xref: /openbmc/qemu/target/riscv/vcrypto_helper.c (revision 1f2355f5)
1 /*
2  * RISC-V Vector Crypto Extension Helpers for QEMU.
3  *
4  * Copyright (C) 2023 SiFive, Inc.
5  * Written by Codethink Ltd and SiFive.
6  *
7  * This program is free software; you can redistribute it and/or modify it
8  * under the terms and conditions of the GNU General Public License,
9  * version 2 or later, as published by the Free Software Foundation.
10  *
11  * This program is distributed in the hope it will be useful, but WITHOUT
12  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
14  * more details.
15  *
16  * You should have received a copy of the GNU General Public License along with
17  * this program.  If not, see <http://www.gnu.org/licenses/>.
18  */
19 
20 #include "qemu/osdep.h"
21 #include "qemu/host-utils.h"
22 #include "qemu/bitops.h"
23 #include "qemu/bswap.h"
24 #include "cpu.h"
25 #include "crypto/aes.h"
26 #include "crypto/aes-round.h"
27 #include "crypto/sm4.h"
28 #include "exec/memop.h"
29 #include "exec/exec-all.h"
30 #include "exec/helper-proto.h"
31 #include "internals.h"
32 #include "vector_internals.h"
33 
34 static uint64_t clmul64(uint64_t y, uint64_t x)
35 {
36     uint64_t result = 0;
37     for (int j = 63; j >= 0; j--) {
38         if ((y >> j) & 1) {
39             result ^= (x << j);
40         }
41     }
42     return result;
43 }
44 
45 static uint64_t clmulh64(uint64_t y, uint64_t x)
46 {
47     uint64_t result = 0;
48     for (int j = 63; j >= 1; j--) {
49         if ((y >> j) & 1) {
50             result ^= (x >> (64 - j));
51         }
52     }
53     return result;
54 }
55 
56 RVVCALL(OPIVV2, vclmul_vv, OP_UUU_D, H8, H8, H8, clmul64)
57 GEN_VEXT_VV(vclmul_vv, 8)
58 RVVCALL(OPIVX2, vclmul_vx, OP_UUU_D, H8, H8, clmul64)
59 GEN_VEXT_VX(vclmul_vx, 8)
60 RVVCALL(OPIVV2, vclmulh_vv, OP_UUU_D, H8, H8, H8, clmulh64)
61 GEN_VEXT_VV(vclmulh_vv, 8)
62 RVVCALL(OPIVX2, vclmulh_vx, OP_UUU_D, H8, H8, clmulh64)
63 GEN_VEXT_VX(vclmulh_vx, 8)
64 
65 RVVCALL(OPIVV2, vror_vv_b, OP_UUU_B, H1, H1, H1, ror8)
66 RVVCALL(OPIVV2, vror_vv_h, OP_UUU_H, H2, H2, H2, ror16)
67 RVVCALL(OPIVV2, vror_vv_w, OP_UUU_W, H4, H4, H4, ror32)
68 RVVCALL(OPIVV2, vror_vv_d, OP_UUU_D, H8, H8, H8, ror64)
69 GEN_VEXT_VV(vror_vv_b, 1)
70 GEN_VEXT_VV(vror_vv_h, 2)
71 GEN_VEXT_VV(vror_vv_w, 4)
72 GEN_VEXT_VV(vror_vv_d, 8)
73 
74 RVVCALL(OPIVX2, vror_vx_b, OP_UUU_B, H1, H1, ror8)
75 RVVCALL(OPIVX2, vror_vx_h, OP_UUU_H, H2, H2, ror16)
76 RVVCALL(OPIVX2, vror_vx_w, OP_UUU_W, H4, H4, ror32)
77 RVVCALL(OPIVX2, vror_vx_d, OP_UUU_D, H8, H8, ror64)
78 GEN_VEXT_VX(vror_vx_b, 1)
79 GEN_VEXT_VX(vror_vx_h, 2)
80 GEN_VEXT_VX(vror_vx_w, 4)
81 GEN_VEXT_VX(vror_vx_d, 8)
82 
83 RVVCALL(OPIVV2, vrol_vv_b, OP_UUU_B, H1, H1, H1, rol8)
84 RVVCALL(OPIVV2, vrol_vv_h, OP_UUU_H, H2, H2, H2, rol16)
85 RVVCALL(OPIVV2, vrol_vv_w, OP_UUU_W, H4, H4, H4, rol32)
86 RVVCALL(OPIVV2, vrol_vv_d, OP_UUU_D, H8, H8, H8, rol64)
87 GEN_VEXT_VV(vrol_vv_b, 1)
88 GEN_VEXT_VV(vrol_vv_h, 2)
89 GEN_VEXT_VV(vrol_vv_w, 4)
90 GEN_VEXT_VV(vrol_vv_d, 8)
91 
92 RVVCALL(OPIVX2, vrol_vx_b, OP_UUU_B, H1, H1, rol8)
93 RVVCALL(OPIVX2, vrol_vx_h, OP_UUU_H, H2, H2, rol16)
94 RVVCALL(OPIVX2, vrol_vx_w, OP_UUU_W, H4, H4, rol32)
95 RVVCALL(OPIVX2, vrol_vx_d, OP_UUU_D, H8, H8, rol64)
96 GEN_VEXT_VX(vrol_vx_b, 1)
97 GEN_VEXT_VX(vrol_vx_h, 2)
98 GEN_VEXT_VX(vrol_vx_w, 4)
99 GEN_VEXT_VX(vrol_vx_d, 8)
100 
101 static uint64_t brev8(uint64_t val)
102 {
103     val = ((val & 0x5555555555555555ull) << 1) |
104           ((val & 0xAAAAAAAAAAAAAAAAull) >> 1);
105     val = ((val & 0x3333333333333333ull) << 2) |
106           ((val & 0xCCCCCCCCCCCCCCCCull) >> 2);
107     val = ((val & 0x0F0F0F0F0F0F0F0Full) << 4) |
108           ((val & 0xF0F0F0F0F0F0F0F0ull) >> 4);
109 
110     return val;
111 }
112 
113 RVVCALL(OPIVV1, vbrev8_v_b, OP_UU_B, H1, H1, brev8)
114 RVVCALL(OPIVV1, vbrev8_v_h, OP_UU_H, H2, H2, brev8)
115 RVVCALL(OPIVV1, vbrev8_v_w, OP_UU_W, H4, H4, brev8)
116 RVVCALL(OPIVV1, vbrev8_v_d, OP_UU_D, H8, H8, brev8)
117 GEN_VEXT_V(vbrev8_v_b, 1)
118 GEN_VEXT_V(vbrev8_v_h, 2)
119 GEN_VEXT_V(vbrev8_v_w, 4)
120 GEN_VEXT_V(vbrev8_v_d, 8)
121 
122 #define DO_IDENTITY(a) (a)
123 RVVCALL(OPIVV1, vrev8_v_b, OP_UU_B, H1, H1, DO_IDENTITY)
124 RVVCALL(OPIVV1, vrev8_v_h, OP_UU_H, H2, H2, bswap16)
125 RVVCALL(OPIVV1, vrev8_v_w, OP_UU_W, H4, H4, bswap32)
126 RVVCALL(OPIVV1, vrev8_v_d, OP_UU_D, H8, H8, bswap64)
127 GEN_VEXT_V(vrev8_v_b, 1)
128 GEN_VEXT_V(vrev8_v_h, 2)
129 GEN_VEXT_V(vrev8_v_w, 4)
130 GEN_VEXT_V(vrev8_v_d, 8)
131 
132 #define DO_ANDN(a, b) ((a) & ~(b))
133 RVVCALL(OPIVV2, vandn_vv_b, OP_UUU_B, H1, H1, H1, DO_ANDN)
134 RVVCALL(OPIVV2, vandn_vv_h, OP_UUU_H, H2, H2, H2, DO_ANDN)
