projects / mesa.git / blob
? search:


7db1285f11d452f10ba654f13d5419e71770859b
[mesa.git] / 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 &v3d->constbuf[PIPE_SHADER_FRAGMENT],
150 &v3d->fragtex);
151 struct v3d_cl_reloc vs_uniforms =
152 v3d_write_uniforms(v3d, v3d->prog.vs,
153 &v3d->constbuf[PIPE_SHADER_VERTEX],
154 &v3d->verttex);
155 struct v3d_cl_reloc cs_uniforms =
156 v3d_write_uniforms(v3d, v3d->prog.cs,
157 &v3d->constbuf[PIPE_SHADER_VERTEX],
158 &v3d->verttex);
159
160 /* See GFXH-930 workaround below */
161 uint32_t num_elements_to_emit = MAX2(vtx->num_elements, 1);
162 uint32_t shader_rec_offset =
163 v3d_cl_ensure_space(&job->indirect,
164 cl_packet_length(GL_SHADER_STATE_RECORD) +
165 num_elements_to_emit *
166 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
167 32);
168
169 cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
170 shader.enable_clipping = true;
171 /* VC5_DIRTY_PRIM_MODE | VC5_DIRTY_RASTERIZER */
172 shader.point_size_in_shaded_vertex_data =
173 (info->mode == PIPE_PRIM_POINTS &&
174 v3d->rasterizer->base.point_size_per_vertex);
175
176 /* Must be set if the shader modifies Z, discards, or modifies
177 * the sample mask. For any of these cases, the fragment
178 * shader needs to write the Z value (even just discards).
179 */
180 shader.fragment_shader_does_z_writes =
181 (v3d->prog.fs->prog_data.fs->writes_z ||
182 v3d->prog.fs->prog_data.fs->discard);
183
184 shader.fragment_shader_uses_real_pixel_centre_w_in_addition_to_centroid_w2 =
185 v3d->prog.fs->prog_data.fs->uses_center_w;
186
187 shader.number_of_varyings_in_fragment_shader =
188 v3d->prog.fs->prog_data.base->num_inputs;
189
190 shader.coordinate_shader_propagate_nans = true;
191 shader.vertex_shader_propagate_nans = true;
192 shader.fragment_shader_propagate_nans = true;
193
194 shader.coordinate_shader_code_address =
195 cl_address(v3d->prog.cs->bo, 0);
196 shader.vertex_shader_code_address =
197 cl_address(v3d->prog.vs->bo, 0);
198 shader.fragment_shader_code_address =
199 cl_address(v3d->prog.fs->bo, 0);
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(vtx->default_attribute_values, 0);
266 }
267
268 for (int i = 0; i < vtx->num_elements; i++) {
269 struct pipe_vertex_element *elem = &vtx->pipe[i];
270 struct pipe_vertex_buffer *vb =
271 &vertexbuf->vb[elem->vertex_buffer_index];
272 struct v3d_resource *rsc = v3d_resource(vb->buffer.resource);
273
274 const uint32_t size =
275 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
276 cl_emit_with_prepacked(&job->indirect,
277 GL_SHADER_STATE_ATTRIBUTE_RECORD,
278 &vtx->attrs[i * size], attr) {
279 attr.stride = vb->stride;
280 attr.address = cl_address(rsc->bo,
281 vb->buffer_offset +
282 elem->src_offset);
283 attr.number_of_values_read_by_coordinate_shader =
284 v3d->prog.cs->prog_data.vs->vattr_sizes[i];
285 attr.number_of_values_read_by_vertex_shader =
286 v3d->prog.vs->prog_data.vs->vattr_sizes[i];
287 #if V3D_VERSION >= 41
288 attr.maximum_index = 0xffffff;
289 #endif
290 }
291 STATIC_ASSERT(sizeof(vtx->attrs) >= VC5_MAX_ATTRIBUTES * size);
292 }
293
294 if (vtx->num_elements == 0) {
295 /* GFXH-930: At least one attribute must be enabled and read
296 * by CS and VS. If we have no attributes being consumed by
297 * the shader, set up a dummy to be loaded into the VPM.
