xref: /openbmc/qemu/target/riscv/vector_helper.c (revision 8fcdf776)
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 "cpu.h"
21 #include "exec/memop.h"
22 #include "exec/exec-all.h"
23 #include "exec/helper-proto.h"
24 #include "tcg/tcg-gvec-desc.h"
25 #include "internals.h"
26 #include <math.h>
27 
28 target_ulong HELPER(vsetvl)(CPURISCVState *env, target_ulong s1,
29                             target_ulong s2)
30 {
31     int vlmax, vl;
32     RISCVCPU *cpu = env_archcpu(env);
33     uint16_t sew = 8 << FIELD_EX64(s2, VTYPE, VSEW);
34     uint8_t ediv = FIELD_EX64(s2, VTYPE, VEDIV);
35     bool vill = FIELD_EX64(s2, VTYPE, VILL);
36     target_ulong reserved = FIELD_EX64(s2, VTYPE, RESERVED);
37 
38     if ((sew > cpu->cfg.elen) || vill || (ediv != 0) || (reserved != 0)) {
39         /* only set vill bit. */
40         env->vtype = FIELD_DP64(0, VTYPE, VILL, 1);
41         env->vl = 0;
42         env->vstart = 0;
43         return 0;
44     }
45 
46     vlmax = vext_get_vlmax(cpu, s2);
47     if (s1 <= vlmax) {
48         vl = s1;
49     } else {
50         vl = vlmax;
51     }
52     env->vl = vl;
53     env->vtype = s2;
54     env->vstart = 0;
55     return vl;
56 }
57 
58 /*
59  * Note that vector data is stored in host-endian 64-bit chunks,
60  * so addressing units smaller than that needs a host-endian fixup.
61  */
62 #ifdef HOST_WORDS_BIGENDIAN
63 #define H1(x)   ((x) ^ 7)
64 #define H1_2(x) ((x) ^ 6)
65 #define H1_4(x) ((x) ^ 4)
66 #define H2(x)   ((x) ^ 3)
67 #define H4(x)   ((x) ^ 1)
68 #define H8(x)   ((x))
69 #else
70 #define H1(x)   (x)
71 #define H1_2(x) (x)
72 #define H1_4(x) (x)
73 #define H2(x)   (x)
74 #define H4(x)   (x)
75 #define H8(x)   (x)
76 #endif
77 
78 static inline uint32_t vext_nf(uint32_t desc)
79 {
80     return FIELD_EX32(simd_data(desc), VDATA, NF);
81 }
82 
83 static inline uint32_t vext_mlen(uint32_t desc)
84 {
85     return FIELD_EX32(simd_data(desc), VDATA, MLEN);
86 }
87 
88 static inline uint32_t vext_vm(uint32_t desc)
89 {
90     return FIELD_EX32(simd_data(desc), VDATA, VM);
91 }
92 
93 static inline uint32_t vext_lmul(uint32_t desc)
94 {
95     return FIELD_EX32(simd_data(desc), VDATA, LMUL);
96 }
97 
98 static uint32_t vext_wd(uint32_t desc)
99 {
100     return (simd_data(desc) >> 11) & 0x1;
101 }
102 
103 /*
104  * Get vector group length in bytes. Its range is [64, 2048].
105  *
106  * As simd_desc support at most 256, the max vlen is 512 bits.
107  * So vlen in bytes is encoded as maxsz.
108  */
109 static inline uint32_t vext_maxsz(uint32_t desc)
110 {
111     return simd_maxsz(desc) << vext_lmul(desc);
112 }
113 
114 /*
115  * This function checks watchpoint before real load operation.
116  *
117  * In softmmu mode, the TLB API probe_access is enough for watchpoint check.
118  * In user mode, there is no watchpoint support now.
119  *
120  * It will trigger an exception if there is no mapping in TLB
121  * and page table walk can't fill the TLB entry. Then the guest
122  * software can return here after process the exception or never return.
123  */
124 static void probe_pages(CPURISCVState *env, target_ulong addr,
125                         target_ulong len, uintptr_t ra,
126                         MMUAccessType access_type)
127 {
128     target_ulong pagelen = -(addr | TARGET_PAGE_MASK);
129     target_ulong curlen = MIN(pagelen, len);
130 
131     probe_access(env, addr, curlen, access_type,
132                  cpu_mmu_index(env, false), ra);
133     if (len > curlen) {
134         addr += curlen;
135         curlen = len - curlen;
136         probe_access(env, addr, curlen, access_type,
137                      cpu_mmu_index(env, false), ra);
138     }
139 }
140 
141 #ifdef HOST_WORDS_BIGENDIAN
142 static void vext_clear(void *tail, uint32_t cnt, uint32_t tot)
143 {
144     /*
145      * Split the remaining range to two parts.
146      * The first part is in the last uint64_t unit.
147      * The second part start from the next uint64_t unit.
148      */
149     int part1 = 0, part2 = tot - cnt;
150     if (cnt % 8) {
151         part1 = 8 - (cnt % 8);
152         part2 = tot - cnt - part1;
153         memset((void *)((uintptr_t)tail & ~(7ULL)), 0, part1);
154         memset((void *)(((uintptr_t)tail + 8) & ~(7ULL)), 0, part2);
155     } else {
156         memset(tail, 0, part2);
157     }
158 }
159 #else
160 static void vext_clear(void *tail, uint32_t cnt, uint32_t tot)
161 {
162     memset(tail, 0, tot - cnt);
163 }
164 #endif
165 
166 static void clearb(void *vd, uint32_t idx, uint32_t cnt, uint32_t tot)
167 {
168     int8_t *cur = ((int8_t *)vd + H1(idx));
169     vext_clear(cur, cnt, tot);
170 }
171 
172 static void clearh(void *vd, uint32_t idx, uint32_t cnt, uint32_t tot)
173 {
174     int16_t *cur = ((int16_t *)vd + H2(idx));
175     vext_clear(cur, cnt, tot);
176 }
177 
178 static void clearl(void *vd, uint32_t idx, uint32_t cnt, uint32_t tot)
179 {
180     int32_t *cur = ((int32_t *)vd + H4(idx));
181     vext_clear(cur, cnt, tot);
182 }
183 
184 static void clearq(void *vd, uint32_t idx, uint32_t cnt, uint32_t tot)
185 {
186     int64_t *cur = (int64_t *)vd + idx;
187     vext_clear(cur, cnt, tot);
188 }
189 
190 
191 static inline int vext_elem_mask(void *v0, int mlen, int index)
192 {
193     int idx = (index * mlen) / 64;
194     int pos = (index * mlen) % 64;
195     return (((uint64_t *)v0)[idx] >> pos) & 1;
196 }
197 
198 /* elements operations for load and store */
199 typedef void vext_ldst_elem_fn(CPURISCVState *env, target_ulong addr,
200                                uint32_t idx, void *vd, uintptr_t retaddr);
201 typedef void clear_fn(void *vd, uint32_t idx, uint32_t cnt, uint32_t tot);
202 
203 #define GEN_VEXT_LD_ELEM(NAME, MTYPE, ETYPE, H, LDSUF)     \
204 static void NAME(CPURISCVState *env, abi_ptr addr,         \
205                  uint32_t idx, void *vd, uintptr_t retaddr)\
206 {                                                          \
207     MTYPE data;                                            \
208     ETYPE *cur = ((ETYPE *)vd + H(idx));                   \
209     data = cpu_##LDSUF##_data_ra(env, addr, retaddr);      \
210     *cur = data;                                           \
211 }                                                          \
212 
213 GEN_VEXT_LD_ELEM(ldb_b, int8_t,  int8_t,  H1, ldsb)
214 GEN_VEXT_LD_ELEM(ldb_h, int8_t,  int16_t, H2, ldsb)
215 GEN_VEXT_LD_ELEM(ldb_w, int8_t,  int32_t, H4, ldsb)
216 GEN_VEXT_LD_ELEM(ldb_d, int8_t,  int64_t, H8, ldsb)
217 GEN_VEXT_LD_ELEM(ldh_h, int16_t, int16_t, H2, ldsw)
218 GEN_VEXT_LD_ELEM(ldh_w, int16_t, int32_t, H4, ldsw)
219 GEN_VEXT_LD_ELEM(ldh_d, int16_t, int64_t, H8, ldsw)
220 GEN_VEXT_LD_ELEM(ldw_w, int32_t, int32_t, H4, ldl)
