radeonsi: move scissor and viewport states into gallium/radeon
[mesa.git] / src / gallium / drivers / radeonsi / si_blit.c
1 /*
2 * Copyright 2010 Jerome Glisse <glisse@freedesktop.org>
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 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the 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 NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE.
22 */
23
24 #include "si_pipe.h"
25 #include "util/u_format.h"
26 #include "util/u_surface.h"
27
28 enum si_blitter_op /* bitmask */
29 {
30 SI_SAVE_TEXTURES = 1,
31 SI_SAVE_FRAMEBUFFER = 2,
32 SI_SAVE_FRAGMENT_STATE = 4,
33 SI_DISABLE_RENDER_COND = 8,
34
35 SI_CLEAR = SI_SAVE_FRAGMENT_STATE,
36
37 SI_CLEAR_SURFACE = SI_SAVE_FRAMEBUFFER | SI_SAVE_FRAGMENT_STATE,
38
39 SI_COPY = SI_SAVE_FRAMEBUFFER | SI_SAVE_TEXTURES |
40 SI_SAVE_FRAGMENT_STATE | SI_DISABLE_RENDER_COND,
41
42 SI_BLIT = SI_SAVE_FRAMEBUFFER | SI_SAVE_TEXTURES |
43 SI_SAVE_FRAGMENT_STATE,
44
45 SI_DECOMPRESS = SI_SAVE_FRAMEBUFFER | SI_SAVE_FRAGMENT_STATE |
46 SI_DISABLE_RENDER_COND,
47
48 SI_COLOR_RESOLVE = SI_SAVE_FRAMEBUFFER | SI_SAVE_FRAGMENT_STATE
49 };
50
51 static void si_blitter_begin(struct pipe_context *ctx, enum si_blitter_op op)
52 {
53 struct si_context *sctx = (struct si_context *)ctx;
54
55 util_blitter_save_vertex_buffer_slot(sctx->blitter, sctx->vertex_buffer);
56 util_blitter_save_vertex_elements(sctx->blitter, sctx->vertex_elements);
57 util_blitter_save_vertex_shader(sctx->blitter, sctx->vs_shader.cso);
58 util_blitter_save_tessctrl_shader(sctx->blitter, sctx->tcs_shader.cso);
59 util_blitter_save_tesseval_shader(sctx->blitter, sctx->tes_shader.cso);
60 util_blitter_save_geometry_shader(sctx->blitter, sctx->gs_shader.cso);
61 util_blitter_save_so_targets(sctx->blitter, sctx->b.streamout.num_targets,
62 (struct pipe_stream_output_target**)sctx->b.streamout.targets);
63 util_blitter_save_rasterizer(sctx->blitter, sctx->queued.named.rasterizer);
64
65 if (op & SI_SAVE_FRAGMENT_STATE) {
66 util_blitter_save_blend(sctx->blitter, sctx->queued.named.blend);
67 util_blitter_save_depth_stencil_alpha(sctx->blitter, sctx->queued.named.dsa);
68 util_blitter_save_stencil_ref(sctx->blitter, &sctx->stencil_ref.state);
69 util_blitter_save_fragment_shader(sctx->blitter, sctx->ps_shader.cso);
70 util_blitter_save_sample_mask(sctx->blitter, sctx->sample_mask.sample_mask);
71 util_blitter_save_viewport(sctx->blitter, &sctx->b.viewports.states[0]);
72 util_blitter_save_scissor(sctx->blitter, &sctx->b.scissors.states[0]);
73 }
74
75 if (op & SI_SAVE_FRAMEBUFFER)
76 util_blitter_save_framebuffer(sctx->blitter, &sctx->framebuffer.state);
77
78 if (op & SI_SAVE_TEXTURES) {
79 util_blitter_save_fragment_sampler_states(
80 sctx->blitter, 2,
81 sctx->samplers[PIPE_SHADER_FRAGMENT].views.sampler_states);
82
83 util_blitter_save_fragment_sampler_views(sctx->blitter, 2,
84 sctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
85 }
86
87 if (op & SI_DISABLE_RENDER_COND)
88 sctx->b.render_cond_force_off = true;
89 }
90
91 static void si_blitter_end(struct pipe_context *ctx)
92 {
93 struct si_context *sctx = (struct si_context *)ctx;
94
95 sctx->b.render_cond_force_off = false;
96 }
97
98 static unsigned u_max_sample(struct pipe_resource *r)
99 {
100 return r->nr_samples ? r->nr_samples - 1 : 0;
101 }
102
103 static void si_blit_decompress_depth(struct pipe_context *ctx,
104 struct r600_texture *texture,
105 struct r600_texture *staging,
106 unsigned first_level, unsigned last_level,
107 unsigned first_layer, unsigned last_layer,
108 unsigned first_sample, unsigned last_sample)
109 {
110 struct si_context *sctx = (struct si_context *)ctx;
111 unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
112 float depth = 1.0f;
113 const struct util_format_description *desc;
114 struct r600_texture *flushed_depth_texture = staging ?
