src/gallium/drivers/vc5/vc5_program.c

   1 /*
   2  * Copyright © 2014-2017 Broadcom
   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 (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 NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <inttypes.h>
  25 #include "util/u_format.h"
  26 #include "util/u_math.h"
  27 #include "util/u_memory.h"
  28 #include "util/ralloc.h"
  29 #include "util/hash_table.h"
  30 #include "tgsi/tgsi_dump.h"
  31 #include "tgsi/tgsi_parse.h"
  32 #include "compiler/nir/nir.h"
  33 #include "compiler/nir/nir_builder.h"
  34 #include "nir/tgsi_to_nir.h"
  35 #include "compiler/v3d_compiler.h"
  36 #include "vc5_context.h"
  37 #include "broadcom/cle/v3d_packet_v33_pack.h"
  38
  39 static gl_varying_slot
  40 vc5_get_slot_for_driver_location(nir_shader *s, uint32_t driver_location)
  41 {
  42         nir_foreach_variable(var, &s->outputs) {
  43                 if (var->data.driver_location == driver_location) {
  44                         return var->data.location;
  45                 }
  46         }
  47
  48         return -1;
  49 }
  50
  51 static void
  52 vc5_set_transform_feedback_outputs(struct vc5_uncompiled_shader *so,
  53                                    const struct pipe_stream_output_info *stream_output)
  54 {
  55         if (!stream_output->num_outputs)
  56                 return;
  57
  58         struct v3d_varying_slot slots[PIPE_MAX_SO_OUTPUTS * 4];
  59         int slot_count = 0;
  60
  61         for (int buffer = 0; buffer < PIPE_MAX_SO_BUFFERS; buffer++) {
  62                 uint32_t buffer_offset = 0;
  63                 uint32_t vpm_start = slot_count;
  64
  65                 for (int i = 0; i < stream_output->num_outputs; i++) {
  66                         const struct pipe_stream_output *output =
  67                                 &stream_output->output[i];
  68
  69                         if (output->output_buffer != buffer)
  70                                 continue;
  71
  72                         /* We assume that the SO outputs appear in increasing
  73                          * order in the buffer.
  74                          */
  75                         assert(output->dst_offset >= buffer_offset);
  76
  77                         /* Pad any undefined slots in the output */
  78                         for (int j = buffer_offset; j < output->dst_offset; j++) {
  79                                 slots[slot_count] =
  80                                         v3d_slot_from_slot_and_component(VARYING_SLOT_POS, 0);
  81                                 slot_count++;
  82                                 buffer_offset++;
  83                         }
  84
  85                         /* Set the coordinate shader up to output the
  86                          * components of this varying.
  87                          */
  88                         for (int j = 0; j < output->num_components; j++) {
  89                                 gl_varying_slot slot =
  90                                         vc5_get_slot_for_driver_location(so->base.ir.nir, output->register_index);
  91
  92                                 slots[slot_count] =
  93                                         v3d_slot_from_slot_and_component(slot,
  94                                                                          output->start_component + j);
  95                                 slot_count++;
  96                                 buffer_offset++;
  97                         }
  98                 }
  99
 100                 uint32_t vpm_size = slot_count - vpm_start;
 101                 if (!vpm_size)
 102                         continue;
 103
 104                 struct V3D33_TRANSFORM_FEEDBACK_OUTPUT_DATA_SPEC unpacked = {
 105                         /* We need the offset from the coordinate shader's VPM
 106                          * output block, which has the [X, Y, Z, W, Xs, Ys]
 107                          * values at the start.  Note that this will need some
 108                          * shifting when PSIZ is also present.