221 GEN_VEXT_LD_ELEM(ldw_d, int32_t, int64_t, H8, ldl)
222 GEN_VEXT_LD_ELEM(lde_b, int8_t,  int8_t,  H1, ldsb)
223 GEN_VEXT_LD_ELEM(lde_h, int16_t, int16_t, H2, ldsw)
224 GEN_VEXT_LD_ELEM(lde_w, int32_t, int32_t, H4, ldl)
225 GEN_VEXT_LD_ELEM(lde_d, int64_t, int64_t, H8, ldq)
226 GEN_VEXT_LD_ELEM(ldbu_b, uint8_t,  uint8_t,  H1, ldub)
227 GEN_VEXT_LD_ELEM(ldbu_h, uint8_t,  uint16_t, H2, ldub)
228 GEN_VEXT_LD_ELEM(ldbu_w, uint8_t,  uint32_t, H4, ldub)
229 GEN_VEXT_LD_ELEM(ldbu_d, uint8_t,  uint64_t, H8, ldub)
230 GEN_VEXT_LD_ELEM(ldhu_h, uint16_t, uint16_t, H2, lduw)
231 GEN_VEXT_LD_ELEM(ldhu_w, uint16_t, uint32_t, H4, lduw)
232 GEN_VEXT_LD_ELEM(ldhu_d, uint16_t, uint64_t, H8, lduw)
233 GEN_VEXT_LD_ELEM(ldwu_w, uint32_t, uint32_t, H4, ldl)
234 GEN_VEXT_LD_ELEM(ldwu_d, uint32_t, uint64_t, H8, ldl)
235 
236 #define GEN_VEXT_ST_ELEM(NAME, ETYPE, H, STSUF)            \
237 static void NAME(CPURISCVState *env, abi_ptr addr,         \
238                  uint32_t idx, void *vd, uintptr_t retaddr)\
239 {                                                          \
240     ETYPE data = *((ETYPE *)vd + H(idx));                  \
241     cpu_##STSUF##_data_ra(env, addr, data, retaddr);       \
242 }
243 
244 GEN_VEXT_ST_ELEM(stb_b, int8_t,  H1, stb)
245 GEN_VEXT_ST_ELEM(stb_h, int16_t, H2, stb)
246 GEN_VEXT_ST_ELEM(stb_w, int32_t, H4, stb)
247 GEN_VEXT_ST_ELEM(stb_d, int64_t, H8, stb)
248 GEN_VEXT_ST_ELEM(sth_h, int16_t, H2, stw)
249 GEN_VEXT_ST_ELEM(sth_w, int32_t, H4, stw)
250 GEN_VEXT_ST_ELEM(sth_d, int64_t, H8, stw)
251 GEN_VEXT_ST_ELEM(stw_w, int32_t, H4, stl)
252 GEN_VEXT_ST_ELEM(stw_d, int64_t, H8, stl)
253 GEN_VEXT_ST_ELEM(ste_b, int8_t,  H1, stb)
254 GEN_VEXT_ST_ELEM(ste_h, int16_t, H2, stw)
255 GEN_VEXT_ST_ELEM(ste_w, int32_t, H4, stl)
256 GEN_VEXT_ST_ELEM(ste_d, int64_t, H8, stq)
257 
258 /*
259  *** stride: access vector element from strided memory
260  */
261 static void
262 vext_ldst_stride(void *vd, void *v0, target_ulong base,
263                  target_ulong stride, CPURISCVState *env,
264                  uint32_t desc, uint32_t vm,
265                  vext_ldst_elem_fn *ldst_elem, clear_fn *clear_elem,
266                  uint32_t esz, uint32_t msz, uintptr_t ra,
267                  MMUAccessType access_type)
268 {
269     uint32_t i, k;
270     uint32_t nf = vext_nf(desc);
271     uint32_t mlen = vext_mlen(desc);
272     uint32_t vlmax = vext_maxsz(desc) / esz;
273 
274     /* probe every access*/
275     for (i = 0; i < env->vl; i++) {
276         if (!vm && !vext_elem_mask(v0, mlen, i)) {
277             continue;
278         }
279         probe_pages(env, base + stride * i, nf * msz, ra, access_type);
280     }
281     /* do real access */
282     for (i = 0; i < env->vl; i++) {
283         k = 0;
284         if (!vm && !vext_elem_mask(v0, mlen, i)) {
285             continue;
286         }
287         while (k < nf) {
288             target_ulong addr = base + stride * i + k * msz;
289             ldst_elem(env, addr, i + k * vlmax, vd, ra);
290             k++;
291         }
292     }
293     /* clear tail elements */
294     if (clear_elem) {
295         for (k = 0; k < nf; k++) {
296             clear_elem(vd, env->vl + k * vlmax, env->vl * esz, vlmax * esz);
297         }
298     }
299 }
300 
301 #define GEN_VEXT_LD_STRIDE(NAME, MTYPE, ETYPE, LOAD_FN, CLEAR_FN)       \
302 void HELPER(NAME)(void *vd, void * v0, target_ulong base,               \
303                   target_ulong stride, CPURISCVState *env,              \
304                   uint32_t desc)                                        \
305 {                                                                       \
306     uint32_t vm = vext_vm(desc);                                        \
307     vext_ldst_stride(vd, v0, base, stride, env, desc, vm, LOAD_FN,      \
308                      CLEAR_FN, sizeof(ETYPE), sizeof(MTYPE),            \
309                      GETPC(), MMU_DATA_LOAD);                           \
310 }
311 
312 GEN_VEXT_LD_STRIDE(vlsb_v_b,  int8_t,   int8_t,   ldb_b,  clearb)
313 GEN_VEXT_LD_STRIDE(vlsb_v_h,  int8_t,   int16_t,  ldb_h,  clearh)
314 GEN_VEXT_LD_STRIDE(vlsb_v_w,  int8_t,   int32_t,  ldb_w,  clearl)
315 GEN_VEXT_LD_STRIDE(vlsb_v_d,  int8_t,   int64_t,  ldb_d,  clearq)
316 GEN_VEXT_LD_STRIDE(vlsh_v_h,  int16_t,  int16_t,  ldh_h,  clearh)
317 GEN_VEXT_LD_STRIDE(vlsh_v_w,  int16_t,  int32_t,  ldh_w,  clearl)
318 GEN_VEXT_LD_STRIDE(vlsh_v_d,  int16_t,  int64_t,  ldh_d,  clearq)
319 GEN_VEXT_LD_STRIDE(vlsw_v_w,  int32_t,  int32_t,  ldw_w,  clearl)
320 GEN_VEXT_LD_STRIDE(vlsw_v_d,  int32_t,  int64_t,  ldw_d,  clearq)
321 GEN_VEXT_LD_STRIDE(vlse_v_b,  int8_t,   int8_t,   lde_b,  clearb)
322 GEN_VEXT_LD_STRIDE(vlse_v_h,  int16_t,  int16_t,  lde_h,  clearh)
323 GEN_VEXT_LD_STRIDE(vlse_v_w,  int32_t,  int32_t,  lde_w,  clearl)
324 GEN_VEXT_LD_STRIDE(vlse_v_d,  int64_t,  int64_t,  lde_d,  clearq)
325 GEN_VEXT_LD_STRIDE(vlsbu_v_b, uint8_t,  uint8_t,  ldbu_b, clearb)
326 GEN_VEXT_LD_STRIDE(vlsbu_v_h, uint8_t,  uint16_t, ldbu_h, clearh)
327 GEN_VEXT_LD_STRIDE(vlsbu_v_w, uint8_t,  uint32_t, ldbu_w, clearl)
328 GEN_VEXT_LD_STRIDE(vlsbu_v_d, uint8_t,  uint64_t, ldbu_d, clearq)
329 GEN_VEXT_LD_STRIDE(vlshu_v_h, uint16_t, uint16_t, ldhu_h, clearh)
330 GEN_VEXT_LD_STRIDE(vlshu_v_w, uint16_t, uint32_t, ldhu_w, clearl)
331 GEN_VEXT_LD_STRIDE(vlshu_v_d, uint16_t, uint64_t, ldhu_d, clearq)
332 GEN_VEXT_LD_STRIDE(vlswu_v_w, uint32_t, uint32_t, ldwu_w, clearl)
333 GEN_VEXT_LD_STRIDE(vlswu_v_d, uint32_t, uint64_t, ldwu_d, clearq)
334 
335 #define GEN_VEXT_ST_STRIDE(NAME, MTYPE, ETYPE, STORE_FN)                \
336 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
337                   target_ulong stride, CPURISCVState *env,              \
338                   uint32_t desc)                                        \
339 {                                                                       \
340     uint32_t vm = vext_vm(desc);                                        \
341     vext_ldst_stride(vd, v0, base, stride, env, desc, vm, STORE_FN,     \
342                      NULL, sizeof(ETYPE), sizeof(MTYPE),                \
343                      GETPC(), MMU_DATA_STORE);                          \
344 }
345 
346 GEN_VEXT_ST_STRIDE(vssb_v_b, int8_t,  int8_t,  stb_b)
347 GEN_VEXT_ST_STRIDE(vssb_v_h, int8_t,  int16_t, stb_h)
348 GEN_VEXT_ST_STRIDE(vssb_v_w, int8_t,  int32_t, stb_w)
349 GEN_VEXT_ST_STRIDE(vssb_v_d, int8_t,  int64_t, stb_d)
350 GEN_VEXT_ST_STRIDE(vssh_v_h, int16_t, int16_t, sth_h)
351 GEN_VEXT_ST_STRIDE(vssh_v_w, int16_t, int32_t, sth_w)
352 GEN_VEXT_ST_STRIDE(vssh_v_d, int16_t, int64_t, sth_d)
353 GEN_VEXT_ST_STRIDE(vssw_v_w, int32_t, int32_t, stw_w)
354 GEN_VEXT_ST_STRIDE(vssw_v_d, int32_t, int64_t, stw_d)
