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         /* "Binning mode lists start with a Tile Binning Mode Configuration
  68          * item (120)"
  69          *
  70          * Part1 signals the end of binning config setup.
  71          */
  72         cl_emit(&job->bcl, TILE_BINNING_MODE_CONFIGURATION_PART2, config) {
  73                 config.tile_allocation_memory_address =
  74                         cl_address(job->tile_alloc, 0);
  75                 config.tile_allocation_memory_size = job->tile_alloc->size;
  76         }
  77 #endif
  78
  79         cl_emit(&job->bcl, TILE_BINNING_MODE_CONFIGURATION_PART1, config) {
  80 #if V3D_VERSION >= 40
  81                 config.width_in_pixels = v3d->framebuffer.width;
  82                 config.height_in_pixels = v3d->framebuffer.height;
  83                 config.number_of_render_targets =
  84                         MAX2(v3d->framebuffer.nr_cbufs, 1);
  85 #else /* V3D_VERSION < 40 */
  86                 config.tile_state_data_array_base_address =
  87                         cl_address(job->tile_state, 0);
  88
  89                 config.width_in_tiles = job->draw_tiles_x;
  90                 config.height_in_tiles = job->draw_tiles_y;
  91                 /* Must be >= 1 */
  92                 config.number_of_render_targets =
  93                         MAX2(v3d->framebuffer.nr_cbufs, 1);
  94 #endif /* V3D_VERSION < 40 */
  95
  96                 config.multisample_mode_4x = job->msaa;
  97
  98                 config.maximum_bpp_of_all_render_targets = job->internal_bpp;
  99         }
 100
 101         /* There's definitely nothing in the VCD cache we want. */
 102         cl_emit(&job->bcl, FLUSH_VCD_CACHE, bin);
 103
 104         /* Disable any leftover OQ state from another job. */
 105         cl_emit(&job->bcl, OCCLUSION_QUERY_COUNTER, counter);
 106
 107         /* "Binning mode lists must have a Start Tile Binning item (6) after
 108          *  any prefix state data before the binning list proper starts."
 109          */
 110         cl_emit(&job->bcl, START_TILE_BINNING, bin);
 111
 112         job->needs_flush = true;
 113         job->draw_width = v3d->framebuffer.width;
 114         job->draw_height = v3d->framebuffer.height;
 115 }
 116
 117 static void
 118 v3d_predraw_check_textures(struct pipe_context *pctx,
 119                            struct v3d_texture_stateobj *stage_tex)
 120 {
 121         struct v3d_context *v3d = v3d_context(pctx);
 122
 123         for (int i = 0; i < stage_tex->num_textures; i++) {
 124                 struct pipe_sampler_view *view = stage_tex->textures[i];
 125                 if (!view)
 126                         continue;
 127
 128                 v3d_flush_jobs_writing_resource(v3d, view->texture);
 129         }
 130 }
 131
 132 static void
 133 v3d_emit_gl_shader_state(struct v3d_context *v3d,
 134                          const struct pipe_draw_info *info)
 135 {
 136         struct v3d_job *job = v3d->job;
 137         /* VC5_DIRTY_VTXSTATE */
 138         struct v3d_vertex_stateobj *vtx = v3d->vtx;
 139         /* VC5_DIRTY_VTXBUF */
 140         struct v3d_vertexbuf_stateobj *vertexbuf = &v3d->vertexbuf;
 141
 142         /* Upload the uniforms to the indirect CL first */
 143         struct v3d_cl_reloc fs_uniforms =
 144                 v3d_write_uniforms(v3d, v3d->prog.fs,
 145                                    &v3d->constbuf[PIPE_SHADER_FRAGMENT],
 146                                    &v3d->fragtex);
 147         struct v3d_cl_reloc vs_uniforms =
 148                 v3d_write_uniforms(v3d, v3d->prog.vs,
 149                                    &v3d->constbuf[PIPE_SHADER_VERTEX],
 150                                    &v3d->verttex);
 151         struct v3d_cl_reloc cs_uniforms =
 152                 v3d_write_uniforms(v3d, v3d->prog.cs,
 153                                    &v3d->constbuf[PIPE_SHADER_VERTEX],
 154                                    &v3d->verttex);
 155
 156         /* See GFXH-930 workaround below */
 157         uint32_t num_elements_to_emit = MAX2(vtx->num_elements, 1);
 158         uint32_t shader_rec_offset =
 159                 v3d_cl_ensure_space(&job->indirect,
 160                                     cl_packet_length(GL_SHADER_STATE_RECORD) +
 161                                     num_elements_to_emit *
 162                                     cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
 163                                     32);
 164
 165         cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
 166                 shader.enable_clipping = true;
 167                 /* VC5_DIRTY_PRIM_MODE | VC5_DIRTY_RASTERIZER */
 168                 shader.point_size_in_shaded_vertex_data =
 169                         (info->mode == PIPE_PRIM_POINTS &&
 170                          v3d->rasterizer->base.point_size_per_vertex);
 171
 172                 /* Must be set if the shader modifies Z, discards, or modifies
 173                  * the sample mask.  For any of these cases, the fragment
 174                  * shader needs to write the Z value (even just discards).