135 RVVCALL(OPIVV2, vandn_vv_w, OP_UUU_W, H4, H4, H4, DO_ANDN)
136 RVVCALL(OPIVV2, vandn_vv_d, OP_UUU_D, H8, H8, H8, DO_ANDN)
137 GEN_VEXT_VV(vandn_vv_b, 1)
138 GEN_VEXT_VV(vandn_vv_h, 2)
139 GEN_VEXT_VV(vandn_vv_w, 4)
140 GEN_VEXT_VV(vandn_vv_d, 8)
141 
142 RVVCALL(OPIVX2, vandn_vx_b, OP_UUU_B, H1, H1, DO_ANDN)
143 RVVCALL(OPIVX2, vandn_vx_h, OP_UUU_H, H2, H2, DO_ANDN)
144 RVVCALL(OPIVX2, vandn_vx_w, OP_UUU_W, H4, H4, DO_ANDN)
145 RVVCALL(OPIVX2, vandn_vx_d, OP_UUU_D, H8, H8, DO_ANDN)
146 GEN_VEXT_VX(vandn_vx_b, 1)
147 GEN_VEXT_VX(vandn_vx_h, 2)
148 GEN_VEXT_VX(vandn_vx_w, 4)
149 GEN_VEXT_VX(vandn_vx_d, 8)
150 
151 RVVCALL(OPIVV1, vbrev_v_b, OP_UU_B, H1, H1, revbit8)
152 RVVCALL(OPIVV1, vbrev_v_h, OP_UU_H, H2, H2, revbit16)
153 RVVCALL(OPIVV1, vbrev_v_w, OP_UU_W, H4, H4, revbit32)
154 RVVCALL(OPIVV1, vbrev_v_d, OP_UU_D, H8, H8, revbit64)
155 GEN_VEXT_V(vbrev_v_b, 1)
156 GEN_VEXT_V(vbrev_v_h, 2)
157 GEN_VEXT_V(vbrev_v_w, 4)
158 GEN_VEXT_V(vbrev_v_d, 8)
159 
160 RVVCALL(OPIVV1, vclz_v_b, OP_UU_B, H1, H1, clz8)
161 RVVCALL(OPIVV1, vclz_v_h, OP_UU_H, H2, H2, clz16)
162 RVVCALL(OPIVV1, vclz_v_w, OP_UU_W, H4, H4, clz32)
163 RVVCALL(OPIVV1, vclz_v_d, OP_UU_D, H8, H8, clz64)
164 GEN_VEXT_V(vclz_v_b, 1)
165 GEN_VEXT_V(vclz_v_h, 2)
166 GEN_VEXT_V(vclz_v_w, 4)
167 GEN_VEXT_V(vclz_v_d, 8)
168 
169 RVVCALL(OPIVV1, vctz_v_b, OP_UU_B, H1, H1, ctz8)
170 RVVCALL(OPIVV1, vctz_v_h, OP_UU_H, H2, H2, ctz16)
171 RVVCALL(OPIVV1, vctz_v_w, OP_UU_W, H4, H4, ctz32)
172 RVVCALL(OPIVV1, vctz_v_d, OP_UU_D, H8, H8, ctz64)
173 GEN_VEXT_V(vctz_v_b, 1)
174 GEN_VEXT_V(vctz_v_h, 2)
175 GEN_VEXT_V(vctz_v_w, 4)
176 GEN_VEXT_V(vctz_v_d, 8)
177 
178 RVVCALL(OPIVV1, vcpop_v_b, OP_UU_B, H1, H1, ctpop8)
179 RVVCALL(OPIVV1, vcpop_v_h, OP_UU_H, H2, H2, ctpop16)
180 RVVCALL(OPIVV1, vcpop_v_w, OP_UU_W, H4, H4, ctpop32)
181 RVVCALL(OPIVV1, vcpop_v_d, OP_UU_D, H8, H8, ctpop64)
182 GEN_VEXT_V(vcpop_v_b, 1)
183 GEN_VEXT_V(vcpop_v_h, 2)
184 GEN_VEXT_V(vcpop_v_w, 4)
185 GEN_VEXT_V(vcpop_v_d, 8)
186 
187 #define DO_SLL(N, M) (N << (M & (sizeof(N) * 8 - 1)))
188 RVVCALL(OPIVV2, vwsll_vv_b, WOP_UUU_B, H2, H1, H1, DO_SLL)
189 RVVCALL(OPIVV2, vwsll_vv_h, WOP_UUU_H, H4, H2, H2, DO_SLL)
190 RVVCALL(OPIVV2, vwsll_vv_w, WOP_UUU_W, H8, H4, H4, DO_SLL)
191 GEN_VEXT_VV(vwsll_vv_b, 2)
192 GEN_VEXT_VV(vwsll_vv_h, 4)
193 GEN_VEXT_VV(vwsll_vv_w, 8)
194 
195 RVVCALL(OPIVX2, vwsll_vx_b, WOP_UUU_B, H2, H1, DO_SLL)
196 RVVCALL(OPIVX2, vwsll_vx_h, WOP_UUU_H, H4, H2, DO_SLL)
197 RVVCALL(OPIVX2, vwsll_vx_w, WOP_UUU_W, H8, H4, DO_SLL)
198 GEN_VEXT_VX(vwsll_vx_b, 2)
199 GEN_VEXT_VX(vwsll_vx_h, 4)
200 GEN_VEXT_VX(vwsll_vx_w, 8)
201 
202 void HELPER(egs_check)(uint32_t egs, CPURISCVState *env)
203 {
204     uint32_t vl = env->vl;
205     uint32_t vstart = env->vstart;
206 
207     if (vl % egs != 0 || vstart % egs != 0) {
208         riscv_raise_exception(env, RISCV_EXCP_ILLEGAL_INST, GETPC());
209     }
210 }
211 
212 static inline void xor_round_key(AESState *round_state, AESState *round_key)
213 {
214     round_state->v = round_state->v ^ round_key->v;
215 }
216 
217 #define GEN_ZVKNED_HELPER_VV(NAME, ...)                                   \
218     void HELPER(NAME)(void *vd, void *vs2, CPURISCVState *env,            \
219                       uint32_t desc)                                      \
220     {                                                                     \
221         uint32_t vl = env->vl;                                            \
222         uint32_t total_elems = vext_get_total_elems(env, desc, 4);        \
223         uint32_t vta = vext_vta(desc);                                    \
224                                                                           \
225         VSTART_CHECK_EARLY_EXIT(env);                                     \
226                                                                           \
227         for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {        \
228             AESState round_key;                                           \
229             round_key.d[0] = *((uint64_t *)vs2 + H8(i * 2 + 0));          \
230             round_key.d[1] = *((uint64_t *)vs2 + H8(i * 2 + 1));          \
231             AESState round_state;                                         \
232             round_state.d[0] = *((uint64_t *)vd + H8(i * 2 + 0));         \
233             round_state.d[1] = *((uint64_t *)vd + H8(i * 2 + 1));         \
234             __VA_ARGS__;                                                  \
235             *((uint64_t *)vd + H8(i * 2 + 0)) = round_state.d[0];         \
236             *((uint64_t *)vd + H8(i * 2 + 1)) = round_state.d[1];         \
