v3d: Emit the VCM_CACHE_SIZE packet.
[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 = true;
205 shader.vertex_shader_has_separate_input_and_output_vpm_blocks = true;
206 shader.coordinate_shader_input_vpm_segment_size =
207 MAX2(v3d->prog.cs->prog_data.vs->vpm_input_size, 1);
208 shader.vertex_shader_input_vpm_segment_size =
209 MAX2(v3d->prog.vs->prog_data.vs->vpm_input_size, 1);
210
211 shader.coordinate_shader_output_vpm_segment_size =
212 v3d->prog.cs->prog_data.vs->vpm_output_size;
213 shader.vertex_shader_output_vpm_segment_size =
214 v3d->prog.vs->prog_data.vs->vpm_output_size;
215
216 shader.coordinate_shader_uniforms_address = cs_uniforms;
217 shader.vertex_shader_uniforms_address = vs_uniforms;
218 shader.fragment_shader_uniforms_address = fs_uniforms;
219
220 #if V3D_VERSION >= 41
221 shader.min_coord_shader_input_segments_required_in_play = 1;
222 shader.min_vertex_shader_input_segments_required_in_play = 1;
223
224 shader.coordinate_shader_4_way_threadable =
225 v3d->prog.cs->prog_data.vs->base.threads == 4;
226 shader.vertex_shader_4_way_threadable =
227 v3d->prog.vs->prog_data.vs->base.threads == 4;
228 shader.fragment_shader_4_way_threadable =
229 v3d->prog.fs->prog_data.fs->base.threads == 4;
230
231 shader.coordinate_shader_start_in_final_thread_section =
232 v3d->prog.cs->prog_data.vs->base.single_seg;
233 shader.vertex_shader_start_in_final_thread_section =
234 v3d->prog.vs->prog_data.vs->base.single_seg;
235 shader.fragment_shader_start_in_final_thread_section =
236 v3d->prog.fs->prog_data.fs->base.single_seg;
237 #else
238 shader.coordinate_shader_4_way_threadable =
239 v3d->prog.cs->prog_data.vs->base.threads == 4;
240 shader.coordinate_shader_2_way_threadable =
241 v3d->prog.cs->prog_data.vs->base.threads == 2;
242 shader.vertex_shader_4_way_threadable =
243 v3d->prog.vs->prog_data.vs->base.threads == 4;
244 shader.vertex_shader_2_way_threadable =
245 v3d->prog.vs->prog_data.vs->base.threads == 2;
246 shader.fragment_shader_4_way_threadable =
247 v3d->prog.fs->prog_data.fs->base.threads == 4;
248 shader.fragment_shader_2_way_threadable =
249 v3d->prog.fs->prog_data.fs->base.threads == 2;
250 #endif
251
252 shader.vertex_id_read_by_coordinate_shader =
253 v3d->prog.cs->prog_data.vs->uses_vid;
254 shader.instance_id_read_by_coordinate_shader =
255 v3d->prog.cs->prog_data.vs->uses_iid;
256 shader.vertex_id_read_by_vertex_shader =
257 v3d->prog.vs->prog_data.vs->uses_vid;
258 shader.instance_id_read_by_vertex_shader =
259 v3d->prog.vs->prog_data.vs->uses_iid;
260
261 shader.address_of_default_attribute_values =
262 cl_address(vtx->default_attribute_values, 0);
263 }
264
265 for (int i = 0; i < vtx->num_elements; i++) {
266 struct pipe_vertex_element *elem = &vtx->pipe[i];
267 struct pipe_vertex_buffer *vb =
268 &vertexbuf->vb[elem->vertex_buffer_index];
269 struct v3d_resource *rsc = v3d_resource(vb->buffer.resource);
270
271 const uint32_t size =
272 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
273 cl_emit_with_prepacked(&job->indirect,
274 GL_SHADER_STATE_ATTRIBUTE_RECORD,
275 &vtx->attrs[i * size], attr) {
276 attr.stride = vb->stride;
277 attr.address = cl_address(rsc->bo,
278 vb->buffer_offset +
279 elem->src_offset);
280 attr.number_of_values_read_by_coordinate_shader =
281 v3d->prog.cs->prog_data.vs->vattr_sizes[i];
282 attr.number_of_values_read_by_vertex_shader =
283 v3d->prog.vs->prog_data.vs->vattr_sizes[i];
284 #if V3D_VERSION >= 41
285 attr.maximum_index = 0xffffff;
286 #endif
287 }
288 STATIC_ASSERT(sizeof(vtx->attrs) >= VC5_MAX_ATTRIBUTES * size);
289 }
290
291 if (vtx->num_elements == 0) {
292 /* GFXH-930: At least one attribute must be enabled and read
293 * by CS and VS. If we have no attributes being consumed by
294 * the shader, set up a dummy to be loaded into the VPM.
