vc4: Add support for enabling early Z discards.
[mesa.git] / src / gallium / drivers / vc4 / vc4_state.c
1 /*
2 * Copyright © 2014 Broadcom
3 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
22 * IN THE SOFTWARE.
23 */
24
25 #include "pipe/p_state.h"
26 #include "util/u_inlines.h"
27 #include "util/u_math.h"
28 #include "util/u_memory.h"
29 #include "util/u_helpers.h"
30
31 #include "vc4_context.h"
32
33 static void *
34 vc4_generic_cso_state_create(const void *src, uint32_t size)
35 {
36 void *dst = calloc(1, size);
37 if (!dst)
38 return NULL;
39 memcpy(dst, src, size);
40 return dst;
41 }
42
43 static void
44 vc4_generic_cso_state_delete(struct pipe_context *pctx, void *hwcso)
45 {
46 free(hwcso);
47 }
48
49 static void
50 vc4_set_blend_color(struct pipe_context *pctx,
51 const struct pipe_blend_color *blend_color)
52 {
53 struct vc4_context *vc4 = vc4_context(pctx);
54 vc4->blend_color = *blend_color;
55 vc4->dirty |= VC4_DIRTY_BLEND_COLOR;
56 }
57
58 static void
59 vc4_set_stencil_ref(struct pipe_context *pctx,
60 const struct pipe_stencil_ref *stencil_ref)
61 {
62 struct vc4_context *vc4 = vc4_context(pctx);
63 vc4->stencil_ref =* stencil_ref;
64 vc4->dirty |= VC4_DIRTY_STENCIL_REF;
65 }
66
67 static void
68 vc4_set_clip_state(struct pipe_context *pctx,
69 const struct pipe_clip_state *clip)
70 {
71 struct vc4_context *vc4 = vc4_context(pctx);
72 vc4->clip = *clip;
73 vc4->dirty |= VC4_DIRTY_CLIP;
74 }
75
76 static void
77 vc4_set_sample_mask(struct pipe_context *pctx, unsigned sample_mask)
78 {
79 struct vc4_context *vc4 = vc4_context(pctx);
80 vc4->sample_mask = (uint16_t)sample_mask;
81 vc4->dirty |= VC4_DIRTY_SAMPLE_MASK;
82 }
83
84 static uint16_t
85 float_to_187_half(float f)
86 {
87 return fui(f) >> 16;
88 }
89
90 static void *
91 vc4_create_rasterizer_state(struct pipe_context *pctx,
92 const struct pipe_rasterizer_state *cso)
93 {
94 struct vc4_rasterizer_state *so;
95
96 so = CALLOC_STRUCT(vc4_rasterizer_state);
97 if (!so)
98 return NULL;
99
100 so->base = *cso;
101
102 if (!(cso->cull_face & PIPE_FACE_FRONT))
103 so->config_bits[0] |= VC4_CONFIG_BITS_ENABLE_PRIM_FRONT;
104 if (!(cso->cull_face & PIPE_FACE_BACK))
105 so->config_bits[0] |= VC4_CONFIG_BITS_ENABLE_PRIM_BACK;
106
107 /* Workaround: HW-2726 PTB does not handle zero-size points (BCM2835,
108 * BCM21553).
109 */
110 so->point_size = MAX2(cso->point_size, .125);
111
112 if (cso->front_ccw)
113 so->config_bits[0] |= VC4_CONFIG_BITS_CW_PRIMITIVES;
114
115 if (cso->offset_tri) {
116 so->config_bits[0] |= VC4_CONFIG_BITS_ENABLE_DEPTH_OFFSET;
117
118 so->offset_units = float_to_187_half(cso->offset_units);
119 so->offset_factor = float_to_187_half(cso->offset_scale);
120 }
121
122 return so;
123 }
124
125 /* Blend state is baked into shaders. */
126 static void *
127 vc4_create_blend_state(struct pipe_context *pctx,
128 const struct pipe_blend_state *cso)
129 {
130 return vc4_generic_cso_state_create(cso, sizeof(*cso));
131 }
132
133 /**
134 * The TLB_STENCIL_SETUP data has a little bitfield for common writemask
135 * values, so you don't have to do a separate writemask setup.
