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    ralloc_free(keybox);
 142
 143    return entry ? entry->data : NULL;
 144 }
 145
 146 /**
 147  * Looks for a program in the cache and binds it.
 148  *
 149  * If no program was found, returns false and leaves the binding alone.
 150  */
 151 bool
 152 iris_bind_cached_shader(struct iris_context *ice,
 153                         enum iris_program_cache_id cache_id,
 154                         const void *key)
 155 {
 156    unsigned key_size = key_size_for_cache(cache_id);
 157    struct iris_compiled_shader *shader =
 158       iris_find_cached_shader(ice, cache_id, key, key_size);
 159
 160    if (!shader)
 161       return false;
 162
 163    if (memcmp(shader, ice->shaders.prog[cache_id], sizeof(*shader)) != 0) {
 164       ice->shaders.prog[cache_id] = shader;
 165       ice->state.dirty |= dirty_flag_for_cache(cache_id);
 166    }
 167
 168    return true;
 169 }
 170
 171 const void *
 172 iris_find_previous_compile(const struct iris_context *ice,
 173                            enum iris_program_cache_id cache_id,
 174                            unsigned program_string_id)
 175 {
 176    hash_table_foreach(ice->shaders.cache, entry) {
 177       const struct keybox *keybox = entry->key;
 178       if (keybox->cache_id == cache_id &&
 179           get_program_string_id(cache_id, keybox->data) == program_string_id) {
 180          return keybox->data;
 181       }
 182    }
 183
 184    return NULL;
 185 }
 186
 187 /**
 188  * Look for an existing entry in the cache that has identical assembly code.
 189  *
 190  * This is useful for programs generating shaders at runtime, where multiple
 191  * distinct shaders (from an API perspective) may compile to the same assembly
 192  * in our backend.  This saves space in the program cache buffer.
 193  */
 194 static const struct iris_compiled_shader *
 195 find_existing_assembly(struct hash_table *cache,
 196                        const void *assembly,
 197                        unsigned assembly_size)
 198 {
 199    hash_table_foreach(cache, entry) {
 200       const struct iris_compiled_shader *existing = entry->data;
 201       if (existing->prog_data->program_size == assembly_size &&
 202           memcmp(existing->map, assembly, assembly_size) == 0)
 203          return existing;
 204    }
 205    return NULL;
 206 }
 207
 208 static struct iris_compiled_shader *
 209 iris_upload_shader(struct iris_context *ice,
 210                    enum iris_program_cache_id cache_id,
 211                    uint32_t key_size,
 212                    const void *key,
 213                    const void *assembly,
 214                    struct brw_stage_prog_data *prog_data)
 215 {
 216    struct iris_screen *screen = (void *) ice->ctx.screen;
 217    struct gen_device_info *devinfo = &screen->devinfo;
 218    struct hash_table *cache = ice->shaders.cache;
 219    struct iris_compiled_shader *shader =
 220       rzalloc_size(cache, sizeof(struct iris_compiled_shader) +
 221                    ice->vtbl.derived_program_state_size(cache_id));
 222    const struct iris_compiled_shader *existing =
 223       find_existing_assembly(cache, assembly, prog_data->program_size);
 224
 225    /* If we can find a matching prog in the cache already, then reuse the
 226     * existing stuff without creating new copy into the underlying buffer
 227     * object.  This is notably useful for programs generating shaders at
 228     * runtime, where multiple shaders may compile to the same thing in our
 229     * backend.
 230     */
 231    if (existing) {
 232       pipe_resource_reference(&shader->buffer, existing->buffer);
 233       shader->offset = existing->offset;
 234       shader->map = existing->map;
 235    } else {
 236       shader->buffer = NULL;
 237       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size,
 238                      64, &shader->offset, &shader->buffer, &shader->map);
 239       memcpy(shader->map, assembly, prog_data->program_size);
 240    }
 241
 242    shader->prog_data = prog_data;
 243
 244    ralloc_steal(shader, shader->prog_data);
 245    ralloc_steal(shader->prog_data, prog_data->param);
 246    ralloc_steal(shader->prog_data, prog_data->pull_param);
 247
 248    /* Store the 3DSTATE shader packets and other derived state. */
 249    ice->vtbl.store_derived_program_state(devinfo, cache_id, shader);
 250
 251    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 252    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 253
 254    return shader;
 255 }
 256
 257 /**
 258  * Upload a new shader to the program cache, and bind it for use.
