src/broadcom/qpu/qpu_instr.c

   1 /*
   2  * Copyright © 2016 Broadcom
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <stdlib.h>
  25 #include "util/macros.h"
  26 #include "broadcom/common/v3d_device_info.h"
  27 #include "qpu_instr.h"
  28
  29 const char *
  30 v3d_qpu_magic_waddr_name(enum v3d_qpu_waddr waddr)
  31 {
  32         static const char *waddr_magic[] = {
  33                 [V3D_QPU_WADDR_R0] = "r0",
  34                 [V3D_QPU_WADDR_R1] = "r1",
  35                 [V3D_QPU_WADDR_R2] = "r2",
  36                 [V3D_QPU_WADDR_R3] = "r3",
  37                 [V3D_QPU_WADDR_R4] = "r4",
  38                 [V3D_QPU_WADDR_R5] = "r5",
  39                 [V3D_QPU_WADDR_NOP] = "-",
  40                 [V3D_QPU_WADDR_TLB] = "tlb",
  41                 [V3D_QPU_WADDR_TLBU] = "tlbu",
  42                 [V3D_QPU_WADDR_TMU] = "tmu",
  43                 [V3D_QPU_WADDR_TMUL] = "tmul",
  44                 [V3D_QPU_WADDR_TMUD] = "tmud",
  45                 [V3D_QPU_WADDR_TMUA] = "tmua",
  46                 [V3D_QPU_WADDR_TMUAU] = "tmuau",
  47                 [V3D_QPU_WADDR_VPM] = "vpm",
  48                 [V3D_QPU_WADDR_VPMU] = "vpmu",
  49                 [V3D_QPU_WADDR_SYNC] = "sync",
  50                 [V3D_QPU_WADDR_SYNCU] = "syncu",
  51                 [V3D_QPU_WADDR_SYNCB] = "syncb",
  52                 [V3D_QPU_WADDR_RECIP] = "recip",
  53                 [V3D_QPU_WADDR_RSQRT] = "rsqrt",
  54                 [V3D_QPU_WADDR_EXP] = "exp",
  55                 [V3D_QPU_WADDR_LOG] = "log",
  56                 [V3D_QPU_WADDR_SIN] = "sin",
  57                 [V3D_QPU_WADDR_RSQRT2] = "rsqrt2",
  58                 [V3D_QPU_WADDR_TMUC] = "tmuc",
  59                 [V3D_QPU_WADDR_TMUS] = "tmus",
  60                 [V3D_QPU_WADDR_TMUT] = "tmut",
  61                 [V3D_QPU_WADDR_TMUR] = "tmur",
  62                 [V3D_QPU_WADDR_TMUI] = "tmui",
  63                 [V3D_QPU_WADDR_TMUB] = "tmub",
  64                 [V3D_QPU_WADDR_TMUDREF] = "tmudref",
  65                 [V3D_QPU_WADDR_TMUOFF] = "tmuoff",
  66                 [V3D_QPU_WADDR_TMUSCM] = "tmuscm",
  67                 [V3D_QPU_WADDR_TMUSF] = "tmusf",
  68                 [V3D_QPU_WADDR_TMUSLOD] = "tmuslod",
  69                 [V3D_QPU_WADDR_TMUHS] = "tmuhs",
  70                 [V3D_QPU_WADDR_TMUHSCM] = "tmuscm",
  71                 [V3D_QPU_WADDR_TMUHSF] = "tmuhsf",
  72                 [V3D_QPU_WADDR_TMUHSLOD] = "tmuhslod",
  73                 [V3D_QPU_WADDR_R5REP] = "r5rep",
  74         };
  75
  76         return waddr_magic[waddr];
  77 }
  78
  79 const char *
  80 v3d_qpu_add_op_name(enum v3d_qpu_add_op op)
  81 {
  82         static const char *op_names[] = {
  83                 [V3D_QPU_A_FADD] = "fadd",
  84                 [V3D_QPU_A_FADDNF] = "faddnf",
  85                 [V3D_QPU_A_VFPACK] = "vfpack",
  86                 [V3D_QPU_A_ADD] = "add",
  87                 [V3D_QPU_A_SUB] = "sub",
  88                 [V3D_QPU_A_FSUB] = "fsub",
  89                 [V3D_QPU_A_MIN] = "min",
  90                 [V3D_QPU_A_MAX] = "max",
  91                 [V3D_QPU_A_UMIN] = "umin",
  92                 [V3D_QPU_A_UMAX] = "umax",
  93                 [V3D_QPU_A_SHL] = "shl",
  94                 [V3D_QPU_A_SHR] = "shr",
  95                 [V3D_QPU_A_ASR] = "asr",
  96                 [V3D_QPU_A_ROR] = "ror",
  97                 [V3D_QPU_A_FMIN] = "fmin",
