src/gallium/drivers/vc4/vc4_job.c

   1 /*
   2  * Copyright © 2014-2015 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 /** @file vc4_job.c
  25  *
  26  * Functions for submitting VC4 render jobs to the kernel.
  27  */
  28
  29 #include <xf86drm.h>
  30 #include "vc4_cl_dump.h"
  31 #include "vc4_context.h"
  32 #include "util/hash_table.h"
  33
  34 static void
  35 remove_from_ht(struct hash_table *ht, void *key)
  36 {
  37         struct hash_entry *entry = _mesa_hash_table_search(ht, key);
  38         _mesa_hash_table_remove(ht, entry);
  39 }
  40
  41 static void
  42 vc4_job_free(struct vc4_context *vc4, struct vc4_job *job)
  43 {
  44         struct vc4_bo **referenced_bos = job->bo_pointers.base;
  45         for (int i = 0; i < cl_offset(&job->bo_handles) / 4; i++) {
  46                 vc4_bo_unreference(&referenced_bos[i]);
  47         }
  48
  49         remove_from_ht(vc4->jobs, &job->key);
  50
  51         if (job->color_write) {
  52                 remove_from_ht(vc4->write_jobs, job->color_write->texture);
  53                 pipe_surface_reference(&job->color_write, NULL);
  54         }
  55         if (job->msaa_color_write) {
  56                 remove_from_ht(vc4->write_jobs, job->msaa_color_write->texture);
  57                 pipe_surface_reference(&job->msaa_color_write, NULL);
  58         }
  59         if (job->zs_write) {
  60                 remove_from_ht(vc4->write_jobs, job->zs_write->texture);
  61                 pipe_surface_reference(&job->zs_write, NULL);
  62         }
  63         if (job->msaa_zs_write) {
  64                 remove_from_ht(vc4->write_jobs, job->msaa_zs_write->texture);
  65                 pipe_surface_reference(&job->msaa_zs_write, NULL);
  66         }
  67
  68         pipe_surface_reference(&job->color_read, NULL);
  69         pipe_surface_reference(&job->zs_read, NULL);
  70
  71         if (vc4->job == job)
  72                 vc4->job = NULL;
  73
  74         ralloc_free(job);
  75 }
  76
  77 static struct vc4_job *
  78 vc4_job_create(struct vc4_context *vc4)
  79 {
  80         struct vc4_job *job = rzalloc(vc4, struct vc4_job);
  81
  82         vc4_init_cl(job, &job->bcl);
  83         vc4_init_cl(job, &job->shader_rec);
  84         vc4_init_cl(job, &job->uniforms);
  85         vc4_init_cl(job, &job->bo_handles);
  86         vc4_init_cl(job, &job->bo_pointers);
  87
  88         job->draw_min_x = ~0;
  89         job->draw_min_y = ~0;
  90         job->draw_max_x = 0;
  91         job->draw_max_y = 0;
  92
  93         job->last_gem_handle_hindex = ~0;
  94
  95         if (vc4->perfmon)
  96                 job->perfmon = vc4->perfmon;
  97
  98         return job;
  99 }
 100
 101 void
 102 vc4_flush_jobs_writing_resource(struct vc4_context *vc4,
 103                                 struct pipe_resource *prsc)
 104 {
 105         struct hash_entry *entry = _mesa_hash_table_search(vc4->write_jobs,
 106                                                            prsc);
 107         if (entry) {
 108                 struct vc4_job *job = entry->data;
 109                 vc4_job_submit(vc4, job);
 110         }
 111 }
 112
 113 void
 114 vc4_flush_jobs_reading_resource(struct vc4_context *vc4,
 115                                 struct pipe_resource *prsc)
 116 {
 117         struct vc4_resource *rsc = vc4_resource(prsc);
 118
 119         vc4_flush_jobs_writing_resource(vc4, prsc);
 120
 121         struct hash_entry *entry;
 122         hash_table_foreach(vc4->jobs, entry) {
 123                 struct vc4_job *job = entry->data;
 124
 125                 struct vc4_bo **referenced_bos = job->bo_pointers.base;
 126                 bool found = false;
 127                 for (int i = 0; i < cl_offset(&job->bo_handles) / 4; i++) {
 128                         if (referenced_bos[i] == rsc->bo) {
 129                                 found = true;
 130                                 break;
 131                         }
 132                 }
 133                 if (found) {
 134                         vc4_job_submit(vc4, job);
 135                         continue;
 136                 }
 137
 138                 /* Also check for the Z/color buffers, since the references to
 139                  * those are only added immediately before submit.
