src/gallium/drivers/v3d/v3dx_draw.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 #include "util/u_blitter.h"
  25 #include "util/u_prim.h"
  26 #include "util/u_format.h"
  27 #include "util/u_pack_color.h"
  28 #include "util/u_prim_restart.h"
  29 #include "util/u_upload_mgr.h"
  30 #include "indices/u_primconvert.h"
  31
  32 #include "v3d_context.h"
  33 #include "v3d_resource.h"
  34 #include "v3d_cl.h"
  35 #include "broadcom/compiler/v3d_compiler.h"
  36 #include "broadcom/common/v3d_macros.h"
  37 #include "broadcom/cle/v3dx_pack.h"
  38
  39 /**
  40  * Does the initial bining command list setup for drawing to a given FBO.
  41  */
  42 static void
  43 v3d_start_draw(struct v3d_context *v3d)
  44 {
  45         struct v3d_job *job = v3d->job;
  46
  47         if (job->needs_flush)
  48                 return;
  49
  50         /* Get space to emit our BCL state, using a branch to jump to a new BO
  51          * if necessary.
  52          */
  53         v3d_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
  54
  55         job->submit.bcl_start = job->bcl.bo->offset;
  56         v3d_job_add_bo(job, job->bcl.bo);
  57
  58         job->tile_alloc = v3d_bo_alloc(v3d->screen, 1024 * 1024, "tile_alloc");
  59         uint32_t tsda_per_tile_size = v3d->screen->devinfo.ver >= 40 ? 256 : 64;
  60         job->tile_state = v3d_bo_alloc(v3d->screen,
  61                                        job->draw_tiles_y *
  62                                        job->draw_tiles_x *
  63                                        tsda_per_tile_size,
  64                                        "TSDA");
  65
  66 #if V3D_VERSION >= 40
  67         cl_emit(&job->bcl, TILE_BINNING_MODE_CFG, config) {
  68                 config.width_in_pixels = v3d->framebuffer.width;
  69                 config.height_in_pixels = v3d->framebuffer.height;
  70                 config.number_of_render_targets =
  71                         MAX2(v3d->framebuffer.nr_cbufs, 1);
  72
  73                 config.multisample_mode_4x = job->msaa;
  74
  75                 config.maximum_bpp_of_all_render_targets = job->internal_bpp;
  76         }
  77 #else /* V3D_VERSION < 40 */
  78         /* "Binning mode lists start with a Tile Binning Mode Configuration
  79          * item (120)"
  80          *
  81          * Part1 signals the end of binning config setup.
  82          */
  83         cl_emit(&job->bcl, TILE_BINNING_MODE_CFG_PART2, config) {
  84                 config.tile_allocation_memory_address =
  85                         cl_address(job->tile_alloc, 0);
  86                 config.tile_allocation_memory_size = job->tile_alloc->size;
  87         }
  88
  89         cl_emit(&job->bcl, TILE_BINNING_MODE_CFG_PART1, config) {
  90                 config.tile_state_data_array_base_address =
  91                         cl_address(job->tile_state, 0);
  92
  93                 config.width_in_tiles = job->draw_tiles_x;
  94                 config.height_in_tiles = job->draw_tiles_y;
  95                 /* Must be >= 1 */
  96                 config.number_of_render_targets =
  97                         MAX2(v3d->framebuffer.nr_cbufs, 1);
  98
  99                 config.multisample_mode_4x = job->msaa;
 100
 101                 config.maximum_bpp_of_all_render_targets = job->internal_bpp;
 102         }
 103 #endif /* V3D_VERSION < 40 */
 104
 105         /* There's definitely nothing in the VCD cache we want. */
 106         cl_emit(&job->bcl, FLUSH_VCD_CACHE, bin);
 107
 108         /* Disable any leftover OQ state from another job. */
 109         cl_emit(&job->bcl, OCCLUSION_QUERY_COUNTER, counter);
 110
 111         /* "Binning mode lists must have a Start Tile Binning item (6) after
 112          *  any prefix state data before the binning list proper starts."
