| 1 | /* |
| 2 | * MMX/3DNow!/SSE/SSE2/SSE3/SSSE3/SSE4/PNI support |
| 3 | * |
| 4 | * Copyright (c) 2005 Fabrice Bellard |
| 5 | * |
| 6 | * This library is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU Lesser General Public |
| 8 | * License as published by the Free Software Foundation; either |
| 9 | * version 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This library is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. |
| 18 | */ |
| 19 | #if SHIFT == 0 |
| 20 | #define Reg MMXReg |
| 21 | #define SUFFIX _mmx |
| 22 | #else |
| 23 | #define Reg ZMMReg |
| 24 | #if SHIFT == 1 |
| 25 | #define SUFFIX _xmm |
| 26 | #else |
| 27 | #define SUFFIX _ymm |
| 28 | #endif |
| 29 | #endif |
| 30 | |
| 31 | #define dh_alias_Reg ptr |
| 32 | #define dh_alias_ZMMReg ptr |
| 33 | #define dh_alias_MMXReg ptr |
| 34 | #define dh_ctype_Reg Reg * |
| 35 | #define dh_ctype_ZMMReg ZMMReg * |
| 36 | #define dh_ctype_MMXReg MMXReg * |
| 37 | #define dh_typecode_Reg dh_typecode_ptr |
| 38 | #define dh_typecode_ZMMReg dh_typecode_ptr |
| 39 | #define dh_typecode_MMXReg dh_typecode_ptr |
| 40 | |
| 41 | DEF_HELPER_4(glue(psrlw, SUFFIX), void, env, Reg, Reg, Reg) |
| 42 | DEF_HELPER_4(glue(psraw, SUFFIX), void, env, Reg, Reg, Reg) |
| 43 | DEF_HELPER_4(glue(psllw, SUFFIX), void, env, Reg, Reg, Reg) |
| 44 | DEF_HELPER_4(glue(psrld, SUFFIX), void, env, Reg, Reg, Reg) |
| 45 | DEF_HELPER_4(glue(psrad, SUFFIX), void, env, Reg, Reg, Reg) |
| 46 | DEF_HELPER_4(glue(pslld, SUFFIX), void, env, Reg, Reg, Reg) |
| 47 | DEF_HELPER_4(glue(psrlq, SUFFIX), void, env, Reg, Reg, Reg) |
| 48 | DEF_HELPER_4(glue(psllq, SUFFIX), void, env, Reg, Reg, Reg) |
| 49 | |
| 50 | #if SHIFT >= 1 |
| 51 | DEF_HELPER_4(glue(psrldq, SUFFIX), void, env, Reg, Reg, Reg) |
| 52 | DEF_HELPER_4(glue(pslldq, SUFFIX), void, env, Reg, Reg, Reg) |
| 53 | #endif |
| 54 | |
| 55 | #define SSE_HELPER_B(name, F)\ |
| 56 | DEF_HELPER_4(glue(name, SUFFIX), void, env, Reg, Reg, Reg) |
| 57 | |
| 58 | #define SSE_HELPER_W(name, F)\ |
| 59 | DEF_HELPER_4(glue(name, SUFFIX), void, env, Reg, Reg, Reg) |
| 60 | |
| 61 | #define SSE_HELPER_L(name, F)\ |
| 62 | DEF_HELPER_4(glue(name, SUFFIX), void, env, Reg, Reg, Reg) |
| 63 | |
| 64 | #define SSE_HELPER_Q(name, F)\ |
| 65 | DEF_HELPER_4(glue(name, SUFFIX), void, env, Reg, Reg, Reg) |
| 66 | |
| 67 | #if SHIFT == 0 |
| 68 | DEF_HELPER_3(glue(pmulhrw, SUFFIX), void, env, Reg, Reg) |
| 69 | #endif |
| 70 | SSE_HELPER_W(pmulhuw, FMULHUW) |
