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