  98                 [V3D_QPU_A_FMAX] = "fmax",
  99                 [V3D_QPU_A_VFMIN] = "vfmin",
 100                 [V3D_QPU_A_AND] = "and",
 101                 [V3D_QPU_A_OR] = "or",
 102                 [V3D_QPU_A_XOR] = "xor",
 103                 [V3D_QPU_A_VADD] = "vadd",
 104                 [V3D_QPU_A_VSUB] = "vsub",
 105                 [V3D_QPU_A_NOT] = "not",
 106                 [V3D_QPU_A_NEG] = "neg",
 107                 [V3D_QPU_A_FLAPUSH] = "flapush",
 108                 [V3D_QPU_A_FLBPUSH] = "flbpush",
 109                 [V3D_QPU_A_FLPOP] = "flpop",
 110                 [V3D_QPU_A_RECIP] = "recip",
 111                 [V3D_QPU_A_SETMSF] = "setmsf",
 112                 [V3D_QPU_A_SETREVF] = "setrevf",
 113                 [V3D_QPU_A_NOP] = "nop",
 114                 [V3D_QPU_A_TIDX] = "tidx",
 115                 [V3D_QPU_A_EIDX] = "eidx",
 116                 [V3D_QPU_A_LR] = "lr",
 117                 [V3D_QPU_A_VFLA] = "vfla",
 118                 [V3D_QPU_A_VFLNA] = "vflna",
 119                 [V3D_QPU_A_VFLB] = "vflb",
 120                 [V3D_QPU_A_VFLNB] = "vflnb",
 121                 [V3D_QPU_A_FXCD] = "fxcd",
 122                 [V3D_QPU_A_XCD] = "xcd",
 123                 [V3D_QPU_A_FYCD] = "fycd",
 124                 [V3D_QPU_A_YCD] = "ycd",
 125                 [V3D_QPU_A_MSF] = "msf",
 126                 [V3D_QPU_A_REVF] = "revf",
 127                 [V3D_QPU_A_VDWWT] = "vdwwt",
 128                 [V3D_QPU_A_IID] = "iid",
 129                 [V3D_QPU_A_SAMPID] = "sampid",
 130                 [V3D_QPU_A_BARRIERID] = "barrierid",
 131                 [V3D_QPU_A_TMUWT] = "tmuwt",
 132                 [V3D_QPU_A_VPMSETUP] = "vpmsetup",
 133                 [V3D_QPU_A_VPMWT] = "vpmwt",
 134                 [V3D_QPU_A_LDVPMV_IN] = "ldvpmv_in",
 135                 [V3D_QPU_A_LDVPMV_OUT] = "ldvpmv_out",
 136                 [V3D_QPU_A_LDVPMD_IN] = "ldvpmd_in",
 137                 [V3D_QPU_A_LDVPMD_OUT] = "ldvpmd_out",
 138                 [V3D_QPU_A_LDVPMP] = "ldvpmp",
 139                 [V3D_QPU_A_RSQRT] = "rsqrt",
 140                 [V3D_QPU_A_EXP] = "exp",
 141                 [V3D_QPU_A_LOG] = "log",
 142                 [V3D_QPU_A_SIN] = "sin",
 143                 [V3D_QPU_A_RSQRT2] = "rsqrt2",
 144                 [V3D_QPU_A_LDVPMG_IN] = "ldvpmg_in",
 145                 [V3D_QPU_A_LDVPMG_OUT] = "ldvpmg_out",
 146                 [V3D_QPU_A_FCMP] = "fcmp",
 147                 [V3D_QPU_A_VFMAX] = "vfmax",
 148                 [V3D_QPU_A_FROUND] = "fround",
 149                 [V3D_QPU_A_FTOIN] = "ftoin",
 150                 [V3D_QPU_A_FTRUNC] = "ftrunc",
 151                 [V3D_QPU_A_FTOIZ] = "ftoiz",
 152                 [V3D_QPU_A_FFLOOR] = "ffloor",
 153                 [V3D_QPU_A_FTOUZ] = "ftouz",
 154                 [V3D_QPU_A_FCEIL] = "fceil",
 155                 [V3D_QPU_A_FTOC] = "ftoc",
 156                 [V3D_QPU_A_FDX] = "fdx",
 157                 [V3D_QPU_A_FDY] = "fdy",
 158                 [V3D_QPU_A_STVPMV] = "stvpmv",
 159                 [V3D_QPU_A_STVPMD] = "stvpmd",
 160                 [V3D_QPU_A_STVPMP] = "stvpmp",
 161                 [V3D_QPU_A_ITOF] = "itof",
 162                 [V3D_QPU_A_CLZ] = "clz",
 163                 [V3D_QPU_A_UTOF] = "utof",
 164         };
 165
 166         if (op >= ARRAY_SIZE(op_names))
 167                 return NULL;
 168
 169         return op_names[op];
 170 }
 171
