1 /*
2 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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 "r300_context.h"
24 #include "r300_emit.h"
25 #include "r300_texture.h"
26 #include "r300_reg.h"
27
28 #include "util/format/u_format.h"
29 #include "util/half_float.h"
30 #include "util/u_pack_color.h"
31 #include "util/u_surface.h"
32
33 enum r300_blitter_op /* bitmask */
34 {
35 R300_STOP_QUERY = 1,
36 R300_SAVE_TEXTURES = 2,
37 R300_SAVE_FRAMEBUFFER = 4,
38 R300_IGNORE_RENDER_COND = 8,
39
40 R300_CLEAR = R300_STOP_QUERY,
41
42 R300_CLEAR_SURFACE = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER,
43
44 R300_COPY = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
45 R300_SAVE_TEXTURES | R300_IGNORE_RENDER_COND,
46
47 R300_BLIT = R300_STOP_QUERY | R300_SAVE_FRAMEBUFFER |
48 R300_SAVE_TEXTURES,
49
50 R300_DECOMPRESS = R300_STOP_QUERY | R300_IGNORE_RENDER_COND,
51 };
52
r300_blitter_begin(struct r300_context * r300,enum r300_blitter_op op)53 static void r300_blitter_begin(struct r300_context* r300, enum r300_blitter_op op)
54 {
55 if ((op & R300_STOP_QUERY) && r300->query_current) {
56 r300->blitter_saved_query = r300->query_current;
57 r300_stop_query(r300);
58 }
59
60 /* Yeah we have to save all those states to ensure the blitter operation
61 * is really transparent. The states will be restored by the blitter once
62 * copying is done. */
63 util_blitter_save_blend(r300->blitter, r300->blend_state.state);
64 util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
65 util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
66 util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
67 util_blitter_save_fragment_shader(r300->blitter, r300->fs.state);
68 util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
69 util_blitter_save_viewport(r300->blitter, &r300->viewport);
70 util_blitter_save_scissor(r300->blitter, r300->scissor_state.state);
71 util_blitter_save_sample_mask(r300->blitter, *(unsigned*)r300->sample_mask.state, 0);
72 util_blitter_save_vertex_buffers(r300->blitter, r300->vertex_buffer,
73 r300->nr_vertex_buffers);
74 util_blitter_save_vertex_elements(r300->blitter, r300->velems);
75
76 struct pipe_constant_buffer cb = {
77 /* r300 doesn't use the size for FS at all. The shader determines it.
78 * Set something for blitter.
79 */
80 .buffer_size = 4,
81 .user_buffer = ((struct r300_constant_buffer*)r300->fs_constants.state)->ptr,
82 };
83 util_blitter_save_fragment_constant_buffer_slot(r300->blitter, &cb);
84
85 if (op & R300_SAVE_FRAMEBUFFER) {
86 util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
87 }
88
89 if (op & R300_SAVE_TEXTURES) {
90 struct r300_textures_state* state =
91 (struct r300_textures_state*)r300->textures_state.state;
92
93 util_blitter_save_fragment_sampler_states(
94 r300->blitter, state->sampler_state_count,
95 (void**)state->sampler_states);
96
97 util_blitter_save_fragment_sampler_views(
98 r300->blitter, state->sampler_view_count,
99 (struct pipe_sampler_view**)state->sampler_views);
100 }
101
102 if (op & R300_IGNORE_RENDER_COND) {
103 /* Save the flag. */
104 r300->blitter_saved_skip_rendering = r300->skip_rendering+1;
105 r300->skip_rendering = false;
106 } else {
107 r300->blitter_saved_skip_rendering = 0;
108 }
109 }
110
r300_blitter_end(struct r300_context * r300)111 static void r300_blitter_end(struct r300_context *r300)
112 {
113 if (r300->blitter_saved_query) {
114 r300_resume_query(r300, r300->blitter_saved_query);
115 r300->blitter_saved_query = NULL;
116 }
117
118 if (r300->blitter_saved_skip_rendering) {
119 /* Restore the flag. */
120 r300->skip_rendering = r300->blitter_saved_skip_rendering-1;
121 }
122 }
123
r300_depth_clear_cb_value(enum pipe_format format,const float * rgba)124 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
125 const float* rgba)