115 staging : texture->flushed_depth_texture;
116
117 if (!staging && !texture->dirty_level_mask)
118 return;
119
120 max_sample = u_max_sample(&texture->resource.b.b);
121
122 desc = util_format_description(flushed_depth_texture->resource.b.b.format);
123
124 if (util_format_has_depth(desc))
125 sctx->dbcb_depth_copy_enabled = true;
126 if (util_format_has_stencil(desc))
127 sctx->dbcb_stencil_copy_enabled = true;
128
129 assert(sctx->dbcb_depth_copy_enabled || sctx->dbcb_stencil_copy_enabled);
130
131 for (level = first_level; level <= last_level; level++) {
132 if (!staging && !(texture->dirty_level_mask & (1 << level)))
133 continue;
134
135 /* The smaller the mipmap level, the less layers there are
136 * as far as 3D textures are concerned. */
137 max_layer = util_max_layer(&texture->resource.b.b, level);
138 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
139
140 for (layer = first_layer; layer <= checked_last_layer; layer++) {
141 for (sample = first_sample; sample <= last_sample; sample++) {
142 struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
143
144 sctx->dbcb_copy_sample = sample;
145 si_mark_atom_dirty(sctx, &sctx->db_render_state);
146
147 surf_tmpl.format = texture->resource.b.b.format;
148 surf_tmpl.u.tex.level = level;
149 surf_tmpl.u.tex.first_layer = layer;
150 surf_tmpl.u.tex.last_layer = layer;
151
152 zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
153
154 surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
155 cbsurf = ctx->create_surface(ctx,
156 (struct pipe_resource*)flushed_depth_texture, &surf_tmpl);
157
158 si_blitter_begin(ctx, SI_DECOMPRESS);
159 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, cbsurf, 1 << sample,
160 sctx->custom_dsa_flush, depth);
161 si_blitter_end(ctx);
162
163 pipe_surface_reference(&zsurf, NULL);
164 pipe_surface_reference(&cbsurf, NULL);
165 }
166 }
167
168 /* The texture will always be dirty if some layers aren't flushed.
169 * I don't think this case can occur though. */
170 if (!staging &&
171 first_layer == 0 && last_layer == max_layer &&
172 first_sample == 0 && last_sample == max_sample) {
173 texture->dirty_level_mask &= ~(1 << level);
174 }
175 }
176
177 sctx->dbcb_depth_copy_enabled = false;
178 sctx->dbcb_stencil_copy_enabled = false;
179 si_mark_atom_dirty(sctx, &sctx->db_render_state);
180 }
181
182 static void si_blit_decompress_depth_in_place(struct si_context *sctx,
183 struct r600_texture *texture,
184 bool is_stencil_sampler,
185 unsigned first_level, unsigned last_level,
186 unsigned first_layer, unsigned last_layer)
187 {
188 struct pipe_surface *zsurf, surf_tmpl = {{0}};
189 unsigned layer, max_layer, checked_last_layer, level;
190 unsigned *dirty_level_mask;
191
192 if (is_stencil_sampler) {
193 sctx->db_flush_stencil_inplace = true;
194 dirty_level_mask = &texture->stencil_dirty_level_mask;
195 } else {
196 sctx->db_flush_depth_inplace = true;
197 dirty_level_mask = &texture->dirty_level_mask;
198 }
199 si_mark_atom_dirty(sctx, &sctx->db_render_state);
200
201 surf_tmpl.format = texture->resource.b.b.format;
202
203 for (level = first_level; level <= last_level; level++) {
204 if (!(*dirty_level_mask & (1 << level)))
205 continue;
206
207 surf_tmpl.u.tex.level = level;
208
209 /* The smaller the mipmap level, the less layers there are
210 * as far as 3D textures are concerned. */
211 max_layer = util_max_layer(&texture->resource.b.b, level);
212 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
213
214 for (layer = first_layer; layer <= checked_last_layer; layer++) {
215 surf_tmpl.u.tex.first_layer = layer;
216 surf_tmpl.u.tex.last_layer = layer;
217
218 zsurf = sctx->b.b.create_surface(&sctx->b.b, &texture->resource.b.b, &surf_tmpl);
219
220 si_blitter_begin(&sctx->b.b, SI_DECOMPRESS);
221 util_blitter_custom_depth_stencil(sctx->blitter, zsurf, NULL, ~0,
222 sctx->custom_dsa_flush,
223 1.0f);
224 si_blitter_end(&sctx->b.b);
225
226 pipe_surface_reference(&zsurf, NULL);
227 }
228
229 /* The texture will always be dirty if some layers aren't flushed.