295 */
296 cl_emit(&job->indirect, GL_SHADER_STATE_ATTRIBUTE_RECORD, attr) {
297 /* Valid address of data whose value will be unused. */
298 attr.address = cl_address(job->indirect.bo, 0);
299
300 attr.type = ATTRIBUTE_FLOAT;
301 attr.stride = 0;
302 attr.vec_size = 1;
303
304 attr.number_of_values_read_by_coordinate_shader = 1;
305 attr.number_of_values_read_by_vertex_shader = 1;
306 }
307 }
308
309 cl_emit(&job->bcl, VCM_CACHE_SIZE, vcm) {
310 vcm.number_of_16_vertex_batches_for_binning =
311 v3d->prog.cs->prog_data.vs->vcm_cache_size;
312 vcm.number_of_16_vertex_batches_for_rendering =
313 v3d->prog.vs->prog_data.vs->vcm_cache_size;
314 }
315
316 cl_emit(&job->bcl, GL_SHADER_STATE, state) {
317 state.address = cl_address(job->indirect.bo, shader_rec_offset);
318 state.number_of_attribute_arrays = num_elements_to_emit;
319 }
320
321 v3d_bo_unreference(&cs_uniforms.bo);
322 v3d_bo_unreference(&vs_uniforms.bo);
323 v3d_bo_unreference(&fs_uniforms.bo);
324
325 job->shader_rec_count++;
326 }
327
328 /**
329 * Computes the various transform feedback statistics, since they can't be
330 * recorded by CL packets.
331 */
332 static void
333 v3d_tf_statistics_record(struct v3d_context *v3d,
334 const struct pipe_draw_info *info,
335 bool prim_tf)
336 {
337 if (!v3d->active_queries)
338 return;
339
340 uint32_t prims = u_prims_for_vertices(info->mode, info->count);
341 v3d->prims_generated += prims;
342
343 if (prim_tf) {
344 /* XXX: Only count if we didn't overflow. */
345 v3d->tf_prims_generated += prims;
346 }
347 }
348
349 static void
350 v3d_update_job_ez(struct v3d_context *v3d, struct v3d_job *job)
351 {
352 switch (v3d->zsa->ez_state) {
353 case VC5_EZ_UNDECIDED:
354 /* If the Z/S state didn't pick a direction but didn't
355 * disable, then go along with the current EZ state. This
356 * allows EZ optimization for Z func == EQUAL or NEVER.
357 */
358 break;
359
360 case VC5_EZ_LT_LE:
361 case VC5_EZ_GT_GE:
362 /* If the Z/S state picked a direction, then it needs to match
363 * the current direction if we've decided on one.
364 */
365 if (job->ez_state == VC5_EZ_UNDECIDED)
366 job->ez_state = v3d->zsa->ez_state;
367 else if (job->ez_state != v3d->zsa->ez_state)
368 job->ez_state = VC5_EZ_DISABLED;
369 break;
370
371 case VC5_EZ_DISABLED:
372 /* If the current Z/S state disables EZ because of a bad Z
373 * func or stencil operation, then we can't do any more EZ in
374 * this frame.
