_t, uint16_t, H4, H2, fwadd16)
4584 GEN_VEXT_FRED(vfwredosum_vs_w, uint64_t, uint32_t, H8, H4, fwadd32)
4585 
4586 /*
4587  * Vector Mask Operations
4588  */
4589 /* Vector Mask-Register Logical Instructions */
4590 #define GEN_VEXT_MASK_VV(NAME, OP)                        \
4591 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
4592                   void *vs2, CPURISCVState *env,          \
4593                   uint32_t desc)                          \
4594 {                                                         \
4595     uint32_t vl = env->vl;                                \
4596     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;\
4597     uint32_t vta_all_1s = vext_vta_all_1s(desc);          \
4598     uint32_t i;                                           \
4599     int a, b;                                             \
4600                                                           \
4601     VSTART_CHECK_EARLY_EXIT(env);                         \
4602                                                           \
4603     for (i = env->vstart; i < vl; i++) {                  \
4604         a = vext_elem_mask(vs1, i);                       \
4605         b = vext_elem_mask(vs2, i);                       \
4606         vext_set_elem_mask(vd, i, OP(b, a));              \
4607     }                                                     \
4608     env->vstart = 0;                                      \
4609     /*
4610      * mask destination register are always tail-agnostic
4611      * set tail elements to 1s
4612      */                                                   \
4613     if (vta_all_1s) {                                     \
4614         for (; i < total_elems; i++) {                    \
4615             vext_set_elem_mask(vd, i, 1);                 \
4616         }                                                 \
4617     }                                                     \
4618 }
4619 
4620 #define DO_NAND(N, M)  (!(N & M))
4621 #define DO_ANDNOT(N, M)  (N & !M)
4622 #define DO_NOR(N, M)  (!(N | M))
4623 #define DO_ORNOT(N, M)  (N | !M)
4624 #define DO_XNOR(N, M)  (!(N ^ M))
4625 
4626 GEN_VEXT_MASK_VV(vmand_mm, DO_AND)
4627 GEN_VEXT_MASK_VV(vmnand_mm, DO_NAND)
4628 GEN_VEXT_MASK_VV(vmandn_mm, DO_ANDNOT)
4629 GEN_VEXT_MASK_VV(vmxor_mm, DO_XOR)
4630 GEN_VEXT_MASK_VV(vmor_mm, DO_OR)
4631 GEN_VEXT_MASK_VV(vmnor_mm, DO_NOR)
4632 GEN_VEXT_MASK_VV(vmorn_mm, DO_ORNOT)
4633 GEN_VEXT_MASK_VV(vmxnor_mm, DO_XNOR)
4634 
4635 /* Vector count population in mask vcpop */
4636 target_ulong HELPER(vcpop_m)(void *v0, void *vs2, CPURISCVState *env,
4637                              uint32_t desc)
4638 {
4639     target_ulong cnt = 0;
4640     uint32_t vm = vext_vm(desc);
4641     uint32_t vl = env->vl;
4642     int i;
4643 
4644     for (i = env->vstart; i < vl; i++) {
4645         if (vm || vext_elem_mask(v0, i)) {
4646             if (vext_elem_mask(vs2, i)) {
4647                 cnt++;
4648             }
4649         }
4650     }
4651     env->vstart = 0;
4652     return cnt;
4653 }
4654 
4655 /* vfirst find-first-set mask bit */
4656 target_ulong HELPER(vfirst_m)(void *v0, void *vs2, CPURISCVState *env,
4657                               uint32_t desc)
4658 {
4659     uint32_t vm = vext_vm(desc);
4660     uint32_t vl = env->vl;
4661     int i;
4662 
4663     for (i = env->vstart; i < vl; i++) {
4664         if (vm || vext_elem_mask(v0, i)) {
4665             if (vext_elem_mask(vs2, i)) {
4666                 return i;
4667             }
4668         }
4669     }
4670     env->vstart = 0;
4671     return -1LL;
4672 }
4673 
4674 enum set_mask_type {
4675     ONLY_FIRST = 1,
4676     INCLUDE_FIRST,
4677     BEFORE_FIRST,
4678 };
4679 
4680 static void vmsetm(void *vd, void *v0, void *vs2, CPURISCVState *env,
4681                    uint32_t desc, enum set_mask_type type)
4682 {
4683     uint32_t vm = vext_vm(desc);
4684     uint32_t vl = env->vl;
4685     uint32_t total_elems = riscv_cpu_cfg(env)->vlenb << 3;
4686     uint32_t vta_all_1s = vext_vta_all_1s(desc);
4687     uint32_t vma = vext_vma(desc);
4688     int i;
4689     bool first_mask_bit = false;
4690 
4691     for (i = env->vstart; i < vl; i++) {
4692         if (!vm && !vext_elem_mask(v0, i)) {
4693             /* set masked-off elements to 1s */
4694             if (vma) {
4695                 vext_set_elem_mask(vd, i, 1);
4696             }
4697             continue;
