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_uint32(blob, state->stream_output.num_outputs);
  48    if (state->stream_output.num_outputs) {
  49       blob_write_bytes(blob, &state->stream_output.stride,
  50                        sizeof(state->stream_output.stride));
  51       blob_write_bytes(blob, &state->stream_output.output,
  52                        sizeof(state->stream_output.output));
  53    }
  54 }
  55
  56 static void
  57 copy_blob_to_driver_cache_blob(struct blob *blob, struct gl_program *prog)
  58 {
  59    prog->driver_cache_blob = ralloc_size(NULL, blob->size);
  60    memcpy(prog->driver_cache_blob, blob->data, blob->size);
  61    prog->driver_cache_blob_size = blob->size;
  62 }
  63
  64 static void
  65 write_tgsi_to_cache(struct blob *blob, const struct tgsi_token *tokens,
  66                     struct gl_program *prog)
  67 {
  68    unsigned num_tokens = tgsi_num_tokens(tokens);
  69
  70    blob_write_uint32(blob, num_tokens);
  71    blob_write_bytes(blob, tokens, num_tokens * sizeof(struct tgsi_token));
  72    copy_blob_to_driver_cache_blob(blob, prog);
  73 }
  74
  75 static void
  76 write_nir_to_cache(struct blob *blob, struct gl_program *prog)
  77 {
  78    nir_serialize(blob, prog->nir, false);
  79    copy_blob_to_driver_cache_blob(blob, prog);
  80 }
  81
  82 static void
  83 st_serialise_ir_program(struct gl_context *ctx, struct gl_program *prog,
  84                         bool nir)
  85 {
  86    if (prog->driver_cache_blob)
  87       return;
  88
  89    struct st_program *stp = (struct st_program *)prog;
  90    struct blob blob;
  91    blob_init(&blob);
  92
  93    if (prog->info.stage == MESA_SHADER_VERTEX) {
  94       struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
  95
  96       blob_write_uint32(&blob, stvp->num_inputs);
  97       blob_write_bytes(&blob, stvp->index_to_input,
  98                        sizeof(stvp->index_to_input));
  99       blob_write_bytes(&blob, stvp->input_to_index,
 100                        sizeof(stvp->input_to_index));
 101       blob_write_bytes(&blob, stvp->result_to_output,
 102                        sizeof(stvp->result_to_output));
 103    }
 104
 105    if (prog->info.stage == MESA_SHADER_VERTEX ||
 106        prog->info.stage == MESA_SHADER_TESS_EVAL ||
 107        prog->info.stage == MESA_SHADER_GEOMETRY)
 108       write_stream_out_to_cache(&blob, &stp->state);
 109
 110    if (nir)
 111       write_nir_to_cache(&blob, prog);
 112    else
 113       write_tgsi_to_cache(&blob, stp->state.tokens, prog);
 114
 115    blob_finish(&blob);
 116 }
 117
 118 /**
 119  * Store TGSI or NIR and any other required state in on-disk shader cache.
 120  */
 121 void
 122 st_store_ir_in_disk_cache(struct st_context *st, struct gl_program *prog,
 123                           bool nir)
 124 {
 125    if (!st->ctx->Cache)
 126       return;
 127
 128    /* Exit early when we are dealing with a ff shader with no source file to
 129     * generate a source from.
 130     */
 131    static const char zero[sizeof(prog->sh.data->sha1)] = {0};
 132    if (memcmp(prog->sh.data->sha1, zero, sizeof(prog->sh.data->sha1)) == 0)
 133       return;
 134
 135    st_serialise_ir_program(st->ctx, prog, nir);
 136
 137    if (st->ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 138       fprintf(stderr, "putting %s state tracker IR in cache\n",
 139               _mesa_shader_stage_to_string(prog->info.stage));
 140    }
 141 }
 142
 143 static void
 144 read_stream_out_from_cache(struct blob_reader *blob_reader,
 145                            struct pipe_shader_state *state)
 146 {
 147    memset(&state->stream_output, 0, sizeof(state->stream_output));
 148    state->stream_output.num_outputs = blob_read_uint32(blob_reader);
 149    if (state->stream_output.num_outputs) {
 150       blob_copy_bytes(blob_reader, &state->stream_output.stride,
 151                       sizeof(state->stream_output.stride));
 152       blob_copy_bytes(blob_reader, &state->stream_output.output,
 153                       sizeof(state->stream_output.output));
 154    }
 155 }
 156
 157 static void
 158 read_tgsi_from_cache(struct blob_reader *blob_reader,
 159                      const struct tgsi_token **tokens)
 160 {
 161    unsigned num_tokens  = blob_read_uint32(blob_reader);
 162    unsigned tokens_size = num_tokens * sizeof(struct tgsi_token);
 163    *tokens = (const struct tgsi_token*) MALLOC(tokens_size);
 164    blob_copy_bytes(blob_reader, (uint8_t *) *tokens, tokens_size);
 165 }
 166
 167 static void
 168 st_deserialise_ir_program(struct gl_context *ctx,
 169                           struct gl_shader_program *shProg,
 170                           struct gl_program *prog, bool nir)
 171 {
 172    struct st_context *st = st_context(ctx);
 173    size_t size = prog->driver_cache_blob_size;
 174    uint8_t *buffer = (uint8_t *) prog->driver_cache_blob;
