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