355 GEN_VEXT_ST_STRIDE(vsse_v_b, int8_t,  int8_t,  ste_b)
356 GEN_VEXT_ST_STRIDE(vsse_v_h, int16_t, int16_t, ste_h)
357 GEN_VEXT_ST_STRIDE(vsse_v_w, int32_t, int32_t, ste_w)
358 GEN_VEXT_ST_STRIDE(vsse_v_d, int64_t, int64_t, ste_d)
359 
360 /*
361  *** unit-stride: access elements stored contiguously in memory
362  */
363 
364 /* unmasked unit-stride load and store operation*/
365 static void
366 vext_ldst_us(void *vd, target_ulong base, CPURISCVState *env, uint32_t desc,
367              vext_ldst_elem_fn *ldst_elem, clear_fn *clear_elem,
368              uint32_t esz, uint32_t msz, uintptr_t ra,
369              MMUAccessType access_type)
370 {
371     uint32_t i, k;
372     uint32_t nf = vext_nf(desc);
373     uint32_t vlmax = vext_maxsz(desc) / esz;
374 
375     /* probe every access */
376     probe_pages(env, base, env->vl * nf * msz, ra, access_type);
377     /* load bytes from guest memory */
378     for (i = 0; i < env->vl; i++) {
379         k = 0;
380         while (k < nf) {
381             target_ulong addr = base + (i * nf + k) * msz;
382             ldst_elem(env, addr, i + k * vlmax, vd, ra);
383             k++;
384         }
385     }
386     /* clear tail elements */
387     if (clear_elem) {
388         for (k = 0; k < nf; k++) {
389             clear_elem(vd, env->vl + k * vlmax, env->vl * esz, vlmax * esz);
390         }
391     }
392 }
393 
394 /*
395  * masked unit-stride load and store operation will be a special case of stride,
396  * stride = NF * sizeof (MTYPE)
397  */
398 
399 #define GEN_VEXT_LD_US(NAME, MTYPE, ETYPE, LOAD_FN, CLEAR_FN)           \
400 void HELPER(NAME##_mask)(void *vd, void *v0, target_ulong base,         \
401                          CPURISCVState *env, uint32_t desc)             \
402 {                                                                       \
403     uint32_t stride = vext_nf(desc) * sizeof(MTYPE);                    \
404     vext_ldst_stride(vd, v0, base, stride, env, desc, false, LOAD_FN,   \
405                      CLEAR_FN, sizeof(ETYPE), sizeof(MTYPE),            \
406                      GETPC(), MMU_DATA_LOAD);                           \
407 }                                                                       \
408                                                                         \
409 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
410                   CPURISCVState *env, uint32_t desc)                    \
411 {                                                                       \
412     vext_ldst_us(vd, base, env, desc, LOAD_FN, CLEAR_FN,                \
413                  sizeof(ETYPE), sizeof(MTYPE), GETPC(), MMU_DATA_LOAD); \
414 }
415 
416 GEN_VEXT_LD_US(vlb_v_b,  int8_t,   int8_t,   ldb_b,  clearb)
417 GEN_VEXT_LD_US(vlb_v_h,  int8_t,   int16_t,  ldb_h,  clearh)
418 GEN_VEXT_LD_US(vlb_v_w,  int8_t,   int32_t,  ldb_w,  clearl)
419 GEN_VEXT_LD_US(vlb_v_d,  int8_t,   int64_t,  ldb_d,  clearq)
420 GEN_VEXT_LD_US(vlh_v_h,  int16_t,  int16_t,  ldh_h,  clearh)
421 GEN_VEXT_LD_US(vlh_v_w,  int16_t,  int32_t,  ldh_w,  clearl)
422 GEN_VEXT_LD_US(vlh_v_d,  int16_t,  int64_t,  ldh_d,  clearq)
423 GEN_VEXT_LD_US(vlw_v_w,  int32_t,  int32_t,  ldw_w,  clearl)
424 GEN_VEXT_LD_US(vlw_v_d,  int32_t,  int64_t,  ldw_d,  clearq)
425 GEN_VEXT_LD_US(vle_v_b,  int8_t,   int8_t,   lde_b,  clearb)
426 GEN_VEXT_LD_US(vle_v_h,  int16_t,  int16_t,  lde_h,  clearh)
427 GEN_VEXT_LD_US(vle_v_w,  int32_t,  int32_t,  lde_w,  clearl)
428 GEN_VEXT_LD_US(vle_v_d,  int64_t,  int64_t,  lde_d,  clearq)
429 GEN_VEXT_LD_US(vlbu_v_b, uint8_t,  uint8_t,  ldbu_b, clearb)
430 GEN_VEXT_LD_US(vlbu_v_h, uint8_t,  uint16_t, ldbu_h, clearh)
431 GEN_VEXT_LD_US(vlbu_v_w, uint8_t,  uint32_t, ldbu_w, clearl)
432 GEN_VEXT_LD_US(vlbu_v_d, uint8_t,  uint64_t, ldbu_d, clearq)
433 GEN_VEXT_LD_US(vlhu_v_h, uint16_t, uint16_t, ldhu_h, clearh)
434 GEN_VEXT_LD_US(vlhu_v_w, uint16_t, uint32_t, ldhu_w, clearl)
435 GEN_VEXT_LD_US(vlhu_v_d, uint16_t, uint64_t, ldhu_d, clearq)
436 GEN_VEXT_LD_US(vlwu_v_w, uint32_t, uint32_t, ldwu_w, clearl)
437 GEN_VEXT_LD_US(vlwu_v_d, uint32_t, uint64_t, ldwu_d, clearq)
438 
439 #define GEN_VEXT_ST_US(NAME, MTYPE, ETYPE, STORE_FN)                    \
440 void HELPER(NAME##_mask)(void *vd, void *v0, target_ulong base,         \
441                          CPURISCVState *env, uint32_t desc)             \
442 {                                                                       \
443     uint32_t stride = vext_nf(desc) * sizeof(MTYPE);                    \
444     vext_ldst_stride(vd, v0, base, stride, env, desc, false, STORE_FN,  \
445                      NULL, sizeof(ETYPE), sizeof(MTYPE),                \
446                      GETPC(), MMU_DATA_STORE);                          \
447 }                                                                       \
448                                                                         \
449 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                \
450                   CPURISCVState *env, uint32_t desc)                    \
451 {                                                                       \
452     vext_ldst_us(vd, base, env, desc, STORE_FN, NULL,                   \
453                  sizeof(ETYPE), sizeof(MTYPE), GETPC(), MMU_DATA_STORE);\
454 }
455 
456 GEN_VEXT_ST_US(vsb_v_b, int8_t,  int8_t , stb_b)
457 GEN_VEXT_ST_US(vsb_v_h, int8_t,  int16_t, stb_h)
458 GEN_VEXT_ST_US(vsb_v_w, int8_t,  int32_t, stb_w)
459 GEN_VEXT_ST_US(vsb_v_d, int8_t,  int64_t, stb_d)
460 GEN_VEXT_ST_US(vsh_v_h, int16_t, int16_t, sth_h)
461 GEN_VEXT_ST_US(vsh_v_w, int16_t, int32_t, sth_w)
462 GEN_VEXT_ST_US(vsh_v_d, int16_t, int64_t, sth_d)
463 GEN_VEXT_ST_US(vsw_v_w, int32_t, int32_t, stw_w)
464 GEN_VEXT_ST_US(vsw_v_d, int32_t, int64_t, stw_d)
465 GEN_VEXT_ST_US(vse_v_b, int8_t,  int8_t , ste_b)
466 GEN_VEXT_ST_US(vse_v_h, int16_t, int16_t, ste_h)
467 GEN_VEXT_ST_US(vse_v_w, int32_t, int32_t, ste_w)
468 GEN_VEXT_ST_US(vse_v_d, int64_t, int64_t, ste_d)
469 
470 /*
471  *** index: access vector element from indexed memory
472  */
473 typedef target_ulong vext_get_index_addr(target_ulong base,
474         uint32_t idx, void *vs2);
475 
476 #define GEN_VEXT_GET_INDEX_ADDR(NAME, ETYPE, H)        \
477 static target_ulong NAME(target_ulong base,            \
478                          uint32_t idx, void *vs2)      \
479 {                                                      \
480     return (base + *((ETYPE *)vs2 + H(idx)));          \
481 }
482 
483 GEN_VEXT_GET_INDEX_ADDR(idx_b, int8_t,  H1)
484 GEN_VEXT_GET_INDEX_ADDR(idx_h, int16_t, H2)
485 GEN_VEXT_GET_INDEX_ADDR(idx_w, int32_t, H4)
486 GEN_VEXT_GET_INDEX_ADDR(idx_d, int64_t, H8)
487 
488 static inline void
489 vext_ldst_index(void *vd, void *v0, target_ulong base,
