broadcom/vc5: Move default attribute value setup to the CSO and fix them.
[mesa.git] / src / gallium / drivers / vc5 / vc5_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 "vc5_context.h"
33 #include "vc5_resource.h"
34 #include "vc5_cl.h"
35 #include "broadcom/cle/v3d_packet_v33_pack.h"
36 #include "broadcom/compiler/v3d_compiler.h"
37
38 /**
39 * Does the initial bining command list setup for drawing to a given FBO.
40 */
41 static void
42 vc5_start_draw(struct vc5_context *vc5)
43 {
44 struct vc5_job *job = vc5->job;
45
46 if (job->needs_flush)
47 return;
48
49 /* Get space to emit our BCL state, using a branch to jump to a new BO
50 * if necessary.
51 */
52 vc5_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
53
54 job->submit.bcl_start = job->bcl.bo->offset;
55 vc5_job_add_bo(job, job->bcl.bo);
56
57 job->tile_alloc = vc5_bo_alloc(vc5->screen, 1024 * 1024, "tile alloc");
58 struct vc5_bo *tsda = vc5_bo_alloc(vc5->screen,
59 job->draw_tiles_y *
60 job->draw_tiles_x *
61 64,
62 "TSDA");
63
64 /* "Binning mode lists start with a Tile Binning Mode Configuration
65 * item (120)"
66 *
67 * Part1 signals the end of binning config setup.
68 */
69 cl_emit(&job->bcl, TILE_BINNING_MODE_CONFIGURATION_PART2, config) {
70 config.tile_allocation_memory_address =
71 cl_address(job->tile_alloc, 0);
72 config.tile_allocation_memory_size = job->tile_alloc->size;
73 }
74
75 cl_emit(&job->bcl, TILE_BINNING_MODE_CONFIGURATION_PART1, config) {
76 config.tile_state_data_array_base_address =
77 cl_address(tsda, 0);
78
79 config.width_in_tiles = job->draw_tiles_x;
80 config.height_in_tiles = job->draw_tiles_y;
81
82 /* Must be >= 1 */
83 config.number_of_render_targets =
84 MAX2(vc5->framebuffer.nr_cbufs, 1);
85
86 config.multisample_mode_4x = job->msaa;
87
88 config.maximum_bpp_of_all_render_targets = job->internal_bpp;
89 }
90
91 vc5_bo_unreference(&tsda);
92
93 /* There's definitely nothing in the VCD cache we want. */
94 cl_emit(&job->bcl, FLUSH_VCD_CACHE, bin);
95
96 /* "Binning mode lists must have a Start Tile Binning item (6) after
97 * any prefix state data before the binning list proper starts."
98 */
99 cl_emit(&job->bcl, START_TILE_BINNING, bin);
100
101 cl_emit(&job->bcl, PRIMITIVE_LIST_FORMAT, fmt) {
102 fmt.data_type = LIST_INDEXED;
103 fmt.primitive_type = LIST_TRIANGLES;
104 }
105
106 job->needs_flush = true;
107 job->draw_width = vc5->framebuffer.width;
108 job->draw_height = vc5->framebuffer.height;
109 }
110
111 static void
112 vc5_predraw_check_textures(struct pipe_context *pctx,
113 struct vc5_texture_stateobj *stage_tex)
114 {
115 struct vc5_context *vc5 = vc5_context(pctx);
116
117 for (int i = 0; i < stage_tex->num_textures; i++) {
118 struct pipe_sampler_view *view = stage_tex->textures[i];
119 if (!view)
120 continue;
121
122 vc5_flush_jobs_writing_resource(vc5, view->texture);
123 }
124 }
125
126 static void
127 vc5_emit_gl_shader_state(struct vc5_context *vc5,
128 const struct pipe_draw_info *info)
129 {
130 struct vc5_job *job = vc5->job;
131 /* VC5_DIRTY_VTXSTATE */
132 struct vc5_vertex_stateobj *vtx = vc5->vtx;
133 /* VC5_DIRTY_VTXBUF */
134 struct vc5_vertexbuf_stateobj *vertexbuf = &vc5->vertexbuf;
135
136 /* Upload the uniforms to the indirect CL first */
