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 #include "r600_pipe.h"
24 #include "compute_memory_pool.h"
25 #include "evergreen_compute.h"
26 #include "util/u_surface.h"
27 #include "util/format/u_format.h"
28 #include "evergreend.h"
29 
30 enum r600_blitter_op /* bitmask */
31 {
32 	R600_SAVE_FRAGMENT_STATE = 1,
33 	R600_SAVE_TEXTURES       = 2,
34 	R600_SAVE_FRAMEBUFFER    = 4,
35 	R600_DISABLE_RENDER_COND = 8,
36 
37 	R600_CLEAR         = R600_SAVE_FRAGMENT_STATE,
38 
39 	R600_CLEAR_SURFACE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER,
40 
41 	R600_COPY_BUFFER   = R600_DISABLE_RENDER_COND,
42 
43 	R600_COPY_TEXTURE  = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES |
44 			     R600_DISABLE_RENDER_COND,
45 
46 	R600_BLIT          = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_SAVE_TEXTURES,
47 
48 	R600_DECOMPRESS    = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER | R600_DISABLE_RENDER_COND,
49 
50 	R600_COLOR_RESOLVE = R600_SAVE_FRAGMENT_STATE | R600_SAVE_FRAMEBUFFER
51 };
52 
r600_blitter_begin(struct pipe_context * ctx,enum r600_blitter_op op)53 static void r600_blitter_begin(struct pipe_context *ctx, enum r600_blitter_op op)
54 {
55 	struct r600_context *rctx = (struct r600_context *)ctx;
56 
57 	if (rctx->cmd_buf_is_compute) {
58 		rctx->b.gfx.flush(rctx, PIPE_FLUSH_ASYNC, NULL);
59 		rctx->cmd_buf_is_compute = false;
60 	}
61 
62 	util_blitter_save_vertex_buffer_slot(rctx->blitter, rctx->vertex_buffer_state.vb);
63 	util_blitter_save_vertex_elements(rctx->blitter, rctx->vertex_fetch_shader.cso);
64 	util_blitter_save_vertex_shader(rctx->blitter, rctx->vs_shader);
65 	util_blitter_save_geometry_shader(rctx->blitter, rctx->gs_shader);
66 	util_blitter_save_tessctrl_shader(rctx->blitter, rctx->tcs_shader);
67 	util_blitter_save_tesseval_shader(rctx->blitter, rctx->tes_shader);
68 	util_blitter_save_so_targets(rctx->blitter, rctx->b.streamout.num_targets,
69 				     (struct pipe_stream_output_target**)rctx->b.streamout.targets);
70 	util_blitter_save_rasterizer(rctx->blitter, rctx->rasterizer_state.cso);
71 
72 	if (op & R600_SAVE_FRAGMENT_STATE) {
73 		util_blitter_save_viewport(rctx->blitter, &rctx->b.viewports.states[0]);
74 		util_blitter_save_scissor(rctx->blitter, &rctx->b.scissors.states[0]);
75 		util_blitter_save_fragment_shader(rctx->blitter, rctx->ps_shader);
76 		util_blitter_save_blend(rctx->blitter, rctx->blend_state.cso);
77 		util_blitter_save_depth_stencil_alpha(rctx->blitter, rctx->dsa_state.cso);
78 		util_blitter_save_stencil_ref(rctx->blitter, &rctx->stencil_ref.pipe_state);
79                 util_blitter_save_sample_mask(rctx->blitter, rctx->sample_mask.sample_mask);
80 	}
81 
82 	if (op & R600_SAVE_FRAMEBUFFER)
83 		util_blitter_save_framebuffer(rctx->blitter, &rctx->framebuffer.state);
84 
85 	if (op & R600_SAVE_TEXTURES) {
86 		util_blitter_save_fragment_sampler_states(
87 			rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].states.enabled_mask),
88 			(void**)rctx->samplers[PIPE_SHADER_FRAGMENT].states.states);
89 
90 		util_blitter_save_fragment_sampler_views(
91 			rctx->blitter, util_last_bit(rctx->samplers[PIPE_SHADER_FRAGMENT].views.enabled_mask),
92 			(struct pipe_sampler_view**)rctx->samplers[PIPE_SHADER_FRAGMENT].views.views);
93 	}
94 
95 	if (op & R600_DISABLE_RENDER_COND)
96 		rctx->b.render_cond_force_off = true;
97 }
98 
r600_blitter_end(struct pipe_context * ctx)99 static void r600_blitter_end(struct pipe_context *ctx)
100 {
101 	struct r600_context *rctx = (struct r600_context *)ctx;
102 
103 	rctx->b.render_cond_force_off = false;
104 }
105 
u_max_sample(struct pipe_resource * r)106 static unsigned u_max_sample(struct pipe_resource *r)
107 {
108 	return r->nr_samples ? r->nr_samples - 1 : 0;
109 }
110 
r600_blit_decompress_depth(struct pipe_context * ctx,struct r600_texture * texture,struct r600_texture * staging,unsigned first_level,unsigned last_level,unsigned first_layer,unsigned last_layer,unsigned first_sample,unsigned last_sample)111 static void r600_blit_decompress_depth(struct pipe_context *ctx,
112 				       struct r600_texture *texture,
113 				       struct r600_texture *staging,
114 				       unsigned first_level, unsigned last_level,
115 				       unsigned first_layer, unsigned last_layer,
116 				       unsigned first_sample, unsigned last_sample)
117 {
118 	struct r600_context *rctx = (struct r600_context *)ctx;
119 	unsigned layer, level, sample, checked_last_layer, max_layer, max_sample;
120 	struct r600_texture *flushed_depth_texture = staging ?
