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