230 * I don't think this case occurs often though. */
231 if (first_layer == 0 && last_layer == max_layer) {
232 *dirty_level_mask &= ~(1 << level);
233 }
234 }
235
236 sctx->db_flush_depth_inplace = false;
237 sctx->db_flush_stencil_inplace = false;
238 si_mark_atom_dirty(sctx, &sctx->db_render_state);
239 }
240
241 static void
242 si_flush_depth_textures(struct si_context *sctx,
243 struct si_textures_info *textures)
244 {
245 unsigned i;
246 uint64_t mask = textures->depth_texture_mask;
247
248 while (mask) {
249 struct pipe_sampler_view *view;
250 struct si_sampler_view *sview;
251 struct r600_texture *tex;
252
253 i = u_bit_scan64(&mask);
254
255 view = textures->views.views[i];
256 assert(view);
257 sview = (struct si_sampler_view*)view;
258
259 tex = (struct r600_texture *)view->texture;
260 assert(tex->is_depth && !tex->is_flushing_texture);
261
262 si_blit_decompress_depth_in_place(sctx, tex,
263 sview->is_stencil_sampler,
264 view->u.tex.first_level, view->u.tex.last_level,
265 0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
266 }
267 }
268
269 static void si_blit_decompress_color(struct pipe_context *ctx,
270 struct r600_texture *rtex,
271 unsigned first_level, unsigned last_level,
272 unsigned first_layer, unsigned last_layer,
273 bool need_dcc_decompress)
274 {
275 struct si_context *sctx = (struct si_context *)ctx;
276 unsigned layer, level, checked_last_layer, max_layer;
277
278 if (!rtex->dirty_level_mask && !need_dcc_decompress)
279 return;
280
281 for (level = first_level; level <= last_level; level++) {
282 void* custom_blend;
283
284 if (!(rtex->dirty_level_mask & (1 << level)) && !need_dcc_decompress)
285 continue;
286
287 if (rtex->dcc_offset && need_dcc_decompress) {
288 custom_blend = sctx->custom_blend_dcc_decompress;
289 } else if (rtex->fmask.size) {
290 custom_blend = sctx->custom_blend_decompress;
291 } else {
292 custom_blend = sctx->custom_blend_fastclear;
293 }
294
295 /* The smaller the mipmap level, the less layers there are
296 * as far as 3D textures are concerned. */
297 max_layer = util_max_layer(&rtex->resource.b.b, level);
298 checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
299
300 for (layer = first_layer; layer <= checked_last_layer; layer++) {
301 struct pipe_surface *cbsurf, surf_tmpl;
302
303 surf_tmpl.format = rtex->resource.b.b.format;
304 surf_tmpl.u.tex.level = level;
305 surf_tmpl.u.tex.first_layer = layer;
306 surf_tmpl.u.tex.last_layer = layer;
307 cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
308
309 si_blitter_begin(ctx, SI_DECOMPRESS);
310 util_blitter_custom_color(sctx->blitter, cbsurf, custom_blend);
311 si_blitter_end(ctx);
312
313 pipe_surface_reference(&cbsurf, NULL);
314 }
315
316 /* The texture will always be dirty if some layers aren't flushed.
