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 v3d_job_free(struct v3d_context *v3d, struct v3d_job *job)
  42 {
  43         set_foreach(job->bos, entry) {
  44                 struct v3d_bo *bo = (struct v3d_bo *)entry->key;
  45                 v3d_bo_unreference(&bo);
  46         }
  47
  48         _mesa_hash_table_remove_key(v3d->jobs, &job->key);
  49
  50         if (job->write_prscs) {
  51                 set_foreach(job->write_prscs, entry) {
  52                         const struct pipe_resource *prsc = entry->key;
  53
  54                         _mesa_hash_table_remove_key(v3d->write_jobs, prsc);
  55                 }
  56         }
  57
  58         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
  59                 if (job->cbufs[i]) {
  60                         _mesa_hash_table_remove_key(v3d->write_jobs,
  61                                                     job->cbufs[i]->texture);
  62                         pipe_surface_reference(&job->cbufs[i], NULL);
  63                 }
  64         }
  65         if (job->zsbuf) {
  66                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
  67                 if (rsc->separate_stencil)
  68                         _mesa_hash_table_remove_key(v3d->write_jobs,
  69                                                     &rsc->separate_stencil->base);
  70
  71                 _mesa_hash_table_remove_key(v3d->write_jobs,
  72                                             job->zsbuf->texture);
  73                 pipe_surface_reference(&job->zsbuf, NULL);
  74         }
  75
  76         if (v3d->job == job)
  77                 v3d->job = NULL;
  78
  79         v3d_destroy_cl(&job->bcl);
  80         v3d_destroy_cl(&job->rcl);
  81         v3d_destroy_cl(&job->indirect);
  82         v3d_bo_unreference(&job->tile_alloc);
  83         v3d_bo_unreference(&job->tile_state);
  84
  85         ralloc_free(job);
  86 }
  87
  88 static struct v3d_job *
  89 v3d_job_create(struct v3d_context *v3d)
  90 {
  91         struct v3d_job *job = rzalloc(v3d, struct v3d_job);
  92
  93         job->v3d = v3d;
  94
  95         v3d_init_cl(job, &job->bcl);
  96         v3d_init_cl(job, &job->rcl);
  97         v3d_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 v3d_job_add_bo(struct v3d_job *job, struct v3d_bo *bo)
 112 {
 113         if (!bo)
 114                 return;
 115
 116         if (_mesa_set_search(job->bos, bo))
 117                 return;
 118
 119         v3d_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 v3d_job_add_write_resource(struct v3d_job *job, struct pipe_resource *prsc)
 136 {
 137         struct v3d_context *v3d = job->v3d;
 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(v3d->write_jobs, prsc, job);
 147 }
 148
 149 void
 150 v3d_flush_jobs_using_bo(struct v3d_context *v3d, struct v3d_bo *bo)
 151 {
 152         hash_table_foreach(v3d->jobs, entry) {
 153                 struct v3d_job *job = entry->data;
 154
 155                 if (_mesa_set_search(job->bos, bo))
 156                         v3d_job_submit(v3d, job);
 157         }
 158 }
 159
 160 void
 161 v3d_flush_jobs_writing_resource(struct v3d_context *v3d,
 162                                 struct pipe_resource *prsc)
 163 {
 164         struct hash_entry *entry = _mesa_hash_table_search(v3d->write_jobs,
 165                                                            prsc);
 166         if (entry) {
 167                 struct v3d_job *job = entry->data;
 168                 v3d_job_submit(v3d, job);
 169         }
 170 }
 171
 172 void
 173 v3d_flush_jobs_reading_resource(struct v3d_context *v3d,
 174                                 struct pipe_resource *prsc)
 175 {
 176         struct v3d_resource *rsc = v3d_resource(prsc);
 177
 178         v3d_flush_jobs_writing_resource(v3d, prsc);
 179
 180         hash_table_foreach(v3d->jobs, entry) {
 181                 struct v3d_job *job = entry->data;
 182
 183                 if (_mesa_set_search(job->bos, rsc->bo)) {
 184                         v3d_job_submit(v3d, job);
 185                         /* Reminder: v3d->jobs is safe to keep iterating even
 186                          * after deletion of an entry.
