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