 172 const char *
 173 v3d_qpu_mul_op_name(enum v3d_qpu_mul_op op)
 174 {
 175         static const char *op_names[] = {
 176                 [V3D_QPU_M_ADD] = "add",
 177                 [V3D_QPU_M_SUB] = "sub",
 178                 [V3D_QPU_M_UMUL24] = "umul24",
 179                 [V3D_QPU_M_VFMUL] = "vfmul",
 180                 [V3D_QPU_M_SMUL24] = "smul24",
 181                 [V3D_QPU_M_MULTOP] = "multop",
 182                 [V3D_QPU_M_FMOV] = "fmov",
 183                 [V3D_QPU_M_MOV] = "mov",
 184                 [V3D_QPU_M_NOP] = "nop",
 185                 [V3D_QPU_M_FMUL] = "fmul",
 186         };
 187
 188         if (op >= ARRAY_SIZE(op_names))
 189                 return NULL;
 190
 191         return op_names[op];
 192 }
 193
 194 const char *
 195 v3d_qpu_cond_name(enum v3d_qpu_cond cond)
 196 {
 197         switch (cond) {
 198         case V3D_QPU_COND_NONE:
 199                 return "";
 200         case V3D_QPU_COND_IFA:
 201                 return ".ifa";
 202         case V3D_QPU_COND_IFB:
 203                 return ".ifb";
 204         case V3D_QPU_COND_IFNA:
 205                 return ".ifna";
 206         case V3D_QPU_COND_IFNB:
 207                 return ".ifnb";
 208         default:
 209                 unreachable("bad cond value");
 210         }
 211 }
 212
 213 const char *
 214 v3d_qpu_branch_cond_name(enum v3d_qpu_branch_cond cond)
 215 {
 216         switch (cond) {
 217         case V3D_QPU_BRANCH_COND_ALWAYS:
 218                 return "";
 219         case V3D_QPU_BRANCH_COND_A0:
 220                 return ".a0";
 221         case V3D_QPU_BRANCH_COND_NA0:
 222                 return ".na0";
 223         case V3D_QPU_BRANCH_COND_ALLA:
 224                 return ".alla";
 225         case V3D_QPU_BRANCH_COND_ANYNA:
 226                 return ".anyna";
 227         case V3D_QPU_BRANCH_COND_ANYA:
 228                 return ".anya";
 229         case V3D_QPU_BRANCH_COND_ALLNA:
 230                 return ".allna";
 231         default:
 232                 unreachable("bad branch cond value");
 233         }
 234 }
 235
 236 const char *
 237 v3d_qpu_msfign_name(enum v3d_qpu_msfign msfign)
 238 {
 239         switch (msfign) {
 240         case V3D_QPU_MSFIGN_NONE:
 241                 return "";
 242         case V3D_QPU_MSFIGN_P:
 243                 return "p";
 244         case V3D_QPU_MSFIGN_Q:
 245                 return "q";
 246         default:
 247                 unreachable("bad branch cond value");
 248         }
 249 }
 250
 251 const char *
 252 v3d_qpu_pf_name(enum v3d_qpu_pf pf)
 253 {
 254         switch (pf) {
 255         case V3D_QPU_PF_NONE:
 256                 return "";
 257         case V3D_QPU_PF_PUSHZ:
 258                 return ".pushz";
 259         case V3D_QPU_PF_PUSHN:
 260                 return ".pushn";
 261         case V3D_QPU_PF_PUSHC:
 262                 return ".pushc";
 263         default:
 264                 unreachable("bad pf value");
 265         }
 266 }
 267
 268 const char *
 269 v3d_qpu_uf_name(enum v3d_qpu_uf uf)
 270 {
 271         switch (uf) {
 272         case V3D_QPU_UF_NONE:
 273                 return "";
 274         case V3D_QPU_UF_ANDZ:
 275                 return ".andz";
 276         case V3D_QPU_UF_ANDNZ:
 277                 return ".andnz";
 278         case V3D_QPU_UF_NORZ:
 279                 return ".norz";
 280         case V3D_QPU_UF_NORNZ:
 281                 return ".nornz";