126 {
127 union util_color uc;
128 util_pack_color(rgba, format, &uc);
129
130 if (util_format_get_blocksizebits(format) == 32)
131 return uc.ui[0];
132 else
133 return uc.us | (uc.us << 16);
134 }
135
r300_cbzb_clear_allowed(struct r300_context * r300,unsigned clear_buffers)136 static bool r300_cbzb_clear_allowed(struct r300_context *r300,
137 unsigned clear_buffers)
138 {
139 struct pipe_framebuffer_state *fb =
140 (struct pipe_framebuffer_state*)r300->fb_state.state;
141
142 /* Only color clear allowed, and only one colorbuffer. */
143 if ((clear_buffers & ~PIPE_CLEAR_COLOR) != 0 || fb->nr_cbufs != 1 || !fb->cbufs[0])
144 return false;
145
146 return r300_surface(fb->cbufs[0])->cbzb_allowed;
147 }
148
r300_fast_zclear_allowed(struct r300_context * r300,unsigned clear_buffers)149 static bool r300_fast_zclear_allowed(struct r300_context *r300,
150 unsigned clear_buffers)
151 {
152 struct pipe_framebuffer_state *fb =
153 (struct pipe_framebuffer_state*)r300->fb_state.state;
154
155 return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level] != 0;
156 }
157
r300_hiz_clear_allowed(struct r300_context * r300)158 static bool r300_hiz_clear_allowed(struct r300_context *r300)
159 {
160 struct pipe_framebuffer_state *fb =
161 (struct pipe_framebuffer_state*)r300->fb_state.state;
162
163 return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level] != 0;
164 }
165
r300_depth_clear_value(enum pipe_format format,double depth,unsigned stencil)166 static uint32_t r300_depth_clear_value(enum pipe_format format,
167 double depth, unsigned stencil)
168 {
169 switch (format) {
170 case PIPE_FORMAT_Z16_UNORM:
171 case PIPE_FORMAT_X8Z24_UNORM:
172 return util_pack_z(format, depth);
173
174 case PIPE_FORMAT_S8_UINT_Z24_UNORM:
175 return util_pack_z_stencil(format, depth, stencil);
176
177 default:
178 assert(0);
179 return 0;
180 }
181 }
182
r300_hiz_clear_value(double depth)183 static uint32_t r300_hiz_clear_value(double depth)
184 {
185 uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
186 assert(r <= 255);
187 return r | (r << 8) | (r << 16) | (r << 24);
188 }
189
r300_set_clear_color(struct r300_context * r300,const union pipe_color_union * color)190 static void r300_set_clear_color(struct r300_context *r300,
191 const union pipe_color_union *color)
192 {
193 struct pipe_framebuffer_state *fb =
194 (struct pipe_framebuffer_state*)r300->fb_state.state;
195 union util_color uc;
196
197 memset(&uc, 0, sizeof(uc));
198 util_pack_color(color->f, fb->cbufs[0]->format, &uc);
199
200 if (fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
201 fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16X16_FLOAT) {
202 /* (0,1,2,3) maps to (B,G,R,A) */
203 r300->color_clear_value_gb = uc.h[0] | ((uint32_t)uc.h[1] << 16);
204 r300->color_clear_value_ar = uc.h[2] | ((uint32_t)uc.h[3] << 16);
205 } else {
206 r300->color_clear_value = uc.ui[0];
207 }
208 }
209
210 DEBUG_GET_ONCE_BOOL_OPTION(hyperz, "RADEON_HYPERZ", false)
211
212 /* Clear currently bound buffers. */
r300_clear(struct pipe_context * pipe,unsigned buffers,const struct pipe_scissor_state * scissor_state,const union pipe_color_union * color,double depth,unsigned stencil)213 static void r300_clear(struct pipe_context* pipe,
214 unsigned buffers,
215 const struct pipe_scissor_state *scissor_state,
216 const union pipe_color_union *color,
217 double depth,
218 unsigned stencil)
219 {
220 /* My notes about Zbuffer compression:
221 *
222 * 1) The zbuffer must be micro-tiled and whole microtiles must be
223 * written if compression is enabled. If microtiling is disabled,
224 * it locks up.
225 *
226 * 2) There is ZMASK RAM which contains a compressed zbuffer.
227 * Each dword of the Z Mask contains compression information
228 * for 16 4x4 pixel tiles, that is 2 bits for each tile.