375 */
376 job->ez_state = VC5_EZ_DISABLED;
377 break;
378 }
379
380 /* If the FS affects the Z of the pixels, then it may update against
381 * the chosen EZ direction (though we could use
382 * ARB_conservative_depth's hints to avoid this)
383 */
384 if (v3d->prog.fs->prog_data.fs->writes_z) {
385 job->ez_state = VC5_EZ_DISABLED;
386 }
387
388 if (job->first_ez_state == VC5_EZ_UNDECIDED &&
389 (job->ez_state != VC5_EZ_DISABLED || job->draw_calls_queued == 0))
390 job->first_ez_state = job->ez_state;
391 }
392
393 static void
394 v3d_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
395 {
396 struct v3d_context *v3d = v3d_context(pctx);
397
398 if (!info->count_from_stream_output && !info->indirect &&
399 !info->primitive_restart &&
400 !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
401 return;
402
403 /* Fall back for weird desktop GL primitive restart values. */
404 if (info->primitive_restart &&
405 info->index_size) {
406 uint32_t mask = ~0;
407
408 switch (info->index_size) {
409 case 2:
410 mask = 0xffff;
411 break;
412 case 1:
413 mask = 0xff;
414 break;
415 }
416
417 if (info->restart_index != mask) {
418 util_draw_vbo_without_prim_restart(pctx, info);
419 return;
420 }
421 }
422
423 if (info->mode >= PIPE_PRIM_QUADS) {
424 util_primconvert_save_rasterizer_state(v3d->primconvert, &v3d->rasterizer->base);
425 util_primconvert_draw_vbo(v3d->primconvert, info);
426 perf_debug("Fallback conversion for %d %s vertices\n",
427 info->count, u_prim_name(info->mode));
428 return;
429 }
430
431 /* Before setting up the draw, flush anything writing to the textures
432 * that we read from.
433 */
434 v3d_predraw_check_textures(pctx, &v3d->verttex);
435 v3d_predraw_check_textures(pctx, &v3d->fragtex);
436
437 struct v3d_job *job = v3d_get_job_for_fbo(v3d);
438
439 /* If vertex texturing depends on the output of rendering, we need to
440 * ensure that that rendering is complete before we run a coordinate
441 * shader that depends on it.
442 *
443 * Given that doing that is unusual, for now we just block the binner
444 * on the last submitted render, rather than tracking the last
445 * rendering to each texture's BO.
446 */
447 if (v3d->verttex.num_textures) {
448 perf_debug("Blocking binner on last render "
449 "due to vertex texturing.\n");
450 job->submit.in_sync_bcl = v3d->out_sync;
451 }
452
453 /* Get space to emit our draw call into the BCL, using a branch to
454 * jump to a new BO if necessary.
455 */
456 v3d_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
457
458 if (v3d->prim_mode != info->mode) {
459 v3d->prim_mode = info->mode;
460 v3d->dirty |= VC5_DIRTY_PRIM_MODE;
461 }
462
463 v3d_start_draw(v3d);
464 v3d_update_compiled_shaders(v3d, info->mode);
465 v3d_update_job_ez(v3d, job);
466
467 #if V3D_VERSION >= 41
468 v3d41_emit_state(pctx);
469 #else
470 v3d33_emit_state(pctx);
471 #endif
472
473 if (v3d->dirty & (VC5_DIRTY_VTXBUF |
474 VC5_DIRTY_VTXSTATE |
475 VC5_DIRTY_PRIM_MODE |
476 VC5_DIRTY_RASTERIZER |
477 VC5_DIRTY_COMPILED_CS |
478 VC5_DIRTY_COMPILED_VS |
479 VC5_DIRTY_COMPILED_FS |
480 v3d->prog.cs->uniform_dirty_bits |
481 v3d->prog.vs->uniform_dirty_bits |
482 v3d->prog.fs->uniform_dirty_bits)) {
483 v3d_emit_gl_shader_state(v3d, info);
484 }
485
486 v3d->dirty = 0;
487
488 /* The Base Vertex/Base Instance packet sets those values to nonzero
489 * for the next draw call only.
490 */
491 if (info->index_bias || info->start_instance) {
492 cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
493 base.base_instance = info->start_instance;
494 base.base_vertex = info->index_bias;
495 }
496 }
497
498 uint32_t prim_tf_enable = 0;
499 #if V3D_VERSION < 40
500 /* V3D 3.x: The HW only processes transform feedback on primitives
501 * with the flag set.