490                 void *vs2, CPURISCVState *env, uint32_t desc,
491                 vext_get_index_addr get_index_addr,
492                 vext_ldst_elem_fn *ldst_elem,
493                 clear_fn *clear_elem,
494                 uint32_t esz, uint32_t msz, uintptr_t ra,
495                 MMUAccessType access_type)
496 {
497     uint32_t i, k;
498     uint32_t nf = vext_nf(desc);
499     uint32_t vm = vext_vm(desc);
500     uint32_t mlen = vext_mlen(desc);
501     uint32_t vlmax = vext_maxsz(desc) / esz;
502 
503     /* probe every access*/
504     for (i = 0; i < env->vl; i++) {
505         if (!vm && !vext_elem_mask(v0, mlen, i)) {
506             continue;
507         }
508         probe_pages(env, get_index_addr(base, i, vs2), nf * msz, ra,
509                     access_type);
510     }
511     /* load bytes from guest memory */
512     for (i = 0; i < env->vl; i++) {
513         k = 0;
514         if (!vm && !vext_elem_mask(v0, mlen, i)) {
515             continue;
516         }
517         while (k < nf) {
518             abi_ptr addr = get_index_addr(base, i, vs2) + k * msz;
519             ldst_elem(env, addr, i + k * vlmax, vd, ra);
520             k++;
521         }
522     }
523     /* clear tail elements */
524     if (clear_elem) {
525         for (k = 0; k < nf; k++) {
526             clear_elem(vd, env->vl + k * vlmax, env->vl * esz, vlmax * esz);
527         }
528     }
529 }
530 
531 #define GEN_VEXT_LD_INDEX(NAME, MTYPE, ETYPE, INDEX_FN, LOAD_FN, CLEAR_FN) \
532 void HELPER(NAME)(void *vd, void *v0, target_ulong base,                   \
533                   void *vs2, CPURISCVState *env, uint32_t desc)            \
534 {                                                                          \
535     vext_ldst_index(vd, v0, base, vs2, env, desc, INDEX_FN,                \
536                     LOAD_FN, CLEAR_FN, sizeof(ETYPE), sizeof(MTYPE),       \
537                     GETPC(), MMU_DATA_LOAD);                               \
538 }
539 
540 GEN_VEXT_LD_INDEX(vlxb_v_b,  int8_t,   int8_t,   idx_b, ldb_b,  clearb)
541 GEN_VEXT_LD_INDEX(vlxb_v_h,  int8_t,   int16_t,  idx_h, ldb_h,  clearh)
542 GEN_VEXT_LD_INDEX(vlxb_v_w,  int8_t,   int32_t,  idx_w, ldb_w,  clearl)
543 GEN_VEXT_LD_INDEX(vlxb_v_d,  int8_t,   int64_t,  idx_d, ldb_d,  clearq)
544 GEN_VEXT_LD_INDEX(vlxh_v_h,  int16_t,  int16_t,  idx_h, ldh_h,  clearh)
545 GEN_VEXT_LD_INDEX(vlxh_v_w,  int16_t,  int32_t,  idx_w, ldh_w,  clearl)
546 GEN_VEXT_LD_INDEX(vlxh_v_d,  int16_t,  int64_t,  idx_d, ldh_d,  clearq)
547 GEN_VEXT_LD_INDEX(vlxw_v_w,  int32_t,  int32_t,  idx_w, ldw_w,  clearl)
548 GEN_VEXT_LD_INDEX(vlxw_v_d,  int32_t,  int64_t,  idx_d, ldw_d,  clearq)
549 GEN_VEXT_LD_INDEX(vlxe_v_b,  int8_t,   int8_t,   idx_b, lde_b,  clearb)
550 GEN_VEXT_LD_INDEX(vlxe_v_h,  int16_t,  int16_t,  idx_h, lde_h,  clearh)
551 GEN_VEXT_LD_INDEX(vlxe_v_w,  int32_t,  int32_t,  idx_w, lde_w,  clearl)
552 GEN_VEXT_LD_INDEX(vlxe_v_d,  int64_t,  int64_t,  idx_d, lde_d,  clearq)
553 GEN_VEXT_LD_INDEX(vlxbu_v_b, uint8_t,  uint8_t,  idx_b, ldbu_b, clearb)
554 GEN_VEXT_LD_INDEX(vlxbu_v_h, uint8_t,  uint16_t, idx_h, ldbu_h, clearh)
555 GEN_VEXT_LD_INDEX(vlxbu_v_w, uint8_t,  uint32_t, idx_w, ldbu_w, clearl)
556 GEN_VEXT_LD_INDEX(vlxbu_v_d, uint8_t,  uint64_t, idx_d, ldbu_d, clearq)
557 GEN_VEXT_LD_INDEX(vlxhu_v_h, uint16_t, uint16_t, idx_h, ldhu_h, clearh)
558 GEN_VEXT_LD_INDEX(vlxhu_v_w, uint16_t, uint32_t, idx_w, ldhu_w, clearl)
559 GEN_VEXT_LD_INDEX(vlxhu_v_d, uint16_t, uint64_t, idx_d, ldhu_d, clearq)
560 GEN_VEXT_LD_INDEX(vlxwu_v_w, uint32_t, uint32_t, idx_w, ldwu_w, clearl)
561 GEN_VEXT_LD_INDEX(vlxwu_v_d, uint32_t, uint64_t, idx_d, ldwu_d, clearq)
562 
563 #define GEN_VEXT_ST_INDEX(NAME, MTYPE, ETYPE, INDEX_FN, STORE_FN)\
564 void HELPER(NAME)(void *vd, void *v0, target_ulong base,         \
565                   void *vs2, CPURISCVState *env, uint32_t desc)  \
566 {                                                                \
567     vext_ldst_index(vd, v0, base, vs2, env, desc, INDEX_FN,      \
568                     STORE_FN, NULL, sizeof(ETYPE), sizeof(MTYPE),\
569                     GETPC(), MMU_DATA_STORE);                    \
570 }
571 
572 GEN_VEXT_ST_INDEX(vsxb_v_b, int8_t,  int8_t,  idx_b, stb_b)
573 GEN_VEXT_ST_INDEX(vsxb_v_h, int8_t,  int16_t, idx_h, stb_h)
574 GEN_VEXT_ST_INDEX(vsxb_v_w, int8_t,  int32_t, idx_w, stb_w)
575 GEN_VEXT_ST_INDEX(vsxb_v_d, int8_t,  int64_t, idx_d, stb_d)
576 GEN_VEXT_ST_INDEX(vsxh_v_h, int16_t, int16_t, idx_h, sth_h)
577 GEN_VEXT_ST_INDEX(vsxh_v_w, int16_t, int32_t, idx_w, sth_w)
578 GEN_VEXT_ST_INDEX(vsxh_v_d, int16_t, int64_t, idx_d, sth_d)
579 GEN_VEXT_ST_INDEX(vsxw_v_w, int32_t, int32_t, idx_w, stw_w)
580 GEN_VEXT_ST_INDEX(vsxw_v_d, int32_t, int64_t, idx_d, stw_d)
581 GEN_VEXT_ST_INDEX(vsxe_v_b, int8_t,  int8_t,  idx_b, ste_b)
582 GEN_VEXT_ST_INDEX(vsxe_v_h, int16_t, int16_t, idx_h, ste_h)
583 GEN_VEXT_ST_INDEX(vsxe_v_w, int32_t, int32_t, idx_w, ste_w)
584 GEN_VEXT_ST_INDEX(vsxe_v_d, int64_t, int64_t, idx_d, ste_d)
585 
586 /*
587  *** unit-stride fault-only-fisrt load instructions
588  */
589 static inline void
590 vext_ldff(void *vd, void *v0, target_ulong base,
591           CPURISCVState *env, uint32_t desc,
592           vext_ldst_elem_fn *ldst_elem,
593           clear_fn *clear_elem,
594           uint32_t esz, uint32_t msz, uintptr_t ra)
595 {
596     void *host;
597     uint32_t i, k, vl = 0;
598     uint32_t mlen = vext_mlen(desc);
599     uint32_t nf = vext_nf(desc);
600     uint32_t vm = vext_vm(desc);
601     uint32_t vlmax = vext_maxsz(desc) / esz;
602     target_ulong addr, offset, remain;
603 
604     /* probe every access*/
605     for (i = 0; i < env->vl; i++) {
606         if (!vm && !vext_elem_mask(v0, mlen, i)) {
607             continue;
608         }
609         addr = base + nf * i * msz;
610         if (i == 0) {
611             probe_pages(env, addr, nf * msz, ra, MMU_DATA_LOAD);
612         } else {
613             /* if it triggers an exception, no need to check watchpoint */
614             remain = nf * msz;
615             while (remain > 0) {
616                 offset = -(addr | TARGET_PAGE_MASK);
617                 host = tlb_vaddr_to_host(env, addr, MMU_DATA_LOAD,
618                                          cpu_mmu_index(env, false));
619                 if (host) {
620 #ifdef CONFIG_USER_ONLY
621                     if (page_check_range(addr, nf * msz, PAGE_READ) < 0) {
622                         vl = i;
623                         goto ProbeSuccess;
624                     }
625 #else
626                     probe_pages(env, addr, nf * msz, ra, MMU_DATA_LOAD);
627 #endif
628                 } else {
629                     vl = i;