121 			staging : texture->flushed_depth_texture;
122 	const struct util_format_description *desc =
123 		util_format_description(texture->resource.b.b.format);
124 	float depth;
125 
126 	if (!staging && !texture->dirty_level_mask)
127 		return;
128 
129 	max_sample = u_max_sample(&texture->resource.b.b);
130 
131 	/* XXX Decompressing MSAA depth textures is broken on R6xx.
132 	 * There is also a hardlock if CMASK and FMASK are not present.
133 	 * Just skip this until we find out how to fix it. */
134 	if (rctx->b.chip_class == R600 && max_sample > 0) {
135 		texture->dirty_level_mask = 0;
136 		return;
137 	}
138 
139 	if (rctx->b.family == CHIP_RV610 || rctx->b.family == CHIP_RV630 ||
140 	    rctx->b.family == CHIP_RV620 || rctx->b.family == CHIP_RV635)
141 		depth = 0.0f;
142 	else
143 		depth = 1.0f;
144 
145 	/* Enable decompression in DB_RENDER_CONTROL */
146 	rctx->db_misc_state.flush_depthstencil_through_cb = true;
147 	rctx->db_misc_state.copy_depth = util_format_has_depth(desc);
148 	rctx->db_misc_state.copy_stencil = util_format_has_stencil(desc);
149 	rctx->db_misc_state.copy_sample = first_sample;
150 	r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
151 
152 	for (level = first_level; level <= last_level; level++) {
153 		if (!staging && !(texture->dirty_level_mask & (1 << level)))
154 			continue;
155 
156 		/* The smaller the mipmap level, the less layers there are
157 		 * as far as 3D textures are concerned. */
158 		max_layer = util_max_layer(&texture->resource.b.b, level);
159 		checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
160 
161 		for (layer = first_layer; layer <= checked_last_layer; layer++) {
162 			for (sample = first_sample; sample <= last_sample; sample++) {
163 				struct pipe_surface *zsurf, *cbsurf, surf_tmpl;
164 
165 				if (sample != rctx->db_misc_state.copy_sample) {
166 					rctx->db_misc_state.copy_sample = sample;
167 					r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
168 				}
169 
170 				surf_tmpl.format = texture->resource.b.b.format;
171 				surf_tmpl.u.tex.level = level;
172 				surf_tmpl.u.tex.first_layer = layer;
173 				surf_tmpl.u.tex.last_layer = layer;
174 
175 				zsurf = ctx->create_surface(ctx, &texture->resource.b.b, &surf_tmpl);
176 
177 				surf_tmpl.format = flushed_depth_texture->resource.b.b.format;
178 				cbsurf = ctx->create_surface(ctx,
179 						&flushed_depth_texture->resource.b.b, &surf_tmpl);
180 
181 				r600_blitter_begin(ctx, R600_DECOMPRESS);
182 				util_blitter_custom_depth_stencil(rctx->blitter, zsurf, cbsurf, 1 << sample,
183 								  rctx->custom_dsa_flush, depth);
184 				r600_blitter_end(ctx);
185 
186 				pipe_surface_reference(&zsurf, NULL);
187 				pipe_surface_reference(&cbsurf, NULL);
188 			}
189 		}
190 
191 		/* The texture will always be dirty if some layers or samples aren't flushed.
192 		 * I don't think this case occurs often though. */
193 		if (!staging &&
194 		    first_layer == 0 && last_layer == max_layer &&
195 		    first_sample == 0 && last_sample == max_sample) {
196 			texture->dirty_level_mask &= ~(1 << level);
197 		}
198 	}
199 
200 	/* reenable compression in DB_RENDER_CONTROL */
201 	rctx->db_misc_state.flush_depthstencil_through_cb = false;
202 	r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
203 }
204 
r600_blit_decompress_depth_in_place(struct r600_context * rctx,struct r600_texture * texture,bool is_stencil_sampler,unsigned first_level,unsigned last_level,unsigned first_layer,unsigned last_layer)205 static void r600_blit_decompress_depth_in_place(struct r600_context *rctx,
206                                                 struct r600_texture *texture,
207 						bool is_stencil_sampler,
208                                                 unsigned first_level, unsigned last_level,
209                                                 unsigned first_layer, unsigned last_layer)
210 {
211 	struct pipe_surface *zsurf, surf_tmpl = {{0}};
212 	unsigned layer, max_layer, checked_last_layer, level;
213 	unsigned *dirty_level_mask;
214 
215 	/* Enable decompression in DB_RENDER_CONTROL */
216 	if (is_stencil_sampler) {
217 		rctx->db_misc_state.flush_stencil_inplace = true;
218 		dirty_level_mask = &texture->stencil_dirty_level_mask;
219 	} else {
220 		rctx->db_misc_state.flush_depth_inplace = true;
221 		dirty_level_mask = &texture->dirty_level_mask;
222 	}
223 	r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
224 
225 	surf_tmpl.format = texture->resource.b.b.format;
226 
227 	for (level = first_level; level <= last_level; level++) {
228 		if (!(*dirty_level_mask & (1 << level)))
229 			continue;
230 
231 		surf_tmpl.u.tex.level = level;
232 
233 		/* The smaller the mipmap level, the less layers there are
234 		 * as far as 3D textures are concerned. */
235 		max_layer = util_max_layer(&texture->resource.b.b, level);
236 		checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
237 
238 		for (layer = first_layer; layer <= checked_last_layer; layer++) {
239 			surf_tmpl.u.tex.first_layer = layer;
240 			surf_tmpl.u.tex.last_layer = layer;
241 
242 			zsurf = rctx->b.b.create_surface(&rctx->b.b, &texture->resource.b.b, &surf_tmpl);
243 
244 			r600_blitter_begin(&rctx->b.b, R600_DECOMPRESS);
245 			util_blitter_custom_depth_stencil(rctx->blitter, zsurf, NULL, ~0,
246 							  rctx->custom_dsa_flush, 1.0f);
247 			r600_blitter_end(&rctx->b.b);
248 
249 			pipe_surface_reference(&zsurf, NULL);
250 		}
251 
252 		/* The texture will always be dirty if some layers or samples aren't flushed.
