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_stage_inputs(struct pipe_context *pctx,
 123                                enum pipe_shader_type s)
 124 {
 125         struct v3d_context *v3d = v3d_context(pctx);
 126
 127         /* XXX perf: If we're reading from the output of TF in this job, we
 128          * should instead be using the wait for transform feedback
 129          * functionality.
 130          */
 131
 132         /* Flush writes to textures we're sampling. */
 133         for (int i = 0; i < v3d->tex[s].num_textures; i++) {
 134                 struct pipe_sampler_view *pview = v3d->tex[s].textures[i];
 135                 if (!pview)
 136                         continue;
 137                 struct v3d_sampler_view *view = v3d_sampler_view(pview);
 138
 139                 if (view->texture != view->base.texture)
 140                         v3d_update_shadow_texture(pctx, &view->base);
 141
 142                 v3d_flush_jobs_writing_resource(v3d, view->texture);
 143         }
 144
 145         /* Flush writes to UBOs. */
 146         foreach_bit(i, v3d->constbuf[s].enabled_mask) {
 147                 struct pipe_constant_buffer *cb = &v3d->constbuf[s].cb[i];
 148                 if (cb->buffer)
 149                         v3d_flush_jobs_writing_resource(v3d, cb->buffer);
 150         }
 151 }
 152
 153 static void
 154 v3d_emit_gl_shader_state(struct v3d_context *v3d,
 155                          const struct pipe_draw_info *info)
 156 {
 157         struct v3d_job *job = v3d->job;
 158         /* VC5_DIRTY_VTXSTATE */
 159         struct v3d_vertex_stateobj *vtx = v3d->vtx;
 160         /* VC5_DIRTY_VTXBUF */
 161         struct v3d_vertexbuf_stateobj *vertexbuf = &v3d->vertexbuf;
 162
 163         /* Upload the uniforms to the indirect CL first */
 164         struct v3d_cl_reloc fs_uniforms =
 165                 v3d_write_uniforms(v3d, v3d->prog.fs,
 166                                    PIPE_SHADER_FRAGMENT);
 167         struct v3d_cl_reloc vs_uniforms =
 168                 v3d_write_uniforms(v3d, v3d->prog.vs,
 169                                    PIPE_SHADER_VERTEX);
 170         struct v3d_cl_reloc cs_uniforms =
 171                 v3d_write_uniforms(v3d, v3d->prog.cs,
 172                                    PIPE_SHADER_VERTEX);
 173
 174         /* See GFXH-930 workaround below */
 175         uint32_t num_elements_to_emit = MAX2(vtx->num_elements, 1);
 176         uint32_t shader_rec_offset =
 177                 v3d_cl_ensure_space(&job->indirect,
 178                                     cl_packet_length(GL_SHADER_STATE_RECORD) +
 179                                     num_elements_to_emit *
 180                                     cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
 181                                     32);
 182
 183         /* XXX perf: We should move most of the SHADER_STATE_RECORD setup to
 184          * compile time, so that we mostly just have to OR the VS and FS
 185          * records together at draw time.
 186          */
 187         cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
 188                 shader.enable_clipping = true;
 189                 /* VC5_DIRTY_PRIM_MODE | VC5_DIRTY_RASTERIZER */
 190                 shader.point_size_in_shaded_vertex_data =
 191                         (info->mode == PIPE_PRIM_POINTS &&
 192                          v3d->rasterizer->base.point_size_per_vertex);
 193
 194                 /* Must be set if the shader modifies Z, discards, or modifies
 195                  * the sample mask.  For any of these cases, the fragment
 196                  * shader needs to write the Z value (even just discards).
