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    uint16_t size;
  49    enum iris_program_cache_id cache_id;
  50    uint8_t data[0];
  51 };
  52
  53 static struct keybox *
  54 make_keybox(void *mem_ctx,
  55             enum iris_program_cache_id cache_id,
  56             const void *key,
  57             uint32_t key_size)
  58 {
  59    struct keybox *keybox =
  60       ralloc_size(mem_ctx, sizeof(struct keybox) + key_size);
  61
  62    keybox->cache_id = cache_id;
  63    keybox->size = key_size;
  64    memcpy(keybox->data, key, key_size);
  65
  66    return keybox;
  67 }
  68
  69 static uint32_t
  70 keybox_hash(const void *void_key)
  71 {
  72    const struct keybox *key = void_key;
  73    return _mesa_hash_data(&key->cache_id, key->size + sizeof(key->cache_id));
  74 }
  75
  76 static bool
  77 keybox_equals(const void *void_a, const void *void_b)
  78 {
  79    const struct keybox *a = void_a, *b = void_b;
  80    if (a->size != b->size)
  81       return false;
  82
  83    return memcmp(a->data, b->data, a->size) == 0;
  84 }
  85
  86 struct iris_compiled_shader *
  87 iris_find_cached_shader(struct iris_context *ice,
  88                         enum iris_program_cache_id cache_id,
  89                         uint32_t key_size,
  90                         const void *key)
  91 {
  92    struct keybox *keybox = make_keybox(NULL, cache_id, key, key_size);
  93    struct hash_entry *entry =
  94       _mesa_hash_table_search(ice->shaders.cache, keybox);
  95
  96    ralloc_free(keybox);
  97
  98    return entry ? entry->data : NULL;
  99 }
 100
 101 const void *
 102 iris_find_previous_compile(const struct iris_context *ice,
 103                            enum iris_program_cache_id cache_id,
 104                            unsigned program_string_id)
 105 {
 106    hash_table_foreach(ice->shaders.cache, entry) {
 107       const struct keybox *keybox = entry->key;
 108       const struct brw_base_prog_key *key = (const void *)keybox->data;
 109       if (keybox->cache_id == cache_id &&
 110           key->program_string_id == program_string_id) {
 111          return keybox->data;
 112       }
 113    }
 114
 115    return NULL;
 116 }
 117
 118 void
 119 iris_delete_shader_variants(struct iris_context *ice,
 120                             struct iris_uncompiled_shader *ish)
 121 {
 122    struct hash_table *cache = ice->shaders.cache;
 123    gl_shader_stage stage = ish->nir->info.stage;
 124    enum iris_program_cache_id cache_id = stage;
 125
 126    hash_table_foreach(cache, entry) {
 127       const struct keybox *keybox = entry->key;
 128       const struct brw_base_prog_key *key = (const void *)keybox->data;
 129
 130       if (keybox->cache_id == cache_id &&
 131           key->program_string_id == ish->program_id) {
 132          struct iris_compiled_shader *shader = entry->data;
 133
 134          _mesa_hash_table_remove(cache, entry);
 135
 136          /* Shader variants may still be bound in the context even after
 137           * the API-facing shader has been deleted.  In particular, a draw
 138           * may not have triggered iris_update_compiled_shaders() yet.  In
 139           * that case, we may be referring to that shader's VUE map, stream
 140           * output settings, and so on.  We also like to compare the old and
 141           * new shader programs when swapping them out to flag dirty state.
 142           *
 143           * So, it's hazardous to delete a bound shader variant.  We avoid
 144           * doing so, choosing to instead move "deleted" shader variants to
 145           * a list, deferring the actual deletion until they're not bound.
 146           *
 147           * For simplicity, we always move deleted variants to the list,
 148           * even if we could delete them immediately.  We'll then process
 149           * the list, catching both these variants and any others.