 109                          */
 110                         .first_shaded_vertex_value_to_output = vpm_start + 6,
 111                         .number_of_consecutive_vertex_values_to_output_as_32_bit_values_minus_1 = vpm_size - 1,
 112                         .output_buffer_to_write_to = buffer,
 113                 };
 114                 V3D33_TRANSFORM_FEEDBACK_OUTPUT_DATA_SPEC_pack(NULL,
 115                                                                (void *)&so->tf_specs[so->num_tf_specs++],
 116                                                                &unpacked);
 117         }
 118
 119         so->num_tf_outputs = slot_count;
 120         so->tf_outputs = ralloc_array(so->base.ir.nir, struct v3d_varying_slot,
 121                                       slot_count);
 122         memcpy(so->tf_outputs, slots, sizeof(*slots) * slot_count);
 123 }
 124
 125 static int
 126 type_size(const struct glsl_type *type)
 127 {
 128         return glsl_count_attribute_slots(type, false);
 129 }
 130
 131 static void *
 132 vc5_shader_state_create(struct pipe_context *pctx,
 133                         const struct pipe_shader_state *cso)
 134 {
 135         struct vc5_context *vc5 = vc5_context(pctx);
 136         struct vc5_uncompiled_shader *so = CALLOC_STRUCT(vc5_uncompiled_shader);
 137         if (!so)
 138                 return NULL;
 139
 140         so->program_id = vc5->next_uncompiled_program_id++;
 141
 142         nir_shader *s;
 143
 144         if (cso->type == PIPE_SHADER_IR_NIR) {
 145                 /* The backend takes ownership of the NIR shader on state
 146                  * creation.
 147                  */
 148                 s = cso->ir.nir;
 149
 150                 NIR_PASS_V(s, nir_lower_io, nir_var_all, type_size,
 151                            (nir_lower_io_options)0);
 152         } else {
 153                 assert(cso->type == PIPE_SHADER_IR_TGSI);
 154
 155                 if (V3D_DEBUG & V3D_DEBUG_TGSI) {
 156                         fprintf(stderr, "prog %d TGSI:\n",
 157                                 so->program_id);
 158                         tgsi_dump(cso->tokens, 0);
 159                         fprintf(stderr, "\n");
 160                 }
 161                 s = tgsi_to_nir(cso->tokens, &v3d_nir_options);
 162         }
 163
 164         NIR_PASS_V(s, nir_opt_global_to_local);
 165         NIR_PASS_V(s, nir_lower_regs_to_ssa);
 166         NIR_PASS_V(s, nir_normalize_cubemap_coords);
 167
 168         NIR_PASS_V(s, nir_lower_load_const_to_scalar);
 169
 170         v3d_optimize_nir(s);
 171
 172         NIR_PASS_V(s, nir_remove_dead_variables, nir_var_local);
 173
 174         /* Garbage collect dead instructions */
 175         nir_sweep(s);
 176
 177         so->base.type = PIPE_SHADER_IR_NIR;
 178         so->base.ir.nir = s;
 179
 180         vc5_set_transform_feedback_outputs(so, &cso->stream_output);
 181
 182         if (V3D_DEBUG & (V3D_DEBUG_NIR |
 183                          v3d_debug_flag_for_shader_stage(s->info.stage))) {
 184                 fprintf(stderr, "%s prog %d NIR:\n",
 185                         gl_shader_stage_name(s->info.stage),
 186                         so->program_id);
 187                 nir_print_shader(s, stderr);
 188                 fprintf(stderr, "\n");
 189         }
 190
 191         return so;
 192 }
 193
 194 static struct vc5_compiled_shader *
 195 vc5_get_compiled_shader(struct vc5_context *vc5, struct v3d_key *key)
 196 {
 197         struct vc5_uncompiled_shader *shader_state = key->shader_state;
 198         nir_shader *s = shader_state->base.ir.nir;
 199
 200         struct hash_table *ht;
 201         uint32_t key_size;
 202         if (s->info.stage == MESA_SHADER_FRAGMENT) {
 203                 ht = vc5->fs_cache;
 204                 key_size = sizeof(struct v3d_fs_key);
 205         } else {
 206                 ht = vc5->vs_cache;
 207                 key_size = sizeof(struct v3d_vs_key);
 208         }
 209
 210         struct hash_entry *entry = _mesa_hash_table_search(ht, key);
 211         if (entry)
 212                 return entry->data;
 213