253 		 * I don't think this case occurs often though. */
254 		if (first_layer == 0 && last_layer == max_layer) {
255 			*dirty_level_mask &= ~(1 << level);
256 		}
257 	}
258 
259 	/* Disable decompression in DB_RENDER_CONTROL */
260 	rctx->db_misc_state.flush_depth_inplace = false;
261 	rctx->db_misc_state.flush_stencil_inplace = false;
262 	r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
263 }
264 
r600_decompress_depth_textures(struct r600_context * rctx,struct r600_samplerview_state * textures)265 void r600_decompress_depth_textures(struct r600_context *rctx,
266 			       struct r600_samplerview_state *textures)
267 {
268 	unsigned i;
269 	unsigned depth_texture_mask = textures->compressed_depthtex_mask;
270 
271 	while (depth_texture_mask) {
272 		struct pipe_sampler_view *view;
273 		struct r600_pipe_sampler_view *rview;
274 		struct r600_texture *tex;
275 
276 		i = u_bit_scan(&depth_texture_mask);
277 
278 		view = &textures->views[i]->base;
279 		assert(view);
280 		rview = (struct r600_pipe_sampler_view*)view;
281 
282 		tex = (struct r600_texture *)view->texture;
283 		assert(tex->db_compatible);
284 
285 		if (r600_can_sample_zs(tex, rview->is_stencil_sampler)) {
286 			r600_blit_decompress_depth_in_place(rctx, tex,
287 						   rview->is_stencil_sampler,
288 						   view->u.tex.first_level, view->u.tex.last_level,
289 						   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
290 		} else {
291 			r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
292 						   view->u.tex.first_level, view->u.tex.last_level,
293 						   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level),
294 						   0, u_max_sample(&tex->resource.b.b));
295 		}
296 	}
297 }
298 
r600_decompress_depth_images(struct r600_context * rctx,struct r600_image_state * images)299 void r600_decompress_depth_images(struct r600_context *rctx,
300 				  struct r600_image_state *images)
301 {
302 	unsigned i;
303 	unsigned depth_texture_mask = images->compressed_depthtex_mask;
304 
305 	while (depth_texture_mask) {
306 		struct r600_image_view *view;
307 		struct r600_texture *tex;
308 
309 		i = u_bit_scan(&depth_texture_mask);
310 
311 		view = &images->views[i];
312 		assert(view);
313 
314 		tex = (struct r600_texture *)view->base.resource;
315 		assert(tex->db_compatible);
316 
317 		if (r600_can_sample_zs(tex, false)) {
318 			r600_blit_decompress_depth_in_place(rctx, tex,
319 							    false,
320 							    view->base.u.tex.level,
321 							    view->base.u.tex.level,
322 							    0, util_max_layer(&tex->resource.b.b, view->base.u.tex.level));
323 		} else {
324 			r600_blit_decompress_depth(&rctx->b.b, tex, NULL,
325 						   view->base.u.tex.level,
326 						   view->base.u.tex.level,
327 						   0, util_max_layer(&tex->resource.b.b, view->base.u.tex.level),
328 						   0, u_max_sample(&tex->resource.b.b));
329 		}
330 	}
331 }
332 
r600_blit_decompress_color(struct pipe_context * ctx,struct r600_texture * rtex,unsigned first_level,unsigned last_level,unsigned first_layer,unsigned last_layer)333 static void r600_blit_decompress_color(struct pipe_context *ctx,
334 		struct r600_texture *rtex,
335 		unsigned first_level, unsigned last_level,
336 		unsigned first_layer, unsigned last_layer)
337 {
338 	struct r600_context *rctx = (struct r600_context *)ctx;
339 	unsigned layer, level, checked_last_layer, max_layer;
340 
341 	if (!rtex->dirty_level_mask)
342 		return;
343 
344 	for (level = first_level; level <= last_level; level++) {
345 		if (!(rtex->dirty_level_mask & (1 << level)))
346 			continue;
347 
348 		/* The smaller the mipmap level, the less layers there are
349 		 * as far as 3D textures are concerned. */
350 		max_layer = util_max_layer(&rtex->resource.b.b, level);
351 		checked_last_layer = last_layer < max_layer ? last_layer : max_layer;
352 
353 		for (layer = first_layer; layer <= checked_last_layer; layer++) {
354 			struct pipe_surface *cbsurf, surf_tmpl;
355 
356 			surf_tmpl.format = rtex->resource.b.b.format;
357 			surf_tmpl.u.tex.level = level;
358 			surf_tmpl.u.tex.first_layer = layer;
359 			surf_tmpl.u.tex.last_layer = layer;
360 			cbsurf = ctx->create_surface(ctx, &rtex->resource.b.b, &surf_tmpl);
361 
362 			r600_blitter_begin(ctx, R600_DECOMPRESS);
363 			util_blitter_custom_color(rctx->blitter, cbsurf,
364 				rtex->fmask.size ? rctx->custom_blend_decompress : rctx->custom_blend_fastclear);
365 			r600_blitter_end(ctx);
366 
367 			pipe_surface_reference(&cbsurf, NULL);
368 		}
369 
370 		/* The texture will always be dirty if some layers aren't flushed.