 282         case V3D_QPU_UF_ANDN:
 283                 return ".andn";
 284         case V3D_QPU_UF_ANDNN:
 285                 return ".andnn";
 286         case V3D_QPU_UF_NORN:
 287                 return ".norn";
 288         case V3D_QPU_UF_NORNN:
 289                 return ".nornn";
 290         case V3D_QPU_UF_ANDC:
 291                 return ".andc";
 292         case V3D_QPU_UF_ANDNC:
 293                 return ".andnc";
 294         case V3D_QPU_UF_NORC:
 295                 return ".norc";
 296         case V3D_QPU_UF_NORNC:
 297                 return ".nornc";
 298         default:
 299                 unreachable("bad pf value");
 300         }
 301 }
 302
 303 const char *
 304 v3d_qpu_pack_name(enum v3d_qpu_output_pack pack)
 305 {
 306         switch (pack) {
 307         case V3D_QPU_PACK_NONE:
 308                 return "";
 309         case V3D_QPU_PACK_L:
 310                 return ".l";
 311         case V3D_QPU_PACK_H:
 312                 return ".h";
 313         default:
 314                 unreachable("bad pack value");
 315         }
 316 }
 317
 318 const char *
 319 v3d_qpu_unpack_name(enum v3d_qpu_input_unpack unpack)
 320 {
 321         switch (unpack) {
 322         case V3D_QPU_UNPACK_NONE:
 323                 return "";
 324         case V3D_QPU_UNPACK_L:
 325                 return ".l";
 326         case V3D_QPU_UNPACK_H:
 327                 return ".h";
 328         case V3D_QPU_UNPACK_ABS:
 329                 return ".abs";
 330         case V3D_QPU_UNPACK_REPLICATE_32F_16:
 331                 return ".ff";
 332         case V3D_QPU_UNPACK_REPLICATE_L_16:
 333                 return ".ll";
 334         case V3D_QPU_UNPACK_REPLICATE_H_16:
 335                 return ".hh";
 336         case V3D_QPU_UNPACK_SWAP_16:
 337                 return ".swp";
 338         default:
 339                 unreachable("bad unpack value");
 340         }
 341 }
 342
 343 #define D       1
 344 #define A       2
 345 #define B       4
 346 static const uint8_t add_op_args[] = {
 347         [V3D_QPU_A_FADD] = D | A | B,
 348         [V3D_QPU_A_FADDNF] = D | A | B,
 349         [V3D_QPU_A_VFPACK] = D | A | B,
 350         [V3D_QPU_A_ADD] = D | A | B,
 351         [V3D_QPU_A_VFPACK] = D | A | B,
 352         [V3D_QPU_A_SUB] = D | A | B,
 353         [V3D_QPU_A_VFPACK] = D | A | B,
 354         [V3D_QPU_A_FSUB] = D | A | B,
 355         [V3D_QPU_A_MIN] = D | A | B,
 356         [V3D_QPU_A_MAX] = D | A | B,
 357         [V3D_QPU_A_UMIN] = D | A | B,
 358         [V3D_QPU_A_UMAX] = D | A | B,
 359         [V3D_QPU_A_SHL] = D | A | B,
 360         [V3D_QPU_A_SHR] = D | A | B,
 361         [V3D_QPU_A_ASR] = D | A | B,
 362         [V3D_QPU_A_ROR] = D | A | B,
 363         [V3D_QPU_A_FMIN] = D | A | B,
 364         [V3D_QPU_A_FMAX] = D | A | B,
 365         [V3D_QPU_A_VFMIN] = D | A | B,
 366
 367         [V3D_QPU_A_AND] = D | A | B,
 368         [V3D_QPU_A_OR] = D | A | B,
 369         [V3D_QPU_A_XOR] = D | A | B,
 370
 371         [V3D_QPU_A_VADD] = D | A | B,
 372         [V3D_QPU_A_VSUB] = D | A | B,
 373         [V3D_QPU_A_NOT] = D | A,
 374         [V3D_QPU_A_NEG] = D | A,
 375         [V3D_QPU_A_FLAPUSH] = D | A,
 376         [V3D_QPU_A_FLBPUSH] = D | A,
 377         [V3D_QPU_A_FLPOP] = D | A,
 378         [V3D_QPU_A_RECIP] = D | A,
 379         [V3D_QPU_A_SETMSF] = D | A,
 380         [V3D_QPU_A_SETREVF] = D | A,
 381         [V3D_QPU_A_NOP] = 0,