229 * On chips with 2 Z pipes, every other dword maps to a different
230 * pipe. On newer chipsets, there is a new compression mode
231 * with 8x8 pixel tiles per 2 bits.
232 *
233 * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
234 * it should look in the ZMASK RAM first before fetching from a real
235 * zbuffer.
236 *
237 * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
238 * during zbuffer reads instead of the value that is actually stored
239 * in the zbuffer memory. A pixel is in a cleared state when its ZMASK
240 * is equal to 0. Therefore, if you clear ZMASK with zeros, you may
241 * leave the zbuffer memory uninitialized, but then you must enable
242 * compression, so that the ZMASK RAM is actually used.
243 *
244 * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
245 * during zbuffer updates. A special decompressing operation should be
246 * used to fully decompress a zbuffer, which basically just stores all
247 * compressed tiles in ZMASK to the zbuffer memory.
248 *
249 * 6) For a 16-bit zbuffer, compression causes a hung with one or
250 * two samples and should not be used.
251 *
252 * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
253 * to avoid needless decompression.
254 *
255 * 8) Fastfill must not be used if reading of compressed Z data is disabled
256 * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
257 * i.e. it cannot be used to compress the zbuffer.
258 *
259 * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
260 *
261 * - Marek
262 */
263
264 struct r300_context* r300 = r300_context(pipe);
265 struct pipe_framebuffer_state *fb =
266 (struct pipe_framebuffer_state*)r300->fb_state.state;
267 struct r300_hyperz_state *hyperz =
268 (struct r300_hyperz_state*)r300->hyperz_state.state;
269 uint32_t width = fb->width;
270 uint32_t height = fb->height;
271 uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
272
273 /* Use fast Z clear.
274 * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
275 if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
276 bool zmask_clear, hiz_clear;
277
278 /* If both depth and stencil are present, they must be cleared together. */
279 if (fb->zsbuf->texture->format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
280 (buffers & PIPE_CLEAR_DEPTHSTENCIL) != PIPE_CLEAR_DEPTHSTENCIL) {
281 zmask_clear = false;
282 hiz_clear = false;
283 } else {
284 zmask_clear = r300_fast_zclear_allowed(r300, buffers);
285 hiz_clear = r300_hiz_clear_allowed(r300);
286 }
287
288 /* If we need Hyper-Z. */
289 if (zmask_clear || hiz_clear) {
290 /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
291 if (!r300->hyperz_enabled &&
292 (r300->screen->caps.is_r500 || debug_get_option_hyperz())) {
293 r300->hyperz_enabled =
294 r300->rws->cs_request_feature(&r300->cs,
295 RADEON_FID_R300_HYPERZ_ACCESS,
296 true);
297 if (r300->hyperz_enabled) {
298 /* Need to emit HyperZ buffer regs for the first time. */
299 r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
300 }
301 }
302
303 /* Setup Hyper-Z clears. */
304 if (r300->hyperz_enabled) {
305 if (zmask_clear) {
306 hyperz_dcv = hyperz->zb_depthclearvalue =
307 r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
308
309 r300_mark_atom_dirty(r300, &r300->zmask_clear);
310 r300_mark_atom_dirty(r300, &r300->gpu_flush);
311 buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
312 }
313
314 if (hiz_clear) {
315 r300->hiz_clear_value = r300_hiz_clear_value(depth);
316 r300_mark_atom_dirty(r300, &r300->hiz_clear);
317 r300_mark_atom_dirty(r300, &r300->gpu_flush);
318 }
319 r300->num_z_clears++;
320 }
321 }
322 }
323
324 /* Use fast color clear for an AA colorbuffer.
325 * The CMASK is shared between all colorbuffers, so we use it
326 * if there is only one colorbuffer bound. */
327 if ((buffers & PIPE_CLEAR_COLOR) && fb->nr_cbufs == 1 && fb->cbufs[0] &&
328 r300_resource(fb->cbufs[0]->texture)->tex.cmask_dwords) {
329 /* Try to obtain the access to the CMASK if we don't have one. */
330 if (!r300->cmask_access) {
331 r300->cmask_access =
332 r300->rws->cs_request_feature(&r300->cs,
333 RADEON_FID_R300_CMASK_ACCESS,
334 true);
335 }
336
337 /* Setup the clear. */
338 if (r300->cmask_access) {
339 /* Pair the resource with the CMASK to avoid other resources
340 * accessing it. */
341 if (!r300->screen->cmask_resource) {
342 mtx_lock(&r300->screen->cmask_mutex);
343 /* Double checking (first unlocked, then locked). */
344 if (!r300->screen->cmask_resource) {
345 /* Don't reference this, so that the texture can be
346 * destroyed while set in cmask_resource.
