src/gallium/drivers/vc5/vc5_job.c

   1 /*
   2  * Copyright © 2014-2017 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 vc5_job.c
  25  *
  26  * Functions for submitting VC5 render jobs to the kernel.
  27  */
  28
  29 #include <xf86drm.h>
  30 #include "vc5_context.h"
  31 /* The OQ/semaphore packets are the same across V3D versions. */
  32 #define V3D_VERSION 33
  33 #include "broadcom/cle/v3dx_pack.h"
  34 #include "broadcom/common/v3d_macros.h"
  35 #include "util/hash_table.h"
  36 #include "util/ralloc.h"
  37 #include "util/set.h"
  38 #include "broadcom/clif/clif_dump.h"
  39
  40 static void
  41 remove_from_ht(struct hash_table *ht, void *key)
  42 {
  43         struct hash_entry *entry = _mesa_hash_table_search(ht, key);
  44         _mesa_hash_table_remove(ht, entry);
  45 }
  46
  47 static void
  48 vc5_job_free(struct vc5_context *vc5, struct vc5_job *job)
  49 {
  50         struct set_entry *entry;
  51
  52         set_foreach(job->bos, entry) {
  53                 struct vc5_bo *bo = (struct vc5_bo *)entry->key;
  54                 vc5_bo_unreference(&bo);
  55         }
  56
  57         remove_from_ht(vc5->jobs, &job->key);
  58
  59         if (job->write_prscs) {
  60                 struct set_entry *entry;
  61
  62                 set_foreach(job->write_prscs, entry) {
  63                         const struct pipe_resource *prsc = entry->key;
  64
  65                         remove_from_ht(vc5->write_jobs, (void *)prsc);
  66                 }
  67         }
  68
  69         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
  70                 if (job->cbufs[i]) {
  71                         remove_from_ht(vc5->write_jobs, job->cbufs[i]->texture);
  72                         pipe_surface_reference(&job->cbufs[i], NULL);
  73                 }
  74         }
  75         if (job->zsbuf) {
  76                 remove_from_ht(vc5->write_jobs, job->zsbuf->texture);
  77                 pipe_surface_reference(&job->zsbuf, NULL);
  78         }
  79
  80         if (vc5->job == job)
  81                 vc5->job = NULL;
  82
  83         vc5_destroy_cl(&job->bcl);
  84         vc5_destroy_cl(&job->rcl);
  85         vc5_destroy_cl(&job->indirect);
  86         vc5_bo_unreference(&job->tile_alloc);
  87
  88         ralloc_free(job);
  89 }
  90
  91 static struct vc5_job *
  92 vc5_job_create(struct vc5_context *vc5)
  93 {
  94         struct vc5_job *job = rzalloc(vc5, struct vc5_job);
  95
  96         job->vc5 = vc5;
  97
  98         vc5_init_cl(job, &job->bcl);
  99         vc5_init_cl(job, &job->rcl);
 100         vc5_init_cl(job, &job->indirect);
 101
 102         job->draw_min_x = ~0;
 103         job->draw_min_y = ~0;
 104         job->draw_max_x = 0;
 105         job->draw_max_y = 0;
 106
 107         job->bos = _mesa_set_create(job,
 108                                     _mesa_hash_pointer,
 109                                     _mesa_key_pointer_equal);
 110         return job;
 111 }
 112
 113 void
 114 vc5_job_add_bo(struct vc5_job *job, struct vc5_bo *bo)
 115 {
 116         if (!bo)
 117                 return;
 118
 119         if (_mesa_set_search(job->bos, bo))
 120                 return;
 121
 122         vc5_bo_reference(bo);
 123         _mesa_set_add(job->bos, bo);
 124         job->referenced_size += bo->size;
 125
 126         uint32_t *bo_handles = (void *)(uintptr_t)job->submit.bo_handles;
 127
 128         if (job->submit.bo_handle_count >= job->bo_handles_size) {
 129                 job->bo_handles_size = MAX2(4, job->bo_handles_size * 2);
 130                 bo_handles = reralloc(job, bo_handles,
 131                                       uint32_t, job->bo_handles_size);
 132                 job->submit.bo_handles = (uintptr_t)(void *)bo_handles;
 133         }
 134         bo_handles[job->submit.bo_handle_count++] = bo->handle;
 135 }
 136
 137 void
 138 vc5_job_add_write_resource(struct vc5_job *job, struct pipe_resource *prsc)
 139 {
 140         struct vc5_context *vc5 = job->vc5;
 141
 142         if (!job->write_prscs) {
 143                 job->write_prscs = _mesa_set_create(job,