4698         }
4699         /* write a zero to all following active elements */
4700         if (first_mask_bit) {
4701             vext_set_elem_mask(vd, i, 0);
4702             continue;
4703         }
4704         if (vext_elem_mask(vs2, i)) {
4705             first_mask_bit = true;
4706             if (type == BEFORE_FIRST) {
4707                 vext_set_elem_mask(vd, i, 0);
4708             } else {
4709                 vext_set_elem_mask(vd, i, 1);
4710             }
4711         } else {
4712             if (type == ONLY_FIRST) {
4713                 vext_set_elem_mask(vd, i, 0);
4714             } else {
4715                 vext_set_elem_mask(vd, i, 1);
4716             }
4717         }
4718     }
4719     env->vstart = 0;
4720     /*
4721      * mask destination register are always tail-agnostic
4722      * set tail elements to 1s
4723      */
4724     if (vta_all_1s) {
4725         for (; i < total_elems; i++) {
4726             vext_set_elem_mask(vd, i, 1);
4727         }
4728     }
4729 }
4730 
4731 void HELPER(vmsbf_m)(void *vd, void *v0, void *vs2, CPURISCVState *env,
4732                      uint32_t desc)
4733 {
4734     vmsetm(vd, v0, vs2, env, desc, BEFORE_FIRST);
4735 }
4736 
4737 void HELPER(vmsif_m)(void *vd, void *v0, void *vs2, CPURISCVState *env,
4738                      uint32_t desc)
4739 {
4740     vmsetm(vd, v0, vs2, env, desc, INCLUDE_FIRST);
4741 }
4742 
4743 void HELPER(vmsof_m)(void *vd, void *v0, void *vs2, CPURISCVState *env,
4744                      uint32_t desc)
4745 {
4746     vmsetm(vd, v0, vs2, env, desc, ONLY_FIRST);
4747 }
4748 
4749 /* Vector Iota Instruction */
4750 #define GEN_VEXT_VIOTA_M(NAME, ETYPE, H)                                  \
4751 void HELPER(NAME)(void *vd, void *v0, void *vs2, CPURISCVState *env,      \
4752                   uint32_t desc)                                          \
4753 {                                                                         \
4754     uint32_t vm = vext_vm(desc);                                          \
4755     uint32_t vl = env->vl;                                                \
4756     uint32_t esz = sizeof(ETYPE);                                         \
4757     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4758     uint32_t vta = vext_vta(desc);                                        \
4759     uint32_t vma = vext_vma(desc);                                        \
4760     uint32_t sum = 0;                                                     \
4761     int i;                                                                \
4762                                                                           \
4763     for (i = env->vstart; i < vl; i++) {                                  \
4764         if (!vm && !vext_elem_mask(v0, i)) {                              \
4765             /* set masked-off elements to 1s */                           \
4766             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4767             continue;                                                     \
4768         }                                                                 \
4769         *((ETYPE *)vd + H(i)) = sum;                                      \
4770         if (vext_elem_mask(vs2, i)) {                                     \
4771             sum++;                                                        \
4772         }                                                                 \
4773     }                                                                     \
4774     env->vstart = 0;                                                      \
4775     /* set tail elements to 1s */                                         \
4776     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4777 }
4778 
4779 GEN_VEXT_VIOTA_M(viota_m_b, uint8_t,  H1)
4780 GEN_VEXT_VIOTA_M(viota_m_h, uint16_t, H2)
4781 GEN_VEXT_VIOTA_M(viota_m_w, uint32_t, H4)
4782 GEN_VEXT_VIOTA_M(viota_m_d, uint64_t, H8)
4783 
4784 /* Vector Element Index Instruction */
4785 #define GEN_VEXT_VID_V(NAME, ETYPE, H)                                    \
4786 void HELPER(NAME)(void *vd, void *v0, CPURISCVState *env, uint32_t desc)  \
4787 {                                                                         \
4788     uint32_t vm = vext_vm(desc);                                          \
4789     uint32_t vl = env->vl;                                                \