347 * Then in texture_destroy, we set cmask_resource to NULL. */
348 r300->screen->cmask_resource = fb->cbufs[0]->texture;
349 }
350 mtx_unlock(&r300->screen->cmask_mutex);
351 }
352
353 if (r300->screen->cmask_resource == fb->cbufs[0]->texture) {
354 r300_set_clear_color(r300, color);
355 r300_mark_atom_dirty(r300, &r300->cmask_clear);
356 r300_mark_atom_dirty(r300, &r300->gpu_flush);
357 buffers &= ~PIPE_CLEAR_COLOR;
358 }
359 }
360 }
361 /* Enable CBZB clear. */
362 else if (r300_cbzb_clear_allowed(r300, buffers)) {
363 struct r300_surface *surf = r300_surface(fb->cbufs[0]);
364
365 hyperz->zb_depthclearvalue =
366 r300_depth_clear_cb_value(surf->base.format, color->f);
367
368 width = surf->cbzb_width;
369 height = surf->cbzb_height;
370
371 r300->cbzb_clear = true;
372 r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
373 }
374
375 /* Clear. */
376 if (buffers) {
377 /* Clear using the blitter. */
378 r300_blitter_begin(r300, R300_CLEAR);
379 util_blitter_clear(r300->blitter, width, height, 1,
380 buffers, color, depth, stencil,
381 util_framebuffer_get_num_samples(fb) > 1);
382 r300_blitter_end(r300);
383 } else if (r300->zmask_clear.dirty ||
384 r300->hiz_clear.dirty ||
385 r300->cmask_clear.dirty) {
386 /* Just clear zmask and hiz now, this does not use the standard draw
387 * procedure. */
388 /* Calculate zmask_clear and hiz_clear atom sizes. */
389 unsigned dwords =
390 r300->gpu_flush.size +
391 (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
392 (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
393 (r300->cmask_clear.dirty ? r300->cmask_clear.size : 0) +
394 r300_get_num_cs_end_dwords(r300);
395
396 /* Reserve CS space. */
397 if (!r300->rws->cs_check_space(&r300->cs, dwords)) {
398 r300_flush(&r300->context, PIPE_FLUSH_ASYNC, NULL);
399 }
400
401 /* Emit clear packets. */
402 r300_emit_gpu_flush(r300, r300->gpu_flush.size, r300->gpu_flush.state);
403 r300->gpu_flush.dirty = false;
404
405 if (r300->zmask_clear.dirty) {
406 r300_emit_zmask_clear(r300, r300->zmask_clear.size,
407 r300->zmask_clear.state);
408 r300->zmask_clear.dirty = false;
409 }
410 if (r300->hiz_clear.dirty) {
411 r300_emit_hiz_clear(r300, r300->hiz_clear.size,
412 r300->hiz_clear.state);
413 r300->hiz_clear.dirty = false;
414 }
415 if (r300->cmask_clear.dirty) {
416 r300_emit_cmask_clear(r300, r300->cmask_clear.size,
417 r300->cmask_clear.state);
418 r300->cmask_clear.dirty = false;
419 }
420 } else {
421 assert(0);
422 }
423
424 /* Disable CBZB clear. */
425 if (r300->cbzb_clear) {
426 r300->cbzb_clear = false;
427 hyperz->zb_depthclearvalue = hyperz_dcv;
428 r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
429 }
430
431 /* Enable fastfill and/or hiz.
