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 <i915_drm.h>
  26 #include "pipe/p_defines.h"
  27 #include "pipe/p_state.h"
  28 #include "pipe/p_context.h"
  29 #include "pipe/p_screen.h"
  30 #include "util/u_atomic.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
  38 struct keybox {
  39    uint8_t size;
  40    enum iris_program_cache_id cache_id;
  41    uint8_t data[0];
  42 };
  43
  44 // XXX: put packets here, expose this somehow and simplify the interfaces
  45 struct iris_program_cache_item {
  46    uint32_t assembly_offset;
  47    uint32_t assembly_size;
  48    void *prog_data;
  49 };
  50
  51 static struct keybox *
  52 make_keybox(struct iris_program_cache *cache,
  53             enum iris_program_cache_id cache_id,
  54             const void *key, unsigned key_size)
  55 {
  56    struct keybox *keybox =
  57       ralloc_size(cache->table, sizeof(struct keybox) + key_size);
  58
  59    keybox->cache_id = cache_id;
  60    keybox->size = key_size;
  61    memcpy(keybox->data, key, key_size);
  62
  63    return keybox;
  64 }
  65
  66 static uint32_t
  67 keybox_hash(const void *void_key)
  68 {
  69    const struct keybox *key = void_key;
  70    return _mesa_hash_data(&key->cache_id, key->size + sizeof(key->cache_id));
  71 }
  72
  73 static bool
  74 keybox_equals(const void *void_a, const void *void_b)
  75 {
  76    const struct keybox *a = void_a, *b = void_b;
  77    if (a->size != b->size)
  78       return false;
  79
  80    return memcmp(a->data, b->data, a->size) == 0;
  81 }
  82
  83 static unsigned
  84 get_program_string_id(enum iris_program_cache_id cache_id, const void *key)
  85 {
  86    switch (cache_id) {
  87    case IRIS_CACHE_VS:
  88       return ((struct brw_vs_prog_key *) key)->program_string_id;
  89    case IRIS_CACHE_TCS:
  90       return ((struct brw_tcs_prog_key *) key)->program_string_id;
  91    case IRIS_CACHE_TES:
  92       return ((struct brw_tes_prog_key *) key)->program_string_id;
  93    case IRIS_CACHE_GS:
  94       return ((struct brw_gs_prog_key *) key)->program_string_id;
  95    case IRIS_CACHE_CS:
  96       return ((struct brw_cs_prog_key *) key)->program_string_id;
  97    case IRIS_CACHE_FS:
  98       return ((struct brw_wm_prog_key *) key)->program_string_id;
  99    default:
 100       unreachable("no program string id for this kind of program");
 101    }
 102 }
 103
 104 /**
 105  * Returns the buffer object matching cache_id and key, or NULL.
 106  */
 107 // XXX: rename to iris_bind_cached_shader?
 108 bool
 109 iris_search_cache(struct iris_context *ice,
 110                   enum iris_program_cache_id cache_id,
 111                   const void *key,
 112                   unsigned key_size,
 113                   uint64_t dirty_flag,
 114                   uint32_t *inout_assembly_offset,
 115                   void *inout_prog_data)
 116 {
 117    struct iris_program_cache *cache = &ice->shaders.cache;
 118
 119    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 120
 121    struct hash_entry *entry = _mesa_hash_table_search(cache->table, keybox);
 122
 123    if (entry == NULL)
 124       return false;
 125
 126    struct iris_program_cache_item *item = entry->data;
 127
 128    if (item->assembly_offset != *inout_assembly_offset ||
 129        item->prog_data != *((void **) inout_prog_data)) {
 130       *inout_assembly_offset = item->assembly_offset;
 131       *((void **) inout_prog_data) = item->prog_data;
 132       ice->state.dirty |= dirty_flag;
 133    }
 134
 135    return true;
 136 }
 137
 138 static void
 139 recreate_cache_bo(struct iris_context *ice, uint32_t size)
 140 {
 141    struct iris_program_cache *cache = &ice->shaders.cache;
 142    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 143    struct iris_bo *old_bo = cache->bo;
 144    void *old_map = cache->map;
 145
 146    cache->bo = iris_bo_alloc(screen->bufmgr, "program cache", size, 64);
 147    cache->bo->kflags = EXEC_OBJECT_CAPTURE;
 148    cache->map = iris_bo_map(&ice->dbg, cache->bo,
 149                             MAP_READ | MAP_WRITE | MAP_ASYNC | MAP_PERSISTENT);
 150
 151    /* Copy any existing data that needs to be saved. */
 152    if (old_bo) {
 153       perf_debug(&ice->dbg,
 154                  "Copying to larger program cache: %u kB -> %u kB\n",
 155                  (unsigned) old_bo->size / 1024,
 156                  (unsigned) cache->bo->size / 1024);
 157
 158       memcpy(cache->map, old_map, cache->next_offset);
 159
 160       iris_bo_unreference(old_bo);
 161       iris_bo_unmap(old_bo);
 162    }
 163 }
 164
 165 const void *
 166 iris_find_previous_compile(struct iris_program_cache *cache,
 167                            enum iris_program_cache_id cache_id,
 168                            unsigned program_string_id)
 169 {
 170    hash_table_foreach(cache->table, entry) {
 171       const struct keybox *keybox = entry->key;
 172       if (keybox->cache_id == cache_id &&
 173           get_program_string_id(cache_id, keybox->data) == program_string_id) {
 174          return keybox->data;
 175       }
 176    }
 177
 178    return NULL;
 179 }
 180
 181 /**
 182  * Look for an existing entry in the cache that has identical assembly code.