4790     uint32_t esz = sizeof(ETYPE);                                         \
4791     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4792     uint32_t vta = vext_vta(desc);                                        \
4793     uint32_t vma = vext_vma(desc);                                        \
4794     int i;                                                                \
4795                                                                           \
4796     VSTART_CHECK_EARLY_EXIT(env);                                         \
4797                                                                           \
4798     for (i = env->vstart; i < vl; i++) {                                  \
4799         if (!vm && !vext_elem_mask(v0, i)) {                              \
4800             /* set masked-off elements to 1s */                           \
4801             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4802             continue;                                                     \
4803         }                                                                 \
4804         *((ETYPE *)vd + H(i)) = i;                                        \
4805     }                                                                     \
4806     env->vstart = 0;                                                      \
4807     /* set tail elements to 1s */                                         \
4808     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4809 }
4810 
4811 GEN_VEXT_VID_V(vid_v_b, uint8_t,  H1)
4812 GEN_VEXT_VID_V(vid_v_h, uint16_t, H2)
4813 GEN_VEXT_VID_V(vid_v_w, uint32_t, H4)
4814 GEN_VEXT_VID_V(vid_v_d, uint64_t, H8)
4815 
4816 /*
4817  * Vector Permutation Instructions
4818  */
4819 
4820 /* Vector Slide Instructions */
4821 #define GEN_VEXT_VSLIDEUP_VX(NAME, ETYPE, H)                              \
4822 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,         \
4823                   CPURISCVState *env, uint32_t desc)                      \
4824 {                                                                         \
4825     uint32_t vm = vext_vm(desc);                                          \
4826     uint32_t vl = env->vl;                                                \
4827     uint32_t esz = sizeof(ETYPE);                                         \
4828     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4829     uint32_t vta = vext_vta(desc);                                        \
4830     uint32_t vma = vext_vma(desc);                                        \
4831     target_ulong offset = s1, i_min, i;                                   \
4832                                                                           \
4833     VSTART_CHECK_EARLY_EXIT(env);                                         \
4834                                                                           \
4835     i_min = MAX(env->vstart, offset);                                     \
4836     for (i = i_min; i < vl; i++) {                                        \
4837         if (!vm && !vext_elem_mask(v0, i)) {                              \
4838             /* set masked-off elements to 1s */                           \
4839             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4840             continue;                                                     \
4841         }                                                                 \
4842         *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i - offset));          \
4843     }                                                                     \
4844     env->vstart = 0;                                                      \
4845     /* set tail elements to 1s */                                         \
4846     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4847 }
4848 
4849 /* vslideup.vx vd, vs2, rs1, vm # vd[i+rs1] = vs2[i] */
4850 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_b, uint8_t,  H1)
4851 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_h, uint16_t, H2)
4852 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_w, uint32_t, H4)
4853 GEN_VEXT_VSLIDEUP_VX(vslideup_vx_d, uint64_t, H8)
4854 
4855 #define GEN_VEXT_VSLIDEDOWN_VX(NAME, ETYPE, H)                            \
4856 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,         \
4857                   CPURISCVState *env, uint32_t desc)                      \