 150           */
 151          list_addtail(&shader->link, &ice->shaders.deleted_variants[stage]);
 152       }
 153    }
 154
 155    /* Process any pending deferred variant deletions. */
 156    list_for_each_entry_safe(struct iris_compiled_shader, shader,
 157                             &ice->shaders.deleted_variants[stage], link) {
 158       /* If the shader is still bound, defer deletion. */
 159       if (ice->shaders.prog[stage] == shader)
 160          continue;
 161
 162       list_del(&shader->link);
 163
 164       /* Actually delete the variant. */
 165       pipe_resource_reference(&shader->assembly.res, NULL);
 166       ralloc_free(shader);
 167    }
 168 }
 169
 170
 171 /**
 172  * Look for an existing entry in the cache that has identical assembly code.
 173  *
 174  * This is useful for programs generating shaders at runtime, where multiple
 175  * distinct shaders (from an API perspective) may compile to the same assembly
 176  * in our backend.  This saves space in the program cache buffer.
 177  */
 178 static const struct iris_compiled_shader *
 179 find_existing_assembly(struct hash_table *cache,
 180                        const void *assembly,
 181                        unsigned assembly_size)
 182 {
 183    hash_table_foreach(cache, entry) {
 184       const struct iris_compiled_shader *existing = entry->data;
 185       if (existing->prog_data->program_size == assembly_size &&
 186           memcmp(existing->map, assembly, assembly_size) == 0)
 187          return existing;
 188    }
 189    return NULL;
 190 }
 191
 192 struct iris_compiled_shader *
 193 iris_upload_shader(struct iris_context *ice,
 194                    enum iris_program_cache_id cache_id,
 195                    uint32_t key_size,
 196                    const void *key,
 197                    const void *assembly,
 198                    struct brw_stage_prog_data *prog_data,
 199                    uint32_t *streamout,
 200                    enum brw_param_builtin *system_values,
 201                    unsigned num_system_values,
 202                    unsigned kernel_input_size,
 203                    unsigned num_cbufs,
 204                    const struct iris_binding_table *bt)
 205 {
 206    struct hash_table *cache = ice->shaders.cache;
 207    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 208    struct iris_compiled_shader *shader =
 209       rzalloc_size(cache, sizeof(struct iris_compiled_shader) +
 210                    screen->vtbl.derived_program_state_size(cache_id));
 211    const struct iris_compiled_shader *existing =
 212       find_existing_assembly(cache, assembly, prog_data->program_size);
 213
 214    /* If we can find a matching prog in the cache already, then reuse the
 215     * existing stuff without creating new copy into the underlying buffer
 216     * object.  This is notably useful for programs generating shaders at
 217     * runtime, where multiple shaders may compile to the same thing in our
 218     * backend.
 219     */
 220    if (existing) {
 221       pipe_resource_reference(&shader->assembly.res, existing->assembly.res);
 222       shader->assembly.offset = existing->assembly.offset;
 223       shader->map = existing->map;
 224    } else {
 225       shader->assembly.res = NULL;
 226       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size, 64,
 227                      &shader->assembly.offset, &shader->assembly.res,
 228                      &shader->map);
 229       memcpy(shader->map, assembly, prog_data->program_size);
 230    }
 231
 232    list_inithead(&shader->link);
 233
 234    shader->prog_data = prog_data;
 235    shader->streamout = streamout;
 236    shader->system_values = system_values;
 237    shader->num_system_values = num_system_values;
 238    shader->kernel_input_size = kernel_input_size;
 239    shader->num_cbufs = num_cbufs;
 240    shader->bt = *bt;
 241
 242    ralloc_steal(shader, shader->prog_data);
 243    ralloc_steal(shader->prog_data, prog_data->param);
 244    ralloc_steal(shader->prog_data, prog_data->pull_param);
 245    ralloc_steal(shader, shader->streamout);
 246    ralloc_steal(shader, shader->system_values);
 247
 248    /* Store the 3DSTATE shader packets and other derived state. */
 249    screen->vtbl.store_derived_program_state(ice, cache_id, shader);
 250
