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         set_foreach(job->bos, entry) {
  51                 struct v3d_bo *bo = (struct v3d_bo *)entry->key;
  52                 v3d_bo_unreference(&bo);
  53         }
  54
  55         remove_from_ht(v3d->jobs, &job->key);
  56
  57         if (job->write_prscs) {
  58                 set_foreach(job->write_prscs, entry) {
  59                         const struct pipe_resource *prsc = entry->key;
  60
  61                         remove_from_ht(v3d->write_jobs, (void *)prsc);
  62                 }
  63         }
  64
  65         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
  66                 if (job->cbufs[i]) {
  67                         remove_from_ht(v3d->write_jobs, job->cbufs[i]->texture);
  68                         pipe_surface_reference(&job->cbufs[i], NULL);
  69                 }
  70         }
  71         if (job->zsbuf) {
  72                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
  73                 if (rsc->separate_stencil)
  74                         remove_from_ht(v3d->write_jobs,
  75                                        &rsc->separate_stencil->base);
  76
  77                 remove_from_ht(v3d->write_jobs, job->zsbuf->texture);
  78                 pipe_surface_reference(&job->zsbuf, NULL);
  79         }
  80
  81         if (v3d->job == job)
  82                 v3d->job = NULL;
  83
  84         v3d_destroy_cl(&job->bcl);
  85         v3d_destroy_cl(&job->rcl);
  86         v3d_destroy_cl(&job->indirect);
  87         v3d_bo_unreference(&job->tile_alloc);
  88         v3d_bo_unreference(&job->tile_state);
  89
  90         ralloc_free(job);
  91 }
  92
  93 static struct v3d_job *
  94 v3d_job_create(struct v3d_context *v3d)
  95 {
  96         struct v3d_job *job = rzalloc(v3d, struct v3d_job);
  97
  98         job->v3d = v3d;
  99
 100         v3d_init_cl(job, &job->bcl);
 101         v3d_init_cl(job, &job->rcl);
 102         v3d_init_cl(job, &job->indirect);
 103
 104         job->draw_min_x = ~0;
 105         job->draw_min_y = ~0;
 106         job->draw_max_x = 0;
 107         job->draw_max_y = 0;
 108
 109         job->bos = _mesa_set_create(job,
 110                                     _mesa_hash_pointer,
 111                                     _mesa_key_pointer_equal);
 112         return job;
 113 }
 114
 115 void
 116 v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo)
 117 {
 118         if (!bo)
 119                 return;
 120
 121         if (_mesa_set_search(job->bos, bo))
 122                 return;
 123
 124         v3d_bo_reference(bo);
 125         _mesa_set_add(job->bos, bo);
 126         job->referenced_size += bo->size;
 127
 128         uint32_t *bo_handles = (void *)(uintptr_t)job->submit.bo_handles;
 129
 130         if (job->submit.bo_handle_count >= job->bo_handles_size) {
 131                 job->bo_handles_size = MAX2(4, job->bo_handles_size * 2);
 132                 bo_handles = reralloc(job, bo_handles,
 133                                       uint32_t, job->bo_handles_size);
 134                 job->submit.bo_handles = (uintptr_t)(void *)bo_handles;
 135         }
 136         bo_handles[job->submit.bo_handle_count++] = bo->handle;
 137 }
 138
 139 void
 140 v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc)
 141 {
 142         struct v3d_context *v3d = job->v3d;
 143
 144         if (!job->write_prscs) {
 145                 job->write_prscs = _mesa_set_create(job,
 146                                                     _mesa_hash_pointer,
 147                                                     _mesa_key_pointer_equal);
 148         }
 149
 150         _mesa_set_add(job->write_prscs, prsc);
 151         _mesa_hash_table_insert(v3d->write_jobs, prsc, job);
 152 }
 153
 154 void
 155 v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
 156                                 struct pipe_resource *prsc)
 157 {
 158         struct hash_entry *entry = _mesa_hash_table_search(v3d->write_jobs,
 159                                                            prsc);
 160         if (entry) {
 161                 struct v3d_job *job = entry->data;
 162                 v3d_job_submit(v3d, job);
 163         }
 164 }
 165
 166 void
 167 v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
 168                                 struct pipe_resource *prsc)
 169 {
 170         struct v3d_resource *rsc = v3d_resource(prsc);
 171
 172         v3d_flush_jobs_writing_resource(v3d, prsc);
 173
 174         hash_table_foreach(v3d->jobs, entry) {
 175                 struct v3d_job *job = entry->data;
 176
 177                 if (_mesa_set_search(job->bos, rsc->bo)) {
 178                         v3d_job_submit(v3d, job);
 179                         /* Reminder: v3d->jobs is safe to keep iterating even
 180                          * after deletion of an entry.
