src/gallium/drivers/iris/iris_program_cache.c

   1 /*
   2  * Copyright © 2017 Intel Corporation
   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 shall be included
  12  * in all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  15  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  18  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  19  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  20  * DEALINGS IN THE SOFTWARE.
  21  */
  22
  23 /**
  24  * @file iris_program_cache.c
  25  *
  26  * The in-memory program cache.  This is basically a hash table mapping
  27  * API-specified shaders and a state key to a compiled variant.  It also
  28  * takes care of uploading shader assembly into a BO for use on the GPU.
  29  */
  30
  31 #include <stdio.h>
  32 #include <errno.h>
  33 #include "pipe/p_defines.h"
  34 #include "pipe/p_state.h"
  35 #include "pipe/p_context.h"
  36 #include "pipe/p_screen.h"
  37 #include "util/u_atomic.h"
  38 #include "util/u_upload_mgr.h"
  39 #include "compiler/nir/nir.h"
  40 #include "compiler/nir/nir_builder.h"
  41 #include "intel/compiler/brw_compiler.h"
  42 #include "intel/compiler/brw_eu.h"
  43 #include "intel/compiler/brw_nir.h"
  44 #include "iris_context.h"
  45 #include "iris_resource.h"
  46
  47 struct keybox {
  48    uint8_t size;
  49    enum iris_program_cache_id cache_id;
  50    uint8_t data[0];
  51 };
  52
  53 static uint32_t
  54 key_size_for_cache(enum iris_program_cache_id cache_id)
  55 {
  56    static const unsigned key_sizes[] = {
  57       [IRIS_CACHE_VS]         = sizeof(struct brw_vs_prog_key),
  58       [IRIS_CACHE_TCS]        = sizeof(struct brw_tcs_prog_key),
  59       [IRIS_CACHE_TES]        = sizeof(struct brw_tes_prog_key),
  60       [IRIS_CACHE_GS]         = sizeof(struct brw_gs_prog_key),
  61       [IRIS_CACHE_FS]         = sizeof(struct brw_wm_prog_key),
  62       [IRIS_CACHE_CS]         = sizeof(struct brw_cs_prog_key),
  63    };
  64
  65    /* BLORP keys aren't all the same size. */
  66    assert(cache_id != IRIS_CACHE_BLORP);
  67
  68    return key_sizes[cache_id];
  69 }
  70
  71 static struct keybox *
  72 make_keybox(void *mem_ctx,
  73             enum iris_program_cache_id cache_id,
  74             const void *key,
  75             uint32_t key_size)
  76 {
  77    struct keybox *keybox =
  78       ralloc_size(mem_ctx, sizeof(struct keybox) + key_size);
  79
  80    keybox->cache_id = cache_id;
  81    keybox->size = key_size;
  82    memcpy(keybox->data, key, key_size);
  83
  84    return keybox;
  85 }
  86
  87 static uint32_t
  88 keybox_hash(const void *void_key)
  89 {
  90    const struct keybox *key = void_key;
  91    return _mesa_hash_data(&key->cache_id, key->size + sizeof(key->cache_id));
  92 }
  93
  94 static bool
  95 keybox_equals(const void *void_a, const void *void_b)
  96 {
  97    const struct keybox *a = void_a, *b = void_b;
  98    if (a->size != b->size)
  99       return false;
 100
 101    return memcmp(a->data, b->data, a->size) == 0;
 102 }
 103
 104 static uint64_t
 105 dirty_flag_for_cache(enum iris_program_cache_id cache_id)
 106 {
 107    assert(cache_id <= MESA_SHADER_STAGES);
 108
 109    uint64_t flags = (IRIS_DIRTY_VS |
 110                      IRIS_DIRTY_BINDINGS_VS |
 111                      IRIS_DIRTY_CONSTANTS_VS) << cache_id;
 112    // XXX: ugly...
 113    // XXX: move this flagging out to a higher level, allow comparison of
 114    // XXX: new and old programs to decide what bits to twiddle
 115    // XXX: CLIP: toggle if barycentric modes has any NONPERSPECTIVE or not
 116    if (cache_id == IRIS_CACHE_FS)
 117       flags |= IRIS_DIRTY_WM | IRIS_DIRTY_CLIP | IRIS_DIRTY_SBE;
 118    if (cache_id == IRIS_CACHE_VS)
 119       flags |= IRIS_DIRTY_VF_SGVS;
 120
 121    return flags;
 122 }
 123
 124 static unsigned
 125 get_program_string_id(enum iris_program_cache_id cache_id, const void *key)
 126 {
 127    switch (cache_id) {
 128    case IRIS_CACHE_VS:
 129       return ((struct brw_vs_prog_key *) key)->program_string_id;
 130    case IRIS_CACHE_TCS:
 131       return ((struct brw_tcs_prog_key *) key)->program_string_id;
 132    case IRIS_CACHE_TES:
 133       return ((struct brw_tes_prog_key *) key)->program_string_id;
 134    case IRIS_CACHE_GS:
 135       return ((struct brw_gs_prog_key *) key)->program_string_id;
 136    case IRIS_CACHE_CS:
 137       return ((struct brw_cs_prog_key *) key)->program_string_id;
 138    case IRIS_CACHE_FS:
 139       return ((struct brw_wm_prog_key *) key)->program_string_id;
 140    default:
 141       unreachable("no program string id for this kind of program");
 142    }
 143 }
 144
 145 static struct iris_compiled_shader *
 146 iris_find_cached_shader(struct iris_context *ice,
 147                         enum iris_program_cache_id cache_id,
 148                         const void *key,
 149                         uint32_t key_size)
 150 {
 151    struct keybox *keybox =
 152       make_keybox(ice->shaders.cache, cache_id, key, key_size);
 153    struct hash_entry *entry =
 154       _mesa_hash_table_search(ice->shaders.cache, keybox);
 155
 156    ralloc_free(keybox);
 157
 158    return entry ? entry->data : NULL;
 159 }
 160
 161 /**
 162  * Looks for a program in the cache and binds it.
