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 "pipe/p_defines.h"
  26 #include "pipe/p_state.h"
  27 #include "pipe/p_context.h"
  28 #include "pipe/p_screen.h"
  29 #include "util/u_atomic.h"
  30 #include "util/u_upload_mgr.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 #include "iris_resource.h"
  38
  39 struct keybox {
  40    uint8_t size;
  41    enum iris_program_cache_id cache_id;
  42    uint8_t data[0];
  43 };
  44
  45 static uint32_t
  46 key_size_for_cache(enum iris_program_cache_id cache_id)
  47 {
  48    static const unsigned key_sizes[] = {
  49       [IRIS_CACHE_VS]         = sizeof(struct brw_vs_prog_key),
  50       [IRIS_CACHE_TCS]        = sizeof(struct brw_tcs_prog_key),
  51       [IRIS_CACHE_TES]        = sizeof(struct brw_tes_prog_key),
  52       [IRIS_CACHE_GS]         = sizeof(struct brw_gs_prog_key),
  53       [IRIS_CACHE_FS]         = sizeof(struct brw_wm_prog_key),
  54       [IRIS_CACHE_CS]         = sizeof(struct brw_cs_prog_key),
  55    };
  56
  57    /* BLORP keys aren't all the same size. */
  58    assert(cache_id != IRIS_CACHE_BLORP);
  59
  60    return key_sizes[cache_id];
  61 }
  62
  63 static struct keybox *
  64 make_keybox(void *mem_ctx,
  65             enum iris_program_cache_id cache_id,
  66             const void *key,
  67             uint32_t key_size)
  68 {
  69    struct keybox *keybox =
  70       ralloc_size(mem_ctx, sizeof(struct keybox) + key_size);
  71
  72    keybox->cache_id = cache_id;
  73    keybox->size = key_size;
  74    memcpy(keybox->data, key, key_size);
  75
  76    return keybox;
  77 }
  78
  79 static uint32_t
  80 keybox_hash(const void *void_key)
  81 {
  82    const struct keybox *key = void_key;
  83    return _mesa_hash_data(&key->cache_id, key->size + sizeof(key->cache_id));
  84 }
  85
  86 static bool
  87 keybox_equals(const void *void_a, const void *void_b)
  88 {
  89    const struct keybox *a = void_a, *b = void_b;
  90    if (a->size != b->size)
  91       return false;
  92
  93    return memcmp(a->data, b->data, a->size) == 0;
  94 }
  95
  96 static uint64_t
  97 dirty_flag_for_cache(enum iris_program_cache_id cache_id)
  98 {
  99    assert(cache_id <= MESA_SHADER_STAGES);
 100    return IRIS_DIRTY_VS << cache_id;
 101 }
 102
 103 static unsigned
 104 get_program_string_id(enum iris_program_cache_id cache_id, const void *key)
 105 {
 106    switch (cache_id) {
 107    case IRIS_CACHE_VS:
 108       return ((struct brw_vs_prog_key *) key)->program_string_id;
 109    case IRIS_CACHE_TCS:
 110       return ((struct brw_tcs_prog_key *) key)->program_string_id;
 111    case IRIS_CACHE_TES:
 112       return ((struct brw_tes_prog_key *) key)->program_string_id;
 113    case IRIS_CACHE_GS:
 114       return ((struct brw_gs_prog_key *) key)->program_string_id;
 115    case IRIS_CACHE_CS:
 116       return ((struct brw_cs_prog_key *) key)->program_string_id;
 117    case IRIS_CACHE_FS:
 118       return ((struct brw_wm_prog_key *) key)->program_string_id;
 119    default:
 120       unreachable("no program string id for this kind of program");
 121    }
 122 }
 123
 124 static struct iris_compiled_shader *
 125 iris_find_cached_shader(struct iris_context *ice,
 126                         enum iris_program_cache_id cache_id,
 127                         const void *key,
 128                         uint32_t key_size)
 129 {
 130    struct keybox *keybox =
 131       make_keybox(ice->shaders.cache, cache_id, key, key_size);
 132    struct hash_entry *entry =
 133       _mesa_hash_table_search(ice->shaders.cache, keybox);
 134
 135    return entry ? entry->data : NULL;
 136 }
 137
 138 /**
 139  * Looks for a program in the cache and binds it.