 382         [V3D_QPU_A_TIDX] = D,
 383         [V3D_QPU_A_EIDX] = D,
 384         [V3D_QPU_A_LR] = D,
 385         [V3D_QPU_A_VFLA] = D,
 386         [V3D_QPU_A_VFLNA] = D,
 387         [V3D_QPU_A_VFLB] = D,
 388         [V3D_QPU_A_VFLNB] = D,
 389
 390         [V3D_QPU_A_FXCD] = D,
 391         [V3D_QPU_A_XCD] = D,
 392         [V3D_QPU_A_FYCD] = D,
 393         [V3D_QPU_A_YCD] = D,
 394
 395         [V3D_QPU_A_MSF] = D,
 396         [V3D_QPU_A_REVF] = D,
 397         [V3D_QPU_A_VDWWT] = D,
 398         [V3D_QPU_A_IID] = D,
 399         [V3D_QPU_A_SAMPID] = D,
 400         [V3D_QPU_A_BARRIERID] = D,
 401         [V3D_QPU_A_TMUWT] = D,
 402         [V3D_QPU_A_VPMWT] = D,
 403
 404         [V3D_QPU_A_VPMSETUP] = D | A,
 405
 406         [V3D_QPU_A_LDVPMV_IN] = D | A,
 407         [V3D_QPU_A_LDVPMV_OUT] = D | A,
 408         [V3D_QPU_A_LDVPMD_IN] = D | A,
 409         [V3D_QPU_A_LDVPMD_OUT] = D | A,
 410         [V3D_QPU_A_LDVPMP] = D | A,
 411         [V3D_QPU_A_RSQRT] = D | A,
 412         [V3D_QPU_A_EXP] = D | A,
 413         [V3D_QPU_A_LOG] = D | A,
 414         [V3D_QPU_A_SIN] = D | A,
 415         [V3D_QPU_A_RSQRT2] = D | A,
 416         [V3D_QPU_A_LDVPMG_IN] = D | A | B,
 417         [V3D_QPU_A_LDVPMG_OUT] = D | A | B,
 418
 419         /* FIXME: MOVABSNEG */
 420
 421         [V3D_QPU_A_FCMP] = D | A | B,
 422         [V3D_QPU_A_VFMAX] = D | A | B,
 423
 424         [V3D_QPU_A_FROUND] = D | A,
 425         [V3D_QPU_A_FTOIN] = D | A,
 426         [V3D_QPU_A_FTRUNC] = D | A,
 427         [V3D_QPU_A_FTOIZ] = D | A,
 428         [V3D_QPU_A_FFLOOR] = D | A,
 429         [V3D_QPU_A_FTOUZ] = D | A,
 430         [V3D_QPU_A_FCEIL] = D | A,
 431         [V3D_QPU_A_FTOC] = D | A,
 432
 433         [V3D_QPU_A_FDX] = D | A,
 434         [V3D_QPU_A_FDY] = D | A,
 435
 436         [V3D_QPU_A_STVPMV] = A | B,
 437         [V3D_QPU_A_STVPMD] = A | B,
 438         [V3D_QPU_A_STVPMP] = A | B,
 439
 440         [V3D_QPU_A_ITOF] = D | A,
 441         [V3D_QPU_A_CLZ] = D | A,
 442         [V3D_QPU_A_UTOF] = D | A,
 443 };
 444
 445 static const uint8_t mul_op_args[] = {
 446         [V3D_QPU_M_ADD] = D | A | B,
 447         [V3D_QPU_M_SUB] = D | A | B,
 448         [V3D_QPU_M_UMUL24] = D | A | B,
 449         [V3D_QPU_M_VFMUL] = D | A | B,
 450         [V3D_QPU_M_SMUL24] = D | A | B,
 451         [V3D_QPU_M_MULTOP] = D | A | B,
 452         [V3D_QPU_M_FMOV] = D | A,
 453         [V3D_QPU_M_NOP] = 0,
 454         [V3D_QPU_M_MOV] = D | A,
 455         [V3D_QPU_M_FMUL] = D | A | B,
 456 };
 457
 458 bool
 459 v3d_qpu_add_op_has_dst(enum v3d_qpu_add_op op)
 460 {
 461         assert(op < ARRAY_SIZE(add_op_args));
 462
 463         return add_op_args[op] & D;
 464 }
 465
 466 bool
 467 v3d_qpu_mul_op_has_dst(enum v3d_qpu_mul_op op)
 468 {
 469         assert(op < ARRAY_SIZE(mul_op_args));
 470
 471         return mul_op_args[op] & D;
 472 }
 473
 474 int
 475 v3d_qpu_add_op_num_src(enum v3d_qpu_add_op op)
 476 {
 477         assert(op < ARRAY_SIZE(add_op_args));
 478
 479         uint8_t args = add_op_args[op];
 480         if (args & B)
 481                 return 2;
 482         else if (args & A)
 483                 return 1;
 484         else
 485                 return 0;
 486 }
 487
 488 int
 489 v3d_qpu_mul_op_num_src(enum v3d_qpu_mul_op op)
 490 {
 491         assert(op < ARRAY_SIZE(mul_op_args));
 492
 493         uint8_t args = mul_op_args[op];
 494         if (args & B)
 495                 return 2;