298 */
299 cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
300 /* Valid address of data whose value will be unused. */
301 attr.address = cl_address(job->indirect.bo, 0);
302
303 attr.type = ATTRIBUTE_FLOAT;
304 attr.stride = 0;
305 attr.vec_size = 1;
306
307 attr.number_of_values_read_by_coordinate_shader = 1;
308 attr.number_of_values_read_by_vertex_shader = 1;
309 }
310 }
311
312 cl_emit(&job->bcl, VCM_CACHE_SIZE, vcm) {
313 vcm.number_of_16_vertex_batches_for_binning =
314 v3d->prog.cs->prog_data.vs->vcm_cache_size;
315 vcm.number_of_16_vertex_batches_for_rendering =
316 v3d->prog.vs->prog_data.vs->vcm_cache_size;
317 }
318
319 cl_emit(&job->bcl, GL_SHADER_STATE, state) {
320 state.address = cl_address(job->indirect.bo, shader_rec_offset);
321 state.number_of_attribute_arrays = num_elements_to_emit;
322 }
323
324 v3d_bo_unreference(&cs_uniforms.bo);
325 v3d_bo_unreference(&vs_uniforms.bo);
326 v3d_bo_unreference(&fs_uniforms.bo);
327
328 job->shader_rec_count++;
329 }
330
331 /**
332 * Computes the various transform feedback statistics, since they can't be
333 * recorded by CL packets.
334 */
335 static void
336 v3d_tf_statistics_record(struct v3d_context *v3d,
337 const struct pipe_draw_info *info,
338 bool prim_tf)
339 {
340 if (!v3d->active_queries)
341 return;
342
343 uint32_t prims = u_prims_for_vertices(info->mode, info->count);
344 v3d->prims_generated += prims;
345
346 if (prim_tf) {
347 /* XXX: Only count if we didn't overflow. */
348 v3d->tf_prims_generated += prims;
349 }
350 }
351
352 static void
353 v3d_update_job_ez(struct v3d_context *v3d, struct v3d_job *job)
354 {
355 switch (v3d->zsa->ez_state) {
356 case VC5_EZ_UNDECIDED:
357 /* If the Z/S state didn't pick a direction but didn't
358 * disable, then go along with the current EZ state. This
359 * allows EZ optimization for Z func == EQUAL or NEVER.
360 */
361 break;
362
363 case VC5_EZ_LT_LE:
364 case VC5_EZ_GT_GE:
365 /* If the Z/S state picked a direction, then it needs to match
366 * the current direction if we've decided on one.
367 */
368 if (job->ez_state == VC5_EZ_UNDECIDED)
369 job->ez_state = v3d->zsa->ez_state;
370 else if (job->ez_state != v3d->zsa->ez_state)
371 job->ez_state = VC5_EZ_DISABLED;
372 break;
373
374 case VC5_EZ_DISABLED:
375 /* If the current Z/S state disables EZ because of a bad Z
376 * func or stencil operation, then we can't do any more EZ in
377 * this frame.