4858 {                                                                         \
4859     uint32_t vlmax = vext_max_elems(desc, ctzl(sizeof(ETYPE)));           \
4860     uint32_t vm = vext_vm(desc);                                          \
4861     uint32_t vl = env->vl;                                                \
4862     uint32_t esz = sizeof(ETYPE);                                         \
4863     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
4864     uint32_t vta = vext_vta(desc);                                        \
4865     uint32_t vma = vext_vma(desc);                                        \
4866     target_ulong i_max, i_min, i;                                         \
4867                                                                           \
4868     VSTART_CHECK_EARLY_EXIT(env);                                         \
4869                                                                           \
4870     i_min = MIN(s1 < vlmax ? vlmax - s1 : 0, vl);                         \
4871     i_max = MAX(i_min, env->vstart);                                      \
4872     for (i = env->vstart; i < i_max; ++i) {                               \
4873         if (!vm && !vext_elem_mask(v0, i)) {                              \
4874             /* set masked-off elements to 1s */                           \
4875             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
4876             continue;                                                     \
4877         }                                                                 \
4878         *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i + s1));              \
4879     }                                                                     \
4880                                                                           \
4881     for (i = i_max; i < vl; ++i) {                                        \
4882         if (vm || vext_elem_mask(v0, i)) {                                \
4883             *((ETYPE *)vd + H(i)) = 0;                                    \
4884         }                                                                 \
4885     }                                                                     \
4886                                                                           \
4887     env->vstart = 0;                                                      \
4888     /* set tail elements to 1s */                                         \
4889     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
4890 }
4891 
4892 /* vslidedown.vx vd, vs2, rs1, vm # vd[i] = vs2[i+rs1] */
4893 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_b, uint8_t,  H1)
4894 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_h, uint16_t, H2)
4895 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_w, uint32_t, H4)
4896 GEN_VEXT_VSLIDEDOWN_VX(vslidedown_vx_d, uint64_t, H8)
4897 
4898 #define GEN_VEXT_VSLIE1UP(BITWIDTH, H)                                      \
4899 static void vslide1up_##BITWIDTH(void *vd, void *v0, uint64_t s1,           \
4900                                  void *vs2, CPURISCVState *env,             \
4901                                  uint32_t desc)                             \
4902 {                                                                           \
4903     typedef uint##BITWIDTH##_t ETYPE;                                       \
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 i;                                                             \
4911                                                                             \
4912     VSTART_CHECK_EARLY_EXIT(env);                                           \
4913                                                                             \
4914     for (i = env->vstart; i < vl; i++) {                                    \
4915         if (!vm && !vext_elem_mask(v0, i)) {                                \
4916             /* set masked-off elements to 1s */                             \
4917             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);             \
4918             continue;                                                       \
4919         }                                                                   \
4920         if (i == 0) {                                                       \
4921             *((ETYPE *)vd + H(i)) = s1;                                     \
