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_context.h"
  31 #include "util/hash_table.h"
  32
  33 static void
  34 remove_from_ht(struct hash_table *ht, void *key)
  35 {
  36         struct hash_entry *entry = _mesa_hash_table_search(ht, key);
  37         _mesa_hash_table_remove(ht, entry);
  38 }
  39
  40 static void
  41 vc4_job_free(struct vc4_context *vc4, struct vc4_job *job)
  42 {
  43         struct vc4_bo **referenced_bos = job->bo_pointers.base;
  44         for (int i = 0; i < cl_offset(&job->bo_handles) / 4; i++) {
  45                 vc4_bo_unreference(&referenced_bos[i]);
  46         }
  47
  48         remove_from_ht(vc4->jobs, &job->key);
  49
  50         if (job->color_write) {
  51                 remove_from_ht(vc4->write_jobs, job->color_write->texture);
  52                 pipe_surface_reference(&job->color_write, NULL);
  53         }
  54         if (job->msaa_color_write) {
  55                 remove_from_ht(vc4->write_jobs, job->msaa_color_write->texture);
  56                 pipe_surface_reference(&job->msaa_color_write, NULL);
  57         }
  58         if (job->zs_write) {
  59                 remove_from_ht(vc4->write_jobs, job->zs_write->texture);
  60                 pipe_surface_reference(&job->zs_write, NULL);
  61         }
  62         if (job->msaa_zs_write) {
  63                 remove_from_ht(vc4->write_jobs, job->msaa_zs_write->texture);
  64                 pipe_surface_reference(&job->msaa_zs_write, NULL);
  65         }
  66
  67         pipe_surface_reference(&job->color_read, NULL);
  68         pipe_surface_reference(&job->zs_read, NULL);
  69
  70         if (vc4->job == job)
  71                 vc4->job = NULL;
  72
  73         ralloc_free(job);
  74 }
  75
  76 static struct vc4_job *
  77 vc4_job_create(struct vc4_context *vc4)
  78 {
  79         struct vc4_job *job = rzalloc(vc4, struct vc4_job);
  80
  81         vc4_init_cl(job, &job->bcl);
  82         vc4_init_cl(job, &job->shader_rec);
  83         vc4_init_cl(job, &job->uniforms);
  84         vc4_init_cl(job, &job->bo_handles);
  85         vc4_init_cl(job, &job->bo_pointers);
  86
  87         job->draw_min_x = ~0;
  88         job->draw_min_y = ~0;
  89         job->draw_max_x = 0;
  90         job->draw_max_y = 0;
  91
  92         return job;
  93 }
  94
  95 void
  96 vc4_flush_jobs_writing_resource(struct vc4_context *vc4,
  97                                 struct pipe_resource *prsc)
  98 {
  99         struct hash_entry *entry = _mesa_hash_table_search(vc4->write_jobs,
 100                                                            prsc);
 101         if (entry) {
 102                 struct vc4_job *job = entry->data;
 103                 vc4_job_submit(vc4, job);
 104         }
 105 }
 106
 107 void
 108 vc4_flush_jobs_reading_resource(struct vc4_context *vc4,
 109                                 struct pipe_resource *prsc)
 110 {
 111         struct vc4_resource *rsc = vc4_resource(prsc);
 112
 113         vc4_flush_jobs_writing_resource(vc4, prsc);
 114
 115         struct hash_entry *entry;
 116         hash_table_foreach(vc4->jobs, entry) {
 117                 struct vc4_job *job = entry->data;
 118
 119                 struct vc4_bo **referenced_bos = job->bo_pointers.base;
 120                 for (int i = 0; i < cl_offset(&job->bo_handles) / 4; i++) {
 121                         if (referenced_bos[i] == rsc->bo) {
 122                                 vc4_job_submit(vc4, job);
 123                                 continue;
 124                         }
 125                 }
 126
 127                 /* Also check for the Z/color buffers, since the references to
 128                  * those are only added immediately before submit.