 113          */
 114         cl_emit(&job->bcl, START_TILE_BINNING, bin);
 115
 116         job->needs_flush = true;
 117         job->draw_width = v3d->framebuffer.width;
 118         job->draw_height = v3d->framebuffer.height;
 119 }
 120
 121 static void
 122 v3d_predraw_check_textures(struct pipe_context *pctx,
 123                            struct v3d_texture_stateobj *stage_tex)
 124 {
 125         struct v3d_context *v3d = v3d_context(pctx);
 126
 127         for (int i = 0; i < stage_tex->num_textures; i++) {
 128                 struct pipe_sampler_view *view = stage_tex->textures[i];
 129                 if (!view)
 130                         continue;
 131
 132                 v3d_flush_jobs_writing_resource(v3d, view->texture);
 133         }
 134 }
 135
 136 static void
 137 v3d_emit_gl_shader_state(struct v3d_context *v3d,
 138                          const struct pipe_draw_info *info)
 139 {
 140         struct v3d_job *job = v3d->job;
 141         /* VC5_DIRTY_VTXSTATE */
 142         struct v3d_vertex_stateobj *vtx = v3d->vtx;
 143         /* VC5_DIRTY_VTXBUF */
 144         struct v3d_vertexbuf_stateobj *vertexbuf = &v3d->vertexbuf;
 145
 146         /* Upload the uniforms to the indirect CL first */
 147         struct v3d_cl_reloc fs_uniforms =
 148                 v3d_write_uniforms(v3d, v3d->prog.fs,
 149                                    PIPE_SHADER_FRAGMENT);
 150         struct v3d_cl_reloc vs_uniforms =
 151                 v3d_write_uniforms(v3d, v3d->prog.vs,
 152                                    PIPE_SHADER_VERTEX);
 153         struct v3d_cl_reloc cs_uniforms =
 154                 v3d_write_uniforms(v3d, v3d->prog.cs,
 155                                    PIPE_SHADER_VERTEX);
 156
 157         /* See GFXH-930 workaround below */
 158         uint32_t num_elements_to_emit = MAX2(vtx->num_elements, 1);
 159         uint32_t shader_rec_offset =
 160                 v3d_cl_ensure_space(&job->indirect,
 161                                     cl_packet_length(GL_SHADER_STATE_RECORD) +
 162                                     num_elements_to_emit *
 163                                     cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
 164                                     32);
 165
 166         cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
 167                 shader.enable_clipping = true;
 168                 /* VC5_DIRTY_PRIM_MODE | VC5_DIRTY_RASTERIZER */
 169                 shader.point_size_in_shaded_vertex_data =
 170                         (info->mode == PIPE_PRIM_POINTS &&
 171                          v3d->rasterizer->base.point_size_per_vertex);
 172
 173                 /* Must be set if the shader modifies Z, discards, or modifies
 174                  * the sample mask.  For any of these cases, the fragment
 175                  * shader needs to write the Z value (even just discards).
 176                  */
 177                 shader.fragment_shader_does_z_writes =
 178                         (v3d->prog.fs->prog_data.fs->writes_z ||
 179                          v3d->prog.fs->prog_data.fs->discard);
 180
 181                 shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
 182                         v3d->prog.fs->prog_data.fs->uses_center_w;
 183
 184                 shader.number_of_varyings_in_fragment_shader =
 185                         v3d->prog.fs->prog_data.base->num_inputs;
 186
 187                 shader.coordinate_shader_propagate_nans = true;
 188                 shader.vertex_shader_propagate_nans = true;
 189                 shader.fragment_shader_propagate_nans = true;
 190
 191                 shader.coordinate_shader_code_address =
 192                         cl_address(v3d_resource(v3d->prog.cs->resource)->bo,
 193                                    v3d->prog.cs->offset);
 194                 shader.vertex_shader_code_address =
 195                         cl_address(v3d_resource(v3d->prog.vs->resource)->bo,
 196                                    v3d->prog.vs->offset);
 197                 shader.fragment_shader_code_address =
 198                         cl_address(v3d_resource(v3d->prog.fs->resource)->bo,
 199                                    v3d->prog.fs->offset);
 200
 201                 /* XXX: Use combined input/output size flag in the common
 202                  * case.