 181                          */
 182                         continue;
 183                 }
 184         }
 185 }
 186
 187 static void
 188 v3d_job_set_tile_buffer_size(struct v3d_job *job)
 189 {
 190         static const uint8_t tile_sizes[] = {
 191                 64, 64,
 192                 64, 32,
 193                 32, 32,
 194                 32, 16,
 195                 16, 16,
 196         };
 197         int tile_size_index = 0;
 198         if (job->msaa)
 199                 tile_size_index += 2;
 200
 201         if (job->cbufs[3] || job->cbufs[2])
 202                 tile_size_index += 2;
 203         else if (job->cbufs[1])
 204                 tile_size_index++;
 205
 206         int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
 207         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
 208                 if (job->cbufs[i]) {
 209                         struct v3d_surface *surf = v3d_surface(job->cbufs[i]);
 210                         max_bpp = MAX2(max_bpp, surf->internal_bpp);
 211                 }
 212         }
 213         job->internal_bpp = max_bpp;
 214         STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
 215         tile_size_index += max_bpp;
 216
 217         assert(tile_size_index < ARRAY_SIZE(tile_sizes));
 218         job->tile_width = tile_sizes[tile_size_index * 2 + 0];
 219         job->tile_height = tile_sizes[tile_size_index * 2 + 1];
 220 }
 221
 222 /**
 223  * Returns a v3d_job struture for tracking V3D rendering to a particular FBO.
 224  *
 225  * If we've already started rendering to this FBO, then return the same job,
 226  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 227  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 228  * have been flushed.
 229  */
 230 struct v3d_job *
 231 v3d_get_job(struct v3d_context *v3d,
 232             struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
 233 {
 234         /* Return the existing job for this FBO if we have one */
 235         struct v3d_job_key local_key = {
 236                 .cbufs = {
 237                         cbufs[0],
 238                         cbufs[1],
 239                         cbufs[2],
 240                         cbufs[3],
 241                 },
 242                 .zsbuf = zsbuf,
 243         };
 244         struct hash_entry *entry = _mesa_hash_table_search(v3d->jobs,
 245                                                            &local_key);
 246         if (entry)
 247                 return entry->data;
 248
 249         /* Creating a new job.  Make sure that any previous jobs reading or
 250          * writing these buffers are flushed.
 251          */
 252         struct v3d_job *job = v3d_job_create(v3d);
 253
 254         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
 255                 if (cbufs[i]) {
 256                         v3d_flush_jobs_reading_resource(v3d, cbufs[i]->texture);
 257                         pipe_surface_reference(&job->cbufs[i], cbufs[i]);
 258
 259                         if (cbufs[i]->texture->nr_samples > 1)
 260                                 job->msaa = true;
 261                 }
 262         }
 263         if (zsbuf) {
 264                 v3d_flush_jobs_reading_resource(v3d, zsbuf->texture);
 265                 pipe_surface_reference(&job->zsbuf, zsbuf);
 266                 if (zsbuf->texture->nr_samples > 1)
 267                         job->msaa = true;
 268         }
 269
 270         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
 271                 if (cbufs[i])
 272                         _mesa_hash_table_insert(v3d->write_jobs,
 273                                                 cbufs[i]->texture, job);
 274         }
 275         if (zsbuf) {
 276                 _mesa_hash_table_insert(v3d->write_jobs, zsbuf->texture, job);
 277
 278                 struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
 279                 if (rsc->separate_stencil) {
 280                         v3d_flush_jobs_reading_resource(v3d,
 281                                                         &rsc->separate_stencil->base);
 282                         _mesa_hash_table_insert(v3d->write_jobs,
 283                                                 &rsc->separate_stencil->base,
 284                                                 job);
 285                 }
 286         }
 287
 288         memcpy(&job->key, &local_key, sizeof(local_key));
 289         _mesa_hash_table_insert(v3d->jobs, &job->key, job);
 290
 291         return job;
 292 }
 293
 294 struct v3d_job *
 295 v3d_get_job_for_fbo(struct v3d_context *v3d)
 296 {
 297         if (v3d->job)
 298                 return v3d->job;
 299
 300         struct pipe_surface **cbufs = v3d->framebuffer.cbufs;
 301         struct pipe_surface *zsbuf = v3d->framebuffer.zsbuf;
 302         struct v3d_job *job = v3d_get_job(v3d, cbufs, zsbuf);
 303
 304         if (v3d->framebuffer.samples >= 1)
 305                 job->msaa = true;
 306
 307         v3d_job_set_tile_buffer_size(job);
 308
 309         /* The dirty flags are tracking what's been updated while v3d->job has
 310          * been bound, so set them all to ~0 when switching between jobs.  We
 311          * also need to reset all state at the start of rendering.
 312          */
 313         v3d->dirty = ~0;
 314
 315         /* If we're binding to uninitialized buffers, no need to load their
 316          * contents before drawing.