 163  *
 164  * If no program was found, returns false and leaves the binding alone.
 165  */
 166 bool
 167 iris_bind_cached_shader(struct iris_context *ice,
 168                         enum iris_program_cache_id cache_id,
 169                         const void *key)
 170 {
 171    unsigned key_size = key_size_for_cache(cache_id);
 172    struct iris_compiled_shader *shader =
 173       iris_find_cached_shader(ice, cache_id, key, key_size);
 174
 175    if (!shader)
 176       return false;
 177
 178    // XXX: why memcmp?
 179    if (!ice->shaders.prog[cache_id] ||
 180        memcmp(shader, ice->shaders.prog[cache_id], sizeof(*shader)) != 0) {
 181       ice->shaders.prog[cache_id] = shader;
 182       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 183    }
 184
 185    return true;
 186 }
 187
 188 void
 189 iris_unbind_shader(struct iris_context *ice,
 190                    enum iris_program_cache_id cache_id)
 191 {
 192    if (ice->shaders.prog[cache_id]) {
 193       ice->shaders.prog[cache_id] = NULL;
 194       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 195    }
 196 }
 197
 198 const void *
 199 iris_find_previous_compile(const struct iris_context *ice,
 200                            enum iris_program_cache_id cache_id,
 201                            unsigned program_string_id)
 202 {
 203    hash_table_foreach(ice->shaders.cache, entry) {
 204       const struct keybox *keybox = entry->key;
 205       if (keybox->cache_id == cache_id &&
 206           get_program_string_id(cache_id, keybox->data) == program_string_id) {
 207          return keybox->data;
 208       }
 209    }
 210
 211    return NULL;
 212 }
 213
 214 /**
 215  * Look for an existing entry in the cache that has identical assembly code.
 216  *
 217  * This is useful for programs generating shaders at runtime, where multiple
 218  * distinct shaders (from an API perspective) may compile to the same assembly
 219  * in our backend.  This saves space in the program cache buffer.
 220  */
 221 static const struct iris_compiled_shader *
 222 find_existing_assembly(struct hash_table *cache,
 223                        const void *assembly,
 224                        unsigned assembly_size)
 225 {
 226    hash_table_foreach(cache, entry) {
 227       const struct iris_compiled_shader *existing = entry->data;
 228       if (existing->prog_data->program_size == assembly_size &&
 229           memcmp(existing->map, assembly, assembly_size) == 0)
 230          return existing;
 231    }
 232    return NULL;
 233 }
 234
 235 static struct iris_compiled_shader *
 236 iris_upload_shader(struct iris_context *ice,
 237                    enum iris_program_cache_id cache_id,
 238                    uint32_t key_size,
 239                    const void *key,
 240                    const void *assembly,
 241                    struct brw_stage_prog_data *prog_data,
 242                    uint32_t *streamout)
 243 {
 244    struct iris_screen *screen = (void *) ice->ctx.screen;
 245    struct gen_device_info *devinfo = &screen->devinfo;
 246    struct hash_table *cache = ice->shaders.cache;
 247    struct iris_compiled_shader *shader =
 248       rzalloc_size(cache, sizeof(struct iris_compiled_shader) +
 249                    ice->vtbl.derived_program_state_size(cache_id));
 250    const struct iris_compiled_shader *existing =
 251       find_existing_assembly(cache, assembly, prog_data->program_size);
 252
 253    /* If we can find a matching prog in the cache already, then reuse the
 254     * existing stuff without creating new copy into the underlying buffer
 255     * object.  This is notably useful for programs generating shaders at
 256     * runtime, where multiple shaders may compile to the same thing in our
 257     * backend.