 129                  */
 130                 if (job->color_read && !(job->cleared & PIPE_CLEAR_COLOR)) {
 131                         struct vc4_resource *ctex =
 132                                 vc4_resource(job->color_read->texture);
 133                         if (ctex->bo == rsc->bo) {
 134                                 vc4_job_submit(vc4, job);
 135                                 continue;
 136                         }
 137                 }
 138
 139                 if (job->zs_read && !(job->cleared &
 140                                       (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
 141                         struct vc4_resource *ztex =
 142                                 vc4_resource(job->zs_read->texture);
 143                         if (ztex->bo == rsc->bo) {
 144                                 vc4_job_submit(vc4, job);
 145                                 continue;
 146                         }
 147                 }
 148         }
 149 }
 150
 151 /**
 152  * Returns a vc4_job struture for tracking V3D rendering to a particular FBO.
 153  *
 154  * If we've already started rendering to this FBO, then return old same job,
 155  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 156  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 157  * have been flushed.
 158  */
 159 struct vc4_job *
 160 vc4_get_job(struct vc4_context *vc4,
 161             struct pipe_surface *cbuf, struct pipe_surface *zsbuf)
 162 {
 163         /* Return the existing job for this FBO if we have one */
 164         struct vc4_job_key local_key = {.cbuf = cbuf, .zsbuf = zsbuf};
 165         struct hash_entry *entry = _mesa_hash_table_search(vc4->jobs,
 166                                                            &local_key);
 167         if (entry)
 168                 return entry->data;
 169
 170         /* Creating a new job.  Make sure that any previous jobs reading or
 171          * writing these buffers are flushed.
 172          */
 173         if (cbuf)
 174                 vc4_flush_jobs_reading_resource(vc4, cbuf->texture);
 175         if (zsbuf)
 176                 vc4_flush_jobs_reading_resource(vc4, zsbuf->texture);
 177
 178         struct vc4_job *job = vc4_job_create(vc4);
 179
 180         if (cbuf) {
 181                 if (cbuf->texture->nr_samples > 1) {
 182                         job->msaa = true;
 183                         pipe_surface_reference(&job->msaa_color_write, cbuf);
 184                 } else {
 185                         pipe_surface_reference(&job->color_write, cbuf);
 186                 }
 187         }
 188
 189         if (zsbuf) {
 190                 if (zsbuf->texture->nr_samples > 1) {
 191                         job->msaa = true;
 192                         pipe_surface_reference(&job->msaa_zs_write, zsbuf);
 193                 } else {
 194                         pipe_surface_reference(&job->zs_write, zsbuf);
 195                 }
 196         }
 197
 198         if (job->msaa) {
 199                 job->tile_width = 32;
 200                 job->tile_height = 32;
 201         } else {
 202                 job->tile_width = 64;
 203                 job->tile_height = 64;
 204         }
 205
 206         if (cbuf)
 207                 _mesa_hash_table_insert(vc4->write_jobs, cbuf->texture, job);
 208         if (zsbuf)
 209                 _mesa_hash_table_insert(vc4->write_jobs, zsbuf->texture, job);
 210
 211         job->key.cbuf = cbuf;
 212         job->key.zsbuf = zsbuf;
 213         _mesa_hash_table_insert(vc4->jobs, &job->key, job);
 214
 215         return job;
 216 }
 217
 218 struct vc4_job *
 219 vc4_get_job_for_fbo(struct vc4_context *vc4)
 220 {
 221         if (vc4->job)
 222                 return vc4->job;
 223
 224         struct pipe_surface *cbuf = vc4->framebuffer.cbufs[0];
 225         struct pipe_surface *zsbuf = vc4->framebuffer.zsbuf;
 226         struct vc4_job *job = vc4_get_job(vc4, cbuf, zsbuf);
 227
 228         /* The dirty flags are tracking what's been updated while vc4->job has
 229          * been bound, so set them all to ~0 when switching between jobs.  We
 230          * also need to reset all state at the start of rendering.
 231          */
 232         vc4->dirty = ~0;
 233
 234         /* Set up the read surfaces in the job.  If they aren't actually
 235          * getting read (due to a clear starting the frame), job->cleared will
 236          * mask out the read.
 237          */
 238         pipe_surface_reference(&job->color_read, cbuf);
 239         pipe_surface_reference(&job->zs_read, zsbuf);
 240
 241         /* If we're binding to uninitialized buffers, no need to load their
 242          * contents before drawing.