317 * I don't think this case occurs often though. */
318 if (first_layer == 0 && last_layer == max_layer) {
319 rtex->dirty_level_mask &= ~(1 << level);
320 }
321 }
322 }
323
324 static void
325 si_decompress_sampler_color_textures(struct si_context *sctx,
326 struct si_textures_info *textures)
327 {
328 unsigned i;
329 uint64_t mask = textures->compressed_colortex_mask;
330
331 while (mask) {
332 struct pipe_sampler_view *view;
333 struct r600_texture *tex;
334
335 i = u_bit_scan64(&mask);
336
337 view = textures->views.views[i];
338 assert(view);
339
340 tex = (struct r600_texture *)view->texture;
341 assert(tex->cmask.size || tex->fmask.size || tex->dcc_offset);
342
343 si_blit_decompress_color(&sctx->b.b, tex,
344 view->u.tex.first_level, view->u.tex.last_level,
345 0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
346 false);
347 }
348 }
349
350 static void
351 si_decompress_image_color_textures(struct si_context *sctx,
352 struct si_images_info *images)
353 {
354 unsigned i;
355 uint64_t mask = images->compressed_colortex_mask;
356
357 while (mask) {
358 const struct pipe_image_view *view;
359 struct r600_texture *tex;
360
361 i = u_bit_scan64(&mask);
362
363 view = &images->views[i];
364 assert(view->resource->target != PIPE_BUFFER);
365
366 tex = (struct r600_texture *)view->resource;
367 if (!tex->cmask.size && !tex->fmask.size && !tex->dcc_offset)
368 continue;
369
370 si_blit_decompress_color(&sctx->b.b, tex,
371 view->u.tex.level, view->u.tex.level,
372 0, util_max_layer(&tex->resource.b.b, view->u.tex.level),
373 false);
374 }
375 }
376
377 void si_decompress_textures(struct si_context *sctx)
378 {
379 unsigned compressed_colortex_counter;
380
381 if (sctx->blitter->running)
382 return;
383
384 /* Update the compressed_colortex_mask if necessary. */
385 compressed_colortex_counter = p_atomic_read(&sctx->screen->b.compressed_colortex_counter);
386 if (compressed_colortex_counter != sctx->b.last_compressed_colortex_counter) {
387 sctx->b.last_compressed_colortex_counter = compressed_colortex_counter;
388 si_update_compressed_colortex_masks(sctx);
389 }
390
391 /* Flush depth textures which need to be flushed. */
392 for (int i = 0; i < SI_NUM_SHADERS; i++) {
393 if (sctx->samplers[i].depth_texture_mask) {
394 si_flush_depth_textures(sctx, &sctx->samplers[i]);
395 }
396 if (sctx->samplers[i].compressed_colortex_mask) {
397 si_decompress_sampler_color_textures(sctx, &sctx->samplers[i]);
398 }
399 if (sctx->images[i].compressed_colortex_mask) {
400 si_decompress_image_color_textures(sctx, &sctx->images[i]);
401 }
402 }
403 }
404
405 static void si_clear(struct pipe_context *ctx, unsigned buffers,
406 const union pipe_color_union *color,
407 double depth, unsigned stencil)
408 {
409 struct si_context *sctx = (struct si_context *)ctx;
410 struct pipe_framebuffer_state *fb = &sctx->framebuffer.state;
411 struct pipe_surface *zsbuf = fb->zsbuf;
412 struct r600_texture *zstex =
413 zsbuf ? (struct r600_texture*)zsbuf->texture : NULL;
414
415 if (buffers & PIPE_CLEAR_COLOR) {
416 evergreen_do_fast_color_clear(&sctx->b, fb,
417 &sctx->framebuffer.atom, &buffers,
418 &sctx->framebuffer.dirty_cbufs,
419 color);
420 if (!buffers)
421 return; /* all buffers have been fast cleared */
422 }
423
424 if (buffers & PIPE_CLEAR_COLOR) {
425 int i;
426
427 /* These buffers cannot use fast clear, make sure to disable expansion. */