 203                  */
 204                 shader.coordinate_shader_has_separate_input_and_output_vpm_blocks =
 205                         v3d->prog.cs->prog_data.vs->separate_segments;
 206                 shader.vertex_shader_has_separate_input_and_output_vpm_blocks =
 207                         v3d->prog.vs->prog_data.vs->separate_segments;
 208
 209                 shader.coordinate_shader_input_vpm_segment_size =
 210                         v3d->prog.cs->prog_data.vs->vpm_input_size;
 211                 shader.vertex_shader_input_vpm_segment_size =
 212                         v3d->prog.vs->prog_data.vs->vpm_input_size;
 213
 214                 shader.coordinate_shader_output_vpm_segment_size =
 215                         v3d->prog.cs->prog_data.vs->vpm_output_size;
 216                 shader.vertex_shader_output_vpm_segment_size =
 217                         v3d->prog.vs->prog_data.vs->vpm_output_size;
 218
 219                 shader.coordinate_shader_uniforms_address = cs_uniforms;
 220                 shader.vertex_shader_uniforms_address = vs_uniforms;
 221                 shader.fragment_shader_uniforms_address = fs_uniforms;
 222
 223 #if V3D_VERSION >= 41
 224                 shader.min_coord_shader_input_segments_required_in_play = 1;
 225                 shader.min_vertex_shader_input_segments_required_in_play = 1;
 226
 227                 shader.coordinate_shader_4_way_threadable =
 228                         v3d->prog.cs->prog_data.vs->base.threads == 4;
 229                 shader.vertex_shader_4_way_threadable =
 230                         v3d->prog.vs->prog_data.vs->base.threads == 4;
 231                 shader.fragment_shader_4_way_threadable =
 232                         v3d->prog.fs->prog_data.fs->base.threads == 4;
 233
 234                 shader.coordinate_shader_start_in_final_thread_section =
 235                         v3d->prog.cs->prog_data.vs->base.single_seg;
 236                 shader.vertex_shader_start_in_final_thread_section =
 237                         v3d->prog.vs->prog_data.vs->base.single_seg;
 238                 shader.fragment_shader_start_in_final_thread_section =
 239                         v3d->prog.fs->prog_data.fs->base.single_seg;
 240 #else
 241                 shader.coordinate_shader_4_way_threadable =
 242                         v3d->prog.cs->prog_data.vs->base.threads == 4;
 243                 shader.coordinate_shader_2_way_threadable =
 244                         v3d->prog.cs->prog_data.vs->base.threads == 2;
 245                 shader.vertex_shader_4_way_threadable =
 246                         v3d->prog.vs->prog_data.vs->base.threads == 4;
 247                 shader.vertex_shader_2_way_threadable =
 248                         v3d->prog.vs->prog_data.vs->base.threads == 2;
 249                 shader.fragment_shader_4_way_threadable =
 250                         v3d->prog.fs->prog_data.fs->base.threads == 4;
 251                 shader.fragment_shader_2_way_threadable =
 252                         v3d->prog.fs->prog_data.fs->base.threads == 2;
 253 #endif
 254
 255                 shader.vertex_id_read_by_coordinate_shader =
 256                         v3d->prog.cs->prog_data.vs->uses_vid;
 257                 shader.instance_id_read_by_coordinate_shader =
 258                         v3d->prog.cs->prog_data.vs->uses_iid;
 259                 shader.vertex_id_read_by_vertex_shader =
 260                         v3d->prog.vs->prog_data.vs->uses_vid;
 261                 shader.instance_id_read_by_vertex_shader =
 262                         v3d->prog.vs->prog_data.vs->uses_iid;
 263
 264                 shader.address_of_default_attribute_values =
 265                         cl_address(v3d_resource(vtx->defaults)->bo,
 266                                    vtx->defaults_offset);
 267         }
 268
 269         for (int i = 0; i < vtx->num_elements; i++) {
 270                 struct pipe_vertex_element *elem = &vtx->pipe[i];
 271                 struct pipe_vertex_buffer *vb =
 272                         &vertexbuf->vb[elem->vertex_buffer_index];
 273                 struct v3d_resource *rsc = v3d_resource(vb->buffer.resource);
 274
 275                 const uint32_t size =
 276                         cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
 277                 cl_emit_with_prepacked(&job->indirect,
 278                                        GL_SHADER_STATE_ATTRIBUTE_RECORD,
 279                                        &vtx->attrs[i * size], attr) {
 280                         attr.stride = vb->stride;
 281                         attr.address = cl_address(rsc->bo,
 282                                                   vb->buffer_offset +
 283                                                   elem->src_offset);
 284                         attr.number_of_values_read_by_coordinate_shader =
 285                                 v3d->prog.cs->prog_data.vs->vattr_sizes[i];
 286                         attr.number_of_values_read_by_vertex_shader =
 287                                 v3d->prog.vs->prog_data.vs->vattr_sizes[i];
 288 #if V3D_VERSION >= 41
 289                         attr.maximum_index = 0xffffff;
 290 #endif
 291                 }
 292                 STATIC_ASSERT(sizeof(vtx->attrs) >= VC5_MAX_ATTRIBUTES * size);
 293         }
 294
 295         if (vtx->num_elements == 0) {
 296                 /* GFXH-930: At least one attribute must be enabled and read
 297                  * by CS and VS.  If we have no attributes being consumed by
 298                  * the shader, set up a dummy to be loaded into the VPM.