 175                  */
 176                 shader.fragment_shader_does_z_writes =
 177                         (v3d->prog.fs->prog_data.fs->writes_z ||
 178                          v3d->prog.fs->prog_data.fs->discard);
 179
 180                 shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
 181                         v3d->prog.fs->prog_data.fs->uses_center_w;
 182
 183                 shader.number_of_varyings_in_fragment_shader =
 184                         v3d->prog.fs->prog_data.base->num_inputs;
 185
 186                 shader.coordinate_shader_propagate_nans = true;
 187                 shader.vertex_shader_propagate_nans = true;
 188                 shader.fragment_shader_propagate_nans = true;
 189
 190                 shader.coordinate_shader_code_address =
 191                         cl_address(v3d->prog.cs->bo, 0);
 192                 shader.vertex_shader_code_address =
 193                         cl_address(v3d->prog.vs->bo, 0);
 194                 shader.fragment_shader_code_address =
 195                         cl_address(v3d->prog.fs->bo, 0);
 196
 197                 /* XXX: Use combined input/output size flag in the common
 198                  * case.
 199                  */
 200                 shader.coordinate_shader_has_separate_input_and_output_vpm_blocks = true;
 201                 shader.vertex_shader_has_separate_input_and_output_vpm_blocks = true;
 202                 shader.coordinate_shader_input_vpm_segment_size =
 203                         MAX2(v3d->prog.cs->prog_data.vs->vpm_input_size, 1);
 204                 shader.vertex_shader_input_vpm_segment_size =
 205                         MAX2(v3d->prog.vs->prog_data.vs->vpm_input_size, 1);
 206
 207                 shader.coordinate_shader_output_vpm_segment_size =
 208                         v3d->prog.cs->prog_data.vs->vpm_output_size;
 209                 shader.vertex_shader_output_vpm_segment_size =
 210                         v3d->prog.vs->prog_data.vs->vpm_output_size;
 211
 212                 shader.coordinate_shader_uniforms_address = cs_uniforms;
 213                 shader.vertex_shader_uniforms_address = vs_uniforms;
 214                 shader.fragment_shader_uniforms_address = fs_uniforms;
 215
 216 #if V3D_VERSION >= 41
 217                 shader.min_coord_shader_input_segments_required_in_play = 1;
 218                 shader.min_vertex_shader_input_segments_required_in_play = 1;
 219
 220                 shader.coordinate_shader_4_way_threadable =
 221                         v3d->prog.cs->prog_data.vs->base.threads == 4;
 222                 shader.vertex_shader_4_way_threadable =
 223                         v3d->prog.vs->prog_data.vs->base.threads == 4;
 224                 shader.fragment_shader_4_way_threadable =
 225                         v3d->prog.fs->prog_data.fs->base.threads == 4;
 226
 227                 shader.coordinate_shader_start_in_final_thread_section =
 228                         v3d->prog.cs->prog_data.vs->base.single_seg;
 229                 shader.vertex_shader_start_in_final_thread_section =
 230                         v3d->prog.vs->prog_data.vs->base.single_seg;
 231                 shader.fragment_shader_start_in_final_thread_section =
 232                         v3d->prog.fs->prog_data.fs->base.single_seg;
 233 #else
 234                 shader.coordinate_shader_4_way_threadable =
 235                         v3d->prog.cs->prog_data.vs->base.threads == 4;
 236                 shader.coordinate_shader_2_way_threadable =
 237                         v3d->prog.cs->prog_data.vs->base.threads == 2;
 238                 shader.vertex_shader_4_way_threadable =
 239                         v3d->prog.vs->prog_data.vs->base.threads == 4;
 240                 shader.vertex_shader_2_way_threadable =
 241                         v3d->prog.vs->prog_data.vs->base.threads == 2;
 242                 shader.fragment_shader_4_way_threadable =