237         }                                                                 \
238         env->vstart = 0;                                                  \
239         /* set tail elements to 1s */                                     \
240         vext_set_elems_1s(vd, vta, vl * 4, total_elems * 4);              \
241     }
242 
243 #define GEN_ZVKNED_HELPER_VS(NAME, ...)                                   \
244     void HELPER(NAME)(void *vd, void *vs2, CPURISCVState *env,            \
245                       uint32_t desc)                                      \
246     {                                                                     \
247         uint32_t vl = env->vl;                                            \
248         uint32_t total_elems = vext_get_total_elems(env, desc, 4);        \
249         uint32_t vta = vext_vta(desc);                                    \
250                                                                           \
251         VSTART_CHECK_EARLY_EXIT(env);                                     \
252                                                                           \
253         for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {        \
254             AESState round_key;                                           \
255             round_key.d[0] = *((uint64_t *)vs2 + H8(0));                  \
256             round_key.d[1] = *((uint64_t *)vs2 + H8(1));                  \
257             AESState round_state;                                         \
258             round_state.d[0] = *((uint64_t *)vd + H8(i * 2 + 0));         \
259             round_state.d[1] = *((uint64_t *)vd + H8(i * 2 + 1));         \
260             __VA_ARGS__;                                                  \
261             *((uint64_t *)vd + H8(i * 2 + 0)) = round_state.d[0];         \
262             *((uint64_t *)vd + H8(i * 2 + 1)) = round_state.d[1];         \
263         }                                                                 \
264         env->vstart = 0;                                                  \
265         /* set tail elements to 1s */                                     \
266         vext_set_elems_1s(vd, vta, vl * 4, total_elems * 4);              \
267     }
268 
269 GEN_ZVKNED_HELPER_VV(vaesef_vv, aesenc_SB_SR_AK(&round_state,
270                                                 &round_state,
271                                                 &round_key,
272                                                 false);)
273 GEN_ZVKNED_HELPER_VS(vaesef_vs, aesenc_SB_SR_AK(&round_state,
274                                                 &round_state,
275                                                 &round_key,
276                                                 false);)
277 GEN_ZVKNED_HELPER_VV(vaesdf_vv, aesdec_ISB_ISR_AK(&round_state,
278                                                   &round_state,
279                                                   &round_key,
280                                                   false);)
281 GEN_ZVKNED_HELPER_VS(vaesdf_vs, aesdec_ISB_ISR_AK(&round_state,
282                                                   &round_state,
283                                                   &round_key,
284                                                   false);)
285 GEN_ZVKNED_HELPER_VV(vaesem_vv, aesenc_SB_SR_MC_AK(&round_state,
286                                                    &round_state,
287                                                    &round_key,
288                                                    false);)
289 GEN_ZVKNED_HELPER_VS(vaesem_vs, aesenc_SB_SR_MC_AK(&round_state,
290                                                    &round_state,
291                                                    &round_key,
292                                                    false);)
293 GEN_ZVKNED_HELPER_VV(vaesdm_vv, aesdec_ISB_ISR_AK_IMC(&round_state,
294                                                       &round_state,
295                                                       &round_key,
296                                                       false);)
297 GEN_ZVKNED_HELPER_VS(vaesdm_vs, aesdec_ISB_ISR_AK_IMC(&round_state,
298                                                       &round_state,
299                                                       &round_key,
300                                                       false);)
301 GEN_ZVKNED_HELPER_VS(vaesz_vs, xor_round_key(&round_state, &round_key);)
302 
303 void HELPER(vaeskf1_vi)(void *vd_vptr, void *vs2_vptr, uint32_t uimm,
304                         CPURISCVState *env, uint32_t desc)