 299                  */
 300                 cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
 301                         /* Valid address of data whose value will be unused. */
 302                         attr.address = cl_address(job->indirect.bo, 0);
 303
 304                         attr.type = ATTRIBUTE_FLOAT;
 305                         attr.stride = 0;
 306                         attr.vec_size = 1;
 307
 308                         attr.number_of_values_read_by_coordinate_shader = 1;
 309                         attr.number_of_values_read_by_vertex_shader = 1;
 310                 }
 311         }
 312
 313         cl_emit(&job->bcl, VCM_CACHE_SIZE, vcm) {
 314                 vcm.number_of_16_vertex_batches_for_binning =
 315                         v3d->prog.cs->prog_data.vs->vcm_cache_size;
 316                 vcm.number_of_16_vertex_batches_for_rendering =
 317                         v3d->prog.vs->prog_data.vs->vcm_cache_size;
 318         }
 319
 320         cl_emit(&job->bcl, GL_SHADER_STATE, state) {
 321                 state.address = cl_address(job->indirect.bo, shader_rec_offset);
 322                 state.number_of_attribute_arrays = num_elements_to_emit;
 323         }
 324
 325         v3d_bo_unreference(&cs_uniforms.bo);
 326         v3d_bo_unreference(&vs_uniforms.bo);
 327         v3d_bo_unreference(&fs_uniforms.bo);
 328
 329         job->shader_rec_count++;
 330 }
 331
 332 /**
 333  * Computes the various transform feedback statistics, since they can't be
 334  * recorded by CL packets.
 335  */
 336 static void
 337 v3d_tf_statistics_record(struct v3d_context *v3d,
 338                          const struct pipe_draw_info *info,
 339                          bool prim_tf)
 340 {
 341         if (!v3d->active_queries)
 342                 return;
 343
 344         uint32_t prims = u_prims_for_vertices(info->mode, info->count);
 345         v3d->prims_generated += prims;
 346
 347         if (prim_tf) {
 348                 /* XXX: Only count if we didn't overflow. */
 349                 v3d->tf_prims_generated += prims;
 350         }
 351 }
 352
 353 static void
 354 v3d_update_job_ez(struct v3d_context *v3d, struct v3d_job *job)
 355 {
 356         switch (v3d->zsa->ez_state) {
 357         case VC5_EZ_UNDECIDED:
 358                 /* If the Z/S state didn't pick a direction but didn't
 359                  * disable, then go along with the current EZ state.  This
 360                  * allows EZ optimization for Z func == EQUAL or NEVER.
 361                  */
 362                 break;
 363
 364         case VC5_EZ_LT_LE:
 365         case VC5_EZ_GT_GE:
 366                 /* If the Z/S state picked a direction, then it needs to match
 367                  * the current direction if we've decided on one.
 368                  */
 369                 if (job->ez_state == VC5_EZ_UNDECIDED)
 370                         job->ez_state = v3d->zsa->ez_state;
 371                 else if (job->ez_state != v3d->zsa->ez_state)
 372                         job->ez_state = VC5_EZ_DISABLED;
 373                 break;
 374
 375         case VC5_EZ_DISABLED:
 376                 /* If the current Z/S state disables EZ because of a bad Z
 377                  * func or stencil operation, then we can't do any more EZ in
 378                  * this frame.