 214         struct vc5_compiled_shader *shader =
 215                 rzalloc(NULL, struct vc5_compiled_shader);
 216
 217         int program_id = shader_state->program_id;
 218         int variant_id =
 219                 p_atomic_inc_return(&shader_state->compiled_variant_count);
 220         uint64_t *qpu_insts;
 221         uint32_t shader_size;
 222
 223         switch (s->info.stage) {
 224         case MESA_SHADER_VERTEX:
 225                 shader->prog_data.vs = rzalloc(shader, struct v3d_vs_prog_data);
 226
 227                 qpu_insts = v3d_compile_vs(vc5->screen->compiler,
 228                                            (struct v3d_vs_key *)key,
 229                                            shader->prog_data.vs, s,
 230                                            program_id, variant_id,
 231                                            &shader_size);
 232                 break;
 233         case MESA_SHADER_FRAGMENT:
 234                 shader->prog_data.fs = rzalloc(shader, struct v3d_fs_prog_data);
 235
 236                 qpu_insts = v3d_compile_fs(vc5->screen->compiler,
 237                                            (struct v3d_fs_key *)key,
 238                                            shader->prog_data.fs, s,
 239                                            program_id, variant_id,
 240                                            &shader_size);
 241                 break;
 242         default:
 243                 unreachable("bad stage");
 244         }
 245
 246         vc5_set_shader_uniform_dirty_flags(shader);
 247
 248         shader->bo = vc5_bo_alloc(vc5->screen, shader_size, "shader");
 249         vc5_bo_map(shader->bo);
 250         memcpy(shader->bo->map, qpu_insts, shader_size);
 251
 252         free(qpu_insts);
 253
 254         struct vc5_key *dup_key;
 255         dup_key = ralloc_size(shader, key_size);
 256         memcpy(dup_key, key, key_size);
 257         _mesa_hash_table_insert(ht, dup_key, shader);
 258
 259         return shader;
 260 }
 261
 262 static void
 263 vc5_setup_shared_key(struct vc5_context *vc5, struct v3d_key *key,
 264                      struct vc5_texture_stateobj *texstate)
 265 {
 266         const struct v3d_device_info *devinfo = &vc5->screen->devinfo;
 267
 268         for (int i = 0; i < texstate->num_textures; i++) {
 269                 struct pipe_sampler_view *sampler = texstate->textures[i];
 270                 struct vc5_sampler_view *vc5_sampler = vc5_sampler_view(sampler);
 271                 struct pipe_sampler_state *sampler_state =
 272                         texstate->samplers[i];
 273
 274                 if (!sampler)
 275                         continue;
 276
 277                 key->tex[i].return_size =
 278                         vc5_get_tex_return_size(devinfo,
 279                                                 sampler->format,
 280                                                 sampler_state->compare_mode);
 281
 282                 /* For 16-bit, we set up the sampler to always return 2
 283                  * channels (meaning no recompiles for most statechanges),
 284                  * while for 32 we actually scale the returns with channels.
 285                  */
 286                 if (key->tex[i].return_size == 16) {
 287                         key->tex[i].return_channels = 2;
 288                 } else if (devinfo->ver > 40) {
 289                         key->tex[i].return_channels = 4;
 290                 } else {
 291                         key->tex[i].return_channels =
 292                                 vc5_get_tex_return_channels(devinfo,
 293                                                             sampler->format);
 294                 }
 295
 296                 if (key->tex[i].return_size == 32 && devinfo->ver < 40) {
 297                         memcpy(key->tex[i].swizzle,
 298                                vc5_sampler->swizzle,
 299                                sizeof(vc5_sampler->swizzle));
 300                 } else {
 301                         /* For 16-bit returns, we let the sampler state handle
 302                          * the swizzle.