 197                  */
 198                 shader.fragment_shader_does_z_writes =
 199                         (v3d->prog.fs->prog_data.fs->writes_z ||
 200                          v3d->prog.fs->prog_data.fs->discard);
 201
 202                 shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
 203                         v3d->prog.fs->prog_data.fs->uses_center_w;
 204
 205                 shader.number_of_varyings_in_fragment_shader =
 206                         v3d->prog.fs->prog_data.base->num_inputs;
 207
 208                 shader.coordinate_shader_propagate_nans = true;
 209                 shader.vertex_shader_propagate_nans = true;
 210                 shader.fragment_shader_propagate_nans = true;
 211
 212                 shader.coordinate_shader_code_address =
 213                         cl_address(v3d_resource(v3d->prog.cs->resource)->bo,
 214                                    v3d->prog.cs->offset);
 215                 shader.vertex_shader_code_address =
 216                         cl_address(v3d_resource(v3d->prog.vs->resource)->bo,
 217                                    v3d->prog.vs->offset);
 218                 shader.fragment_shader_code_address =
 219                         cl_address(v3d_resource(v3d->prog.fs->resource)->bo,
 220                                    v3d->prog.fs->offset);
 221
 222                 /* XXX: Use combined input/output size flag in the common
 223                  * case.
 224                  */
 225                 shader.coordinate_shader_has_separate_input_and_output_vpm_blocks =
 226                         v3d->prog.cs->prog_data.vs->separate_segments;
 227                 shader.vertex_shader_has_separate_input_and_output_vpm_blocks =
 228                         v3d->prog.vs->prog_data.vs->separate_segments;
 229
 230                 shader.coordinate_shader_input_vpm_segment_size =
 231                         v3d->prog.cs->prog_data.vs->vpm_input_size;
 232                 shader.vertex_shader_input_vpm_segment_size =
 233                         v3d->prog.vs->prog_data.vs->vpm_input_size;
 234
 235                 shader.coordinate_shader_output_vpm_segment_size =
 236                         v3d->prog.cs->prog_data.vs->vpm_output_size;
 237                 shader.vertex_shader_output_vpm_segment_size =
 238                         v3d->prog.vs->prog_data.vs->vpm_output_size;
 239
 240                 shader.coordinate_shader_uniforms_address = cs_uniforms;
 241                 shader.vertex_shader_uniforms_address = vs_uniforms;
 242                 shader.fragment_shader_uniforms_address = fs_uniforms;
 243
 244 #if V3D_VERSION >= 41
 245                 shader.min_coord_shader_input_segments_required_in_play = 1;
 246                 shader.min_vertex_shader_input_segments_required_in_play = 1;
 247
 248                 shader.coordinate_shader_4_way_threadable =
 249                         v3d->prog.cs->prog_data.vs->base.threads == 4;
 250                 shader.vertex_shader_4_way_threadable =
 251                         v3d->prog.vs->prog_data.vs->base.threads == 4;
 252                 shader.fragment_shader_4_way_threadable =
 253                         v3d->prog.fs->prog_data.fs->base.threads == 4;
 254
 255                 shader.coordinate_shader_start_in_final_thread_section =
 256                         v3d->prog.cs->prog_data.vs->base.single_seg;
 257                 shader.vertex_shader_start_in_final_thread_section =
 258                         v3d->prog.vs->prog_data.vs->base.single_seg;
 259                 shader.fragment_shader_start_in_final_thread_section =
 260                         v3d->prog.fs->prog_data.fs->base.single_seg;
 261 #else
 262                 shader.coordinate_shader_4_way_threadable =
 263                         v3d->prog.cs->prog_data.vs->base.threads == 4;
 264                 shader.coordinate_shader_2_way_threadable =
 265                         v3d->prog.cs->prog_data.vs->base.threads == 2;
 266                 shader.vertex_shader_4_way_threadable =
 267                         v3d->prog.vs->prog_data.vs->base.threads == 4;
 268                 shader.vertex_shader_2_way_threadable =
 269                         v3d->prog.vs->prog_data.vs->base.threads == 2;
 270                 shader.fragment_shader_4_way_threadable =
 271                         v3d->prog.fs->prog_data.fs->base.threads == 4;
 272                 shader.fragment_shader_2_way_threadable =
 273                         v3d->prog.fs->prog_data.fs->base.threads == 2;
 274 #endif
 275
 276                 shader.vertex_id_read_by_coordinate_shader =
 277                         v3d->prog.cs->prog_data.vs->uses_vid;
 278                 shader.instance_id_read_by_coordinate_shader =
 279                         v3d->prog.cs->prog_data.vs->uses_iid;