 496         else if (args & A)
 497                 return 1;
 498         else
 499                 return 0;
 500 }
 501
 502 bool
 503 v3d_qpu_magic_waddr_is_sfu(enum v3d_qpu_waddr waddr)
 504 {
 505         switch (waddr) {
 506         case V3D_QPU_WADDR_RECIP:
 507         case V3D_QPU_WADDR_RSQRT:
 508         case V3D_QPU_WADDR_EXP:
 509         case V3D_QPU_WADDR_LOG:
 510         case V3D_QPU_WADDR_SIN:
 511         case V3D_QPU_WADDR_RSQRT2:
 512                 return true;
 513         default:
 514                 return false;
 515         }
 516 }
 517
 518 bool
 519 v3d_qpu_magic_waddr_is_tmu(enum v3d_qpu_waddr waddr)
 520 {
 521         /* XXX: WADDR_TMU changed to UNIFA on 4.x */
 522         return ((waddr >= V3D_QPU_WADDR_TMU &&
 523                  waddr <= V3D_QPU_WADDR_TMUAU) ||
 524                 (waddr >= V3D_QPU_WADDR_TMUC &&
 525                  waddr <= V3D_QPU_WADDR_TMUHSLOD));
 526 }
 527
 528 bool
 529 v3d_qpu_waits_on_tmu(const struct v3d_qpu_instr *inst)
 530 {
 531         return (inst->sig.ldtmu ||
 532                 (inst->type == V3D_QPU_INSTR_TYPE_ALU &&
 533                  inst->alu.add.op == V3D_QPU_A_TMUWT));
 534 }
 535
 536 bool
 537 v3d_qpu_magic_waddr_is_tlb(enum v3d_qpu_waddr waddr)
 538 {
 539         return (waddr == V3D_QPU_WADDR_TLB ||
 540                 waddr == V3D_QPU_WADDR_TLBU);
 541 }
 542
 543 bool
 544 v3d_qpu_magic_waddr_is_vpm(enum v3d_qpu_waddr waddr)
 545 {
 546         return (waddr == V3D_QPU_WADDR_VPM ||
 547                 waddr == V3D_QPU_WADDR_VPMU);
 548 }
 549
 550 bool
 551 v3d_qpu_magic_waddr_is_tsy(enum v3d_qpu_waddr waddr)
 552 {
 553         return (waddr == V3D_QPU_WADDR_SYNC ||
 554                 waddr == V3D_QPU_WADDR_SYNCB ||
 555                 waddr == V3D_QPU_WADDR_SYNCU);
 556 }
 557
 558 bool
 559 v3d_qpu_magic_waddr_loads_unif(enum v3d_qpu_waddr waddr)
 560 {
 561         switch (waddr) {
 562         case V3D_QPU_WADDR_VPMU:
 563         case V3D_QPU_WADDR_TLBU:
 564         case V3D_QPU_WADDR_TMUAU:
 565         case V3D_QPU_WADDR_SYNCU:
 566                 return true;
 567         default:
 568                 return false;
 569         }
 570 }
 571
 572 static bool
 573 v3d_qpu_add_op_reads_vpm(enum  v3d_qpu_add_op op)
 574 {
 575         switch (op) {
 576         case V3D_QPU_A_VPMSETUP:
 577         case V3D_QPU_A_VPMWT:
 578         case V3D_QPU_A_LDVPMV_IN:
 579         case V3D_QPU_A_LDVPMV_OUT:
 580         case V3D_QPU_A_LDVPMD_IN:
 581         case V3D_QPU_A_LDVPMD_OUT:
 582         case V3D_QPU_A_LDVPMP:
 583         case V3D_QPU_A_LDVPMG_IN:
 584         case V3D_QPU_A_LDVPMG_OUT:
 585                 return true;
 586         default:
 587                 return false;
 588         }
 589 }
 590
 591 static bool
 592 v3d_qpu_add_op_writes_vpm(enum  v3d_qpu_add_op op)
 593 {
 594         switch (op) {
 595         case V3D_QPU_A_VPMSETUP:
 596         case V3D_QPU_A_VPMWT:
 597         case V3D_QPU_A_STVPMV:
 598         case V3D_QPU_A_STVPMD:
 599         case V3D_QPU_A_STVPMP:
 600                 return true;
 601         default:
 602                 return false;
 603         }
 604 }
 605
 606 bool
 607 v3d_qpu_uses_tlb(const struct v3d_qpu_instr *inst)
 608 {
 609         if (inst->sig.ldtlb ||
 610             inst->sig.ldtlbu)
 611                 return true;
 612
 613         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 614                 if (inst->alu.add.magic_write &&
