src/mesa/state_tracker/st_shader_cache.c

   1 /*
   2  * Copyright © 2017 Timothy Arceri
   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
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include <stdio.h>
  25 #include "st_debug.h"
  26 #include "st_program.h"
  27 #include "st_shader_cache.h"
  28 #include "st_util.h"
  29 #include "compiler/glsl/program.h"
  30 #include "compiler/nir/nir.h"
  31 #include "compiler/nir/nir_serialize.h"
  32 #include "pipe/p_shader_tokens.h"
  33 #include "program/ir_to_mesa.h"
  34 #include "util/u_memory.h"
  35
  36 void
  37 st_get_program_binary_driver_sha1(struct gl_context *ctx, uint8_t *sha1)
  38 {
  39    disk_cache_compute_key(ctx->Cache, NULL, 0, sha1);
  40 }
  41
  42 static void
  43 write_stream_out_to_cache(struct blob *blob,
  44                           struct pipe_shader_state *tgsi)
  45 {
  46    blob_write_bytes(blob, &tgsi->stream_output,
  47                     sizeof(tgsi->stream_output));
  48 }
  49
  50 static void
  51 copy_blob_to_driver_cache_blob(struct blob *blob, struct gl_program *prog)
  52 {
  53    prog->driver_cache_blob = ralloc_size(NULL, blob->size);
  54    memcpy(prog->driver_cache_blob, blob->data, blob->size);
  55    prog->driver_cache_blob_size = blob->size;
  56 }
  57
  58 static void
  59 write_tgsi_to_cache(struct blob *blob, const struct tgsi_token *tokens,
  60                     struct gl_program *prog, unsigned num_tokens)
  61 {
  62    blob_write_uint32(blob, num_tokens);
  63    blob_write_bytes(blob, tokens, num_tokens * sizeof(struct tgsi_token));
  64    copy_blob_to_driver_cache_blob(blob, prog);
  65 }
  66
  67 static void
  68 write_nir_to_cache(struct blob *blob, struct gl_program *prog)
  69 {
  70    nir_serialize(blob, prog->nir, false);
  71    copy_blob_to_driver_cache_blob(blob, prog);
  72 }
  73
  74 static void
  75 st_serialise_ir_program(struct gl_context *ctx, struct gl_program *prog,
  76                         bool nir)
  77 {
  78    if (prog->driver_cache_blob)
  79       return;
  80
  81    struct blob blob;
  82    blob_init(&blob);
  83
  84    switch (prog->info.stage) {
  85    case MESA_SHADER_VERTEX: {
  86       struct st_vertex_program *stvp = (struct st_vertex_program *) prog;
  87
  88       blob_write_uint32(&blob, stvp->num_inputs);
  89       blob_write_bytes(&blob, stvp->index_to_input,
  90                        sizeof(stvp->index_to_input));
  91       blob_write_bytes(&blob, stvp->input_to_index,
  92                        sizeof(stvp->input_to_index));
  93       blob_write_bytes(&blob, stvp->result_to_output,
  94                        sizeof(stvp->result_to_output));
  95
  96       write_stream_out_to_cache(&blob, &stvp->tgsi);
  97
  98       if (nir)
  99          write_nir_to_cache(&blob, prog);
 100       else
 101          write_tgsi_to_cache(&blob, stvp->tgsi.tokens, prog,
 102                              stvp->num_tgsi_tokens);
 103       break;
 104    }
 105    case MESA_SHADER_TESS_CTRL:
 106    case MESA_SHADER_TESS_EVAL:
 107    case MESA_SHADER_GEOMETRY: {
 108       struct st_common_program *stcp = (struct st_common_program *) prog;
 109
 110       write_stream_out_to_cache(&blob, &stcp->tgsi);
 111
 112       if (nir)
 113          write_nir_to_cache(&blob, prog);
 114       else
 115          write_tgsi_to_cache(&blob, stcp->tgsi.tokens, prog,
 116                              stcp->num_tgsi_tokens);
 117       break;
 118    }
 119    case MESA_SHADER_FRAGMENT: {
 120       struct st_fragment_program *stfp = (struct st_fragment_program *) prog;
 121
 122       if (nir)
 123          write_nir_to_cache(&blob, prog);
 124       else
 125          write_tgsi_to_cache(&blob, stfp->tgsi.tokens, prog,
 126                              stfp->num_tgsi_tokens);
 127       break;
 128    }
 129    case MESA_SHADER_COMPUTE: {
 130       struct st_compute_program *stcp = (struct st_compute_program *) prog;
 131
 132       if (nir)
 133          write_nir_to_cache(&blob, prog);
 134       else
 135          write_tgsi_to_cache(&blob, stcp->tgsi.prog, prog,
 136                              stcp->num_tgsi_tokens);
 137       break;
 138    }
 139    default:
 140       unreachable("Unsupported stage");
 141    }
 142
 143    blob_finish(&blob);
 144 }
 145
 146 /**
 147  * Store tgsi and any other required state in on-disk shader cache.