 258     */
 259    if (existing) {
 260       pipe_resource_reference(&shader->assembly.res, existing->assembly.res);
 261       shader->assembly.offset = existing->assembly.offset;
 262       shader->map = existing->map;
 263    } else {
 264       shader->assembly.res = NULL;
 265       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size, 64,
 266                      &shader->assembly.offset, &shader->assembly.res,
 267                      &shader->map);
 268       memcpy(shader->map, assembly, prog_data->program_size);
 269    }
 270
 271    shader->prog_data = prog_data;
 272    shader->streamout = streamout;
 273
 274    ralloc_steal(shader, shader->prog_data);
 275    ralloc_steal(shader->prog_data, prog_data->param);
 276    ralloc_steal(shader->prog_data, prog_data->pull_param);
 277    ralloc_steal(shader, shader->streamout);
 278
 279    /* Store the 3DSTATE shader packets and other derived state. */
 280    ice->vtbl.store_derived_program_state(devinfo, cache_id, shader);
 281
 282    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 283    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 284
 285    return shader;
 286 }
 287
 288 /**
 289  * Upload a new shader to the program cache, and bind it for use.
 290  *
 291  * \param prog_data must be ralloc'd and will be stolen.
 292  */
 293 void
 294 iris_upload_and_bind_shader(struct iris_context *ice,
 295                             enum iris_program_cache_id cache_id,
 296                             const void *key,
 297                             const void *assembly,
 298                             struct brw_stage_prog_data *prog_data,
 299                             uint32_t *streamout)
 300 {
 301    assert(cache_id != IRIS_CACHE_BLORP);
 302
 303    struct iris_compiled_shader *shader =
 304       iris_upload_shader(ice, cache_id, key_size_for_cache(cache_id), key,
 305                          assembly, prog_data, streamout);
 306
 307    ice->shaders.prog[cache_id] = shader;
 308    ice->state.dirty |= dirty_flag_for_cache(cache_id);
 309 }
 310
 311 bool
 312 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 313                          const void *key, uint32_t key_size,
 314                          uint32_t *kernel_out, void *prog_data_out)
 315 {
 316    struct blorp_context *blorp = blorp_batch->blorp;
 317    struct iris_context *ice = blorp->driver_ctx;
 318    struct iris_batch *batch = blorp_batch->driver_batch;
 319    struct iris_compiled_shader *shader =
 320       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key, key_size);
 321
 322    if (!shader)
 323       return false;
 324
 325    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 326    *kernel_out =
 327       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 328    *((void **) prog_data_out) = shader->prog_data;
 329
 330    iris_use_pinned_bo(batch, bo, false);
 331
 332    return true;
 333 }
 334
 335 bool
 336 iris_blorp_upload_shader(struct blorp_batch *blorp_batch,
 337                          const void *key, uint32_t key_size,
 338                          const void *kernel, UNUSED uint32_t kernel_size,
 339                          const struct brw_stage_prog_data *prog_data_templ,
 340                          UNUSED uint32_t prog_data_size,
 341                          uint32_t *kernel_out, void *prog_data_out)
 342 {
 343    struct blorp_context *blorp = blorp_batch->blorp;
 344    struct iris_context *ice = blorp->driver_ctx;
 345    struct iris_batch *batch = blorp_batch->driver_batch;
 346
 347    void *prog_data = ralloc_size(NULL, prog_data_size);
 348    memcpy(prog_data, prog_data_templ, prog_data_size);
 349
 350    struct iris_compiled_shader *shader =
 351       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 352                          prog_data, NULL);
 353
 354    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 355    *kernel_out =
 356       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 357    *((void **) prog_data_out) = shader->prog_data;
 358
 359    iris_use_pinned_bo(batch, bo, false);
 360
 361    return true;
 362 }
 363
 364 void
 365 iris_init_program_cache(struct iris_context *ice)
 366 {
 367    ice->shaders.cache =
 368       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 369
 370    ice->shaders.uploader =
 371       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 372                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 373 }
 374
 375 void
 376 iris_destroy_program_cache(struct iris_context *ice)
 377 {
 378    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 379       ice->shaders.prog[i] = NULL;
 380    }
 381
 382    hash_table_foreach(ice->shaders.cache, entry) {
 383       struct iris_compiled_shader *shader = entry->data;
 384       pipe_resource_reference(&shader->assembly.res, NULL);
 385    }
 386
 387    u_upload_destroy(ice->shaders.uploader);
 388
 389    ralloc_free(ice->shaders.cache);
 390 }
 391
 392 static const char *
 393 cache_name(enum iris_program_cache_id cache_id)
 394 {
 395    if (cache_id == IRIS_CACHE_BLORP)
 396       return "BLORP";
 397
 398    return _mesa_shader_stage_to_string(cache_id);
 399 }
 400
 401 void
 402 iris_print_program_cache(struct iris_context *ice)
 403 {
 404    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 405    const struct gen_device_info *devinfo = &screen->devinfo;
 406
 407    hash_table_foreach(ice->shaders.cache, entry) {
 408       const struct keybox *keybox = entry->key;
 409       struct iris_compiled_shader *shader = entry->data;
 410       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 411       brw_disassemble(devinfo, shader->map, 0,
 412                       shader->prog_data->program_size, stderr);
 413    }
 414 }