502 */
503 if (v3d->streamout.num_targets)
504 prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
505 #endif
506
507 v3d_tf_statistics_record(v3d, info, v3d->streamout.num_targets);
508
509 /* Note that the primitive type fields match with OpenGL/gallium
510 * definitions, up to but not including QUADS.
511 */
512 if (info->index_size) {
513 uint32_t index_size = info->index_size;
514 uint32_t offset = info->start * index_size;
515 struct pipe_resource *prsc;
516 if (info->has_user_indices) {
517 prsc = NULL;
518 u_upload_data(v3d->uploader, 0,
519 info->count * info->index_size, 4,
520 info->index.user,
521 &offset, &prsc);
522 } else {
523 prsc = info->index.resource;
524 }
525 struct v3d_resource *rsc = v3d_resource(prsc);
526
527 #if V3D_VERSION >= 40
528 cl_emit(&job->bcl, INDEX_BUFFER_SETUP, ib) {
529 ib.address = cl_address(rsc->bo, 0);
530 ib.size = rsc->bo->size;
531 }
532 #endif
533
534 if (info->instance_count > 1) {
535 cl_emit(&job->bcl, INDEXED_INSTANCED_PRIM_LIST, prim) {
536 prim.index_type = ffs(info->index_size) - 1;
537 #if V3D_VERSION >= 40
538 prim.index_offset = offset;
539 #else /* V3D_VERSION < 40 */
540 prim.maximum_index = (1u << 31) - 1; /* XXX */
541 prim.address_of_indices_list =
542 cl_address(rsc->bo, offset);
543 #endif /* V3D_VERSION < 40 */
544 prim.mode = info->mode | prim_tf_enable;
545 prim.enable_primitive_restarts = info->primitive_restart;
546
547 prim.number_of_instances = info->instance_count;
548 prim.instance_length = info->count;
549 }
550 } else {
551 cl_emit(&job->bcl, INDEXED_PRIM_LIST, prim) {
552 prim.index_type = ffs(info->index_size) - 1;
553 prim.length = info->count;
554 #if V3D_VERSION >= 40
555 prim.index_offset = offset;
556 #else /* V3D_VERSION < 40 */
557 prim.maximum_index = (1u << 31) - 1; /* XXX */
558 prim.address_of_indices_list =
559 cl_address(rsc->bo, offset);
560 #endif /* V3D_VERSION < 40 */
561 prim.mode = info->mode | prim_tf_enable;
562 prim.enable_primitive_restarts = info->primitive_restart;
563 }
564 }
565
566 job->draw_calls_queued++;
567
568 if (info->has_user_indices)
569 pipe_resource_reference(&prsc, NULL);
570 } else {
571 if (info->instance_count > 1) {
572 cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMS, prim) {
573 prim.mode = info->mode | prim_tf_enable;
574 prim.index_of_first_vertex = info->start;
575 prim.number_of_instances = info->instance_count;
576 prim.instance_length = info->count;
577 }
578 } else {
579 cl_emit(&job->bcl, VERTEX_ARRAY_PRIMS, prim) {
580 prim.mode = info->mode | prim_tf_enable;
581 prim.length = info->count;
582 prim.index_of_first_vertex = info->start;
583 }
584 }
585 }
586
587 /* A flush is required in between a TF draw and any following TF specs
588 * packet, or the GPU may hang. Just flush each time for now.
589 */
590 if (v3d->streamout.num_targets)
591 cl_emit(&job->bcl, TRANSFORM_FEEDBACK_FLUSH_AND_COUNT, flush);
592
593 job->draw_calls_queued++;
594
595 /* Increment the TF offsets by how many verts we wrote. XXX: This
596 * needs some clamping to the buffer size.