 259  *
 260  * \param prog_data must be ralloc'd and will be stolen.
 261  */
 262 void
 263 iris_upload_and_bind_shader(struct iris_context *ice,
 264                             enum iris_program_cache_id cache_id,
 265                             const void *key,
 266                             const void *assembly,
 267                             struct brw_stage_prog_data *prog_data)
 268 {
 269    assert(cache_id != IRIS_CACHE_BLORP);
 270
 271    struct iris_compiled_shader *shader =
 272       iris_upload_shader(ice, cache_id, key_size_for_cache(cache_id), key,
 273                          assembly, prog_data);
 274
 275    ice->shaders.prog[cache_id] = shader;
 276    ice->state.dirty |= dirty_flag_for_cache(cache_id);
 277 }
 278
 279 bool
 280 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 281                          const void *key, uint32_t key_size,
 282                          uint32_t *kernel_out, void *prog_data_out)
 283 {
 284    struct blorp_context *blorp = blorp_batch->blorp;
 285    struct iris_context *ice = blorp->driver_ctx;
 286    struct iris_batch *batch = blorp_batch->driver_batch;
 287    struct iris_compiled_shader *shader =
 288       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key, key_size);
 289
 290    if (!shader)
 291       return false;
 292
 293    struct iris_bo *bo = iris_resource_bo(shader->buffer);
 294    *kernel_out = iris_bo_offset_from_base_address(bo) + shader->offset;
 295    *((void **) prog_data_out) = shader->prog_data;
 296
 297    iris_use_pinned_bo(batch, bo, false);
 298
 299    return true;
 300 }
 301
 302 bool
 303 iris_blorp_upload_shader(struct blorp_batch *blorp_batch,
 304                          const void *key, uint32_t key_size,
 305                          const void *kernel, UNUSED uint32_t kernel_size,
 306                          const struct brw_stage_prog_data *prog_data_templ,
 307                          UNUSED uint32_t prog_data_size,
 308                          uint32_t *kernel_out, void *prog_data_out)
 309 {
 310    struct blorp_context *blorp = blorp_batch->blorp;
 311    struct iris_context *ice = blorp->driver_ctx;
 312    struct iris_batch *batch = blorp_batch->driver_batch;
 313
 314    void *prog_data = ralloc_size(NULL, prog_data_size);
 315    memcpy(prog_data, prog_data_templ, prog_data_size);
 316
 317    struct iris_compiled_shader *shader =
 318       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 319                          prog_data);
 320
 321    struct iris_bo *bo = iris_resource_bo(shader->buffer);
 322    *kernel_out = iris_bo_offset_from_base_address(bo) + shader->offset;
 323    *((void **) prog_data_out) = shader->prog_data;
 324
 325    iris_use_pinned_bo(batch, bo, false);
 326
 327    return true;
 328 }
 329
 330 void
 331 iris_init_program_cache(struct iris_context *ice)
 332 {
 333    ice->shaders.cache =
 334       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 335
 336    ice->shaders.uploader =
 337       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 338                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 339 }
 340
 341 void
 342 iris_destroy_program_cache(struct iris_context *ice)
 343 {
 344    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 345       ice->shaders.prog[i] = NULL;
 346    }
 347
 348    hash_table_foreach(ice->shaders.cache, entry) {
 349       struct iris_compiled_shader *shader = entry->data;
 350       pipe_resource_reference(&shader->buffer, NULL);
 351    }
 352
 353    u_upload_destroy(ice->shaders.uploader);
 354
 355    ralloc_free(ice->shaders.cache);
 356 }
 357
 358 static const char *
 359 cache_name(enum iris_program_cache_id cache_id)
 360 {
 361    if (cache_id == IRIS_CACHE_BLORP)
 362       return "BLORP";
 363
 364    return _mesa_shader_stage_to_string(cache_id);
 365 }
 366
 367 void
 368 iris_print_program_cache(struct iris_context *ice)
 369 {
 370    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 371    const struct gen_device_info *devinfo = &screen->devinfo;
 372
 373    hash_table_foreach(ice->shaders.cache, entry) {
 374       const struct keybox *keybox = entry->key;
 375       struct iris_compiled_shader *shader = entry->data;
 376       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 377       brw_disassemble(devinfo, shader->map, 0,
 378                       shader->prog_data->program_size, stderr);
 379    }
 380 }