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 #include "util/hash_table.h"
  32 #include "util/ralloc.h"
  33 #include "util/set.h"
  34 #include "broadcom/clif/clif_dump.h"
  35 #include "broadcom/cle/v3d_packet_v33_pack.h"
  36
  37 static void
  38 remove_from_ht(struct hash_table *ht, void *key)
  39 {
  40         struct hash_entry *entry = _mesa_hash_table_search(ht, key);
  41         _mesa_hash_table_remove(ht, entry);
  42 }
  43
  44 static void
  45 vc5_job_free(struct vc5_context *vc5, struct vc5_job *job)
  46 {
  47         struct set_entry *entry;
  48
  49         set_foreach(job->bos, entry) {
  50                 struct vc5_bo *bo = (struct vc5_bo *)entry->key;
  51                 vc5_bo_unreference(&bo);
  52         }
  53
  54         remove_from_ht(vc5->jobs, &job->key);
  55
  56         if (job->write_prscs) {
  57                 struct set_entry *entry;
  58
  59                 set_foreach(job->write_prscs, entry) {
  60                         const struct pipe_resource *prsc = entry->key;
  61
  62                         remove_from_ht(vc5->write_jobs, (void *)prsc);
  63                 }
  64         }
  65
  66         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
  67                 if (job->cbufs[i]) {
  68                         remove_from_ht(vc5->write_jobs, job->cbufs[i]->texture);
  69                         pipe_surface_reference(&job->cbufs[i], NULL);
  70                 }
  71         }
  72         if (job->zsbuf) {
  73                 remove_from_ht(vc5->write_jobs, job->zsbuf->texture);
  74                 pipe_surface_reference(&job->zsbuf, NULL);
  75         }
  76
  77         if (vc5->job == job)
  78                 vc5->job = NULL;
  79
  80         vc5_destroy_cl(&job->bcl);
  81         vc5_destroy_cl(&job->rcl);
  82         vc5_destroy_cl(&job->indirect);
  83         vc5_bo_unreference(&job->tile_alloc);
  84
  85         ralloc_free(job);
  86 }
  87
  88 static struct vc5_job *
  89 vc5_job_create(struct vc5_context *vc5)
  90 {
  91         struct vc5_job *job = rzalloc(vc5, struct vc5_job);
  92
  93         job->vc5 = vc5;
  94
  95         vc5_init_cl(job, &job->bcl);
  96         vc5_init_cl(job, &job->rcl);
  97         vc5_init_cl(job, &job->indirect);
  98
  99         job->draw_min_x = ~0;
 100         job->draw_min_y = ~0;
 101         job->draw_max_x = 0;
 102         job->draw_max_y = 0;
 103
 104         job->bos = _mesa_set_create(job,
 105                                     _mesa_hash_pointer,
 106                                     _mesa_key_pointer_equal);
 107         return job;
 108 }
 109
 110 void
 111 vc5_job_add_bo(struct vc5_job *job, struct vc5_bo *bo)
 112 {
 113         if (!bo)
 114                 return;
 115
 116         if (_mesa_set_search(job->bos, bo))
 117                 return;
 118
 119         vc5_bo_reference(bo);
 120         _mesa_set_add(job->bos, bo);
 121         job->referenced_size += bo->size;
 122
 123         uint32_t *bo_handles = (void *)(uintptr_t)job->submit.bo_handles;
 124
 125         if (job->submit.bo_handle_count >= job->bo_handles_size) {
 126                 job->bo_handles_size = MAX2(4, job->bo_handles_size * 2);
 127                 bo_handles = reralloc(job, bo_handles,
 128                                       uint32_t, job->bo_handles_size);
 129                 job->submit.bo_handles = (uintptr_t)(void *)bo_handles;
 130         }
 131         bo_handles[job->submit.bo_handle_count++] = bo->handle;
 132 }
 133
 134 void
 135 vc5_job_add_write_resource(struct vc5_job *job, struct pipe_resource *prsc)
 136 {
 137         struct vc5_context *vc5 = job->vc5;
 138
 139         if (!job->write_prscs) {
 140                 job->write_prscs = _mesa_set_create(job,
 141                                                     _mesa_hash_pointer,
 142                                                     _mesa_key_pointer_equal);