| 71 | SSE_HELPER_W(pmulhw, FMULHW) |
| 72 | |
| 73 | SSE_HELPER_B(pavgb, FAVG) |
| 74 | SSE_HELPER_W(pavgw, FAVG) |
| 75 | |
| 76 | DEF_HELPER_4(glue(pmuludq, SUFFIX), void, env, Reg, Reg, Reg) |
| 77 | DEF_HELPER_4(glue(pmaddwd, SUFFIX), void, env, Reg, Reg, Reg) |
| 78 | |
| 79 | DEF_HELPER_4(glue(psadbw, SUFFIX), void, env, Reg, Reg, Reg) |
| 80 | #if SHIFT < 2 |
| 81 | DEF_HELPER_4(glue(maskmov, SUFFIX), void, env, Reg, Reg, tl) |
| 82 | #endif |
| 83 | |
| 84 | #if SHIFT == 0 |
| 85 | DEF_HELPER_3(glue(pshufw, SUFFIX), void, Reg, Reg, int) |
| 86 | #else |
| 87 | DEF_HELPER_3(glue(pshufd, SUFFIX), void, Reg, Reg, int) |
| 88 | DEF_HELPER_3(glue(pshuflw, SUFFIX), void, Reg, Reg, int) |
| 89 | DEF_HELPER_3(glue(pshufhw, SUFFIX), void, Reg, Reg, int) |
| 90 | #endif |
| 91 | |
| 92 | #if SHIFT >= 1 |
| 93 | /* FPU ops */ |
| 94 | /* XXX: not accurate */ |
| 95 | |
| 96 | #define SSE_HELPER_P4(name) \ |
| 97 | DEF_HELPER_4(glue(name ## ps, SUFFIX), void, env, Reg, Reg, Reg) \ |
| 98 | DEF_HELPER_4(glue(name ## pd, SUFFIX), void, env, Reg, Reg, Reg) |
| 99 | |
| 100 | #define SSE_HELPER_P3(name, ...) \ |
| 101 | DEF_HELPER_3(glue(name ## ps, SUFFIX), void, env, Reg, Reg) \ |
| 102 | DEF_HELPER_3(glue(name ## pd, SUFFIX), void, env, Reg, Reg) |
| 103 | |
| 104 | #if SHIFT == 1 |
| 105 | #define SSE_HELPER_S4(name) \ |
| 106 | SSE_HELPER_P4(name) \ |
| 107 | DEF_HELPER_4(name ## ss, void, env, Reg, Reg, Reg) \ |
| 108 | DEF_HELPER_4(name ## sd, void, env, Reg, Reg, Reg) |
| 109 | #define SSE_HELPER_S3(name) \ |
| 110 | SSE_HELPER_P3(name) \ |
| 111 | DEF_HELPER_4(name ## ss, void, env, Reg, Reg, Reg) \ |
| 112 | DEF_HELPER_4(name ## sd, void, env, Reg, Reg, Reg) |
| 113 | #else |
| 114 | #define SSE_HELPER_S4(name, ...) SSE_HELPER_P4(name) |
| 115 | #define SSE_HELPER_S3(name, ...) SSE_HELPER_P3(name) |
| 116 | #endif |
| 117 | |
| 118 | DEF_HELPER_4(glue(shufps, SUFFIX), void, Reg, Reg, Reg, int) |
| 119 | DEF_HELPER_4(glue(shufpd, SUFFIX), void, Reg, Reg, Reg, int) |
| 120 | |
| 121 | SSE_HELPER_S4(add) |
| 122 | SSE_HELPER_S4(sub) |
| 123 | SSE_HELPER_S4(mul) |
| 124 | SSE_HELPER_S4(div) |
| 125 | SSE_HELPER_S4(min) |
| 126 | SSE_HELPER_S4(max) |
| 127 | |
| 128 | SSE_HELPER_S3(sqrt) |
| 129 | |
| 130 | DEF_HELPER_3(glue(cvtps2pd, SUFFIX), void, env, Reg, Reg) |
| 131 | DEF_HELPER_3(glue(cvtpd2ps, SUFFIX), void, env, Reg, Reg) |
| 132 | DEF_HELPER_3(glue(cvtdq2ps, SUFFIX), void, env, Reg, Reg) |