 303                          */
 304                         key->tex[i].swizzle[0] = PIPE_SWIZZLE_X;
 305                         key->tex[i].swizzle[1] = PIPE_SWIZZLE_Y;
 306                         key->tex[i].swizzle[2] = PIPE_SWIZZLE_Z;
 307                         key->tex[i].swizzle[3] = PIPE_SWIZZLE_W;
 308                 }
 309
 310                 if (sampler->texture->nr_samples > 1) {
 311                         key->tex[i].msaa_width = sampler->texture->width0;
 312                         key->tex[i].msaa_height = sampler->texture->height0;
 313                 } else if (sampler){
 314                         key->tex[i].compare_mode = sampler_state->compare_mode;
 315                         key->tex[i].compare_func = sampler_state->compare_func;
 316                         key->tex[i].clamp_s =
 317                                 sampler_state->wrap_s == PIPE_TEX_WRAP_CLAMP;
 318                         key->tex[i].clamp_t =
 319                                 sampler_state->wrap_t == PIPE_TEX_WRAP_CLAMP;
 320                         key->tex[i].clamp_r =
 321                                 sampler_state->wrap_r == PIPE_TEX_WRAP_CLAMP;
 322                 }
 323         }
 324
 325         key->ucp_enables = vc5->rasterizer->base.clip_plane_enable;
 326 }
 327
 328 static void
 329 vc5_update_compiled_fs(struct vc5_context *vc5, uint8_t prim_mode)
 330 {
 331         struct vc5_job *job = vc5->job;
 332         struct v3d_fs_key local_key;
 333         struct v3d_fs_key *key = &local_key;
 334
 335         if (!(vc5->dirty & (VC5_DIRTY_PRIM_MODE |
 336                             VC5_DIRTY_BLEND |
 337                             VC5_DIRTY_FRAMEBUFFER |
 338                             VC5_DIRTY_ZSA |
 339                             VC5_DIRTY_RASTERIZER |
 340                             VC5_DIRTY_SAMPLE_MASK |
 341                             VC5_DIRTY_FRAGTEX |
 342                             VC5_DIRTY_UNCOMPILED_FS))) {
 343                 return;
 344         }
 345
 346         memset(key, 0, sizeof(*key));
 347         vc5_setup_shared_key(vc5, &key->base, &vc5->fragtex);
 348         key->base.shader_state = vc5->prog.bind_fs;
 349         key->is_points = (prim_mode == PIPE_PRIM_POINTS);
 350         key->is_lines = (prim_mode >= PIPE_PRIM_LINES &&
 351                          prim_mode <= PIPE_PRIM_LINE_STRIP);
 352         key->clamp_color = vc5->rasterizer->base.clamp_fragment_color;
 353         if (vc5->blend->logicop_enable) {
 354                 key->logicop_func = vc5->blend->logicop_func;
 355         } else {
 356                 key->logicop_func = PIPE_LOGICOP_COPY;
 357         }
 358         if (job->msaa) {
 359                 key->msaa = vc5->rasterizer->base.multisample;
 360                 key->sample_coverage = (vc5->rasterizer->base.multisample &&
 361                                         vc5->sample_mask != (1 << VC5_MAX_SAMPLES) - 1);
 362                 key->sample_alpha_to_coverage = vc5->blend->alpha_to_coverage;
 363                 key->sample_alpha_to_one = vc5->blend->alpha_to_one;
 364         }
 365
 366         key->depth_enabled = (vc5->zsa->base.depth.enabled ||
 367                               vc5->zsa->base.stencil[0].enabled);
 368         if (vc5->zsa->base.alpha.enabled) {
 369                 key->alpha_test = true;
 370                 key->alpha_test_func = vc5->zsa->base.alpha.func;
 371         }
 372
 373         /* gl_FragColor's propagation to however many bound color buffers
 374          * there are means that the buffer count needs to be in the key.
 375          */
 376         key->nr_cbufs = vc5->framebuffer.nr_cbufs;
 377         key->swap_color_rb = vc5->swap_color_rb;
 378
 379         for (int i = 0; i < key->nr_cbufs; i++) {
 380                 struct pipe_surface *cbuf = vc5->framebuffer.cbufs[i];
 381                 const struct util_format_description *desc =
 382                         util_format_description(cbuf->format);
 383
 384                 if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT &&
 385                     desc->channel[0].size == 32) {
 386                         key->f32_color_rb |= 1 << i;
