1 /* 2 * QEMU TCG support -- s390x vector floating point instruction support 3 * 4 * Copyright (C) 2019 Red Hat Inc 5 * 6 * Authors: 7 * David Hildenbrand <david@redhat.com> 8 * 9 * This work is licensed under the terms of the GNU GPL, version 2 or later. 10 * See the COPYING file in the top-level directory. 11 */ 12 #include "qemu/osdep.h" 13 #include "qemu-common.h" 14 #include "cpu.h" 15 #include "internal.h" 16 #include "vec.h" 17 #include "tcg_s390x.h" 18 #include "tcg/tcg-gvec-desc.h" 19 #include "exec/exec-all.h" 20 #include "exec/helper-proto.h" 21 #include "fpu/softfloat.h" 22 23 #define VIC_INVALID 0x1 24 #define VIC_DIVBYZERO 0x2 25 #define VIC_OVERFLOW 0x3 26 #define VIC_UNDERFLOW 0x4 27 #define VIC_INEXACT 0x5 28 29 /* returns the VEX. If the VEX is 0, there is no trap */ 30 static uint8_t check_ieee_exc(CPUS390XState *env, uint8_t enr, bool XxC, 31 uint8_t *vec_exc) 32 { 33 uint8_t vece_exc = 0, trap_exc; 34 unsigned qemu_exc; 35 36 /* Retrieve and clear the softfloat exceptions */ 37 qemu_exc = env->fpu_status.float_exception_flags; 38 if (qemu_exc == 0) { 39 return 0; 40 } 41 env->fpu_status.float_exception_flags = 0; 42 43 vece_exc = s390_softfloat_exc_to_ieee(qemu_exc); 44 45 /* Add them to the vector-wide s390x exception bits */ 46 *vec_exc |= vece_exc; 47 48 /* Check for traps and construct the VXC */ 49 trap_exc = vece_exc & env->fpc >> 24; 50 if (trap_exc) { 51 if (trap_exc & S390_IEEE_MASK_INVALID) { 52 return enr << 4 | VIC_INVALID; 53 } else if (trap_exc & S390_IEEE_MASK_DIVBYZERO) { 54 return enr << 4 | VIC_DIVBYZERO; 55 } else if (trap_exc & S390_IEEE_MASK_OVERFLOW) { 56 return enr << 4 | VIC_OVERFLOW; 57 } else if (trap_exc & S390_IEEE_MASK_UNDERFLOW) { 58 return enr << 4 | VIC_UNDERFLOW; 59 } else if (!XxC) { 60 g_assert(trap_exc & S390_IEEE_MASK_INEXACT); 61 /* inexact has lowest priority on traps */ 62 return enr << 4 | VIC_INEXACT; 63 } 64 } 65 return 0; 66 } 67 68 static void handle_ieee_exc(CPUS390XState *env, uint8_t vxc, uint8_t vec_exc, 69 uintptr_t retaddr) 70 { 71 if (vxc) { 72 /* on traps, the fpc flags are not updated, instruction is suppressed */ 73 tcg_s390_vector_exception(env, vxc, retaddr); 74 } 75 if (vec_exc) { 76 /* indicate exceptions for all elements combined */ 77 env->fpc |= vec_exc << 16; 78 } 79 } 80 81 static float64 s390_vec_read_float64(const S390Vector *v, uint8_t enr) 82 { 83 return make_float64(s390_vec_read_element64(v, enr)); 84 } 85 86 static void s390_vec_write_float64(S390Vector *v, uint8_t enr, float64 data) 87 { 88 return s390_vec_write_element64(v, enr, data); 89 } 90 91 typedef float64 (*vop64_2_fn)(float64 a, float_status *s); 92 static void vop64_2(S390Vector *v1, const S390Vector *v2, CPUS390XState *env, 93 bool s, bool XxC, uint8_t erm, vop64_2_fn fn, 94 uintptr_t retaddr) 95 { 96 uint8_t vxc, vec_exc = 0; 97 S390Vector tmp = {}; 98 int i, old_mode; 99 100 old_mode = s390_swap_bfp_rounding_mode(env, erm); 101 for (i = 0; i < 2; i++) { 102 const float64 a = s390_vec_read_float64(v2, i); 103 104 s390_vec_write_float64(&tmp, i, fn(a, &env->fpu_status)); 