 280                 shader.vertex_id_read_by_vertex_shader =
 281                         v3d->prog.vs->prog_data.vs->uses_vid;
 282                 shader.instance_id_read_by_vertex_shader =
 283                         v3d->prog.vs->prog_data.vs->uses_iid;
 284
 285                 shader.address_of_default_attribute_values =
 286                         cl_address(v3d_resource(vtx->defaults)->bo,
 287                                    vtx->defaults_offset);
 288         }
 289
 290         for (int i = 0; i < vtx->num_elements; i++) {
 291                 struct pipe_vertex_element *elem = &vtx->pipe[i];
 292                 struct pipe_vertex_buffer *vb =
 293                         &vertexbuf->vb[elem->vertex_buffer_index];
 294                 struct v3d_resource *rsc = v3d_resource(vb->buffer.resource);
 295
 296                 const uint32_t size =
 297                         cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
 298                 cl_emit_with_prepacked(&job->indirect,
 299                                        GL_SHADER_STATE_ATTRIBUTE_RECORD,
 300                                        &vtx->attrs[i * size], attr) {
 301                         attr.stride = vb->stride;
 302                         attr.address = cl_address(rsc->bo,
 303                                                   vb->buffer_offset +
 304                                                   elem->src_offset);
 305                         attr.number_of_values_read_by_coordinate_shader =
 306                                 v3d->prog.cs->prog_data.vs->vattr_sizes[i];
 307                         attr.number_of_values_read_by_vertex_shader =
 308                                 v3d->prog.vs->prog_data.vs->vattr_sizes[i];
 309 #if V3D_VERSION >= 41
 310                         attr.maximum_index = 0xffffff;
 311 #endif
 312                 }
 313                 STATIC_ASSERT(sizeof(vtx->attrs) >= VC5_MAX_ATTRIBUTES * size);
 314         }
 315
 316         if (vtx->num_elements == 0) {
 317                 /* GFXH-930: At least one attribute must be enabled and read
 318                  * by CS and VS.  If we have no attributes being consumed by
 319                  * the shader, set up a dummy to be loaded into the VPM.
 320                  */
 321                 cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
 322                         /* Valid address of data whose value will be unused. */
 323                         attr.address = cl_address(job->indirect.bo, 0);
 324
 325                         attr.type = ATTRIBUTE_FLOAT;
 326                         attr.stride = 0;
 327                         attr.vec_size = 1;
 328
 329                         attr.number_of_values_read_by_coordinate_shader = 1;
 330                         attr.number_of_values_read_by_vertex_shader = 1;
 331                 }
 332         }
 333
 334         cl_emit(&job->bcl, VCM_CACHE_SIZE, vcm) {
 335                 vcm.number_of_16_vertex_batches_for_binning =
 336                         v3d->prog.cs->prog_data.vs->vcm_cache_size;
 337                 vcm.number_of_16_vertex_batches_for_rendering =
 338                         v3d->prog.vs->prog_data.vs->vcm_cache_size;
 339         }
 340
 341         cl_emit(&job->bcl, GL_SHADER_STATE, state) {
 342                 state.address = cl_address(job->indirect.bo, shader_rec_offset);
 343                 state.number_of_attribute_arrays = num_elements_to_emit;
 344         }
 345
 346         v3d_bo_unreference(&cs_uniforms.bo);
 347         v3d_bo_unreference(&vs_uniforms.bo);
 348         v3d_bo_unreference(&fs_uniforms.bo);
 349
 350         job->shader_rec_count++;
 351 }
 352
 353 /**
 354  * Computes the various transform feedback statistics, since they can't be
 355  * recorded by CL packets.
 356  */
 357 static void
 358 v3d_tf_statistics_record(struct v3d_context *v3d,
 359                          const struct pipe_draw_info *info,
 360                          bool prim_tf)
 361 {
 362         if (!v3d->active_queries)
 363                 return;
 364
 365         uint32_t prims = u_prims_for_vertices(info->mode, info->count);
 366         v3d->prims_generated += prims;
 367
 368         if (prim_tf) {
 369                 /* XXX: Only count if we didn't overflow. */
 370                 v3d->tf_prims_generated += prims;
 371         }
 372 }
 373
 374 static void
 375 v3d_update_job_ez(struct v3d_context *v3d, struct v3d_job *job)
 376 {
 377         switch (v3d->zsa->ez_state) {
 378         case VC5_EZ_UNDECIDED:
 379                 /* If the Z/S state didn't pick a direction but didn't
 380                  * disable, then go along with the current EZ state.  This
 381                  * allows EZ optimization for Z func == EQUAL or NEVER.