428 for (i = 0; i < fb->nr_cbufs; i++) {
429 struct r600_texture *tex;
430
431 /* If not clearing this buffer, skip. */
432 if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
433 continue;
434
435 if (!fb->cbufs[i])
436 continue;
437
438 tex = (struct r600_texture *)fb->cbufs[i]->texture;
439 if (tex->fmask.size == 0)
440 tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
441 }
442 }
443
444 if (zstex && zstex->htile_buffer &&
445 zsbuf->u.tex.level == 0 &&
446 zsbuf->u.tex.first_layer == 0 &&
447 zsbuf->u.tex.last_layer == util_max_layer(&zstex->resource.b.b, 0)) {
448 if (buffers & PIPE_CLEAR_DEPTH) {
449 /* Need to disable EXPCLEAR temporarily if clearing
450 * to a new value. */
451 if (zstex->depth_cleared && zstex->depth_clear_value != depth) {
452 sctx->db_depth_disable_expclear = true;
453 }
454
455 zstex->depth_clear_value = depth;
456 sctx->framebuffer.dirty_zsbuf = true;
457 si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_DEPTH_CLEAR */
458 sctx->db_depth_clear = true;
459 si_mark_atom_dirty(sctx, &sctx->db_render_state);
460 }
461
462 if (buffers & PIPE_CLEAR_STENCIL) {
463 stencil &= 0xff;
464
465 /* Need to disable EXPCLEAR temporarily if clearing
466 * to a new value. */
467 if (zstex->stencil_cleared && zstex->stencil_clear_value != stencil) {
468 sctx->db_stencil_disable_expclear = true;
469 }
470
471 zstex->stencil_clear_value = stencil;
472 sctx->framebuffer.dirty_zsbuf = true;
473 si_mark_atom_dirty(sctx, &sctx->framebuffer.atom); /* updates DB_STENCIL_CLEAR */
474 sctx->db_stencil_clear = true;
475 si_mark_atom_dirty(sctx, &sctx->db_render_state);
476 }
477 }
478
479 si_blitter_begin(ctx, SI_CLEAR);
480 util_blitter_clear(sctx->blitter, fb->width, fb->height,
481 util_framebuffer_get_num_layers(fb),
482 buffers, color, depth, stencil);
483 si_blitter_end(ctx);
484
485 if (sctx->db_depth_clear) {
486 sctx->db_depth_clear = false;
487 sctx->db_depth_disable_expclear = false;
488 zstex->depth_cleared = true;
489 si_mark_atom_dirty(sctx, &sctx->db_render_state);
490 }
491
492 if (sctx->db_stencil_clear) {
493 sctx->db_stencil_clear = false;
494 sctx->db_stencil_disable_expclear = false;
495 zstex->stencil_cleared = true;
496 si_mark_atom_dirty(sctx, &sctx->db_render_state);
497 }
498 }
499
500 static void si_clear_render_target(struct pipe_context *ctx,
501 struct pipe_surface *dst,
502 const union pipe_color_union *color,
503 unsigned dstx, unsigned dsty,
504 unsigned width, unsigned height)
505 {
506 struct si_context *sctx = (struct si_context *)ctx;
507
508 si_blitter_begin(ctx, SI_CLEAR_SURFACE);
509 util_blitter_clear_render_target(sctx->blitter, dst, color,
510 dstx, dsty, width, height);
511 si_blitter_end(ctx);
512 }
513
514 static void si_clear_depth_stencil(struct pipe_context *ctx,
515 struct pipe_surface *dst,
516 unsigned clear_flags,
517 double depth,
518 unsigned stencil,
519 unsigned dstx, unsigned dsty,
520 unsigned width, unsigned height)
521 {
522 struct si_context *sctx = (struct si_context *)ctx;
523
524 si_blitter_begin(ctx, SI_CLEAR_SURFACE);
525 util_blitter_clear_depth_stencil(sctx->blitter, dst, clear_flags, depth, stencil,
526 dstx, dsty, width, height);
527 si_blitter_end(ctx);
528 }
529
530 /* Helper for decompressing a portion of a color or depth resource before
531 * blitting if any decompression is needed.