597 */
598 for (int i = 0; i < v3d->streamout.num_targets; i++)
599 v3d->streamout.offsets[i] += info->count;
600
601 if (v3d->zsa && job->zsbuf && v3d->zsa->base.depth.enabled) {
602 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
603 v3d_job_add_bo(job, rsc->bo);
604
605 job->load |= PIPE_CLEAR_DEPTH & ~job->clear;
606 if (v3d->zsa->base.depth.writemask)
607 job->store |= PIPE_CLEAR_DEPTH;
608 rsc->initialized_buffers = PIPE_CLEAR_DEPTH;
609 }
610
611 if (v3d->zsa && job->zsbuf && v3d->zsa->base.stencil[0].enabled) {
612 struct v3d_resource *rsc = v3d_resource(job->zsbuf->texture);
613 if (rsc->separate_stencil)
614 rsc = rsc->separate_stencil;
615
616 v3d_job_add_bo(job, rsc->bo);
617
618 job->load |= PIPE_CLEAR_STENCIL & ~job->clear;
619 if (v3d->zsa->base.stencil[0].writemask ||
620 v3d->zsa->base.stencil[1].writemask) {
621 job->store |= PIPE_CLEAR_STENCIL;
622 }
623 rsc->initialized_buffers |= PIPE_CLEAR_STENCIL;
624 }
625
626 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
627 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
628 int blend_rt = v3d->blend->base.independent_blend_enable ? i : 0;
629
630 if (job->store & bit || !job->cbufs[i])
631 continue;
632 struct v3d_resource *rsc = v3d_resource(job->cbufs[i]->texture);
633
634 job->load |= bit & ~job->clear;
635 if (v3d->blend->base.rt[blend_rt].colormask)
636 job->store |= bit;
637 v3d_job_add_bo(job, rsc->bo);
638 }
639
640 if (job->referenced_size > 768 * 1024 * 1024) {
641 perf_debug("Flushing job with %dkb to try to free up memory\n",
642 job->referenced_size / 1024);
643 v3d_flush(pctx);
644 }
645
646 if (V3D_DEBUG & V3D_DEBUG_ALWAYS_FLUSH)
647 v3d_flush(pctx);
648 }
649
650 /**
651 * Implements gallium's clear() hook (glClear()) by drawing a pair of triangles.
652 */
653 static void
654 v3d_draw_clear(struct v3d_context *v3d,
655 unsigned buffers,
656 const union pipe_color_union *color,
657 double depth, unsigned stencil)
658 {
659 static const union pipe_color_union dummy_color = {};
660
661 /* The blitter util dereferences the color regardless, even though the
662 * gallium clear API may not pass one in when only Z/S are cleared.
663 */
664 if (!color)
665 color = &dummy_color;
666
667 v3d_blitter_save(v3d);
668 util_blitter_clear(v3d->blitter,
669 v3d->framebuffer.width,
670 v3d->framebuffer.height,
671 util_framebuffer_get_num_layers(&v3d->framebuffer),
672 buffers, color, depth, stencil);
673 }
674
675 /**
676 * Attempts to perform the GL clear by using the TLB's fast clear at the start
677 * of the frame.
678 */
679 static unsigned
680 v3d_tlb_clear(struct v3d_job *job, unsigned buffers,
681 const union pipe_color_union *color,
682 double depth, unsigned stencil)
683 {
684 struct v3d_context *v3d = job->v3d;
685
686 if (job->draw_calls_queued) {
687 /* If anything in the CL has drawn using the buffer, then the
688 * TLB clear we're trying to add now would happen before that
689 * drawing.
690 */
691 buffers &= ~(job->load | job->store);
692 }
693
694 /* GFXH-1461: If we were to emit a load of just depth or just stencil,
695 * then the clear for the other may get lost. We need to decide now
696 * if it would be possible to need to emit a load of just one after
697 * we've set up our TLB clears.
