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 "tgsi/tgsi_parse.h"
  35 #include "util/u_memory.h"
  36
  37 void
  38 st_get_program_binary_driver_sha1(struct gl_context *ctx, uint8_t *sha1)
  39 {
  40    disk_cache_compute_key(ctx->Cache, NULL, 0, sha1);
  41 }
  42
  43 static void
  44 write_stream_out_to_cache(struct blob *blob,
  45                           struct pipe_shader_state *state)
  46 {
  47    blob_write_bytes(blob, &state->stream_output,
  48                     sizeof(state->stream_output));
  49 }
  50
  51 static void
  52 copy_blob_to_driver_cache_blob(struct blob *blob, struct gl_program *prog)
  53 {
  54    prog->driver_cache_blob = ralloc_size(NULL, blob->size);
  55    memcpy(prog->driver_cache_blob, blob->data, blob->size);
  56    prog->driver_cache_blob_size = blob->size;
  57 }
  58
  59 static void
  60 write_tgsi_to_cache(struct blob *blob, const struct tgsi_token *tokens,
  61                     struct gl_program *prog)
  62 {
  63    unsigned num_tokens = tgsi_num_tokens(tokens);
  64
  65    blob_write_uint32(blob, num_tokens);
  66    blob_write_bytes(blob, tokens, num_tokens * sizeof(struct tgsi_token));
  67    copy_blob_to_driver_cache_blob(blob, prog);
  68 }
  69
  70 static void
  71 write_nir_to_cache(struct blob *blob, struct gl_program *prog)
  72 {
  73    nir_serialize(blob, prog->nir, false);
  74    copy_blob_to_driver_cache_blob(blob, prog);
  75 }
  76
  77 static void
  78 st_serialise_ir_program(struct gl_context *ctx, struct gl_program *prog,
  79                         bool nir)
  80 {
  81    if (prog->driver_cache_blob)
  82       return;
  83
  84    struct st_program *stp = (struct st_program *)prog;
  85    struct blob blob;
  86    blob_init(&blob);
  87
  88    if (prog->info.stage == MESA_SHADER_VERTEX) {
  89       struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
  90
  91       blob_write_uint32(&blob, stvp->num_inputs);
  92       blob_write_bytes(&blob, stvp->index_to_input,
  93                        sizeof(stvp->index_to_input));
  94       blob_write_bytes(&blob, stvp->input_to_index,
  95                        sizeof(stvp->input_to_index));
  96       blob_write_bytes(&blob, stvp->result_to_output,
  97                        sizeof(stvp->result_to_output));
  98    }
  99
 100    if (prog->info.stage == MESA_SHADER_VERTEX ||
 101        prog->info.stage == MESA_SHADER_TESS_EVAL ||
 102        prog->info.stage == MESA_SHADER_GEOMETRY)
 103       write_stream_out_to_cache(&blob, &stp->state);
 104
 105    if (nir)
 106       write_nir_to_cache(&blob, prog);
 107    else
 108       write_tgsi_to_cache(&blob, stp->state.tokens, prog);
 109
 110    blob_finish(&blob);
 111 }
 112
 113 /**
 114  * Store TGSI or NIR and any other required state in on-disk shader cache.
 115  */
 116 void
 117 st_store_ir_in_disk_cache(struct st_context *st, struct gl_program *prog,
 118                           bool nir)
 119 {
 120    if (!st->ctx->Cache)
 121       return;
 122
 123    /* Exit early when we are dealing with a ff shader with no source file to
 124     * generate a source from.
 125     */
 126    static const char zero[sizeof(prog->sh.data->sha1)] = {0};
 127    if (memcmp(prog->sh.data->sha1, zero, sizeof(prog->sh.data->sha1)) == 0)
 128       return;
 129
 130    st_serialise_ir_program(st->ctx, prog, nir);
 131
 132    if (st->ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 133       fprintf(stderr, "putting %s state tracker IR in cache\n",
 134               _mesa_shader_stage_to_string(prog->info.stage));
 135    }
 136 }
 137
 138 static void
 139 read_stream_out_from_cache(struct blob_reader *blob_reader,
 140                            struct pipe_shader_state *state)
 141 {
 142    blob_copy_bytes(blob_reader, (uint8_t *) &state->stream_output,
 143                     sizeof(state->stream_output));
 144 }
 145
 146 static void
 147 read_tgsi_from_cache(struct blob_reader *blob_reader,
 148                      const struct tgsi_token **tokens)
 149 {
 150    unsigned num_tokens  = blob_read_uint32(blob_reader);
 151    unsigned tokens_size = num_tokens * sizeof(struct tgsi_token);
 152    *tokens = (const struct tgsi_token*) MALLOC(tokens_size);
 153    blob_copy_bytes(blob_reader, (uint8_t *) *tokens, tokens_size);
 154 }
 155
 156 static void
 157 st_deserialise_ir_program(struct gl_context *ctx,
 158                           struct gl_shader_program *shProg,
 159                           struct gl_program *prog, bool nir)
 160 {
 161    struct st_context *st = st_context(ctx);
 162    size_t size = prog->driver_cache_blob_size;
 163    uint8_t *buffer = (uint8_t *) prog->driver_cache_blob;
 164    const struct nir_shader_compiler_options *options =
 165       ctx->Const.ShaderCompilerOptions[prog->info.stage].NirOptions;
 166
 167    st_set_prog_affected_state_flags(prog);