 382                  */
 383                 break;
 384
 385         case VC5_EZ_LT_LE:
 386         case VC5_EZ_GT_GE:
 387                 /* If the Z/S state picked a direction, then it needs to match
 388                  * the current direction if we've decided on one.
 389                  */
 390                 if (job->ez_state == VC5_EZ_UNDECIDED)
 391                         job->ez_state = v3d->zsa->ez_state;
 392                 else if (job->ez_state != v3d->zsa->ez_state)
 393                         job->ez_state = VC5_EZ_DISABLED;
 394                 break;
 395
 396         case VC5_EZ_DISABLED:
 397                 /* If the current Z/S state disables EZ because of a bad Z
 398                  * func or stencil operation, then we can't do any more EZ in
 399                  * this frame.
 400                  */
 401                 job->ez_state = VC5_EZ_DISABLED;
 402                 break;
 403         }
 404
 405         /* If the FS affects the Z of the pixels, then it may update against
 406          * the chosen EZ direction (though we could use
 407          * ARB_conservative_depth's hints to avoid this)
 408          */
 409         if (v3d->prog.fs->prog_data.fs->writes_z) {
 410                 job->ez_state = VC5_EZ_DISABLED;
 411         }
 412
 413         if (job->first_ez_state == VC5_EZ_UNDECIDED &&
 414             (job->ez_state != VC5_EZ_DISABLED || job->draw_calls_queued == 0))
 415                 job->first_ez_state = job->ez_state;
 416 }
 417
 418 static void
 419 v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
 420 {
 421         struct v3d_context *v3d = v3d_context(pctx);
 422
 423         if (!info->count_from_stream_output && !info->indirect &&
 424             !info->primitive_restart &&
 425             !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
 426                 return;
 427
 428         /* Fall back for weird desktop GL primitive restart values. */
 429         if (info->primitive_restart &&
 430             info->index_size) {
 431                 uint32_t mask = ~0;
 432
 433                 switch (info->index_size) {
 434                 case 2:
 435                         mask = 0xffff;
 436                         break;
 437                 case 1:
 438                         mask = 0xff;
 439                         break;
 440                 }
 441
 442                 if (info->restart_index != mask) {
 443                         util_draw_vbo_without_prim_restart(pctx, info);
 444                         return;
 445                 }
 446         }
 447
 448         if (info->mode >= PIPE_PRIM_QUADS) {
 449                 util_primconvert_save_rasterizer_state(v3d->primconvert, &v3d->rasterizer->base);
 450                 util_primconvert_draw_vbo(v3d->primconvert, info);
 451                 perf_debug("Fallback conversion for %d %s vertices\n",
 452                            info->count, u_prim_name(info->mode));
 453                 return;
 454         }
 455
 456         /* Before setting up the draw, flush anything writing to the textures
 457          * that we read from.
 458          */
 459         for (int s = 0; s < PIPE_SHADER_TYPES; s++)
 460                 v3d_predraw_check_stage_inputs(pctx, s);
 461
 462         if (info->indirect)
 463                 v3d_flush_jobs_writing_resource(v3d, info->indirect->buffer);
 464
 465         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 466
 467         /* If vertex texturing depends on the output of rendering, we need to
 468          * ensure that that rendering is complete before we run a coordinate
 469          * shader that depends on it.
 470          *
 471          * Given that doing that is unusual, for now we just block the binner
 472          * on the last submitted render, rather than tracking the last
 473          * rendering to each texture's BO.
 474          */
 475         if (v3d->tex[PIPE_SHADER_VERTEX].num_textures) {
 476                 perf_debug("Blocking binner on last render "
 477                            "due to vertex texturing.\n");
 478                 job->submit.in_sync_bcl = v3d->out_sync;
 479         }
 480
 481         /* Get space to emit our draw call into the BCL, using a branch to
 482          * jump to a new BO if necessary.