305 {
306     uint32_t *vd = vd_vptr;
307     uint32_t *vs2 = vs2_vptr;
308     uint32_t vl = env->vl;
309     uint32_t total_elems = vext_get_total_elems(env, desc, 4);
310     uint32_t vta = vext_vta(desc);
311 
312     VSTART_CHECK_EARLY_EXIT(env);
313 
314     uimm &= 0b1111;
315     if (uimm > 10 || uimm == 0) {
316         uimm ^= 0b1000;
317     }
318 
319     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
320         uint32_t rk[8], tmp;
321         static const uint32_t rcon[] = {
322             0x00000001, 0x00000002, 0x00000004, 0x00000008, 0x00000010,
323             0x00000020, 0x00000040, 0x00000080, 0x0000001B, 0x00000036,
324         };
325 
326         rk[0] = vs2[i * 4 + H4(0)];
327         rk[1] = vs2[i * 4 + H4(1)];
328         rk[2] = vs2[i * 4 + H4(2)];
329         rk[3] = vs2[i * 4 + H4(3)];
330         tmp = ror32(rk[3], 8);
331 
332         rk[4] = rk[0] ^ (((uint32_t)AES_sbox[(tmp >> 24) & 0xff] << 24) |
333                          ((uint32_t)AES_sbox[(tmp >> 16) & 0xff] << 16) |
334                          ((uint32_t)AES_sbox[(tmp >> 8) & 0xff] << 8) |
335                          ((uint32_t)AES_sbox[(tmp >> 0) & 0xff] << 0))
336                       ^ rcon[uimm - 1];
337         rk[5] = rk[1] ^ rk[4];
338         rk[6] = rk[2] ^ rk[5];
339         rk[7] = rk[3] ^ rk[6];
340 
341         vd[i * 4 + H4(0)] = rk[4];
342         vd[i * 4 + H4(1)] = rk[5];
343         vd[i * 4 + H4(2)] = rk[6];
344         vd[i * 4 + H4(3)] = rk[7];
345     }
346     env->vstart = 0;
347     /* set tail elements to 1s */
348     vext_set_elems_1s(vd, vta, vl * 4, total_elems * 4);
349 }
350 
351 void HELPER(vaeskf2_vi)(void *vd_vptr, void *vs2_vptr, uint32_t uimm,
352                         CPURISCVState *env, uint32_t desc)
353 {
354     uint32_t *vd = vd_vptr;
355     uint32_t *vs2 = vs2_vptr;
356     uint32_t vl = env->vl;
357     uint32_t total_elems = vext_get_total_elems(env, desc, 4);
358     uint32_t vta = vext_vta(desc);
359 
360     VSTART_CHECK_EARLY_EXIT(env);
361 
362     uimm &= 0b1111;
363     if (uimm > 14 || uimm < 2) {
364         uimm ^= 0b1000;
365     }
366 
367     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
368         uint32_t rk[12], tmp;
369         static const uint32_t rcon[] = {
370             0x00000001, 0x00000002, 0x00000004, 0x00000008, 0x00000010,
371             0x00000020, 0x00000040, 0x00000080, 0x0000001B, 0x00000036,
372         };
373 
374         rk[0] = vd[i * 4 + H4(0)];
375         rk[1] = vd[i * 4 + H4(1)];
376         rk[2] = vd[i * 4 + H4(2)];
377         rk[3] = vd[i * 4 + H4(3)];
378         rk[4] = vs2[i * 4 + H4(0)];
379         rk[5] = vs2[i * 4 + H4(1)];
380         rk[6] = vs2[i * 4 + H4(2)];
381         rk[7] = vs2[i * 4 + H4(3)];
382 
383         if (uimm % 2 == 0) {
384             tmp = ror32(rk[7], 8);
385             rk[8] = rk[0] ^ (((uint32_t)AES_sbox[(tmp >> 24) & 0xff] << 24) |
386                              ((uint32_t)AES_sbox[(tmp >> 16) & 0xff] << 16) |
387                              ((uint32_t)AES_sbox[(tmp >> 8) & 0xff] << 8) |
388                              ((uint32_t)AES_sbox[(tmp >> 0) & 0xff] << 0))
389                           ^ rcon[(uimm - 1) / 2];
390         } else {
391             rk[8] = rk[0] ^ (((uint32_t)AES_sbox[(rk[7] >> 24) & 0xff] << 24) |
392                              ((uint32_t)AES_sbox[(rk[7] >> 16) & 0xff] << 16) |
393                              ((uint32_t)AES_sbox[(rk[7] >> 8) & 0xff] << 8) |
394                              ((uint32_t)AES_sbox[(rk[7] >> 0) & 0xff] << 0));
395         }
396         rk[9] = rk[1] ^ rk[8];
397         rk[10] = rk[2] ^ rk[9];
398         rk[11] = rk[3] ^ rk[10];
399 
400         vd[i * 4 + H4(0)] = rk[8];
401         vd[i * 4 + H4(1)] = rk[9];
402         vd[i * 4 + H4(2)] = rk[10];
403         vd[i * 4 + H4(3)] = rk[11];
404     }
405     env->vstart = 0;
406     /* set tail elements to 1s */
407     vext_set_elems_1s(vd, vta, vl * 4, total_elems * 4);
408 }
409 
410 static inline uint32_t sig0_sha256(uint32_t x)
411 {
412     return ror32(x, 7) ^ ror32(x, 18) ^ (x >> 3);
413 }
414 
415 static inline uint32_t sig1_sha256(uint32_t x)
416 {
417     return ror32(x, 17) ^ ror32(x, 19) ^ (x >> 10);
418 }
419 
420 static inline uint64_t sig0_sha512(uint64_t x)
421 {
422     return ror64(x, 1) ^ ror64(x, 8) ^ (x >> 7);
423 }
424 
425 static inline uint64_t sig1_sha512(uint64_t x)
426 {
427     return ror64(x, 19) ^ ror64(x, 61) ^ (x >> 6);
428 }
429 
430 static inline void vsha2ms_e32(uint32_t *vd, uint32_t *vs1, uint32_t *vs2)
431 {
432     uint32_t res[4];