 140                  */
 141                 if (job->color_read && !(job->cleared & PIPE_CLEAR_COLOR)) {
 142                         struct vc4_resource *ctex =
 143                                 vc4_resource(job->color_read->texture);
 144                         if (ctex->bo == rsc->bo) {
 145                                 vc4_job_submit(vc4, job);
 146                                 continue;
 147                         }
 148                 }
 149
 150                 if (job->zs_read && !(job->cleared &
 151                                       (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
 152                         struct vc4_resource *ztex =
 153                                 vc4_resource(job->zs_read->texture);
 154                         if (ztex->bo == rsc->bo) {
 155                                 vc4_job_submit(vc4, job);
 156                                 continue;
 157                         }
 158                 }
 159         }
 160 }
 161
 162 /**
 163  * Returns a vc4_job struture for tracking V3D rendering to a particular FBO.
 164  *
 165  * If we've already started rendering to this FBO, then return old same job,
 166  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 167  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 168  * have been flushed.
 169  */
 170 struct vc4_job *
 171 vc4_get_job(struct vc4_context *vc4,
 172             struct pipe_surface *cbuf, struct pipe_surface *zsbuf)
 173 {
 174         /* Return the existing job for this FBO if we have one */
 175         struct vc4_job_key local_key = {.cbuf = cbuf, .zsbuf = zsbuf};
 176         struct hash_entry *entry = _mesa_hash_table_search(vc4->jobs,
 177                                                            &local_key);
 178         if (entry)
 179                 return entry->data;
 180
 181         /* Creating a new job.  Make sure that any previous jobs reading or
 182          * writing these buffers are flushed.
 183          */
 184         if (cbuf)
 185                 vc4_flush_jobs_reading_resource(vc4, cbuf->texture);
 186         if (zsbuf)
 187                 vc4_flush_jobs_reading_resource(vc4, zsbuf->texture);
 188
 189         struct vc4_job *job = vc4_job_create(vc4);
 190
 191         if (cbuf) {
 192                 if (cbuf->texture->nr_samples > 1) {
 193                         job->msaa = true;
 194                         pipe_surface_reference(&job->msaa_color_write, cbuf);
 195                 } else {
 196                         pipe_surface_reference(&job->color_write, cbuf);
 197                 }
 198         }
 199
 200         if (zsbuf) {
 201                 if (zsbuf->texture->nr_samples > 1) {
 202                         job->msaa = true;
 203                         pipe_surface_reference(&job->msaa_zs_write, zsbuf);
 204                 } else {
 205                         pipe_surface_reference(&job->zs_write, zsbuf);
 206                 }
 207         }
 208
 209         if (job->msaa) {
 210                 job->tile_width = 32;
 211                 job->tile_height = 32;
 212         } else {
 213                 job->tile_width = 64;
 214                 job->tile_height = 64;
 215         }
 216
 217         if (cbuf)
 218                 _mesa_hash_table_insert(vc4->write_jobs, cbuf->texture, job);
 219         if (zsbuf)
 220                 _mesa_hash_table_insert(vc4->write_jobs, zsbuf->texture, job);
 221
 222         job->key.cbuf = cbuf;
 223         job->key.zsbuf = zsbuf;
 224         _mesa_hash_table_insert(vc4->jobs, &job->key, job);
 225
 226         return job;
 227 }
 228
 229 struct vc4_job *
 230 vc4_get_job_for_fbo(struct vc4_context *vc4)
 231 {
 232         if (vc4->job)
 233                 return vc4->job;
 234
 235         struct pipe_surface *cbuf = vc4->framebuffer.cbufs[0];
 236         struct pipe_surface *zsbuf = vc4->framebuffer.zsbuf;
 237         struct vc4_job *job = vc4_get_job(vc4, cbuf, zsbuf);
 238
 239         /* The dirty flags are tracking what's been updated while vc4->job has
 240          * been bound, so set them all to ~0 when switching between jobs.  We
 241          * also need to reset all state at the start of rendering.