378 */
379 job->ez_state = VC5_EZ_DISABLED;
380 break;
381 }
382
383 /* If the FS affects the Z of the pixels, then it may update against
384 * the chosen EZ direction (though we could use
385 * ARB_conservative_depth's hints to avoid this)
386 */
387 if (v3d->prog.fs->prog_data.fs->writes_z) {
388 job->ez_state = VC5_EZ_DISABLED;
389 }
390
391 if (job->first_ez_state == VC5_EZ_UNDECIDED &&
392 (job->ez_state != VC5_EZ_DISABLED || job->draw_calls_queued == 0))
393 job->first_ez_state = job->ez_state;
394 }
395
396 static void
397 v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
398 {
399 struct v3d_context *v3d = v3d_context(pctx);
400
401 if (!info->count_from_stream_output && !info->indirect &&
402 !info->primitive_restart &&
403 !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
404 return;
405
406 /* Fall back for weird desktop GL primitive restart values. */
407 if (info->primitive_restart &&
408 info->index_size) {
409 uint32_t mask = ~0;
410
411 switch (info->index_size) {
412 case 2:
413 mask = 0xffff;
414 break;
415 case 1:
416 mask = 0xff;
417 break;
418 }
419
420 if (info->restart_index != mask) {
421 util_draw_vbo_without_prim_restart(pctx, info);
422 return;
423 }
424 }
425
426 if (info->mode >= PIPE_PRIM_QUADS) {
427 util_primconvert_save_rasterizer_state(v3d->primconvert, &v3d->rasterizer->base);
428 util_primconvert_draw_vbo(v3d->primconvert, info);
429 perf_debug("Fallback conversion for %d %s vertices\n",
430 info->count, u_prim_name(info->mode));
431 return;
432 }
433
434 /* Before setting up the draw, flush anything writing to the textures
435 * that we read from.
436 */
437 v3d_predraw_check_textures(pctx, &v3d->verttex);
438 v3d_predraw_check_textures(pctx, &v3d->fragtex);
439
440 struct v3d_job *job = v3d_get_job_for_fbo(v3d);
441
442 /* If vertex texturing depends on the output of rendering, we need to
443 * ensure that that rendering is complete before we run a coordinate
444 * shader that depends on it.
445 *
446 * Given that doing that is unusual, for now we just block the binner
447 * on the last submitted render, rather than tracking the last
448 * rendering to each texture's BO.
449 */
450 if (v3d->verttex.num_textures) {
451 perf_debug("Blocking binner on last render "
452 "due to vertex texturing.\n");
453 job->submit.in_sync_bcl = v3d->out_sync;
454 }
455
456 /* Get space to emit our draw call into the BCL, using a branch to
457 * jump to a new BO if necessary.
458 */
459 v3d_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
460
461 if (v3d->prim_mode != info->mode) {
462 v3d->prim_mode = info->mode;
463 v3d->dirty |= VC5_DIRTY_PRIM_MODE;
464 }
465
466 v3d_start_draw(v3d);
467 v3d_update_compiled_shaders(v3d, info->mode);
468 v3d_update_job_ez(v3d, job);
469
470 #if V3D_VERSION >= 41
471 v3d41_emit_state(pctx);
472 #else
473 v3d33_emit_state(pctx);
474 #endif
475
476 if (v3d->dirty & (VC5_DIRTY_VTXBUF |
477 VC5_DIRTY_VTXSTATE |
478 VC5_DIRTY_PRIM_MODE |
479 VC5_DIRTY_RASTERIZER |
480 VC5_DIRTY_COMPILED_CS |
481 VC5_DIRTY_COMPILED_VS |
482 VC5_DIRTY_COMPILED_FS |
483 v3d->prog.cs->uniform_dirty_bits |
484 v3d->prog.vs->uniform_dirty_bits |
485 v3d->prog.fs->uniform_dirty_bits)) {
486 v3d_emit_gl_shader_state(v3d, info);
487 }
488
489 v3d->dirty = 0;
490
491 /* The Base Vertex/Base Instance packet sets those values to nonzero
492 * for the next draw call only.
493 */
494 if (info->index_bias || info->start_instance) {
495 cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
496 base.base_instance = info->start_instance;
497 base.base_vertex = info->index_bias;
498 }
499 }
500
501 uint32_t prim_tf_enable = 0;
502 #if V3D_VERSION < 40
503 /* V3D 3.x: The HW only processes transform feedback on primitives
504 * with the flag set.
505 */
506 if (v3d->streamout.num_targets)
507 prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
508 #endif
509
510 v3d_tf_statistics_record(v3d, info, v3d->streamout.num_targets);
511
512 /* Note that the primitive type fields match with OpenGL/gallium
513 * definitions, up to but not including QUADS.