 243                         v3d->prog.fs->prog_data.fs->base.threads == 4;
 244                 shader.fragment_shader_2_way_threadable =
 245                         v3d->prog.fs->prog_data.fs->base.threads == 2;
 246 #endif
 247
 248                 shader.vertex_id_read_by_coordinate_shader =
 249                         v3d->prog.cs->prog_data.vs->uses_vid;
 250                 shader.instance_id_read_by_coordinate_shader =
 251                         v3d->prog.cs->prog_data.vs->uses_iid;
 252                 shader.vertex_id_read_by_vertex_shader =
 253                         v3d->prog.vs->prog_data.vs->uses_vid;
 254                 shader.instance_id_read_by_vertex_shader =
 255                         v3d->prog.vs->prog_data.vs->uses_iid;
 256
 257                 shader.address_of_default_attribute_values =
 258                         cl_address(vtx->default_attribute_values, 0);
 259         }
 260
 261         for (int i = 0; i < vtx->num_elements; i++) {
 262                 struct pipe_vertex_element *elem = &vtx->pipe[i];
 263                 struct pipe_vertex_buffer *vb =
 264                         &vertexbuf->vb[elem->vertex_buffer_index];
 265                 struct v3d_resource *rsc = v3d_resource(vb->buffer.resource);
 266
 267                 const uint32_t size =
 268                         cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
 269                 cl_emit_with_prepacked(&job->indirect,
 270                                        GL_SHADER_STATE_ATTRIBUTE_RECORD,
 271                                        &vtx->attrs[i * size], attr) {
 272                         attr.stride = vb->stride;
 273                         attr.address = cl_address(rsc->bo,
 274                                                   vb->buffer_offset +
 275                                                   elem->src_offset);
 276                         attr.number_of_values_read_by_coordinate_shader =
 277                                 v3d->prog.cs->prog_data.vs->vattr_sizes[i];
 278                         attr.number_of_values_read_by_vertex_shader =
 279                                 v3d->prog.vs->prog_data.vs->vattr_sizes[i];
 280 #if V3D_VERSION >= 41
 281                         attr.maximum_index = 0xffffff;
 282 #endif
 283                 }
 284                 STATIC_ASSERT(sizeof(vtx->attrs) >= VC5_MAX_ATTRIBUTES * size);
 285         }
 286
 287         if (vtx->num_elements == 0) {
 288                 /* GFXH-930: At least one attribute must be enabled and read
 289                  * by CS and VS.  If we have no attributes being consumed by
 290                  * the shader, set up a dummy to be loaded into the VPM.
 291                  */
 292                 cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
 293                         /* Valid address of data whose value will be unused. */
 294                         attr.address = cl_address(job->indirect.bo, 0);
 295
 296                         attr.type = ATTRIBUTE_FLOAT;
 297                         attr.stride = 0;
 298                         attr.vec_size = 1;
 299
 300                         attr.number_of_values_read_by_coordinate_shader = 1;
 301                         attr.number_of_values_read_by_vertex_shader = 1;
 302                 }
 303         }
 304
 305         cl_emit(&job->bcl, GL_SHADER_STATE, state) {
 306                 state.address = cl_address(job->indirect.bo, shader_rec_offset);
 307                 state.number_of_attribute_arrays = num_elements_to_emit;
 308         }
 309
 310         v3d_bo_unreference(&cs_uniforms.bo);
 311         v3d_bo_unreference(&vs_uniforms.bo);
 312         v3d_bo_unreference(&fs_uniforms.bo);
 313
 314         job->shader_rec_count++;
 315 }
 316
 317 /**
 318  * Computes the various transform feedback statistics, since they can't be
 319  * recorded by CL packets.