136 */
137 static uint8_t
138 tlb_stencil_setup_writemask(uint8_t mask)
139 {
140 switch (mask) {
141 case 0x1: return 0;
142 case 0x3: return 1;
143 case 0xf: return 2;
144 case 0xff: return 3;
145 default: return 0xff;
146 }
147 }
148
149 static uint32_t
150 tlb_stencil_setup_bits(const struct pipe_stencil_state *state,
151 uint8_t writemask_bits)
152 {
153 static const uint8_t op_map[] = {
154 [PIPE_STENCIL_OP_ZERO] = 0,
155 [PIPE_STENCIL_OP_KEEP] = 1,
156 [PIPE_STENCIL_OP_REPLACE] = 2,
157 [PIPE_STENCIL_OP_INCR] = 3,
158 [PIPE_STENCIL_OP_DECR] = 4,
159 [PIPE_STENCIL_OP_INVERT] = 5,
160 [PIPE_STENCIL_OP_INCR_WRAP] = 6,
161 [PIPE_STENCIL_OP_DECR_WRAP] = 7,
162 };
163 uint32_t bits = 0;
164
165 if (writemask_bits != 0xff)
166 bits |= writemask_bits << 28;
167 bits |= op_map[state->zfail_op] << 25;
168 bits |= op_map[state->zpass_op] << 22;
169 bits |= op_map[state->fail_op] << 19;
170 bits |= state->func << 16;
171 /* Ref is filled in at uniform upload time */
172 bits |= state->valuemask << 0;
173
174 return bits;
175 }
176
177 static void *
178 vc4_create_depth_stencil_alpha_state(struct pipe_context *pctx,
179 const struct pipe_depth_stencil_alpha_state *cso)
180 {
181 struct vc4_depth_stencil_alpha_state *so;
182
183 so = CALLOC_STRUCT(vc4_depth_stencil_alpha_state);
184 if (!so)
185 return NULL;
186
187 so->base = *cso;
188
189 /* We always keep the early Z state correct, since a later state using
190 * early Z may want it.
191 */
192 so->config_bits[2] |= VC4_CONFIG_BITS_EARLY_Z_UPDATE;
193
194 if (cso->depth.enabled) {
195 if (cso->depth.writemask) {
196 so->config_bits[1] |= VC4_CONFIG_BITS_Z_UPDATE;
197 }
198 so->config_bits[1] |= (cso->depth.func <<
199 VC4_CONFIG_BITS_DEPTH_FUNC_SHIFT);
200
201 /* We only handle early Z in the < direction because otherwise
202 * we'd have to runtime guess which direction to set in the
203 * render config.
204 */
205 if ((cso->depth.func == PIPE_FUNC_LESS ||
206 cso->depth.func == PIPE_FUNC_LEQUAL) &&
207 (!cso->stencil[0].enabled ||
208 (cso->stencil[0].zfail_op == PIPE_STENCIL_OP_KEEP &&
209 (!cso->stencil[1].enabled ||
210 cso->stencil[1].zfail_op == PIPE_STENCIL_OP_KEEP)))) {
211 so->config_bits[2] |= VC4_CONFIG_BITS_EARLY_Z;
212 }
213 } else {
214 so->config_bits[1] |= (PIPE_FUNC_ALWAYS <<
215 VC4_CONFIG_BITS_DEPTH_FUNC_SHIFT);
216 }
217
218 if (cso->stencil[0].enabled) {
219 const struct pipe_stencil_state *front = &cso->stencil[0];
220 const struct pipe_stencil_state *back = &cso->stencil[1];
221
222 uint8_t front_writemask_bits =
223 tlb_stencil_setup_writemask(front->writemask);
224 uint8_t back_writemask = front->writemask;
225 uint8_t back_writemask_bits = front_writemask_bits;
226
227 so->stencil_uniforms[0] =
228 tlb_stencil_setup_bits(front, front_writemask_bits);
229 if (back->enabled) {
230 back_writemask = back->writemask;
231 back_writemask_bits =
232 tlb_stencil_setup_writemask(back->writemask);
233
234 so->stencil_uniforms[0] |= (1 << 30);
235 so->stencil_uniforms[1] =