4922         } else {                                                            \
4923             *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i - 1));             \
4924         }                                                                   \
4925     }                                                                       \
4926     env->vstart = 0;                                                        \
4927     /* set tail elements to 1s */                                           \
4928     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);                \
4929 }
4930 
4931 GEN_VEXT_VSLIE1UP(8,  H1)
4932 GEN_VEXT_VSLIE1UP(16, H2)
4933 GEN_VEXT_VSLIE1UP(32, H4)
4934 GEN_VEXT_VSLIE1UP(64, H8)
4935 
4936 #define GEN_VEXT_VSLIDE1UP_VX(NAME, BITWIDTH)                     \
4937 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2, \
4938                   CPURISCVState *env, uint32_t desc)              \
4939 {                                                                 \
4940     vslide1up_##BITWIDTH(vd, v0, s1, vs2, env, desc);             \
4941 }
4942 
4943 /* vslide1up.vx vd, vs2, rs1, vm # vd[0]=x[rs1], vd[i+1] = vs2[i] */
4944 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_b, 8)
4945 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_h, 16)
4946 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_w, 32)
4947 GEN_VEXT_VSLIDE1UP_VX(vslide1up_vx_d, 64)
4948 
4949 #define GEN_VEXT_VSLIDE1DOWN(BITWIDTH, H)                                     \
4950 static void vslide1down_##BITWIDTH(void *vd, void *v0, uint64_t s1,           \
4951                                    void *vs2, CPURISCVState *env,             \
4952                                    uint32_t desc)                             \
4953 {                                                                             \
4954     typedef uint##BITWIDTH##_t ETYPE;                                         \
4955     uint32_t vm = vext_vm(desc);                                              \
4956     uint32_t vl = env->vl;                                                    \
4957     uint32_t esz = sizeof(ETYPE);                                             \
4958     uint32_t total_elems = vext_get_total_elems(env, desc, esz);              \
4959     uint32_t vta = vext_vta(desc);                                            \
4960     uint32_t vma = vext_vma(desc);                                            \
4961     uint32_t i;                                                               \
4962                                                                               \
4963     VSTART_CHECK_EARLY_EXIT(env);                                             \
4964                                                                               \
4965     for (i = env->vstart; i < vl; i++) {                                      \
4966         if (!vm && !vext_elem_mask(v0, i)) {                                  \
4967             /* set masked-off elements to 1s */                               \
4968             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);               \
4969             continue;                                                         \
4970         }                                                                     \
4971         if (i == vl - 1) {                                                    \
4972             *((ETYPE *)vd + H(i)) = s1;                                       \
4973         } else {                                                              \
4974             *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(i + 1));               \
4975         }                                                                     \
4976     }                                                                         \
4977     env->vstart = 0;                                                          \
4978     /* set tail elements to 1s */                                             \
4979     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);                  \
4980 }
4981 
4982 GEN_VEXT_VSLIDE1DOWN(8,  H1)
4983 GEN_VEXT_VSLIDE1DOWN(16, H2)
4984 GEN_VEXT_VSLIDE1DOWN(32, H4)
4985 GEN_VEXT_VSLIDE1DOWN(64, H8)
4986 
4987 #define GEN_VEXT_VSLIDE1DOWN_VX(NAME, BITWIDTH)                   \
4988 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2, \