532 * The driver doesn't decompress resources automatically while u_blitter is
533 * rendering. */
534 static void si_decompress_subresource(struct pipe_context *ctx,
535 struct pipe_resource *tex,
536 unsigned level,
537 unsigned first_layer, unsigned last_layer)
538 {
539 struct si_context *sctx = (struct si_context *)ctx;
540 struct r600_texture *rtex = (struct r600_texture*)tex;
541
542 if (rtex->is_depth && !rtex->is_flushing_texture) {
543 si_blit_decompress_depth_in_place(sctx, rtex, false,
544 level, level,
545 first_layer, last_layer);
546 if (rtex->surface.flags & RADEON_SURF_SBUFFER)
547 si_blit_decompress_depth_in_place(sctx, rtex, true,
548 level, level,
549 first_layer, last_layer);
550 } else if (rtex->fmask.size || rtex->cmask.size || rtex->dcc_offset) {
551 si_blit_decompress_color(ctx, rtex, level, level,
552 first_layer, last_layer, false);
553 }
554 }
555
556 struct texture_orig_info {
557 unsigned format;
558 unsigned width0;
559 unsigned height0;
560 unsigned npix_x;
561 unsigned npix_y;
562 unsigned npix0_x;
563 unsigned npix0_y;
564 };
565
566 void si_resource_copy_region(struct pipe_context *ctx,
567 struct pipe_resource *dst,
568 unsigned dst_level,
569 unsigned dstx, unsigned dsty, unsigned dstz,
570 struct pipe_resource *src,
571 unsigned src_level,
572 const struct pipe_box *src_box)
573 {
574 struct si_context *sctx = (struct si_context *)ctx;
575 struct pipe_surface *dst_view, dst_templ;
576 struct pipe_sampler_view src_templ, *src_view;
577 unsigned dst_width, dst_height, src_width0, src_height0;
578 unsigned src_force_level = 0;
579 struct pipe_box sbox, dstbox;
580
581 /* Handle buffers first. */
582 if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
583 si_copy_buffer(sctx, dst, src, dstx, src_box->x, src_box->width, false);
584 return;
585 }
586
587 assert(u_max_sample(dst) == u_max_sample(src));
588
589 /* The driver doesn't decompress resources automatically while
590 * u_blitter is rendering. */
591 si_decompress_subresource(ctx, src, src_level,
592 src_box->z, src_box->z + src_box->depth - 1);
593
594 dst_width = u_minify(dst->width0, dst_level);
595 dst_height = u_minify(dst->height0, dst_level);
596 src_width0 = src->width0;
597 src_height0 = src->height0;
598
599 util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
600 util_blitter_default_src_texture(&src_templ, src, src_level);
601
602 if (util_format_is_compressed(src->format) ||
603 util_format_is_compressed(dst->format)) {
604 unsigned blocksize = util_format_get_blocksize(src->format);
605
606 if (blocksize == 8)
607 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
608 else
609 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
610 dst_templ.format = src_templ.format;
611
612 dst_width = util_format_get_nblocksx(dst->format, dst_width);
613 dst_height = util_format_get_nblocksy(dst->format, dst_height);
614 src_width0 = util_format_get_nblocksx(src->format, src_width0);
615 src_height0 = util_format_get_nblocksy(src->format, src_height0);
616
617 dstx = util_format_get_nblocksx(dst->format, dstx);
618 dsty = util_format_get_nblocksy(dst->format, dsty);
619
620 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
621 sbox.y = util_format_get_nblocksy(src->format, src_box->y);
622 sbox.z = src_box->z;
623 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
624 sbox.height = util_format_get_nblocksy(src->format, src_box->height);
625 sbox.depth = src_box->depth;
626 src_box = &sbox;
627
628 src_force_level = src_level;
629 } else if (!util_blitter_is_copy_supported(sctx->blitter, dst, src) ||
630 /* also *8_SNORM has precision issues, use UNORM instead */
631 util_format_is_snorm8(src->format)) {
632 if (util_format_is_subsampled_422(src->format)) {
633 src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
634 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
635
636 dst_width = util_format_get_nblocksx(dst->format, dst_width);
637 src_width0 = util_format_get_nblocksx(src->format, src_width0);
638
639 dstx = util_format_get_nblocksx(dst->format, dstx);
640
641 sbox = *src_box;
642 sbox.x = util_format_get_nblocksx(src->format, src_box->x);
643 sbox.width = util_format_get_nblocksx(src->format, src_box->width);
644 src_box = &sbox;
645 } else {
646 unsigned blocksize = util_format_get_blocksize(src->format);
