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 = make_keybox(NULL, cache_id, key, key_size);
  93    struct hash_entry *entry =
  94       _mesa_hash_table_search(ice->shaders.cache, keybox);
  95
  96    ralloc_free(keybox);
  97
  98    return entry ? entry->data : NULL;
  99 }
 100
 101 const void *
 102 iris_find_previous_compile(const struct iris_context *ice,
 103                            enum iris_program_cache_id cache_id,
 104                            unsigned program_string_id)
 105 {
 106    hash_table_foreach(ice->shaders.cache, entry) {
 107       const struct keybox *keybox = entry->key;
 108       const struct brw_base_prog_key *key = (const void *)keybox->data;
 109       if (keybox->cache_id == cache_id &&
 110           key->program_string_id == program_string_id) {
 111          return keybox->data;
 112       }
 113    }
 114
 115    return NULL;
 116 }
 117
 118 /**
 119  * Look for an existing entry in the cache that has identical assembly code.
 120  *
 121  * This is useful for programs generating shaders at runtime, where multiple
 122  * distinct shaders (from an API perspective) may compile to the same assembly
 123  * in our backend.  This saves space in the program cache buffer.
 124  */
 125 static const struct iris_compiled_shader *
 126 find_existing_assembly(struct hash_table *cache,
 127                        const void *assembly,
 128                        unsigned assembly_size)
 129 {
 130    hash_table_foreach(cache, entry) {
 131       const struct iris_compiled_shader *existing = entry->data;
 132       if (existing->prog_data->program_size == assembly_size &&
 133           memcmp(existing->map, assembly, assembly_size) == 0)
 134          return existing;
 135    }
 136    return NULL;
 137 }
 138
 139 struct iris_compiled_shader *
 140 iris_upload_shader(struct iris_context *ice,
 141                    enum iris_program_cache_id cache_id,
 142                    uint32_t key_size,
 143                    const void *key,
 144                    const void *assembly,
 145                    struct brw_stage_prog_data *prog_data,
 146                    uint32_t *streamout,
 147                    enum brw_param_builtin *system_values,
 148                    unsigned num_system_values,
 149                    unsigned num_cbufs,
 150                    const struct iris_binding_table *bt)
 151 {
 152    struct hash_table *cache = ice->shaders.cache;
 153    struct iris_compiled_shader *shader =
 154       rzalloc_size(cache, sizeof(struct iris_compiled_shader) +
 155                    ice->vtbl.derived_program_state_size(cache_id));
 156    const struct iris_compiled_shader *existing =
 157       find_existing_assembly(cache, assembly, prog_data->program_size);
 158
 159    /* If we can find a matching prog in the cache already, then reuse the
 160     * existing stuff without creating new copy into the underlying buffer
 161     * object.  This is notably useful for programs generating shaders at
 162     * runtime, where multiple shaders may compile to the same thing in our
 163     * backend.
 164     */
 165    if (existing) {
 166       pipe_resource_reference(&shader->assembly.res, existing->assembly.res);
 167       shader->assembly.offset = existing->assembly.offset;
 168       shader->map = existing->map;
 169    } else {
 170       shader->assembly.res = NULL;
 171       u_upload_alloc(ice->shaders.uploader, 0, prog_data->program_size, 64,
 172                      &shader->assembly.offset, &shader->assembly.res,
 173                      &shader->map);
 174       memcpy(shader->map, assembly, prog_data->program_size);
 175    }
 176
 177    shader->prog_data = prog_data;
 178    shader->streamout = streamout;
 179    shader->system_values = system_values;
 180    shader->num_system_values = num_system_values;
 181    shader->num_cbufs = num_cbufs;
 182    shader->bt = *bt;
 183
 184    ralloc_steal(shader, shader->prog_data);
 185    ralloc_steal(shader->prog_data, prog_data->param);
 186    ralloc_steal(shader->prog_data, prog_data->pull_param);
 187    ralloc_steal(shader, shader->streamout);
 188    ralloc_steal(shader, shader->system_values);
 189