4989                   CPURISCVState *env, uint32_t desc)              \
4990 {                                                                 \
4991     vslide1down_##BITWIDTH(vd, v0, s1, vs2, env, desc);           \
4992 }
4993 
4994 /* vslide1down.vx vd, vs2, rs1, vm # vd[i] = vs2[i+1], vd[vl-1]=x[rs1] */
4995 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_b, 8)
4996 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_h, 16)
4997 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_w, 32)
4998 GEN_VEXT_VSLIDE1DOWN_VX(vslide1down_vx_d, 64)
4999 
5000 /* Vector Floating-Point Slide Instructions */
5001 #define GEN_VEXT_VFSLIDE1UP_VF(NAME, BITWIDTH)                \
5002 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2, \
5003                   CPURISCVState *env, uint32_t desc)          \
5004 {                                                             \
5005     vslide1up_##BITWIDTH(vd, v0, s1, vs2, env, desc);         \
5006 }
5007 
5008 /* vfslide1up.vf vd, vs2, rs1, vm # vd[0]=f[rs1], vd[i+1] = vs2[i] */
5009 GEN_VEXT_VFSLIDE1UP_VF(vfslide1up_vf_h, 16)
5010 GEN_VEXT_VFSLIDE1UP_VF(vfslide1up_vf_w, 32)
5011 GEN_VEXT_VFSLIDE1UP_VF(vfslide1up_vf_d, 64)
5012 
5013 #define GEN_VEXT_VFSLIDE1DOWN_VF(NAME, BITWIDTH)              \
5014 void HELPER(NAME)(void *vd, void *v0, uint64_t s1, void *vs2, \
5015                   CPURISCVState *env, uint32_t desc)          \
5016 {                                                             \
5017     vslide1down_##BITWIDTH(vd, v0, s1, vs2, env, desc);       \
5018 }
5019 
5020 /* vfslide1down.vf vd, vs2, rs1, vm # vd[i] = vs2[i+1], vd[vl-1]=f[rs1] */
5021 GEN_VEXT_VFSLIDE1DOWN_VF(vfslide1down_vf_h, 16)
5022 GEN_VEXT_VFSLIDE1DOWN_VF(vfslide1down_vf_w, 32)
5023 GEN_VEXT_VFSLIDE1DOWN_VF(vfslide1down_vf_d, 64)
5024 
5025 /* Vector Register Gather Instruction */
5026 #define GEN_VEXT_VRGATHER_VV(NAME, TS1, TS2, HS1, HS2)                    \
5027 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,               \
5028                   CPURISCVState *env, uint32_t desc)                      \
5029 {                                                                         \
5030     uint32_t vlmax = vext_max_elems(desc, ctzl(sizeof(TS2)));             \
5031     uint32_t vm = vext_vm(desc);                                          \
5032     uint32_t vl = env->vl;                                                \
5033     uint32_t esz = sizeof(TS2);                                           \
5034     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5035     uint32_t vta = vext_vta(desc);                                        \
5036     uint32_t vma = vext_vma(desc);                                        \
5037     uint64_t index;                                                       \
5038     uint32_t i;                                                           \
5039                                                                           \
5040     VSTART_CHECK_EARLY_EXIT(env);                                         \
5041                                                                           \
5042     for (i = env->vstart; i < vl; i++) {                                  \
5043         if (!vm && !vext_elem_mask(v0, i)) {                              \
5044             /* set masked-off elements to 1s */                           \
5045             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
5046             continue;                                                     \
5047         }                                                                 \
5048         index = *((TS1 *)vs1 + HS1(i));                                   \
5049         if (index >= vlmax) {                                             \
5050             *((TS2 *)vd + HS2(i)) = 0;                                    \
5051         } else {                                                          \
5052             *((TS2 *)vd + HS2(i)) = *((TS2 *)vs2 + HS2(index));           \
5053         }                                                                 \
5054     }                                                                     \
5055     env->vstart = 0;                                                      \
5056     /* set tail elements to 1s */                                         \
5057     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5058 }
5059 
5060 /* vd[i] = (vs1[i] >= VLMAX) ? 0 : vs2[vs1[i]]; */