 387                 }
 388         }
 389
 390         if (key->is_points) {
 391                 key->point_sprite_mask =
 392                         vc5->rasterizer->base.sprite_coord_enable;
 393                 key->point_coord_upper_left =
 394                         (vc5->rasterizer->base.sprite_coord_mode ==
 395                          PIPE_SPRITE_COORD_UPPER_LEFT);
 396         }
 397
 398         key->light_twoside = vc5->rasterizer->base.light_twoside;
 399         key->shade_model_flat = vc5->rasterizer->base.flatshade;
 400
 401         struct vc5_compiled_shader *old_fs = vc5->prog.fs;
 402         vc5->prog.fs = vc5_get_compiled_shader(vc5, &key->base);
 403         if (vc5->prog.fs == old_fs)
 404                 return;
 405
 406         vc5->dirty |= VC5_DIRTY_COMPILED_FS;
 407
 408         if (old_fs &&
 409             vc5->prog.fs->prog_data.fs->flat_shade_flags !=
 410             old_fs->prog_data.fs->flat_shade_flags) {
 411                 vc5->dirty |= VC5_DIRTY_FLAT_SHADE_FLAGS;
 412         }
 413
 414         if (old_fs && memcmp(vc5->prog.fs->prog_data.fs->input_slots,
 415                              old_fs->prog_data.fs->input_slots,
 416                              sizeof(vc5->prog.fs->prog_data.fs->input_slots))) {
 417                 vc5->dirty |= VC5_DIRTY_FS_INPUTS;
 418         }
 419 }
 420
 421 static void
 422 vc5_update_compiled_vs(struct vc5_context *vc5, uint8_t prim_mode)
 423 {
 424         struct v3d_vs_key local_key;
 425         struct v3d_vs_key *key = &local_key;
 426
 427         if (!(vc5->dirty & (VC5_DIRTY_PRIM_MODE |
 428                             VC5_DIRTY_RASTERIZER |
 429                             VC5_DIRTY_VERTTEX |
 430                             VC5_DIRTY_VTXSTATE |
 431                             VC5_DIRTY_UNCOMPILED_VS |
 432                             VC5_DIRTY_FS_INPUTS))) {
 433                 return;
 434         }
 435
 436         memset(key, 0, sizeof(*key));
 437         vc5_setup_shared_key(vc5, &key->base, &vc5->verttex);
 438         key->base.shader_state = vc5->prog.bind_vs;
 439         key->num_fs_inputs = vc5->prog.fs->prog_data.fs->base.num_inputs;
 440         STATIC_ASSERT(sizeof(key->fs_inputs) ==
 441                       sizeof(vc5->prog.fs->prog_data.fs->input_slots));
 442         memcpy(key->fs_inputs, vc5->prog.fs->prog_data.fs->input_slots,
 443                sizeof(key->fs_inputs));
 444         key->clamp_color = vc5->rasterizer->base.clamp_vertex_color;
 445
 446         key->per_vertex_point_size =
 447                 (prim_mode == PIPE_PRIM_POINTS &&
 448                  vc5->rasterizer->base.point_size_per_vertex);
 449
 450         struct vc5_compiled_shader *vs =
 451                 vc5_get_compiled_shader(vc5, &key->base);
 452         if (vs != vc5->prog.vs) {
 453                 vc5->prog.vs = vs;
 454                 vc5->dirty |= VC5_DIRTY_COMPILED_VS;
 455         }
 456
 457         key->is_coord = true;
 458         /* Coord shaders only output varyings used by transform feedback. */
 459         struct vc5_uncompiled_shader *shader_state = key->base.shader_state;
 460         memcpy(key->fs_inputs, shader_state->tf_outputs,
 461                sizeof(*key->fs_inputs) * shader_state->num_tf_outputs);
 462         if (shader_state->num_tf_outputs < key->num_fs_inputs) {
 463                 memset(&key->fs_inputs[shader_state->num_tf_outputs],
 464                        0,
 465                        sizeof(*key->fs_inputs) * (key->num_fs_inputs -
 466                                                   shader_state->num_tf_outputs));
 467         }
 468         key->num_fs_inputs = shader_state->num_tf_outputs;
 469
 470         struct vc5_compiled_shader *cs =
 471                 vc5_get_compiled_shader(vc5, &key->base);
 472         if (cs != vc5->prog.cs) {
 473                 vc5->prog.cs = cs;
 474                 vc5->dirty |= VC5_DIRTY_COMPILED_CS;
 475         }
 476 }
 477
 478 void
 479 vc5_update_compiled_shaders(struct vc5_context *vc5, uint8_t prim_mode)
 480 {
 481         vc5_update_compiled_fs(vc5, prim_mode);
 482         vc5_update_compiled_vs(vc5, prim_mode);
 483 }
 484
 485 static uint32_t