105 vxc = check_ieee_exc(env, i, XxC, &vec_exc); 106 if (s || vxc) { 107 break; 108 } 109 } 110 s390_restore_bfp_rounding_mode(env, old_mode); 111 handle_ieee_exc(env, vxc, vec_exc, retaddr); 112 *v1 = tmp; 113 } 114 115 static float64 vcdg64(float64 a, float_status *s) 116 { 117 return int64_to_float64(a, s); 118 } 119 120 static float64 vcdlg64(float64 a, float_status *s) 121 { 122 return uint64_to_float64(a, s); 123 } 124 125 static float64 vcgd64(float64 a, float_status *s) 126 { 127 const float64 tmp = float64_to_int64(a, s); 128 129 return float64_is_any_nan(a) ? INT64_MIN : tmp; 130 } 131 132 static float64 vclgd64(float64 a, float_status *s) 133 { 134 const float64 tmp = float64_to_uint64(a, s); 135 136 return float64_is_any_nan(a) ? 0 : tmp; 137 } 138 139 #define DEF_GVEC_VOP2_FN(NAME, FN, BITS) \ 140 void HELPER(gvec_##NAME##BITS)(void *v1, const void *v2, CPUS390XState *env, \ 141 uint32_t desc) \ 142 { \ 143 const uint8_t erm = extract32(simd_data(desc), 4, 4); \ 144 const bool se = extract32(simd_data(desc), 3, 1); \ 145 const bool XxC = extract32(simd_data(desc), 2, 1); \ 146 \ 147 vop##BITS##_2(v1, v2, env, se, XxC, erm, FN, GETPC()); \ 148 } 149 150 #define DEF_GVEC_VOP2_64(NAME) \ 151 DEF_GVEC_VOP2_FN(NAME, NAME##64, 64) 152 153 #define DEF_GVEC_VOP2(NAME, OP) \ 154 DEF_GVEC_VOP2_FN(NAME, float64_##OP, 64) 155 156 DEF_GVEC_VOP2_64(vcdg) 157 DEF_GVEC_VOP2_64(vcdlg) 158 DEF_GVEC_VOP2_64(vcgd) 159 DEF_GVEC_VOP2_64(vclgd) 160 DEF_GVEC_VOP2(vfi, round_to_int) 161 DEF_GVEC_VOP2(vfsq, sqrt) 162 163 typedef float64 (*vop64_3_fn)(float64 a, float64 b, float_status *s); 164 static void vop64_3(S390Vector *v1, const S390Vector *v2, const S390Vector *v3, 165 CPUS390XState *env, bool s, vop64_3_fn fn, 166 uintptr_t retaddr) 167 { 168 uint8_t vxc, vec_exc = 0; 169 S390Vector tmp = {}; 170 int i; 171 172 for (i = 0; i < 2; i++) { 173 const float64 a = s390_vec_read_float64(v2, i); 174 const float64 b = s390_vec_read_float64(v3, i); 175 176 s390_vec_write_float64(&tmp, i, fn(a, b, &env->fpu_status)); 177 vxc = check_ieee_exc(env, i, false, &vec_exc); 178 if (s || vxc) { 179 break; 180 } 181 } 182 handle_ieee_exc(env, vxc, vec_exc, retaddr); 183 *v1 = tmp; 184 } 185 186 #define DEF_GVEC_VOP3(NAME, OP) \ 187 void HELPER(gvec_##NAME##64)(void *v1, const void *v2, const void *v3, \ 188 CPUS390XState *env, uint32_t desc) \ 189 { \ 190 const bool se = extract32(simd_data(desc), 3, 1); \ 191 \ 192 vop64_3(v1, v2, v3, env, se, float64_##OP, GETPC()); \ 193 } 194 195 DEF_GVEC_VOP3(vfa, add) 196 DEF_GVEC_VOP3(vfs, sub) 197 DEF_GVEC_VOP3(vfd, div) 198 DEF_GVEC_VOP3(vfm, mul) 199 200 static int wfc64(const S390Vector *v1, const S390Vector *v2, 201 CPUS390XState *env, bool signal, uintptr_t retaddr) 202 { 203 /* only the zero-indexed elements are compared */ 204 const float64 a = s390_vec_read_element64(v1, 0); 205 const float64 b = s390_vec_read_element64(v2, 0); 206 uint8_t vxc, vec_exc = 0; 207 int cmp; 208 209 if (signal) { 210 cmp = float64_compare(a, b, &env->fpu_status); 211 } else { 212 cmp = float64_compare_quiet(a, b, &env->fpu_status); 213 } 214 vxc = check_ieee_exc(env, 0, false, &vec_exc); 215 handle_ieee_exc(env, vxc, vec_exc, retaddr); 216 217 return float_comp_to_cc(env, cmp); 218 } 219 220 void HELPER(gvec_wfc64)(const void *v1, const void *v2, CPUS390XState *env, 221 uint32_t desc) 222 { 223 env->cc_op = wfc64(v1, v2, env, false, GETPC()); 224 } 225 226 void HELPER(gvec_wfk64)(const void *v1, const void *v2, CPUS390XState *env, 227 uint32_t desc) 228 { 229 env->cc_op = wfc64(v1, v2, env, true, GETPC()); 230 } 231 232 typedef bool (*vfc64_fn)(float64 a, float64 b, float_status *status); 233 static int vfc64(S390Vector *v1, const S390Vector *v2, const S390Vector *v3, 234 CPUS390XState *env, bool s, vfc64_fn fn, uintptr_t retaddr) 235 { 236 uint8_t vxc, vec_exc = 0; 237 S390Vector tmp = {}; 238 int match = 0; 239 int i; 240 241 for (i = 0; i < 2; i++) { 242 const float64 a = s390_vec_read_float64(v2, i); 243 const float64 b = s390_vec_read_float64(v3, i); 244 245 /* swap the order of the parameters, so we can use existing functions */ 246 if (fn(b, a, &env->fpu_status)) { 247 match++; 248 s390_vec_write_element64(&tmp, i, -1ull); 249 } 250 vxc = check_ieee_exc(env, i, false, &vec_exc); 251 if (s || vxc) { 252 break; 253 } 254 } 255 256 handle_ieee_exc(env, vxc, vec_exc, retaddr); 257 *v1 = tmp; 258 if (match) { 259 return s || match == 2 ? 0 : 1; 260 } 261 return 3; 262 } 263 264 #define DEF_GVEC_VFC_B(NAME, OP, BITS) \ 265 void HELPER(gvec_##NAME##BITS)(void *v1, const void *v2, const void *v3, \ 266 CPUS390XState *env, uint32_t desc) \ 267 { \ 268 const bool se = extract32(simd_data(desc), 3, 1); \ 269 vfc##BITS##_fn fn = float##BITS##_##OP##_quiet; \ 270 \ 271 vfc##BITS(v1, v2, v3, env, se, fn, GETPC()); \ 272 } \ 273 \ 274 void HELPER(gvec_##NAME##BITS##_cc)(void *v1, const void *v2, const void *v3, \ 275 CPUS390XState *env, uint32_t desc) \ 276 { \ 277 const bool se = extract32(simd_data(desc), 3, 1); \ 278 vfc##BITS##_fn fn = float##BITS##_##OP##_quiet; \ 279 \ 280 env->cc_op = vfc##BITS(v1, v2, v3, env, se, fn, GETPC()); \ 281 } 282 283 #define DEF_GVEC_VFC(NAME, OP) \ 284 DEF_GVEC_VFC_B(NAME, OP, 64) 285 286 DEF_GVEC_VFC(vfce, eq) 287 DEF_GVEC_VFC(vfch, lt) 288 DEF_GVEC_VFC(vfche, le) 289 290 void HELPER(gvec_vfll32)(void *v1, const void *v2, CPUS390XState *env, 291 uint32_t desc) 292 { 293 const bool s = extract32(simd_data(desc), 3, 1); 294 uint8_t vxc, vec_exc = 0; 295 S390Vector tmp = {}; 296 int i; 297 298 for (i = 0; i < 2; i++) { 299 /* load from even element */ 300 const float32 a = s390_vec_read_element32(v2, i * 2); 301 const uint64_t ret = float32_to_float64(a, &env->fpu_status); 302 303 s390_vec_write_element64(&tmp, i, ret); 304 /* indicate the source element */ 305 vxc = check_ieee_exc(env, i * 2, false, &vec_exc); 306 if (s || vxc) { 307 break; 308 } 309 } 310 handle_ieee_exc(env, vxc, vec_exc, GETPC()); 311 *(S390Vector *)v1 = tmp; 312 } 313 314 static void vflr64(S390Vector *v1, const S390Vector *v2, CPUS390XState *env, 315 bool s, bool XxC, uint8_t erm, uintptr_t