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  * on the rights to use, copy, modify, merge, publish, distribute, sub
   8  * license, and/or sell copies of the Software, and to permit persons to whom
   9  * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
  18  * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
  19  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
  20  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  21  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  22  */
  23 #include <stdio.h>
  24 #include <errno.h>
  25 #include <i915_drm.h>
  26 #include "pipe/p_defines.h"
  27 #include "pipe/p_state.h"
  28 #include "pipe/p_context.h"
  29 #include "pipe/p_screen.h"
  30 #include "util/u_atomic.h"
  31 #include "compiler/nir/nir.h"
  32 #include "compiler/nir/nir_builder.h"
  33 #include "intel/compiler/brw_compiler.h"
  34 #include "intel/compiler/brw_eu.h"
  35 #include "intel/compiler/brw_nir.h"
  36 #include "iris_context.h"
  37
  38 struct keybox {
  39    uint8_t size;
  40    enum iris_program_cache_id cache_id;
  41    uint8_t data[0];
  42 };
  43
  44 static struct keybox *
  45 make_keybox(struct iris_program_cache *cache,
  46             enum iris_program_cache_id cache_id,
  47             const void *key)
  48 {
  49    static const unsigned key_sizes[] = {
  50       [IRIS_CACHE_VS]         = sizeof(struct brw_vs_prog_key),
  51       [IRIS_CACHE_TCS]        = sizeof(struct brw_tcs_prog_key),
  52       [IRIS_CACHE_TES]        = sizeof(struct brw_tes_prog_key),
  53       [IRIS_CACHE_GS]         = sizeof(struct brw_gs_prog_key),
  54       [IRIS_CACHE_FS]         = sizeof(struct brw_wm_prog_key),
  55       [IRIS_CACHE_CS]         = sizeof(struct brw_cs_prog_key),
  56       //[IRIS_CACHE_BLORP_BLIT] = sizeof(struct brw_blorp_blit_prog_key),
  57    };
  58
  59    struct keybox *keybox =
  60       ralloc_size(cache->table, sizeof(struct keybox) + key_sizes[cache_id]);
  61
  62    keybox->cache_id = cache_id;
  63    keybox->size = key_sizes[cache_id];
  64    memcpy(keybox->data, key, key_sizes[cache_id]);
  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 static uint64_t
  87 dirty_flag_for_cache(enum iris_program_cache_id cache_id)
  88 {
  89    assert(cache_id <= MESA_SHADER_STAGES);
  90    return IRIS_DIRTY_VS << cache_id;
  91 }
  92
  93 static unsigned
  94 get_program_string_id(enum iris_program_cache_id cache_id, const void *key)
  95 {
  96    switch (cache_id) {
  97    case IRIS_CACHE_VS:
  98       return ((struct brw_vs_prog_key *) key)->program_string_id;
  99    case IRIS_CACHE_TCS:
 100       return ((struct brw_tcs_prog_key *) key)->program_string_id;
 101    case IRIS_CACHE_TES:
 102       return ((struct brw_tes_prog_key *) key)->program_string_id;
 103    case IRIS_CACHE_GS:
 104       return ((struct brw_gs_prog_key *) key)->program_string_id;
 105    case IRIS_CACHE_CS:
 106       return ((struct brw_cs_prog_key *) key)->program_string_id;
 107    case IRIS_CACHE_FS:
 108       return ((struct brw_wm_prog_key *) key)->program_string_id;
 109    default:
 110       unreachable("no program string id for this kind of program");
 111    }
 112 }
 113
 114 /**
 115  * Looks for a program in the cache and binds it.
 116  *
 117  * If no program was found, returns false and leaves the binding alone.