630                     goto ProbeSuccess;
631                 }
632                 if (remain <=  offset) {
633                     break;
634                 }
635                 remain -= offset;
636                 addr += offset;
637             }
638         }
639     }
640 ProbeSuccess:
641     /* load bytes from guest memory */
642     if (vl != 0) {
643         env->vl = vl;
644     }
645     for (i = 0; i < env->vl; i++) {
646         k = 0;
647         if (!vm && !vext_elem_mask(v0, mlen, i)) {
648             continue;
649         }
650         while (k < nf) {
651             target_ulong addr = base + (i * nf + k) * msz;
652             ldst_elem(env, addr, i + k * vlmax, vd, ra);
653             k++;
654         }
655     }
656     /* clear tail elements */
657     if (vl != 0) {
658         return;
659     }
660     for (k = 0; k < nf; k++) {
661         clear_elem(vd, env->vl + k * vlmax, env->vl * esz, vlmax * esz);
662     }
663 }
664 
665 #define GEN_VEXT_LDFF(NAME, MTYPE, ETYPE, LOAD_FN, CLEAR_FN)     \
666 void HELPER(NAME)(void *vd, void *v0, target_ulong base,         \
667                   CPURISCVState *env, uint32_t desc)             \
668 {                                                                \
669     vext_ldff(vd, v0, base, env, desc, LOAD_FN, CLEAR_FN,        \
670               sizeof(ETYPE), sizeof(MTYPE), GETPC());            \
671 }
672 
673 GEN_VEXT_LDFF(vlbff_v_b,  int8_t,   int8_t,   ldb_b,  clearb)
674 GEN_VEXT_LDFF(vlbff_v_h,  int8_t,   int16_t,  ldb_h,  clearh)
675 GEN_VEXT_LDFF(vlbff_v_w,  int8_t,   int32_t,  ldb_w,  clearl)
676 GEN_VEXT_LDFF(vlbff_v_d,  int8_t,   int64_t,  ldb_d,  clearq)
677 GEN_VEXT_LDFF(vlhff_v_h,  int16_t,  int16_t,  ldh_h,  clearh)
678 GEN_VEXT_LDFF(vlhff_v_w,  int16_t,  int32_t,  ldh_w,  clearl)
679 GEN_VEXT_LDFF(vlhff_v_d,  int16_t,  int64_t,  ldh_d,  clearq)
680 GEN_VEXT_LDFF(vlwff_v_w,  int32_t,  int32_t,  ldw_w,  clearl)
681 GEN_VEXT_LDFF(vlwff_v_d,  int32_t,  int64_t,  ldw_d,  clearq)
682 GEN_VEXT_LDFF(vleff_v_b,  int8_t,   int8_t,   lde_b,  clearb)
683 GEN_VEXT_LDFF(vleff_v_h,  int16_t,  int16_t,  lde_h,  clearh)
684 GEN_VEXT_LDFF(vleff_v_w,  int32_t,  int32_t,  lde_w,  clearl)
685 GEN_VEXT_LDFF(vleff_v_d,  int64_t,  int64_t,  lde_d,  clearq)
686 GEN_VEXT_LDFF(vlbuff_v_b, uint8_t,  uint8_t,  ldbu_b, clearb)
687 GEN_VEXT_LDFF(vlbuff_v_h, uint8_t,  uint16_t, ldbu_h, clearh)
688 GEN_VEXT_LDFF(vlbuff_v_w, uint8_t,  uint32_t, ldbu_w, clearl)
689 GEN_VEXT_LDFF(vlbuff_v_d, uint8_t,  uint64_t, ldbu_d, clearq)
690 GEN_VEXT_LDFF(vlhuff_v_h, uint16_t, uint16_t, ldhu_h, clearh)
691 GEN_VEXT_LDFF(vlhuff_v_w, uint16_t, uint32_t, ldhu_w, clearl)
692 GEN_VEXT_LDFF(vlhuff_v_d, uint16_t, uint64_t, ldhu_d, clearq)
693 GEN_VEXT_LDFF(vlwuff_v_w, uint32_t, uint32_t, ldwu_w, clearl)
694 GEN_VEXT_LDFF(vlwuff_v_d, uint32_t, uint64_t, ldwu_d, clearq)
695 
696 /*
697  *** Vector AMO Operations (Zvamo)
698  */
699 typedef void vext_amo_noatomic_fn(void *vs3, target_ulong addr,
700                                   uint32_t wd, uint32_t idx, CPURISCVState *env,
701                                   uintptr_t retaddr);
702 
703 /* no atomic opreation for vector atomic insructions */
704 #define DO_SWAP(N, M) (M)
705 #define DO_AND(N, M)  (N & M)
706 #define DO_XOR(N, M)  (N ^ M)
707 #define DO_OR(N, M)   (N | M)
708 #define DO_ADD(N, M)  (N + M)
709 
710 #define GEN_VEXT_AMO_NOATOMIC_OP(NAME, ESZ, MSZ, H, DO_OP, SUF) \
711 static void                                                     \
712 vext_##NAME##_noatomic_op(void *vs3, target_ulong addr,         \
713                           uint32_t wd, uint32_t idx,            \
714                           CPURISCVState *env, uintptr_t retaddr)\
715 {                                                               \
716     typedef int##ESZ##_t ETYPE;                                 \
717     typedef int##MSZ##_t MTYPE;                                 \
718     typedef uint##MSZ##_t UMTYPE __attribute__((unused));       \
719     ETYPE *pe3 = (ETYPE *)vs3 + H(idx);                         \
720     MTYPE  a = cpu_ld##SUF##_data(env, addr), b = *pe3;         \
721                                                                 \
722     cpu_st##SUF##_data(env, addr, DO_OP(a, b));                 \
723     if (wd) {                                                   \
724         *pe3 = a;                                               \
725     }                                                           \
726 }
727 
728 /* Signed min/max */
729 #define DO_MAX(N, M)  ((N) >= (M) ? (N) : (M))
730 #define DO_MIN(N, M)  ((N) >= (M) ? (M) : (N))
731 
732 /* Unsigned min/max */
733 #define DO_MAXU(N, M) DO_MAX((UMTYPE)N, (UMTYPE)M)
734 #define DO_MINU(N, M) DO_MIN((UMTYPE)N, (UMTYPE)M)
735 
736 GEN_VEXT_AMO_NOATOMIC_OP(vamoswapw_v_w, 32, 32, H4, DO_SWAP, l)
737 GEN_VEXT_AMO_NOATOMIC_OP(vamoaddw_v_w,  32, 32, H4, DO_ADD,  l)
738 GEN_VEXT_AMO_NOATOMIC_OP(vamoxorw_v_w,  32, 32, H4, DO_XOR,  l)
739 GEN_VEXT_AMO_NOATOMIC_OP(vamoandw_v_w,  32, 32, H4, DO_AND,  l)
740 GEN_VEXT_AMO_NOATOMIC_OP(vamoorw_v_w,   32, 32, H4, DO_OR,   l)
741 GEN_VEXT_AMO_NOATOMIC_OP(vamominw_v_w,  32, 32, H4, DO_MIN,  l)
742 GEN_VEXT_AMO_NOATOMIC_OP(vamomaxw_v_w,  32, 32, H4, DO_MAX,  l)
743 GEN_VEXT_AMO_NOATOMIC_OP(vamominuw_v_w, 32, 32, H4, DO_MINU, l)
744 GEN_VEXT_AMO_NOATOMIC_OP(vamomaxuw_v_w, 32, 32, H4, DO_MAXU, l)
745 #ifdef TARGET_RISCV64
746 GEN_VEXT_AMO_NOATOMIC_OP(vamoswapw_v_d, 64, 32, H8, DO_SWAP, l)
747 GEN_VEXT_AMO_NOATOMIC_OP(vamoswapd_v_d, 64, 64, H8, DO_SWAP, q)
748 GEN_VEXT_AMO_NOATOMIC_OP(vamoaddw_v_d,  64, 32, H8, DO_ADD,  l)
749 GEN_VEXT_AMO_NOATOMIC_OP(vamoaddd_v_d,  64, 64, H8, DO_ADD,  q)
750 GEN_VEXT_AMO_NOATOMIC_OP(vamoxorw_v_d,  64, 32, H8, DO_XOR,  l)
751 GEN_VEXT_AMO_NOATOMIC_OP(vamoxord_v_d,  64, 64, H8, DO_XOR,  q)
752 GEN_VEXT_AMO_NOATOMIC_OP(vamoandw_v_d,  64, 32, H8, DO_AND,  l)
753 GEN_VEXT_AMO_NOATOMIC_OP(vamoandd_v_d,  64, 64, H8, DO_AND,  q)
754 GEN_VEXT_AMO_NOATOMIC_OP(vamoorw_v_d,   64, 32, H8, DO_OR,   l)
755 GEN_VEXT_AMO_NOATOMIC_OP(vamoord_v_d,   64, 64, H8, DO_OR,   q)
756 GEN_VEXT_AMO_NOATOMIC_OP(vamominw_v_d,  64, 32, H8, DO_MIN,  l)
757 GEN_VEXT_AMO_NOATOMIC_OP(vamomind_v_d,  64, 64, H8, DO_MIN,  q)
758 GEN_VEXT_AMO_NOATOMIC_OP(vamomaxw_v_d,  64, 32, H8, DO_MAX,  l)
759 GEN_VEXT_AMO_NOATOMIC_OP(vamomaxd_v_d,  64, 64, H8, DO_MAX,  q)
760 GEN_VEXT_AMO_NOATOMIC_OP(vamominuw_v_d, 64, 32, H8, DO_MINU, l)
761 GEN_VEXT_AMO_NOATOMIC_OP(vamominud_v_d, 64, 64, H8, DO_MINU, q)
762 GEN_VEXT_AMO_NOATOMIC_OP(vamomaxuw_v_d, 64, 32, H8, DO_MAXU, l)
763 GEN_VEXT_AMO_NOATOMIC_OP(vamomaxud_v_d, 64, 64, H8, DO_MAXU, q)
764 #endif
765 
766 static inline void
767 vext_amo_noatomic(void *vs3, void *v0, target_ulong base,
