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         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
  57                 if (job->cbufs[i]) {
  58                         remove_from_ht(vc5->write_jobs, job->cbufs[i]->texture);
  59                         pipe_surface_reference(&job->cbufs[i], NULL);
  60                 }
  61         }
  62         if (job->zsbuf) {
  63                 remove_from_ht(vc5->write_jobs, job->zsbuf->texture);
  64                 pipe_surface_reference(&job->zsbuf, NULL);
  65         }
  66
  67         if (vc5->job == job)
  68                 vc5->job = NULL;
  69
  70         vc5_destroy_cl(&job->bcl);
  71         vc5_destroy_cl(&job->rcl);
  72         vc5_destroy_cl(&job->indirect);
  73         vc5_bo_unreference(&job->tile_alloc);
  74
  75         ralloc_free(job);
  76 }
  77
  78 static struct vc5_job *
  79 vc5_job_create(struct vc5_context *vc5)
  80 {
  81         struct vc5_job *job = rzalloc(vc5, struct vc5_job);
  82
  83         job->vc5 = vc5;
  84
  85         vc5_init_cl(job, &job->bcl);
  86         vc5_init_cl(job, &job->rcl);
  87         vc5_init_cl(job, &job->indirect);
  88
  89         job->draw_min_x = ~0;
  90         job->draw_min_y = ~0;
  91         job->draw_max_x = 0;
  92         job->draw_max_y = 0;
  93
  94         job->bos = _mesa_set_create(job,
  95                                     _mesa_hash_pointer,
  96                                     _mesa_key_pointer_equal);
  97         return job;
  98 }
  99
 100 void
 101 vc5_job_add_bo(struct vc5_job *job, struct vc5_bo *bo)
 102 {
 103         if (!bo)
 104                 return;
 105
 106         if (_mesa_set_search(job->bos, bo))
 107                 return;
 108
 109         vc5_bo_reference(bo);
 110         _mesa_set_add(job->bos, bo);
 111
 112         uint32_t *bo_handles = (void *)(uintptr_t)job->submit.bo_handles;
 113
 114         if (job->submit.bo_handle_count >= job->bo_handles_size) {
 115                 job->bo_handles_size = MAX2(4, job->bo_handles_size * 2);
 116                 bo_handles = reralloc(job, bo_handles,
 117                                       uint32_t, job->bo_handles_size);
 118                 job->submit.bo_handles = (uintptr_t)(void *)bo_handles;
 119         }
 120         bo_handles[job->submit.bo_handle_count++] = bo->handle;
 121 }
 122
 123 void
 124 vc5_flush_jobs_writing_resource(struct vc5_context *vc5,
 125                                 struct pipe_resource *prsc)
 126 {
 127         struct hash_entry *entry = _mesa_hash_table_search(vc5->write_jobs,
 128                                                            prsc);
 129         if (entry) {
 130                 struct vc5_job *job = entry->data;
 131                 vc5_job_submit(vc5, job);
 132         }
 133 }
 134
 135 void
 136 vc5_flush_jobs_reading_resource(struct vc5_context *vc5,
 137                                 struct pipe_resource *prsc)
 138 {
 139         struct vc5_resource *rsc = vc5_resource(prsc);
 140
 141         vc5_flush_jobs_writing_resource(vc5, prsc);
 142
 143         struct hash_entry *entry;
 144         hash_table_foreach(vc5->jobs, entry) {
 145                 struct vc5_job *job = entry->data;
 146
 147                 if (_mesa_set_search(job->bos, rsc->bo)) {
 148                         vc5_job_submit(vc5, job);
 149                         /* Reminder: vc5->jobs is safe to keep iterating even
 150                          * after deletion of an entry.
 151                          */
 152                         continue;
 153                 }
 154         }
 155 }
 156
 157 static void
 158 vc5_job_set_tile_buffer_size(struct vc5_job *job)
 159 {
 160         static const uint8_t tile_sizes[] = {
 161                 64, 64,
 162                 64, 32,
 163                 32, 32,
 164                 32, 16,
 165                 16, 16,
 166         };
 167         int tile_size_index = 0;
 168         if (job->msaa)
 169                 tile_size_index += 2;
 170
 171         if (job->cbufs[3])
 172                 tile_size_index += 2;
 173         else if (job->cbufs[2])
 174                 tile_size_index++;
 175
 176         int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
 177         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 178                 if (job->cbufs[i]) {
 179                         struct vc5_surface *surf = vc5_surface(job->cbufs[i]);
 180                         max_bpp = MAX2(max_bpp, surf->internal_bpp);
 181                 }
 182         }
 183         job->internal_bpp = max_bpp;
 184         STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
 185         tile_size_index += max_bpp;
 186
 187         assert(tile_size_index < ARRAY_SIZE(tile_sizes));
 188         job->tile_width = tile_sizes[tile_size_index * 2 + 0];
 189         job->tile_height = tile_sizes[tile_size_index * 2 + 1];
 190 }
 191
 192 /**
 193  * Returns a vc5_job struture for tracking V3D rendering to a particular FBO.