 143         }
 144
 145         _mesa_set_add(job->write_prscs, prsc);
 146         _mesa_hash_table_insert(vc5->write_jobs, prsc, job);
 147 }
 148
 149 void
 150 vc5_flush_jobs_writing_resource(struct vc5_context *vc5,
 151                                 struct pipe_resource *prsc)
 152 {
 153         struct hash_entry *entry = _mesa_hash_table_search(vc5->write_jobs,
 154                                                            prsc);
 155         if (entry) {
 156                 struct vc5_job *job = entry->data;
 157                 vc5_job_submit(vc5, job);
 158         }
 159 }
 160
 161 void
 162 vc5_flush_jobs_reading_resource(struct vc5_context *vc5,
 163                                 struct pipe_resource *prsc)
 164 {
 165         struct vc5_resource *rsc = vc5_resource(prsc);
 166
 167         vc5_flush_jobs_writing_resource(vc5, prsc);
 168
 169         struct hash_entry *entry;
 170         hash_table_foreach(vc5->jobs, entry) {
 171                 struct vc5_job *job = entry->data;
 172
 173                 if (_mesa_set_search(job->bos, rsc->bo)) {
 174                         vc5_job_submit(vc5, job);
 175                         /* Reminder: vc5->jobs is safe to keep iterating even
 176                          * after deletion of an entry.
 177                          */
 178                         continue;
 179                 }
 180         }
 181 }
 182
 183 static void
 184 vc5_job_set_tile_buffer_size(struct vc5_job *job)
 185 {
 186         static const uint8_t tile_sizes[] = {
 187                 64, 64,
 188                 64, 32,
 189                 32, 32,
 190                 32, 16,
 191                 16, 16,
 192         };
 193         int tile_size_index = 0;
 194         if (job->msaa)
 195                 tile_size_index += 2;
 196
 197         if (job->cbufs[3] || job->cbufs[2])
 198                 tile_size_index += 2;
 199         else if (job->cbufs[1])
 200                 tile_size_index++;
 201
 202         int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
 203         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 204                 if (job->cbufs[i]) {
 205                         struct vc5_surface *surf = vc5_surface(job->cbufs[i]);
 206                         max_bpp = MAX2(max_bpp, surf->internal_bpp);
 207                 }
 208         }
 209         job->internal_bpp = max_bpp;
 210         STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
 211         tile_size_index += max_bpp;
 212
 213         assert(tile_size_index < ARRAY_SIZE(tile_sizes));
 214         job->tile_width = tile_sizes[tile_size_index * 2 + 0];
 215         job->tile_height = tile_sizes[tile_size_index * 2 + 1];
 216 }
 217
 218 /**
 219  * Returns a vc5_job struture for tracking V3D rendering to a particular FBO.
 220  *
 221  * If we've already started rendering to this FBO, then return old same job,
 222  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 223  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 224  * have been flushed.
 225  */
 226 struct vc5_job *
 227 vc5_get_job(struct vc5_context *vc5,
 228             struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
 229 {
 230         /* Return the existing job for this FBO if we have one */
 231         struct vc5_job_key local_key = {
 232                 .cbufs = {
 233                         cbufs[0],
 234                         cbufs[1],
 235                         cbufs[2],
 236                         cbufs[3],
 237                 },
 238                 .zsbuf = zsbuf,
 239         };
 240         struct hash_entry *entry = _mesa_hash_table_search(vc5->jobs,
 241                                                            &local_key);
 242         if (entry)
 243                 return entry->data;
 244
 245         /* Creating a new job.  Make sure that any previous jobs reading or
 246          * writing these buffers are flushed.