retaddr) 316 { 317 uint8_t vxc, vec_exc = 0; 318 S390Vector tmp = {}; 319 int i, old_mode; 320 321 old_mode = s390_swap_bfp_rounding_mode(env, erm); 322 for (i = 0; i < 2; i++) { 323 float64 a = s390_vec_read_element64(v2, i); 324 uint32_t ret = float64_to_float32(a, &env->fpu_status); 325 326 /* place at even element */ 327 s390_vec_write_element32(&tmp, i * 2, ret); 328 /* indicate the source element */ 329 vxc = check_ieee_exc(env, i, XxC, &vec_exc); 330 if (s || vxc) { 331 break; 332 } 333 } 334 s390_restore_bfp_rounding_mode(env, old_mode); 335 handle_ieee_exc(env, vxc, vec_exc, retaddr); 336 *v1 = tmp; 337 } 338 339 void HELPER(gvec_vflr64)(void *v1, const void *v2, CPUS390XState *env, 340 uint32_t desc) 341 { 342 const uint8_t erm = extract32(simd_data(desc), 4, 4); 343 const bool XxC = extract32(simd_data(desc), 2, 1); 344 345 vflr64(v1, v2, env, false, XxC, erm, GETPC()); 346 } 347 348 void HELPER(gvec_vflr64s)(void *v1, const void *v2, CPUS390XState *env, 349 uint32_t desc) 350 { 351 const uint8_t erm = extract32(simd_data(desc), 4, 4); 352 const bool XxC = extract32(simd_data(desc), 2, 1); 353 354 vflr64(v1, v2, env, true, XxC, erm, GETPC()); 355 } 356 357 static void vfma64(S390Vector *v1, const S390Vector *v2, const S390Vector *v3, 358 const S390Vector *v4, CPUS390XState *env, bool s, int flags, 359 uintptr_t retaddr) 360 { 361 uint8_t vxc, vec_exc = 0; 362 S390Vector tmp = {}; 363 int i; 364 365 for (i = 0; i < 2; i++) { 366 const float64 a = s390_vec_read_float64(v2, i); 367 const float64 b = s390_vec_read_float64(v3, i); 368 const float64 c = s390_vec_read_float64(v4, i); 369 const float64 ret = float64_muladd(a, b, c, flags, &env->fpu_status); 370 371 s390_vec_write_float64(&tmp, i, ret); 372 vxc = check_ieee_exc(env, i, false, &vec_exc); 373 if (s || vxc) { 374 break; 375 } 376 } 377 handle_ieee_exc(env, vxc, vec_exc, retaddr); 378 *v1 = tmp; 379 } 380 381 #define DEF_GVEC_VFMA_B(NAME, FLAGS, BITS) \ 382 void HELPER(gvec_##NAME##BITS)(void *v1, const void *v2, const void *v3, \ 383 const void *v4, CPUS390XState *env, \ 384 uint32_t desc) \ 385 { \ 386 const bool se = extract32(simd_data(desc), 3, 1); \ 387 \ 388 vfma##BITS(v1, v2, v3, v4, env, se, FLAGS, GETPC()); \ 389 } 390 391 #define DEF_GVEC_VFMA(NAME, FLAGS) \ 392 DEF_GVEC_VFMA_B(NAME, FLAGS, 64) 393 394 DEF_GVEC_VFMA(vfma, 0) 395 DEF_GVEC_VFMA(vfms, float_muladd_negate_c) 396 397 void HELPER(gvec_vftci64)(void *v1, const void *v2, CPUS390XState *env, 398 uint32_t desc) 399 { 400 const uint16_t i3 = extract32(simd_data(desc), 4, 12); 401 const bool s = extract32(simd_data(desc), 3, 1); 402 int i, match = 0; 403 404 for (i = 0; i < 2; i++) { 405 const float64 a = s390_vec_read_float64(v2, i); 406 407 if (float64_dcmask(env, a) & i3) { 408 match++; 409 s390_vec_write_element64(v1, i, -1ull); 410 } else { 411 s390_vec_write_element64(v1, i, 0); 412 } 413 if (s) { 414 break; 415 } 416 } 417 418 if (match == 2 || (s && match)) { 419 env->cc_op = 0; 420 } else if (match) { 421 env->cc_op = 1; 422 } else { 423 env->cc_op = 3; 424 } 425 } 426