 320  */
 321 static void
 322 v3d_tf_statistics_record(struct v3d_context *v3d,
 323                          const struct pipe_draw_info *info,
 324                          bool prim_tf)
 325 {
 326         if (!v3d->active_queries)
 327                 return;
 328
 329         uint32_t prims = u_prims_for_vertices(info->mode, info->count);
 330         v3d->prims_generated += prims;
 331
 332         if (prim_tf) {
 333                 /* XXX: Only count if we didn't overflow. */
 334                 v3d->tf_prims_generated += prims;
 335         }
 336 }
 337
 338 static void
 339 v3d_update_job_ez(struct v3d_context *v3d, struct v3d_job *job)
 340 {
 341         switch (v3d->zsa->ez_state) {
 342         case VC5_EZ_UNDECIDED:
 343                 /* If the Z/S state didn't pick a direction but didn't
 344                  * disable, then go along with the current EZ state.  This
 345                  * allows EZ optimization for Z func == EQUAL or NEVER.
 346                  */
 347                 break;
 348
 349         case VC5_EZ_LT_LE:
 350         case VC5_EZ_GT_GE:
 351                 /* If the Z/S state picked a direction, then it needs to match
 352                  * the current direction if we've decided on one.
 353                  */
 354                 if (job->ez_state == VC5_EZ_UNDECIDED)
 355                         job->ez_state = v3d->zsa->ez_state;
 356                 else if (job->ez_state != v3d->zsa->ez_state)
 357                         job->ez_state = VC5_EZ_DISABLED;
 358                 break;
 359
 360         case VC5_EZ_DISABLED:
 361                 /* If the current Z/S state disables EZ because of a bad Z
 362                  * func or stencil operation, then we can't do any more EZ in
 363                  * this frame.
 364                  */
 365                 job->ez_state = VC5_EZ_DISABLED;
 366                 break;
 367         }
 368
 369         /* If the FS affects the Z of the pixels, then it may update against
 370          * the chosen EZ direction (though we could use
 371          * ARB_conservative_depth's hints to avoid this)
 372          */
 373         if (v3d->prog.fs->prog_data.fs->writes_z) {
 374                 job->ez_state = VC5_EZ_DISABLED;
 375         }
 376
 377         if (job->first_ez_state == VC5_EZ_UNDECIDED &&
 378             (job->ez_state != VC5_EZ_DISABLED || job->draw_calls_queued == 0))
 379                 job->first_ez_state = job->ez_state;
 380 }
 381
 382 static void
 383 v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 384 {
 385         struct v3d_context *v3d = v3d_context(pctx);
 386
 387         if (!info->count_from_stream_output && !info->indirect &&
 388             !info->primitive_restart &&
 389             !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
 390                 return;
 391
 392         /* Fall back for weird desktop GL primitive restart values. */
 393         if (info->primitive_restart &&
 394             info->index_size) {
 395                 uint32_t mask = ~0;
 396
 397                 switch (info->index_size) {
 398                 case 2:
 399                         mask = 0xffff;
 400                         break;
 401                 case 1:
 402                         mask = 0xff;
 403                         break;
 404                 }
 405
 406                 if (info->restart_index != mask) {
 407                         util_draw_vbo_without_prim_restart(pctx, info);
 408                         return;
 409                 }
 410         }
 411
 412         if (info->mode >= PIPE_PRIM_QUADS) {
 413                 util_primconvert_save_rasterizer_state(v3d->primconvert, &v3d->rasterizer->base);
 414                 util_primconvert_draw_vbo(v3d->primconvert, info);
 415                 perf_debug("Fallback conversion for %d %s vertices\n",
 416                            info->count, u_prim_name(info->mode));
 417                 return;
 418         }
 419
 420         /* Before setting up the draw, flush anything writing to the textures
 421          * that we read from.