433     res[0] = sig1_sha256(vs1[H4(2)]) + vs2[H4(1)] + sig0_sha256(vd[H4(1)]) +
434              vd[H4(0)];
435     res[1] = sig1_sha256(vs1[H4(3)]) + vs2[H4(2)] + sig0_sha256(vd[H4(2)]) +
436              vd[H4(1)];
437     res[2] =
438         sig1_sha256(res[0]) + vs2[H4(3)] + sig0_sha256(vd[H4(3)]) + vd[H4(2)];
439     res[3] =
440         sig1_sha256(res[1]) + vs1[H4(0)] + sig0_sha256(vs2[H4(0)]) + vd[H4(3)];
441     vd[H4(3)] = res[3];
442     vd[H4(2)] = res[2];
443     vd[H4(1)] = res[1];
444     vd[H4(0)] = res[0];
445 }
446 
447 static inline void vsha2ms_e64(uint64_t *vd, uint64_t *vs1, uint64_t *vs2)
448 {
449     uint64_t res[4];
450     res[0] = sig1_sha512(vs1[2]) + vs2[1] + sig0_sha512(vd[1]) + vd[0];
451     res[1] = sig1_sha512(vs1[3]) + vs2[2] + sig0_sha512(vd[2]) + vd[1];
452     res[2] = sig1_sha512(res[0]) + vs2[3] + sig0_sha512(vd[3]) + vd[2];
453     res[3] = sig1_sha512(res[1]) + vs1[0] + sig0_sha512(vs2[0]) + vd[3];
454     vd[3] = res[3];
455     vd[2] = res[2];
456     vd[1] = res[1];
457     vd[0] = res[0];
458 }
459 
460 void HELPER(vsha2ms_vv)(void *vd, void *vs1, void *vs2, CPURISCVState *env,
461                         uint32_t desc)
462 {
463     uint32_t sew = FIELD_EX64(env->vtype, VTYPE, VSEW);
464     uint32_t esz = sew == MO_32 ? 4 : 8;
465     uint32_t total_elems;
466     uint32_t vta = vext_vta(desc);
467 
468     VSTART_CHECK_EARLY_EXIT(env);
469 
470     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
471         if (sew == MO_32) {
472             vsha2ms_e32(((uint32_t *)vd) + i * 4, ((uint32_t *)vs1) + i * 4,
473                         ((uint32_t *)vs2) + i * 4);
474         } else {
475             /* If not 32 then SEW should be 64 */
476             vsha2ms_e64(((uint64_t *)vd) + i * 4, ((uint64_t *)vs1) + i * 4,
477                         ((uint64_t *)vs2) + i * 4);
478         }
479     }
480     /* set tail elements to 1s */
481     total_elems = vext_get_total_elems(env, desc, esz);
482     vext_set_elems_1s(vd, vta, env->vl * esz, total_elems * esz);
483     env->vstart = 0;
484 }
485 
486 static inline uint64_t sum0_64(uint64_t x)
487 {
488     return ror64(x, 28) ^ ror64(x, 34) ^ ror64(x, 39);
489 }
490 
491 static inline uint32_t sum0_32(uint32_t x)
492 {
493     return ror32(x, 2) ^ ror32(x, 13) ^ ror32(x, 22);
494 }
495 
496 static inline uint64_t sum1_64(uint64_t x)
497 {
498     return ror64(x, 14) ^ ror64(x, 18) ^ ror64(x, 41);
499 }
500 
501 static inline uint32_t sum1_32(uint32_t x)
502 {
503     return ror32(x, 6) ^ ror32(x, 11) ^ ror32(x, 25);
504 }
505 
506 #define ch(x, y, z) ((x & y) ^ ((~x) & z))
507 
508 #define maj(x, y, z) ((x & y) ^ (x & z) ^ (y & z))
509 
510 static void vsha2c_64(uint64_t *vs2, uint64_t *vd, uint64_t *vs1)
511 {
512     uint64_t a = vs2[3], b = vs2[2], e = vs2[1], f = vs2[0];
513     uint64_t c = vd[3], d = vd[2], g = vd[1], h = vd[0];
514     uint64_t W0 = vs1[0], W1 = vs1[1];
515     uint64_t T1 = h + sum1_64(e) + ch(e, f, g) + W0;
516     uint64_t T2 = sum0_64(a) + maj(a, b, c);
517 
518     h = g;
519     g = f;
520     f = e;
521     e = d + T1;
522     d = c;
523     c = b;
524     b = a;
525     a = T1 + T2;
526 
527     T1 = h + sum1_64(e) + ch(e, f, g) + W1;
528     T2 = sum0_64(a) + maj(a, b, c);
529     h = g;
530     g = f;
531     f = e;
532     e = d + T1;
533     d = c;
534     c = b;
535     b = a;
536     a = T1 + T2;
537 
538     vd[0] = f;
539     vd[1] = e;
540     vd[2] = b;
541     vd[3] = a;
542 }
543 
544 static void vsha2c_32(uint32_t *vs2, uint32_t *vd, uint32_t *vs1)
545 {
546     uint32_t a = vs2[H4(3)], b = vs2[H4(2)], e = vs2[H4(1)], f = vs2[H4(0)];
547     uint32_t c = vd[H4(3)], d = vd[H4(2)], g = vd[H4(1)], h = vd[H4(0)];
548     uint32_t W0 = vs1[H4(0)], W1 = vs1[H4(1)];
549     uint32_t T1 = h + sum1_32(e) + ch(e, f, g) + W0;
550     uint32_t T2 = sum0_32(a) + maj(a, b, c);
551 
552     h = g;
553     g = f;
554     f = e;
555     e = d + T1;
556     d = c;
557     c = b;
558     b = a;
559     a = T1 + T2;
560 
561     T1 = h + sum1_32(e) + ch(e, f, g) + W1;
562     T2 = sum0_32(a) + maj(a, b, c);
563     h = g;
564     g = f;
565     f = e;
566     e = d + T1;
567     d = c;
568     c = b;
569     b = a;
570     a = T1 + T2;
571 
572     vd[H4(0)] = f;
573     vd[H4(1)] = e;
574     vd[H4(2)] = b;
575     vd[H4(3)] = a;
576 }
577 
578 void HELPER(vsha2ch32_vv)(void *vd, void *vs1, void *vs2, CPURISCVState *env,
579                           uint32_t desc)
580 {
581     const uint32_t esz = 4;
582     uint32_t total_elems;
583     uint32_t vta = vext_vta(desc);
584 
585     VSTART_CHECK_EARLY_EXIT(env);
586 
587     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