371 		 * I don't think this case occurs often though. */
372 		if (first_layer == 0 && last_layer == max_layer) {
373 			rtex->dirty_level_mask &= ~(1 << level);
374 		}
375 	}
376 }
377 
r600_decompress_color_textures(struct r600_context * rctx,struct r600_samplerview_state * textures)378 void r600_decompress_color_textures(struct r600_context *rctx,
379 				    struct r600_samplerview_state *textures)
380 {
381 	unsigned i;
382 	unsigned mask = textures->compressed_colortex_mask;
383 
384 	while (mask) {
385 		struct pipe_sampler_view *view;
386 		struct r600_texture *tex;
387 
388 		i = u_bit_scan(&mask);
389 
390 		view = &textures->views[i]->base;
391 		assert(view);
392 
393 		tex = (struct r600_texture *)view->texture;
394 		assert(tex->cmask.size);
395 
396 		r600_blit_decompress_color(&rctx->b.b, tex,
397 					   view->u.tex.first_level, view->u.tex.last_level,
398 					   0, util_max_layer(&tex->resource.b.b, view->u.tex.first_level));
399 	}
400 }
401 
r600_decompress_color_images(struct r600_context * rctx,struct r600_image_state * images)402 void r600_decompress_color_images(struct r600_context *rctx,
403 				  struct r600_image_state *images)
404 {
405 	unsigned i;
406 	unsigned mask = images->compressed_colortex_mask;
407 
408 	while (mask) {
409 		struct r600_image_view *view;
410 		struct r600_texture *tex;
411 
412 		i = u_bit_scan(&mask);
413 
414 		view = &images->views[i];
415 		assert(view);
416 
417 		tex = (struct r600_texture *)view->base.resource;
418 		assert(tex->cmask.size);
419 
420 		r600_blit_decompress_color(&rctx->b.b, tex,
421 					   view->base.u.tex.level, view->base.u.tex.level,
422 					   view->base.u.tex.first_layer,
423 					   view->base.u.tex.last_layer);
424 	}
425 }
426 
427 /* Helper for decompressing a portion of a color or depth resource before
428  * blitting if any decompression is needed.
429  * The driver doesn't decompress resources automatically while u_blitter is
430  * rendering. */
r600_decompress_subresource(struct pipe_context * ctx,struct pipe_resource * tex,unsigned level,unsigned first_layer,unsigned last_layer)431 static bool r600_decompress_subresource(struct pipe_context *ctx,
432 					struct pipe_resource *tex,
433 					unsigned level,
434 					unsigned first_layer, unsigned last_layer)
435 {
436 	struct r600_context *rctx = (struct r600_context *)ctx;
437 	struct r600_texture *rtex = (struct r600_texture*)tex;
438 
439 	if (rtex->db_compatible) {
440 		if (r600_can_sample_zs(rtex, false)) {
441 			r600_blit_decompress_depth_in_place(rctx, rtex, false,
442 						   level, level,
443 						   first_layer, last_layer);
444 			if (rtex->surface.has_stencil) {
445 				r600_blit_decompress_depth_in_place(rctx, rtex, true,
446 							   level, level,
447 							   first_layer, last_layer);
448 			}
449 		} else {
450 			if (!r600_init_flushed_depth_texture(ctx, tex, NULL))
451 				return false; /* error */
452 
453 			r600_blit_decompress_depth(ctx, rtex, NULL,
454 						   level, level,
455 						   first_layer, last_layer,
456 						   0, u_max_sample(tex));
457 		}
458 	} else if (rtex->cmask.size) {
459 		r600_blit_decompress_color(ctx, rtex, level, level,
460 					   first_layer, last_layer);
461 	}
462 	return true;
463 }
464 
r600_clear(struct pipe_context * ctx,unsigned buffers,const struct pipe_scissor_state * scissor_state,const union pipe_color_union * color,double depth,unsigned stencil)465 static void r600_clear(struct pipe_context *ctx, unsigned buffers,
466 		       const struct pipe_scissor_state *scissor_state,
467 		       const union pipe_color_union *color,
468 		       double depth, unsigned stencil)
469 {
470 	struct r600_context *rctx = (struct r600_context *)ctx;
471 	struct pipe_framebuffer_state *fb = &rctx->framebuffer.state;
472 
473 	if (buffers & PIPE_CLEAR_COLOR && rctx->b.chip_class >= EVERGREEN) {
474 		evergreen_do_fast_color_clear(&rctx->b, fb, &rctx->framebuffer.atom,
475 					      &buffers, NULL, color);
476 		if (!buffers)
477 			return; /* all buffers have been fast cleared */
478 	}
479 
480 	if (buffers & PIPE_CLEAR_COLOR) {
481 		int i;
482 
483 		/* These buffers cannot use fast clear, make sure to disable expansion. */
484 		for (i = 0; i < fb->nr_cbufs; i++) {
485 			struct r600_texture *tex;
486 
487 			/* If not clearing this buffer, skip. */
488 			if (!(buffers & (PIPE_CLEAR_COLOR0 << i)))
489 				continue;
490 
491 			if (!fb->cbufs[i])
492 				continue;
493 
494 			tex = (struct r600_texture *)fb->cbufs[i]->texture;
495 			if (tex->fmask.size == 0)
496 				tex->dirty_level_mask &= ~(1 << fb->cbufs[i]->u.tex.level);
497 		}
498 	}
499 
500 	/* if hyperz enabled just clear hyperz */
501 	if (fb->zsbuf && (buffers & PIPE_CLEAR_DEPTH)) {
502 		struct r600_texture *rtex;
503 		unsigned level = fb->zsbuf->u.tex.level;
504 
505 		rtex = (struct r600_texture*)fb->zsbuf->texture;
506 
507 		/* We can't use hyperz fast clear if each slice of a texture
508 		 * array are clear to different value. To simplify code just
509 		 * disable fast clear for texture array.
