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
 101    // XXX: ugly...
 102    // XXX: move this flagging out to a higher level, allow comparison of
 103    // XXX: new and old programs to decide what bits to twiddle
 104    // XXX: CLIP: toggle if barycentric modes has any NONPERSPECTIVE or not
 105    if (cache_id == IRIS_CACHE_FS)
 106       return IRIS_DIRTY_WM | IRIS_DIRTY_FS | IRIS_DIRTY_CLIP;
 107    if (cache_id == IRIS_CACHE_VS)
 108       return IRIS_DIRTY_VS | IRIS_DIRTY_VF_SGVS;
 109
 110    return IRIS_DIRTY_VS << cache_id;
 111 }
 112
 113 static unsigned
 114 get_program_string_id(enum iris_program_cache_id cache_id, const void *key)
 115 {
 116    switch (cache_id) {
 117    case IRIS_CACHE_VS:
 118       return ((struct brw_vs_prog_key *) key)->program_string_id;
 119    case IRIS_CACHE_TCS:
 120       return ((struct brw_tcs_prog_key *) key)->program_string_id;
 121    case IRIS_CACHE_TES:
 122       return ((struct brw_tes_prog_key *) key)->program_string_id;
 123    case IRIS_CACHE_GS:
 124       return ((struct brw_gs_prog_key *) key)->program_string_id;
 125    case IRIS_CACHE_CS:
 126       return ((struct brw_cs_prog_key *) key)->program_string_id;
 127    case IRIS_CACHE_FS:
 128       return ((struct brw_wm_prog_key *) key)->program_string_id;
 129    default:
 130       unreachable("no program string id for this kind of program");
 131    }
 132 }
 133
 134 static struct iris_compiled_shader *
 135 iris_find_cached_shader(struct iris_context *ice,
 136                         enum iris_program_cache_id cache_id,
 137                         const void *key,
 138                         uint32_t key_size)
 139 {
 140    struct keybox *keybox =
 141       make_keybox(ice->shaders.cache, cache_id, key, key_size);
 142    struct hash_entry *entry =
 143       _mesa_hash_table_search(ice->shaders.cache, keybox);
 144
 145    ralloc_free(keybox);
 146
 147    return entry ? entry->data : NULL;
 148 }
 149
 150 /**
 151  * Looks for a program in the cache and binds it.
 152  *
 153  * If no program was found, returns false and leaves the binding alone.
 154  */
 155 bool
 156 iris_bind_cached_shader(struct iris_context *ice,
 157                         enum iris_program_cache_id cache_id,
 158                         const void *key)
 159 {
 160    unsigned key_size = key_size_for_cache(cache_id);
 161    struct iris_compiled_shader *shader =
 162       iris_find_cached_shader(ice, cache_id, key, key_size);
 163
 164    if (!shader)
 165       return false;
 166
 167    if (memcmp(shader, ice->shaders.prog[cache_id], sizeof(*shader)) != 0) {
 168       ice->shaders.prog[cache_id] = shader;
 169       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 170    }
 171
 172    return true;
 173 }
 174
 175 const void *
 176 iris_find_previous_compile(const struct iris_context *ice,
 177                            enum iris_program_cache_id cache_id,
 178                            unsigned program_string_id)
 179 {
 180    hash_table_foreach(ice->shaders.cache, entry) {
 181       const struct keybox *keybox = entry->key;
 182       if (keybox->cache_id == cache_id &&
 183           get_program_string_id(cache_id, keybox->data) == program_string_id) {
 184          return keybox->data;
 185       }
 186    }
 187
 188    return NULL;
 189 }
 190
 191 /**
 192  * Look for an existing entry in the cache that has identical assembly code.
 193  *
 194  * This is useful for programs generating shaders at runtime, where multiple
 195  * distinct shaders (from an API perspective) may compile to the same assembly
 196  * in our backend.  This saves space in the program cache buffer.
 197  */
 198 static const struct iris_compiled_shader *
 199 find_existing_assembly(struct hash_table *cache,
 200                        const void *assembly,
 201                        unsigned assembly_size)
 202 {
 203    hash_table_foreach(cache, entry) {
 204       const struct iris_compiled_shader *existing = entry->data;
 205       if (existing->prog_data->program_size == assembly_size &&
 206           memcmp(existing->map, assembly, assembly_size) == 0)
 207          return existing;
 208    }
 209    return NULL;
 210 }
 211
 212 static struct iris_compiled_shader *
 213 iris_upload_shader(struct iris_context *ice,
 214                    enum iris_program_cache_id cache_id,
 215                    uint32_t key_size,
 216                    const void *key,
 217                    const void *assembly,
 218                    struct brw_stage_prog_data *prog_data,
 219                    uint32_t *streamout)
 220 {
 221    struct iris_screen *screen = (void *) ice->ctx.screen;
 222    struct gen_device_info *devinfo = &screen->devinfo;
 223    struct hash_table *cache = ice->shaders.cache;
 224    struct iris_compiled_shader *shader =
 225       rzalloc_size(cache, sizeof(struct iris_compiled_shader) +
 226                    ice->vtbl.derived_program_state_size(cache_id));
 227    const struct iris_compiled_shader *existing =
 228       find_existing_assembly(cache, assembly, prog_data->program_size);
 229
 230    /* If we can find a matching prog in the cache already, then reuse the
 231     * existing stuff without creating new copy into the underlying buffer
 232     * object.  This is notably useful for programs generating shaders at
 233     * runtime, where multiple shaders may compile to the same thing in our
 234     * backend.