647
648 switch (blocksize) {
649 case 1:
650 dst_templ.format = PIPE_FORMAT_R8_UNORM;
651 src_templ.format = PIPE_FORMAT_R8_UNORM;
652 break;
653 case 2:
654 dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
655 src_templ.format = PIPE_FORMAT_R8G8_UNORM;
656 break;
657 case 4:
658 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
659 src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
660 break;
661 case 8:
662 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
663 src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
664 break;
665 case 16:
666 dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
667 src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
668 break;
669 default:
670 fprintf(stderr, "Unhandled format %s with blocksize %u\n",
671 util_format_short_name(src->format), blocksize);
672 assert(0);
673 }
674 }
675 }
676
677 /* Initialize the surface. */
678 dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
679 dst_width, dst_height);
680
681 /* Initialize the sampler view. */
682 src_view = si_create_sampler_view_custom(ctx, src, &src_templ,
683 src_width0, src_height0,
684 src_force_level);
685
686 u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
687 abs(src_box->depth), &dstbox);
688
689 /* Copy. */
690 si_blitter_begin(ctx, SI_COPY);
691 util_blitter_blit_generic(sctx->blitter, dst_view, &dstbox,
692 src_view, src_box, src_width0, src_height0,
693 PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
694 FALSE);
695 si_blitter_end(ctx);
696
697 pipe_surface_reference(&dst_view, NULL);
698 pipe_sampler_view_reference(&src_view, NULL);
699 }
700
701 /* For MSAA integer resolving to work, we change the format to NORM using this function. */
702 static enum pipe_format int_to_norm_format(enum pipe_format format)
703 {
704 switch (format) {
705 #define REPLACE_FORMAT_SIGN(format,sign) \
706 case PIPE_FORMAT_##format##_##sign##INT: \
707 return PIPE_FORMAT_##format##_##sign##NORM
708 #define REPLACE_FORMAT(format) \
709 REPLACE_FORMAT_SIGN(format, U); \
710 REPLACE_FORMAT_SIGN(format, S)
711
712 REPLACE_FORMAT_SIGN(B10G10R10A2, U);
713 REPLACE_FORMAT(R8);
714 REPLACE_FORMAT(R8G8);
715 REPLACE_FORMAT(R8G8B8X8);
716 REPLACE_FORMAT(R8G8B8A8);
717 REPLACE_FORMAT(A8);
718 REPLACE_FORMAT(I8);
719 REPLACE_FORMAT(L8);
720 REPLACE_FORMAT(L8A8);
721 REPLACE_FORMAT(R16);
722 REPLACE_FORMAT(R16G16);
723 REPLACE_FORMAT(R16G16B16X16);
724 REPLACE_FORMAT(R16G16B16A16);
725 REPLACE_FORMAT(A16);
726 REPLACE_FORMAT(I16);
727 REPLACE_FORMAT(L16);
728 REPLACE_FORMAT(L16A16);
729
730 #undef REPLACE_FORMAT
731 #undef REPLACE_FORMAT_SIGN
732 default:
733 return format;
734 }
735 }
736
737 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
738 const struct pipe_blit_info *info)
739 {
740 struct si_context *sctx = (struct si_context*)ctx;
741 struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
742 unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
743 unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
744 enum pipe_format format = int_to_norm_format(info->dst.format);
745 unsigned sample_mask = ~0;
746
747 /* Hardware MSAA resolve doesn't work if SPI format = NORM16_ABGR and
748 * the format is R16G16. Use R16A16, which does work.
749 */
750 if (format == PIPE_FORMAT_R16G16_UNORM)
751 format = PIPE_FORMAT_R16A16_UNORM;
752 if (format == PIPE_FORMAT_R16G16_SNORM)
753 format = PIPE_FORMAT_R16A16_SNORM;
754
755 if (info->src.resource->nr_samples > 1 &&
756 info->dst.resource->nr_samples <= 1 &&
757 util_max_layer(info->src.resource, 0) == 0 &&
758 util_max_layer(info->dst.resource, info->dst.level) == 0 &&
759 info->dst.format == info->src.format &&
760 !util_format_is_pure_integer(format) &&
761 !util_format_is_depth_or_stencil(format) &&
762 !info->scissor_enable &&
763 (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
764 dst_width == info->src.resource->width0 &&
765 dst_height == info->src.resource->height0 &&
766 info->dst.box.x == 0 &&
767 info->dst.box.y == 0 &&
768 info->dst.box.width == dst_width &&
769 info->dst.box.height == dst_height &&
770 info->dst.box.depth == 1 &&
771 info->src.box.x == 0 &&
772 info->src.box.y == 0 &&
773 info->src.box.width == dst_width &&
774 info->src.box.height == dst_height &&
775 info->src.box.depth == 1 &&