 247          */
 248         struct vc5_job *job = vc5_job_create(vc5);
 249
 250         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 251                 if (cbufs[i]) {
 252                         vc5_flush_jobs_reading_resource(vc5, cbufs[i]->texture);
 253                         pipe_surface_reference(&job->cbufs[i], cbufs[i]);
 254
 255                         if (cbufs[i]->texture->nr_samples > 1)
 256                                 job->msaa = true;
 257                 }
 258         }
 259         if (zsbuf) {
 260                 vc5_flush_jobs_reading_resource(vc5, zsbuf->texture);
 261                 pipe_surface_reference(&job->zsbuf, zsbuf);
 262                 if (zsbuf->texture->nr_samples > 1)
 263                         job->msaa = true;
 264         }
 265
 266         vc5_job_set_tile_buffer_size(job);
 267
 268         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 269                 if (cbufs[i])
 270                         _mesa_hash_table_insert(vc5->write_jobs,
 271                                                 cbufs[i]->texture, job);
 272         }
 273         if (zsbuf)
 274                 _mesa_hash_table_insert(vc5->write_jobs, zsbuf->texture, job);
 275
 276         memcpy(&job->key, &local_key, sizeof(local_key));
 277         _mesa_hash_table_insert(vc5->jobs, &job->key, job);
 278
 279         return job;
 280 }
 281
 282 struct vc5_job *
 283 vc5_get_job_for_fbo(struct vc5_context *vc5)
 284 {
 285         if (vc5->job)
 286                 return vc5->job;
 287
 288         struct pipe_surface **cbufs = vc5->framebuffer.cbufs;
 289         struct pipe_surface *zsbuf = vc5->framebuffer.zsbuf;
 290         struct vc5_job *job = vc5_get_job(vc5, cbufs, zsbuf);
 291
 292         /* The dirty flags are tracking what's been updated while vc5->job has
 293          * been bound, so set them all to ~0 when switching between jobs.  We
 294          * also need to reset all state at the start of rendering.
 295          */
 296         vc5->dirty = ~0;
 297
 298         /* If we're binding to uninitialized buffers, no need to load their
 299          * contents before drawing.
 300          */
 301         for (int i = 0; i < 4; i++) {
 302                 if (cbufs[i]) {
 303                         struct vc5_resource *rsc = vc5_resource(cbufs[i]->texture);
 304                         if (!rsc->writes)
 305                                 job->cleared |= PIPE_CLEAR_COLOR0 << i;
 306                 }
 307         }
 308
 309         if (zsbuf) {
 310                 struct vc5_resource *rsc = vc5_resource(zsbuf->texture);
 311                 if (!rsc->writes)
 312                         job->cleared |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 313         }
 314
 315         job->draw_tiles_x = DIV_ROUND_UP(vc5->framebuffer.width,
 316                                          job->tile_width);
 317         job->draw_tiles_y = DIV_ROUND_UP(vc5->framebuffer.height,
 318                                          job->tile_height);
 319
 320         vc5->job = job;
 321
 322         return job;
 323 }
 324
 325 static bool
 326 vc5_clif_dump_lookup(void *data, uint32_t addr, void **vaddr)
 327 {
 328         struct vc5_job *job = data;
 329         struct set_entry *entry;
 330
 331         set_foreach(job->bos, entry) {
 332                 struct vc5_bo *bo = (void *)entry->key;
 333
 334                 if (addr >= bo->offset &&
 335                     addr < bo->offset + bo->size) {
 336                         vc5_bo_map(bo);
 337                         *vaddr = bo->map + addr - bo->offset;
 338                         return true;
 339                 }
 340         }
 341
 342         return false;
 343 }
 344
 345 static void
 346 vc5_clif_dump(struct vc5_context *vc5, struct vc5_job *job)
 347 {
 348         if (!(V3D_DEBUG & V3D_DEBUG_CL))
 349                 return;
 350
 351         struct clif_dump *clif = clif_dump_init(&vc5->screen->devinfo,
 352                                                 stderr, vc5_clif_dump_lookup,
 353                                                 job);
 354
 355         fprintf(stderr, "BCL: 0x%08x..0x%08x\n",
 356                 job->submit.bcl_start, job->submit.bcl_end);
 357
 358         clif_dump_add_cl(clif, job->submit.bcl_start, job->submit.bcl_end);
 359
 360         fprintf(stderr, "RCL: 0x%08x..0x%08x\n",
 361                 job->submit.rcl_start, job->submit.rcl_end);
 362         clif_dump_add_cl(clif, job->submit.rcl_start, job->submit.rcl_end);
 363 }
 364
 365 /**
 366  * Submits the job to the kernel and then reinitializes it.