 183  *
 184  * This is useful for programs generating shaders at runtime, where multiple
 185  * distinct shaders (from an API perspective) may compile to the same assembly
 186  * in our backend.
 187  */
 188 static const struct iris_program_cache_item *
 189 find_existing_assembly(const struct iris_program_cache *cache,
 190                        const void *assembly,
 191                        unsigned assembly_size)
 192 {
 193    hash_table_foreach(cache->table, entry) {
 194       const struct iris_program_cache_item *item = entry->data;
 195       if (item->assembly_size == assembly_size &&
 196           memcmp(cache->map + item->assembly_offset,
 197                  assembly, assembly_size) == 0)
 198          return item;
 199    }
 200    return NULL;
 201 }
 202
 203 static uint32_t
 204 upload_new_assembly(struct iris_context *ice,
 205                     const void *assembly,
 206                     unsigned size)
 207 {
 208    struct iris_program_cache *cache = &ice->shaders.cache;
 209
 210    /* Allocate space in the cache BO for our new program. */
 211    if (cache->next_offset + size > cache->bo->size) {
 212       uint32_t new_size = cache->bo->size * 2;
 213
 214       while (cache->next_offset + size > new_size)
 215          new_size *= 2;
 216
 217       recreate_cache_bo(ice, new_size);
 218    }
 219
 220    uint32_t offset = cache->next_offset;
 221
 222    /* Programs are always 64-byte aligned, so set up the next one now */
 223    cache->next_offset = ALIGN(offset + size, 64);
 224
 225    /* Copy data to the buffer */
 226    memcpy(cache->map + offset, assembly, size);
 227
 228    return offset;
 229 }
 230
 231 void
 232 iris_upload_cache(struct iris_context *ice,
 233                   enum iris_program_cache_id cache_id,
 234                   const void *key,
 235                   unsigned key_size,
 236                   const void *assembly,
 237                   unsigned assembly_size,
 238                   const void *prog_data,
 239                   unsigned prog_data_size,
 240                   uint32_t *out_assembly_offset,
 241                   void *out_prog_data)
 242 {
 243    struct iris_program_cache *cache = &ice->shaders.cache;
 244    struct iris_program_cache_item *item =
 245       ralloc(cache->table, struct iris_program_cache_item);
 246    const struct iris_program_cache_item *matching_data =
 247       find_existing_assembly(cache, assembly, assembly_size);
 248
 249    /* If we can find a matching prog in the cache already, then reuse the
 250     * existing stuff without creating new copy into the underlying buffer
 251     * object.  This is notably useful for programs generating shaders at
 252     * runtime, where multiple shaders may compile to the same thing in our
 253     * backend.
 254     */
 255    if (matching_data) {
 256       item->assembly_offset = matching_data->assembly_offset;
 257    } else {
 258       item->assembly_offset =
 259          upload_new_assembly(ice, assembly, assembly_size);
 260    }
 261
 262    item->assembly_size = assembly_size;
 263    item->prog_data = ralloc_size(item, prog_data_size);
 264    memcpy(item->prog_data, prog_data, prog_data_size);
 265
 266    if (cache_id != IRIS_CACHE_BLORP) {
 267       struct brw_stage_prog_data *stage_prog_data = prog_data;
 268       ralloc_steal(item->prog_data, stage_prog_data->param);
 269       ralloc_steal(item->prog_data, stage_prog_data->pull_param);
 270    }
 271
 272    struct keybox *keybox = make_keybox(cache, cache_id, key, key_size);
 273    _mesa_hash_table_insert(cache->table, keybox, item);
 274
 275    *out_assembly_offset = item->assembly_offset;
 276    *(void **)out_prog_data = item->prog_data;
 277 }
 278
 279 void
 280 iris_init_program_cache(struct iris_context *ice)
 281 {
 282    struct iris_program_cache *cache = &ice->shaders.cache;
 283
 284    cache->table = _mesa_hash_table_create(ice, keybox_hash, keybox_equals);
 285
 286    recreate_cache_bo(ice, 16384);
 287 }
 288
 289 void
 290 iris_destroy_program_cache(struct iris_context *ice)
 291 {
 292    struct iris_program_cache *cache = &ice->shaders.cache;
 293
 294    /* This can be NULL if context creation failed early on. */
 295    if (cache->bo) {
 296       iris_bo_unmap(cache->bo);
 297       iris_bo_unreference(cache->bo);
 298       cache->bo = NULL;
 299       cache->map = NULL;
 300    }
 301
 302    cache->next_offset = 0;
 303
 304    /* Also, NULL out any stale program pointers. */
 305    for (int i = 0; i < MESA_SHADER_STAGES; i++) {
 306       ice->shaders.prog_data[i] = NULL;
 307    }
 308 }
 309
 310 static const char *
 311 cache_name(enum iris_program_cache_id cache_id)
 312 {
 313    if (cache_id == IRIS_CACHE_BLORP)
 314       return "BLORP";
 315
 316    return _mesa_shader_stage_to_string(cache_id);
 317 }
 318
 319 void
 320 iris_print_program_cache(struct iris_context *ice)
 321 {
 322    struct iris_program_cache *cache = &ice->shaders.cache;
 323    struct iris_screen *screen = (struct iris_screen *)ice->ctx.screen;
 324    const struct gen_device_info *devinfo = &screen->devinfo;
 325
 326    hash_table_foreach(cache->table, entry) {
 327       const struct keybox *keybox = entry->key;
 328       struct iris_program_cache_item *item = entry->data;
 329       fprintf(stderr, "%s:\n", cache_name(keybox->cache_id));
 330       brw_disassemble(devinfo, cache->map,
 331                       item->assembly_offset, item->assembly_size, stderr);
 332    }
 333 }