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  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice shall be included
  12  * in all copies or substantial portions of the Software.
  13  *
  14  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  15  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  17  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  18  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  19  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  20  * DEALINGS IN THE SOFTWARE.
  21  */
  22
  23 /**
  24  * @file iris_program_cache.c
  25  *
  26  * The in-memory program cache.  This is basically a hash table mapping
  27  * API-specified shaders and a state key to a compiled variant.  It also
  28  * takes care of uploading shader assembly into a BO for use on the GPU.
  29  */
  30
  31 #include <stdio.h>
  32 #include <errno.h>
  33 #include "pipe/p_defines.h"
  34 #include "pipe/p_state.h"
  35 #include "pipe/p_context.h"
  36 #include "pipe/p_screen.h"
  37 #include "util/u_atomic.h"
  38 #include "util/u_upload_mgr.h"
  39 #include "compiler/nir/nir.h"
  40 #include "compiler/nir/nir_builder.h"
  41 #include "intel/compiler/brw_compiler.h"
  42 #include "intel/compiler/brw_eu.h"
  43 #include "intel/compiler/brw_nir.h"
  44 #include "iris_context.h"
  45 #include "iris_resource.h"
  46
  47 struct keybox {
  48    uint16_t size;
  49    enum iris_program_cache_id cache_id;
  50    uint8_t data[0];
  51 };
  52
  53 static struct keybox *
  54 make_keybox(void *mem_ctx,
  55             enum iris_program_cache_id cache_id,
  56             const void *key,
  57             uint32_t key_size)
  58 {
  59    struct keybox *keybox =
  60       ralloc_size(mem_ctx, sizeof(struct keybox) + key_size);
  61
  62    keybox->cache_id = cache_id;
  63    keybox->size = key_size;
  64    memcpy(keybox->data, key, key_size);
  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 struct iris_compiled_shader *
  87 iris_find_cached_shader(struct iris_context *ice,
  88                         enum iris_program_cache_id cache_id,
  89                         uint32_t key_size,
  90                         const void *key)
  91 {
  92    struct keybox *keybox =
  93       make_keybox(ice->shaders.cache, cache_id, key, key_size);
  94    struct hash_entry *entry =
  95       _mesa_hash_table_search(ice->shaders.cache, keybox);
  96
  97    ralloc_free(keybox);
  98
  99    return entry ? entry->data : NULL;
 100 }
 101
 102 const void *
 103 iris_find_previous_compile(const struct iris_context *ice,
 104                            enum iris_program_cache_id cache_id,
 105                            unsigned program_string_id)
 106 {
 107    hash_table_foreach(ice->shaders.cache, entry) {
 108       const struct keybox *keybox = entry->key;
 109       const struct brw_base_prog_key *key = (const void *)keybox->data;
 110       if (keybox->cache_id == cache_id &&
 111           key->program_string_id == program_string_id) {
 112          return keybox->data;
 113       }
 114    }
 115
 116    return NULL;
 117 }
 118
 119 /**
 120  * Look for an existing entry in the cache that has identical assembly code.
 121  *
 122  * This is useful for programs generating shaders at runtime, where multiple
 123  * distinct shaders (from an API perspective) may compile to the same assembly
 124  * in our backend.  This saves space in the program cache buffer.
 125  */
 126 static const struct iris_compiled_shader *
 127 find_existing_assembly(struct hash_table *cache,
 128                        const void *assembly,
 129                        unsigned assembly_size)
 130 {
 131    hash_table_foreach(cache, entry) {
 132       const struct iris_compiled_shader *existing = entry->data;
 133       if (existing->prog_data->program_size == assembly_size &&
 134           memcmp(existing->map, assembly, assembly_size) == 0)
 135          return existing;
 136    }
 137    return NULL;
 138 }
 139
 140 struct iris_compiled_shader *
 141 iris_upload_shader(struct iris_context *ice,
 142                    enum iris_program_cache_id cache_id,
 143                    uint32_t key_size,
 144                    const void *key,
 145                    const void *assembly,
 146                    struct brw_stage_prog_data *prog_data,
 147                    uint32_t *streamout,
 148                    enum brw_param_builtin *system_values,
 149                    unsigned num_system_values,
 150                    unsigned num_cbufs,
 151                    const struct iris_binding_table *bt)
 152 {
 153    struct hash_table *cache = ice->shaders.cache;
 154    struct iris_compiled_shader *shader =
 155       rzalloc_size(cache, sizeof(struct iris_compiled_shader) +
 156                    ice->vtbl.derived_program_state_size(cache_id));
 157    const struct iris_compiled_shader *existing =
 158       find_existing_assembly(cache, assembly, prog_data->program_size);
 159
 160    /* If we can find a matching prog in the cache already, then reuse the
 161     * existing stuff without creating new copy into the underlying buffer
 162     * object.  This is notably useful for programs generating shaders at
 163     * runtime, where multiple shaders may compile to the same thing in our
 164     * backend.