510 		 */
511 		if (r600_htile_enabled(rtex, level) &&
512                    fb->zsbuf->u.tex.first_layer == 0 &&
513                    fb->zsbuf->u.tex.last_layer == util_max_layer(&rtex->resource.b.b, level)) {
514 			if (rtex->depth_clear_value != depth) {
515 				rtex->depth_clear_value = depth;
516 				r600_mark_atom_dirty(rctx, &rctx->db_state.atom);
517 			}
518 			rctx->db_misc_state.htile_clear = true;
519 			r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
520 		}
521 	}
522 
523 	r600_blitter_begin(ctx, R600_CLEAR);
524 	util_blitter_clear(rctx->blitter, fb->width, fb->height,
525 			   util_framebuffer_get_num_layers(fb),
526 			   buffers, color, depth, stencil,
527 			   util_framebuffer_get_num_samples(fb) > 1);
528 	r600_blitter_end(ctx);
529 
530 	/* disable fast clear */
531 	if (rctx->db_misc_state.htile_clear) {
532 		rctx->db_misc_state.htile_clear = false;
533 		r600_mark_atom_dirty(rctx, &rctx->db_misc_state.atom);
534 	}
535 }
536 
r600_clear_render_target(struct pipe_context * ctx,struct pipe_surface * dst,const union pipe_color_union * color,unsigned dstx,unsigned dsty,unsigned width,unsigned height,bool render_condition_enabled)537 static void r600_clear_render_target(struct pipe_context *ctx,
538 				     struct pipe_surface *dst,
539 				     const union pipe_color_union *color,
540 				     unsigned dstx, unsigned dsty,
541 				     unsigned width, unsigned height,
542 				     bool render_condition_enabled)
543 {
544 	struct r600_context *rctx = (struct r600_context *)ctx;
545 
546 	r600_blitter_begin(ctx, R600_CLEAR_SURFACE |
547 			   (render_condition_enabled ? 0 : R600_DISABLE_RENDER_COND));
548 	util_blitter_clear_render_target(rctx->blitter, dst, color,
549 					 dstx, dsty, width, height);
550 	r600_blitter_end(ctx);
551 }
552 
r600_clear_depth_stencil(struct pipe_context * ctx,struct pipe_surface * dst,unsigned clear_flags,double depth,unsigned stencil,unsigned dstx,unsigned dsty,unsigned width,unsigned height,bool render_condition_enabled)553 static void r600_clear_depth_stencil(struct pipe_context *ctx,
554 				     struct pipe_surface *dst,
555 				     unsigned clear_flags,
556 				     double depth,
557 				     unsigned stencil,
558 				     unsigned dstx, unsigned dsty,
559 				     unsigned width, unsigned height,
560 				     bool render_condition_enabled)
561 {
562 	struct r600_context *rctx = (struct r600_context *)ctx;
563 
564 	r600_blitter_begin(ctx, R600_CLEAR_SURFACE |
565 			   (render_condition_enabled ? 0 : R600_DISABLE_RENDER_COND));
566 	util_blitter_clear_depth_stencil(rctx->blitter, dst, clear_flags, depth, stencil,
567 					 dstx, dsty, width, height);
568 	r600_blitter_end(ctx);
569 }
570 
r600_copy_buffer(struct pipe_context * ctx,struct pipe_resource * dst,unsigned dstx,struct pipe_resource * src,const struct pipe_box * src_box)571 static void r600_copy_buffer(struct pipe_context *ctx, struct pipe_resource *dst, unsigned dstx,
572 			     struct pipe_resource *src, const struct pipe_box *src_box)
573 {
574 	struct r600_context *rctx = (struct r600_context*)ctx;
575 
576 	if (rctx->screen->b.has_cp_dma) {
577 		r600_cp_dma_copy_buffer(rctx, dst, dstx, src, src_box->x, src_box->width);
578 	}
579 	else if (rctx->screen->b.has_streamout &&
580 		 /* Require 4-byte alignment. */
581 		 dstx % 4 == 0 && src_box->x % 4 == 0 && src_box->width % 4 == 0) {
582 
583 		r600_blitter_begin(ctx, R600_COPY_BUFFER);
584 		util_blitter_copy_buffer(rctx->blitter, dst, dstx, src, src_box->x, src_box->width);
585 		r600_blitter_end(ctx);
586 	} else {
587 		util_resource_copy_region(ctx, dst, 0, dstx, 0, 0, src, 0, src_box);
588 	}
589 }
590 
591 /**
592  * Global buffers are not really resources, they are are actually offsets
593  * into a single global resource (r600_screen::global_pool).  The means
594  * they don't have their own buf handle, so they cannot be passed
595  * to r600_copy_buffer() and must be handled separately.