 483          */
 484         v3d_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
 485
 486         if (v3d->prim_mode != info->mode) {
 487                 v3d->prim_mode = info->mode;
 488                 v3d->dirty |= VC5_DIRTY_PRIM_MODE;
 489         }
 490
 491         v3d_start_draw(v3d);
 492         v3d_update_compiled_shaders(v3d, info->mode);
 493         v3d_update_job_ez(v3d, job);
 494
 495 #if V3D_VERSION >= 41
 496         v3d41_emit_state(pctx);
 497 #else
 498         v3d33_emit_state(pctx);
 499 #endif
 500
 501         if (v3d->dirty & (VC5_DIRTY_VTXBUF |
 502                           VC5_DIRTY_VTXSTATE |
 503                           VC5_DIRTY_PRIM_MODE |
 504                           VC5_DIRTY_RASTERIZER |
 505                           VC5_DIRTY_COMPILED_CS |
 506                           VC5_DIRTY_COMPILED_VS |
 507                           VC5_DIRTY_COMPILED_FS |
 508                           v3d->prog.cs->uniform_dirty_bits |
 509                           v3d->prog.vs->uniform_dirty_bits |
 510                           v3d->prog.fs->uniform_dirty_bits)) {
 511                 v3d_emit_gl_shader_state(v3d, info);
 512         }
 513
 514         v3d->dirty = 0;
 515
 516         /* The Base Vertex/Base Instance packet sets those values to nonzero
 517          * for the next draw call only.
 518          */
 519         if (info->index_bias || info->start_instance) {
 520                 cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
 521                         base.base_instance = info->start_instance;
 522                         base.base_vertex = info->index_bias;
 523                 }
 524         }
 525
 526         uint32_t prim_tf_enable = 0;
 527 #if V3D_VERSION < 40
 528         /* V3D 3.x: The HW only processes transform feedback on primitives
 529          * with the flag set.
 530          */
 531         if (v3d->streamout.num_targets)
 532                 prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
 533 #endif
 534
 535         v3d_tf_statistics_record(v3d, info, v3d->streamout.num_targets);
 536
 537         /* Note that the primitive type fields match with OpenGL/gallium
 538          * definitions, up to but not including QUADS.
 539          */
 540         if (info->index_size) {
 541                 uint32_t index_size = info->index_size;
 542                 uint32_t offset = info->start * index_size;
 543                 struct pipe_resource *prsc;
 544                 if (info->has_user_indices) {
 545                         prsc = NULL;
 546                         u_upload_data(v3d->uploader, 0,
 547                                       info->count * info->index_size, 4,
 548                                       info->index.user,
 549                                       &offset, &prsc);
 550                 } else {
 551                         prsc = info->index.resource;
 552                 }
 553                 struct v3d_resource *rsc = v3d_resource(prsc);
 554
 555 #if V3D_VERSION >= 40
 556                 cl_emit(&job->bcl, INDEX_BUFFER_SETUP, ib) {
 557                         ib.address = cl_address(rsc->bo, 0);
 558                         ib.size = rsc->bo->size;
 559                 }
 560 #endif
 561
 562                 if (info->indirect) {
 563                         cl_emit(&job->bcl, INDIRECT_INDEXED_INSTANCED_PRIM_LIST, prim) {
 564                                 prim.index_type = ffs(info->index_size) - 1;
 565 #if V3D_VERSION < 40
 566                                 prim.address_of_indices_list =
 567                                         cl_address(rsc->bo, offset);
 568 #endif /* V3D_VERSION < 40 */
 569                                 prim.mode = info->mode | prim_tf_enable;
 570                                 prim.enable_primitive_restarts = info->primitive_restart;
 571
 572                                 prim.number_of_draw_indirect_indexed_records = info->indirect->draw_count;
 573
 574                                 prim.stride_in_multiples_of_4_bytes = info->indirect->stride >> 2;
 575                                 prim.address = cl_address(v3d_resource(info->indirect->buffer)->bo,