236 tlb_stencil_setup_bits(back, back_writemask_bits);
237 so->stencil_uniforms[1] |= (2 << 30);
238 } else {
239 so->stencil_uniforms[0] |= (3 << 30);
240 }
241
242 if (front_writemask_bits == 0xff ||
243 back_writemask_bits == 0xff) {
244 so->stencil_uniforms[2] = (front->writemask |
245 (back_writemask << 8));
246 }
247 }
248
249 return so;
250 }
251
252 static void
253 vc4_set_polygon_stipple(struct pipe_context *pctx,
254 const struct pipe_poly_stipple *stipple)
255 {
256 struct vc4_context *vc4 = vc4_context(pctx);
257 vc4->stipple = *stipple;
258 vc4->dirty |= VC4_DIRTY_STIPPLE;
259 }
260
261 static void
262 vc4_set_scissor_states(struct pipe_context *pctx,
263 unsigned start_slot,
264 unsigned num_scissors,
265 const struct pipe_scissor_state *scissor)
266 {
267 struct vc4_context *vc4 = vc4_context(pctx);
268
269 vc4->scissor = *scissor;
270 vc4->dirty |= VC4_DIRTY_SCISSOR;
271 }
272
273 static void
274 vc4_set_viewport_states(struct pipe_context *pctx,
275 unsigned start_slot,
276 unsigned num_viewports,
277 const struct pipe_viewport_state *viewport)
278 {
279 struct vc4_context *vc4 = vc4_context(pctx);
280 vc4->viewport = *viewport;
281 vc4->dirty |= VC4_DIRTY_VIEWPORT;
282 }
283
284 static void
285 vc4_set_vertex_buffers(struct pipe_context *pctx,
286 unsigned start_slot, unsigned count,
287 const struct pipe_vertex_buffer *vb)
288 {
289 struct vc4_context *vc4 = vc4_context(pctx);
290 struct vc4_vertexbuf_stateobj *so = &vc4->vertexbuf;
291
292 util_set_vertex_buffers_mask(so->vb, &so->enabled_mask, vb,
293 start_slot, count);
294 so->count = util_last_bit(so->enabled_mask);
295
296 vc4->dirty |= VC4_DIRTY_VTXBUF;
297 }
298
299 static void
300 vc4_set_index_buffer(struct pipe_context *pctx,
301 const struct pipe_index_buffer *ib)
302 {
303 struct vc4_context *vc4 = vc4_context(pctx);
304
305 if (ib) {
306 assert(!ib->user_buffer);
307
308 if (ib->index_size == 4) {
309 struct pipe_resource tmpl = *ib->buffer;
310 assert(tmpl.format == PIPE_FORMAT_R8_UNORM);
311 assert(tmpl.height0 == 1);
312 tmpl.width0 = (tmpl.width0 - ib->offset) / 2;
313 struct pipe_resource *pshadow =
314 vc4_resource_create(&vc4->screen->base, &tmpl);
315 struct vc4_resource *shadow = vc4_resource(pshadow);
316 pipe_resource_reference(&shadow->shadow_parent, ib->buffer);
317
318 pipe_resource_reference(&vc4->indexbuf.buffer, NULL);
319 vc4->indexbuf.buffer = pshadow;
320 vc4->indexbuf.index_size = 2;
321 } else {
322 pipe_resource_reference(&vc4->indexbuf.buffer, ib->buffer);
323 vc4->indexbuf.index_size = ib->index_size;
324 }
325 vc4->indexbuf.offset = ib->offset;
326 } else {
327 pipe_resource_reference(&vc4->indexbuf.buffer, NULL);
328 }
329
330 vc4->dirty |= VC4_DIRTY_INDEXBUF;
331 }
332
333 static void
334 vc4_blend_state_bind(struct pipe_context *pctx, void *hwcso)
335 {
336 struct vc4_context *vc4 = vc4_context(pctx);
337 vc4->blend = hwcso;
338 vc4->dirty |= VC4_DIRTY_BLEND;
339 }
340
341 static void
342 vc4_rasterizer_state_bind(struct pipe_context *pctx, void *hwcso)