 422          */
 423         v3d_predraw_check_textures(pctx, &v3d->verttex);
 424         v3d_predraw_check_textures(pctx, &v3d->fragtex);
 425
 426         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 427
 428         /* If vertex texturing depends on the output of rendering, we need to
 429          * ensure that that rendering is complete before we run a coordinate
 430          * shader that depends on it.
 431          *
 432          * Given that doing that is unusual, for now we just block the binner
 433          * on the last submitted render, rather than tracking the last
 434          * rendering to each texture's BO.
 435          */
 436         if (v3d->verttex.num_textures) {
 437                 perf_debug("Blocking binner on last render "
 438                            "due to vertex texturing.\n");
 439                 job->submit.in_sync_bcl = v3d->out_sync;
 440         }
 441
 442         /* Get space to emit our draw call into the BCL, using a branch to
 443          * jump to a new BO if necessary.
 444          */
 445         v3d_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
 446
 447         if (v3d->prim_mode != info->mode) {
 448                 v3d->prim_mode = info->mode;
 449                 v3d->dirty |= VC5_DIRTY_PRIM_MODE;
 450         }
 451
 452         v3d_start_draw(v3d);
 453         v3d_update_compiled_shaders(v3d, info->mode);
 454         v3d_update_job_ez(v3d, job);
 455
 456 #if V3D_VERSION >= 41
 457         v3d41_emit_state(pctx);
 458 #else
 459         v3d33_emit_state(pctx);
 460 #endif
 461
 462         if (v3d->dirty & (VC5_DIRTY_VTXBUF |
 463                           VC5_DIRTY_VTXSTATE |
 464                           VC5_DIRTY_PRIM_MODE |
 465                           VC5_DIRTY_RASTERIZER |
 466                           VC5_DIRTY_COMPILED_CS |
 467                           VC5_DIRTY_COMPILED_VS |
 468                           VC5_DIRTY_COMPILED_FS |
 469                           v3d->prog.cs->uniform_dirty_bits |
 470                           v3d->prog.vs->uniform_dirty_bits |
 471                           v3d->prog.fs->uniform_dirty_bits)) {
 472                 v3d_emit_gl_shader_state(v3d, info);
 473         }
 474
 475         v3d->dirty = 0;
 476
 477         /* The Base Vertex/Base Instance packet sets those values to nonzero
 478          * for the next draw call only.
 479          */
 480         if (info->index_bias || info->start_instance) {
 481                 cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
 482                         base.base_instance = info->start_instance;
 483                         base.base_vertex = info->index_bias;
 484                 }
 485         }
 486
 487         uint32_t prim_tf_enable = 0;
 488 #if V3D_VERSION < 40
 489         /* V3D 3.x: The HW only processes transform feedback on primitives
 490          * with the flag set.
 491          */
 492         if (v3d->streamout.num_targets)
 493                 prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
 494 #endif
 495
 496         v3d_tf_statistics_record(v3d, info, v3d->streamout.num_targets);
 497
 498         /* Note that the primitive type fields match with OpenGL/gallium
 499          * definitions, up to but not including QUADS.