 379                  */
 380                 job->ez_state = VC5_EZ_DISABLED;
 381                 break;
 382         }
 383
 384         /* If the FS affects the Z of the pixels, then it may update against
 385          * the chosen EZ direction (though we could use
 386          * ARB_conservative_depth's hints to avoid this)
 387          */
 388         if (v3d->prog.fs->prog_data.fs->writes_z) {
 389                 job->ez_state = VC5_EZ_DISABLED;
 390         }
 391
 392         if (job->first_ez_state == VC5_EZ_UNDECIDED &&
 393             (job->ez_state != VC5_EZ_DISABLED || job->draw_calls_queued == 0))
 394                 job->first_ez_state = job->ez_state;
 395 }
 396
 397 static void
 398 v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 399 {
 400         struct v3d_context *v3d = v3d_context(pctx);
 401
 402         if (!info->count_from_stream_output && !info->indirect &&
 403             !info->primitive_restart &&
 404             !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
 405                 return;
 406
 407         /* Fall back for weird desktop GL primitive restart values. */
 408         if (info->primitive_restart &&
 409             info->index_size) {
 410                 uint32_t mask = ~0;
 411
 412                 switch (info->index_size) {
 413                 case 2:
 414                         mask = 0xffff;
 415                         break;
 416                 case 1:
 417                         mask = 0xff;
 418                         break;
 419                 }
 420
 421                 if (info->restart_index != mask) {
 422                         util_draw_vbo_without_prim_restart(pctx, info);
 423                         return;
 424                 }
 425         }
 426
 427         if (info->mode >= PIPE_PRIM_QUADS) {
 428                 util_primconvert_save_rasterizer_state(v3d->primconvert, &v3d->rasterizer->base);
 429                 util_primconvert_draw_vbo(v3d->primconvert, info);
 430                 perf_debug("Fallback conversion for %d %s vertices\n",
 431                            info->count, u_prim_name(info->mode));
 432                 return;
 433         }
 434
 435         /* Before setting up the draw, flush anything writing to the textures
 436          * that we read from.
 437          */
 438         for (int s = 0; s < PIPE_SHADER_TYPES; s++)
 439                 v3d_predraw_check_textures(pctx, &v3d->tex[s]);
 440
 441         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 442
 443         /* If vertex texturing depends on the output of rendering, we need to
 444          * ensure that that rendering is complete before we run a coordinate
 445          * shader that depends on it.
 446          *
 447          * Given that doing that is unusual, for now we just block the binner
 448          * on the last submitted render, rather than tracking the last
 449          * rendering to each texture's BO.
 450          */
 451         if (v3d->tex[PIPE_SHADER_VERTEX].num_textures) {
 452                 perf_debug("Blocking binner on last render "
 453                            "due to vertex texturing.\n");
 454                 job->submit.in_sync_bcl = v3d->out_sync;
 455         }
 456
 457         /* Get space to emit our draw call into the BCL, using a branch to
 458          * jump to a new BO if necessary.
 459          */
 460         v3d_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
 461
 462         if (v3d->prim_mode != info->mode) {
 463                 v3d->prim_mode = info->mode;
 464                 v3d->dirty |= VC5_DIRTY_PRIM_MODE;
 465         }
 466
 467         v3d_start_draw(v3d);
 468         v3d_update_compiled_shaders(v3d, info->mode);
 469         v3d_update_job_ez(v3d, job);
 470
 471 #if V3D_VERSION >= 41
 472         v3d41_emit_state(pctx);
 473 #else
 474         v3d33_emit_state(pctx);
 475 #endif
 476
 477         if (v3d->dirty & (VC5_DIRTY_VTXBUF |
 478                           VC5_DIRTY_VTXSTATE |
 479                           VC5_DIRTY_PRIM_MODE |
 480                           VC5_DIRTY_RASTERIZER |
 481                           VC5_DIRTY_COMPILED_CS |
 482                           VC5_DIRTY_COMPILED_VS |
 483                           VC5_DIRTY_COMPILED_FS |
 484                           v3d->prog.cs->uniform_dirty_bits |
 485                           v3d->prog.vs->uniform_dirty_bits |
 486                           v3d->prog.fs->uniform_dirty_bits)) {
 487                 v3d_emit_gl_shader_state(v3d, info);
 488         }
 489
 490         v3d->dirty = 0;
 491
 492         /* The Base Vertex/Base Instance packet sets those values to nonzero
 493          * for the next draw call only.
 494          */
 495         if (info->index_bias || info->start_instance) {
 496                 cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
 497                         base.base_instance = info->start_instance;
 498                         base.base_vertex = info->index_bias;
 499                 }
 500         }
 501
 502         uint32_t prim_tf_enable = 0;
 503 #if V3D_VERSION < 40
 504         /* V3D 3.x: The HW only processes transform feedback on primitives
 505          * with the flag set.
 506          */
 507         if (v3d->streamout.num_targets)
 508                 prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
 509 #endif
 510
 511         v3d_tf_statistics_record(v3d, info, v3d->streamout.num_targets);
 512
 513         /* Note that the primitive type fields match with OpenGL/gallium
 514          * definitions, up to but not including QUADS.