 148  */
 149 void
 150 st_store_ir_in_disk_cache(struct st_context *st, struct gl_program *prog,
 151                           bool nir)
 152 {
 153    if (!st->ctx->Cache)
 154       return;
 155
 156    /* Exit early when we are dealing with a ff shader with no source file to
 157     * generate a source from.
 158     */
 159    static const char zero[sizeof(prog->sh.data->sha1)] = {0};
 160    if (memcmp(prog->sh.data->sha1, zero, sizeof(prog->sh.data->sha1)) == 0)
 161       return;
 162
 163    st_serialise_ir_program(st->ctx, prog, nir);
 164
 165    if (st->ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 166       fprintf(stderr, "putting %s state tracker IR in cache\n",
 167               _mesa_shader_stage_to_string(prog->info.stage));
 168    }
 169 }
 170
 171 static void
 172 read_stream_out_from_cache(struct blob_reader *blob_reader,
 173                            struct pipe_shader_state *tgsi)
 174 {
 175    blob_copy_bytes(blob_reader, (uint8_t *) &tgsi->stream_output,
 176                     sizeof(tgsi->stream_output));
 177 }
 178
 179 static void
 180 read_tgsi_from_cache(struct blob_reader *blob_reader,
 181                      const struct tgsi_token **tokens,
 182                      unsigned *num_tokens)
 183 {
 184    *num_tokens  = blob_read_uint32(blob_reader);
 185    unsigned tokens_size = *num_tokens * sizeof(struct tgsi_token);
 186    *tokens = (const struct tgsi_token*) MALLOC(tokens_size);
 187    blob_copy_bytes(blob_reader, (uint8_t *) *tokens, tokens_size);
 188 }
 189
 190 static void
 191 st_deserialise_ir_program(struct gl_context *ctx,
 192                           struct gl_shader_program *shProg,
 193                           struct gl_program *prog, bool nir)
 194 {
 195    struct st_context *st = st_context(ctx);
 196    size_t size = prog->driver_cache_blob_size;
 197    uint8_t *buffer = (uint8_t *) prog->driver_cache_blob;
 198    const struct nir_shader_compiler_options *options =
 199       ctx->Const.ShaderCompilerOptions[prog->info.stage].NirOptions;
 200
 201    assert(prog->driver_cache_blob && prog->driver_cache_blob_size > 0);
 202
 203    struct blob_reader blob_reader;
 204    blob_reader_init(&blob_reader, buffer, size);
 205
 206    switch (prog->info.stage) {
 207    case MESA_SHADER_VERTEX: {
 208       struct st_vertex_program *stvp = (struct st_vertex_program *) prog;
 209
 210       st_release_vp_variants(st, stvp);
 211
 212       stvp->num_inputs = blob_read_uint32(&blob_reader);
 213       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->index_to_input,
 214                       sizeof(stvp->index_to_input));
 215       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->input_to_index,
 216                       sizeof(stvp->input_to_index));
 217       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->result_to_output,
 218                       sizeof(stvp->result_to_output));
 219
 220       read_stream_out_from_cache(&blob_reader, &stvp->tgsi);
 221
 222       if (nir) {
 223          stvp->tgsi.type = PIPE_SHADER_IR_NIR;
 224          stvp->shader_program = shProg;
 225          stvp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
 226          prog->nir = stvp->tgsi.ir.nir;
 227       } else {
 228          read_tgsi_from_cache(&blob_reader, &stvp->tgsi.tokens,
 229                               &stvp->num_tgsi_tokens);
 230       }
 231
 232       if (st->vp == stvp)
 233          st->dirty |= ST_NEW_VERTEX_PROGRAM(st, stvp);
 234
 235       break;
 236    }
 237    case MESA_SHADER_TESS_CTRL: {
 238       struct st_common_program *sttcp = st_common_program(prog);
 239
 240       st_release_basic_variants(st, sttcp);
 241       read_stream_out_from_cache(&blob_reader, &sttcp->tgsi);
 242
 243       if (nir) {
 244          sttcp->tgsi.type = PIPE_SHADER_IR_NIR;