 500          */
 501         if (info->index_size) {
 502                 uint32_t index_size = info->index_size;
 503                 uint32_t offset = info->start * index_size;
 504                 struct pipe_resource *prsc;
 505                 if (info->has_user_indices) {
 506                         prsc = NULL;
 507                         u_upload_data(v3d->uploader, 0,
 508                                       info->count * info->index_size, 4,
 509                                       info->index.user,
 510                                       &offset, &prsc);
 511                 } else {
 512                         prsc = info->index.resource;
 513                 }
 514                 struct v3d_resource *rsc = v3d_resource(prsc);
 515
 516 #if V3D_VERSION >= 40
 517                 cl_emit(&job->bcl, INDEX_BUFFER_SETUP, ib) {
 518                         ib.address = cl_address(rsc->bo, 0);
 519                         ib.size = rsc->bo->size;
 520                 }
 521 #endif
 522
 523                 if (info->instance_count > 1) {
 524                         cl_emit(&job->bcl, INDEXED_INSTANCED_PRIMITIVE_LIST, prim) {
 525                                 prim.index_type = ffs(info->index_size) - 1;
 526 #if V3D_VERSION >= 40
 527                                 prim.index_offset = offset;
 528 #else /* V3D_VERSION < 40 */
 529                                 prim.maximum_index = (1u << 31) - 1; /* XXX */
 530                                 prim.address_of_indices_list =
 531                                         cl_address(rsc->bo, offset);
 532 #endif /* V3D_VERSION < 40 */
 533                                 prim.mode = info->mode | prim_tf_enable;
 534                                 prim.enable_primitive_restarts = info->primitive_restart;
 535
 536                                 prim.number_of_instances = info->instance_count;
 537                                 prim.instance_length = info->count;
 538                         }
 539                 } else {
 540                         cl_emit(&job->bcl, INDEXED_PRIMITIVE_LIST, prim) {
 541                                 prim.index_type = ffs(info->index_size) - 1;
 542                                 prim.length = info->count;
 543 #if V3D_VERSION >= 40
 544                                 prim.index_offset = offset;
 545 #else /* V3D_VERSION < 40 */
 546                                 prim.maximum_index = (1u << 31) - 1; /* XXX */
 547                                 prim.address_of_indices_list =
 548                                         cl_address(rsc->bo, offset);
 549 #endif /* V3D_VERSION < 40 */
 550                                 prim.mode = info->mode | prim_tf_enable;
 551                                 prim.enable_primitive_restarts = info->primitive_restart;
 552                         }
 553                 }
 554
 555                 job->draw_calls_queued++;
 556
 557                 if (info->has_user_indices)
 558                         pipe_resource_reference(&prsc, NULL);
 559         } else {
 560                 if (info->instance_count > 1) {
 561                         cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMITIVES, prim) {
 562                                 prim.mode = info->mode | prim_tf_enable;
 563                                 prim.index_of_first_vertex = info->start;
 564                                 prim.number_of_instances = info->instance_count;
 565                                 prim.instance_length = info->count;
 566                         }
 567                 } else {
 568                         cl_emit(&job->bcl, VERTEX_ARRAY_PRIMITIVES, prim) {
 569                                 prim.mode = info->mode | prim_tf_enable;
 570                                 prim.length = info->count;
 571                                 prim.index_of_first_vertex = info->start;
 572                         }
 573                 }
 574         }
 575
 576         /* A flush is required in between a TF draw and any following TF specs
 577          * packet, or the GPU may hang.  Just flush each time for now.
 578          */
 579         if (v3d->streamout.num_targets)
 580                 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_FLUSH_AND_COUNT, flush);
 581
 582         job->draw_calls_queued++;
 583
 584         /* Increment the TF offsets by how many verts we wrote.  XXX: This
 585          * needs some clamping to the buffer size.