343 {
344 struct vc4_context *vc4 = vc4_context(pctx);
345 struct vc4_rasterizer_state *rast = hwcso;
346
347 if (vc4->rasterizer && rast &&
348 vc4->rasterizer->base.flatshade != rast->base.flatshade) {
349 vc4->dirty |= VC4_DIRTY_FLAT_SHADE_FLAGS;
350 }
351
352 vc4->rasterizer = hwcso;
353 vc4->dirty |= VC4_DIRTY_RASTERIZER;
354 }
355
356 static void
357 vc4_zsa_state_bind(struct pipe_context *pctx, void *hwcso)
358 {
359 struct vc4_context *vc4 = vc4_context(pctx);
360 vc4->zsa = hwcso;
361 vc4->dirty |= VC4_DIRTY_ZSA;
362 }
363
364 static void *
365 vc4_vertex_state_create(struct pipe_context *pctx, unsigned num_elements,
366 const struct pipe_vertex_element *elements)
367 {
368 struct vc4_vertex_stateobj *so = CALLOC_STRUCT(vc4_vertex_stateobj);
369
370 if (!so)
371 return NULL;
372
373 memcpy(so->pipe, elements, sizeof(*elements) * num_elements);
374 so->num_elements = num_elements;
375
376 return so;
377 }
378
379 static void
380 vc4_vertex_state_bind(struct pipe_context *pctx, void *hwcso)
381 {
382 struct vc4_context *vc4 = vc4_context(pctx);
383 vc4->vtx = hwcso;
384 vc4->dirty |= VC4_DIRTY_VTXSTATE;
385 }
386
387 static void
388 vc4_set_constant_buffer(struct pipe_context *pctx, uint shader, uint index,
389 struct pipe_constant_buffer *cb)
390 {
391 struct vc4_context *vc4 = vc4_context(pctx);
392 struct vc4_constbuf_stateobj *so = &vc4->constbuf[shader];
393
394 assert(index == 0);
395
396 /* Note that the state tracker can unbind constant buffers by
397 * passing NULL here.
398 */
399 if (unlikely(!cb)) {
400 so->enabled_mask &= ~(1 << index);
401 so->dirty_mask &= ~(1 << index);
402 return;
403 }
404
405 assert(!cb->buffer);
406 so->cb[index].buffer_offset = cb->buffer_offset;
407 so->cb[index].buffer_size = cb->buffer_size;
408 so->cb[index].user_buffer = cb->user_buffer;
409
410 so->enabled_mask |= 1 << index;
411 so->dirty_mask |= 1 << index;
412 vc4->dirty |= VC4_DIRTY_CONSTBUF;
413 }
414
415 static void
416 vc4_set_framebuffer_state(struct pipe_context *pctx,
417 const struct pipe_framebuffer_state *framebuffer)
418 {
419 struct vc4_context *vc4 = vc4_context(pctx);
420 struct pipe_framebuffer_state *cso = &vc4->framebuffer;
421 unsigned i;
422
423 vc4_flush(pctx);
424
425 for (i = 0; i < framebuffer->nr_cbufs; i++)
426 pipe_surface_reference(&cso->cbufs[i], framebuffer->cbufs[i]);
427 for (; i < vc4->framebuffer.nr_cbufs; i++)
428 pipe_surface_reference(&cso->cbufs[i], NULL);
429
430 cso->nr_cbufs = framebuffer->nr_cbufs;
431
432 pipe_surface_reference(&cso->zsbuf, framebuffer->zsbuf);
433
434 cso->width = framebuffer->width;
435 cso->height = framebuffer->height;
436
437 /* Nonzero texture mipmap levels are laid out as if they were in
438 * power-of-two-sized spaces. The renderbuffer config infers its
439 * stride from the width parameter, so we need to configure our
440 * framebuffer. Note that if the z/color buffers were mismatched
441 * sizes, we wouldn't be able to do this.