 187                          */
 188                         continue;
 189                 }
 190         }
 191 }
 192
 193 static void
 194 v3d_job_set_tile_buffer_size(struct v3d_job *job)
 195 {
 196         static const uint8_t tile_sizes[] = {
 197                 64, 64,
 198                 64, 32,
 199                 32, 32,
 200                 32, 16,
 201                 16, 16,
 202         };
 203         int tile_size_index = 0;
 204         if (job->msaa)
 205                 tile_size_index += 2;
 206
 207         if (job->cbufs[3] || job->cbufs[2])
 208                 tile_size_index += 2;
 209         else if (job->cbufs[1])
 210                 tile_size_index++;
 211
 212         int max_bpp = RENDER_TARGET_MAXIMUM_32BPP;
 213         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
 214                 if (job->cbufs[i]) {
 215                         struct v3d_surface *surf = v3d_surface(job->cbufs[i]);
 216                         max_bpp = MAX2(max_bpp, surf->internal_bpp);
 217                 }
 218         }
 219         job->internal_bpp = max_bpp;
 220         STATIC_ASSERT(RENDER_TARGET_MAXIMUM_32BPP == 0);
 221         tile_size_index += max_bpp;
 222
 223         assert(tile_size_index < ARRAY_SIZE(tile_sizes));
 224         job->tile_width = tile_sizes[tile_size_index * 2 + 0];
 225         job->tile_height = tile_sizes[tile_size_index * 2 + 1];
 226 }
 227
 228 /**
 229  * Returns a v3d_job struture for tracking V3D rendering to a particular FBO.
 230  *
 231  * If we've already started rendering to this FBO, then return the same job,
 232  * otherwise make a new one.  If we're beginning rendering to an FBO, make
 233  * sure that any previous reads of the FBO (or writes to its color/Z surfaces)
 234  * have been flushed.
 235  */
 236 struct v3d_job *
 237 v3d_get_job(struct v3d_context *v3d,
 238             struct pipe_surface **cbufs, struct pipe_surface *zsbuf)
 239 {
 240         /* Return the existing job for this FBO if we have one */
 241         struct v3d_job_key local_key = {
 242                 .cbufs = {
 243                         cbufs[0],
 244                         cbufs[1],
 245                         cbufs[2],
 246                         cbufs[3],
 247                 },
 248                 .zsbuf = zsbuf,
 249         };
 250         struct hash_entry *entry = _mesa_hash_table_search(v3d->jobs,
 251                                                            &local_key);
 252         if (entry)
 253                 return entry->data;
 254
 255         /* Creating a new job.  Make sure that any previous jobs reading or
 256          * writing these buffers are flushed.
 257          */
 258         struct v3d_job *job = v3d_job_create(v3d);
 259
 260         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
 261                 if (cbufs[i]) {
 262                         v3d_flush_jobs_reading_resource(v3d, cbufs[i]->texture);
 263                         pipe_surface_reference(&job->cbufs[i], cbufs[i]);
 264
 265                         if (cbufs[i]->texture->nr_samples > 1)
 266                                 job->msaa = true;
 267                 }
 268         }
 269         if (zsbuf) {
 270                 v3d_flush_jobs_reading_resource(v3d, zsbuf->texture);
 271                 pipe_surface_reference(&job->zsbuf, zsbuf);
 272                 if (zsbuf->texture->nr_samples > 1)
 273                         job->msaa = true;
 274         }
 275
 276         for (int i = 0; i < V3D_MAX_DRAW_BUFFERS; i++) {
 277                 if (cbufs[i])
 278                         _mesa_hash_table_insert(v3d->write_jobs,
 279                                                 cbufs[i]->texture, job);
 280         }
 281         if (zsbuf) {
 282                 _mesa_hash_table_insert(v3d->write_jobs, zsbuf->texture, job);
 283
 284                 struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
 285                 if (rsc->separate_stencil) {
 286                         v3d_flush_jobs_reading_resource(v3d,
 287                                                         &rsc->separate_stencil->base);
 288                         _mesa_hash_table_insert(v3d->write_jobs,
 289                                                 &rsc->separate_stencil->base,
 290                                                 job);
 291                 }
 292         }
 293
 294         memcpy(&job->key, &local_key, sizeof(local_key));
 295         _mesa_hash_table_insert(v3d->jobs, &job->key, job);
 296
 297         return job;
 298 }
 299
 300 struct v3d_job *
 301 v3d_get_job_for_fbo(struct v3d_context *v3d)
 302 {
 303         if (v3d->job)
 304                 return v3d->job;
 305
 306         struct pipe_surface **cbufs = v3d->framebuffer.cbufs;
 307         struct pipe_surface *zsbuf = v3d->framebuffer.zsbuf;
 308         struct v3d_job *job = v3d_get_job(v3d, cbufs, zsbuf);
 309
 310         if (v3d->framebuffer.samples >= 1)
 311                 job->msaa = true;
 312
 313         v3d_job_set_tile_buffer_size(job);
 314
 315         /* The dirty flags are tracking what's been updated while v3d->job has
 316          * been bound, so set them all to ~0 when switching between jobs.  We
 317          * also need to reset all state at the start of rendering.
 318          */
 319         v3d->dirty = ~0;
 320
 321         /* If we're binding to uninitialized buffers, no need to load their
 322          * contents before drawing.