 586          */
 587         for (int i = 0; i < v3d->streamout.num_targets; i++)
 588                 v3d->streamout.offsets[i] += info->count;
 589
 590         if (v3d->zsa && job->zsbuf && v3d->zsa->base.depth.enabled) {
 591                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
 592                 v3d_job_add_bo(job, rsc->bo);
 593
 594                 job->load |= PIPE_CLEAR_DEPTH & ~job->clear;
 595                 if (v3d->zsa->base.depth.writemask)
 596                         job->store |= PIPE_CLEAR_DEPTH;
 597                 rsc->initialized_buffers = PIPE_CLEAR_DEPTH;
 598         }
 599
 600         if (v3d->zsa && job->zsbuf && v3d->zsa->base.stencil[0].enabled) {
 601                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
 602                 if (rsc->separate_stencil)
 603                         rsc = rsc->separate_stencil;
 604
 605                 v3d_job_add_bo(job, rsc->bo);
 606
 607                 job->load |= PIPE_CLEAR_STENCIL & ~job->clear;
 608                 if (v3d->zsa->base.stencil[0].writemask ||
 609                     v3d->zsa->base.stencil[1].writemask) {
 610                         job->store |= PIPE_CLEAR_STENCIL;
 611                 }
 612                 rsc->initialized_buffers |= PIPE_CLEAR_STENCIL;
 613         }
 614
 615         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 616                 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
 617                 int blend_rt = v3d->blend->base.independent_blend_enable ? i : 0;
 618
 619                 if (job->store & bit || !job->cbufs[i])
 620                         continue;
 621                 struct v3d_resource *rsc = v3d_resource(job->cbufs[i]->texture);
 622
 623                 job->load |= bit & ~job->clear;
 624                 if (v3d->blend->base.rt[blend_rt].colormask)
 625                         job->store |= bit;
 626                 v3d_job_add_bo(job, rsc->bo);
 627         }
 628
 629         if (job->referenced_size > 768 * 1024 * 1024) {
 630                 perf_debug("Flushing job with %dkb to try to free up memory\n",
 631                         job->referenced_size / 1024);
 632                 v3d_flush(pctx);
 633         }
 634
 635         if (V3D_DEBUG & V3D_DEBUG_ALWAYS_FLUSH)
 636                 v3d_flush(pctx);
 637 }
 638
 639 /**
 640  * Implements gallium's clear() hook (glClear()) by drawing a pair of triangles.
 641  */
 642 static void
 643 v3d_draw_clear(struct v3d_context *v3d,
 644                unsigned buffers,
 645                const union pipe_color_union *color,
 646                double depth, unsigned stencil)
 647 {
 648         static const union pipe_color_union dummy_color = {};
 649
 650         /* The blitter util dereferences the color regardless, even though the
 651          * gallium clear API may not pass one in when only Z/S are cleared.
 652          */
 653         if (!color)
 654                 color = &dummy_color;
 655
 656         v3d_blitter_save(v3d);
 657         util_blitter_clear(v3d->blitter,
 658                            v3d->framebuffer.width,
 659                            v3d->framebuffer.height,
 660                            util_framebuffer_get_num_layers(&v3d->framebuffer),
 661                            buffers, color, depth, stencil);
 662 }
 663
 664 /**
 665  * Attempts to perform the GL clear by using the TLB's fast clear at the start
 666  * of the frame.
 667  */
 668 static unsigned
 669 v3d_tlb_clear(struct v3d_job *job, unsigned buffers,
 670               const union pipe_color_union *color,
 671               double depth, unsigned stencil)
 672 {
 673         struct v3d_context *v3d = job->v3d;
 674
 675         if (job->draw_calls_queued) {
 676                 /* If anything in the CL has drawn using the buffer, then the
 677                  * TLB clear we're trying to add now would happen before that
 678                  * drawing.
 679                  */
 680                 buffers &= ~(job->load | job->store);
 681         }
 682
 683         /* GFXH-1461: If we were to emit a load of just depth or just stencil,
 684          * then the clear for the other may get lost.  We need to decide now
 685          * if it would be possible to need to emit a load of just one after
 686          * we've set up our TLB clears.