776 dst->surface.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
777 !(dst->surface.flags & RADEON_SURF_SCANOUT) &&
778 (!dst->cmask.size || !dst->dirty_level_mask) && /* dst cannot be fast-cleared */
779 !dst->dcc_offset) {
780 si_blitter_begin(ctx, SI_COLOR_RESOLVE |
781 (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
782 util_blitter_custom_resolve_color(sctx->blitter,
783 info->dst.resource, info->dst.level,
784 info->dst.box.z,
785 info->src.resource, info->src.box.z,
786 sample_mask, sctx->custom_blend_resolve,
787 format);
788 si_blitter_end(ctx);
789 return true;
790 }
791 return false;
792 }
793
794 static void si_blit(struct pipe_context *ctx,
795 const struct pipe_blit_info *info)
796 {
797 struct si_context *sctx = (struct si_context*)ctx;
798
799 if (do_hardware_msaa_resolve(ctx, info)) {
800 return;
801 }
802
803 assert(util_blitter_is_blit_supported(sctx->blitter, info));
804
805 /* The driver doesn't decompress resources automatically while
806 * u_blitter is rendering. */
807 si_decompress_subresource(ctx, info->src.resource, info->src.level,
808 info->src.box.z,
809 info->src.box.z + info->src.box.depth - 1);
810
811 if (sctx->screen->b.debug_flags & DBG_FORCE_DMA &&
812 util_try_blit_via_copy_region(ctx, info))
813 return;
814
815 si_blitter_begin(ctx, SI_BLIT |
816 (info->render_condition_enable ? 0 : SI_DISABLE_RENDER_COND));
817 util_blitter_blit(sctx->blitter, info);
818 si_blitter_end(ctx);
819 }
820
821 static void si_flush_resource(struct pipe_context *ctx,
822 struct pipe_resource *res)
823 {
824 struct r600_texture *rtex = (struct r600_texture*)res;
825
826 assert(res->target != PIPE_BUFFER);
827
828 if (!rtex->is_depth && (rtex->cmask.size || rtex->dcc_offset)) {
829 si_blit_decompress_color(ctx, rtex, 0, res->last_level,
830 0, util_max_layer(res, 0), false);
831 }
832 }
833
834 static void si_decompress_dcc(struct pipe_context *ctx,
835 struct r600_texture *rtex)
836 {
837 if (!rtex->dcc_offset)
838 return;
839
840 si_blit_decompress_color(ctx, rtex, 0, rtex->resource.b.b.last_level,
841 0, util_max_layer(&rtex->resource.b.b, 0),
842 true);
843 }
844
845 static void si_pipe_clear_buffer(struct pipe_context *ctx,
846 struct pipe_resource *dst,
847 unsigned offset, unsigned size,
848 const void *clear_value_ptr,
849 int clear_value_size)
850 {
851 struct si_context *sctx = (struct si_context*)ctx;
852 uint32_t dword_value;
853 unsigned i;
854
855 assert(offset % clear_value_size == 0);
856 assert(size % clear_value_size == 0);
857
858 if (clear_value_size > 4) {
859 const uint32_t *u32 = clear_value_ptr;
860 bool clear_dword_duplicated = true;
861
862 /* See if we can lower large fills to dword fills. */
863 for (i = 1; i < clear_value_size / 4; i++)
864 if (u32[0] != u32[i]) {
865 clear_dword_duplicated = false;
866 break;
867 }
868
869 if (!clear_dword_duplicated) {
870 /* Use transform feedback for 64-bit, 96-bit, and
871 * 128-bit fills.
872 */
873 union pipe_color_union clear_value;
874
875 memcpy(&clear_value, clear_value_ptr, clear_value_size);
876 si_blitter_begin(ctx, SI_DISABLE_RENDER_COND);
877 util_blitter_clear_buffer(sctx->blitter, dst, offset,
878 size, clear_value_size / 4,
879 &clear_value);
880 si_blitter_end(ctx);
881 return;
882 }
883 }
884
885 /* Expand the clear value to a dword. */
886 switch (clear_value_size) {
887 case 1:
888 dword_value = *(uint8_t*)clear_value_ptr;
889 dword_value |= (dword_value << 8) |
890 (dword_value << 16) |
891 (dword_value << 24);
892 break;
893 case 2:
894 dword_value = *(uint16_t*)clear_value_ptr;
895 dword_value |= dword_value << 16;
896 break;
897 default:
898 dword_value = *(uint32_t*)clear_value_ptr;
899 }
900
901 sctx->b.clear_buffer(ctx, dst, offset, size, dword_value, false);
902 }
903
904 void si_init_blit_functions(struct si_context *sctx)
905 {
906 sctx->b.b.clear = si_clear;
907 sctx->b.b.clear_buffer = si_pipe_clear_buffer;
908 sctx->b.b.clear_render_target = si_clear_render_target;
909 sctx->b.b.clear_depth_stencil = si_clear_depth_stencil;
910 sctx->b.b.resource_copy_region = si_resource_copy_region;
911 sctx->b.b.blit = si_blit;
912 sctx->b.b.flush_resource = si_flush_resource;
913 sctx->b.blit_decompress_depth = si_blit_decompress_depth;
914 sctx->b.decompress_dcc = si_decompress_dcc;
915 }