 515          */
 516         if (info->index_size) {
 517                 uint32_t index_size = info->index_size;
 518                 uint32_t offset = info->start * index_size;
 519                 struct pipe_resource *prsc;
 520                 if (info->has_user_indices) {
 521                         prsc = NULL;
 522                         u_upload_data(v3d->uploader, 0,
 523                                       info->count * info->index_size, 4,
 524                                       info->index.user,
 525                                       &offset, &prsc);
 526                 } else {
 527                         prsc = info->index.resource;
 528                 }
 529                 struct v3d_resource *rsc = v3d_resource(prsc);
 530
 531 #if V3D_VERSION >= 40
 532                 cl_emit(&job->bcl, INDEX_BUFFER_SETUP, ib) {
 533                         ib.address = cl_address(rsc->bo, 0);
 534                         ib.size = rsc->bo->size;
 535                 }
 536 #endif
 537
 538                 if (info->instance_count > 1) {
 539                         cl_emit(&job->bcl, INDEXED_INSTANCED_PRIM_LIST, prim) {
 540                                 prim.index_type = ffs(info->index_size) - 1;
 541 #if V3D_VERSION >= 40
 542                                 prim.index_offset = offset;
 543 #else /* V3D_VERSION < 40 */
 544                                 prim.maximum_index = (1u << 31) - 1; /* XXX */
 545                                 prim.address_of_indices_list =
 546                                         cl_address(rsc->bo, offset);
 547 #endif /* V3D_VERSION < 40 */
 548                                 prim.mode = info->mode | prim_tf_enable;
 549                                 prim.enable_primitive_restarts = info->primitive_restart;
 550
 551                                 prim.number_of_instances = info->instance_count;
 552                                 prim.instance_length = info->count;
 553                         }
 554                 } else {
 555                         cl_emit(&job->bcl, INDEXED_PRIM_LIST, prim) {
 556                                 prim.index_type = ffs(info->index_size) - 1;
 557                                 prim.length = info->count;
 558 #if V3D_VERSION >= 40
 559                                 prim.index_offset = offset;
 560 #else /* V3D_VERSION < 40 */
 561                                 prim.maximum_index = (1u << 31) - 1; /* XXX */
 562                                 prim.address_of_indices_list =
 563                                         cl_address(rsc->bo, offset);
 564 #endif /* V3D_VERSION < 40 */
 565                                 prim.mode = info->mode | prim_tf_enable;
 566                                 prim.enable_primitive_restarts = info->primitive_restart;
 567                         }
 568                 }
 569
 570                 job->draw_calls_queued++;
 571
 572                 if (info->has_user_indices)
 573                         pipe_resource_reference(&prsc, NULL);
 574         } else {
 575                 if (info->instance_count > 1) {
 576                         cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMS, prim) {
 577                                 prim.mode = info->mode | prim_tf_enable;
 578                                 prim.index_of_first_vertex = info->start;
 579                                 prim.number_of_instances = info->instance_count;
 580                                 prim.instance_length = info->count;
 581                         }
 582                 } else {
 583                         cl_emit(&job->bcl, VERTEX_ARRAY_PRIMS, prim) {
 584                                 prim.mode = info->mode | prim_tf_enable;
 585                                 prim.length = info->count;
 586                                 prim.index_of_first_vertex = info->start;
 587                         }
 588                 }
 589         }
 590
 591         /* A flush is required in between a TF draw and any following TF specs
 592          * packet, or the GPU may hang.  Just flush each time for now.
 593          */
 594         if (v3d->streamout.num_targets)
 595                 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_FLUSH_AND_COUNT, flush);
 596
 597         job->draw_calls_queued++;
 598
 599         /* Increment the TF offsets by how many verts we wrote.  XXX: This
 600          * needs some clamping to the buffer size.