 317          */
 318         for (int i = 0; i < 4; i++) {
 319                 if (cbufs[i]) {
 320                         struct v3d_resource *rsc = v3d_resource(cbufs[i]->texture);
 321                         if (!rsc->writes)
 322                                 job->clear |= PIPE_CLEAR_COLOR0 << i;
 323                 }
 324         }
 325
 326         if (zsbuf) {
 327                 struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
 328                 if (!rsc->writes)
 329                         job->clear |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 330         }
 331
 332         job->draw_tiles_x = DIV_ROUND_UP(v3d->framebuffer.width,
 333                                          job->tile_width);
 334         job->draw_tiles_y = DIV_ROUND_UP(v3d->framebuffer.height,
 335                                          job->tile_height);
 336
 337         v3d->job = job;
 338
 339         return job;
 340 }
 341
 342 static void
 343 v3d_clif_dump(struct v3d_context *v3d, struct v3d_job *job)
 344 {
 345         if (!(V3D_DEBUG & (V3D_DEBUG_CL | V3D_DEBUG_CLIF)))
 346                 return;
 347
 348         struct clif_dump *clif = clif_dump_init(&v3d->screen->devinfo,
 349                                                 stderr,
 350                                                 V3D_DEBUG & V3D_DEBUG_CL);
 351
 352         set_foreach(job->bos, entry) {
 353                 struct v3d_bo *bo = (void *)entry->key;
 354                 char *name = ralloc_asprintf(NULL, "%s_0x%x",
 355                                              bo->name, bo->offset);
 356
 357                 v3d_bo_map(bo);
 358                 clif_dump_add_bo(clif, name, bo->offset, bo->size, bo->map);
 359
 360                 ralloc_free(name);
 361         }
 362
 363         clif_dump(clif, &job->submit);
 364
 365         clif_dump_destroy(clif);
 366 }
 367
 368 /**
 369  * Submits the job to the kernel and then reinitializes it.
 370  */
 371 void
 372 v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job)
 373 {
 374         MAYBE_UNUSED struct v3d_screen *screen = v3d->screen;
 375
 376         if (!job->needs_flush)
 377                 goto done;
 378
 379         if (v3d->screen->devinfo.ver >= 41)
 380                 v3d41_emit_rcl(job);
 381         else
 382                 v3d33_emit_rcl(job);
 383
 384         if (cl_offset(&job->bcl) > 0) {
 385                 if (screen->devinfo.ver >= 41)
 386                         v3d41_bcl_epilogue(v3d, job);
 387                 else
 388                         v3d33_bcl_epilogue(v3d, job);
 389         }
 390
 391         /* While the RCL will implicitly depend on the last RCL to have
 392          * finished, we also need to block on any previous TFU job we may have
 393          * dispatched.
 394          */
 395         job->submit.in_sync_rcl = v3d->out_sync;
 396
 397         /* Update the sync object for the last rendering by our context. */
 398         job->submit.out_sync = v3d->out_sync;
 399
 400         job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
 401         job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);
 402
 403         /* On V3D 4.1, the tile alloc/state setup moved to register writes
 404          * instead of binner packets.
 405          */
 406         if (screen->devinfo.ver >= 41) {
 407                 v3d_job_add_bo(job, job->tile_alloc);
 408                 job->submit.qma = job->tile_alloc->offset;
 409                 job->submit.qms = job->tile_alloc->size;
 410
 411                 v3d_job_add_bo(job, job->tile_state);
 412                 job->submit.qts = job->tile_state->offset;
 413         }
 414
 415         v3d_clif_dump(v3d, job);
 416
 417         if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
 418                 int ret;
 419
 420                 ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_SUBMIT_CL, &job->submit);
 421                 static bool warned = false;
 422                 if (ret && !warned) {
 423                         fprintf(stderr, "Draw call returned %s.  "
 424                                         "Expect corruption.\n", strerror(errno));
 425                         warned = true;
 426                 }
 427         }
 428
 429 done:
 430         v3d_job_free(v3d, job);
 431 }
 432
 433 static bool
 434 v3d_job_compare(const void *a, const void *b)
 435 {
 436         return memcmp(a, b, sizeof(struct v3d_job_key)) == 0;
 437 }
 438
 439 static uint32_t
 440 v3d_job_hash(const void *key)
 441 {
 442         return _mesa_hash_data(key, sizeof(struct v3d_job_key));
 443 }
 444
 445 void
 446 v3d_job_init(struct v3d_context *v3d)
 447 {
 448         v3d->jobs = _mesa_hash_table_create(v3d,
 449                                             v3d_job_hash,
 450                                             v3d_job_compare);
 451         v3d->write_jobs = _mesa_hash_table_create(v3d,
 452                                                   _mesa_hash_pointer,
 453                                                   _mesa_key_pointer_equal);
 454 }
 455