5061 GEN_VEXT_VRGATHER_VV(vrgather_vv_b, uint8_t,  uint8_t,  H1, H1)
5062 GEN_VEXT_VRGATHER_VV(vrgather_vv_h, uint16_t, uint16_t, H2, H2)
5063 GEN_VEXT_VRGATHER_VV(vrgather_vv_w, uint32_t, uint32_t, H4, H4)
5064 GEN_VEXT_VRGATHER_VV(vrgather_vv_d, uint64_t, uint64_t, H8, H8)
5065 
5066 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_b, uint16_t, uint8_t,  H2, H1)
5067 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_h, uint16_t, uint16_t, H2, H2)
5068 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_w, uint16_t, uint32_t, H2, H4)
5069 GEN_VEXT_VRGATHER_VV(vrgatherei16_vv_d, uint16_t, uint64_t, H2, H8)
5070 
5071 #define GEN_VEXT_VRGATHER_VX(NAME, ETYPE, H)                              \
5072 void HELPER(NAME)(void *vd, void *v0, target_ulong s1, void *vs2,         \
5073                   CPURISCVState *env, uint32_t desc)                      \
5074 {                                                                         \
5075     uint32_t vlmax = vext_max_elems(desc, ctzl(sizeof(ETYPE)));           \
5076     uint32_t vm = vext_vm(desc);                                          \
5077     uint32_t vl = env->vl;                                                \
5078     uint32_t esz = sizeof(ETYPE);                                         \
5079     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5080     uint32_t vta = vext_vta(desc);                                        \
5081     uint32_t vma = vext_vma(desc);                                        \
5082     uint64_t index = s1;                                                  \
5083     uint32_t i;                                                           \
5084                                                                           \
5085     VSTART_CHECK_EARLY_EXIT(env);                                         \
5086                                                                           \
5087     for (i = env->vstart; i < vl; i++) {                                  \
5088         if (!vm && !vext_elem_mask(v0, i)) {                              \
5089             /* set masked-off elements to 1s */                           \
5090             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);           \
5091             continue;                                                     \
5092         }                                                                 \
5093         if (index >= vlmax) {                                             \
5094             *((ETYPE *)vd + H(i)) = 0;                                    \
5095         } else {                                                          \
5096             *((ETYPE *)vd + H(i)) = *((ETYPE *)vs2 + H(index));           \
5097         }                                                                 \
5098     }                                                                     \
5099     env->vstart = 0;                                                      \
5100     /* set tail elements to 1s */                                         \
5101     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5102 }
5103 
5104 /* vd[i] = (x[rs1] >= VLMAX) ? 0 : vs2[rs1] */
5105 GEN_VEXT_VRGATHER_VX(vrgather_vx_b, uint8_t,  H1)
5106 GEN_VEXT_VRGATHER_VX(vrgather_vx_h, uint16_t, H2)
5107 GEN_VEXT_VRGATHER_VX(vrgather_vx_w, uint32_t, H4)
5108 GEN_VEXT_VRGATHER_VX(vrgather_vx_d, uint64_t, H8)
5109 
5110 /* Vector Compress Instruction */
5111 #define GEN_VEXT_VCOMPRESS_VM(NAME, ETYPE, H)                             \
5112 void HELPER(NAME)(void *vd, void *v0, void *vs1, void *vs2,               \
5113                   CPURISCVState *env, uint32_t desc)                      \
5114 {                                                                         \
5115     uint32_t vl = env->vl;                                                \
5116     uint32_t esz = sizeof(ETYPE);                                         \
5117     uint32_t total_elems = vext_get_total_elems(env, desc, esz);          \
5118     uint32_t vta = vext_vta(desc);                                        \
5119     uint32_t num = 0, i;                                                  \
5120                                                                           \
5121     for (i = env->vstart; i < vl; i++) {                                  \
5122         if (!vext_elem_mask(vs1, i)) {                                    \