 175    const struct nir_shader_compiler_options *options =
 176       ctx->Const.ShaderCompilerOptions[prog->info.stage].NirOptions;
 177
 178    st_set_prog_affected_state_flags(prog);
 179    _mesa_associate_uniform_storage(ctx, shProg, prog);
 180
 181    assert(prog->driver_cache_blob && prog->driver_cache_blob_size > 0);
 182
 183    struct st_program *stp = st_program(prog);
 184    struct blob_reader blob_reader;
 185    blob_reader_init(&blob_reader, buffer, size);
 186
 187    st_release_variants(st, stp);
 188
 189    if (prog->info.stage == MESA_SHADER_VERTEX) {
 190       struct st_vertex_program *stvp = (struct st_vertex_program *)stp;
 191       stvp->num_inputs = blob_read_uint32(&blob_reader);
 192       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->index_to_input,
 193                       sizeof(stvp->index_to_input));
 194       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->input_to_index,
 195                       sizeof(stvp->input_to_index));
 196       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->result_to_output,
 197                       sizeof(stvp->result_to_output));
 198    }
 199
 200    if (prog->info.stage == MESA_SHADER_VERTEX ||
 201        prog->info.stage == MESA_SHADER_TESS_EVAL ||
 202        prog->info.stage == MESA_SHADER_GEOMETRY)
 203       read_stream_out_from_cache(&blob_reader, &stp->state);
 204
 205    if (nir) {
 206       stp->state.type = PIPE_SHADER_IR_NIR;
 207       stp->shader_program = shProg;
 208       prog->nir = nir_deserialize(NULL, options, &blob_reader);
 209    } else {
 210       read_tgsi_from_cache(&blob_reader, &stp->state.tokens);
 211    }
 212
 213    /* Make sure we don't try to read more data than we wrote. This should
 214     * never happen in release builds but its useful to have this check to
 215     * catch development bugs.
 216     */
 217    if (blob_reader.current != blob_reader.end || blob_reader.overrun) {
 218       assert(!"Invalid TGSI shader disk cache item!");
 219
 220       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 221          fprintf(stderr, "Error reading program from cache (invalid "
 222                  "TGSI cache item)\n");
 223       }
 224    }
 225
 226    st_finalize_program(st, prog);
 227 }
 228
 229 bool
 230 st_load_ir_from_disk_cache(struct gl_context *ctx,
 231                            struct gl_shader_program *prog,
 232                            bool nir)
 233 {
 234    if (!ctx->Cache)
 235       return false;
 236
 237    /* If we didn't load the GLSL metadata from cache then we could not have
 238     * loaded TGSI or NIR either.
 239     */
 240    if (prog->data->LinkStatus != LINKING_SKIPPED)
 241       return false;
 242
 243    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
 244       if (prog->_LinkedShaders[i] == NULL)
 245          continue;
 246
 247       struct gl_program *glprog = prog->_LinkedShaders[i]->Program;
 248       st_deserialise_ir_program(ctx, prog, glprog, nir);
 249
 250       /* We don't need the cached blob anymore so free it */
 251       ralloc_free(glprog->driver_cache_blob);
 252       glprog->driver_cache_blob = NULL;
 253       glprog->driver_cache_blob_size = 0;
 254
 255       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
 256          fprintf(stderr, "%s state tracker IR retrieved from cache\n",
 257                  _mesa_shader_stage_to_string(i));
 258       }
 259    }
 260
 261    return true;
 262 }
 263
 264 void
 265 st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
 266 {
 267    st_serialise_ir_program(ctx, prog, false);
 268 }
 269
 270 void
 271 st_serialise_tgsi_program_binary(struct gl_context *ctx,
 272                                  struct gl_shader_program *shProg,
 273                                  struct gl_program *prog)
 274 {
 275    st_serialise_ir_program(ctx, prog, false);
 276 }
 277
 278 void
 279 st_deserialise_tgsi_program(struct gl_context *ctx,
 280                             struct gl_shader_program *shProg,
 281                             struct gl_program *prog)
 282 {
 283    st_deserialise_ir_program(ctx, shProg, prog, false);
 284 }
 285
 286 void
 287 st_serialise_nir_program(struct gl_context *ctx, struct gl_program *prog)
 288 {
 289    st_serialise_ir_program(ctx, prog, true);
 290 }
 291
 292 void
 293 st_serialise_nir_program_binary(struct gl_context *ctx,
 294                                 struct gl_shader_program *shProg,
 295                                 struct gl_program *prog)
 296 {
 297    st_serialise_ir_program(ctx, prog, true);
 298 }
 299
 300 void
 301 st_deserialise_nir_program(struct gl_context *ctx,
 302                            struct gl_shader_program *shProg,
 303                            struct gl_program *prog)
 304 {
 305    st_deserialise_ir_program(ctx, shProg, prog, true);
 306 }