137 struct vc5_cl_reloc fs_uniforms =
138 vc5_write_uniforms(vc5, vc5->prog.fs,
139 &vc5->constbuf[PIPE_SHADER_FRAGMENT],
140 &vc5->fragtex);
141 struct vc5_cl_reloc vs_uniforms =
142 vc5_write_uniforms(vc5, vc5->prog.vs,
143 &vc5->constbuf[PIPE_SHADER_VERTEX],
144 &vc5->verttex);
145 struct vc5_cl_reloc cs_uniforms =
146 vc5_write_uniforms(vc5, vc5->prog.cs,
147 &vc5->constbuf[PIPE_SHADER_VERTEX],
148 &vc5->verttex);
149
150 uint32_t shader_rec_offset =
151 vc5_cl_ensure_space(&job->indirect,
152 cl_packet_length(GL_SHADER_STATE_RECORD) +
153 vtx->num_elements *
154 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD),
155 32);
156
157 cl_emit(&job->indirect, GL_SHADER_STATE_RECORD, shader) {
158 shader.enable_clipping = true;
159 /* VC5_DIRTY_PRIM_MODE | VC5_DIRTY_RASTERIZER */
160 shader.point_size_in_shaded_vertex_data =
161 (info->mode == PIPE_PRIM_POINTS &&
162 vc5->rasterizer->base.point_size_per_vertex);
163
164 /* Must be set if the shader modifies Z, discards, or modifies
165 * the sample mask. For any of these cases, the fragment
166 * shader needs to write the Z value (even just discards).
167 */
168 shader.fragment_shader_does_z_writes =
169 (vc5->prog.fs->prog_data.fs->writes_z ||
170 vc5->prog.fs->prog_data.fs->discard);
171
172 shader.number_of_varyings_in_fragment_shader =
173 vc5->prog.fs->prog_data.base->num_inputs;
174
175 shader.propagate_nans = true;
176
177 shader.coordinate_shader_code_address =
178 cl_address(vc5->prog.cs->bo, 0);
179 shader.vertex_shader_code_address =
180 cl_address(vc5->prog.vs->bo, 0);
181 shader.fragment_shader_code_address =
182 cl_address(vc5->prog.fs->bo, 0);
183
184 /* XXX: Use combined input/output size flag in the common
185 * case.
186 */
187 shader.coordinate_shader_has_separate_input_and_output_vpm_blocks = true;
188 shader.vertex_shader_has_separate_input_and_output_vpm_blocks = true;
189 shader.coordinate_shader_input_vpm_segment_size =
190 vc5->prog.cs->prog_data.vs->vpm_input_size;
191 shader.vertex_shader_input_vpm_segment_size =
192 vc5->prog.vs->prog_data.vs->vpm_input_size;
193
194 shader.coordinate_shader_output_vpm_segment_size =
195 vc5->prog.cs->prog_data.vs->vpm_output_size;
196 shader.vertex_shader_output_vpm_segment_size =
197 vc5->prog.vs->prog_data.vs->vpm_output_size;
198
199 shader.coordinate_shader_uniforms_address = cs_uniforms;
200 shader.vertex_shader_uniforms_address = vs_uniforms;
201 shader.fragment_shader_uniforms_address = fs_uniforms;
202
203 shader.vertex_id_read_by_coordinate_shader =
204 vc5->prog.cs->prog_data.vs->uses_vid;
205 shader.instance_id_read_by_coordinate_shader =
206 vc5->prog.cs->prog_data.vs->uses_iid;
207 shader.vertex_id_read_by_vertex_shader =
208 vc5->prog.vs->prog_data.vs->uses_vid;
209 shader.instance_id_read_by_vertex_shader =
210 vc5->prog.vs->prog_data.vs->uses_iid;
211
212 shader.address_of_default_attribute_values =
213 cl_address(vtx->default_attribute_values, 0);
214 }
215
216 for (int i = 0; i < vtx->num_elements; i++) {
217 struct pipe_vertex_element *elem = &vtx->pipe[i];
218 struct pipe_vertex_buffer *vb =
219 &vertexbuf->vb[elem->vertex_buffer_index];
220 struct vc5_resource *rsc = vc5_resource(vb->buffer.resource);
221
222 struct V3D33_GL_SHADER_STATE_ATTRIBUTE_RECORD attr_unpacked = {