| 133 | DEF_HELPER_3(glue(cvtdq2pd, SUFFIX), void, env, Reg, Reg) |
| 134 | |
| 135 | DEF_HELPER_3(glue(cvtps2dq, SUFFIX), void, env, ZMMReg, ZMMReg) |
| 136 | DEF_HELPER_3(glue(cvtpd2dq, SUFFIX), void, env, ZMMReg, ZMMReg) |
| 137 | |
| 138 | DEF_HELPER_3(glue(cvttps2dq, SUFFIX), void, env, ZMMReg, ZMMReg) |
| 139 | DEF_HELPER_3(glue(cvttpd2dq, SUFFIX), void, env, ZMMReg, ZMMReg) |
| 140 | |
| 141 | #if SHIFT == 1 |
| 142 | DEF_HELPER_4(cvtss2sd, void, env, Reg, Reg, Reg) |
| 143 | DEF_HELPER_4(cvtsd2ss, void, env, Reg, Reg, Reg) |
| 144 | DEF_HELPER_3(cvtpi2ps, void, env, ZMMReg, MMXReg) |
| 145 | DEF_HELPER_3(cvtpi2pd, void, env, ZMMReg, MMXReg) |
| 146 | DEF_HELPER_3(cvtsi2ss, void, env, ZMMReg, i32) |
| 147 | DEF_HELPER_3(cvtsi2sd, void, env, ZMMReg, i32) |
| 148 | |
| 149 | #ifdef TARGET_X86_64 |
| 150 | DEF_HELPER_3(cvtsq2ss, void, env, ZMMReg, i64) |
| 151 | DEF_HELPER_3(cvtsq2sd, void, env, ZMMReg, i64) |
| 152 | #endif |
| 153 | |
| 154 | DEF_HELPER_3(cvtps2pi, void, env, MMXReg, ZMMReg) |
| 155 | DEF_HELPER_3(cvtpd2pi, void, env, MMXReg, ZMMReg) |
| 156 | DEF_HELPER_2(cvtss2si, s32, env, ZMMReg) |
| 157 | DEF_HELPER_2(cvtsd2si, s32, env, ZMMReg) |
| 158 | #ifdef TARGET_X86_64 |
| 159 | DEF_HELPER_2(cvtss2sq, s64, env, ZMMReg) |
| 160 | DEF_HELPER_2(cvtsd2sq, s64, env, ZMMReg) |
| 161 | #endif |
| 162 | |
| 163 | DEF_HELPER_3(cvttps2pi, void, env, MMXReg, ZMMReg) |
| 164 | DEF_HELPER_3(cvttpd2pi, void, env, MMXReg, ZMMReg) |
| 165 | DEF_HELPER_2(cvttss2si, s32, env, ZMMReg) |
| 166 | DEF_HELPER_2(cvttsd2si, s32, env, ZMMReg) |
| 167 | #ifdef TARGET_X86_64 |
| 168 | DEF_HELPER_2(cvttss2sq, s64, env, ZMMReg) |
| 169 | DEF_HELPER_2(cvttsd2sq, s64, env, ZMMReg) |
| 170 | #endif |
| 171 | #endif |
| 172 | |
| 173 | DEF_HELPER_3(glue(rsqrtps, SUFFIX), void, env, ZMMReg, ZMMReg) |
| 174 | DEF_HELPER_3(glue(rcpps, SUFFIX), void, env, ZMMReg, ZMMReg) |
| 175 | |
| 176 | #if SHIFT == 1 |
| 177 | DEF_HELPER_4(rsqrtss, void, env, ZMMReg, ZMMReg, ZMMReg) |
| 178 | DEF_HELPER_4(rcpss, void, env, ZMMReg, ZMMReg, ZMMReg) |
| 179 | DEF_HELPER_3(extrq_r, void, env, ZMMReg, ZMMReg) |
| 180 | DEF_HELPER_4(extrq_i, void, env, ZMMReg, int, int) |
| 181 | DEF_HELPER_3(insertq_r, void, env, ZMMReg, ZMMReg) |
| 182 | DEF_HELPER_5(insertq_i, void, env, ZMMReg, ZMMReg, int, int) |
| 183 | #endif |
| 184 | |
| 185 | SSE_HELPER_P4(hadd) |
| 186 | SSE_HELPER_P4(hsub) |
| 187 | SSE_HELPER_P4(addsub) |
| 188 | |