596  */
r600_copy_global_buffer(struct pipe_context * ctx,struct pipe_resource * dst,unsigned dstx,struct pipe_resource * src,const struct pipe_box * src_box)597 static void r600_copy_global_buffer(struct pipe_context *ctx,
598 				    struct pipe_resource *dst, unsigned
599 				    dstx, struct pipe_resource *src,
600 				    const struct pipe_box *src_box)
601 {
602 	struct r600_context *rctx = (struct r600_context*)ctx;
603 	struct compute_memory_pool *pool = rctx->screen->global_pool;
604 	struct pipe_box new_src_box = *src_box;
605 
606 	if (src->bind & PIPE_BIND_GLOBAL) {
607 		struct r600_resource_global *rsrc =
608 			(struct r600_resource_global *)src;
609 		struct compute_memory_item *item = rsrc->chunk;
610 
611 		if (is_item_in_pool(item)) {
612 			new_src_box.x += 4 * item->start_in_dw;
613 			src = (struct pipe_resource *)pool->bo;
614 		} else {
615 			if (item->real_buffer == NULL) {
616 				item->real_buffer =
617 					r600_compute_buffer_alloc_vram(pool->screen,
618 								       item->size_in_dw * 4);
619 			}
620 			src = (struct pipe_resource*)item->real_buffer;
621 		}
622 	}
623 	if (dst->bind & PIPE_BIND_GLOBAL) {
624 		struct r600_resource_global *rdst =
625 			(struct r600_resource_global *)dst;
626 		struct compute_memory_item *item = rdst->chunk;
627 
628 		if (is_item_in_pool(item)) {
629 			dstx += 4 * item->start_in_dw;
630 			dst = (struct pipe_resource *)pool->bo;
631 		} else {
632 			if (item->real_buffer == NULL) {
633 				item->real_buffer =
634 					r600_compute_buffer_alloc_vram(pool->screen,
635 								       item->size_in_dw * 4);
636 			}
637 			dst = (struct pipe_resource*)item->real_buffer;
638 		}
639 	}
640 
641 	r600_copy_buffer(ctx, dst, dstx, src, &new_src_box);
642 }
643 
r600_clear_buffer(struct pipe_context * ctx,struct pipe_resource * dst,uint64_t offset,uint64_t size,unsigned value,enum r600_coherency coher)644 static void r600_clear_buffer(struct pipe_context *ctx, struct pipe_resource *dst,
645 			      uint64_t offset, uint64_t size, unsigned value,
646 			      enum r600_coherency coher)
647 {
648 	struct r600_context *rctx = (struct r600_context*)ctx;
649 
650 	if (rctx->screen->b.has_cp_dma &&
651 	    rctx->b.chip_class >= EVERGREEN &&
652 	    offset % 4 == 0 && size % 4 == 0) {
653 		evergreen_cp_dma_clear_buffer(rctx, dst, offset, size, value, coher);
654 	} else if (rctx->screen->b.has_streamout && offset % 4 == 0 && size % 4 == 0) {
655 		union pipe_color_union clear_value;
656 		clear_value.ui[0] = value;
657 
658 		r600_blitter_begin(ctx, R600_DISABLE_RENDER_COND);
659 		util_blitter_clear_buffer(rctx->blitter, dst, offset, size,
660 					  1, &clear_value);
661 		r600_blitter_end(ctx);
662 	} else {
663 		uint32_t *map = r600_buffer_map_sync_with_rings(&rctx->b, r600_resource(dst),
664 								 PIPE_MAP_WRITE);
665 		map += offset / 4;
666 		size /= 4;
667 		for (unsigned i = 0; i < size; i++)
668 			*map++ = value;
669 	}
670 }
671 
r600_resource_copy_region(struct pipe_context * ctx,struct pipe_resource * dst,unsigned dst_level,unsigned dstx,unsigned dsty,unsigned dstz,struct pipe_resource * src,unsigned src_level,const struct pipe_box * src_box)672 void r600_resource_copy_region(struct pipe_context *ctx,
673 			       struct pipe_resource *dst,
674 			       unsigned dst_level,
675 			       unsigned dstx, unsigned dsty, unsigned dstz,
676 			       struct pipe_resource *src,
677 			       unsigned src_level,
678 			       const struct pipe_box *src_box)
679 {
680 	struct r600_context *rctx = (struct r600_context *)ctx;
681 	struct pipe_surface *dst_view, dst_templ;
682 	struct pipe_sampler_view src_templ, *src_view;
683 	unsigned dst_width, dst_height, src_width0, src_height0, src_widthFL, src_heightFL;
684 	unsigned src_force_level = 0;
685 	struct pipe_box sbox, dstbox;
686 
687 	/* Handle buffers first. */
688 	if (dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) {
689 		if ((src->bind & PIPE_BIND_GLOBAL) ||
690 					(dst->bind & PIPE_BIND_GLOBAL)) {
691 			r600_copy_global_buffer(ctx, dst, dstx, src, src_box);
692 		} else {
693 			r600_copy_buffer(ctx, dst, dstx, src, src_box);
694 		}
695 		return;
696 	}
697 
698 	assert(u_max_sample(dst) == u_max_sample(src));
699 
700 	/* The driver doesn't decompress resources automatically while
701 	 * u_blitter is rendering. */
702 	if (!r600_decompress_subresource(ctx, src, src_level,
703 					 src_box->z, src_box->z + src_box->depth - 1)) {
704 		return; /* error */
705 	}
706 
707 	dst_width = u_minify(dst->width0, dst_level);
708         dst_height = u_minify(dst->height0, dst_level);
709 	src_width0 = src->width0;
710 	src_height0 = src->height0;
711         src_widthFL = u_minify(src->width0, src_level);
712         src_heightFL = u_minify(src->height0, src_level);
713 
714 	util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
715 	util_blitter_default_src_texture(rctx->blitter, &src_templ, src, src_level);
716 
717 	if (util_format_is_compressed(src->format) ||
718 	    util_format_is_compressed(dst->format)) {
719 		unsigned blocksize = util_format_get_blocksize(src->format);
720 
721 		if (blocksize == 8)
722 			src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT; /* 64-bit block */
723 		else
724 			src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT; /* 128-bit block */
725 		dst_templ.format = src_templ.format;
726 
727 		dst_width = util_format_get_nblocksx(dst->format, dst_width);
728 		dst_height = util_format_get_nblocksy(dst->format, dst_height);
729 		src_width0 = util_format_get_nblocksx(src->format, src_width0);
730 		src_height0 = util_format_get_nblocksy(src->format, src_height0);
731 		src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
732 		src_heightFL = util_format_get_nblocksy(src->format, src_heightFL);
733 
734 		dstx = util_format_get_nblocksx(dst->format, dstx);
735 		dsty = util_format_get_nblocksy(dst->format, dsty);