 367  */
 368 void
 369 vc5_job_submit(struct vc5_context *vc5, struct vc5_job *job)
 370 {
 371         if (!job->needs_flush)
 372                 goto done;
 373
 374         /* The RCL setup would choke if the draw bounds cause no drawing, so
 375          * just drop the drawing if that's the case.
 376          */
 377         if (job->draw_max_x <= job->draw_min_x ||
 378             job->draw_max_y <= job->draw_min_y) {
 379                 goto done;
 380         }
 381
 382         vc5_emit_rcl(job);
 383
 384         if (cl_offset(&job->bcl) > 0) {
 385                 vc5_cl_ensure_space_with_branch(&job->bcl,
 386                                                 7 +
 387                                                 cl_packet_length(OCCLUSION_QUERY_COUNTER));
 388
 389                 if (job->oq_enabled) {
 390                         /* Disable the OQ at the end of the CL, so that the
 391                          * draw calls at the start of the CL don't inherit the
 392                          * OQ counter.
 393                          */
 394                         cl_emit(&job->bcl, OCCLUSION_QUERY_COUNTER, counter);
 395                 }
 396
 397                 /* Increment the semaphore indicating that binning is done and
 398                  * unblocking the render thread.  Note that this doesn't act
 399                  * until the FLUSH completes.
 400                  */
 401                 cl_emit(&job->bcl, INCREMENT_SEMAPHORE, incr);
 402
 403                 /* The FLUSH_ALL emits any unwritten state changes in each
 404                  * tile.  We can use this to reset any state that needs to be
 405                  * present at the start of the next tile, as we do with
 406                  * OCCLUSION_QUERY_COUNTER above.
 407                  */
 408                 cl_emit(&job->bcl, FLUSH_ALL_STATE, flush);
 409         }
 410
 411         job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
 412         job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);
 413
 414         vc5_clif_dump(vc5, job);
 415
 416         if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
 417                 int ret;
 418
 419 #ifndef USE_VC5_SIMULATOR
 420                 ret = drmIoctl(vc5->fd, DRM_IOCTL_VC5_SUBMIT_CL, &job->submit);
 421 #else
 422                 ret = vc5_simulator_flush(vc5, &job->submit, job);
 423 #endif
 424                 static bool warned = false;
 425                 if (ret && !warned) {
 426                         fprintf(stderr, "Draw call returned %s.  "
 427                                         "Expect corruption.\n", strerror(errno));
 428                         warned = true;
 429                 }
 430         }
 431
 432         if (vc5->last_emit_seqno - vc5->screen->finished_seqno > 5) {
 433                 if (!vc5_wait_seqno(vc5->screen,
 434                                     vc5->last_emit_seqno - 5,
 435                                     PIPE_TIMEOUT_INFINITE,
 436                                     "job throttling")) {
 437                         fprintf(stderr, "Job throttling failed\n");
 438                 }
 439         }
 440
 441 done:
 442         vc5_job_free(vc5, job);
 443 }
 444
 445 static bool
 446 vc5_job_compare(const void *a, const void *b)
 447 {
 448         return memcmp(a, b, sizeof(struct vc5_job_key)) == 0;
 449 }
 450
 451 static uint32_t
 452 vc5_job_hash(const void *key)
 453 {
 454         return _mesa_hash_data(key, sizeof(struct vc5_job_key));
 455 }
 456
 457 void
 458 vc5_job_init(struct vc5_context *vc5)
 459 {
 460         vc5->jobs = _mesa_hash_table_create(vc5,
 461                                             vc5_job_hash,
 462                                             vc5_job_compare);
 463         vc5->write_jobs = _mesa_hash_table_create(vc5,
 464                                                   _mesa_hash_pointer,
 465                                                   _mesa_key_pointer_equal);
 466 }
 467