| 189 | #define SSE_HELPER_CMP(name, F, C) SSE_HELPER_S4(name) |
| 190 | |
| 191 | SSE_HELPER_CMP(cmpeq, FPU_CMPQ, FPU_EQ) |
| 192 | SSE_HELPER_CMP(cmplt, FPU_CMPS, FPU_LT) |
| 193 | SSE_HELPER_CMP(cmple, FPU_CMPS, FPU_LE) |
| 194 | SSE_HELPER_CMP(cmpunord, FPU_CMPQ, FPU_UNORD) |
| 195 | SSE_HELPER_CMP(cmpneq, FPU_CMPQ, !FPU_EQ) |
| 196 | SSE_HELPER_CMP(cmpnlt, FPU_CMPS, !FPU_LT) |
| 197 | SSE_HELPER_CMP(cmpnle, FPU_CMPS, !FPU_LE) |
| 198 | SSE_HELPER_CMP(cmpord, FPU_CMPQ, !FPU_UNORD) |
| 199 | |
| 200 | SSE_HELPER_CMP(cmpequ, FPU_CMPQ, FPU_EQU) |
| 201 | SSE_HELPER_CMP(cmpnge, FPU_CMPS, !FPU_GE) |
| 202 | SSE_HELPER_CMP(cmpngt, FPU_CMPS, !FPU_GT) |
| 203 | SSE_HELPER_CMP(cmpfalse, FPU_CMPQ, FPU_FALSE) |
| 204 | SSE_HELPER_CMP(cmpnequ, FPU_CMPQ, !FPU_EQU) |
| 205 | SSE_HELPER_CMP(cmpge, FPU_CMPS, FPU_GE) |
| 206 | SSE_HELPER_CMP(cmpgt, FPU_CMPS, FPU_GT) |
| 207 | SSE_HELPER_CMP(cmptrue, FPU_CMPQ, !FPU_FALSE) |
| 208 | |
| 209 | SSE_HELPER_CMP(cmpeqs, FPU_CMPS, FPU_EQ) |
| 210 | SSE_HELPER_CMP(cmpltq, FPU_CMPQ, FPU_LT) |
| 211 | SSE_HELPER_CMP(cmpleq, FPU_CMPQ, FPU_LE) |
| 212 | SSE_HELPER_CMP(cmpunords, FPU_CMPS, FPU_UNORD) |
| 213 | SSE_HELPER_CMP(cmpneqq, FPU_CMPS, !FPU_EQ) |
| 214 | SSE_HELPER_CMP(cmpnltq, FPU_CMPQ, !FPU_LT) |
| 215 | SSE_HELPER_CMP(cmpnleq, FPU_CMPQ, !FPU_LE) |
| 216 | SSE_HELPER_CMP(cmpords, FPU_CMPS, !FPU_UNORD) |
| 217 | |
| 218 | SSE_HELPER_CMP(cmpequs, FPU_CMPS, FPU_EQU) |
| 219 | SSE_HELPER_CMP(cmpngeq, FPU_CMPQ, !FPU_GE) |
| 220 | SSE_HELPER_CMP(cmpngtq, FPU_CMPQ, !FPU_GT) |
| 221 | SSE_HELPER_CMP(cmpfalses, FPU_CMPS, FPU_FALSE) |
| 222 | SSE_HELPER_CMP(cmpnequs, FPU_CMPS, !FPU_EQU) |
| 223 | SSE_HELPER_CMP(cmpgeq, FPU_CMPQ, FPU_GE) |
| 224 | SSE_HELPER_CMP(cmpgtq, FPU_CMPQ, FPU_GT) |
| 225 | SSE_HELPER_CMP(cmptrues, FPU_CMPS, !FPU_FALSE) |
| 226 | |
| 227 | #if SHIFT == 1 |
| 228 | DEF_HELPER_3(ucomiss, void, env, Reg, Reg) |
| 229 | DEF_HELPER_3(comiss, void, env, Reg, Reg) |
| 230 | DEF_HELPER_3(ucomisd, void, env, Reg, Reg) |
| 231 | DEF_HELPER_3(comisd, void, env, Reg, Reg) |
| 232 | #endif |
| 233 | |
| 234 | DEF_HELPER_2(glue(movmskps, SUFFIX), i32, env, Reg) |
| 235 | DEF_HELPER_2(glue(movmskpd, SUFFIX), i32, env, Reg) |
| 236 | #endif |
| 237 | |
| 238 | DEF_HELPER_4(glue(packsswb, SUFFIX), void, env, Reg, Reg, Reg) |
| 239 | DEF_HELPER_4(glue(packuswb, SUFFIX), void, env, Reg, Reg, Reg) |
| 240 | DEF_HELPER_4(glue(packssdw, SUFFIX), void, env, Reg, Reg, Reg) |
| 241 | #define UNPCK_OP(name, base) \ |