588         vsha2c_32(((uint32_t *)vs2) + 4 * i, ((uint32_t *)vd) + 4 * i,
589                   ((uint32_t *)vs1) + 4 * i + 2);
590     }
591 
592     /* set tail elements to 1s */
593     total_elems = vext_get_total_elems(env, desc, esz);
594     vext_set_elems_1s(vd, vta, env->vl * esz, total_elems * esz);
595     env->vstart = 0;
596 }
597 
598 void HELPER(vsha2ch64_vv)(void *vd, void *vs1, void *vs2, CPURISCVState *env,
599                           uint32_t desc)
600 {
601     const uint32_t esz = 8;
602     uint32_t total_elems;
603     uint32_t vta = vext_vta(desc);
604 
605     VSTART_CHECK_EARLY_EXIT(env);
606 
607     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
608         vsha2c_64(((uint64_t *)vs2) + 4 * i, ((uint64_t *)vd) + 4 * i,
609                   ((uint64_t *)vs1) + 4 * i + 2);
610     }
611 
612     /* set tail elements to 1s */
613     total_elems = vext_get_total_elems(env, desc, esz);
614     vext_set_elems_1s(vd, vta, env->vl * esz, total_elems * esz);
615     env->vstart = 0;
616 }
617 
618 void HELPER(vsha2cl32_vv)(void *vd, void *vs1, void *vs2, CPURISCVState *env,
619                           uint32_t desc)
620 {
621     const uint32_t esz = 4;
622     uint32_t total_elems;
623     uint32_t vta = vext_vta(desc);
624 
625     VSTART_CHECK_EARLY_EXIT(env);
626 
627     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
628         vsha2c_32(((uint32_t *)vs2) + 4 * i, ((uint32_t *)vd) + 4 * i,
629                   (((uint32_t *)vs1) + 4 * i));
630     }
631 
632     /* set tail elements to 1s */
633     total_elems = vext_get_total_elems(env, desc, esz);
634     vext_set_elems_1s(vd, vta, env->vl * esz, total_elems * esz);
635     env->vstart = 0;
636 }
637 
638 void HELPER(vsha2cl64_vv)(void *vd, void *vs1, void *vs2, CPURISCVState *env,
639                           uint32_t desc)
640 {
641     uint32_t esz = 8;
642     uint32_t total_elems;
643     uint32_t vta = vext_vta(desc);
644 
645     VSTART_CHECK_EARLY_EXIT(env);
646 
647     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
648         vsha2c_64(((uint64_t *)vs2) + 4 * i, ((uint64_t *)vd) + 4 * i,
649                   (((uint64_t *)vs1) + 4 * i));
650     }
651 
652     /* set tail elements to 1s */
653     total_elems = vext_get_total_elems(env, desc, esz);
654     vext_set_elems_1s(vd, vta, env->vl * esz, total_elems * esz);
655     env->vstart = 0;
656 }
657 
658 static inline uint32_t p1(uint32_t x)
659 {
660     return x ^ rol32(x, 15) ^ rol32(x, 23);
661 }
662 
663 static inline uint32_t zvksh_w(uint32_t m16, uint32_t m9, uint32_t m3,
664                                uint32_t m13, uint32_t m6)
665 {
666     return p1(m16 ^ m9 ^ rol32(m3, 15)) ^ rol32(m13, 7) ^ m6;
667 }
668 
669 void HELPER(vsm3me_vv)(void *vd_vptr, void *vs1_vptr, void *vs2_vptr,
670                        CPURISCVState *env, uint32_t desc)
671 {
672     uint32_t esz = memop_size(FIELD_EX64(env->vtype, VTYPE, VSEW));
673     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
674     uint32_t vta = vext_vta(desc);
675     uint32_t *vd = vd_vptr;
676     uint32_t *vs1 = vs1_vptr;
677     uint32_t *vs2 = vs2_vptr;
678 
679     VSTART_CHECK_EARLY_EXIT(env);
680 
681     for (int i = env->vstart / 8; i < env->vl / 8; i++) {
682         uint32_t w[24];
683         for (int j = 0; j < 8; j++) {
684             w[j] = bswap32(vs1[H4((i * 8) + j)]);
685             w[j + 8] = bswap32(vs2[H4((i * 8) + j)]);
686         }
687         for (int j = 0; j < 8; j++) {
688             w[j + 16] =
689                 zvksh_w(w[j], w[j + 7], w[j + 13], w[j + 3], w[j + 10]);
690         }
691         for (int j = 0; j < 8; j++) {
692             vd[(i * 8) + j] = bswap32(w[H4(j + 16)]);
693         }
694     }
695     vext_set_elems_1s(vd_vptr, vta, env->vl * esz, total_elems * esz);
696     env->vstart = 0;
697 }
698 
699 static inline uint32_t ff1(uint32_t x, uint32_t y, uint32_t z)
700 {
701     return x ^ y ^ z;
702 }
703 
704 static inline uint32_t ff2(uint32_t x, uint32_t y, uint32_t z)
705 {
706     return (x & y) | (x & z) | (y & z);
707 }
708 
709 static inline uint32_t ff_j(uint32_t x, uint32_t y, uint32_t z, uint32_t j)
710 {
711     return (j <= 15) ? ff1(x, y, z) : ff2(x, y, z);
712 }
713 
714 static inline uint32_t gg1(uint32_t x, uint32_t y, uint32_t z)
715 {
716     return x ^ y ^ z;
717 }
718 
719 static inline uint32_t gg2(uint32_t x, uint32_t y, uint32_t z)
720 {
721     return (x & y) | (~x & z);
722 }
723 
724 static inline uint32_t gg_j(uint32_t x, uint32_t y, uint32_t z, uint32_t j)
725 {
726     return (j <= 15) ? gg1(x, y, z) : gg2(x, y, z);
727 }