 251    struct keybox *keybox = make_keybox(shader, cache_id, key, key_size);
 252    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 253
 254    return shader;
 255 }
 256
 257 bool
 258 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 259                          const void *key, uint32_t key_size,
 260                          uint32_t *kernel_out, void *prog_data_out)
 261 {
 262    struct blorp_context *blorp = blorp_batch->blorp;
 263    struct iris_context *ice = blorp->driver_ctx;
 264    struct iris_batch *batch = blorp_batch->driver_batch;
 265    struct iris_compiled_shader *shader =
 266       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key_size, key);
 267
 268    if (!shader)
 269       return false;
 270
 271    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 272    *kernel_out =
 273       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 274    *((void **) prog_data_out) = shader->prog_data;
 275
 276    iris_use_pinned_bo(batch, bo, false, IRIS_DOMAIN_NONE);
 277
 278    return true;
 279 }
 280
 281 bool
 282 iris_blorp_upload_shader(struct blorp_batch *blorp_batch, uint32_t stage,
 283                          const void *key, uint32_t key_size,
 284                          const void *kernel, UNUSED uint32_t kernel_size,
 285                          const struct brw_stage_prog_data *prog_data_templ,
 286                          UNUSED uint32_t prog_data_size,
 287                          uint32_t *kernel_out, void *prog_data_out)
 288 {
 289    struct blorp_context *blorp = blorp_batch->blorp;
 290    struct iris_context *ice = blorp->driver_ctx;
 291    struct iris_batch *batch = blorp_batch->driver_batch;
 292
 293    void *prog_data = ralloc_size(NULL, prog_data_size);
 294    memcpy(prog_data, prog_data_templ, prog_data_size);
 295
 296    struct iris_binding_table bt;
 297    memset(&bt, 0, sizeof(bt));
 298
 299    struct iris_compiled_shader *shader =
 300       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 301                          prog_data, NULL, NULL, 0, 0, 0, &bt);
 302
 303    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 304    *kernel_out =
 305       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 306    *((void **) prog_data_out) = shader->prog_data;
 307
 308    iris_use_pinned_bo(batch, bo, false, IRIS_DOMAIN_NONE);
 309
 310    return true;
 311 }
 312
 313 void
 314 iris_init_program_cache(struct iris_context *ice)
 315 {
 316    ice->shaders.cache =
 317       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 318
 319    ice->shaders.uploader =
 320       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 321                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 322
 323    for (int i = 0; i < MESA_SHADER_STAGES; i++)
 324       list_inithead(&ice->shaders.deleted_variants[i]);
 325 }
 326
 327 void
 328 iris_destroy_program_cache(struct iris_context *ice)
 329 {
 330    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 331       ice->shaders.prog[i] = NULL;
 332
 333       list_for_each_entry_safe(struct iris_compiled_shader, shader,
 334                                &ice->shaders.deleted_variants[i], link) {
 335          pipe_resource_reference(&shader->assembly.res, NULL);
 336       }
 337    }
 338
 339    hash_table_foreach(ice->shaders.cache, entry) {
 340       struct iris_compiled_shader *shader = entry->data;
 341       pipe_resource_reference(&shader->assembly.res, NULL);
 342    }
 343
 344    u_upload_destroy(ice->shaders.uploader);
 345
 346    ralloc_free(ice->shaders.cache);
 347 }
 348
 349 static const char *
 350 cache_name(enum iris_program_cache_id cache_id)
 351 {
 352    if (cache_id == IRIS_CACHE_BLORP)
 353       return "BLORP";
 354
 355    return _mesa_shader_stage_to_string(cache_id);
 356 }
 357
 358 void
 359 iris_print_program_cache(struct iris_context *ice)
 360 {
 361    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 362    const struct gen_device_info *devinfo = &screen->devinfo;
 363
 364    hash_table_foreach(ice->shaders.cache, entry) {
 365       const struct keybox *keybox = entry->key;
 366       struct iris_compiled_shader *shader = entry->data;
 367       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 368       brw_disassemble(devinfo, shader->map, 0,
 369                       shader->prog_data->program_size, stderr);
 370    }
 371 }