 242          */
 243         vc4->dirty = ~0;
 244
 245         /* Set up the read surfaces in the job.  If they aren't actually
 246          * getting read (due to a clear starting the frame), job->cleared will
 247          * mask out the read.
 248          */
 249         pipe_surface_reference(&job->color_read, cbuf);
 250         pipe_surface_reference(&job->zs_read, zsbuf);
 251
 252         /* If we're binding to uninitialized buffers, no need to load their
 253          * contents before drawing.
 254          */
 255         if (cbuf) {
 256                 struct vc4_resource *rsc = vc4_resource(cbuf->texture);
 257                 if (!rsc->writes)
 258                         job->cleared |= PIPE_CLEAR_COLOR0;
 259         }
 260
 261         if (zsbuf) {
 262                 struct vc4_resource *rsc = vc4_resource(zsbuf->texture);
 263                 if (!rsc->writes)
 264                         job->cleared |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 265         }
 266
 267         job->draw_tiles_x = DIV_ROUND_UP(vc4->framebuffer.width,
 268                                          job->tile_width);
 269         job->draw_tiles_y = DIV_ROUND_UP(vc4->framebuffer.height,
 270                                          job->tile_height);
 271
 272         /* Initialize the job with the raster order flags -- each draw will
 273          * check that we haven't changed the flags, since that requires a
 274          * flush.
 275          */
 276         if (vc4->rasterizer)
 277                 job->flags = vc4->rasterizer->tile_raster_order_flags;
 278
 279         vc4->job = job;
 280
 281         return job;
 282 }
 283
 284 static void
 285 vc4_submit_setup_rcl_surface(struct vc4_job *job,
 286                              struct drm_vc4_submit_rcl_surface *submit_surf,
 287                              struct pipe_surface *psurf,
 288                              bool is_depth, bool is_write)
 289 {
 290         struct vc4_surface *surf = vc4_surface(psurf);
 291
 292         if (!surf)
 293                 return;
 294
 295         struct vc4_resource *rsc = vc4_resource(psurf->texture);
 296         submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
 297         submit_surf->offset = surf->offset;
 298
 299         if (psurf->texture->nr_samples <= 1) {
 300                 if (is_depth) {
 301                         submit_surf->bits =
 302                                 VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_ZS,
 303                                               VC4_LOADSTORE_TILE_BUFFER_BUFFER);
 304
 305                 } else {
 306                         submit_surf->bits =
 307                                 VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_COLOR,
 308                                               VC4_LOADSTORE_TILE_BUFFER_BUFFER) |
 309                                 VC4_SET_FIELD(vc4_rt_format_is_565(psurf->format) ?
 310                                               VC4_LOADSTORE_TILE_BUFFER_BGR565 :
 311                                               VC4_LOADSTORE_TILE_BUFFER_RGBA8888,
 312                                               VC4_LOADSTORE_TILE_BUFFER_FORMAT);
 313                 }
 314                 submit_surf->bits |=
 315                         VC4_SET_FIELD(surf->tiling,
 316                                       VC4_LOADSTORE_TILE_BUFFER_TILING);
 317         } else {
 318                 assert(!is_write);
 319                 submit_surf->flags |= VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES;
 320         }
 321
 322         if (is_write)
 323                 rsc->writes++;
 324 }
 325
 326 static void
 327 vc4_submit_setup_rcl_render_config_surface(struct vc4_job *job,
 328                                            struct drm_vc4_submit_rcl_surface *submit_surf,
 329                                            struct pipe_surface *psurf)
 330 {
 331         struct vc4_surface *surf = vc4_surface(psurf);
 332
 333         if (!surf)
 334                 return;
 335
 336         struct vc4_resource *rsc = vc4_resource(psurf->texture);
 337         submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
 338         submit_surf->offset = surf->offset;