 140  *
 141  * If no program was found, returns false and leaves the binding alone.
 142  */
 143 bool
 144 iris_bind_cached_shader(struct iris_context *ice,
 145                         enum iris_program_cache_id cache_id,
 146                         const void *key)
 147 {
 148    unsigned key_size = key_size_for_cache(cache_id);
 149    struct iris_compiled_shader *shader =
 150       iris_find_cached_shader(ice, cache_id, key, key_size);
 151
 152    if (!shader)
 153       return false;
 154
 155    if (memcmp(shader, ice->shaders.prog[cache_id], sizeof(*shader)) != 0) {
 156       ice->shaders.prog[cache_id] = shader;
 157       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 158    }
 159
 160    return true;
 161 }
 162
 163 const void *
 164 iris_find_previous_compile(const struct iris_context *ice,
 165                            enum iris_program_cache_id cache_id,
 166                            unsigned program_string_id)
 167 {
 168    hash_table_foreach(ice->shaders.cache, entry) {
 169       const struct keybox *keybox = entry->key;
 170       if (keybox->cache_id == cache_id &&
 171           get_program_string_id(cache_id, keybox->data) == program_string_id) {
 172          return keybox->data;
 173       }
 174    }
 175
 176    return NULL;
 177 }
 178
 179 /**
 180  * Look for an existing entry in the cache that has identical assembly code.
 181  *
 182  * This is useful for programs generating shaders at runtime, where multiple
 183  * distinct shaders (from an API perspective) may compile to the same assembly
 184  * in our backend.  This saves space in the program cache buffer.
 185  */
 186 static const struct iris_compiled_shader *
 187 find_existing_assembly(struct hash_table *cache,
 188                        const void *assembly,
 189                        unsigned assembly_size)
 190 {
 191    hash_table_foreach(cache, entry) {
 192       const struct iris_compiled_shader *existing = entry->data;
 193       if (existing->prog_data->program_size == assembly_size &&
 194           memcmp(existing->map, assembly, assembly_size) == 0)
 195          return existing;
 196    }
 197    return NULL;
 198 }
 199
 200 static struct iris_compiled_shader *
 201 iris_upload_shader(struct iris_context *ice,
 202                    enum iris_program_cache_id cache_id,
 203                    uint32_t key_size,
 204                    const void *key,
 205                    const void *assembly,
 206                    const struct brw_stage_prog_data *prog_data)
 207 {
 208    struct iris_screen *screen = (void *) ice->ctx.screen;
 209    struct gen_device_info *devinfo = &screen->devinfo;
 210    struct hash_table *cache = ice->shaders.cache;
 211    struct iris_compiled_shader *shader =
 212       ralloc_size(cache, sizeof(struct iris_compiled_shader) +
 213                   ice->vtbl.derived_program_state_size(cache_id));
 214    const struct iris_compiled_shader *existing =
 215       find_existing_assembly(cache, assembly, prog_data->program_size);
 216
 217    /* If we can find a matching prog in the cache already, then reuse the
 218     * existing stuff without creating new copy into the underlying buffer
 219     * object.  This is notably useful for programs generating shaders at
 220     * runtime, where multiple shaders may compile to the same thing in our
 221     * backend.
 222     */
 223    if (existing) {
 224       shader->offset = existing->offset;
 225    } else {
 226       shader->buffer = NULL;
 227       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size,
 228                      64, &shader->offset, &shader->buffer, &shader->map);
 229       memcpy(shader->map, assembly, prog_data->program_size);
 230    }
 231
 232    shader->prog_data = prog_data;
 233
 234    /* Store the 3DSTATE shader packets and other derived state. */
 235    ice->vtbl.set_derived_program_state(devinfo, cache_id, shader);
 236
 237    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 238    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 239
 240    return shader;
 241 }
 242
 243 /**
 244  * Upload a new shader to the program cache, and bind it for use.
 245  *
 246  * \param prog_data must be ralloc'd and will be stolen.