 165     */
 166    if (existing) {
 167       pipe_resource_reference(&shader->assembly.res, existing->assembly.res);
 168       shader->assembly.offset = existing->assembly.offset;
 169       shader->map = existing->map;
 170    } else {
 171       shader->assembly.res = NULL;
 172       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size, 64,
 173                      &shader->assembly.offset, &shader->assembly.res,
 174                      &shader->map);
 175       memcpy(shader->map, assembly, prog_data->program_size);
 176    }
 177
 178    shader->prog_data = prog_data;
 179    shader->streamout = streamout;
 180    shader->system_values = system_values;
 181    shader->num_system_values = num_system_values;
 182    shader->num_cbufs = num_cbufs;
 183    shader->bt = *bt;
 184
 185    ralloc_steal(shader, shader->prog_data);
 186    ralloc_steal(shader->prog_data, prog_data->param);
 187    ralloc_steal(shader->prog_data, prog_data->pull_param);
 188    ralloc_steal(shader, shader->streamout);
 189    ralloc_steal(shader, shader->system_values);
 190
 191    /* Store the 3DSTATE shader packets and other derived state. */
 192    ice->vtbl.store_derived_program_state(ice, cache_id, shader);
 193
 194    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 195    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 196
 197    return shader;
 198 }
 199
 200 bool
 201 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 202                          const void *key, uint32_t key_size,
 203                          uint32_t *kernel_out, void *prog_data_out)
 204 {
 205    struct blorp_context *blorp = blorp_batch->blorp;
 206    struct iris_context *ice = blorp->driver_ctx;
 207    struct iris_batch *batch = blorp_batch->driver_batch;
 208    struct iris_compiled_shader *shader =
 209       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key_size, key);
 210
 211    if (!shader)
 212       return false;
 213
 214    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 215    *kernel_out =
 216       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 217    *((void **) prog_data_out) = shader->prog_data;
 218
 219    iris_use_pinned_bo(batch, bo, false);
 220
 221    return true;
 222 }
 223
 224 bool
 225 iris_blorp_upload_shader(struct blorp_batch *blorp_batch,
 226                          const void *key, uint32_t key_size,
 227                          const void *kernel, UNUSED uint32_t kernel_size,
 228                          const struct brw_stage_prog_data *prog_data_templ,
 229                          UNUSED uint32_t prog_data_size,
 230                          uint32_t *kernel_out, void *prog_data_out)
 231 {
 232    struct blorp_context *blorp = blorp_batch->blorp;
 233    struct iris_context *ice = blorp->driver_ctx;
 234    struct iris_batch *batch = blorp_batch->driver_batch;
 235
 236    void *prog_data = ralloc_size(NULL, prog_data_size);
 237    memcpy(prog_data, prog_data_templ, prog_data_size);
 238
 239    struct iris_binding_table bt;
 240    memset(&bt, 0, sizeof(bt));
 241
 242    struct iris_compiled_shader *shader =
 243       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 244                          prog_data, NULL, NULL, 0, 0, &bt);
 245
 246    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 247    *kernel_out =
 248       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 249    *((void **) prog_data_out) = shader->prog_data;
 250
 251    iris_use_pinned_bo(batch, bo, false);
 252
 253    return true;
 254 }
 255
 256 void
 257 iris_init_program_cache(struct iris_context *ice)
 258 {
 259    ice->shaders.cache =
 260       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 261
 262    ice->shaders.uploader =
 263       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 264                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 265 }
 266
 267 void
 268 iris_destroy_program_cache(struct iris_context *ice)
 269 {
 270    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 271       ice->shaders.prog[i] = NULL;
 272    }
 273
 274    hash_table_foreach(ice->shaders.cache, entry) {
 275       struct iris_compiled_shader *shader = entry->data;
 276       pipe_resource_reference(&shader->assembly.res, NULL);
 277    }
 278
 279    u_upload_destroy(ice->shaders.uploader);
 280
 281    ralloc_free(ice->shaders.cache);
 282 }
 283
 284 static const char *
 285 cache_name(enum iris_program_cache_id cache_id)
 286 {
 287    if (cache_id == IRIS_CACHE_BLORP)
 288       return "BLORP";
 289
 290    return _mesa_shader_stage_to_string(cache_id);
 291 }
 292
 293 void
 294 iris_print_program_cache(struct iris_context *ice)
 295 {
 296    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 297    const struct gen_device_info *devinfo = &screen->devinfo;
 298
 299    hash_table_foreach(ice->shaders.cache, entry) {
 300       const struct keybox *keybox = entry->key;
 301       struct iris_compiled_shader *shader = entry->data;
 302       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 303       brw_disassemble(devinfo, shader->map, 0,
 304                       shader->prog_data->program_size, stderr);
 305    }
 306 }