768                   void *vs2, CPURISCVState *env, uint32_t desc,
769                   vext_get_index_addr get_index_addr,
770                   vext_amo_noatomic_fn *noatomic_op,
771                   clear_fn *clear_elem,
772                   uint32_t esz, uint32_t msz, uintptr_t ra)
773 {
774     uint32_t i;
775     target_long addr;
776     uint32_t wd = vext_wd(desc);
777     uint32_t vm = vext_vm(desc);
778     uint32_t mlen = vext_mlen(desc);
779     uint32_t vlmax = vext_maxsz(desc) / esz;
780 
781     for (i = 0; i < env->vl; i++) {
782         if (!vm && !vext_elem_mask(v0, mlen, i)) {
783             continue;
784         }
785         probe_pages(env, get_index_addr(base, i, vs2), msz, ra, MMU_DATA_LOAD);
786         probe_pages(env, get_index_addr(base, i, vs2), msz, ra, MMU_DATA_STORE);
787     }
788     for (i = 0; i < env->vl; i++) {
789         if (!vm && !vext_elem_mask(v0, mlen, i)) {
790             continue;
791         }
792         addr = get_index_addr(base, i, vs2);
793         noatomic_op(vs3, addr, wd, i, env, ra);
794     }
795     clear_elem(vs3, env->vl, env->vl * esz, vlmax * esz);
796 }
797 
798 #define GEN_VEXT_AMO(NAME, MTYPE, ETYPE, INDEX_FN, CLEAR_FN)    \
799 void HELPER(NAME)(void *vs3, void *v0, target_ulong base,       \
800                   void *vs2, CPURISCVState *env, uint32_t desc) \
801 {                                                               \
802     vext_amo_noatomic(vs3, v0, base, vs2, env, desc,            \
803                       INDEX_FN, vext_##NAME##_noatomic_op,      \
804                       CLEAR_FN, sizeof(ETYPE), sizeof(MTYPE),   \
805                       GETPC());                                 \
806 }
807 
808 #ifdef TARGET_RISCV64
809 GEN_VEXT_AMO(vamoswapw_v_d, int32_t,  int64_t,  idx_d, clearq)
810 GEN_VEXT_AMO(vamoswapd_v_d, int64_t,  int64_t,  idx_d, clearq)
811 GEN_VEXT_AMO(vamoaddw_v_d,  int32_t,  int64_t,  idx_d, clearq)
812 GEN_VEXT_AMO(vamoaddd_v_d,  int64_t,  int64_t,  idx_d, clearq)
813 GEN_VEXT_AMO(vamoxorw_v_d,  int32_t,  int64_t,  idx_d, clearq)
814 GEN_VEXT_AMO(vamoxord_v_d,  int64_t,  int64_t,  idx_d, clearq)
815 GEN_VEXT_AMO(vamoandw_v_d,  int32_t,  int64_t,  idx_d, clearq)
816 GEN_VEXT_AMO(vamoandd_v_d,  int64_t,  int64_t,  idx_d, clearq)
817 GEN_VEXT_AMO(vamoorw_v_d,   int32_t,  int64_t,  idx_d, clearq)
818 GEN_VEXT_AMO(vamoord_v_d,   int64_t,  int64_t,  idx_d, clearq)
819 GEN_VEXT_AMO(vamominw_v_d,  int32_t,  int64_t,  idx_d, clearq)
820 GEN_VEXT_AMO(vamomind_v_d,  int64_t,  int64_t,  idx_d, clearq)
821 GEN_VEXT_AMO(vamomaxw_v_d,  int32_t,  int64_t,  idx_d, clearq)
822 GEN_VEXT_AMO(vamomaxd_v_d,  int64_t,  int64_t,  idx_d, clearq)
823 GEN_VEXT_AMO(vamominuw_v_d, uint32_t, uint64_t, idx_d, clearq)
824 GEN_VEXT_AMO(vamominud_v_d, uint64_t, uint64_t, idx_d, clearq)
825 GEN_VEXT_AMO(vamomaxuw_v_d, uint32_t, uint64_t, idx_d, clearq)
826 GEN_VEXT_AMO(vamomaxud_v_d, uint64_t, uint64_t, idx_d, clearq)
827 #endif
828 GEN_VEXT_AMO(vamoswapw_v_w, int32_t,  int32_t,  idx_w, clearl)
829 GEN_VEXT_AMO(vamoaddw_v_w,  int32_t,  int32_t,  idx_w, clearl)
830 GEN_VEXT_AMO(vamoxorw_v_w,  int32_t,  int32_t,  idx_w, clearl)
831 GEN_VEXT_AMO(vamoandw_v_w,  int32_t,  int32_t,  idx_w, clearl)
832 GEN_VEXT_AMO(vamoorw_v_w,   int32_t,  int32_t,  idx_w, clearl)
833 GEN_VEXT_AMO(vamominw_v_w,  int32_t,  int32_t,  idx_w, clearl)
834 GEN_VEXT_AMO(vamomaxw_v_w,  int32_t,  int32_t,  idx_w, clearl)
835 GEN_VEXT_AMO(vamominuw_v_w, uint32_t, uint32_t, idx_w, clearl)
836 GEN_VEXT_AMO(vamomaxuw_v_w, uint32_t, uint32_t, idx_w, clearl)
837 
838 /*
839  *** Vector Integer Arithmetic Instructions
840  */
841 
842 /* expand macro args before macro */
843 #define RVVCALL(macro, ...)  macro(__VA_ARGS__)
844 
845 /* (TD, T1, T2, TX1, TX2) */
846 #define OP_SSS_B int8_t, int8_t, int8_t, int8_t, int8_t
847 #define OP_SSS_H int16_t, int16_t, int16_t, int16_t, int16_t
848 #define OP_SSS_W int32_t, int32_t, int32_t, int32_t, int32_t
849 #define OP_SSS_D int64_t, int64_t, int64_t, int64_t, int64_t
850 
851 /* operation of two vector elements */
852 typedef void opivv2_fn(void *vd, void *vs1, void *vs2, int i);
853 
854 #define OPIVV2(NAME, TD, T1, T2, TX1, TX2, HD, HS1, HS2, OP)    \
855 static void do_##NAME(void *vd, void *vs1, void *vs2, int i)    \
856 {                                                               \
857     TX1 s1 = *((T1 *)vs1 + HS1(i));                             \
858     TX2 s2 = *((T2 *)vs2 + HS2(i));                             \
859     *((TD *)vd + HD(i)) = OP(s2, s1);                           \
860 }
861 #define DO_SUB(N, M) (N - M)
862 #define DO_RSUB(N, M) (M - N)
863 
864 RVVCALL(OPIVV2, vadd_vv_b, OP_SSS_B, H1, H1, H1, DO_ADD)
865 RVVCALL(OPIVV2, vadd_vv_h, OP_SSS_H, H2, H2, H2, DO_ADD)
866 RVVCALL(OPIVV2, vadd_vv_w, OP_SSS_W, H4, H4, H4, DO_ADD)
867 RVVCALL(OPIVV2, vadd_vv_d, OP_SSS_D, H8, H8, H8, DO_ADD)
868 RVVCALL(OPIVV2, vsub_vv_b, OP_SSS_B, H1, H1, H1, DO_SUB)
869 RVVCALL(OPIVV2, vsub_vv_h, OP_SSS_H, H2, H2, H2, DO_SUB)
870 RVVCALL(OPIVV2, vsub_vv_w, OP_SSS_W, H4, H4, H4, DO_SUB)
871 RVVCALL(OPIVV2, vsub_vv_d, OP_SSS_D, H8, H8, H8, DO_SUB)
872 
873 static void do_vext_vv(void *vd, void *v0, void *vs1, void *vs2,
874                        CPURISCVState *env, uint32_t desc,
875                        uint32_t esz, uint32_t dsz,
876                        opivv2_fn *fn, clear_fn *clearfn)
877 {
878     uint32_t vlmax = vext_maxsz(desc) / esz;
879     uint32_t mlen = vext_mlen(desc);
880     uint32_t vm = vext_vm(desc);
881     uint32_t vl = env->vl;
882     uint32_t i;
883 
884     for (i = 0; i < vl; i++) {
885         if (!vm && !vext_elem_mask(v0, mlen, i)) {
886             continue;
887         }
888         fn(vd, vs1, vs2, i);
889     }
890     clearfn(vd, vl, vl * dsz,  vlmax * dsz);
891 }
892 
893 /* generate the helpers for OPIVV */
894 #define GEN_VEXT_VV(NAME, ESZ, DSZ, CLEAR_FN)             \
895 void HELPER(NAME)(void *vd, void *v0, void *vs1,          \
896                   void *vs2, CPURISCVState *env,          \
897                   uint32_t desc)                          \
898 {                                                         \
899     do_vext_vv(vd, v0, vs1, vs2, env, desc, ESZ, DSZ,     \
900                do_##NAME, CLEAR_FN);                      \
901 }
902 
903 GEN_VEXT_VV(vadd_vv_b, 1, 1, clearb)
904 GEN_VEXT_VV(vadd_vv_h, 2, 2, clearh)
905 GEN_VEXT_VV(vadd_vv_w, 4, 4, clearl)
906 GEN_VEXT_VV(vadd_vv_d, 8, 8, clearq)
907 GEN_VEXT_VV(vsub_vv_b, 1, 1, clearb)
908 GEN_VEXT_VV(vsub_vv_h, 2, 2, clearh)
909 GEN_VEXT_VV(vsub_vv_w, 4, 4, clearl)
910 GEN_VEXT_VV(vsub_vv_d, 8, 8, clearq)
911 
912 typedef void opivx2_fn(void *vd, target_long s1, void *vs2, int i);
913 
914 /*
915  * (T1)s1 gives the real operator type.