432 *
433 * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
434 * looks if zmask/hiz is in use and programs hardware accordingly. */
435 if (r300->zmask_in_use || r300->hiz_in_use) {
436 r300_mark_atom_dirty(r300, &r300->hyperz_state);
437 }
438 }
439
440 /* Clear a region of a color surface to a constant value. */
r300_clear_render_target(struct pipe_context * pipe,struct pipe_surface * dst,const union pipe_color_union * color,unsigned dstx,unsigned dsty,unsigned width,unsigned height,bool render_condition_enabled)441 static void r300_clear_render_target(struct pipe_context *pipe,
442 struct pipe_surface *dst,
443 const union pipe_color_union *color,
444 unsigned dstx, unsigned dsty,
445 unsigned width, unsigned height,
446 bool render_condition_enabled)
447 {
448 struct r300_context *r300 = r300_context(pipe);
449
450 r300_blitter_begin(r300, R300_CLEAR_SURFACE |
451 (render_condition_enabled ? 0 : R300_IGNORE_RENDER_COND));
452 util_blitter_clear_render_target(r300->blitter, dst, color,
453 dstx, dsty, width, height);
454 r300_blitter_end(r300);
455 }
456
457 /* Clear a region of a depth stencil surface. */
r300_clear_depth_stencil(struct pipe_context * pipe,struct pipe_surface * dst,unsigned clear_flags,double depth,unsigned stencil,unsigned dstx,unsigned dsty,unsigned width,unsigned height,bool render_condition_enabled)458 static void r300_clear_depth_stencil(struct pipe_context *pipe,
459 struct pipe_surface *dst,
460 unsigned clear_flags,
461 double depth,
462 unsigned stencil,
463 unsigned dstx, unsigned dsty,
464 unsigned width, unsigned height,
465 bool render_condition_enabled)
466 {
467 struct r300_context *r300 = r300_context(pipe);
468 struct pipe_framebuffer_state *fb =
469 (struct pipe_framebuffer_state*)r300->fb_state.state;
470
471 if (r300->zmask_in_use && !r300->locked_zbuffer) {
472 if (fb->zsbuf->texture == dst->texture) {
473 r300_decompress_zmask(r300);
474 }
475 }
476
477 /* XXX Do not decompress ZMask of the currently-set zbuffer. */
478 r300_blitter_begin(r300, R300_CLEAR_SURFACE |
479 (render_condition_enabled ? 0 : R300_IGNORE_RENDER_COND));
480 util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
481 dstx, dsty, width, height);
482 r300_blitter_end(r300);
483 }
484
r300_decompress_zmask(struct r300_context * r300)485 void r300_decompress_zmask(struct r300_context *r300)
486 {
487 struct pipe_framebuffer_state *fb =
488 (struct pipe_framebuffer_state*)r300->fb_state.state;
489
490 if (!r300->zmask_in_use || r300->locked_zbuffer)
491 return;
492
493 r300->zmask_decompress = true;
494 r300_mark_atom_dirty(r300, &r300->hyperz_state);
495
496 r300_blitter_begin(r300, R300_DECOMPRESS);
497 util_blitter_custom_clear_depth(r300->blitter, fb->width, fb->height, 0,
498 r300->dsa_decompress_zmask);
499 r300_blitter_end(r300);
500
501 r300->zmask_decompress = false;
502 r300->zmask_in_use = false;
503 r300_mark_atom_dirty(r300, &r300->hyperz_state);
504 }
505
r300_decompress_zmask_locked_unsafe(struct r300_context * r300)506 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300)
507 {
508 struct pipe_framebuffer_state fb;
509
510 memset(&fb, 0, sizeof(fb));
511 fb.width = r300->locked_zbuffer->width;
512 fb.height = r300->locked_zbuffer->height;
513 fb.zsbuf = r300->locked_zbuffer;
514
515 r300->context.set_framebuffer_state(&r300->context, &fb);
516 r300_decompress_zmask(r300);
517 }
518
r300_decompress_zmask_locked(struct r300_context * r300)519 void r300_decompress_zmask_locked(struct r300_context *r300)
520 {
521 struct pipe_framebuffer_state saved_fb;
522
523 memset(&saved_fb, 0, sizeof(saved_fb));
524 util_copy_framebuffer_state(&saved_fb, r300->fb_state.state);
525 r300_decompress_zmask_locked_unsafe(r300);
526 r300->context.set_framebuffer_state(&r300->context, &saved_fb);
527 util_unreference_framebuffer_state(&saved_fb);
528
529 pipe_surface_reference(&r300->locked_zbuffer, NULL);
530 }
531
r300_is_blit_supported(enum pipe_format format)532 bool r300_is_blit_supported(enum pipe_format format)