| 242 | DEF_HELPER_4(glue(punpck ## name ## bw, SUFFIX), void, env, Reg, Reg, Reg) \ |
| 243 | DEF_HELPER_4(glue(punpck ## name ## wd, SUFFIX), void, env, Reg, Reg, Reg) \ |
| 244 | DEF_HELPER_4(glue(punpck ## name ## dq, SUFFIX), void, env, Reg, Reg, Reg) |
| 245 | |
| 246 | UNPCK_OP(l, 0) |
| 247 | UNPCK_OP(h, 1) |
| 248 | |
| 249 | #if SHIFT >= 1 |
| 250 | DEF_HELPER_4(glue(punpcklqdq, SUFFIX), void, env, Reg, Reg, Reg) |
| 251 | DEF_HELPER_4(glue(punpckhqdq, SUFFIX), void, env, Reg, Reg, Reg) |
| 252 | #endif |
| 253 | |
| 254 | /* 3DNow! float ops */ |
| 255 | #if SHIFT == 0 |
| 256 | DEF_HELPER_3(pi2fd, void, env, MMXReg, MMXReg) |
| 257 | DEF_HELPER_3(pi2fw, void, env, MMXReg, MMXReg) |
| 258 | DEF_HELPER_3(pf2id, void, env, MMXReg, MMXReg) |
| 259 | DEF_HELPER_3(pf2iw, void, env, MMXReg, MMXReg) |
| 260 | DEF_HELPER_3(pfacc, void, env, MMXReg, MMXReg) |
| 261 | DEF_HELPER_3(pfadd, void, env, MMXReg, MMXReg) |
| 262 | DEF_HELPER_3(pfcmpeq, void, env, MMXReg, MMXReg) |
| 263 | DEF_HELPER_3(pfcmpge, void, env, MMXReg, MMXReg) |
| 264 | DEF_HELPER_3(pfcmpgt, void, env, MMXReg, MMXReg) |
| 265 | DEF_HELPER_3(pfmax, void, env, MMXReg, MMXReg) |
| 266 | DEF_HELPER_3(pfmin, void, env, MMXReg, MMXReg) |
| 267 | DEF_HELPER_3(pfmul, void, env, MMXReg, MMXReg) |
| 268 | DEF_HELPER_3(pfnacc, void, env, MMXReg, MMXReg) |
| 269 | DEF_HELPER_3(pfpnacc, void, env, MMXReg, MMXReg) |
| 270 | DEF_HELPER_3(pfrcp, void, env, MMXReg, MMXReg) |
| 271 | DEF_HELPER_3(pfrsqrt, void, env, MMXReg, MMXReg) |
| 272 | DEF_HELPER_3(pfsub, void, env, MMXReg, MMXReg) |
| 273 | DEF_HELPER_3(pfsubr, void, env, MMXReg, MMXReg) |
| 274 | DEF_HELPER_3(pswapd, void, env, MMXReg, MMXReg) |
| 275 | #endif |
| 276 | |
| 277 | /* SSSE3 op helpers */ |
| 278 | DEF_HELPER_4(glue(phaddw, SUFFIX), void, env, Reg, Reg, Reg) |
| 279 | DEF_HELPER_4(glue(phaddd, SUFFIX), void, env, Reg, Reg, Reg) |
| 280 | DEF_HELPER_4(glue(phaddsw, SUFFIX), void, env, Reg, Reg, Reg) |
| 281 | DEF_HELPER_4(glue(phsubw, SUFFIX), void, env, Reg, Reg, Reg) |
| 282 | DEF_HELPER_4(glue(phsubd, SUFFIX), void, env, Reg, Reg, Reg) |
| 283 | DEF_HELPER_4(glue(phsubsw, SUFFIX), void, env, Reg, Reg, Reg) |
| 284 | DEF_HELPER_4(glue(pmaddubsw, SUFFIX), void, env, Reg, Reg, Reg) |
| 285 | DEF_HELPER_4(glue(pmulhrsw, SUFFIX), void, env, Reg, Reg, Reg) |
| 286 | DEF_HELPER_4(glue(pshufb, SUFFIX), void, env, Reg, Reg, Reg) |
| 287 | DEF_HELPER_4(glue(psignb, SUFFIX), void, env, Reg, Reg, Reg) |