514 */
515 if (info->index_size) {
516 uint32_t index_size = info->index_size;
517 uint32_t offset = info->start * index_size;
518 struct pipe_resource *prsc;
519 if (info->has_user_indices) {
520 prsc = NULL;
521 u_upload_data(v3d->uploader, 0,
522 info->count * info->index_size, 4,
523 info->index.user,
524 &offset, &prsc);
525 } else {
526 prsc = info->index.resource;
527 }
528 struct v3d_resource *rsc = v3d_resource(prsc);
529
530 #if V3D_VERSION >= 40
531 cl_emit(&job->bcl, INDEX_BUFFER_SETUP, ib) {
532 ib.address = cl_address(rsc->bo, 0);
533 ib.size = rsc->bo->size;
534 }
535 #endif
536
537 if (info->instance_count > 1) {
538 cl_emit(&job->bcl, INDEXED_INSTANCED_PRIM_LIST, prim) {
539 prim.index_type = ffs(info->index_size) - 1;
540 #if V3D_VERSION >= 40
541 prim.index_offset = offset;
542 #else /* V3D_VERSION < 40 */
543 prim.maximum_index = (1u << 31) - 1; /* XXX */
544 prim.address_of_indices_list =
545 cl_address(rsc->bo, offset);
546 #endif /* V3D_VERSION < 40 */
547 prim.mode = info->mode | prim_tf_enable;
548 prim.enable_primitive_restarts = info->primitive_restart;
549
550 prim.number_of_instances = info->instance_count;
551 prim.instance_length = info->count;
552 }
553 } else {
554 cl_emit(&job->bcl, INDEXED_PRIM_LIST, prim) {
555 prim.index_type = ffs(info->index_size) - 1;
556 prim.length = info->count;
557 #if V3D_VERSION >= 40
558 prim.index_offset = offset;
559 #else /* V3D_VERSION < 40 */
560 prim.maximum_index = (1u << 31) - 1; /* XXX */
561 prim.address_of_indices_list =
562 cl_address(rsc->bo, offset);
563 #endif /* V3D_VERSION < 40 */
564 prim.mode = info->mode | prim_tf_enable;
565 prim.enable_primitive_restarts = info->primitive_restart;
566 }
567 }
568
569 job->draw_calls_queued++;
570
571 if (info->has_user_indices)
572 pipe_resource_reference(&prsc, NULL);
573 } else {
574 if (info->instance_count > 1) {
575 cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMS, prim) {
576 prim.mode = info->mode | prim_tf_enable;
577 prim.index_of_first_vertex = info->start;
578 prim.number_of_instances = info->instance_count;
579 prim.instance_length = info->count;
580 }
581 } else {
582 cl_emit(&job->bcl, VERTEX_ARRAY_PRIMS, prim) {
583 prim.mode = info->mode | prim_tf_enable;
584 prim.length = info->count;
585 prim.index_of_first_vertex = info->start;
586 }
587 }
588 }
589
590 /* A flush is required in between a TF draw and any following TF specs
591 * packet, or the GPU may hang. Just flush each time for now.
592 */
593 if (v3d->streamout.num_targets)
594 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_FLUSH_AND_COUNT, flush);
595
596 job->draw_calls_queued++;
597
598 /* Increment the TF offsets by how many verts we wrote. XXX: This
599 * needs some clamping to the buffer size.