 245          sttcp->shader_program = shProg;
 246          sttcp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
 247          prog->nir = sttcp->tgsi.ir.nir;
 248       } else {
 249          read_tgsi_from_cache(&blob_reader, &sttcp->tgsi.tokens,
 250                               &sttcp->num_tgsi_tokens);
 251       }
 252
 253       if (st->tcp == sttcp)
 254          st->dirty |= sttcp->affected_states;
 255
 256       break;
 257    }
 258    case MESA_SHADER_TESS_EVAL: {
 259       struct st_common_program *sttep = st_common_program(prog);
 260
 261       st_release_basic_variants(st, sttep);
 262       read_stream_out_from_cache(&blob_reader, &sttep->tgsi);
 263
 264       if (nir) {
 265          sttep->tgsi.type = PIPE_SHADER_IR_NIR;
 266          sttep->shader_program = shProg;
 267          sttep->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
 268          prog->nir = sttep->tgsi.ir.nir;
 269       } else {
 270          read_tgsi_from_cache(&blob_reader, &sttep->tgsi.tokens,
 271                               &sttep->num_tgsi_tokens);
 272       }
 273
 274       if (st->tep == sttep)
 275          st->dirty |= sttep->affected_states;
 276
 277       break;
 278    }
 279    case MESA_SHADER_GEOMETRY: {
 280       struct st_common_program *stgp = st_common_program(prog);
 281
 282       st_release_basic_variants(st, stgp);
 283       read_stream_out_from_cache(&blob_reader, &stgp->tgsi);
 284
 285       if (nir) {
 286          stgp->tgsi.type = PIPE_SHADER_IR_NIR;
 287          stgp->shader_program = shProg;
 288          stgp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
 289          prog->nir = stgp->tgsi.ir.nir;
 290       } else {
 291          read_tgsi_from_cache(&blob_reader, &stgp->tgsi.tokens,
 292                               &stgp->num_tgsi_tokens);
 293       }
 294
 295       if (st->gp == stgp)
 296          st->dirty |= stgp->affected_states;
 297
 298       break;
 299    }
 300    case MESA_SHADER_FRAGMENT: {
 301       struct st_fragment_program *stfp = (struct st_fragment_program *) prog;
 302
 303       st_release_fp_variants(st, stfp);
 304
 305       if (nir) {
 306          stfp->tgsi.type = PIPE_SHADER_IR_NIR;
 307          stfp->shader_program = shProg;
 308          stfp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
 309          prog->nir = stfp->tgsi.ir.nir;
 310       } else {
 311          read_tgsi_from_cache(&blob_reader, &stfp->tgsi.tokens,
 312                               &stfp->num_tgsi_tokens);
 313       }
 314
 315       if (st->fp == stfp)
 316          st->dirty |= stfp->affected_states;
 317
 318       break;
 319    }
 320    case MESA_SHADER_COMPUTE: {
 321       struct st_compute_program *stcp = (struct st_compute_program *) prog;
 322
 323       st_release_cp_variants(st, stcp);
 324
 325       if (nir) {
 326          stcp->tgsi.ir_type = PIPE_SHADER_IR_NIR;
 327          stcp->shader_program = shProg;
 328          stcp->tgsi.prog = nir_deserialize(NULL, options, &blob_reader);
 329          prog->nir = (nir_shader *) stcp->tgsi.prog;
 330       } else {
 331          read_tgsi_from_cache(&blob_reader,
 332                               (const struct tgsi_token**) &stcp->tgsi.prog,
 333                               &stcp->num_tgsi_tokens);
 334       }
 335
 336       stcp->tgsi.req_local_mem = stcp->Base.info.cs.shared_size;
 337       stcp->tgsi.req_private_mem = 0;
 338       stcp->tgsi.req_input_mem = 0;
 339
 340       if (st->cp == stcp)
 341          st->dirty |= stcp->affected_states;
 342
 343       break;
 344    }
 345    default:
 346       unreachable("Unsupported stage");
 347    }
 348
 349    /* Make sure we don't try to read more data than we wrote. This should
 350     * never happen in release builds but its useful to have this check to
 351     * catch development bugs.