 601          */
 602         for (int i = 0; i < v3d->streamout.num_targets; i++)
 603                 v3d->streamout.offsets[i] += info->count;
 604
 605         if (v3d->zsa && job->zsbuf && v3d->zsa->base.depth.enabled) {
 606                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
 607                 v3d_job_add_bo(job, rsc->bo);
 608
 609                 job->load |= PIPE_CLEAR_DEPTH & ~job->clear;
 610                 if (v3d->zsa->base.depth.writemask)
 611                         job->store |= PIPE_CLEAR_DEPTH;
 612                 rsc->initialized_buffers = PIPE_CLEAR_DEPTH;
 613         }
 614
 615         if (v3d->zsa && job->zsbuf && v3d->zsa->base.stencil[0].enabled) {
 616                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
 617                 if (rsc->separate_stencil)
 618                         rsc = rsc->separate_stencil;
 619
 620                 v3d_job_add_bo(job, rsc->bo);
 621
 622                 job->load |= PIPE_CLEAR_STENCIL & ~job->clear;
 623                 if (v3d->zsa->base.stencil[0].writemask ||
 624                     v3d->zsa->base.stencil[1].writemask) {
 625                         job->store |= PIPE_CLEAR_STENCIL;
 626                 }
 627                 rsc->initialized_buffers |= PIPE_CLEAR_STENCIL;
 628         }
 629
 630         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 631                 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
 632                 int blend_rt = v3d->blend->base.independent_blend_enable ? i : 0;
 633
 634                 if (job->store & bit || !job->cbufs[i])
 635                         continue;
 636                 struct v3d_resource *rsc = v3d_resource(job->cbufs[i]->texture);
 637
 638                 job->load |= bit & ~job->clear;
 639                 if (v3d->blend->base.rt[blend_rt].colormask)
 640                         job->store |= bit;
 641                 v3d_job_add_bo(job, rsc->bo);
 642         }
 643
 644         if (job->referenced_size > 768 * 1024 * 1024) {
 645                 perf_debug("Flushing job with %dkb to try to free up memory\n",
 646                         job->referenced_size / 1024);
 647                 v3d_flush(pctx);
 648         }
 649
 650         if (V3D_DEBUG & V3D_DEBUG_ALWAYS_FLUSH)
 651                 v3d_flush(pctx);
 652 }
 653
 654 /**
 655  * Implements gallium's clear() hook (glClear()) by drawing a pair of triangles.
 656  */
 657 static void
 658 v3d_draw_clear(struct v3d_context *v3d,
 659                unsigned buffers,
 660                const union pipe_color_union *color,
 661                double depth, unsigned stencil)
 662 {
 663         static const union pipe_color_union dummy_color = {};
 664
 665         /* The blitter util dereferences the color regardless, even though the
 666          * gallium clear API may not pass one in when only Z/S are cleared.
 667          */
 668         if (!color)
 669                 color = &dummy_color;
 670
 671         v3d_blitter_save(v3d);
 672         util_blitter_clear(v3d->blitter,
 673                            v3d->framebuffer.width,
 674                            v3d->framebuffer.height,
 675                            util_framebuffer_get_num_layers(&v3d->framebuffer),
 676                            buffers, color, depth, stencil);
 677 }
 678
 679 /**
 680  * Attempts to perform the GL clear by using the TLB's fast clear at the start
 681  * of the frame.
 682  */
 683 static unsigned
 684 v3d_tlb_clear(struct v3d_job *job, unsigned buffers,
 685               const union pipe_color_union *color,
 686               double depth, unsigned stencil)
 687 {
 688         struct v3d_context *v3d = job->v3d;
 689
 690         if (job->draw_calls_queued) {
 691                 /* If anything in the CL has drawn using the buffer, then the
 692                  * TLB clear we're trying to add now would happen before that
 693                  * drawing.
 694                  */
 695                 buffers &= ~(job->load | job->store);
 696         }
 697
 698         /* GFXH-1461: If we were to emit a load of just depth or just stencil,
 699          * then the clear for the other may get lost.  We need to decide now
 700          * if it would be possible to need to emit a load of just one after
 701          * we've set up our TLB clears.