| 288 | DEF_HELPER_4(glue(psignw, SUFFIX), void, env, Reg, Reg, Reg) |
| 289 | DEF_HELPER_4(glue(psignd, SUFFIX), void, env, Reg, Reg, Reg) |
| 290 | DEF_HELPER_5(glue(palignr, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 291 | |
| 292 | /* SSE4.1 op helpers */ |
| 293 | #if SHIFT >= 1 |
| 294 | DEF_HELPER_5(glue(pblendvb, SUFFIX), void, env, Reg, Reg, Reg, Reg) |
| 295 | DEF_HELPER_5(glue(blendvps, SUFFIX), void, env, Reg, Reg, Reg, Reg) |
| 296 | DEF_HELPER_5(glue(blendvpd, SUFFIX), void, env, Reg, Reg, Reg, Reg) |
| 297 | DEF_HELPER_3(glue(ptest, SUFFIX), void, env, Reg, Reg) |
| 298 | DEF_HELPER_3(glue(pmovsxbw, SUFFIX), void, env, Reg, Reg) |
| 299 | DEF_HELPER_3(glue(pmovsxbd, SUFFIX), void, env, Reg, Reg) |
| 300 | DEF_HELPER_3(glue(pmovsxbq, SUFFIX), void, env, Reg, Reg) |
| 301 | DEF_HELPER_3(glue(pmovsxwd, SUFFIX), void, env, Reg, Reg) |
| 302 | DEF_HELPER_3(glue(pmovsxwq, SUFFIX), void, env, Reg, Reg) |
| 303 | DEF_HELPER_3(glue(pmovsxdq, SUFFIX), void, env, Reg, Reg) |
| 304 | DEF_HELPER_3(glue(pmovzxbw, SUFFIX), void, env, Reg, Reg) |
| 305 | DEF_HELPER_3(glue(pmovzxbd, SUFFIX), void, env, Reg, Reg) |
| 306 | DEF_HELPER_3(glue(pmovzxbq, SUFFIX), void, env, Reg, Reg) |
| 307 | DEF_HELPER_3(glue(pmovzxwd, SUFFIX), void, env, Reg, Reg) |
| 308 | DEF_HELPER_3(glue(pmovzxwq, SUFFIX), void, env, Reg, Reg) |
| 309 | DEF_HELPER_3(glue(pmovzxdq, SUFFIX), void, env, Reg, Reg) |
| 310 | DEF_HELPER_3(glue(pmovsldup, SUFFIX), void, env, Reg, Reg) |
| 311 | DEF_HELPER_3(glue(pmovshdup, SUFFIX), void, env, Reg, Reg) |
| 312 | DEF_HELPER_3(glue(pmovdldup, SUFFIX), void, env, Reg, Reg) |
| 313 | DEF_HELPER_4(glue(pmuldq, SUFFIX), void, env, Reg, Reg, Reg) |
| 314 | DEF_HELPER_4(glue(packusdw, SUFFIX), void, env, Reg, Reg, Reg) |
| 315 | #if SHIFT == 1 |
| 316 | DEF_HELPER_3(glue(phminposuw, SUFFIX), void, env, Reg, Reg) |
| 317 | #endif |
| 318 | DEF_HELPER_4(glue(roundps, SUFFIX), void, env, Reg, Reg, i32) |
| 319 | DEF_HELPER_4(glue(roundpd, SUFFIX), void, env, Reg, Reg, i32) |
| 320 | #if SHIFT == 1 |
| 321 | DEF_HELPER_5(roundss_xmm, void, env, Reg, Reg, Reg, i32) |
| 322 | DEF_HELPER_5(roundsd_xmm, void, env, Reg, Reg, Reg, i32) |
| 323 | #endif |
| 324 | DEF_HELPER_5(glue(blendps, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 325 | DEF_HELPER_5(glue(blendpd, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 326 | DEF_HELPER_5(glue(pblendw, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 327 | DEF_HELPER_5(glue(dpps, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 328 | #if SHIFT == 1 |