 352     */
 353    if (blob_reader.current != blob_reader.end || blob_reader.overrun) {
 354       assert(!"Invalid TGSI shader disk cache item!");
 355
 356       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 357          fprintf(stderr, "Error reading program from cache (invalid "
 358                  "TGSI cache item)\n");
 359       }
 360    }
 361
 362    st_set_prog_affected_state_flags(prog);
 363    _mesa_associate_uniform_storage(ctx, shProg, prog);
 364
 365    /* Create Gallium shaders now instead of on demand. */
 366    if (ST_DEBUG & DEBUG_PRECOMPILE ||
 367        st->shader_has_one_variant[prog->info.stage])
 368       st_precompile_shader_variant(st, prog);
 369 }
 370
 371 bool
 372 st_load_ir_from_disk_cache(struct gl_context *ctx,
 373                            struct gl_shader_program *prog,
 374                            bool nir)
 375 {
 376    if (!ctx->Cache)
 377       return false;
 378
 379    /* If we didn't load the GLSL metadata from cache then we could not have
 380     * loaded the tgsi either.
 381     */
 382    if (prog->data->LinkStatus != LINKING_SKIPPED)
 383       return false;
 384
 385    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
 386       if (prog->_LinkedShaders[i] == NULL)
 387          continue;
 388
 389       struct gl_program *glprog = prog->_LinkedShaders[i]->Program;
 390       st_deserialise_ir_program(ctx, prog, glprog, nir);
 391
 392       /* We don't need the cached blob anymore so free it */
 393       ralloc_free(glprog->driver_cache_blob);
 394       glprog->driver_cache_blob = NULL;
 395       glprog->driver_cache_blob_size = 0;
 396
 397       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 398          fprintf(stderr, "%s state tracker IR retrieved from cache\n",
 399                  _mesa_shader_stage_to_string(i));
 400       }
 401    }
 402
 403    return true;
 404 }
 405
 406 void
 407 st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
 408 {
 409    st_serialise_ir_program(ctx, prog, false);
 410 }
 411
 412 void
 413 st_serialise_tgsi_program_binary(struct gl_context *ctx,
 414                                  struct gl_shader_program *shProg,
 415                                  struct gl_program *prog)
 416 {
 417    st_serialise_ir_program(ctx, prog, false);
 418 }
 419
 420 void
 421 st_deserialise_tgsi_program(struct gl_context *ctx,
 422                             struct gl_shader_program *shProg,
 423                             struct gl_program *prog)
 424 {
 425    st_deserialise_ir_program(ctx, shProg, prog, false);
 426 }
 427
 428 void
 429 st_serialise_nir_program(struct gl_context *ctx, struct gl_program *prog)
 430 {
 431    st_serialise_ir_program(ctx, prog, true);
 432 }
 433
 434 void
 435 st_serialise_nir_program_binary(struct gl_context *ctx,
 436                                 struct gl_shader_program *shProg,
 437                                 struct gl_program *prog)
 438 {
 439    st_serialise_ir_program(ctx, prog, true);
 440 }
 441
 442 void
 443 st_deserialise_nir_program(struct gl_context *ctx,
 444                            struct gl_shader_program *shProg,
 445                            struct gl_program *prog)
 446 {
 447    st_deserialise_ir_program(ctx, shProg, prog, true);
 448 }