 194  *
 195  * If we've already started rendering to this FBO, then return old same job,
 196  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 197  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 198  * have been flushed.
 199  */
 200 struct vc5_job *
 201 vc5_get_job(struct vc5_context *vc5,
 202             struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
 203 {
 204         /* Return the existing job for this FBO if we have one */
 205         struct vc5_job_key local_key = {
 206                 .cbufs = {
 207                         cbufs[0],
 208                         cbufs[1],
 209                         cbufs[2],
 210                         cbufs[3],
 211                 },
 212                 .zsbuf = zsbuf,
 213         };
 214         struct hash_entry *entry = _mesa_hash_table_search(vc5->jobs,
 215                                                            &local_key);
 216         if (entry)
 217                 return entry->data;
 218
 219         /* Creating a new job.  Make sure that any previous jobs reading or
 220          * writing these buffers are flushed.
 221          */
 222         struct vc5_job *job = vc5_job_create(vc5);
 223
 224         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 225                 if (cbufs[i]) {
 226                         vc5_flush_jobs_reading_resource(vc5, cbufs[i]->texture);
 227                         pipe_surface_reference(&job->cbufs[i], cbufs[i]);
 228
 229                         if (cbufs[i]->texture->nr_samples > 1)
 230                                 job->msaa = true;
 231                 }
 232         }
 233         if (zsbuf) {
 234                 vc5_flush_jobs_reading_resource(vc5, zsbuf->texture);
 235                 pipe_surface_reference(&job->zsbuf, zsbuf);
 236                 if (zsbuf->texture->nr_samples > 1)
 237                         job->msaa = true;
 238         }
 239
 240         vc5_job_set_tile_buffer_size(job);
 241
 242         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 243                 if (cbufs[i])
 244                         _mesa_hash_table_insert(vc5->write_jobs,
 245                                                 cbufs[i]->texture, job);
 246         }
 247         if (zsbuf)
 248                 _mesa_hash_table_insert(vc5->write_jobs, zsbuf->texture, job);
 249
 250         memcpy(&job->key, &local_key, sizeof(local_key));
 251         _mesa_hash_table_insert(vc5->jobs, &job->key, job);
 252
 253         return job;
 254 }
 255
 256 struct vc5_job *
 257 vc5_get_job_for_fbo(struct vc5_context *vc5)
 258 {
 259         if (vc5->job)
 260                 return vc5->job;
 261
 262         struct pipe_surface **cbufs = vc5->framebuffer.cbufs;
 263         struct pipe_surface *zsbuf = vc5->framebuffer.zsbuf;
 264         struct vc5_job *job = vc5_get_job(vc5, cbufs, zsbuf);
 265
 266         /* The dirty flags are tracking what's been updated while vc5->job has
 267          * been bound, so set them all to ~0 when switching between jobs.  We
 268          * also need to reset all state at the start of rendering.
 269          */
 270         vc5->dirty = ~0;
 271
 272         /* If we're binding to uninitialized buffers, no need to load their
 273          * contents before drawing.