728 
729 static inline uint32_t t_j(uint32_t j)
730 {
731     return (j <= 15) ? 0x79cc4519 : 0x7a879d8a;
732 }
733 
734 static inline uint32_t p_0(uint32_t x)
735 {
736     return x ^ rol32(x, 9) ^ rol32(x, 17);
737 }
738 
739 static void sm3c(uint32_t *vd, uint32_t *vs1, uint32_t *vs2, uint32_t uimm)
740 {
741     uint32_t x0, x1;
742     uint32_t j;
743     uint32_t ss1, ss2, tt1, tt2;
744     x0 = vs2[0] ^ vs2[4];
745     x1 = vs2[1] ^ vs2[5];
746     j = 2 * uimm;
747     ss1 = rol32(rol32(vs1[0], 12) + vs1[4] + rol32(t_j(j), j % 32), 7);
748     ss2 = ss1 ^ rol32(vs1[0], 12);
749     tt1 = ff_j(vs1[0], vs1[1], vs1[2], j) + vs1[3] + ss2 + x0;
750     tt2 = gg_j(vs1[4], vs1[5], vs1[6], j) + vs1[7] + ss1 + vs2[0];
751     vs1[3] = vs1[2];
752     vd[3] = rol32(vs1[1], 9);
753     vs1[1] = vs1[0];
754     vd[1] = tt1;
755     vs1[7] = vs1[6];
756     vd[7] = rol32(vs1[5], 19);
757     vs1[5] = vs1[4];
758     vd[5] = p_0(tt2);
759     j = 2 * uimm + 1;
760     ss1 = rol32(rol32(vd[1], 12) + vd[5] + rol32(t_j(j), j % 32), 7);
761     ss2 = ss1 ^ rol32(vd[1], 12);
762     tt1 = ff_j(vd[1], vs1[1], vd[3], j) + vs1[3] + ss2 + x1;
763     tt2 = gg_j(vd[5], vs1[5], vd[7], j) + vs1[7] + ss1 + vs2[1];
764     vd[2] = rol32(vs1[1], 9);
765     vd[0] = tt1;
766     vd[6] = rol32(vs1[5], 19);
767     vd[4] = p_0(tt2);
768 }
769 
770 void HELPER(vsm3c_vi)(void *vd_vptr, void *vs2_vptr, uint32_t uimm,
771                       CPURISCVState *env, uint32_t desc)
772 {
773     uint32_t esz = memop_size(FIELD_EX64(env->vtype, VTYPE, VSEW));
774     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
775     uint32_t vta = vext_vta(desc);
776     uint32_t *vd = vd_vptr;
777     uint32_t *vs2 = vs2_vptr;
778     uint32_t v1[8], v2[8], v3[8];
779 
780     VSTART_CHECK_EARLY_EXIT(env);
781 
782     for (int i = env->vstart / 8; i < env->vl / 8; i++) {
783         for (int k = 0; k < 8; k++) {
784             v2[k] = bswap32(vd[H4(i * 8 + k)]);
785             v3[k] = bswap32(vs2[H4(i * 8 + k)]);
786         }
787         sm3c(v1, v2, v3, uimm);
788         for (int k = 0; k < 8; k++) {
789             vd[i * 8 + k] = bswap32(v1[H4(k)]);
790         }
791     }
792     vext_set_elems_1s(vd_vptr, vta, env->vl * esz, total_elems * esz);
793     env->vstart = 0;
794 }
795 
796 void HELPER(vghsh_vv)(void *vd_vptr, void *vs1_vptr, void *vs2_vptr,
797                       CPURISCVState *env, uint32_t desc)
798 {
799     uint64_t *vd = vd_vptr;
800     uint64_t *vs1 = vs1_vptr;
801     uint64_t *vs2 = vs2_vptr;
802     uint32_t vta = vext_vta(desc);
803     uint32_t total_elems = vext_get_total_elems(env, desc, 4);
804 
805     VSTART_CHECK_EARLY_EXIT(env);
806 
807     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
808         uint64_t Y[2] = {vd[i * 2 + 0], vd[i * 2 + 1]};
809         uint64_t H[2] = {brev8(vs2[i * 2 + 0]), brev8(vs2[i * 2 + 1])};
810         uint64_t X[2] = {vs1[i * 2 + 0], vs1[i * 2 + 1]};
811         uint64_t Z[2] = {0, 0};
812 
813         uint64_t S[2] = {brev8(Y[0] ^ X[0]), brev8(Y[1] ^ X[1])};
814 
815         for (int j = 0; j < 128; j++) {
816             if ((S[j / 64] >> (j % 64)) & 1) {
817                 Z[0] ^= H[0];
818                 Z[1] ^= H[1];
819             }
820             bool reduce = ((H[1] >> 63) & 1);
821             H[1] = H[1] << 1 | H[0] >> 63;
822             H[0] = H[0] << 1;
823             if (reduce) {
824                 H[0] ^= 0x87;
825             }
826         }
827 
828         vd[i * 2 + 0] = brev8(Z[0]);
829         vd[i * 2 + 1] = brev8(Z[1]);
830     }
831     /* set tail elements to 1s */
832     vext_set_elems_1s(vd, vta, env->vl * 4, total_elems * 4);
833     env->vstart = 0;
834 }
835 
836 void HELPER(vgmul_vv)(void *vd_vptr, void *vs2_vptr, CPURISCVState *env,
837                       uint32_t desc)
838 {
839     uint64_t *vd = vd_vptr;
840     uint64_t *vs2 = vs2_vptr;
841     uint32_t vta = vext_vta(desc);
842     uint32_t total_elems = vext_get_total_elems(env, desc, 4);
843 
844     VSTART_CHECK_EARLY_EXIT(env);
845 
846     for (uint32_t i = env->vstart / 4; i < env->vl / 4; i++) {
847         uint64_t Y[2] = {brev8(vd[i * 2 + 0]), brev8(vd[i * 2 + 1])};
848         uint64_t H[2] = {brev8(vs2[i * 2 + 0]), brev8(vs2[i * 2 + 1])};
849         uint64_t Z[2] = {0, 0};
850 
851         for (int j = 0; j < 128; j++) {
852             if ((Y[j / 64] >> (j % 64)) & 1) {
853                 Z[0] ^= H[0];
854                 Z[1] ^= H[1];
855             }
856             bool reduce = ((H[1] >> 63) & 1);
857             H[1] = H[1] << 1 | H[0] >> 63;
858             H[0] = H[0] << 1;
859             if (reduce) {