 190    /* Store the 3DSTATE shader packets and other derived state. */
 191    ice->vtbl.store_derived_program_state(ice, cache_id, shader);
 192
 193    struct keybox *keybox = make_keybox(shader, cache_id, key, key_size);
 194    _mesa_hash_table_insert(ice->shaders.cache, keybox, shader);
 195
 196    return shader;
 197 }
 198
 199 bool
 200 iris_blorp_lookup_shader(struct blorp_batch *blorp_batch,
 201                          const void *key, uint32_t key_size,
 202                          uint32_t *kernel_out, void *prog_data_out)
 203 {
 204    struct blorp_context *blorp = blorp_batch->blorp;
 205    struct iris_context *ice = blorp->driver_ctx;
 206    struct iris_batch *batch = blorp_batch->driver_batch;
 207    struct iris_compiled_shader *shader =
 208       iris_find_cached_shader(ice, IRIS_CACHE_BLORP, key_size, key);
 209
 210    if (!shader)
 211       return false;
 212
 213    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 214    *kernel_out =
 215       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 216    *((void **) prog_data_out) = shader->prog_data;
 217
 218    iris_use_pinned_bo(batch, bo, false);
 219
 220    return true;
 221 }
 222
 223 bool
 224 iris_blorp_upload_shader(struct blorp_batch *blorp_batch, uint32_t stage,
 225                          const void *key, uint32_t key_size,
 226                          const void *kernel, UNUSED uint32_t kernel_size,
 227                          const struct brw_stage_prog_data *prog_data_templ,
 228                          UNUSED uint32_t prog_data_size,
 229                          uint32_t *kernel_out, void *prog_data_out)
 230 {
 231    struct blorp_context *blorp = blorp_batch->blorp;
 232    struct iris_context *ice = blorp->driver_ctx;
 233    struct iris_batch *batch = blorp_batch->driver_batch;
 234
 235    void *prog_data = ralloc_size(NULL, prog_data_size);
 236    memcpy(prog_data, prog_data_templ, prog_data_size);
 237
 238    struct iris_binding_table bt;
 239    memset(&bt, 0, sizeof(bt));
 240
 241    struct iris_compiled_shader *shader =
 242       iris_upload_shader(ice, IRIS_CACHE_BLORP, key_size, key, kernel,
 243                          prog_data, NULL, NULL, 0, 0, &bt);
 244
 245    struct iris_bo *bo = iris_resource_bo(shader->assembly.res);
 246    *kernel_out =
 247       iris_bo_offset_from_base_address(bo) + shader->assembly.offset;
 248    *((void **) prog_data_out) = shader->prog_data;
 249
 250    iris_use_pinned_bo(batch, bo, false);
 251
 252    return true;
 253 }
 254
 255 void
 256 iris_init_program_cache(struct iris_context *ice)
 257 {
 258    ice->shaders.cache =
 259       _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 260
 261    ice->shaders.uploader =
 262       u_upload_create(&ice->ctx, 16384, PIPE_BIND_CUSTOM, PIPE_USAGE_IMMUTABLE,
 263                       IRIS_RESOURCE_FLAG_SHADER_MEMZONE);
 264 }
 265
 266 void
 267 iris_destroy_program_cache(struct iris_context *ice)
 268 {
 269    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 270       ice->shaders.prog[i] = NULL;
 271    }
 272
 273    hash_table_foreach(ice->shaders.cache, entry) {
 274       struct iris_compiled_shader *shader = entry->data;
 275       pipe_resource_reference(&shader->assembly.res, NULL);
 276    }
 277
 278    u_upload_destroy(ice->shaders.uploader);
 279
 280    ralloc_free(ice->shaders.cache);
 281 }
 282
 283 static const char *
 284 cache_name(enum iris_program_cache_id cache_id)
 285 {
 286    if (cache_id == IRIS_CACHE_BLORP)
 287       return "BLORP";
 288
 289    return _mesa_shader_stage_to_string(cache_id);
 290 }
 291
 292 void
 293 iris_print_program_cache(struct iris_context *ice)
 294 {
 295    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 296    const struct gen_device_info *devinfo = &screen->devinfo;
 297
 298    hash_table_foreach(ice->shaders.cache, entry) {
 299       const struct keybox *keybox = entry->key;
 300       struct iris_compiled_shader *shader = entry->data;
 301       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 302       brw_disassemble(devinfo, shader->map, 0,
 303                       shader->prog_data->program_size, stderr);
 304    }
 305 }