223 .stride = vb->stride,
224 .address = cl_address(rsc->bo,
225 vb->buffer_offset +
226 elem->src_offset),
227 .number_of_values_read_by_coordinate_shader =
228 vc5->prog.cs->prog_data.vs->vattr_sizes[i],
229 .number_of_values_read_by_vertex_shader =
230 vc5->prog.vs->prog_data.vs->vattr_sizes[i],
231 };
232 const uint32_t size =
233 cl_packet_length(GL_SHADER_STATE_ATTRIBUTE_RECORD);
234 uint8_t attr_packed[size];
235 V3D33_GL_SHADER_STATE_ATTRIBUTE_RECORD_pack(&job->indirect,
236 attr_packed,
237 &attr_unpacked);
238 for (int j = 0; j < size; j++)
239 attr_packed[j] |= vtx->attrs[i * size + j];
240 cl_emit_prepacked(&job->indirect, &attr_packed);
241 }
242
243 cl_emit(&job->bcl, GL_SHADER_STATE, state) {
244 state.address = cl_address(job->indirect.bo, shader_rec_offset);
245 state.number_of_attribute_arrays = vtx->num_elements;
246 }
247
248 vc5_bo_unreference(&cs_uniforms.bo);
249 vc5_bo_unreference(&vs_uniforms.bo);
250 vc5_bo_unreference(&fs_uniforms.bo);
251
252 job->shader_rec_count++;
253 }
254
255 static void
256 vc5_draw_vbo(struct pipe_context *pctx, const struct pipe_draw_info *info)
257 {
258 struct vc5_context *vc5 = vc5_context(pctx);
259
260 if (!info->count_from_stream_output && !info->indirect &&
261 !info->primitive_restart &&
262 !u_trim_pipe_prim(info->mode, (unsigned*)&info->count))
263 return;
264
265 /* Fall back for weird desktop GL primitive restart values. */
266 if (info->primitive_restart &&
267 info->index_size) {
268 uint32_t mask = ~0;
269
270 switch (info->index_size) {
271 case 2:
272 mask = 0xffff;
273 break;
274 case 1:
275 mask = 0xff;
276 break;
277 }
278
279 if (info->restart_index != mask) {
280 util_draw_vbo_without_prim_restart(pctx, info);
281 return;
282 }
283 }
284
285 if (info->mode >= PIPE_PRIM_QUADS) {
286 util_primconvert_save_rasterizer_state(vc5->primconvert, &vc5->rasterizer->base);
287 util_primconvert_draw_vbo(vc5->primconvert, info);
288 perf_debug("Fallback conversion for %d %s vertices\n",
289 info->count, u_prim_name(info->mode));
290 return;
291 }
292
293 /* Before setting up the draw, flush anything writing to the textures
294 * that we read from.
295 */
296 vc5_predraw_check_textures(pctx, &vc5->verttex);
297 vc5_predraw_check_textures(pctx, &vc5->fragtex);
298
299 struct vc5_job *job = vc5_get_job_for_fbo(vc5);
300
301 /* Get space to emit our draw call into the BCL, using a branch to
302 * jump to a new BO if necessary.
303 */
304 vc5_cl_ensure_space_with_branch(&job->bcl, 256 /* XXX */);
305
306 if (vc5->prim_mode != info->mode) {
307 vc5->prim_mode = info->mode;
308 vc5->dirty |= VC5_DIRTY_PRIM_MODE;
309 }
310
311 vc5_start_draw(vc5);
312 vc5_update_compiled_shaders(vc5, info->mode);
313
314 vc5_emit_state(pctx);
315
316 if (vc5->dirty & (VC5_DIRTY_VTXBUF |
317 VC5_DIRTY_VTXSTATE |
318 VC5_DIRTY_PRIM_MODE |
319 VC5_DIRTY_RASTERIZER |
320 VC5_DIRTY_COMPILED_CS |
321 VC5_DIRTY_COMPILED_VS |
322 VC5_DIRTY_COMPILED_FS |
323 vc5->prog.cs->uniform_dirty_bits |
324 vc5->prog.vs->uniform_dirty_bits |
325 vc5->prog.fs->uniform_dirty_bits)) {
326 vc5_emit_gl_shader_state(vc5, info);
327 }
328
329 vc5->dirty = 0;
330
331 /* The Base Vertex/Base Instance packet sets those values to nonzero
332 * for the next draw call only.