600 */
601 for (int i = 0; i < v3d->streamout.num_targets; i++)
602 v3d->streamout.offsets[i] += info->count;
603
604 if (v3d->zsa && job->zsbuf && v3d->zsa->base.depth.enabled) {
605 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
606 v3d_job_add_bo(job, rsc->bo);
607
608 job->load |= PIPE_CLEAR_DEPTH & ~job->clear;
609 if (v3d->zsa->base.depth.writemask)
610 job->store |= PIPE_CLEAR_DEPTH;
611 rsc->initialized_buffers = PIPE_CLEAR_DEPTH;
612 }
613
614 if (v3d->zsa && job->zsbuf && v3d->zsa->base.stencil[0].enabled) {
615 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
616 if (rsc->separate_stencil)
617 rsc = rsc->separate_stencil;
618
619 v3d_job_add_bo(job, rsc->bo);
620
621 job->load |= PIPE_CLEAR_STENCIL & ~job->clear;
622 if (v3d->zsa->base.stencil[0].writemask ||
623 v3d->zsa->base.stencil[1].writemask) {
624 job->store |= PIPE_CLEAR_STENCIL;
625 }
626 rsc->initialized_buffers |= PIPE_CLEAR_STENCIL;
627 }
628
629 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
630 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
631 int blend_rt = v3d->blend->base.independent_blend_enable ? i : 0;
632
633 if (job->store & bit || !job->cbufs[i])
634 continue;
635 struct v3d_resource *rsc = v3d_resource(job->cbufs[i]->texture);
636
637 job->load |= bit & ~job->clear;
638 if (v3d->blend->base.rt[blend_rt].colormask)
639 job->store |= bit;
640 v3d_job_add_bo(job, rsc->bo);
641 }
642
643 if (job->referenced_size > 768 * 1024 * 1024) {
644 perf_debug("Flushing job with %dkb to try to free up memory\n",
645 job->referenced_size / 1024);
646 v3d_flush(pctx);
647 }
648
649 if (V3D_DEBUG & V3D_DEBUG_ALWAYS_FLUSH)
650 v3d_flush(pctx);
651 }
652
653 /**
654 * Implements gallium's clear() hook (glClear()) by drawing a pair of triangles.
655 */
656 static void
657 v3d_draw_clear(struct v3d_context *v3d,
658 unsigned buffers,
659 const union pipe_color_union *color,
660 double depth, unsigned stencil)
661 {
662 static const union pipe_color_union dummy_color = {};
663
664 /* The blitter util dereferences the color regardless, even though the
665 * gallium clear API may not pass one in when only Z/S are cleared.
666 */
667 if (!color)
668 color = &dummy_color;
669
670 v3d_blitter_save(v3d);
671 util_blitter_clear(v3d->blitter,
672 v3d->framebuffer.width,
673 v3d->framebuffer.height,
674 util_framebuffer_get_num_layers(&v3d->framebuffer),
675 buffers, color, depth, stencil);
676 }
677
678 /**
679 * Attempts to perform the GL clear by using the TLB's fast clear at the start
680 * of the frame.
681 */
682 static unsigned
683 v3d_tlb_clear(struct v3d_job *job, unsigned buffers,
684 const union pipe_color_union *color,
685 double depth, unsigned stencil)
686 {
687 struct v3d_context *v3d = job->v3d;
688
689 if (job->draw_calls_queued) {
690 /* If anything in the CL has drawn using the buffer, then the
691 * TLB clear we're trying to add now would happen before that
692 * drawing.
693 */
694 buffers &= ~(job->load | job->store);
695 }
696
697 /* GFXH-1461: If we were to emit a load of just depth or just stencil,
698 * then the clear for the other may get lost. We need to decide now
699 * if it would be possible to need to emit a load of just one after
700 * we've set up our TLB clears.