442 */
443 if (cso->cbufs[0] && cso->cbufs[0]->u.tex.level) {
444 struct vc4_resource *rsc =
445 vc4_resource(cso->cbufs[0]->texture);
446 cso->width =
447 (rsc->slices[cso->cbufs[0]->u.tex.level].stride /
448 rsc->cpp);
449 } else if (cso->zsbuf && cso->zsbuf->u.tex.level){
450 struct vc4_resource *rsc =
451 vc4_resource(cso->zsbuf->texture);
452 cso->width =
453 (rsc->slices[cso->zsbuf->u.tex.level].stride /
454 rsc->cpp);
455 }
456
457 vc4->dirty |= VC4_DIRTY_FRAMEBUFFER;
458 }
459
460 static struct vc4_texture_stateobj *
461 vc4_get_stage_tex(struct vc4_context *vc4, unsigned shader)
462 {
463 vc4->dirty |= VC4_DIRTY_TEXSTATE;
464
465 switch (shader) {
466 case PIPE_SHADER_FRAGMENT:
467 vc4->dirty |= VC4_DIRTY_FRAGTEX;
468 return &vc4->fragtex;
469 break;
470 case PIPE_SHADER_VERTEX:
471 vc4->dirty |= VC4_DIRTY_VERTTEX;
472 return &vc4->verttex;
473 break;
474 default:
475 fprintf(stderr, "Unknown shader target %d\n", shader);
476 abort();
477 }
478 }
479
480 static void *
481 vc4_create_sampler_state(struct pipe_context *pctx,
482 const struct pipe_sampler_state *cso)
483 {
484 return vc4_generic_cso_state_create(cso, sizeof(*cso));
485 }
486
487 static void
488 vc4_sampler_states_bind(struct pipe_context *pctx,
489 unsigned shader, unsigned start,
490 unsigned nr, void **hwcso)
491 {
492 struct vc4_context *vc4 = vc4_context(pctx);
493 struct vc4_texture_stateobj *stage_tex = vc4_get_stage_tex(vc4, shader);
494
495 assert(start == 0);
496 unsigned i;
497 unsigned new_nr = 0;
498
499 for (i = 0; i < nr; i++) {
500 if (hwcso[i])
501 new_nr = i + 1;
502 stage_tex->samplers[i] = hwcso[i];
503 stage_tex->dirty_samplers |= (1 << i);
504 }
505
506 for (; i < stage_tex->num_samplers; i++) {
507 stage_tex->samplers[i] = NULL;
508 stage_tex->dirty_samplers |= (1 << i);
509 }
510
511 stage_tex->num_samplers = new_nr;
512 }
513
514 static struct pipe_sampler_view *
515 vc4_create_sampler_view(struct pipe_context *pctx, struct pipe_resource *prsc,
516 const struct pipe_sampler_view *cso)
517 {
518 struct pipe_sampler_view *so = malloc(sizeof(*so));
519
520 if (!so)
521 return NULL;
522
523 *so = *cso;
524
525 pipe_reference(NULL, &prsc->reference);
526
527 /* There is no hardware level clamping, and the start address of a
528 * texture may be misaligned, so in that case we have to copy to a
529 * temporary.