 615                     v3d_qpu_magic_waddr_is_tlb(inst->alu.add.waddr)) {
 616                         return true;
 617                 }
 618
 619                 if (inst->alu.mul.magic_write &&
 620                     v3d_qpu_magic_waddr_is_tlb(inst->alu.mul.waddr)) {
 621                         return true;
 622                 }
 623         }
 624
 625         return false;
 626 }
 627
 628 bool
 629 v3d_qpu_uses_sfu(const struct v3d_qpu_instr *inst)
 630 {
 631         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 632                 switch (inst->alu.add.op) {
 633                 case V3D_QPU_A_RECIP:
 634                 case V3D_QPU_A_RSQRT:
 635                 case V3D_QPU_A_EXP:
 636                 case V3D_QPU_A_LOG:
 637                 case V3D_QPU_A_SIN:
 638                 case V3D_QPU_A_RSQRT2:
 639                         return true;
 640                 default:
 641                         break;
 642                 }
 643
 644                 if (inst->alu.add.magic_write &&
 645                     v3d_qpu_magic_waddr_is_sfu(inst->alu.add.waddr)) {
 646                         return true;
 647                 }
 648
 649                 if (inst->alu.mul.magic_write &&
 650                     v3d_qpu_magic_waddr_is_sfu(inst->alu.mul.waddr)) {
 651                         return true;
 652                 }
 653         }
 654
 655         return false;
 656 }
 657
 658 bool
 659 v3d_qpu_writes_tmu(const struct v3d_qpu_instr *inst)
 660 {
 661         return (inst->type == V3D_QPU_INSTR_TYPE_ALU &&
 662                 ((inst->alu.add.magic_write &&
 663                   v3d_qpu_magic_waddr_is_tmu(inst->alu.add.waddr)) ||
 664                  (inst->alu.mul.magic_write &&
 665                   v3d_qpu_magic_waddr_is_tmu(inst->alu.mul.waddr))));
 666 }
 667
 668 bool
 669 v3d_qpu_reads_vpm(const struct v3d_qpu_instr *inst)
 670 {
 671         if (inst->sig.ldvpm)
 672                 return true;
 673
 674         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 675                 if (v3d_qpu_add_op_reads_vpm(inst->alu.add.op))
 676                         return true;
 677         }
 678
 679         return false;
 680 }
 681
 682 bool
 683 v3d_qpu_writes_vpm(const struct v3d_qpu_instr *inst)
 684 {
 685         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 686                 if (v3d_qpu_add_op_writes_vpm(inst->alu.add.op))
 687                         return true;
 688
 689                 if (inst->alu.add.magic_write &&
 690                     v3d_qpu_magic_waddr_is_vpm(inst->alu.add.waddr)) {
 691                         return true;
 692                 }
 693
 694                 if (inst->alu.mul.magic_write &&
 695                     v3d_qpu_magic_waddr_is_vpm(inst->alu.mul.waddr)) {
 696                         return true;
 697                 }
 698         }
 699
 700         return false;
 701 }
 702
 703 bool
 704 v3d_qpu_uses_vpm(const struct v3d_qpu_instr *inst)
 705 {
 706         return v3d_qpu_reads_vpm(inst) || v3d_qpu_writes_vpm(inst);
 707 }
 708
 709 bool
 710 v3d_qpu_writes_r3(const struct v3d_device_info *devinfo,
 711                   const struct v3d_qpu_instr *inst)
 712 {
 713         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 714                 if (inst->alu.add.magic_write &&
 715                     inst->alu.add.waddr == V3D_QPU_WADDR_R3) {