 168    _mesa_associate_uniform_storage(ctx, shProg, prog);
 169
 170    assert(prog->driver_cache_blob && prog->driver_cache_blob_size > 0);
 171
 172    struct st_program *stp = st_program(prog);
 173    struct blob_reader blob_reader;
 174    blob_reader_init(&blob_reader, buffer, size);
 175
 176    if (prog->info.stage == MESA_SHADER_VERTEX) {
 177       st_release_vp_variants(st, stp);
 178
 179       struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
 180       stvp->num_inputs = blob_read_uint32(&blob_reader);
 181       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->index_to_input,
 182                       sizeof(stvp->index_to_input));
 183       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->input_to_index,
 184                       sizeof(stvp->input_to_index));
 185       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->result_to_output,
 186                       sizeof(stvp->result_to_output));
 187    } else if (prog->info.stage == MESA_SHADER_FRAGMENT) {
 188       st_release_fp_variants(st, stp);
 189    } else {
 190       st_release_common_variants(st, stp);
 191    }
 192
 193    if (prog->info.stage == MESA_SHADER_VERTEX ||
 194        prog->info.stage == MESA_SHADER_TESS_EVAL ||
 195        prog->info.stage == MESA_SHADER_GEOMETRY)
 196       read_stream_out_from_cache(&blob_reader, &stp->state);
 197
 198    if (nir) {
 199       stp->state.type = PIPE_SHADER_IR_NIR;
 200       stp->state.ir.nir = nir_deserialize(NULL, options, &blob_reader);
 201       stp->shader_program = shProg;
 202       prog->nir = stp->state.ir.nir;
 203    } else {
 204       read_tgsi_from_cache(&blob_reader, &stp->state.tokens);
 205    }
 206
 207    /* Make sure we don't try to read more data than we wrote. This should
 208     * never happen in release builds but its useful to have this check to
 209     * catch development bugs.
 210     */
 211    if (blob_reader.current != blob_reader.end || blob_reader.overrun) {
 212       assert(!"Invalid TGSI shader disk cache item!");
 213
 214       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 215          fprintf(stderr, "Error reading program from cache (invalid "
 216                  "TGSI cache item)\n");
 217       }
 218    }
 219
 220    st_finalize_program(st, prog);
 221 }
 222
 223 bool
 224 st_load_ir_from_disk_cache(struct gl_context *ctx,
 225                            struct gl_shader_program *prog,
 226                            bool nir)
 227 {
 228    if (!ctx->Cache)
 229       return false;
 230
 231    /* If we didn't load the GLSL metadata from cache then we could not have
 232     * loaded TGSI or NIR either.
 233     */
 234    if (prog->data->LinkStatus != LINKING_SKIPPED)
 235       return false;
 236
 237    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
 238       if (prog->_LinkedShaders[i] == NULL)
 239          continue;
 240
 241       struct gl_program *glprog = prog->_LinkedShaders[i]->Program;
 242       st_deserialise_ir_program(ctx, prog, glprog, nir);
 243
 244       /* We don't need the cached blob anymore so free it */
 245       ralloc_free(glprog->driver_cache_blob);
 246       glprog->driver_cache_blob = NULL;
 247       glprog->driver_cache_blob_size = 0;
 248
 249       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 250          fprintf(stderr, "%s state tracker IR retrieved from cache\n",
 251                  _mesa_shader_stage_to_string(i));
 252       }
 253    }
 254
 255    return true;
 256 }
 257
 258 void
 259 st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
 260 {
 261    st_serialise_ir_program(ctx, prog, false);
 262 }
 263
 264 void
 265 st_serialise_tgsi_program_binary(struct gl_context *ctx,
 266                                  struct gl_shader_program *shProg,
 267                                  struct gl_program *prog)
 268 {
 269    st_serialise_ir_program(ctx, prog, false);
 270 }
 271
 272 void
 273 st_deserialise_tgsi_program(struct gl_context *ctx,
 274                             struct gl_shader_program *shProg,
 275                             struct gl_program *prog)
 276 {
 277    st_deserialise_ir_program(ctx, shProg, prog, false);
 278 }
 279
 280 void
 281 st_serialise_nir_program(struct gl_context *ctx, struct gl_program *prog)
 282 {
 283    st_serialise_ir_program(ctx, prog, true);
 284 }
 285
 286 void
 287 st_serialise_nir_program_binary(struct gl_context *ctx,
 288                                 struct gl_shader_program *shProg,
 289                                 struct gl_program *prog)
 290 {
 291    st_serialise_ir_program(ctx, prog, true);
 292 }
 293
 294 void
 295 st_deserialise_nir_program(struct gl_context *ctx,
 296                            struct gl_shader_program *shProg,
 297                            struct gl_program *prog)
 298 {
 299    st_deserialise_ir_program(ctx, shProg, prog, true);
 300 }