 486 fs_cache_hash(const void *key)
 487 {
 488         return _mesa_hash_data(key, sizeof(struct v3d_fs_key));
 489 }
 490
 491 static uint32_t
 492 vs_cache_hash(const void *key)
 493 {
 494         return _mesa_hash_data(key, sizeof(struct v3d_vs_key));
 495 }
 496
 497 static bool
 498 fs_cache_compare(const void *key1, const void *key2)
 499 {
 500         return memcmp(key1, key2, sizeof(struct v3d_fs_key)) == 0;
 501 }
 502
 503 static bool
 504 vs_cache_compare(const void *key1, const void *key2)
 505 {
 506         return memcmp(key1, key2, sizeof(struct v3d_vs_key)) == 0;
 507 }
 508
 509 static void
 510 delete_from_cache_if_matches(struct hash_table *ht,
 511                              struct vc5_compiled_shader **last_compile,
 512                              struct hash_entry *entry,
 513                              struct vc5_uncompiled_shader *so)
 514 {
 515         const struct v3d_key *key = entry->key;
 516
 517         if (key->shader_state == so) {
 518                 struct vc5_compiled_shader *shader = entry->data;
 519                 _mesa_hash_table_remove(ht, entry);
 520                 vc5_bo_unreference(&shader->bo);
 521
 522                 if (shader == *last_compile)
 523                         *last_compile = NULL;
 524
 525                 ralloc_free(shader);
 526         }
 527 }
 528
 529 static void
 530 vc5_shader_state_delete(struct pipe_context *pctx, void *hwcso)
 531 {
 532         struct vc5_context *vc5 = vc5_context(pctx);
 533         struct vc5_uncompiled_shader *so = hwcso;
 534
 535         struct hash_entry *entry;
 536         hash_table_foreach(vc5->fs_cache, entry) {
 537                 delete_from_cache_if_matches(vc5->fs_cache, &vc5->prog.fs,
 538                                              entry, so);
 539         }
 540         hash_table_foreach(vc5->vs_cache, entry) {
 541                 delete_from_cache_if_matches(vc5->vs_cache, &vc5->prog.vs,
 542                                              entry, so);
 543         }
 544
 545         ralloc_free(so->base.ir.nir);
 546         free(so);
 547 }
 548
 549 static void
 550 vc5_fp_state_bind(struct pipe_context *pctx, void *hwcso)
 551 {
 552         struct vc5_context *vc5 = vc5_context(pctx);
 553         vc5->prog.bind_fs = hwcso;
 554         vc5->dirty |= VC5_DIRTY_UNCOMPILED_FS;
 555 }
 556
 557 static void
 558 vc5_vp_state_bind(struct pipe_context *pctx, void *hwcso)
 559 {
 560         struct vc5_context *vc5 = vc5_context(pctx);
 561         vc5->prog.bind_vs = hwcso;
 562         vc5->dirty |= VC5_DIRTY_UNCOMPILED_VS;
 563 }
 564
 565 void
 566 vc5_program_init(struct pipe_context *pctx)
 567 {
 568         struct vc5_context *vc5 = vc5_context(pctx);
 569
 570         pctx->create_vs_state = vc5_shader_state_create;
 571         pctx->delete_vs_state = vc5_shader_state_delete;
 572
 573         pctx->create_fs_state = vc5_shader_state_create;
 574         pctx->delete_fs_state = vc5_shader_state_delete;
 575
 576         pctx->bind_fs_state = vc5_fp_state_bind;
 577         pctx->bind_vs_state = vc5_vp_state_bind;
 578
 579         vc5->fs_cache = _mesa_hash_table_create(pctx, fs_cache_hash,
 580                                                 fs_cache_compare);
 581         vc5->vs_cache = _mesa_hash_table_create(pctx, vs_cache_hash,
 582                                                 vs_cache_compare);
 583 }
 584
 585 void
 586 vc5_program_fini(struct pipe_context *pctx)
 587 {
 588         struct vc5_context *vc5 = vc5_context(pctx);
 589
 590         struct hash_entry *entry;
 591         hash_table_foreach(vc5->fs_cache, entry) {
 592                 struct vc5_compiled_shader *shader = entry->data;
 593                 vc5_bo_unreference(&shader->bo);
 594                 ralloc_free(shader);
 595                 _mesa_hash_table_remove(vc5->fs_cache, entry);
 596         }
 597
 598         hash_table_foreach(vc5->vs_cache, entry) {
 599                 struct vc5_compiled_shader *shader = entry->data;
 600                 vc5_bo_unreference(&shader->bo);
 601                 ralloc_free(shader);
 602                 _mesa_hash_table_remove(vc5->vs_cache, entry);
 603         }
 604 }