533 {
534 const struct util_format_description *desc =
535 util_format_description(format);
536
537 return desc->layout == UTIL_FORMAT_LAYOUT_PLAIN ||
538 desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
539 desc->layout == UTIL_FORMAT_LAYOUT_RGTC;
540 }
541
542 /* Copy a block of pixels from one surface to another. */
r300_resource_copy_region(struct pipe_context * pipe,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)543 static void r300_resource_copy_region(struct pipe_context *pipe,
544 struct pipe_resource *dst,
545 unsigned dst_level,
546 unsigned dstx, unsigned dsty, unsigned dstz,
547 struct pipe_resource *src,
548 unsigned src_level,
549 const struct pipe_box *src_box)
550 {
551 struct pipe_screen *screen = pipe->screen;
552 struct r300_context *r300 = r300_context(pipe);
553 struct pipe_framebuffer_state *fb =
554 (struct pipe_framebuffer_state*)r300->fb_state.state;
555 unsigned src_width0 = r300_resource(src)->tex.width0;
556 unsigned src_height0 = r300_resource(src)->tex.height0;
557 unsigned dst_width0 = r300_resource(dst)->tex.width0;
558 unsigned dst_height0 = r300_resource(dst)->tex.height0;
559 unsigned layout;
560 struct pipe_box box, dstbox;
561 struct pipe_sampler_view src_templ, *src_view;
562 struct pipe_surface dst_templ, *dst_view;
563
564 /* Fallback for buffers. */
565 if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
566 !r300_is_blit_supported(dst->format)) {
567 util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
568 src, src_level, src_box);
569 return;
570 }
571
572 /* Can't read MSAA textures. */
573 if (src->nr_samples > 1 || dst->nr_samples > 1) {
574 return;
575 }
576
577 /* The code below changes the texture format so that the copy can be done
578 * on hardware. E.g. depth-stencil surfaces are copied as RGBA
579 * colorbuffers. */
580
581 util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz);
582 util_blitter_default_src_texture(r300->blitter, &src_templ, src, src_level);
583
584 layout = util_format_description(dst_templ.format)->layout;
585
586 /* Handle non-renderable plain formats. */
587 if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
588 (!screen->is_format_supported(screen, src_templ.format, src->target,
589 src->nr_samples, src->nr_storage_samples,
590 PIPE_BIND_SAMPLER_VIEW) ||
591 !screen->is_format_supported(screen, dst_templ.format, dst->target,
592 dst->nr_samples, dst->nr_storage_samples,
593 PIPE_BIND_RENDER_TARGET))) {
594 switch (util_format_get_blocksize(dst_templ.format)) {
595 case 1:
596 dst_templ.format = PIPE_FORMAT_I8_UNORM;
597 break;
598 case 2:
599 dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
600 break;
601 case 4:
602 dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
603 break;
604 case 8:
605 dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
606 break;
607 default:
608 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
609 "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
610 util_format_short_name(dst_templ.format));
611 }
612 src_templ.format = dst_templ.format;
613 }
614
615 /* Handle compressed formats. */
616 if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
617 layout == UTIL_FORMAT_LAYOUT_RGTC) {
618 assert(src_templ.format == dst_templ.format);
619
620 box = *src_box;
621 src_box = &box;
622
623 dst_width0 = align(dst_width0, 4);
624 dst_height0 = align(dst_height0, 4);
625 src_width0 = align(src_width0, 4);
626 src_height0 = align(src_height0, 4);
627 box.width = align(box.width, 4);
628 box.height = align(box.height, 4);
629
630 switch (util_format_get_blocksize(dst_templ.format)) {
631 case 8:
632 /* one 4x4 pixel block has 8 bytes.
633 * we set 1 pixel = 4 bytes ===> 1 block corresponds to 2 pixels. */
634 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
635 dst_width0 = dst_width0 / 2;
636 src_width0 = src_width0 / 2;
637 dstx /= 2;
638 box.x /= 2;
639 box.width /= 2;
640 break;
641 case 16:
642 /* one 4x4 pixel block has 16 bytes.