 339
 340         if (psurf->texture->nr_samples <= 1) {
 341                 submit_surf->bits =
 342                         VC4_SET_FIELD(vc4_rt_format_is_565(surf->base.format) ?
 343                                       VC4_RENDER_CONFIG_FORMAT_BGR565 :
 344                                       VC4_RENDER_CONFIG_FORMAT_RGBA8888,
 345                                       VC4_RENDER_CONFIG_FORMAT) |
 346                         VC4_SET_FIELD(surf->tiling,
 347                                       VC4_RENDER_CONFIG_MEMORY_FORMAT);
 348         }
 349
 350         rsc->writes++;
 351 }
 352
 353 static void
 354 vc4_submit_setup_rcl_msaa_surface(struct vc4_job *job,
 355                                   struct drm_vc4_submit_rcl_surface *submit_surf,
 356                                   struct pipe_surface *psurf)
 357 {
 358         struct vc4_surface *surf = vc4_surface(psurf);
 359
 360         if (!surf)
 361                 return;
 362
 363         struct vc4_resource *rsc = vc4_resource(psurf->texture);
 364         submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
 365         submit_surf->offset = surf->offset;
 366         submit_surf->bits = 0;
 367         rsc->writes++;
 368 }
 369
 370 /**
 371  * Submits the job to the kernel and then reinitializes it.
 372  */
 373 void
 374 vc4_job_submit(struct vc4_context *vc4, struct vc4_job *job)
 375 {
 376         if (!job->needs_flush)
 377                 goto done;
 378
 379         /* The RCL setup would choke if the draw bounds cause no drawing, so
 380          * just drop the drawing if that's the case.
 381          */
 382         if (job->draw_max_x <= job->draw_min_x ||
 383             job->draw_max_y <= job->draw_min_y) {
 384                 goto done;
 385         }
 386
 387         if (vc4_debug & VC4_DEBUG_CL) {
 388                 fprintf(stderr, "BCL:\n");
 389                 vc4_dump_cl(job->bcl.base, cl_offset(&job->bcl), false);
 390         }
 391
 392         if (cl_offset(&job->bcl) > 0) {
 393                 /* Increment the semaphore indicating that binning is done and
 394                  * unblocking the render thread.  Note that this doesn't act
 395                  * until the FLUSH completes.
 396                  */
 397                 cl_ensure_space(&job->bcl, 8);
 398                 cl_emit(&job->bcl, INCREMENT_SEMAPHORE, incr);
 399                 /* The FLUSH caps all of our bin lists with a
 400                  * VC4_PACKET_RETURN.
 401                  */
 402                 cl_emit(&job->bcl, FLUSH, flush);
 403         }
 404         struct drm_vc4_submit_cl submit = {
 405                 .color_read.hindex = ~0,
 406                 .zs_read.hindex = ~0,
 407                 .color_write.hindex = ~0,
 408                 .msaa_color_write.hindex = ~0,
 409                 .zs_write.hindex = ~0,
 410                 .msaa_zs_write.hindex = ~0,
 411         };
 412
 413         cl_ensure_space(&job->bo_handles, 6 * sizeof(uint32_t));
 414         cl_ensure_space(&job->bo_pointers, 6 * sizeof(struct vc4_bo *));
 415
 416         if (job->resolve & PIPE_CLEAR_COLOR) {
 417                 if (!(job->cleared & PIPE_CLEAR_COLOR)) {
 418                         vc4_submit_setup_rcl_surface(job, &submit.color_read,
 419                                                      job->color_read,
 420                                                      false, false);
 421                 }
 422                 vc4_submit_setup_rcl_render_config_surface(job,
 423                                                            &submit.color_write,
 424                                                            job->color_write);
 425                 vc4_submit_setup_rcl_msaa_surface(job,
 426                                                   &submit.msaa_color_write,
 427                                                   job->msaa_color_write);
 428         }
 429         if (job->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
 430                 if (!(job->cleared & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
 431                         vc4_submit_setup_rcl_surface(job, &submit.zs_read,
 432                                                      job->zs_read, true, false);
 433                 }
 434                 vc4_submit_setup_rcl_surface(job, &submit.zs_write,
 435                                              job->zs_write, true, true);
 436                 vc4_submit_setup_rcl_msaa_surface(job, &submit.msaa_zs_write,
 437                                                   job->msaa_zs_write);
 438         }
 439
 440         if (job->msaa) {
 441                 /* This bit controls how many pixels the general
 442                  * (i.e. subsampled) loads/stores are iterating over
 443                  * (multisample loads replicate out to the other samples).
