src/gallium/drivers/v3d/v3d_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 v3d_job.c
  25  *
  26  * Functions for submitting VC5 render jobs to the kernel.
  27  */
  28
  29 #include <xf86drm.h>
  30 #include "v3d_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 v3d_job_free(struct v3d_context *v3d, struct v3d_job *job)
  49 {
  50         struct set_entry *entry;
  51
  52         set_foreach(job->bos, entry) {
  53                 struct v3d_bo *bo = (struct v3d_bo *)entry->key;
  54                 v3d_bo_unreference(&bo);
  55         }
  56
  57         remove_from_ht(v3d->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(v3d->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(v3d->write_jobs, job->cbufs[i]->texture);
  72                         pipe_surface_reference(&job->cbufs[i], NULL);
  73                 }
  74         }
  75         if (job->zsbuf) {
  76                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
  77                 if (rsc->separate_stencil)
  78                         remove_from_ht(v3d->write_jobs,
  79                                        &rsc->separate_stencil->base);
  80
  81                 remove_from_ht(v3d->write_jobs, job->zsbuf->texture);
  82                 pipe_surface_reference(&job->zsbuf, NULL);
  83         }
  84
  85         if (v3d->job == job)
  86                 v3d->job = NULL;
  87
  88         v3d_destroy_cl(&job->bcl);
  89         v3d_destroy_cl(&job->rcl);
  90         v3d_destroy_cl(&job->indirect);
  91         v3d_bo_unreference(&job->tile_alloc);
  92         v3d_bo_unreference(&job->tile_state);
  93
  94         ralloc_free(job);
  95 }
  96
  97 static struct v3d_job *
  98 v3d_job_create(struct v3d_context *v3d)
  99 {
 100         struct v3d_job *job = rzalloc(v3d, struct v3d_job);
 101
 102         job->v3d = v3d;
 103
 104         v3d_init_cl(job, &job->bcl);
 105         v3d_init_cl(job, &job->rcl);
 106         v3d_init_cl(job, &job->indirect);
 107
 108         job->draw_min_x = ~0;
 109         job->draw_min_y = ~0;
 110         job->draw_max_x = 0;
 111         job->draw_max_y = 0;
 112
 113         job->bos = _mesa_set_create(job,
 114                                     _mesa_hash_pointer,
 115                                     _mesa_key_pointer_equal);
 116         return job;
 117 }
 118
 119 void
 120 v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo)
 121 {
 122         if (!bo)
 123                 return;
 124
 125         if (_mesa_set_search(job->bos, bo))
 126                 return;
 127
 128         v3d_bo_reference(bo);
 129         _mesa_set_add(job->bos, bo);
 130         job->referenced_size += bo->size;
 131
 132         uint32_t *bo_handles = (void *)(uintptr_t)job->submit.bo_handles;
 133
 134         if (job->submit.bo_handle_count >= job->bo_handles_size) {
 135                 job->bo_handles_size = MAX2(4, job->bo_handles_size * 2);
 136                 bo_handles = reralloc(job, bo_handles,
 137                                       uint32_t, job->bo_handles_size);
 138                 job->submit.bo_handles = (uintptr_t)(void *)bo_handles;
 139         }
 140         bo_handles[job->submit.bo_handle_count++] = bo->handle;
 141 }
 142
 143 void
 144 v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc)
 145 {
 146         struct v3d_context *v3d = job->v3d;
 147
 148         if (!job->write_prscs) {
 149                 job->write_prscs = _mesa_set_create(job,
 150                                                     _mesa_hash_pointer,
 151                                                     _mesa_key_pointer_equal);
 152         }
 153
 154         _mesa_set_add(job->write_prscs, prsc);
 155         _mesa_hash_table_insert(v3d->write_jobs, prsc, job);
 156 }
 157
 158 void
 159 v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
 160                                 struct pipe_resource *prsc)
 161 {
 162         struct hash_entry *entry = _mesa_hash_table_search(v3d->write_jobs,
 163                                                            prsc);
 164         if (entry) {
 165                 struct v3d_job *job = entry->data;
 166                 v3d_job_submit(v3d, job);
 167         }
 168 }
 169
 170 void
 171 v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
 172                                 struct pipe_resource *prsc)
 173 {
 174         struct v3d_resource *rsc = v3d_resource(prsc);
 175
 176         v3d_flush_jobs_writing_resource(v3d, prsc);
 177
 178         struct hash_entry *entry;
 179         hash_table_foreach(v3d->jobs, entry) {
 180                 struct v3d_job *job = entry->data;
 181
 182                 if (_mesa_set_search(job->bos, rsc->bo)) {
 183                         v3d_job_submit(v3d, job);
 184                         /* Reminder: v3d->jobs is safe to keep iterating even
 185                          * after deletion of an entry.