 144                                                     _mesa_hash_pointer,
 145                                                     _mesa_key_pointer_equal);
 146         }
 147
 148         _mesa_set_add(job->write_prscs, prsc);
 149         _mesa_hash_table_insert(vc5->write_jobs, prsc, job);
 150 }
 151
 152 void
 153 vc5_flush_jobs_writing_resource(struct vc5_context *vc5,
 154                                 struct pipe_resource *prsc)
 155 {
 156         struct hash_entry *entry = _mesa_hash_table_search(vc5->write_jobs,
 157                                                            prsc);
 158         if (entry) {
 159                 struct vc5_job *job = entry->data;
 160                 vc5_job_submit(vc5, job);
 161         }
 162 }
 163
 164 void
 165 vc5_flush_jobs_reading_resource(struct vc5_context *vc5,
 166                                 struct pipe_resource *prsc)
 167 {
 168         struct vc5_resource *rsc = vc5_resource(prsc);
 169
 170         vc5_flush_jobs_writing_resource(vc5, prsc);
 171
 172         struct hash_entry *entry;
 173         hash_table_foreach(vc5->jobs, entry) {
 174                 struct vc5_job *job = entry->data;
 175
 176                 if (_mesa_set_search(job->bos, rsc->bo)) {
 177                         vc5_job_submit(vc5, job);
 178                         /* Reminder: vc5->jobs is safe to keep iterating even
 179                          * after deletion of an entry.
 180                          */
 181                         continue;
 182                 }
 183         }
 184 }
 185
 186 static void
 187 vc5_job_set_tile_buffer_size(struct vc5_job *job)
 188 {
 189         static const uint8_t tile_sizes[] = {
 190                 64, 64,
 191                 64, 32,
 192                 32, 32,
 193                 32, 16,
 194                 16, 16,
 195         };
 196         int tile_size_index = 0;
 197         if (job->msaa)
 198                 tile_size_index += 2;
 199
 200         if (job->cbufs[3] || job->cbufs[2])
 201                 tile_size_index += 2;
 202         else if (job->cbufs[1])
 203                 tile_size_index++;
 204
 205         int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
 206         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 207                 if (job->cbufs[i]) {
 208                         struct vc5_surface *surf = vc5_surface(job->cbufs[i]);
 209                         max_bpp = MAX2(max_bpp, surf->internal_bpp);
 210                 }
 211         }
 212         job->internal_bpp = max_bpp;
 213         STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
 214         tile_size_index += max_bpp;
 215
 216         assert(tile_size_index < ARRAY_SIZE(tile_sizes));
 217         job->tile_width = tile_sizes[tile_size_index * 2 + 0];
 218         job->tile_height = tile_sizes[tile_size_index * 2 + 1];
 219 }
 220
 221 /**
 222  * Returns a vc5_job struture for tracking V3D rendering to a particular FBO.
 223  *
 224  * If we've already started rendering to this FBO, then return old same job,
 225  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 226  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 227  * have been flushed.
 228  */
 229 struct vc5_job *
 230 vc5_get_job(struct vc5_context *vc5,
 231             struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
 232 {
 233         /* Return the existing job for this FBO if we have one */
 234         struct vc5_job_key local_key = {
 235                 .cbufs = {
 236                         cbufs[0],
 237                         cbufs[1],
 238                         cbufs[2],
 239                         cbufs[3],
 240                 },
 241                 .zsbuf = zsbuf,
 242         };
 243         struct hash_entry *entry = _mesa_hash_table_search(vc5->jobs,
 244                                                            &local_key);
 245         if (entry)
 246                 return entry->data;
 247
 248         /* Creating a new job.  Make sure that any previous jobs reading or
 249          * writing these buffers are flushed.