916  * (TX1)(T1)s1 expands the operator type of widen or narrow operations.
917  */
918 #define OPIVX2(NAME, TD, T1, T2, TX1, TX2, HD, HS2, OP)             \
919 static void do_##NAME(void *vd, target_long s1, void *vs2, int i)   \
920 {                                                                   \
921     TX2 s2 = *((T2 *)vs2 + HS2(i));                                 \
922     *((TD *)vd + HD(i)) = OP(s2, (TX1)(T1)s1);                      \
923 }
924 
925 RVVCALL(OPIVX2, vadd_vx_b, OP_SSS_B, H1, H1, DO_ADD)
926 RVVCALL(OPIVX2, vadd_vx_h, OP_SSS_H, H2, H2, DO_ADD)
927 RVVCALL(OPIVX2, vadd_vx_w, OP_SSS_W, H4, H4, DO_ADD)
928 RVVCALL(OPIVX2, vadd_vx_d, OP_SSS_D, H8, H8, DO_ADD)
929 RVVCALL(OPIVX2, vsub_vx_b, OP_SSS_B, H1, H1, DO_SUB)
930 RVVCALL(OPIVX2, vsub_vx_h, OP_SSS_H, H2, H2, DO_SUB)
931 RVVCALL(OPIVX2, vsub_vx_w, OP_SSS_W, H4, H4, DO_SUB)
932 RVVCALL(OPIVX2, vsub_vx_d, OP_SSS_D, H8, H8, DO_SUB)
933 RVVCALL(OPIVX2, vrsub_vx_b, OP_SSS_B, H1, H1, DO_RSUB)
934 RVVCALL(OPIVX2, vrsub_vx_h, OP_SSS_H, H2, H2, DO_RSUB)
935 RVVCALL(OPIVX2, vrsub_vx_w, OP_SSS_W, H4, H4, DO_RSUB)
936 RVVCALL(OPIVX2, vrsub_vx_d, OP_SSS_D, H8, H8, DO_RSUB)
937 
938 static void do_vext_vx(void *vd, void *v0, target_long s1, void *vs2,
939                        CPURISCVState *env, uint32_t desc,
940                        uint32_t esz, uint32_t dsz,
941                        opivx2_fn fn, clear_fn *clearfn)
942 {
943     uint32_t vlmax = vext_maxsz(desc) / esz;
944     uint32_t mlen = vext_mlen(desc);
945     uint32_t vm = vext_vm(desc);
946     uint32_t vl = env->vl;
947     uint32_t i;
948 
949     for (i = 0; i < vl; i++) {
950         if (!vm && !vext_elem_mask(v0, mlen, i)) {
951             continue;
952         }
953         fn(vd, s1, vs2, i);
954     }
955     clearfn(vd, vl, vl * dsz,  vlmax * dsz);
956 }
957 
958 /* generate the helpers for OPIVX */
959 #define GEN_VEXT_VX(NAME, ESZ, DSZ, CLEAR_FN)             \
960 void HELPER(NAME)(void *vd, void *v0, target_ulong s1,    \
961                   void *vs2, CPURISCVState *env,          \
962                   uint32_t desc)                          \
963 {                                                         \
964     do_vext_vx(vd, v0, s1, vs2, env, desc, ESZ, DSZ,      \
965                do_##NAME, CLEAR_FN);                      \
966 }
967 
968 GEN_VEXT_VX(vadd_vx_b, 1, 1, clearb)
969 GEN_VEXT_VX(vadd_vx_h, 2, 2, clearh)
970 GEN_VEXT_VX(vadd_vx_w, 4, 4, clearl)
971 GEN_VEXT_VX(vadd_vx_d, 8, 8, clearq)
972 GEN_VEXT_VX(vsub_vx_b, 1, 1, clearb)
973 GEN_VEXT_VX(vsub_vx_h, 2, 2, clearh)
974 GEN_VEXT_VX(vsub_vx_w, 4, 4, clearl)
975 GEN_VEXT_VX(vsub_vx_d, 8, 8, clearq)
976 GEN_VEXT_VX(vrsub_vx_b, 1, 1, clearb)
977 GEN_VEXT_VX(vrsub_vx_h, 2, 2, clearh)
978 GEN_VEXT_VX(vrsub_vx_w, 4, 4, clearl)
979 GEN_VEXT_VX(vrsub_vx_d, 8, 8, clearq)
980 
981 void HELPER(vec_rsubs8)(void *d, void *a, uint64_t b, uint32_t desc)
982 {
983     intptr_t oprsz = simd_oprsz(desc);
984     intptr_t i;
985 
986     for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
987         *(uint8_t *)(d + i) = (uint8_t)b - *(uint8_t *)(a + i);
988     }
989 }
990 
991 void HELPER(vec_rsubs16)(void *d, void *a, uint64_t b, uint32_t desc)
992 {
993     intptr_t oprsz = simd_oprsz(desc);
994     intptr_t i;
995 
996     for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
997         *(uint16_t *)(d + i) = (uint16_t)b - *(uint16_t *)(a + i);
998     }
999 }
1000 
1001 void HELPER(vec_rsubs32)(void *d, void *a, uint64_t b, uint32_t desc)
1002 {
1003     intptr_t oprsz = simd_oprsz(desc);
1004     intptr_t i;
1005 
1006     for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
1007         *(uint32_t *)(d + i) = (uint32_t)b - *(uint32_t *)(a + i);
1008     }
1009 }
1010 
1011 void HELPER(vec_rsubs64)(void *d, void *a, uint64_t b, uint32_t desc)
1012 {
1013     intptr_t oprsz = simd_oprsz(desc);
1014     intptr_t i;
1015 
1016     for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
1017         *(uint64_t *)(d + i) = b - *(uint64_t *)(a + i);
1018     }
1019 }
1020 
1021 /* Vector Widening Integer Add/Subtract */
1022 #define WOP_UUU_B uint16_t, uint8_t, uint8_t, uint16_t, uint16_t
1023 #define WOP_UUU_H uint32_t, uint16_t, uint16_t, uint32_t, uint32_t
1024 #define WOP_UUU_W uint64_t, uint32_t, uint32_t, uint64_t, uint64_t
1025 #define WOP_SSS_B int16_t, int8_t, int8_t, int16_t, int16_t
1026 #define WOP_SSS_H int32_t, int16_t, int16_t, int32_t, int32_t
1027 #define WOP_SSS_W int64_t, int32_t, int32_t, int64_t, int64_t
1028 #define WOP_WUUU_B  uint16_t, uint8_t, uint16_t, uint16_t, uint16_t
1029 #define WOP_WUUU_H  uint32_t, uint16_t, uint32_t, uint32_t, uint32_t
1030 #define WOP_WUUU_W  uint64_t, uint32_t, uint64_t, uint64_t, uint64_t
1031 #define WOP_WSSS_B  int16_t, int8_t, int16_t, int16_t, int16_t
1032 #define WOP_WSSS_H  int32_t, int16_t, int32_t, int32_t, int32_t
1033 #define WOP_WSSS_W  int64_t, int32_t, int64_t, int64_t, int64_t
1034 RVVCALL(OPIVV2, vwaddu_vv_b, WOP_UUU_B, H2, H1, H1, DO_ADD)