333 */
334 if (info->index_bias || info->start_instance) {
335 cl_emit(&job->bcl, BASE_VERTEX_BASE_INSTANCE, base) {
336 base.base_instance = info->start_instance;
337 base.base_vertex = info->index_bias;
338 }
339 }
340
341 /* The HW only processes transform feedback on primitives with the
342 * flag set.
343 */
344 uint32_t prim_tf_enable = 0;
345 if (vc5->prog.bind_vs->num_tf_outputs)
346 prim_tf_enable = (V3D_PRIM_POINTS_TF - V3D_PRIM_POINTS);
347
348 /* Note that the primitive type fields match with OpenGL/gallium
349 * definitions, up to but not including QUADS.
350 */
351 if (info->index_size) {
352 uint32_t index_size = info->index_size;
353 uint32_t offset = info->start * index_size;
354 struct pipe_resource *prsc;
355 if (info->has_user_indices) {
356 prsc = NULL;
357 u_upload_data(vc5->uploader, 0,
358 info->count * info->index_size, 4,
359 info->index.user,
360 &offset, &prsc);
361 } else {
362 prsc = info->index.resource;
363 }
364 struct vc5_resource *rsc = vc5_resource(prsc);
365
366 if (info->instance_count > 1) {
367 cl_emit(&job->bcl, INDEXED_INSTANCED_PRIMITIVE_LIST, prim) {
368 prim.index_type = ffs(info->index_size) - 1;
369 prim.maximum_index = (1u << 31) - 1; /* XXX */
370 prim.address_of_indices_list =
371 cl_address(rsc->bo, offset);
372 prim.mode = info->mode | prim_tf_enable;
373 prim.enable_primitive_restarts = info->primitive_restart;
374
375 prim.number_of_instances = info->instance_count;
376 prim.instance_length = info->count;
377 }
378 } else {
379 cl_emit(&job->bcl, INDEXED_PRIMITIVE_LIST, prim) {
380 prim.index_type = ffs(info->index_size) - 1;
381 prim.length = info->count;
382 prim.maximum_index = (1u << 31) - 1; /* XXX */
383 prim.address_of_indices_list =
384 cl_address(rsc->bo, offset);
385 prim.mode = info->mode | prim_tf_enable;
386 prim.enable_primitive_restarts = info->primitive_restart;
387 }
388 }
389
390 job->draw_calls_queued++;
391
392 if (info->has_user_indices)
393 pipe_resource_reference(&prsc, NULL);
394 } else {
395 if (info->instance_count > 1) {
396 cl_emit(&job->bcl, VERTEX_ARRAY_INSTANCED_PRIMITIVES, prim) {
397 prim.mode = info->mode | prim_tf_enable;
398 prim.index_of_first_vertex = info->start;
399 prim.number_of_instances = info->instance_count;
400 prim.instance_length = info->count;
401 }
402 } else {
403 cl_emit(&job->bcl, VERTEX_ARRAY_PRIMITIVES, prim) {
404 prim.mode = info->mode | prim_tf_enable;
405 prim.length = info->count;
406 prim.index_of_first_vertex = info->start;
407 }
408 }
409 }
410 job->draw_calls_queued++;
411
412 if (vc5->zsa && job->zsbuf &&
413 (vc5->zsa->base.depth.enabled ||
414 vc5->zsa->base.stencil[0].enabled)) {
415 struct vc5_resource *rsc = vc5_resource(job->zsbuf->texture);
416 vc5_job_add_bo(job, rsc->bo);
417
418 if (vc5->zsa->base.depth.enabled) {
419 job->resolve |= PIPE_CLEAR_DEPTH;
420 rsc->initialized_buffers = PIPE_CLEAR_DEPTH;
421
422 if (vc5->zsa->early_z_enable)
423 job->uses_early_z = true;
424 }
425
426 if (vc5->zsa->base.stencil[0].enabled) {