 444                  */
 445                 submit.color_write.bits |= VC4_RENDER_CONFIG_MS_MODE_4X;
 446                 /* Controls whether color_write's
 447                  * VC4_PACKET_STORE_MS_TILE_BUFFER does 4x decimation
 448                  */
 449                 submit.color_write.bits |= VC4_RENDER_CONFIG_DECIMATE_MODE_4X;
 450         }
 451
 452         submit.bo_handles = (uintptr_t)job->bo_handles.base;
 453         submit.bo_handle_count = cl_offset(&job->bo_handles) / 4;
 454         submit.bin_cl = (uintptr_t)job->bcl.base;
 455         submit.bin_cl_size = cl_offset(&job->bcl);
 456         submit.shader_rec = (uintptr_t)job->shader_rec.base;
 457         submit.shader_rec_size = cl_offset(&job->shader_rec);
 458         submit.shader_rec_count = job->shader_rec_count;
 459         submit.uniforms = (uintptr_t)job->uniforms.base;
 460         submit.uniforms_size = cl_offset(&job->uniforms);
 461         if (job->perfmon)
 462                 submit.perfmonid = job->perfmon->id;
 463
 464         assert(job->draw_min_x != ~0 && job->draw_min_y != ~0);
 465         submit.min_x_tile = job->draw_min_x / job->tile_width;
 466         submit.min_y_tile = job->draw_min_y / job->tile_height;
 467         submit.max_x_tile = (job->draw_max_x - 1) / job->tile_width;
 468         submit.max_y_tile = (job->draw_max_y - 1) / job->tile_height;
 469         submit.width = job->draw_width;
 470         submit.height = job->draw_height;
 471         if (job->cleared) {
 472                 submit.flags |= VC4_SUBMIT_CL_USE_CLEAR_COLOR;
 473                 submit.clear_color[0] = job->clear_color[0];
 474                 submit.clear_color[1] = job->clear_color[1];
 475                 submit.clear_z = job->clear_depth;
 476                 submit.clear_s = job->clear_stencil;
 477         }
 478         submit.flags |= job->flags;
 479
 480         if (!(vc4_debug & VC4_DEBUG_NORAST)) {
 481                 int ret;
 482
 483 #ifndef USE_VC4_SIMULATOR
 484                 ret = drmIoctl(vc4->fd, DRM_IOCTL_VC4_SUBMIT_CL, &submit);
 485 #else
 486                 ret = vc4_simulator_flush(vc4, &submit, job);
 487 #endif
 488                 static bool warned = false;
 489                 if (ret && !warned) {
 490                         fprintf(stderr, "Draw call returned %s.  "
 491                                         "Expect corruption.\n", strerror(errno));
 492                         warned = true;
 493                 } else if (!ret) {
 494                         vc4->last_emit_seqno = submit.seqno;
 495                         if (job->perfmon)
 496                                 job->perfmon->last_seqno = submit.seqno;
 497                 }
 498         }
 499
 500         if (vc4->last_emit_seqno - vc4->screen->finished_seqno > 5) {
 501                 if (!vc4_wait_seqno(vc4->screen,
 502                                     vc4->last_emit_seqno - 5,
 503                                     PIPE_TIMEOUT_INFINITE,
 504                                     "job throttling")) {
 505                         fprintf(stderr, "Job throttling failed\n");
 506                 }
 507         }
 508
 509         if (vc4_debug & VC4_DEBUG_ALWAYS_SYNC) {
 510                 if (!vc4_wait_seqno(vc4->screen, vc4->last_emit_seqno,
 511                                     PIPE_TIMEOUT_INFINITE, "sync")) {
 512                         fprintf(stderr, "Wait failed.\n");
 513                         abort();
 514                 }
 515         }
 516
 517 done:
 518         vc4_job_free(vc4, job);
 519 }
 520
 521 static bool
 522 vc4_job_compare(const void *a, const void *b)
 523 {
 524         return memcmp(a, b, sizeof(struct vc4_job_key)) == 0;
 525 }
 526
 527 static uint32_t
 528 vc4_job_hash(const void *key)
 529 {
 530         return _mesa_hash_data(key, sizeof(struct vc4_job_key));
 531 }
 532
 533 void
 534 vc4_job_init(struct vc4_context *vc4)
 535 {
 536         vc4->jobs = _mesa_hash_table_create(vc4,
 537                                             vc4_job_hash,
 538                                             vc4_job_compare);
 539         vc4->write_jobs = _mesa_hash_table_create(vc4,
 540                                                   _mesa_hash_pointer,
 541                                                   _mesa_key_pointer_equal);
 542 }
 543