 243          */
 244         if (cbuf) {
 245                 struct vc4_resource *rsc = vc4_resource(cbuf->texture);
 246                 if (!rsc->writes)
 247                         job->cleared |= PIPE_CLEAR_COLOR0;
 248         }
 249
 250         if (zsbuf) {
 251                 struct vc4_resource *rsc = vc4_resource(zsbuf->texture);
 252                 if (!rsc->writes)
 253                         job->cleared |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 254         }
 255
 256         job->draw_tiles_x = DIV_ROUND_UP(vc4->framebuffer.width,
 257                                          job->tile_width);
 258         job->draw_tiles_y = DIV_ROUND_UP(vc4->framebuffer.height,
 259                                          job->tile_height);
 260
 261         vc4->job = job;
 262
 263         return job;
 264 }
 265
 266 static void
 267 vc4_submit_setup_rcl_surface(struct vc4_job *job,
 268                              struct drm_vc4_submit_rcl_surface *submit_surf,
 269                              struct pipe_surface *psurf,
 270                              bool is_depth, bool is_write)
 271 {
 272         struct vc4_surface *surf = vc4_surface(psurf);
 273
 274         if (!surf)
 275                 return;
 276
 277         struct vc4_resource *rsc = vc4_resource(psurf->texture);
 278         submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
 279         submit_surf->offset = surf->offset;
 280
 281         if (psurf->texture->nr_samples <= 1) {
 282                 if (is_depth) {
 283                         submit_surf->bits =
 284                                 VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_ZS,
 285                                               VC4_LOADSTORE_TILE_BUFFER_BUFFER);
 286
 287                 } else {
 288                         submit_surf->bits =
 289                                 VC4_SET_FIELD(VC4_LOADSTORE_TILE_BUFFER_COLOR,
 290                                               VC4_LOADSTORE_TILE_BUFFER_BUFFER) |
 291                                 VC4_SET_FIELD(vc4_rt_format_is_565(psurf->format) ?
 292                                               VC4_LOADSTORE_TILE_BUFFER_BGR565 :
 293                                               VC4_LOADSTORE_TILE_BUFFER_RGBA8888,
 294                                               VC4_LOADSTORE_TILE_BUFFER_FORMAT);
 295                 }
 296                 submit_surf->bits |=
 297                         VC4_SET_FIELD(surf->tiling,
 298                                       VC4_LOADSTORE_TILE_BUFFER_TILING);
 299         } else {
 300                 assert(!is_write);
 301                 submit_surf->flags |= VC4_SUBMIT_RCL_SURFACE_READ_IS_FULL_RES;
 302         }
 303
 304         if (is_write)
 305                 rsc->writes++;
 306 }
 307
 308 static void
 309 vc4_submit_setup_rcl_render_config_surface(struct vc4_job *job,
 310                                            struct drm_vc4_submit_rcl_surface *submit_surf,
 311                                            struct pipe_surface *psurf)
 312 {
 313         struct vc4_surface *surf = vc4_surface(psurf);
 314
 315         if (!surf)
 316                 return;
 317
 318         struct vc4_resource *rsc = vc4_resource(psurf->texture);
 319         submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
 320         submit_surf->offset = surf->offset;
 321
 322         if (psurf->texture->nr_samples <= 1) {
 323                 submit_surf->bits =
 324                         VC4_SET_FIELD(vc4_rt_format_is_565(surf->base.format) ?
 325                                       VC4_RENDER_CONFIG_FORMAT_BGR565 :
 326                                       VC4_RENDER_CONFIG_FORMAT_RGBA8888,
 327                                       VC4_RENDER_CONFIG_FORMAT) |
 328                         VC4_SET_FIELD(surf->tiling,
 329                                       VC4_RENDER_CONFIG_MEMORY_FORMAT);
 330         }
 331
 332         rsc->writes++;
 333 }
 334
 335 static void
 336 vc4_submit_setup_rcl_msaa_surface(struct vc4_job *job,
 337                                   struct drm_vc4_submit_rcl_surface *submit_surf,
 338                                   struct pipe_surface *psurf)
 339 {
 340         struct vc4_surface *surf = vc4_surface(psurf);
 341
 342         if (!surf)
 343                 return;
 344
 345         struct vc4_resource *rsc = vc4_resource(psurf->texture);
 346         submit_surf->hindex = vc4_gem_hindex(job, rsc->bo);
 347         submit_surf->offset = surf->offset;
 348         submit_surf->bits = 0;
 349         rsc->writes++;
 350 }
 351
 352 /**
 353  * Submits the job to the kernel and then reinitializes it.