 716                         return true;
 717                 }
 718
 719                 if (inst->alu.mul.magic_write &&
 720                     inst->alu.mul.waddr == V3D_QPU_WADDR_R3) {
 721                         return true;
 722                 }
 723         }
 724
 725         if (v3d_qpu_sig_writes_address(devinfo, &inst->sig) &&
 726             inst->sig_magic && inst->sig_addr == V3D_QPU_WADDR_R3) {
 727                 return true;
 728         }
 729
 730         return inst->sig.ldvary || inst->sig.ldvpm;
 731 }
 732
 733 bool
 734 v3d_qpu_writes_r4(const struct v3d_device_info *devinfo,
 735                   const struct v3d_qpu_instr *inst)
 736 {
 737         if (inst->sig.ldtmu)
 738                 return true;
 739
 740         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 741                 if (inst->alu.add.magic_write &&
 742                     (inst->alu.add.waddr == V3D_QPU_WADDR_R4 ||
 743                      v3d_qpu_magic_waddr_is_sfu(inst->alu.add.waddr))) {
 744                         return true;
 745                 }
 746
 747                 if (inst->alu.mul.magic_write &&
 748                     (inst->alu.mul.waddr == V3D_QPU_WADDR_R4 ||
 749                      v3d_qpu_magic_waddr_is_sfu(inst->alu.mul.waddr))) {
 750                         return true;
 751                 }
 752         }
 753
 754         if (v3d_qpu_sig_writes_address(devinfo, &inst->sig) &&
 755             inst->sig_magic && inst->sig_addr == V3D_QPU_WADDR_R4) {
 756                 return true;
 757         }
 758
 759         return false;
 760 }
 761
 762 bool
 763 v3d_qpu_writes_r5(const struct v3d_device_info *devinfo,
 764                   const struct v3d_qpu_instr *inst)
 765 {
 766         if (inst->type == V3D_QPU_INSTR_TYPE_ALU) {
 767                 if (inst->alu.add.magic_write &&
 768                     inst->alu.add.waddr == V3D_QPU_WADDR_R5) {
 769                         return true;
 770                 }
 771
 772                 if (inst->alu.mul.magic_write &&
 773                     inst->alu.mul.waddr == V3D_QPU_WADDR_R5) {
 774                         return true;
 775                 }
 776         }
 777
 778         if (v3d_qpu_sig_writes_address(devinfo, &inst->sig) &&
 779             inst->sig_magic && inst->sig_addr == V3D_QPU_WADDR_R5) {
 780                 return true;
 781         }
 782
 783         return inst->sig.ldvary || inst->sig.ldunif || inst->sig.ldunifa;
 784 }
 785
 786 bool
 787 v3d_qpu_uses_mux(const struct v3d_qpu_instr *inst, enum v3d_qpu_mux mux)
 788 {
 789         int add_nsrc = v3d_qpu_add_op_num_src(inst->alu.add.op);
 790         int mul_nsrc = v3d_qpu_mul_op_num_src(inst->alu.mul.op);
 791
 792         return ((add_nsrc > 0 && inst->alu.add.a == mux) ||
 793                 (add_nsrc > 1 && inst->alu.add.b == mux) ||
 794                 (mul_nsrc > 0 && inst->alu.mul.a == mux) ||
 795                 (mul_nsrc > 1 && inst->alu.mul.b == mux));
 796 }
 797
 798 bool
 799 v3d_qpu_sig_writes_address(const struct v3d_device_info *devinfo,
 800                            const struct v3d_qpu_sig *sig)
 801 {
 802         if (devinfo->ver < 41)
 803                 return false;
 804
 805         return (sig->ldunifrf ||
 806                 sig->ldunifarf ||
 807                 sig->ldvary ||
 808                 sig->ldtmu ||
 809                 sig->ldtlb ||
 810                 sig->ldtlbu);
 811 }