 250          */
 251         struct vc5_job *job = vc5_job_create(vc5);
 252
 253         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 254                 if (cbufs[i]) {
 255                         vc5_flush_jobs_reading_resource(vc5, cbufs[i]->texture);
 256                         pipe_surface_reference(&job->cbufs[i], cbufs[i]);
 257
 258                         if (cbufs[i]->texture->nr_samples > 1)
 259                                 job->msaa = true;
 260                 }
 261         }
 262         if (zsbuf) {
 263                 vc5_flush_jobs_reading_resource(vc5, zsbuf->texture);
 264                 pipe_surface_reference(&job->zsbuf, zsbuf);
 265                 if (zsbuf->texture->nr_samples > 1)
 266                         job->msaa = true;
 267         }
 268
 269         vc5_job_set_tile_buffer_size(job);
 270
 271         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 272                 if (cbufs[i])
 273                         _mesa_hash_table_insert(vc5->write_jobs,
 274                                                 cbufs[i]->texture, job);
 275         }
 276         if (zsbuf)
 277                 _mesa_hash_table_insert(vc5->write_jobs, zsbuf->texture, job);
 278
 279         memcpy(&job->key, &local_key, sizeof(local_key));
 280         _mesa_hash_table_insert(vc5->jobs, &job->key, job);
 281
 282         return job;
 283 }
 284
 285 struct vc5_job *
 286 vc5_get_job_for_fbo(struct vc5_context *vc5)
 287 {
 288         if (vc5->job)
 289                 return vc5->job;
 290
 291         struct pipe_surface **cbufs = vc5->framebuffer.cbufs;
 292         struct pipe_surface *zsbuf = vc5->framebuffer.zsbuf;
 293         struct vc5_job *job = vc5_get_job(vc5, cbufs, zsbuf);
 294
 295         /* The dirty flags are tracking what's been updated while vc5->job has
 296          * been bound, so set them all to ~0 when switching between jobs.  We
 297          * also need to reset all state at the start of rendering.
 298          */
 299         vc5->dirty = ~0;
 300
 301         /* If we're binding to uninitialized buffers, no need to load their
 302          * contents before drawing.
 303          */
 304         for (int i = 0; i < 4; i++) {
 305                 if (cbufs[i]) {
 306                         struct vc5_resource *rsc = vc5_resource(cbufs[i]->texture);
 307                         if (!rsc->writes)
 308                                 job->cleared |= PIPE_CLEAR_COLOR0 << i;
 309                 }
 310         }
 311
 312         if (zsbuf) {
 313                 struct vc5_resource *rsc = vc5_resource(zsbuf->texture);
 314                 if (!rsc->writes)
 315                         job->cleared |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 316         }
 317
 318         job->draw_tiles_x = DIV_ROUND_UP(vc5->framebuffer.width,
 319                                          job->tile_width);
 320         job->draw_tiles_y = DIV_ROUND_UP(vc5->framebuffer.height,
 321                                          job->tile_height);
 322
 323         vc5->job = job;
 324
 325         return job;
 326 }
 327
 328 static bool
 329 vc5_clif_dump_lookup(void *data, uint32_t addr, void **vaddr)
 330 {
 331         struct vc5_job *job = data;
 332         struct set_entry *entry;
 333
 334         set_foreach(job->bos, entry) {
 335                 struct vc5_bo *bo = (void *)entry->key;
 336
 337                 if (addr >= bo->offset &&
 338                     addr < bo->offset + bo->size) {
 339                         vc5_bo_map(bo);
 340                         *vaddr = bo->map + addr - bo->offset;
 341                         return true;
 342                 }
 343         }
 344
 345         return false;
 346 }
 347
 348 static void
 349 vc5_clif_dump(struct vc5_context *vc5, struct vc5_job *job)
 350 {
 351         if (!(V3D_DEBUG & V3D_DEBUG_CL))
 352                 return;
 353
 354         struct clif_dump *clif = clif_dump_init(&vc5->screen->devinfo,
 355                                                 stderr, vc5_clif_dump_lookup,
 356                                                 job);
 357
 358         fprintf(stderr, "BCL: 0x%08x..0x%08x\n",
 359                 job->submit.bcl_start, job->submit.bcl_end);
 360
 361         clif_dump_add_cl(clif, job->submit.bcl_start, job->submit.bcl_end);
 362
 363         fprintf(stderr, "RCL: 0x%08x..0x%08x\n",
 364                 job->submit.rcl_start, job->submit.rcl_end);
 365         clif_dump_add_cl(clif, job->submit.rcl_start, job->submit.rcl_end);
 366 }
 367
 368 /**
 369  * Submits the job to the kernel and then reinitializes it.