 354  */
 355 void
 356 vc4_job_submit(struct vc4_context *vc4, struct vc4_job *job)
 357 {
 358         if (!job->needs_flush)
 359                 goto done;
 360
 361         /* The RCL setup would choke if the draw bounds cause no drawing, so
 362          * just drop the drawing if that's the case.
 363          */
 364         if (job->draw_max_x <= job->draw_min_x ||
 365             job->draw_max_y <= job->draw_min_y) {
 366                 goto done;
 367         }
 368
 369         if (vc4_debug & VC4_DEBUG_CL) {
 370                 fprintf(stderr, "BCL:\n");
 371                 vc4_dump_cl(job->bcl.base, cl_offset(&job->bcl), false);
 372         }
 373
 374         if (cl_offset(&job->bcl) > 0) {
 375                 /* Increment the semaphore indicating that binning is done and
 376                  * unblocking the render thread.  Note that this doesn't act
 377                  * until the FLUSH completes.
 378                  */
 379                 cl_ensure_space(&job->bcl, 8);
 380                 struct vc4_cl_out *bcl = cl_start(&job->bcl);
 381                 cl_u8(&bcl, VC4_PACKET_INCREMENT_SEMAPHORE);
 382                 /* The FLUSH caps all of our bin lists with a
 383                  * VC4_PACKET_RETURN.
 384                  */
 385                 cl_u8(&bcl, VC4_PACKET_FLUSH);
 386                 cl_end(&job->bcl, bcl);
 387         }
 388         struct drm_vc4_submit_cl submit = {
 389                 .color_read.hindex = ~0,
 390                 .zs_read.hindex = ~0,
 391                 .color_write.hindex = ~0,
 392                 .msaa_color_write.hindex = ~0,
 393                 .zs_write.hindex = ~0,
 394                 .msaa_zs_write.hindex = ~0,
 395         };
 396
 397         cl_ensure_space(&job->bo_handles, 6 * sizeof(uint32_t));
 398         cl_ensure_space(&job->bo_pointers, 6 * sizeof(struct vc4_bo *));
 399
 400         if (job->resolve & PIPE_CLEAR_COLOR) {
 401                 if (!(job->cleared & PIPE_CLEAR_COLOR)) {
 402                         vc4_submit_setup_rcl_surface(job, &submit.color_read,
 403                                                      job->color_read,
 404                                                      false, false);
 405                 }
 406                 vc4_submit_setup_rcl_render_config_surface(job,
 407                                                            &submit.color_write,
 408                                                            job->color_write);
 409                 vc4_submit_setup_rcl_msaa_surface(job,
 410                                                   &submit.msaa_color_write,
 411                                                   job->msaa_color_write);
 412         }
 413         if (job->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) {
 414                 if (!(job->cleared & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
 415                         vc4_submit_setup_rcl_surface(job, &submit.zs_read,
 416                                                      job->zs_read, true, false);
 417                 }
 418                 vc4_submit_setup_rcl_surface(job, &submit.zs_write,
 419                                              job->zs_write, true, true);
 420                 vc4_submit_setup_rcl_msaa_surface(job, &submit.msaa_zs_write,
 421                                                   job->msaa_zs_write);
 422         }
 423
 424         if (job->msaa) {
 425                 /* This bit controls how many pixels the general
 426                  * (i.e. subsampled) loads/stores are iterating over
 427                  * (multisample loads replicate out to the other samples).