 118  */
 119 bool
 120 iris_bind_cached_shader(struct iris_context *ice,
 121                         enum iris_program_cache_id cache_id,
 122                         const void *key)
 123 {
 124    struct iris_program_cache *cache = &ice->shaders.cache;
 125
 126    struct keybox *keybox = make_keybox(cache, cache_id, key);
 127
 128    struct hash_entry *entry = _mesa_hash_table_search(cache->table, keybox);
 129
 130    if (!entry)
 131       return false;
 132
 133    struct iris_compiled_shader *shader = entry->data;
 134
 135    if (cache_id <= MESA_SHADER_STAGES &&
 136        memcmp(shader, ice->shaders.prog[cache_id], sizeof(*shader)) != 0) {
 137       ice->shaders.prog[cache_id] = shader;
 138       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 139    }
 140
 141    return true;
 142 }
 143
 144 static void
 145 recreate_cache_bo(struct iris_context *ice, uint32_t size)
 146 {
 147    struct iris_program_cache *cache = &ice->shaders.cache;
 148    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 149    struct iris_bo *old_bo = cache->bo;
 150    void *old_map = cache->map;
 151
 152    cache->bo = iris_bo_alloc(screen->bufmgr, "program cache", size, 64);
 153    cache->bo->kflags = EXEC_OBJECT_CAPTURE;
 154    cache->map = iris_bo_map(&ice->dbg, cache->bo,
 155                             MAP_READ | MAP_WRITE | MAP_ASYNC | MAP_PERSISTENT);
 156
 157    /* Copy any existing data that needs to be saved. */
 158    if (old_bo) {
 159       perf_debug(&ice->dbg,
 160                  "Copying to larger program cache: %u kB -> %u kB\n",
 161                  (unsigned) old_bo->size / 1024,
 162                  (unsigned) cache->bo->size / 1024);
 163
 164       memcpy(cache->map, old_map, cache->next_offset);
 165
 166       iris_bo_unreference(old_bo);
 167       iris_bo_unmap(old_bo);
 168    }
 169
 170    ice->state.dirty |= IRIS_DIRTY_STATE_BASE_ADDRESS;
 171 }
 172
 173 const void *
 174 iris_find_previous_compile(struct iris_program_cache *cache,
 175                            enum iris_program_cache_id cache_id,
 176                            unsigned program_string_id)
 177 {
 178    hash_table_foreach(cache->table, entry) {
 179       const struct keybox *keybox = entry->key;
 180       if (keybox->cache_id == cache_id &&
 181           get_program_string_id(cache_id, keybox->data) == program_string_id) {
 182          return keybox->data;
 183       }
 184    }
 185
 186    return NULL;
 187 }
 188
 189 /**
 190  * Look for an existing entry in the cache that has identical assembly code.
 191  *
 192  * This is useful for programs generating shaders at runtime, where multiple
 193  * distinct shaders (from an API perspective) may compile to the same assembly
 194  * in our backend.  This saves space in the program cache buffer.
 195  */
 196 static const struct iris_compiled_shader *
 197 find_existing_assembly(const struct iris_program_cache *cache,
 198                        const void *assembly,
 199                        unsigned assembly_size)
 200 {
 201    hash_table_foreach(cache->table, entry) {
 202       const struct iris_compiled_shader *existing = entry->data;
 203       if (existing->prog_data->program_size == assembly_size &&
 204           memcmp(cache->map + existing->prog_offset,
 205                  assembly, assembly_size) == 0)
 206          return existing;
 207    }
 208    return NULL;
 209 }
 210
 211 static uint32_t
 212 upload_new_assembly(struct iris_context *ice,
 213                     const void *assembly,
 214                     unsigned size)
 215 {
 216    struct iris_program_cache *cache = &ice->shaders.cache;
 217
 218    /* Allocate space in the cache BO for our new program. */
 219    if (cache->next_offset + size > cache->bo->size) {
 220       uint32_t new_size = cache->bo->size * 2;
 221
 222       while (cache->next_offset + size > new_size)
 223          new_size *= 2;
 224
 225       recreate_cache_bo(ice, new_size);
 226    }
 227
 228    uint32_t offset = cache->next_offset;
 229
 230    /* Programs are always 64-byte aligned, so set up the next one now */
 231    cache->next_offset = ALIGN(offset + size, 64);
 232
 233    /* Copy data to the buffer */
 234    memcpy(cache->map + offset, assembly, size);
 235
 236    return offset;
 237 }
 238
 239 /**
 240  * Upload a new shader to the program cache, and bind it for use.