643 * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
644 dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
645 break;
646 }
647 src_templ.format = dst_templ.format;
648
649 dst_height0 = dst_height0 / 4;
650 src_height0 = src_height0 / 4;
651 dsty /= 4;
652 box.y /= 4;
653 box.height /= 4;
654 }
655
656 /* Fallback for textures. */
657 if (!screen->is_format_supported(screen, dst_templ.format,
658 dst->target, dst->nr_samples,
659 dst->nr_storage_samples,
660 PIPE_BIND_RENDER_TARGET) ||
661 !screen->is_format_supported(screen, src_templ.format,
662 src->target, src->nr_samples,
663 src->nr_storage_samples,
664 PIPE_BIND_SAMPLER_VIEW)) {
665 assert(0 && "this shouldn't happen, update r300_is_blit_supported");
666 util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
667 src, src_level, src_box);
668 return;
669 }
670
671 /* Decompress ZMASK. */
672 if (r300->zmask_in_use && !r300->locked_zbuffer) {
673 if (fb->zsbuf->texture == src ||
674 fb->zsbuf->texture == dst) {
675 r300_decompress_zmask(r300);
676 }
677 }
678
679 dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
680 src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
681
682 u_box_3d(dstx, dsty, dstz, abs(src_box->width), abs(src_box->height),
683 abs(src_box->depth), &dstbox);
684
685 r300_blitter_begin(r300, R300_COPY);
686 util_blitter_blit_generic(r300->blitter, dst_view, &dstbox,
687 src_view, src_box, src_width0, src_height0,
688 PIPE_MASK_RGBAZS, PIPE_TEX_FILTER_NEAREST, NULL,
689 false, false, 0);
690 r300_blitter_end(r300);
691
692 pipe_surface_reference(&dst_view, NULL);
693 pipe_sampler_view_reference(&src_view, NULL);
694 }
695
r300_is_simple_msaa_resolve(const struct pipe_blit_info * info)696 static bool r300_is_simple_msaa_resolve(const struct pipe_blit_info *info)
697 {
698 unsigned dst_width = u_minify(info->dst.resource->width0, info->dst.level);
699 unsigned dst_height = u_minify(info->dst.resource->height0, info->dst.level);
700
701 return info->src.resource->nr_samples > 1 &&
702 info->dst.resource->nr_samples <= 1 &&
703 info->dst.resource->format == info->src.resource->format &&
704 info->dst.resource->format == info->dst.format &&
705 info->src.resource->format == info->src.format &&
706 !info->scissor_enable &&
707 info->mask == PIPE_MASK_RGBA &&
708 dst_width == info->src.resource->width0 &&
709 dst_height == info->src.resource->height0 &&
710 info->dst.box.x == 0 &&
711 info->dst.box.y == 0 &&
712 info->dst.box.width == dst_width &&
713 info->dst.box.height == dst_height &&
714 info->src.box.x == 0 &&
715 info->src.box.y == 0 &&
716 info->src.box.width == dst_width &&
717 info->src.box.height == dst_height &&
718 (r300_resource(info->dst.resource)->tex.microtile != RADEON_LAYOUT_LINEAR ||
719 r300_resource(info->dst.resource)->tex.macrotile[info->dst.level] != RADEON_LAYOUT_LINEAR);
720 }
721
r300_simple_msaa_resolve(struct pipe_context * pipe,struct pipe_resource * dst,unsigned dst_level,unsigned dst_layer,struct pipe_resource * src,enum pipe_format format)722 static void r300_simple_msaa_resolve(struct pipe_context *pipe,
723 struct pipe_resource *dst,
724 unsigned dst_level,
725 unsigned dst_layer,
726 struct pipe_resource *src,
727 enum pipe_format format)
728 {
729 struct r300_context *r300 = r300_context(pipe);
730 struct r300_surface *srcsurf, *dstsurf;
731 struct pipe_surface surf_tmpl;
732 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
733
734 memset(&surf_tmpl, 0, sizeof(surf_tmpl));
735 surf_tmpl.format = format;
736 srcsurf = r300_surface(pipe->create_surface(pipe, src, &surf_tmpl));
737
738 surf_tmpl.format = format;
739 surf_tmpl.u.tex.level = dst_level;
740 surf_tmpl.u.tex.first_layer =
741 surf_tmpl.u.tex.last_layer = dst_layer;
742 dstsurf = r300_surface(pipe->create_surface(pipe, dst, &surf_tmpl));
743
744 /* COLORPITCH should contain the tiling info of the resolve buffer.