 428                  */
 429                 submit.color_write.bits |= VC4_RENDER_CONFIG_MS_MODE_4X;
 430                 /* Controls whether color_write's
 431                  * VC4_PACKET_STORE_MS_TILE_BUFFER does 4x decimation
 432                  */
 433                 submit.color_write.bits |= VC4_RENDER_CONFIG_DECIMATE_MODE_4X;
 434         }
 435
 436         submit.bo_handles = (uintptr_t)job->bo_handles.base;
 437         submit.bo_handle_count = cl_offset(&job->bo_handles) / 4;
 438         submit.bin_cl = (uintptr_t)job->bcl.base;
 439         submit.bin_cl_size = cl_offset(&job->bcl);
 440         submit.shader_rec = (uintptr_t)job->shader_rec.base;
 441         submit.shader_rec_size = cl_offset(&job->shader_rec);
 442         submit.shader_rec_count = job->shader_rec_count;
 443         submit.uniforms = (uintptr_t)job->uniforms.base;
 444         submit.uniforms_size = cl_offset(&job->uniforms);
 445
 446         assert(job->draw_min_x != ~0 && job->draw_min_y != ~0);
 447         submit.min_x_tile = job->draw_min_x / job->tile_width;
 448         submit.min_y_tile = job->draw_min_y / job->tile_height;
 449         submit.max_x_tile = (job->draw_max_x - 1) / job->tile_width;
 450         submit.max_y_tile = (job->draw_max_y - 1) / job->tile_height;
 451         submit.width = job->draw_width;
 452         submit.height = job->draw_height;
 453         if (job->cleared) {
 454                 submit.flags |= VC4_SUBMIT_CL_USE_CLEAR_COLOR;
 455                 submit.clear_color[0] = job->clear_color[0];
 456                 submit.clear_color[1] = job->clear_color[1];
 457                 submit.clear_z = job->clear_depth;
 458                 submit.clear_s = job->clear_stencil;
 459         }
 460
 461         if (!(vc4_debug & VC4_DEBUG_NORAST)) {
 462                 int ret;
 463
 464 #ifndef USE_VC4_SIMULATOR
 465                 ret = drmIoctl(vc4->fd, DRM_IOCTL_VC4_SUBMIT_CL, &submit);
 466 #else
 467                 ret = vc4_simulator_flush(vc4, &submit, job);
 468 #endif
 469                 static bool warned = false;
 470                 if (ret && !warned) {
 471                         fprintf(stderr, "Draw call returned %s.  "
 472                                         "Expect corruption.\n", strerror(errno));
 473                         warned = true;
 474                 } else if (!ret) {
 475                         vc4->last_emit_seqno = submit.seqno;
 476                 }
 477         }
 478
 479         if (vc4->last_emit_seqno - vc4->screen->finished_seqno > 5) {
 480                 if (!vc4_wait_seqno(vc4->screen,
 481                                     vc4->last_emit_seqno - 5,
 482                                     PIPE_TIMEOUT_INFINITE,
 483                                     "job throttling")) {
 484                         fprintf(stderr, "Job throttling failed\n");
 485                 }
 486         }
 487
 488         if (vc4_debug & VC4_DEBUG_ALWAYS_SYNC) {
 489                 if (!vc4_wait_seqno(vc4->screen, vc4->last_emit_seqno,
 490                                     PIPE_TIMEOUT_INFINITE, "sync")) {
 491                         fprintf(stderr, "Wait failed.\n");
 492                         abort();
 493                 }
 494         }
 495
 496 done:
 497         vc4_job_free(vc4, job);
 498 }
 499
 500 static bool
 501 vc4_job_compare(const void *a, const void *b)
 502 {
 503         return memcmp(a, b, sizeof(struct vc4_job_key)) == 0;
 504 }
 505
 506 static uint32_t
 507 vc4_job_hash(const void *key)
 508 {
 509         return _mesa_hash_data(key, sizeof(struct vc4_job_key));
 510 }
 511
 512 void
 513 vc4_job_init(struct vc4_context *vc4)
 514 {
 515         vc4->jobs = _mesa_hash_table_create(vc4,
 516                                             vc4_job_hash,
 517                                             vc4_job_compare);
 518         vc4->write_jobs = _mesa_hash_table_create(vc4,
 519                                                   _mesa_hash_pointer,
 520                                                   _mesa_key_pointer_equal);
 521 }
 522