 241  *
 242  * \param prog_data must be ralloc'd and will be stolen.
 243  */
 244 void
 245 iris_upload_and_bind_shader(struct iris_context *ice,
 246                             enum iris_program_cache_id cache_id,
 247                             const void *key,
 248                             const void *assembly,
 249                             const struct brw_stage_prog_data *prog_data)
 250 {
 251    struct iris_screen *screen = (void *) ice->ctx.screen;
 252    struct gen_device_info *devinfo = &screen->devinfo;
 253    struct iris_program_cache *cache = &ice->shaders.cache;
 254    struct iris_compiled_shader *shader =
 255       ralloc_size(cache->table, sizeof(struct iris_compiled_shader) +
 256                   iris_derived_program_state_size(cache_id));
 257    const struct iris_compiled_shader *existing =
 258       find_existing_assembly(cache, assembly, prog_data->program_size);
 259
 260    /* If we can find a matching prog in the cache already, then reuse the
 261     * existing stuff without creating new copy into the underlying buffer
 262     * object.  This is notably useful for programs generating shaders at
 263     * runtime, where multiple shaders may compile to the same thing in our
 264     * backend.
 265     */
 266    if (existing) {
 267       shader->prog_offset = existing->prog_offset;
 268    } else {
 269       shader->prog_offset =
 270          upload_new_assembly(ice, assembly, prog_data->program_size);
 271    }
 272
 273    shader->prog_data = prog_data;
 274
 275    ralloc_steal(shader, shader->prog_data);
 276    ralloc_steal(shader->prog_data, prog_data->param);
 277    ralloc_steal(shader->prog_data, prog_data->pull_param);
 278
 279    /* Store the 3DSTATE shader packets and other derived state. */
 280    iris_set_derived_program_state(devinfo, cache_id, shader);
 281
 282    struct keybox *keybox = make_keybox(cache, cache_id, key);
 283    _mesa_hash_table_insert(cache->table, keybox, shader);
 284
 285    if (cache_id <= MESA_SHADER_STAGES) {
 286       ice->shaders.prog[cache_id] = shader;
 287       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 288    }
 289 }
 290
 291 void
 292 iris_init_program_cache(struct iris_context *ice)
 293 {
 294    struct iris_program_cache *cache = &ice->shaders.cache;
 295
 296    cache->table = _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 297
 298    recreate_cache_bo(ice, 16384);
 299 }
 300
 301 void
 302 iris_destroy_program_cache(struct iris_context *ice)
 303 {
 304    struct iris_program_cache *cache = &ice->shaders.cache;
 305
 306    /* This can be NULL if context creation failed early on. */
 307    if (cache->bo) {
 308       iris_bo_unmap(cache->bo);
 309       iris_bo_unreference(cache->bo);
 310       cache->bo = NULL;
 311       cache->map = NULL;
 312    }
 313
 314    cache->next_offset = 0;
 315
 316    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 317       ice->shaders.prog[i] = NULL;
 318    }
 319
 320    ralloc_free(cache->table);
 321 }
 322
 323 static const char *
 324 cache_name(enum iris_program_cache_id cache_id)
 325 {
 326    if (cache_id == IRIS_CACHE_BLORP_BLIT)
 327       return "BLORP";
 328
 329    return _mesa_shader_stage_to_string(cache_id);
 330 }
 331
 332 void
 333 iris_print_program_cache(struct iris_context *ice)
 334 {
 335    struct iris_program_cache *cache = &ice->shaders.cache;
 336    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 337    const struct gen_device_info *devinfo = &screen->devinfo;
 338
 339    hash_table_foreach(cache->table, entry) {
 340       const struct keybox *keybox = entry->key;
 341       struct iris_compiled_shader *shader = entry->data;
 342       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 343       brw_disassemble(devinfo, cache->map, shader->prog_offset,
 344                       shader->prog_data->program_size, stderr);
 345    }
 346 }