 323          */
 324         for (int i = 0; i < 4; i++) {
 325                 if (cbufs[i]) {
 326                         struct v3d_resource *rsc = v3d_resource(cbufs[i]->texture);
 327                         if (!rsc->writes)
 328                                 job->clear |= PIPE_CLEAR_COLOR0 << i;
 329                 }
 330         }
 331
 332         if (zsbuf) {
 333                 struct v3d_resource *rsc = v3d_resource(zsbuf->texture);
 334                 if (!rsc->writes)
 335                         job->clear |= PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL;
 336         }
 337
 338         job->draw_tiles_x = DIV_ROUND_UP(v3d->framebuffer.width,
 339                                          job->tile_width);
 340         job->draw_tiles_y = DIV_ROUND_UP(v3d->framebuffer.height,
 341                                          job->tile_height);
 342
 343         v3d->job = job;
 344
 345         return job;
 346 }
 347
 348 static void
 349 v3d_clif_dump(struct v3d_context *v3d, struct v3d_job *job)
 350 {
 351         if (!(V3D_DEBUG & (V3D_DEBUG_CL | V3D_DEBUG_CLIF)))
 352                 return;
 353
 354         struct clif_dump *clif = clif_dump_init(&v3d->screen->devinfo,
 355                                                 stderr,
 356                                                 V3D_DEBUG & V3D_DEBUG_CL);
 357
 358         set_foreach(job->bos, entry) {
 359                 struct v3d_bo *bo = (void *)entry->key;
 360                 char *name = ralloc_asprintf(NULL, "%s_0x%x",
 361                                              bo->name, bo->offset);
 362
 363                 v3d_bo_map(bo);
 364                 clif_dump_add_bo(clif, name, bo->offset, bo->size, bo->map);
 365
 366                 ralloc_free(name);
 367         }
 368
 369         clif_dump(clif, &job->submit);
 370
 371         clif_dump_destroy(clif);
 372 }
 373
 374 /**
 375  * Submits the job to the kernel and then reinitializes it.
 376  */
 377 void
 378 v3d_job_submit(struct v3d_context *v3d, struct v3d_job *job)
 379 {
 380         MAYBE_UNUSED struct v3d_screen *screen = v3d->screen;
 381
 382         if (!job->needs_flush)
 383                 goto done;
 384
 385         if (v3d->screen->devinfo.ver >= 41)
 386                 v3d41_emit_rcl(job);
 387         else
 388                 v3d33_emit_rcl(job);
 389
 390         if (cl_offset(&job->bcl) > 0) {
 391                 if (screen->devinfo.ver >= 41)
 392                         v3d41_bcl_epilogue(v3d, job);
 393                 else
 394                         v3d33_bcl_epilogue(v3d, job);
 395         }
 396
 397         /* While the RCL will implicitly depend on the last RCL to have
 398          * finished, we also need to block on any previous TFU job we may have
 399          * dispatched.
 400          */
 401         job->submit.in_sync_rcl = v3d->out_sync;
 402
 403         /* Update the sync object for the last rendering by our context. */
 404         job->submit.out_sync = v3d->out_sync;
 405
 406         job->submit.bcl_end = job->bcl.bo->offset + cl_offset(&job->bcl);
 407         job->submit.rcl_end = job->rcl.bo->offset + cl_offset(&job->rcl);
 408
 409         /* On V3D 4.1, the tile alloc/state setup moved to register writes
 410          * instead of binner packets.
 411          */
 412         if (screen->devinfo.ver >= 41) {
 413                 v3d_job_add_bo(job, job->tile_alloc);
 414                 job->submit.qma = job->tile_alloc->offset;
 415                 job->submit.qms = job->tile_alloc->size;
 416
 417                 v3d_job_add_bo(job, job->tile_state);
 418                 job->submit.qts = job->tile_state->offset;
 419         }
 420
 421         v3d_clif_dump(v3d, job);
 422
 423         if (!(V3D_DEBUG & V3D_DEBUG_NORAST)) {
 424                 int ret;
 425
 426                 ret = v3d_ioctl(v3d->fd, DRM_IOCTL_V3D_SUBMIT_CL, &job->submit);
 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 done:
 436         v3d_job_free(v3d, job);
 437 }
 438
 439 static bool
 440 v3d_job_compare(const void *a, const void *b)
 441 {
 442         return memcmp(a, b, sizeof(struct v3d_job_key)) == 0;
 443 }
 444
 445 static uint32_t
 446 v3d_job_hash(const void *key)
 447 {
 448         return _mesa_hash_data(key, sizeof(struct v3d_job_key));
 449 }
 450
 451 void
 452 v3d_job_init(struct v3d_context *v3d)
 453 {
 454         v3d->jobs = _mesa_hash_table_create(v3d,
 455                                             v3d_job_hash,
 456                                             v3d_job_compare);
 457         v3d->write_jobs = _mesa_hash_table_create(v3d,
 458                                                   _mesa_hash_pointer,
 459                                                   _mesa_key_pointer_equal);
 460 }
 461