 576                                                           info->indirect->offset);
 577                         }
 578                 } else if (info->instance_count > 1) {
 579                         cl_emit(&job->bcl, INDEXED_INSTANCED_PRIM_LIST, prim) {
 580                                 prim.index_type = ffs(info->index_size) - 1;
 581 #if V3D_VERSION >= 40
 582                                 prim.index_offset = offset;
 583 #else /* V3D_VERSION < 40 */
 584                                 prim.maximum_index = (1u << 31) - 1; /* XXX */
 585                                 prim.address_of_indices_list =
 586                                         cl_address(rsc->bo, offset);
 587 #endif /* V3D_VERSION < 40 */
 588                                 prim.mode = info->mode | prim_tf_enable;
 589                                 prim.enable_primitive_restarts = info->primitive_restart;
 590
 591                                 prim.number_of_instances = info->instance_count;
 592                                 prim.instance_length = info->count;
 593                         }
 594                 } else {
 595                         cl_emit(&job->bcl, INDEXED_PRIM_LIST, prim) {
 596                                 prim.index_type = ffs(info->index_size) - 1;
 597                                 prim.length = info->count;
 598 #if V3D_VERSION >= 40
 599                                 prim.index_offset = offset;
 600 #else /* V3D_VERSION < 40 */
 601                                 prim.maximum_index = (1u << 31) - 1; /* XXX */
 602                                 prim.address_of_indices_list =
 603                                         cl_address(rsc->bo, offset);
 604 #endif /* V3D_VERSION < 40 */
 605                                 prim.mode = info->mode | prim_tf_enable;
 606                                 prim.enable_primitive_restarts = info->primitive_restart;
 607                         }
 608                 }
 609
 610                 job->draw_calls_queued++;
 611
 612                 if (info->has_user_indices)
 613                         pipe_resource_reference(&prsc, NULL);
 614         } else {
 615                 if (info->indirect) {
 616                         cl_emit(&job->bcl, INDIRECT_VERTEX_ARRAY_INSTANCED_PRIMS, prim) {
 617                                 prim.mode = info->mode | prim_tf_enable;
 618                                 prim.number_of_draw_indirect_array_records = info->indirect->draw_count;
 619
 620                                 prim.stride_in_multiples_of_4_bytes = info->indirect->stride >> 2;
 621                                 prim.address = cl_address(v3d_resource(info->indirect->buffer)->bo,
 622                                                           info->indirect->offset);
 623                         }
 624                 } else if (info->instance_count > 1) {
 625                         cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMS, prim) {
 626                                 prim.mode = info->mode | prim_tf_enable;
 627                                 prim.index_of_first_vertex = info->start;
 628                                 prim.number_of_instances = info->instance_count;
 629                                 prim.instance_length = info->count;
 630                         }
 631                 } else {
 632                         cl_emit(&job->bcl, VERTEX_ARRAY_PRIMS, prim) {
 633                                 prim.mode = info->mode | prim_tf_enable;
 634                                 prim.length = info->count;
 635                                 prim.index_of_first_vertex = info->start;
 636                         }
 637                 }
 638         }
 639
 640         /* A flush is required in between a TF draw and any following TF specs
 641          * packet, or the GPU may hang.  Just flush each time for now.
 642          */
 643         if (v3d->streamout.num_targets)
 644                 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_FLUSH_AND_COUNT, flush);
 645
 646         job->draw_calls_queued++;
 647
 648         /* Increment the TF offsets by how many verts we wrote.  XXX: This
 649          * needs some clamping to the buffer size.