 702          */
 703         if (buffers & PIPE_CLEAR_DEPTHSTENCIL &&
 704             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL &&
 705             job->zsbuf &&
 706             util_format_is_depth_and_stencil(job->zsbuf->texture->format)) {
 707                 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 708         }
 709
 710         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 711                 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
 712                 if (!(buffers & bit))
 713                         continue;
 714
 715                 struct pipe_surface *psurf = v3d->framebuffer.cbufs[i];
 716                 struct v3d_surface *surf = v3d_surface(psurf);
 717                 struct v3d_resource *rsc = v3d_resource(psurf->texture);
 718
 719                 union util_color uc;
 720                 uint32_t internal_size = 4 << surf->internal_bpp;
 721
 722                 static union pipe_color_union swapped_color;
 723                 if (v3d->swap_color_rb & (1 << i)) {
 724                         swapped_color.f[0] = color->f[2];
 725                         swapped_color.f[1] = color->f[1];
 726                         swapped_color.f[2] = color->f[0];
 727                         swapped_color.f[3] = color->f[3];
 728                         color = &swapped_color;
 729                 }
 730
 731                 switch (surf->internal_type) {
 732                 case V3D_INTERNAL_TYPE_8:
 733                         util_pack_color(color->f, PIPE_FORMAT_R8G8B8A8_UNORM,
 734                                         &uc);
 735                         memcpy(job->clear_color[i], uc.ui, internal_size);
 736                         break;
 737                 case V3D_INTERNAL_TYPE_8I:
 738                 case V3D_INTERNAL_TYPE_8UI:
 739                         job->clear_color[i][0] = ((color->ui[0] & 0xff) |
 740                                                   (color->ui[1] & 0xff) << 8 |
 741                                                   (color->ui[2] & 0xff) << 16 |
 742                                                   (color->ui[3] & 0xff) << 24);
 743                         break;
 744                 case V3D_INTERNAL_TYPE_16F:
 745                         util_pack_color(color->f, PIPE_FORMAT_R16G16B16A16_FLOAT,
 746                                         &uc);
 747                         memcpy(job->clear_color[i], uc.ui, internal_size);
 748                         break;
 749                 case V3D_INTERNAL_TYPE_16I:
 750                 case V3D_INTERNAL_TYPE_16UI:
 751                         job->clear_color[i][0] = ((color->ui[0] & 0xffff) |
 752                                                   color->ui[1] << 16);
 753                         job->clear_color[i][1] = ((color->ui[2] & 0xffff) |
 754                                                   color->ui[3] << 16);
 755                         break;
 756                 case V3D_INTERNAL_TYPE_32F:
 757                 case V3D_INTERNAL_TYPE_32I:
 758                 case V3D_INTERNAL_TYPE_32UI:
 759                         memcpy(job->clear_color[i], color->ui, internal_size);
 760                         break;
 761                 }
 762
 763                 rsc->initialized_buffers |= bit;
 764         }
 765
 766         unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
 767         if (zsclear) {
 768                 struct v3d_resource *rsc =
 769                         v3d_resource(v3d->framebuffer.zsbuf->texture);
 770
 771                 if (zsclear & PIPE_CLEAR_DEPTH)
 772                         job->clear_z = depth;
 773                 if (zsclear & PIPE_CLEAR_STENCIL)
 774                         job->clear_s = stencil;
 775
 776                 rsc->initialized_buffers |= zsclear;
 777         }
 778
 779         job->draw_min_x = 0;
 780         job->draw_min_y = 0;
 781         job->draw_max_x = v3d->framebuffer.width;
 782         job->draw_max_y = v3d->framebuffer.height;
 783         job->clear |= buffers;
 784         job->store |= buffers;
 785
 786         v3d_start_draw(v3d);
 787
 788         return buffers;
 789 }
 790
 791 static void
 792 v3d_clear(struct pipe_context *pctx, unsigned buffers,
 793           const union pipe_color_union *color, double depth, unsigned stencil)
 794 {
 795         struct v3d_context *v3d = v3d_context(pctx);
 796         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 797
 798         buffers &= ~v3d_tlb_clear(job, buffers, color, depth, stencil);
 799
 800         if (buffers)
 801                 v3d_draw_clear(v3d, buffers, color, depth, stencil);
 802 }
 803
 804 static void
 805 v3d_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 806                         const union pipe_color_union *color,
 807                         unsigned x, unsigned y, unsigned w, unsigned h,
 808                         bool render_condition_enabled)
 809 {
 810         fprintf(stderr, "unimpl: clear RT\n");
 811 }
 812
 813 static void
 814 v3d_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 815                         unsigned buffers, double depth, unsigned stencil,
 816                         unsigned x, unsigned y, unsigned w, unsigned h,
 817                         bool render_condition_enabled)
 818 {
 819         fprintf(stderr, "unimpl: clear DS\n");
 820 }
 821
 822 void
 823 v3dX(draw_init)(struct pipe_context *pctx)
 824 {
 825         pctx->draw_vbo = v3d_draw_vbo;
 826         pctx->clear = v3d_clear;
 827         pctx->clear_render_target = v3d_clear_render_target;
 828         pctx->clear_depth_stencil = v3d_clear_depth_stencil;
 829 }