 186                          */
 187                         continue;
 188                 }
 189         }
 190 }
 191
 192 static void
 193 v3d_job_set_tile_buffer_size(struct v3d_job *job)
 194 {
 195         static const uint8_t tile_sizes[] = {
 196                 64, 64,
 197                 64, 32,
 198                 32, 32,
 199                 32, 16,
 200                 16, 16,
 201         };
 202         int tile_size_index = 0;
 203         if (job->msaa)
 204                 tile_size_index += 2;
 205
 206         if (job->cbufs[3] || job->cbufs[2])
 207                 tile_size_index += 2;
 208         else if (job->cbufs[1])
 209                 tile_size_index++;
 210
 211         int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
 212         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 213                 if (job->cbufs[i]) {
 214                         struct v3d_surface *surf = v3d_surface(job->cbufs[i]);
 215                         max_bpp = MAX2(max_bpp, surf->internal_bpp);
 216                 }
 217         }
 218         job->internal_bpp = max_bpp;
 219         STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
 220         tile_size_index += max_bpp;
 221
 222         assert(tile_size_index < ARRAY_SIZE(tile_sizes));
 223         job->tile_width = tile_sizes[tile_size_index * 2 + 0];
 224         job->tile_height = tile_sizes[tile_size_index * 2 + 1];
 225 }
 226
 227 /**
 228  * Returns a v3d_job struture for tracking V3D rendering to a particular FBO.
 229  *
 230  * If we've already started rendering to this FBO, then return old same job,
 231  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 232  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 233  * have been flushed.
 234  */
 235 struct v3d_job *
 236 v3d_get_job(struct v3d_context *v3d,
 237             struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
 238 {
 239         /* Return the existing job for this FBO if we have one */
 240         struct v3d_job_key local_key = {
 241                 .cbufs = {
 242                         cbufs[0],
 243                         cbufs[1],
 244                         cbufs[2],
 245                         cbufs[3],
 246                 },
 247                 .zsbuf = zsbuf,
 248         };
 249         struct hash_entry *entry = _mesa_hash_table_search(v3d->jobs,
 250                                                            &local_key);
 251         if (entry)
 252                 return entry->data;
 253
 254         /* Creating a new job.  Make sure that any previous jobs reading or
 255          * writing these buffers are flushed.
 256          */
 257         struct v3d_job *job = v3d_job_create(v3d);
 258
 259         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 260                 if (cbufs[i]) {
 261                         v3d_flush_jobs_reading_resource(v3d, cbufs[i]->texture);
 262                         pipe_surface_reference(&job->cbufs[i], cbufs[i]);
 263
 264                         if (cbufs[i]->texture->nr_samples > 1)
 265                                 job->msaa = true;
 266                 }
 267         }
 268         if (zsbuf) {
 269                 v3d_flush_jobs_reading_resource(v3d, zsbuf->texture);
 270                 pipe_surface_reference(&job->zsbuf, zsbuf);
 271                 if (zsbuf->texture->nr_samples > 1)
 272                         job->msaa = true;
 273         }
 274
 275         v3d_job_set_tile_buffer_size(job);
 276
 277         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 278                 if (cbufs[i])
 279                         _mesa_hash_table_insert(v3d->write_jobs,
 280                                                 cbufs[i]->texture, job);
 281         }
 282         if (zsbuf) {
 283                 _mesa_hash_table_insert(v3d->write_jobs, zsbuf->texture, job);
 284
 285                 struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
 286                 if (rsc->separate_stencil) {
 287                         v3d_flush_jobs_reading_resource(v3d,
 288                                                         &rsc->separate_stencil->base);
 289                         _mesa_hash_table_insert(v3d->write_jobs,
 290                                                 &rsc->separate_stencil->base,
 291                                                 job);
 292                 }
 293         }
 294
 295         memcpy(&job->key, &local_key, sizeof(local_key));
 296         _mesa_hash_table_insert(v3d->jobs, &job->key, job);
 297
 298         return job;
 299 }
 300
 301 struct v3d_job *
 302 v3d_get_job_for_fbo(struct v3d_context *v3d)
 303 {
 304         if (v3d->job)
 305                 return v3d->job;
 306
 307         struct pipe_surface **cbufs = v3d->framebuffer.cbufs;
 308         struct pipe_surface *zsbuf = v3d->framebuffer.zsbuf;
 309         struct v3d_job *job = v3d_get_job(v3d, cbufs, zsbuf);
 310
 311         /* The dirty flags are tracking what's been updated while v3d->job has
 312          * been bound, so set them all to ~0 when switching between jobs.  We
 313          * also need to reset all state at the start of rendering.
 314          */
 315         v3d->dirty = ~0;
 316
 317         /* If we're binding to uninitialized buffers, no need to load their
 318          * contents before drawing.