 687          */
 688         if (buffers & PIPE_CLEAR_DEPTHSTENCIL &&
 689             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL &&
 690             job->zsbuf &&
 691             util_format_is_depth_and_stencil(job->zsbuf->texture->format)) {
 692                 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 693         }
 694
 695         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 696                 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
 697                 if (!(buffers & bit))
 698                         continue;
 699
 700                 struct pipe_surface *psurf = v3d->framebuffer.cbufs[i];
 701                 struct v3d_surface *surf = v3d_surface(psurf);
 702                 struct v3d_resource *rsc = v3d_resource(psurf->texture);
 703
 704                 union util_color uc;
 705                 uint32_t internal_size = 4 << surf->internal_bpp;
 706
 707                 static union pipe_color_union swapped_color;
 708                 if (v3d->swap_color_rb & (1 << i)) {
 709                         swapped_color.f[0] = color->f[2];
 710                         swapped_color.f[1] = color->f[1];
 711                         swapped_color.f[2] = color->f[0];
 712                         swapped_color.f[3] = color->f[3];
 713                         color = &swapped_color;
 714                 }
 715
 716                 switch (surf->internal_type) {
 717                 case V3D_INTERNAL_TYPE_8:
 718                         util_pack_color(color->f, PIPE_FORMAT_R8G8B8A8_UNORM,
 719                                         &uc);
 720                         memcpy(job->clear_color[i], uc.ui, internal_size);
 721                         break;
 722                 case V3D_INTERNAL_TYPE_8I:
 723                 case V3D_INTERNAL_TYPE_8UI:
 724                         job->clear_color[i][0] = ((color->ui[0] & 0xff) |
 725                                                   (color->ui[1] & 0xff) << 8 |
 726                                                   (color->ui[2] & 0xff) << 16 |
 727                                                   (color->ui[3] & 0xff) << 24);
 728                         break;
 729                 case V3D_INTERNAL_TYPE_16F:
 730                         util_pack_color(color->f, PIPE_FORMAT_R16G16B16A16_FLOAT,
 731                                         &uc);
 732                         memcpy(job->clear_color[i], uc.ui, internal_size);
 733                         break;
 734                 case V3D_INTERNAL_TYPE_16I:
 735                 case V3D_INTERNAL_TYPE_16UI:
 736                         job->clear_color[i][0] = ((color->ui[0] & 0xffff) |
 737                                                   color->ui[1] << 16);
 738                         job->clear_color[i][1] = ((color->ui[2] & 0xffff) |
 739                                                   color->ui[3] << 16);
 740                         break;
 741                 case V3D_INTERNAL_TYPE_32F:
 742                 case V3D_INTERNAL_TYPE_32I:
 743                 case V3D_INTERNAL_TYPE_32UI:
 744                         memcpy(job->clear_color[i], color->ui, internal_size);
 745                         break;
 746                 }
 747
 748                 rsc->initialized_buffers |= bit;
 749         }
 750
 751         unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
 752         if (zsclear) {
 753                 struct v3d_resource *rsc =
 754                         v3d_resource(v3d->framebuffer.zsbuf->texture);
 755
 756                 if (zsclear & PIPE_CLEAR_DEPTH)
 757                         job->clear_z = depth;
 758                 if (zsclear & PIPE_CLEAR_STENCIL)
 759                         job->clear_s = stencil;
 760
 761                 rsc->initialized_buffers |= zsclear;
 762         }
 763
 764         job->draw_min_x = 0;
 765         job->draw_min_y = 0;
 766         job->draw_max_x = v3d->framebuffer.width;
 767         job->draw_max_y = v3d->framebuffer.height;
 768         job->clear |= buffers;
 769         job->store |= buffers;
 770
 771         v3d_start_draw(v3d);
 772
 773         return buffers;
 774 }
 775
 776 static void
 777 v3d_clear(struct pipe_context *pctx, unsigned buffers,
 778           const union pipe_color_union *color, double depth, unsigned stencil)
 779 {
 780         struct v3d_context *v3d = v3d_context(pctx);
 781         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 782
 783         buffers &= ~v3d_tlb_clear(job, buffers, color, depth, stencil);
 784
 785         if (buffers)
 786                 v3d_draw_clear(v3d, buffers, color, depth, stencil);
 787 }
 788
 789 static void
 790 v3d_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 791                         const union pipe_color_union *color,
 792                         unsigned x, unsigned y, unsigned w, unsigned h,
 793                         bool render_condition_enabled)
 794 {
 795         fprintf(stderr, "unimpl: clear RT\n");
 796 }
 797
 798 static void
 799 v3d_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 800                         unsigned buffers, double depth, unsigned stencil,
 801                         unsigned x, unsigned y, unsigned w, unsigned h,
 802                         bool render_condition_enabled)
 803 {
 804         fprintf(stderr, "unimpl: clear DS\n");
 805 }
 806
 807 void
 808 v3dX(draw_init)(struct pipe_context *pctx)
 809 {
 810         pctx->draw_vbo = v3d_draw_vbo;
 811         pctx->clear = v3d_clear;
 812         pctx->clear_render_target = v3d_clear_render_target;
 813         pctx->clear_depth_stencil = v3d_clear_depth_stencil;
 814 }