5123             continue;                                                     \
5124         }                                                                 \
5125         *((ETYPE *)vd + H(num)) = *((ETYPE *)vs2 + H(i));                 \
5126         num++;                                                            \
5127     }                                                                     \
5128     env->vstart = 0;                                                      \
5129     /* set tail elements to 1s */                                         \
5130     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);              \
5131 }
5132 
5133 /* Compress into vd elements of vs2 where vs1 is enabled */
5134 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_b, uint8_t,  H1)
5135 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_h, uint16_t, H2)
5136 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_w, uint32_t, H4)
5137 GEN_VEXT_VCOMPRESS_VM(vcompress_vm_d, uint64_t, H8)
5138 
5139 /* Vector Whole Register Move */
5140 void HELPER(vmvr_v)(void *vd, void *vs2, CPURISCVState *env, uint32_t desc)
5141 {
5142     /* EEW = SEW */
5143     uint32_t maxsz = simd_maxsz(desc);
5144     uint32_t sewb = 1 << FIELD_EX64(env->vtype, VTYPE, VSEW);
5145     uint32_t startb = env->vstart * sewb;
5146     uint32_t i = startb;
5147 
5148     if (startb >= maxsz) {
5149         env->vstart = 0;
5150         return;
5151     }
5152 
5153     if (HOST_BIG_ENDIAN && i % 8 != 0) {
5154         uint32_t j = ROUND_UP(i, 8);
5155         memcpy((uint8_t *)vd + H1(j - 1),
5156                (uint8_t *)vs2 + H1(j - 1),
5157                j - i);
5158         i = j;
5159     }
5160 
5161     memcpy((uint8_t *)vd + H1(i),
5162            (uint8_t *)vs2 + H1(i),
5163            maxsz - i);
5164 
5165     env->vstart = 0;
5166 }
5167 
5168 /* Vector Integer Extension */
5169 #define GEN_VEXT_INT_EXT(NAME, ETYPE, DTYPE, HD, HS1)            \
5170 void HELPER(NAME)(void *vd, void *v0, void *vs2,                 \
5171                   CPURISCVState *env, uint32_t desc)             \
5172 {                                                                \
5173     uint32_t vl = env->vl;                                       \
5174     uint32_t vm = vext_vm(desc);                                 \
5175     uint32_t esz = sizeof(ETYPE);                                \
5176     uint32_t total_elems = vext_get_total_elems(env, desc, esz); \
5177     uint32_t vta = vext_vta(desc);                               \
5178     uint32_t vma = vext_vma(desc);                               \
5179     uint32_t i;                                                  \
5180                                                                  \
5181     VSTART_CHECK_EARLY_EXIT(env);                                \
5182                                                                  \
5183     for (i = env->vstart; i < vl; i++) {                         \
5184         if (!vm && !vext_elem_mask(v0, i)) {                     \
5185             /* set masked-off elements to 1s */                  \
5186             vext_set_elems_1s(vd, vma, i * esz, (i + 1) * esz);  \
5187             continue;                                            \
5188         }                                                        \
5189         *((ETYPE *)vd + HD(i)) = *((DTYPE *)vs2 + HS1(i));       \
5190     }                                                            \
5191     env->vstart = 0;                                             \
5192     /* set tail elements to 1s */                                \
5193     vext_set_elems_1s(vd, vta, vl * esz, total_elems * esz);     \
5194 }
5195 
5196 GEN_VEXT_INT_EXT(vzext_vf2_h, uint16_t, uint8_t,  H2, H1)
5197 GEN_VEXT_INT_EXT(vzext_vf2_w, uint32_t, uint16_t, H4, H2)
5198 GEN_VEXT_INT_EXT(vzext_vf2_d, uint64_t, uint32_t, H8, H4)
5199 GEN_VEXT_INT_EXT(vzext_vf4_w, uint32_t, uint8_t,  H4, H1)
5200 GEN_VEXT_INT_EXT(vzext_vf4_d, uint64_t, uint16_t, H8, H2)
5201 GEN_VEXT_INT_EXT(vzext_vf8_d, uint64_t, uint8_t,  H8, H1)
5202 
5203 GEN_VEXT_INT_EXT(vsext_vf2_h, int16_t, int8_t,  H2, H1)
5204 GEN_VEXT_INT_EXT(vsext_vf2_w, int32_t, int16_t, H4, H2)
5205 GEN_VEXT_INT_EXT(vsext_vf2_d, int64_t, int32_t, H8, H4)
5206 GEN_VEXT_INT_EXT(vsext_vf4_w, int32_t, int8_t,  H4, H1)
5207 GEN_VEXT_INT_EXT(vsext_vf4_d, int64_t, int16_t, H8, H2)
5208 GEN_VEXT_INT_EXT(vsext_vf8_d, int64_t, int8_t,  H8, H1)
5209