530 */
531 if (so->u.tex.first_level) {
532 struct vc4_resource *shadow_parent = vc4_resource(prsc);
533 struct pipe_resource tmpl = shadow_parent->base.b;
534 struct vc4_resource *clone;
535
536 tmpl.width0 = u_minify(tmpl.width0, so->u.tex.first_level);
537 tmpl.height0 = u_minify(tmpl.height0, so->u.tex.first_level);
538 tmpl.last_level = so->u.tex.last_level - so->u.tex.first_level;
539
540 prsc = vc4_resource_create(pctx->screen, &tmpl);
541 clone = vc4_resource(prsc);
542 clone->shadow_parent = &shadow_parent->base.b;
543 /* Flag it as needing update of the contents from the parent. */
544 clone->writes = shadow_parent->writes - 1;
545 }
546 so->texture = prsc;
547 so->reference.count = 1;
548 so->context = pctx;
549
550 return so;
551 }
552
553 static void
554 vc4_sampler_view_destroy(struct pipe_context *pctx,
555 struct pipe_sampler_view *view)
556 {
557 pipe_resource_reference(&view->texture, NULL);
558 free(view);
559 }
560
561 static void
562 vc4_set_sampler_views(struct pipe_context *pctx, unsigned shader,
563 unsigned start, unsigned nr,
564 struct pipe_sampler_view **views)
565 {
566 struct vc4_context *vc4 = vc4_context(pctx);
567 struct vc4_texture_stateobj *stage_tex = vc4_get_stage_tex(vc4, shader);
568 unsigned i;
569 unsigned new_nr = 0;
570
571 assert(start == 0);
572
573 vc4->dirty |= VC4_DIRTY_TEXSTATE;
574
575 for (i = 0; i < nr; i++) {
576 if (views[i]) {
577 new_nr = i + 1;
578 if (views[i]->u.tex.first_level != 0)
579 vc4_update_shadow_baselevel_texture(pctx, views[i]);
580 }
581 pipe_sampler_view_reference(&stage_tex->textures[i], views[i]);
582 stage_tex->dirty_samplers |= (1 << i);
583 }
584
585 for (; i < stage_tex->num_textures; i++) {
586 pipe_sampler_view_reference(&stage_tex->textures[i], NULL);
587 stage_tex->dirty_samplers |= (1 << i);
588 }
589
590 stage_tex->num_textures = new_nr;
591 }
592
593 void
594 vc4_state_init(struct pipe_context *pctx)
595 {
596 pctx->set_blend_color = vc4_set_blend_color;
597 pctx->set_stencil_ref = vc4_set_stencil_ref;
598 pctx->set_clip_state = vc4_set_clip_state;
599 pctx->set_sample_mask = vc4_set_sample_mask;
600 pctx->set_constant_buffer = vc4_set_constant_buffer;
601 pctx->set_framebuffer_state = vc4_set_framebuffer_state;
602 pctx->set_polygon_stipple = vc4_set_polygon_stipple;
603 pctx->set_scissor_states = vc4_set_scissor_states;
604 pctx->set_viewport_states = vc4_set_viewport_states;
605
606 pctx->set_vertex_buffers = vc4_set_vertex_buffers;
607 pctx->set_index_buffer = vc4_set_index_buffer;
608
609 pctx->create_blend_state = vc4_create_blend_state;
610 pctx->bind_blend_state = vc4_blend_state_bind;
611 pctx->delete_blend_state = vc4_generic_cso_state_delete;
612
613 pctx->create_rasterizer_state = vc4_create_rasterizer_state;
614 pctx->bind_rasterizer_state = vc4_rasterizer_state_bind;
615 pctx->delete_rasterizer_state = vc4_generic_cso_state_delete;
616
617 pctx->create_depth_stencil_alpha_state = vc4_create_depth_stencil_alpha_state;
618 pctx->bind_depth_stencil_alpha_state = vc4_zsa_state_bind;
619 pctx->delete_depth_stencil_alpha_state = vc4_generic_cso_state_delete;
620
621 pctx->create_vertex_elements_state = vc4_vertex_state_create;
622 pctx->delete_vertex_elements_state = vc4_generic_cso_state_delete;
623 pctx->bind_vertex_elements_state = vc4_vertex_state_bind;
624
625 pctx->create_sampler_state = vc4_create_sampler_state;
626 pctx->delete_sampler_state = vc4_generic_cso_state_delete;
627 pctx->bind_sampler_states = vc4_sampler_states_bind;
628
629 pctx->create_sampler_view = vc4_create_sampler_view;
630 pctx->sampler_view_destroy = vc4_sampler_view_destroy;
631 pctx->set_sampler_views = vc4_set_sampler_views;
632 }