860                 H[0] ^= 0x87;
861             }
862         }
863 
864         vd[i * 2 + 0] = brev8(Z[0]);
865         vd[i * 2 + 1] = brev8(Z[1]);
866     }
867     /* set tail elements to 1s */
868     vext_set_elems_1s(vd, vta, env->vl * 4, total_elems * 4);
869     env->vstart = 0;
870 }
871 
872 void HELPER(vsm4k_vi)(void *vd, void *vs2, uint32_t uimm5, CPURISCVState *env,
873                       uint32_t desc)
874 {
875     const uint32_t egs = 4;
876     uint32_t rnd = uimm5 & 0x7;
877     uint32_t group_start = env->vstart / egs;
878     uint32_t group_end = env->vl / egs;
879     uint32_t esz = sizeof(uint32_t);
880     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
881 
882     VSTART_CHECK_EARLY_EXIT(env);
883 
884     for (uint32_t i = group_start; i < group_end; ++i) {
885         uint32_t vstart = i * egs;
886         uint32_t vend = (i + 1) * egs;
887         uint32_t rk[4] = {0};
888         uint32_t tmp[8] = {0};
889 
890         for (uint32_t j = vstart; j < vend; ++j) {
891             rk[j - vstart] = *((uint32_t *)vs2 + H4(j));
892         }
893 
894         for (uint32_t j = 0; j < egs; ++j) {
895             tmp[j] = rk[j];
896         }
897 
898         for (uint32_t j = 0; j < egs; ++j) {
899             uint32_t b, s;
900             b = tmp[j + 1] ^ tmp[j + 2] ^ tmp[j + 3] ^ sm4_ck[rnd * 4 + j];
901 
902             s = sm4_subword(b);
903 
904             tmp[j + 4] = tmp[j] ^ (s ^ rol32(s, 13) ^ rol32(s, 23));
905         }
906 
907         for (uint32_t j = vstart; j < vend; ++j) {
908             *((uint32_t *)vd + H4(j)) = tmp[egs + (j - vstart)];
909         }
910     }
911 
912     env->vstart = 0;
913     /* set tail elements to 1s */
914     vext_set_elems_1s(vd, vext_vta(desc), env->vl * esz, total_elems * esz);
915 }
916 
917 static void do_sm4_round(uint32_t *rk, uint32_t *buf)
918 {
919     const uint32_t egs = 4;
920     uint32_t s, b;
921 
922     for (uint32_t j = egs; j < egs * 2; ++j) {
923         b = buf[j - 3] ^ buf[j - 2] ^ buf[j - 1] ^ rk[j - 4];
924 
925         s = sm4_subword(b);
926 
927         buf[j] = buf[j - 4] ^ (s ^ rol32(s, 2) ^ rol32(s, 10) ^ rol32(s, 18) ^
928                                rol32(s, 24));
929     }
930 }
931 
932 void HELPER(vsm4r_vv)(void *vd, void *vs2, CPURISCVState *env, uint32_t desc)
933 {
934     const uint32_t egs = 4;
935     uint32_t group_start = env->vstart / egs;
936     uint32_t group_end = env->vl / egs;
937     uint32_t esz = sizeof(uint32_t);
938     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
939 
940     VSTART_CHECK_EARLY_EXIT(env);
941 
942     for (uint32_t i = group_start; i < group_end; ++i) {
943         uint32_t vstart = i * egs;
944         uint32_t vend = (i + 1) * egs;
945         uint32_t rk[4] = {0};
946         uint32_t tmp[8] = {0};
947 
948         for (uint32_t j = vstart; j < vend; ++j) {
949             rk[j - vstart] = *((uint32_t *)vs2 + H4(j));
950         }
951 
952         for (uint32_t j = vstart; j < vend; ++j) {
953             tmp[j - vstart] = *((uint32_t *)vd + H4(j));
954         }
955 
956         do_sm4_round(rk, tmp);
957 
958         for (uint32_t j = vstart; j < vend; ++j) {
959             *((uint32_t *)vd + H4(j)) = tmp[egs + (j - vstart)];
960         }
961     }
962 
963     env->vstart = 0;
964     /* set tail elements to 1s */
965     vext_set_elems_1s(vd, vext_vta(desc), env->vl * esz, total_elems * esz);
966 }
967 
968 void HELPER(vsm4r_vs)(void *vd, void *vs2, CPURISCVState *env, uint32_t desc)
969 {
970     const uint32_t egs = 4;
971     uint32_t group_start = env->vstart / egs;
972     uint32_t group_end = env->vl / egs;
973     uint32_t esz = sizeof(uint32_t);
974     uint32_t total_elems = vext_get_total_elems(env, desc, esz);
975 
976     VSTART_CHECK_EARLY_EXIT(env);
977 
978     for (uint32_t i = group_start; i < group_end; ++i) {
979         uint32_t vstart = i * egs;
980         uint32_t vend = (i + 1) * egs;
981         uint32_t rk[4] = {0};
982         uint32_t tmp[8] = {0};
983 
984         for (uint32_t j = 0; j < egs; ++j) {
985             rk[j] = *((uint32_t *)vs2 + H4(j));
986         }
987 
988         for (uint32_t j = vstart; j < vend; ++j) {
989             tmp[j - vstart] = *((uint32_t *)vd + H4(j));
990         }
991 
992         do_sm4_round(rk, tmp);
993 
994         for (uint32_t j = vstart; j < vend; ++j) {
995             *((uint32_t *)vd + H4(j)) = tmp[egs + (j - vstart)];
996         }
997     }
998 
999     env->vstart = 0;
1000     /* set tail elements to 1s */
1001     vext_set_elems_1s(vd, vext_vta(desc), env->vl * esz, total_elems * esz);
1002 }
1003