701 */
702 if (buffers & PIPE_CLEAR_DEPTHSTENCIL &&
703 (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL &&
704 job->zsbuf &&
705 util_format_is_depth_and_stencil(job->zsbuf->texture->format)) {
706 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
707 }
708
709 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
710 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
711 if (!(buffers & bit))
712 continue;
713
714 struct pipe_surface *psurf = v3d->framebuffer.cbufs[i];
715 struct v3d_surface *surf = v3d_surface(psurf);
716 struct v3d_resource *rsc = v3d_resource(psurf->texture);
717
718 union util_color uc;
719 uint32_t internal_size = 4 << surf->internal_bpp;
720
721 static union pipe_color_union swapped_color;
722 if (v3d->swap_color_rb & (1 << i)) {
723 swapped_color.f[0] = color->f[2];
724 swapped_color.f[1] = color->f[1];
725 swapped_color.f[2] = color->f[0];
726 swapped_color.f[3] = color->f[3];
727 color = &swapped_color;
728 }
729
730 switch (surf->internal_type) {
731 case V3D_INTERNAL_TYPE_8:
732 util_pack_color(color->f, PIPE_FORMAT_R8G8B8A8_UNORM,
733 &uc);
734 memcpy(job->clear_color[i], uc.ui, internal_size);
735 break;
736 case V3D_INTERNAL_TYPE_8I:
737 case V3D_INTERNAL_TYPE_8UI:
738 job->clear_color[i][0] = ((color->ui[0] & 0xff) |
739 (color->ui[1] & 0xff) << 8 |
740 (color->ui[2] & 0xff) << 16 |
741 (color->ui[3] & 0xff) << 24);
742 break;
743 case V3D_INTERNAL_TYPE_16F:
744 util_pack_color(color->f, PIPE_FORMAT_R16G16B16A16_FLOAT,
745 &uc);
746 memcpy(job->clear_color[i], uc.ui, internal_size);
747 break;
748 case V3D_INTERNAL_TYPE_16I:
749 case V3D_INTERNAL_TYPE_16UI:
750 job->clear_color[i][0] = ((color->ui[0] & 0xffff) |
751 color->ui[1] << 16);
752 job->clear_color[i][1] = ((color->ui[2] & 0xffff) |
753 color->ui[3] << 16);
754 break;
755 case V3D_INTERNAL_TYPE_32F:
756 case V3D_INTERNAL_TYPE_32I:
757 case V3D_INTERNAL_TYPE_32UI:
758 memcpy(job->clear_color[i], color->ui, internal_size);
759 break;
760 }
761
762 rsc->initialized_buffers |= bit;
763 }
764
765 unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
766 if (zsclear) {
767 struct v3d_resource *rsc =
768 v3d_resource(v3d->framebuffer.zsbuf->texture);
769
770 if (zsclear & PIPE_CLEAR_DEPTH)
771 job->clear_z = depth;
772 if (zsclear & PIPE_CLEAR_STENCIL)
773 job->clear_s = stencil;
774
775 rsc->initialized_buffers |= zsclear;
776 }
777
778 job->draw_min_x = 0;
779 job->draw_min_y = 0;
780 job->draw_max_x = v3d->framebuffer.width;
781 job->draw_max_y = v3d->framebuffer.height;
782 job->clear |= buffers;
783 job->store |= buffers;
784
785 v3d_start_draw(v3d);
786
787 return buffers;
788 }
789
790 static void
791 v3d_clear(struct pipe_context *pctx, unsigned buffers,
792 const union pipe_color_union *color, double depth, unsigned stencil)
793 {
794 struct v3d_context *v3d = v3d_context(pctx);
795 struct v3d_job *job = v3d_get_job_for_fbo(v3d);
796
797 buffers &= ~v3d_tlb_clear(job, buffers, color, depth, stencil);
798
799 if (buffers)
800 v3d_draw_clear(v3d, buffers, color, depth, stencil);
801 }
802
803 static void
804 v3d_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
805 const union pipe_color_union *color,
806 unsigned x, unsigned y, unsigned w, unsigned h,
807 bool render_condition_enabled)
808 {
809 fprintf(stderr, "unimpl: clear RT\n");
810 }
811
812 static void
813 v3d_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
814 unsigned buffers, double depth, unsigned stencil,
815 unsigned x, unsigned y, unsigned w, unsigned h,
816 bool render_condition_enabled)
817 {
818 fprintf(stderr, "unimpl: clear DS\n");
819 }
820
821 void
822 v3dX(draw_init)(struct pipe_context *pctx)
823 {
824 pctx->draw_vbo = v3d_draw_vbo;
825 pctx->clear = v3d_clear;
826 pctx->clear_render_target = v3d_clear_render_target;
827 pctx->clear_depth_stencil = v3d_clear_depth_stencil;
828 }