 235     */
 236    if (existing) {
 237       pipe_resource_reference(&shader->assembly.res, existing->assembly.res);
 238       shader->assembly.offset = existing->assembly.offset;
 239       shader->map = existing->map;
 240    } else {
 241       shader->assembly.res = NULL;
 242       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size, 64,
 243                      &shader->assembly.offset, &shader->assembly.res,
 244                      &shader->map);
 245       memcpy(shader->map, assembly, prog_data->program_size);
 246    }
 247
 248    shader->prog_data = prog_data;
 249    shader->streamout = streamout;
 250
 251    ralloc_steal(shader, shader->prog_data);
 252    ralloc_steal(shader->prog_data, prog_data->param);
 253    ralloc_steal(shader->prog_data, prog_data->pull_param);
 254    ralloc_steal(shader, shader->streamout);
 255
 256    /* Store the 3DSTATE shader packets and other derived state. */
 257    ice->vtbl.store_derived_program_state(devinfo, cache_id, shader);
 258
 259    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 260    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 261
 262    return shader;
 263 }
 264
 265 /**
 266  * Upload a new shader to the program cache, and bind it for use.
 267  *
 268  * \param prog_data must be ralloc'd and will be stolen.
 269  */
 270 void
 271 iris_upload_and_bind_shader(struct iris_context *ice,
 272                             enum iris_program_cache_id cache_id,
 273                             const void *key,
 274                             const void *assembly,
 275                             struct brw_stage_prog_data *prog_data,
 276                             uint32_t *streamout)
 277 {
 278    assert(cache_id != IRIS_CACHE_BLORP);
 279
 280    struct iris_compiled_shader *shader =
 281       iris_upload_shader(ice, cache_id, key_size_for_cache(cache_id), key,
 282                          assembly, prog_data, streamout);
 283
 284    ice->shaders.prog[cache_id] = shader;
 285    ice->state.dirty |= dirty_flag_for_cache(cache_id);
 286 }
 287
 288 bool
 289 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 290                          const void *key, uint32_t key_size,
 291                          uint32_t *kernel_out, void *prog_data_out)
 292 {
 293    struct blorp_context *blorp = blorp_batch->blorp;
 294    struct iris_context *ice = blorp->driver_ctx;
 295    struct iris_batch *batch = blorp_batch->driver_batch;
 296    struct iris_compiled_shader *shader =
 297       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key, key_size);
 298
 299    if (!shader)
 300       return false;
 301
 302    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 303    *kernel_out =
 304       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 305    *((void **) prog_data_out) = shader->prog_data;
 306
 307    iris_use_pinned_bo(batch, bo, false);
 308
 309    return true;
 310 }
 311
 312 bool
 313 iris_blorp_upload_shader(struct blorp_batch *blorp_batch,
 314                          const void *key, uint32_t key_size,
 315                          const void *kernel, UNUSED uint32_t kernel_size,
 316                          const struct brw_stage_prog_data *prog_data_templ,
 317                          UNUSED uint32_t prog_data_size,
 318                          uint32_t *kernel_out, void *prog_data_out)
 319 {
 320    struct blorp_context *blorp = blorp_batch->blorp;
 321    struct iris_context *ice = blorp->driver_ctx;
 322    struct iris_batch *batch = blorp_batch->driver_batch;
 323
 324    void *prog_data = ralloc_size(NULL, prog_data_size);
 325    memcpy(prog_data, prog_data_templ, prog_data_size);
 326
 327    struct iris_compiled_shader *shader =
 328       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 329                          prog_data, NULL);
 330
 331    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 332    *kernel_out =
 333       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 334    *((void **) prog_data_out) = shader->prog_data;
 335
 336    iris_use_pinned_bo(batch, bo, false);
 337
 338    return true;
 339 }
 340
 341 void
 342 iris_init_program_cache(struct iris_context *ice)
 343 {
 344    ice->shaders.cache =
 345       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 346
 347    ice->shaders.uploader =
 348       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 349                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 350 }
 351
 352 void
 353 iris_destroy_program_cache(struct iris_context *ice)
 354 {
 355    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 356       ice->shaders.prog[i] = NULL;
 357    }
 358
 359    hash_table_foreach(ice->shaders.cache, entry) {
 360       struct iris_compiled_shader *shader = entry->data;
 361       pipe_resource_reference(&shader->assembly.res, NULL);
 362    }
 363
 364    u_upload_destroy(ice->shaders.uploader);
 365
 366    ralloc_free(ice->shaders.cache);
 367 }
 368
 369 static const char *
 370 cache_name(enum iris_program_cache_id cache_id)
 371 {
 372    if (cache_id == IRIS_CACHE_BLORP)
 373       return "BLORP";
 374
 375    return _mesa_shader_stage_to_string(cache_id);
 376 }
 377
 378 void
 379 iris_print_program_cache(struct iris_context *ice)
 380 {
 381    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 382    const struct gen_device_info *devinfo = &screen->devinfo;
 383
 384    hash_table_foreach(ice->shaders.cache, entry) {
 385       const struct keybox *keybox = entry->key;
 386       struct iris_compiled_shader *shader = entry->data;
 387       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 388       brw_disassemble(devinfo, shader->map, 0,
 389                       shader->prog_data->program_size, stderr);
 390    }
 391 }