 650          */
 651         for (int i = 0; i < v3d->streamout.num_targets; i++)
 652                 v3d->streamout.offsets[i] += info->count;
 653
 654         if (v3d->zsa && job->zsbuf && v3d->zsa->base.depth.enabled) {
 655                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
 656                 v3d_job_add_bo(job, rsc->bo);
 657
 658                 job->load |= PIPE_CLEAR_DEPTH & ~job->clear;
 659                 if (v3d->zsa->base.depth.writemask)
 660                         job->store |= PIPE_CLEAR_DEPTH;
 661                 rsc->initialized_buffers = PIPE_CLEAR_DEPTH;
 662         }
 663
 664         if (v3d->zsa && job->zsbuf && v3d->zsa->base.stencil[0].enabled) {
 665                 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
 666                 if (rsc->separate_stencil)
 667                         rsc = rsc->separate_stencil;
 668
 669                 v3d_job_add_bo(job, rsc->bo);
 670
 671                 job->load |= PIPE_CLEAR_STENCIL & ~job->clear;
 672                 if (v3d->zsa->base.stencil[0].writemask ||
 673                     v3d->zsa->base.stencil[1].writemask) {
 674                         job->store |= PIPE_CLEAR_STENCIL;
 675                 }
 676                 rsc->initialized_buffers |= PIPE_CLEAR_STENCIL;
 677         }
 678
 679         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 680                 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
 681                 int blend_rt = v3d->blend->base.independent_blend_enable ? i : 0;
 682
 683                 if (job->store & bit || !job->cbufs[i])
 684                         continue;
 685                 struct v3d_resource *rsc = v3d_resource(job->cbufs[i]->texture);
 686
 687                 job->load |= bit & ~job->clear;
 688                 if (v3d->blend->base.rt[blend_rt].colormask)
 689                         job->store |= bit;
 690                 v3d_job_add_bo(job, rsc->bo);
 691         }
 692
 693         if (job->referenced_size > 768 * 1024 * 1024) {
 694                 perf_debug("Flushing job with %dkb to try to free up memory\n",
 695                         job->referenced_size / 1024);
 696                 v3d_flush(pctx);
 697         }
 698
 699         if (V3D_DEBUG & V3D_DEBUG_ALWAYS_FLUSH)
 700                 v3d_flush(pctx);
 701 }
 702
 703 /**
 704  * Implements gallium's clear() hook (glClear()) by drawing a pair of triangles.
 705  */
 706 static void
 707 v3d_draw_clear(struct v3d_context *v3d,
 708                unsigned buffers,
 709                const union pipe_color_union *color,
 710                double depth, unsigned stencil)
 711 {
 712         static const union pipe_color_union dummy_color = {};
 713
 714         /* The blitter util dereferences the color regardless, even though the
 715          * gallium clear API may not pass one in when only Z/S are cleared.
 716          */
 717         if (!color)
 718                 color = &dummy_color;
 719
 720         v3d_blitter_save(v3d);
 721         util_blitter_clear(v3d->blitter,
 722                            v3d->framebuffer.width,
 723                            v3d->framebuffer.height,
 724                            util_framebuffer_get_num_layers(&v3d->framebuffer),
 725                            buffers, color, depth, stencil);
 726 }
 727
 728 /**
 729  * Attempts to perform the GL clear by using the TLB's fast clear at the start
 730  * of the frame.
 731  */
 732 static unsigned
 733 v3d_tlb_clear(struct v3d_job *job, unsigned buffers,
 734               const union pipe_color_union *color,
 735               double depth, unsigned stencil)
 736 {
 737         struct v3d_context *v3d = job->v3d;
 738
 739         if (job->draw_calls_queued) {
 740                 /* If anything in the CL has drawn using the buffer, then the
 741                  * TLB clear we're trying to add now would happen before that
 742                  * drawing.
 743                  */
 744                 buffers &= ~(job->load | job->store);
 745         }
 746
 747         /* GFXH-1461: If we were to emit a load of just depth or just stencil,
 748          * then the clear for the other may get lost.  We need to decide now
 749          * if it would be possible to need to emit a load of just one after
 750          * we've set up our TLB clears.