 274          */
 275         for (int i = 0; i < 4; i++) {
 276                 if (cbufs[i]) {
 277                         struct vc5_resource *rsc = vc5_resource(cbufs[i]->texture);
 278                         if (!rsc->writes)
 279                                 job->cleared |= PIPE_CLEAR_COLOR0 << i;
 280                 }
 281         }
 282
 283         if (zsbuf) {
 284                 struct vc5_resource *rsc = vc5_resource(zsbuf->texture);
 285                 if (!rsc->writes)
 286                         job->cleared |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 287         }
 288
 289         job->draw_tiles_x = DIV_ROUND_UP(vc5->framebuffer.width,
 290                                          job->tile_width);
 291         job->draw_tiles_y = DIV_ROUND_UP(vc5->framebuffer.height,
 292                                          job->tile_height);
 293
 294         vc5->job = job;
 295
 296         return job;
 297 }
 298
 299 static bool
 300 vc5_clif_dump_lookup(void *data, uint32_t addr, void **vaddr)
 301 {
 302         struct vc5_job *job = data;
 303         struct set_entry *entry;
 304
 305         set_foreach(job->bos, entry) {
 306                 struct vc5_bo *bo = (void *)entry->key;
 307
 308                 if (addr >= bo->offset &&
 309                     addr < bo->offset + bo->size) {
 310                         vc5_bo_map(bo);
 311                         *vaddr = bo->map + addr - bo->offset;
 312                         return true;
 313                 }
 314         }
 315
 316         return false;
 317 }
 318
 319 static void
 320 vc5_clif_dump(struct vc5_context *vc5, struct vc5_job *job)
 321 {
 322         if (!(V3D_DEBUG & V3D_DEBUG_CL))
 323                 return;
 324
 325         struct clif_dump *clif = clif_dump_init(&vc5->screen->devinfo,
 326                                                 stderr, vc5_clif_dump_lookup,
 327                                                 job);
 328
 329         fprintf(stderr, "BCL: 0x%08x..0x%08x\n",
 330                 job->submit.bcl_start, job->submit.bcl_end);
 331
 332         clif_dump_add_cl(clif, job->submit.bcl_start, job->submit.bcl_end);
 333
 334         fprintf(stderr, "RCL: 0x%08x..0x%08x\n",
 335                 job->submit.rcl_start, job->submit.rcl_end);
 336         clif_dump_add_cl(clif, job->submit.rcl_start, job->submit.rcl_end);
 337 }
 338
 339 /**
 340  * Submits the job to the kernel and then reinitializes it.
 341  */
 342 void
 343 vc5_job_submit(struct vc5_context *vc5, struct vc5_job *job)
 344 {
 345         if (!job->needs_flush)
 346                 goto done;
 347
 348         /* The RCL setup would choke if the draw bounds cause no drawing, so
 349          * just drop the drawing if that's the case.
 350          */
 351         if (job->draw_max_x <= job->draw_min_x ||
 352             job->draw_max_y <= job->draw_min_y) {
 353                 goto done;
 354         }
 355
 356         vc5_emit_rcl(job);
 357
 358         if (cl_offset(&job->bcl) > 0) {
 359                 vc5_cl_ensure_space_with_branch(&job->bcl, 2);
 360
 361                 /* Increment the semaphore indicating that binning is done and
 362                  * unblocking the render thread.  Note that this doesn't act
 363                  * until the FLUSH completes.
 364                  */
 365                 cl_emit(&job->bcl, INCREMENT_SEMAPHORE, incr);
 366
 367                 /* The FLUSH caps all of our bin lists with a
 368                  * VC5_PACKET_RETURN.
 369                  */
 370                 cl_emit(&job->bcl, FLUSH, flush);
 371         }
 372
 373         job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
 374         job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);
 375
 376         vc5_clif_dump(vc5, job);
 377
 378         if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
 379                 int ret;
 380
 381 #ifndef USE_VC5_SIMULATOR
 382                 ret = drmIoctl(vc5->fd, DRM_IOCTL_VC5_SUBMIT_CL, &job->submit);
 383 #else
 384                 ret = vc5_simulator_flush(vc5, &job->submit, job);
 385 #endif
 386                 static bool warned = false;
 387                 if (ret && !warned) {
 388                         fprintf(stderr, "Draw call returned %s.  "
 389                                         "Expect corruption.\n", strerror(errno));
 390                         warned = true;
 391                 }
 392         }
 393
 394         if (vc5->last_emit_seqno - vc5->screen->finished_seqno > 5) {
 395                 if (!vc5_wait_seqno(vc5->screen,
 396                                     vc5->last_emit_seqno - 5,
 397                                     PIPE_TIMEOUT_INFINITE,
 398                                     "job throttling")) {
 399                         fprintf(stderr, "Job throttling failed\n");
 400                 }
 401         }
 402
 403 done:
 404         vc5_job_free(vc5, job);
 405 }
 406
 407 static bool
 408 vc5_job_compare(const void *a, const void *b)
 409 {
 410         return memcmp(a, b, sizeof(struct vc5_job_key)) == 0;
 411 }
 412
 413 static uint32_t
 414 vc5_job_hash(const void *key)
 415 {
 416         return _mesa_hash_data(key, sizeof(struct vc5_job_key));
 417 }
 418
 419 void
 420 vc5_job_init(struct vc5_context *vc5)
 421 {
 422         vc5->jobs = _mesa_hash_table_create(vc5,
 423                                             vc5_job_hash,
 424                                             vc5_job_compare);
 425         vc5->write_jobs = _mesa_hash_table_create(vc5,
 426                                                   _mesa_hash_pointer,
 427                                                   _mesa_key_pointer_equal);
 428 }
 429