427 job->resolve |= PIPE_CLEAR_STENCIL;
428 rsc->initialized_buffers |= PIPE_CLEAR_STENCIL;
429 }
430 }
431
432 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
433 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
434
435 if (job->resolve & bit || !job->cbufs[i])
436 continue;
437 struct vc5_resource *rsc = vc5_resource(job->cbufs[i]->texture);
438
439 job->resolve |= bit;
440 vc5_job_add_bo(job, rsc->bo);
441 }
442
443 if (V3D_DEBUG & V3D_DEBUG_ALWAYS_FLUSH)
444 vc5_flush(pctx);
445 }
446
447 static void
448 vc5_clear(struct pipe_context *pctx, unsigned buffers,
449 const union pipe_color_union *color, double depth, unsigned stencil)
450 {
451 struct vc5_context *vc5 = vc5_context(pctx);
452 struct vc5_job *job = vc5_get_job_for_fbo(vc5);
453
454 /* We can't flag new buffers for clearing once we've queued draws. We
455 * could avoid this by using the 3d engine to clear.
456 */
457 if (job->draw_calls_queued) {
458 perf_debug("Flushing rendering to process new clear.\n");
459 vc5_job_submit(vc5, job);
460 job = vc5_get_job_for_fbo(vc5);
461 }
462
463 for (int i = 0; i < VC5_MAX_DRAW_BUFFERS; i++) {
464 uint32_t bit = PIPE_CLEAR_COLOR0 << i;
465 if (!(buffers & bit))
466 continue;
467
468 struct pipe_surface *cbuf = vc5->framebuffer.cbufs[i];
469 struct vc5_resource *rsc =
470 vc5_resource(cbuf->texture);
471
472 union util_color uc;
473 util_pack_color(color->f, cbuf->format, &uc);
474
475 memcpy(job->clear_color[i], uc.ui,
476 util_format_get_blocksize(cbuf->format));
477
478 rsc->initialized_buffers |= bit;
479 }
480
481 unsigned zsclear = buffers & PIPE_CLEAR_DEPTHSTENCIL;
482 if (zsclear) {
483 struct vc5_resource *rsc =
484 vc5_resource(vc5->framebuffer.zsbuf->texture);
485
486 if (zsclear & PIPE_CLEAR_DEPTH)
487 job->clear_z = depth;
488 if (zsclear & PIPE_CLEAR_STENCIL)
489 job->clear_s = stencil;
490
491 rsc->initialized_buffers |= zsclear;
492 }
493
494 job->draw_min_x = 0;
495 job->draw_min_y = 0;
496 job->draw_max_x = vc5->framebuffer.width;
497 job->draw_max_y = vc5->framebuffer.height;
498 job->cleared |= buffers;
499 job->resolve |= buffers;
500
501 vc5_start_draw(vc5);
502 }
503
504 static void
505 vc5_clear_render_target(struct pipe_context *pctx, struct pipe_surface *ps,
506 const union pipe_color_union *color,
507 unsigned x, unsigned y, unsigned w, unsigned h,
508 bool render_condition_enabled)
509 {
510 fprintf(stderr, "unimpl: clear RT\n");
511 }
512
513 static void
514 vc5_clear_depth_stencil(struct pipe_context *pctx, struct pipe_surface *ps,
515 unsigned buffers, double depth, unsigned stencil,
516 unsigned x, unsigned y, unsigned w, unsigned h,
517 bool render_condition_enabled)
518 {
519 fprintf(stderr, "unimpl: clear DS\n");
520 }
521
522 void
523 vc5_draw_init(struct pipe_context *pctx)
524 {
525 pctx->draw_vbo = vc5_draw_vbo;
526 pctx->clear = vc5_clear;
527 pctx->clear_render_target = vc5_clear_render_target;
528 pctx->clear_depth_stencil = vc5_clear_depth_stencil;
529 }