1035 RVVCALL(OPIVV2, vwaddu_vv_h, WOP_UUU_H, H4, H2, H2, DO_ADD)
1036 RVVCALL(OPIVV2, vwaddu_vv_w, WOP_UUU_W, H8, H4, H4, DO_ADD)
1037 RVVCALL(OPIVV2, vwsubu_vv_b, WOP_UUU_B, H2, H1, H1, DO_SUB)
1038 RVVCALL(OPIVV2, vwsubu_vv_h, WOP_UUU_H, H4, H2, H2, DO_SUB)
1039 RVVCALL(OPIVV2, vwsubu_vv_w, WOP_UUU_W, H8, H4, H4, DO_SUB)
1040 RVVCALL(OPIVV2, vwadd_vv_b, WOP_SSS_B, H2, H1, H1, DO_ADD)
1041 RVVCALL(OPIVV2, vwadd_vv_h, WOP_SSS_H, H4, H2, H2, DO_ADD)
1042 RVVCALL(OPIVV2, vwadd_vv_w, WOP_SSS_W, H8, H4, H4, DO_ADD)
1043 RVVCALL(OPIVV2, vwsub_vv_b, WOP_SSS_B, H2, H1, H1, DO_SUB)
1044 RVVCALL(OPIVV2, vwsub_vv_h, WOP_SSS_H, H4, H2, H2, DO_SUB)
1045 RVVCALL(OPIVV2, vwsub_vv_w, WOP_SSS_W, H8, H4, H4, DO_SUB)
1046 RVVCALL(OPIVV2, vwaddu_wv_b, WOP_WUUU_B, H2, H1, H1, DO_ADD)
1047 RVVCALL(OPIVV2, vwaddu_wv_h, WOP_WUUU_H, H4, H2, H2, DO_ADD)
1048 RVVCALL(OPIVV2, vwaddu_wv_w, WOP_WUUU_W, H8, H4, H4, DO_ADD)
1049 RVVCALL(OPIVV2, vwsubu_wv_b, WOP_WUUU_B, H2, H1, H1, DO_SUB)
1050 RVVCALL(OPIVV2, vwsubu_wv_h, WOP_WUUU_H, H4, H2, H2, DO_SUB)
1051 RVVCALL(OPIVV2, vwsubu_wv_w, WOP_WUUU_W, H8, H4, H4, DO_SUB)
1052 RVVCALL(OPIVV2, vwadd_wv_b, WOP_WSSS_B, H2, H1, H1, DO_ADD)
1053 RVVCALL(OPIVV2, vwadd_wv_h, WOP_WSSS_H, H4, H2, H2, DO_ADD)
1054 RVVCALL(OPIVV2, vwadd_wv_w, WOP_WSSS_W, H8, H4, H4, DO_ADD)
1055 RVVCALL(OPIVV2, vwsub_wv_b, WOP_WSSS_B, H2, H1, H1, DO_SUB)
1056 RVVCALL(OPIVV2, vwsub_wv_h, WOP_WSSS_H, H4, H2, H2, DO_SUB)
1057 RVVCALL(OPIVV2, vwsub_wv_w, WOP_WSSS_W, H8, H4, H4, DO_SUB)
1058 GEN_VEXT_VV(vwaddu_vv_b, 1, 2, clearh)
1059 GEN_VEXT_VV(vwaddu_vv_h, 2, 4, clearl)
1060 GEN_VEXT_VV(vwaddu_vv_w, 4, 8, clearq)
1061 GEN_VEXT_VV(vwsubu_vv_b, 1, 2, clearh)
1062 GEN_VEXT_VV(vwsubu_vv_h, 2, 4, clearl)
1063 GEN_VEXT_VV(vwsubu_vv_w, 4, 8, clearq)
1064 GEN_VEXT_VV(vwadd_vv_b, 1, 2, clearh)
1065 GEN_VEXT_VV(vwadd_vv_h, 2, 4, clearl)
1066 GEN_VEXT_VV(vwadd_vv_w, 4, 8, clearq)
1067 GEN_VEXT_VV(vwsub_vv_b, 1, 2, clearh)
1068 GEN_VEXT_VV(vwsub_vv_h, 2, 4, clearl)
1069 GEN_VEXT_VV(vwsub_vv_w, 4, 8, clearq)
1070 GEN_VEXT_VV(vwaddu_wv_b, 1, 2, clearh)
1071 GEN_VEXT_VV(vwaddu_wv_h, 2, 4, clearl)
1072 GEN_VEXT_VV(vwaddu_wv_w, 4, 8, clearq)
1073 GEN_VEXT_VV(vwsubu_wv_b, 1, 2, clearh)
1074 GEN_VEXT_VV(vwsubu_wv_h, 2, 4, clearl)
1075 GEN_VEXT_VV(vwsubu_wv_w, 4, 8, clearq)
1076 GEN_VEXT_VV(vwadd_wv_b, 1, 2, clearh)
1077 GEN_VEXT_VV(vwadd_wv_h, 2, 4, clearl)
1078 GEN_VEXT_VV(vwadd_wv_w, 4, 8, clearq)
1079 GEN_VEXT_VV(vwsub_wv_b, 1, 2, clearh)
1080 GEN_VEXT_VV(vwsub_wv_h, 2, 4, clearl)
1081 GEN_VEXT_VV(vwsub_wv_w, 4, 8, clearq)
1082 
1083 RVVCALL(OPIVX2, vwaddu_vx_b, WOP_UUU_B, H2, H1, DO_ADD)
1084 RVVCALL(OPIVX2, vwaddu_vx_h, WOP_UUU_H, H4, H2, DO_ADD)
1085 RVVCALL(OPIVX2, vwaddu_vx_w, WOP_UUU_W, H8, H4, DO_ADD)
1086 RVVCALL(OPIVX2, vwsubu_vx_b, WOP_UUU_B, H2, H1, DO_SUB)
1087 RVVCALL(OPIVX2, vwsubu_vx_h, WOP_UUU_H, H4, H2, DO_SUB)
1088 RVVCALL(OPIVX2, vwsubu_vx_w, WOP_UUU_W, H8, H4, DO_SUB)
1089 RVVCALL(OPIVX2, vwadd_vx_b, WOP_SSS_B, H2, H1, DO_ADD)
1090 RVVCALL(OPIVX2, vwadd_vx_h, WOP_SSS_H, H4, H2, DO_ADD)
1091 RVVCALL(OPIVX2, vwadd_vx_w, WOP_SSS_W, H8, H4, DO_ADD)
1092 RVVCALL(OPIVX2, vwsub_vx_b, WOP_SSS_B, H2, H1, DO_SUB)
1093 RVVCALL(OPIVX2, vwsub_vx_h, WOP_SSS_H, H4, H2, DO_SUB)
1094 RVVCALL(OPIVX2, vwsub_vx_w, WOP_SSS_W, H8, H4, DO_SUB)
1095 RVVCALL(OPIVX2, vwaddu_wx_b, WOP_WUUU_B, H2, H1, DO_ADD)
1096 RVVCALL(OPIVX2, vwaddu_wx_h, WOP_WUUU_H, H4, H2, DO_ADD)
1097 RVVCALL(OPIVX2, vwaddu_wx_w, WOP_WUUU_W, H8, H4, DO_ADD)
1098 RVVCALL(OPIVX2, vwsubu_wx_b, WOP_WUUU_B, H2, H1, DO_SUB)
1099 RVVCALL(OPIVX2, vwsubu_wx_h, WOP_WUUU_H, H4, H2, DO_SUB)
1100 RVVCALL(OPIVX2, vwsubu_wx_w, WOP_WUUU_W, H8, H4, DO_SUB)
1101 RVVCALL(OPIVX2, vwadd_wx_b, WOP_WSSS_B, H2, H1, DO_ADD)
1102 RVVCALL(OPIVX2, vwadd_wx_h, WOP_WSSS_H, H4, H2, DO_ADD)
1103 RVVCALL(OPIVX2, vwadd_wx_w, WOP_WSSS_W, H8, H4, DO_ADD)
1104 RVVCALL(OPIVX2, vwsub_wx_b, WOP_WSSS_B, H2, H1, DO_SUB)
1105 RVVCALL(OPIVX2, vwsub_wx_h, WOP_WSSS_H, H4, H2, DO_SUB)
1106 RVVCALL(OPIVX2, vwsub_wx_w, WOP_WSSS_W, H8, H4, DO_SUB)
1107 GEN_VEXT_VX(vwaddu_vx_b, 1, 2, clearh)
1108 GEN_VEXT_VX(vwaddu_vx_h, 2, 4, clearl)
1109 GEN_VEXT_VX(vwaddu_vx_w, 4, 8, clearq)
1110 GEN_VEXT_VX(vwsubu_vx_b, 1, 2, clearh)
1111 GEN_VEXT_VX(vwsubu_vx_h, 2, 4, clearl)
1112 GEN_VEXT_VX(vwsubu_vx_w, 4, 8, clearq)
1113 GEN_VEXT_VX(vwadd_vx_b, 1, 2, clearh)
1114 GEN_VEXT_VX(vwadd_vx_h, 2, 4, clearl)
1115 GEN_VEXT_VX(vwadd_vx_w, 4, 8, clearq)
1116 GEN_VEXT_VX(vwsub_vx_b, 1, 2, clearh)
1117 GEN_VEXT_VX(vwsub_vx_h, 2, 4, clearl)
1118 GEN_VEXT_VX(vwsub_vx_w, 4, 8, clearq)
1119 GEN_VEXT_VX(vwaddu_wx_b, 1, 2, clearh)
1120 GEN_VEXT_VX(vwaddu_wx_h, 2, 4, clearl)
1121 GEN_VEXT_VX(vwaddu_wx_w, 4, 8, clearq)
1122 GEN_VEXT_VX(vwsubu_wx_b, 1, 2, clearh)
1123 GEN_VEXT_VX(vwsubu_wx_h, 2, 4, clearl)
1124 GEN_VEXT_VX(vwsubu_wx_w, 4, 8, clearq)
1125 GEN_VEXT_VX(vwadd_wx_b, 1, 2, clearh)
1126 GEN_VEXT_VX(vwadd_wx_h, 2, 4, clearl)
1127 GEN_VEXT_VX(vwadd_wx_w, 4, 8, clearq)
1128 GEN_VEXT_VX(vwsub_wx_b, 1, 2, clearh)
1129 GEN_VEXT_VX(vwsub_wx_h, 2, 4, clearl)
1130 GEN_VEXT_VX(vwsub_wx_w, 4, 8, clearq)
1131