| 329 | DEF_HELPER_5(glue(dppd, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 330 | #endif |
| 331 | DEF_HELPER_5(glue(mpsadbw, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 332 | #endif |
| 333 | |
| 334 | /* SSE4.2 op helpers */ |
| 335 | #if SHIFT == 1 |
| 336 | DEF_HELPER_4(glue(pcmpestri, SUFFIX), void, env, Reg, Reg, i32) |
| 337 | DEF_HELPER_4(glue(pcmpestrm, SUFFIX), void, env, Reg, Reg, i32) |
| 338 | DEF_HELPER_4(glue(pcmpistri, SUFFIX), void, env, Reg, Reg, i32) |
| 339 | DEF_HELPER_4(glue(pcmpistrm, SUFFIX), void, env, Reg, Reg, i32) |
| 340 | DEF_HELPER_3(crc32, tl, i32, tl, i32) |
| 341 | #endif |
| 342 | |
| 343 | /* AES-NI op helpers */ |
| 344 | #if SHIFT >= 1 |
| 345 | DEF_HELPER_4(glue(aesdec, SUFFIX), void, env, Reg, Reg, Reg) |
| 346 | DEF_HELPER_4(glue(aesdeclast, SUFFIX), void, env, Reg, Reg, Reg) |
| 347 | DEF_HELPER_4(glue(aesenc, SUFFIX), void, env, Reg, Reg, Reg) |
| 348 | DEF_HELPER_4(glue(aesenclast, SUFFIX), void, env, Reg, Reg, Reg) |
| 349 | #if SHIFT == 1 |
| 350 | DEF_HELPER_3(glue(aesimc, SUFFIX), void, env, Reg, Reg) |
| 351 | DEF_HELPER_4(glue(aeskeygenassist, SUFFIX), void, env, Reg, Reg, i32) |
| 352 | #endif |
| 353 | DEF_HELPER_5(glue(pclmulqdq, SUFFIX), void, env, Reg, Reg, Reg, i32) |
| 354 | #endif |
| 355 | |
| 356 | /* F16C helpers */ |
| 357 | #if SHIFT >= 1 |
| 358 | DEF_HELPER_3(glue(cvtph2ps, SUFFIX), void, env, Reg, Reg) |
| 359 | DEF_HELPER_4(glue(cvtps2ph, SUFFIX), void, env, Reg, Reg, int) |
| 360 | #endif |
| 361 | |
| 362 | /* FMA3 helpers */ |
| 363 | #if SHIFT == 1 |
| 364 | DEF_HELPER_6(fma4ss, void, env, Reg, Reg, Reg, Reg, int) |
| 365 | DEF_HELPER_6(fma4sd, void, env, Reg, Reg, Reg, Reg, int) |
| 366 | #endif |
| 367 | |
| 368 | #if SHIFT >= 1 |
| 369 | DEF_HELPER_7(glue(fma4ps, SUFFIX), void, env, Reg, Reg, Reg, Reg, int, int) |
| 370 | DEF_HELPER_7(glue(fma4pd, SUFFIX), void, env, Reg, Reg, Reg, Reg, int, int) |
| 371 | #endif |
| 372 | |
| 373 | /* AVX helpers */ |
| 374 | #if SHIFT >= 1 |
| 375 | DEF_HELPER_4(glue(vpermilpd, SUFFIX), void, env, Reg, Reg, Reg) |
| 376 | DEF_HELPER_4(glue(vpermilps, SUFFIX), void, env, Reg, Reg, Reg) |
| 377 | DEF_HELPER_3(glue(vpermilpd_imm, SUFFIX), void, Reg, Reg, i32) |
| 378 | DEF_HELPER_3(glue(vpermilps_imm, SUFFIX), void, Reg, Reg, i32) |
| 379 | DEF_HELPER_4(glue(vpsrlvd, SUFFIX), void, env, Reg, Reg, Reg) |