 247  */
 248 void
 249 iris_upload_and_bind_shader(struct iris_context *ice,
 250                             enum iris_program_cache_id cache_id,
 251                             const void *key,
 252                             const void *assembly,
 253                             struct brw_stage_prog_data *prog_data)
 254 {
 255    assert(cache_id != IRIS_CACHE_BLORP);
 256
 257    struct iris_compiled_shader *shader =
 258       iris_upload_shader(ice, cache_id, key_size_for_cache(cache_id), key,
 259                          assembly, prog_data);
 260
 261    ralloc_steal(shader, shader->prog_data);
 262    ralloc_steal(shader->prog_data, prog_data->param);
 263    ralloc_steal(shader->prog_data, prog_data->pull_param);
 264
 265    ice->shaders.prog[cache_id] = shader;
 266    ice->state.dirty |= dirty_flag_for_cache(cache_id);
 267 }
 268
 269 bool
 270 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 271                          const void *key, uint32_t key_size,
 272                          uint32_t *kernel_out, void *prog_data_out)
 273 {
 274    struct blorp_context *blorp = blorp_batch->blorp;
 275    struct iris_context *ice = blorp->driver_ctx;
 276    struct iris_batch *batch = blorp_batch->driver_batch;
 277    struct iris_compiled_shader *shader =
 278       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key, key_size);
 279
 280    if (!shader)
 281       return false;
 282
 283    struct iris_bo *bo = iris_resource_bo(shader->buffer);
 284    *kernel_out = iris_bo_offset_from_base_address(bo) + shader->offset;
 285    *((void **) prog_data_out) = shader->prog_data;
 286
 287    iris_use_pinned_bo(batch, bo, false);
 288
 289    return true;
 290 }
 291
 292 bool
 293 iris_blorp_upload_shader(struct blorp_batch *blorp_batch,
 294                          const void *key, uint32_t key_size,
 295                          const void *kernel, UNUSED uint32_t kernel_size,
 296                          const struct brw_stage_prog_data *prog_data,
 297                          UNUSED uint32_t prog_data_size,
 298                          uint32_t *kernel_out, void *prog_data_out)
 299 {
 300    struct blorp_context *blorp = blorp_batch->blorp;
 301    struct iris_context *ice = blorp->driver_ctx;
 302    struct iris_batch *batch = blorp_batch->driver_batch;
 303
 304    struct iris_compiled_shader *shader =
 305       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 306                          prog_data);
 307
 308    struct iris_bo *bo = iris_resource_bo(shader->buffer);
 309    *kernel_out = iris_bo_offset_from_base_address(bo) + shader->offset;
 310    *((void **) prog_data_out) = shader->prog_data;
 311
 312    iris_use_pinned_bo(batch, bo, false);
 313
 314    return true;
 315 }
 316
 317 void
 318 iris_init_program_cache(struct iris_context *ice)
 319 {
 320    ice->shaders.cache =
 321       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 322
 323    ice->shaders.uploader =
 324       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 325                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 326 }
 327
 328 void
 329 iris_destroy_program_cache(struct iris_context *ice)
 330 {
 331    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 332       ice->shaders.prog[i] = NULL;
 333    }
 334
 335    u_upload_destroy(ice->shaders.uploader);
 336
 337    ralloc_free(ice->shaders.cache);
 338 }
 339
 340 static const char *
 341 cache_name(enum iris_program_cache_id cache_id)
 342 {
 343    if (cache_id == IRIS_CACHE_BLORP)
 344       return "BLORP";
 345
 346    return _mesa_shader_stage_to_string(cache_id);
 347 }
 348
 349 void
 350 iris_print_program_cache(struct iris_context *ice)
 351 {
 352    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 353    const struct gen_device_info *devinfo = &screen->devinfo;
 354
 355    hash_table_foreach(ice->shaders.cache, entry) {
 356       const struct keybox *keybox = entry->key;
 357       struct iris_compiled_shader *shader = entry->data;
 358       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 359       brw_disassemble(devinfo, shader->map, 0,
 360                       shader->prog_data->program_size, stderr);
 361    }
 362 }