698 */
699 if (buffers & PIPE_CLEAR_DEPTHSTENCIL &&
700 (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL &&
701 job->zsbuf &&
702 util_format_is_depth_and_stencil(job->zsbuf->texture->format)) {
703 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
704 }
705
706 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
707 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
708 if (!(buffers & bit))
709 continue;
710
711 struct pipe_surface *psurf = v3d->framebuffer.cbufs[i];
712 struct v3d_surface *surf = v3d_surface(psurf);
713 struct v3d_resource *rsc = v3d_resource(psurf->texture);
714
715 union util_color uc;
716 uint32_t internal_size = 4 << surf->internal_bpp;
717
718 static union pipe_color_union swapped_color;
719 if (v3d->swap_color_rb & (1 << i)) {
720 swapped_color.f[0] = color->f[2];
721 swapped_color.f[1] = color->f[1];
722 swapped_color.f[2] = color->f[0];
723 swapped_color.f[3] = color->f[3];
724 color = &swapped_color;
725 }
726
727 switch (surf->internal_type) {
728 case V3D_INTERNAL_TYPE_8:
729 util_pack_color(color->f, PIPE_FORMAT_R8G8B8A8_UNORM,
730 &uc);
731 memcpy(job->clear_color[i], uc.ui, internal_size);
732 break;
733 case V3D_INTERNAL_TYPE_8I:
734 case V3D_INTERNAL_TYPE_8UI:
735 job->clear_color[i][0] = ((color->ui[0] & 0xff) |
736 (color->ui[1] & 0xff) << 8 |
737 (color->ui[2] & 0xff) << 16 |
738 (color->ui[3] & 0xff) << 24);
739 break;
740 case V3D_INTERNAL_TYPE_16F:
741 util_pack_color(color->f, PIPE_FORMAT_R16G16B16A16_FLOAT,
742 &uc);
743 memcpy(job->clear_color[i], uc.ui, internal_size);
744 break;
745 case V3D_INTERNAL_TYPE_16I:
746 case V3D_INTERNAL_TYPE_16UI:
747 job->clear_color[i][0] = ((color->ui[0] & 0xffff) |
748 color->ui[1] << 16);
749 job->clear_color[i][1] = ((color->ui[2] & 0xffff) |
750 color->ui[3] << 16);
751 break;
752 case V3D_INTERNAL_TYPE_32F:
753 case V3D_INTERNAL_TYPE_32I:
754 case V3D_INTERNAL_TYPE_32UI:
755 memcpy(job->clear_color[i], color->ui, internal_size);
756 break;
757 }
758
759 rsc->initialized_buffers |= bit;
760 }
761
762 unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
763 if (zsclear) {
764 struct v3d_resource *rsc =
765 v3d_resource(v3d->framebuffer.zsbuf->texture);
766
767 if (zsclear & PIPE_CLEAR_DEPTH)
768 job->clear_z = depth;
769 if (zsclear & PIPE_CLEAR_STENCIL)
770 job->clear_s = stencil;
771
772 rsc->initialized_buffers |= zsclear;
773 }
774
775 job->draw_min_x = 0;
776 job->draw_min_y = 0;
777 job->draw_max_x = v3d->framebuffer.width;
778 job->draw_max_y = v3d->framebuffer.height;
779 job->clear |= buffers;
780 job->store |= buffers;
781
782 v3d_start_draw(v3d);
783
784 return buffers;
785 }
786
787 static void
788 v3d_clear(struct pipe_context *pctx, unsigned buffers,
789 const union pipe_color_union *color, double depth, unsigned stencil)
790 {
791 struct v3d_context *v3d = v3d_context(pctx);
792 struct v3d_job *job = v3d_get_job_for_fbo(v3d);
793
794 buffers &= ~v3d_tlb_clear(job, buffers, color, depth, stencil);
795
796 if (buffers)
797 v3d_draw_clear(v3d, buffers, color, depth, stencil);
798 }
799
800 static void
801 v3d_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
802 const union pipe_color_union *color,
803 unsigned x, unsigned y, unsigned w, unsigned h,
804 bool render_condition_enabled)
805 {
806 fprintf(stderr, "unimpl: clear RT\n");
807 }
808
809 static void
810 v3d_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
811 unsigned buffers, double depth, unsigned stencil,
812 unsigned x, unsigned y, unsigned w, unsigned h,
813 bool render_condition_enabled)
814 {
815 fprintf(stderr, "unimpl: clear DS\n");
816 }
817
818 void
819 v3dX(draw_init)(struct pipe_context *pctx)
820 {
821 pctx->draw_vbo = v3d_draw_vbo;
822 pctx->clear = v3d_clear;
823 pctx->clear_render_target = v3d_clear_render_target;
824 pctx->clear_depth_stencil = v3d_clear_depth_stencil;
825 }