| 380 | DEF_HELPER_4(glue(vpsravd, SUFFIX), void, env, Reg, Reg, Reg) |
| 381 | DEF_HELPER_4(glue(vpsllvd, SUFFIX), void, env, Reg, Reg, Reg) |
| 382 | DEF_HELPER_4(glue(vpsrlvq, SUFFIX), void, env, Reg, Reg, Reg) |
| 383 | DEF_HELPER_4(glue(vpsravq, SUFFIX), void, env, Reg, Reg, Reg) |
| 384 | DEF_HELPER_4(glue(vpsllvq, SUFFIX), void, env, Reg, Reg, Reg) |
| 385 | DEF_HELPER_3(glue(vtestps, SUFFIX), void, env, Reg, Reg) |
| 386 | DEF_HELPER_3(glue(vtestpd, SUFFIX), void, env, Reg, Reg) |
| 387 | DEF_HELPER_4(glue(vpmaskmovd_st, SUFFIX), void, env, Reg, Reg, tl) |
| 388 | DEF_HELPER_4(glue(vpmaskmovq_st, SUFFIX), void, env, Reg, Reg, tl) |
| 389 | DEF_HELPER_4(glue(vpmaskmovd, SUFFIX), void, env, Reg, Reg, Reg) |
| 390 | DEF_HELPER_4(glue(vpmaskmovq, SUFFIX), void, env, Reg, Reg, Reg) |
| 391 | DEF_HELPER_7(glue(vpgatherdd, SUFFIX), void, env, Reg, Reg, Reg, tl, i32, tl) |
| 392 | DEF_HELPER_7(glue(vpgatherdq, SUFFIX), void, env, Reg, Reg, Reg, tl, i32, tl) |
| 393 | DEF_HELPER_7(glue(vpgatherqd, SUFFIX), void, env, Reg, Reg, Reg, tl, i32, tl) |
| 394 | DEF_HELPER_7(glue(vpgatherqq, SUFFIX), void, env, Reg, Reg, Reg, tl, i32, tl) |
| 395 | #if SHIFT == 2 |
| 396 | DEF_HELPER_3(vpermd_ymm, void, Reg, Reg, Reg) |
| 397 | DEF_HELPER_4(vpermdq_ymm, void, Reg, Reg, Reg, i32) |
| 398 | DEF_HELPER_3(vpermq_ymm, void, Reg, Reg, i32) |
| 399 | #endif |
| 400 | #endif |
| 401 | |
| 402 | /* SHA helpers */ |
| 403 | #if SHIFT == 1 |
| 404 | DEF_HELPER_3(sha1rnds4_f0, void, Reg, Reg, Reg) |
| 405 | DEF_HELPER_3(sha1rnds4_f1, void, Reg, Reg, Reg) |
| 406 | DEF_HELPER_3(sha1rnds4_f2, void, Reg, Reg, Reg) |
| 407 | DEF_HELPER_3(sha1rnds4_f3, void, Reg, Reg, Reg) |
| 408 | DEF_HELPER_3(sha1nexte, void, Reg, Reg, Reg) |
| 409 | DEF_HELPER_3(sha1msg1, void, Reg, Reg, Reg) |
| 410 | DEF_HELPER_3(sha1msg2, void, Reg, Reg, Reg) |
| 411 | DEF_HELPER_5(sha256rnds2, void, Reg, Reg, Reg, i32, i32) |
| 412 | DEF_HELPER_3(sha256msg1, void, Reg, Reg, Reg) |
| 413 | DEF_HELPER_3(sha256msg2, void, Reg, Reg, Reg) |
| 414 | #endif |
| 415 | |
| 416 | #undef SHIFT |
| 417 | #undef Reg |
| 418 | #undef SUFFIX |
| 419 | |
| 420 | #undef SSE_HELPER_B |
| 421 | #undef SSE_HELPER_W |
| 422 | #undef SSE_HELPER_L |
| 423 | #undef SSE_HELPER_Q |
| 424 | #undef SSE_HELPER_S3 |
| 425 | #undef SSE_HELPER_S4 |
| 426 | #undef SSE_HELPER_P3 |
| 427 | #undef SSE_HELPER_P4 |
| 428 | #undef SSE_HELPER_CMP |
| 429 | #undef UNPCK_OP |