736 
737 		sbox.x = util_format_get_nblocksx(src->format, src_box->x);
738 		sbox.y = util_format_get_nblocksy(src->format, src_box->y);
739 		sbox.z = src_box->z;
740 		sbox.width = util_format_get_nblocksx(src->format, src_box->width);
741 		sbox.height = util_format_get_nblocksy(src->format, src_box->height);
742 		sbox.depth = src_box->depth;
743 		src_box = &sbox;
744 
745 		src_force_level = src_level;
746 	} else if (!util_blitter_is_copy_supported(rctx->blitter, dst, src)) {
747 		if (util_format_is_subsampled_422(src->format)) {
748 
749 			src_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
750 			dst_templ.format = PIPE_FORMAT_R8G8B8A8_UINT;
751 
752 			dst_width = util_format_get_nblocksx(dst->format, dst_width);
753 			src_width0 = util_format_get_nblocksx(src->format, src_width0);
754 			src_widthFL = util_format_get_nblocksx(src->format, src_widthFL);
755 
756 			dstx = util_format_get_nblocksx(dst->format, dstx);
757 
758 			sbox = *src_box;
759 			sbox.x = util_format_get_nblocksx(src->format, src_box->x);
760 			sbox.width = util_format_get_nblocksx(src->format, src_box->width);
761 			src_box = &sbox;
762 		} else {
763 			unsigned blocksize = util_format_get_blocksize(src->format);
764 
765 			switch (blocksize) {
766 			case 1:
767 				dst_templ.format = PIPE_FORMAT_R8_UNORM;
768 				src_templ.format = PIPE_FORMAT_R8_UNORM;
769 				break;
770 			case 2:
771 				dst_templ.format = PIPE_FORMAT_R8G8_UNORM;
772 				src_templ.format = PIPE_FORMAT_R8G8_UNORM;
773 				break;
774 			case 4:
775 				dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
776 				src_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
777 				break;
778 			case 8:
779 				dst_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
780 				src_templ.format = PIPE_FORMAT_R16G16B16A16_UINT;
781 				break;
782 			case 16:
783 				dst_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
784 				src_templ.format = PIPE_FORMAT_R32G32B32A32_UINT;
785 				break;
786 			default:
787 				fprintf(stderr, "Unhandled format %s with blocksize %u\n",
788 					util_format_short_name(src->format), blocksize);
789 				assert(0);
790 			}
791 		}
792 	}
793 
794 	dst_view = r600_create_surface_custom(ctx, dst, &dst_templ,
795 					      /* we don't care about these two for r600g */
796 					      dst->width0, dst->height0,
797 					      dst_width, dst_height);
798 
799 	if (rctx->b.chip_class >= EVERGREEN) {
800 		src_view = evergreen_create_sampler_view_custom(ctx, src, &src_templ,
801 								src_width0, src_height0,
802 								src_force_level);
803 	} else {
804 		src_view = r600_create_sampler_view_custom(ctx, src, &src_templ,
805 							   src_widthFL, src_heightFL);
806 	}
807 
808         u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
809                  abs(src_box->depth), &dstbox);
810 
811 	/* Copy. */
812 	r600_blitter_begin(ctx, R600_COPY_TEXTURE);
813 	util_blitter_blit_generic(rctx->blitter, dst_view, &dstbox,
814 				  src_view, src_box, src_width0, src_height0,
815 				  PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
816 				  FALSE, FALSE);
817 	r600_blitter_end(ctx);
818 
819 	pipe_surface_reference(&dst_view, NULL);
820 	pipe_sampler_view_reference(&src_view, NULL);
821 }
822 
do_hardware_msaa_resolve(struct pipe_context * ctx,const struct pipe_blit_info * info)823 static bool do_hardware_msaa_resolve(struct pipe_context *ctx,
824 				     const struct pipe_blit_info *info)
825 {
826 	struct r600_context *rctx = (struct r600_context*)ctx;
827 	struct r600_texture *dst = (struct r600_texture*)info->dst.resource;
828 	unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
829 	unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
830 	enum pipe_format format = info->src.format;
831 	unsigned sample_mask =
832 		rctx->b.chip_class == CAYMAN ? ~0 :
833 		((1ull << MAX2(1, info->src.resource->nr_samples)) - 1);
834 	struct pipe_resource *tmp, templ;
835 	struct pipe_blit_info blit;
836 
837 	/* Check basic requirements for hw resolve. */
838 	if (!(info->src.resource->nr_samples > 1 &&
839 	      info->dst.resource->nr_samples <= 1 &&
840 	      !util_format_is_pure_integer(format) &&
841 	      !util_format_is_depth_or_stencil(format) &&
842 	      util_max_layer(info->src.resource, 0) == 0))
843 		return false;
844 
845 	/* Check the remaining requirements for hw resolve. */
846 	if (util_max_layer(info->dst.resource, info->dst.level) == 0 &&
847 	    util_is_format_compatible(util_format_description(info->src.format),
848 				      util_format_description(info->dst.format)) &&
849 	    !info->scissor_enable &&
850 	    (info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA &&
851 	    dst_width == info->src.resource->width0 &&
852 	    dst_height == info->src.resource->height0 &&
853 	    info->dst.box.x == 0 &&
854 	    info->dst.box.y == 0 &&
855 	    info->dst.box.width == dst_width &&
856 	    info->dst.box.height == dst_height &&
857 	    info->dst.box.depth == 1 &&
858 	    info->src.box.x == 0 &&
859 	    info->src.box.y == 0 &&
860 	    info->src.box.width == dst_width &&
861 	    info->src.box.height == dst_height &&
862 	    info->src.box.depth == 1 &&
863 	    dst->surface.u.legacy.level[info->dst.level].mode >= RADEON_SURF_MODE_1D &&
864 	    (!dst->cmask.size || !dst->dirty_level_mask) /* dst cannot be fast-cleared */) {
865 		r600_blitter_begin(ctx, R600_COLOR_RESOLVE |
866 				   (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
867 		util_blitter_custom_resolve_color(rctx->blitter,
868 						  info->dst.resource, info->dst.level,
869 						  info->dst.box.z,
870 						  info->src.resource, info->src.box.z,
871 						  sample_mask, rctx->custom_blend_resolve,
872 						  format);
873 		r600_blitter_end(ctx);
874 		return true;
875 	}
876 
877 	/* Shader-based resolve is VERY SLOW. Instead, resolve into
878 	 * a temporary texture and blit.