 751          */
 752         if (buffers & PIPE_CLEAR_DEPTHSTENCIL &&
 753             (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL &&
 754             job->zsbuf &&
 755             util_format_is_depth_and_stencil(job->zsbuf->texture->format)) {
 756                 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
 757         }
 758
 759         for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
 760                 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
 761                 if (!(buffers & bit))
 762                         continue;
 763
 764                 struct pipe_surface *psurf = v3d->framebuffer.cbufs[i];
 765                 struct v3d_surface *surf = v3d_surface(psurf);
 766                 struct v3d_resource *rsc = v3d_resource(psurf->texture);
 767
 768                 union util_color uc;
 769                 uint32_t internal_size = 4 << surf->internal_bpp;
 770
 771                 static union pipe_color_union swapped_color;
 772                 if (v3d->swap_color_rb & (1 << i)) {
 773                         swapped_color.f[0] = color->f[2];
 774                         swapped_color.f[1] = color->f[1];
 775                         swapped_color.f[2] = color->f[0];
 776                         swapped_color.f[3] = color->f[3];
 777                         color = &swapped_color;
 778                 }
 779
 780                 switch (surf->internal_type) {
 781                 case V3D_INTERNAL_TYPE_8:
 782                         util_pack_color(color->f, PIPE_FORMAT_R8G8B8A8_UNORM,
 783                                         &uc);
 784                         memcpy(job->clear_color[i], uc.ui, internal_size);
 785                         break;
 786                 case V3D_INTERNAL_TYPE_8I:
 787                 case V3D_INTERNAL_TYPE_8UI:
 788                         job->clear_color[i][0] = ((color->ui[0] & 0xff) |
 789                                                   (color->ui[1] & 0xff) << 8 |
 790                                                   (color->ui[2] & 0xff) << 16 |
 791                                                   (color->ui[3] & 0xff) << 24);
 792                         break;
 793                 case V3D_INTERNAL_TYPE_16F:
 794                         util_pack_color(color->f, PIPE_FORMAT_R16G16B16A16_FLOAT,
 795                                         &uc);
 796                         memcpy(job->clear_color[i], uc.ui, internal_size);
 797                         break;
 798                 case V3D_INTERNAL_TYPE_16I:
 799                 case V3D_INTERNAL_TYPE_16UI:
 800                         job->clear_color[i][0] = ((color->ui[0] & 0xffff) |
 801                                                   color->ui[1] << 16);
 802                         job->clear_color[i][1] = ((color->ui[2] & 0xffff) |
 803                                                   color->ui[3] << 16);
 804                         break;
 805                 case V3D_INTERNAL_TYPE_32F:
 806                 case V3D_INTERNAL_TYPE_32I:
 807                 case V3D_INTERNAL_TYPE_32UI:
 808                         memcpy(job->clear_color[i], color->ui, internal_size);
 809                         break;
 810                 }
 811
 812                 rsc->initialized_buffers |= bit;
 813         }
 814
 815         unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
 816         if (zsclear) {
 817                 struct v3d_resource *rsc =
 818                         v3d_resource(v3d->framebuffer.zsbuf->texture);
 819
 820                 if (zsclear & PIPE_CLEAR_DEPTH)
 821                         job->clear_z = depth;
 822                 if (zsclear & PIPE_CLEAR_STENCIL)
 823                         job->clear_s = stencil;
 824
 825                 rsc->initialized_buffers |= zsclear;
 826         }
 827
 828         job->draw_min_x = 0;
 829         job->draw_min_y = 0;
 830         job->draw_max_x = v3d->framebuffer.width;
 831         job->draw_max_y = v3d->framebuffer.height;
 832         job->clear |= buffers;
 833         job->store |= buffers;
 834
 835         v3d_start_draw(v3d);
 836
 837         return buffers;
 838 }
 839
 840 static void
 841 v3d_clear(struct pipe_context *pctx, unsigned buffers,
 842           const union pipe_color_union *color, double depth, unsigned stencil)
 843 {
 844         struct v3d_context *v3d = v3d_context(pctx);
 845         struct v3d_job *job = v3d_get_job_for_fbo(v3d);
 846
 847         buffers &= ~v3d_tlb_clear(job, buffers, color, depth, stencil);
 848
 849         if (buffers)
 850                 v3d_draw_clear(v3d, buffers, color, depth, stencil);
 851 }
 852
 853 static void
 854 v3d_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
 855                         const union pipe_color_union *color,
 856                         unsigned x, unsigned y, unsigned w, unsigned h,
 857                         bool render_condition_enabled)
 858 {
 859         fprintf(stderr, "unimpl: clear RT\n");
 860 }
 861
 862 static void
 863 v3d_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
 864                         unsigned buffers, double depth, unsigned stencil,
 865                         unsigned x, unsigned y, unsigned w, unsigned h,
 866                         bool render_condition_enabled)
 867 {
 868         fprintf(stderr, "unimpl: clear DS\n");
 869 }
 870
 871 void
 872 v3dX(draw_init)(struct pipe_context *pctx)
 873 {
 874         pctx->draw_vbo = v3d_draw_vbo;
 875         pctx->clear = v3d_clear;
 876         pctx->clear_render_target = v3d_clear_render_target;
 877         pctx->clear_depth_stencil = v3d_clear_depth_stencil;
 878 }