 370  */
 371 void
 372 vc5_job_submit(struct vc5_context *vc5, struct vc5_job *job)
 373 {
 374         if (!job->needs_flush)
 375                 goto done;
 376
 377         /* The RCL setup would choke if the draw bounds cause no drawing, so
 378          * just drop the drawing if that's the case.
 379          */
 380         if (job->draw_max_x <= job->draw_min_x ||
 381             job->draw_max_y <= job->draw_min_y) {
 382                 goto done;
 383         }
 384
 385         vc5_emit_rcl(job);
 386
 387         if (cl_offset(&job->bcl) > 0) {
 388                 vc5_cl_ensure_space_with_branch(&job->bcl,
 389                                                 7 +
 390                                                 cl_packet_length(OCCLUSION_QUERY_COUNTER));
 391
 392                 if (job->oq_enabled) {
 393                         /* Disable the OQ at the end of the CL, so that the
 394                          * draw calls at the start of the CL don't inherit the
 395                          * OQ counter.
 396                          */
 397                         cl_emit(&job->bcl, OCCLUSION_QUERY_COUNTER, counter);
 398                 }
 399
 400                 /* Increment the semaphore indicating that binning is done and
 401                  * unblocking the render thread.  Note that this doesn't act
 402                  * until the FLUSH completes.
 403                  */
 404                 cl_emit(&job->bcl, INCREMENT_SEMAPHORE, incr);
 405
 406                 /* The FLUSH_ALL emits any unwritten state changes in each
 407                  * tile.  We can use this to reset any state that needs to be
 408                  * present at the start of the next tile, as we do with
 409                  * OCCLUSION_QUERY_COUNTER above.
 410                  */
 411                 cl_emit(&job->bcl, FLUSH_ALL_STATE, flush);
 412         }
 413
 414         job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
 415         job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);
 416
 417         vc5_clif_dump(vc5, job);
 418
 419         if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
 420                 int ret;
 421
 422 #ifndef USE_VC5_SIMULATOR
 423                 ret = drmIoctl(vc5->fd, DRM_IOCTL_VC5_SUBMIT_CL, &job->submit);
 424 #else
 425                 ret = vc5_simulator_flush(vc5, &job->submit, job);
 426 #endif
 427                 static bool warned = false;
 428                 if (ret && !warned) {
 429                         fprintf(stderr, "Draw call returned %s.  "
 430                                         "Expect corruption.\n", strerror(errno));
 431                         warned = true;
 432                 }
 433         }
 434
 435         if (vc5->last_emit_seqno - vc5->screen->finished_seqno > 5) {
 436                 if (!vc5_wait_seqno(vc5->screen,
 437                                     vc5->last_emit_seqno - 5,
 438                                     PIPE_TIMEOUT_INFINITE,
 439                                     "job throttling")) {
 440                         fprintf(stderr, "Job throttling failed\n");
 441                 }
 442         }
 443
 444 done:
 445         vc5_job_free(vc5, job);
 446 }
 447
 448 static bool
 449 vc5_job_compare(const void *a, const void *b)
 450 {
 451         return memcmp(a, b, sizeof(struct vc5_job_key)) == 0;
 452 }
 453
 454 static uint32_t
 455 vc5_job_hash(const void *key)
 456 {
 457         return _mesa_hash_data(key, sizeof(struct vc5_job_key));
 458 }
 459
 460 void
 461 vc5_job_init(struct vc5_context *vc5)
 462 {
 463         vc5->jobs = _mesa_hash_table_create(vc5,
 464                                             vc5_job_hash,
 465                                             vc5_job_compare);
 466         vc5->write_jobs = _mesa_hash_table_create(vc5,
 467                                                   _mesa_hash_pointer,
 468                                                   _mesa_key_pointer_equal);
 469 }
 470