 319          */
 320         for (int i = 0; i < 4; i++) {
 321                 if (cbufs[i]) {
 322                         struct v3d_resource *rsc = v3d_resource(cbufs[i]->texture);
 323                         if (!rsc->writes)
 324                                 job->clear |= PIPE_CLEAR_COLOR0 << i;
 325                 }
 326         }
 327
 328         if (zsbuf) {
 329                 struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
 330                 if (!rsc->writes)
 331                         job->clear |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 332         }
 333
 334         job->draw_tiles_x = DIV_ROUND_UP(v3d->framebuffer.width,
 335                                          job->tile_width);
 336         job->draw_tiles_y = DIV_ROUND_UP(v3d->framebuffer.height,
 337                                          job->tile_height);
 338
 339         v3d->job = job;
 340
 341         return job;
 342 }
 343
 344 static void
 345 v3d_clif_dump(struct v3d_context *v3d, struct v3d_job *job)
 346 {
 347         if (!(V3D_DEBUG & (V3D_DEBUG_CL | V3D_DEBUG_CLIF)))
 348                 return;
 349
 350         struct clif_dump *clif = clif_dump_init(&v3d->screen->devinfo,
 351                                                 stderr,
 352                                                 V3D_DEBUG & V3D_DEBUG_CL);
 353
 354         struct set_entry *entry;
 355         set_foreach(job->bos, entry) {
 356                 struct v3d_bo *bo = (void *)entry->key;
 357                 char *name = ralloc_asprintf(NULL, "%s_0x%x",
 358                                              bo->name, bo->offset);
 359
 360                 v3d_bo_map(bo);
 361                 clif_dump_add_bo(clif, name, bo->offset, bo->size, bo->map);
 362
 363                 ralloc_free(name);
 364         }
 365
 366         fprintf(stderr, "BCL: 0x%08x..0x%08x\n",
 367                 job->submit.bcl_start, job->submit.bcl_end);
 368
 369         clif_dump_add_cl(clif, job->submit.bcl_start, job->submit.bcl_end);
 370
 371         fprintf(stderr, "RCL: 0x%08x..0x%08x\n",
 372                 job->submit.rcl_start, job->submit.rcl_end);
 373         clif_dump_add_cl(clif, job->submit.rcl_start, job->submit.rcl_end);
 374
 375         clif_dump(clif);
 376
 377         clif_dump_destroy(clif);
 378 }
 379
 380 /**
 381  * Submits the job to the kernel and then reinitializes it.
 382  */
 383 void
 384 v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job)
 385 {
 386         MAYBE_UNUSED struct v3d_screen *screen = v3d->screen;
 387
 388         if (!job->needs_flush)
 389                 goto done;
 390
 391         if (v3d->screen->devinfo.ver >= 41)
 392                 v3d41_emit_rcl(job);
 393         else
 394                 v3d33_emit_rcl(job);
 395
 396         if (cl_offset(&job->bcl) > 0) {
 397                 if (screen->devinfo.ver >= 41)
 398                         v3d41_bcl_epilogue(v3d, job);
 399                 else
 400                         v3d33_bcl_epilogue(v3d, job);
 401         }
 402
 403         job->submit.out_sync = v3d->out_sync;
 404         job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
 405         job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);
 406
 407         /* On V3D 4.1, the tile alloc/state setup moved to register writes
 408          * instead of binner packets.
 409          */
 410         if (screen->devinfo.ver >= 41) {
 411                 v3d_job_add_bo(job, job->tile_alloc);
 412                 job->submit.qma = job->tile_alloc->offset;
 413                 job->submit.qms = job->tile_alloc->size;
 414
 415                 v3d_job_add_bo(job, job->tile_state);
 416                 job->submit.qts = job->tile_state->offset;
 417         }
 418
 419         v3d_clif_dump(v3d, job);
 420
 421         if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
 422                 int ret;
 423
 424 #ifndef USE_V3D_SIMULATOR
 425                 ret = drmIoctl(v3d->fd, DRM_IOCTL_V3D_SUBMIT_CL, &job->submit);
 426 #else
 427                 ret = v3d_simulator_flush(v3d, &job->submit, job);
 428 #endif
 429                 static bool warned = false;
 430                 if (ret && !warned) {
 431                         fprintf(stderr, "Draw call returned %s.  "
 432                                         "Expect corruption.\n", strerror(errno));
 433                         warned = true;
 434                 }
 435         }
 436
 437 done:
 438         v3d_job_free(v3d, job);
 439 }
 440
 441 static bool
 442 v3d_job_compare(const void *a, const void *b)
 443 {
 444         return memcmp(a, b, sizeof(struct v3d_job_key)) == 0;
 445 }
 446
 447 static uint32_t
 448 v3d_job_hash(const void *key)
 449 {
 450         return _mesa_hash_data(key, sizeof(struct v3d_job_key));
 451 }
 452
 453 void
 454 v3d_job_init(struct v3d_context *v3d)
 455 {
 456         v3d->jobs = _mesa_hash_table_create(v3d,
 457                                             v3d_job_hash,
 458                                             v3d_job_compare);
 459         v3d->write_jobs = _mesa_hash_table_create(v3d,
 460                                                   _mesa_hash_pointer,
 461                                                   _mesa_key_pointer_equal);
 462 }
 463