879 	 */
880 	memset(&templ, 0, sizeof(templ));
881 	templ.target = PIPE_TEXTURE_2D;
882 	templ.format = info->src.resource->format;
883 	templ.width0 = info->src.resource->width0;
884 	templ.height0 = info->src.resource->height0;
885 	templ.depth0 = 1;
886 	templ.array_size = 1;
887 	templ.usage = PIPE_USAGE_DEFAULT;
888 	templ.flags = R600_RESOURCE_FLAG_FORCE_TILING;
889 
890 	tmp = ctx->screen->resource_create(ctx->screen, &templ);
891 	if (!tmp)
892 		return false;
893 
894 	/* resolve */
895 	r600_blitter_begin(ctx, R600_COLOR_RESOLVE |
896 			   (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
897 	util_blitter_custom_resolve_color(rctx->blitter, tmp, 0, 0,
898 					  info->src.resource, info->src.box.z,
899 					  sample_mask, rctx->custom_blend_resolve,
900 					  format);
901 	r600_blitter_end(ctx);
902 
903 	/* blit */
904 	blit = *info;
905 	blit.src.resource = tmp;
906 	blit.src.box.z = 0;
907 
908 	r600_blitter_begin(ctx, R600_BLIT |
909 			   (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
910 	util_blitter_blit(rctx->blitter, &blit);
911 	r600_blitter_end(ctx);
912 
913 	pipe_resource_reference(&tmp, NULL);
914 	return true;
915 }
916 
r600_blit(struct pipe_context * ctx,const struct pipe_blit_info * info)917 static void r600_blit(struct pipe_context *ctx,
918                       const struct pipe_blit_info *info)
919 {
920 	struct r600_context *rctx = (struct r600_context*)ctx;
921 	struct r600_texture *rdst = (struct r600_texture *)info->dst.resource;
922 
923 	if (do_hardware_msaa_resolve(ctx, info)) {
924 		return;
925 	}
926 
927 	/* Using SDMA for copying to a linear texture in GTT is much faster.
928 	 * This improves DRI PRIME performance.
929 	 *
930 	 * resource_copy_region can't do this yet, because dma_copy calls it
931 	 * on failure (recursion).
932 	 */
933 	if (rdst->surface.u.legacy.level[info->dst.level].mode ==
934 	    RADEON_SURF_MODE_LINEAR_ALIGNED &&
935 	    rctx->b.dma_copy &&
936 	    util_can_blit_via_copy_region(info, false)) {
937 		rctx->b.dma_copy(ctx, info->dst.resource, info->dst.level,
938 				 info->dst.box.x, info->dst.box.y,
939 				 info->dst.box.z,
940 				 info->src.resource, info->src.level,
941 				 &info->src.box);
942 		return;
943 	}
944 
945 	assert(util_blitter_is_blit_supported(rctx->blitter, info));
946 
947 	/* The driver doesn't decompress resources automatically while
948 	 * u_blitter is rendering. */
949 	if (!r600_decompress_subresource(ctx, info->src.resource, info->src.level,
950 					 info->src.box.z,
951 					 info->src.box.z + info->src.box.depth - 1)) {
952 		return; /* error */
953 	}
954 
955 	if (rctx->screen->b.debug_flags & DBG_FORCE_DMA &&
956 	    util_try_blit_via_copy_region(ctx, info))
957 		return;
958 
959 	r600_blitter_begin(ctx, R600_BLIT |
960 			   (info->render_condition_enable ? 0 : R600_DISABLE_RENDER_COND));
961 	util_blitter_blit(rctx->blitter, info);
962 	r600_blitter_end(ctx);
963 }
964 
r600_flush_resource(struct pipe_context * ctx,struct pipe_resource * res)965 static void r600_flush_resource(struct pipe_context *ctx,
966 				struct pipe_resource *res)
967 {
968 	struct r600_texture *rtex = (struct r600_texture*)res;
969 
970 	assert(res->target != PIPE_BUFFER);
971 
972 	if (!rtex->is_depth && rtex->cmask.size) {
973 		r600_blit_decompress_color(ctx, rtex, 0, res->last_level,
974 					   0, util_max_layer(res, 0));
975 	}
976 }
977 
r600_init_blit_functions(struct r600_context * rctx)978 void r600_init_blit_functions(struct r600_context *rctx)
979 {
980 	rctx->b.b.clear = r600_clear;
981 	rctx->b.b.clear_render_target = r600_clear_render_target;
982 	rctx->b.b.clear_depth_stencil = r600_clear_depth_stencil;
983 	rctx->b.b.resource_copy_region = r600_resource_copy_region;
984 	rctx->b.b.blit = r600_blit;
985 	rctx->b.b.flush_resource = r600_flush_resource;
986 	rctx->b.clear_buffer = r600_clear_buffer;
987 	rctx->b.blit_decompress_depth = r600_blit_decompress_depth;
988 }
989