745 * The tiling of the AA buffer isn't programmable anyway. */
746 srcsurf->pitch &= ~(R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
747 srcsurf->pitch |= dstsurf->pitch & (R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
748
749 /* Enable AA resolve. */
750 aa->dest = dstsurf;
751 r300->aa_state.size = 8;
752 r300_mark_atom_dirty(r300, &r300->aa_state);
753
754 /* Resolve the surface. */
755 r300_blitter_begin(r300, R300_CLEAR_SURFACE);
756 util_blitter_custom_color(r300->blitter, &srcsurf->base, NULL);
757 r300_blitter_end(r300);
758
759 /* Disable AA resolve. */
760 aa->dest = NULL;
761 r300->aa_state.size = 4;
762 r300_mark_atom_dirty(r300, &r300->aa_state);
763
764 pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
765 pipe_surface_reference((struct pipe_surface**)&dstsurf, NULL);
766 }
767
r300_msaa_resolve(struct pipe_context * pipe,const struct pipe_blit_info * info)768 static void r300_msaa_resolve(struct pipe_context *pipe,
769 const struct pipe_blit_info *info)
770 {
771 struct r300_context *r300 = r300_context(pipe);
772 struct pipe_screen *screen = pipe->screen;
773 struct pipe_resource *tmp, templ;
774 struct pipe_blit_info blit;
775
776 assert(info->src.level == 0);
777 assert(info->src.box.z == 0);
778 assert(info->src.box.depth == 1);
779 assert(info->dst.box.depth == 1);
780
781 if (r300_is_simple_msaa_resolve(info)) {
782 r300_simple_msaa_resolve(pipe, info->dst.resource, info->dst.level,
783 info->dst.box.z, info->src.resource,
784 info->src.format);
785 return;
786 }
787
788 /* resolve into a temporary texture, then blit */
789 memset(&templ, 0, sizeof(templ));
790 templ.target = PIPE_TEXTURE_2D;
791 templ.format = info->src.resource->format;
792 templ.width0 = info->src.resource->width0;
793 templ.height0 = info->src.resource->height0;
794 templ.depth0 = 1;
795 templ.array_size = 1;
796 templ.usage = PIPE_USAGE_DEFAULT;
797 templ.flags = R300_RESOURCE_FORCE_MICROTILING;
798
799 tmp = screen->resource_create(screen, &templ);
800
801 /* resolve */
802 r300_simple_msaa_resolve(pipe, tmp, 0, 0, info->src.resource,
803 info->src.format);
804
805 /* blit */
806 blit = *info;
807 blit.src.resource = tmp;
808 blit.src.box.z = 0;
809
810 r300_blitter_begin(r300, R300_BLIT | R300_IGNORE_RENDER_COND);
811 util_blitter_blit(r300->blitter, &blit);
812 r300_blitter_end(r300);
813
814 pipe_resource_reference(&tmp, NULL);
815 }
816
r300_blit(struct pipe_context * pipe,const struct pipe_blit_info * blit)817 static void r300_blit(struct pipe_context *pipe,
818 const struct pipe_blit_info *blit)
819 {
820 struct r300_context *r300 = r300_context(pipe);
821 struct pipe_framebuffer_state *fb =
822 (struct pipe_framebuffer_state*)r300->fb_state.state;
823 struct pipe_blit_info info = *blit;
824
825 /* The driver supports sRGB textures but not framebuffers. Blitting
826 * from sRGB to sRGB should be the same as blitting from linear
827 * to linear, so use that, This avoids incorrect linearization.
828 */
829 if (util_format_is_srgb(info.src.format)) {
830 info.src.format = util_format_linear(info.src.format);
831 info.dst.format = util_format_linear(info.dst.format);
832 }
833
834 /* MSAA resolve. */
835 if (info.src.resource->nr_samples > 1 &&
836 !util_format_is_depth_or_stencil(info.src.resource->format)) {
837 r300_msaa_resolve(pipe, &info);
838 return;
839 }
840
841 /* Can't read MSAA textures. */
842 if (info.src.resource->nr_samples > 1) {
843 return;
844 }
845
846 /* Blit a combined depth-stencil resource as color.
847 * S8Z24 is the only supported stencil format. */
848 if ((info.mask & PIPE_MASK_S) &&
849 info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
850 info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
851 if (info.dst.resource->nr_samples > 1) {
852 /* Cannot do that with MSAA buffers. */
853 info.mask &= ~PIPE_MASK_S;
854 if (!(info.mask & PIPE_MASK_Z)) {
855 return;
856 }
857 } else {
858 /* Single-sample buffer. */
859 info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
860 info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
861 if (info.mask & PIPE_MASK_Z) {
862 info.mask = PIPE_MASK_RGBA; /* depth+stencil */
863 } else {
864 info.mask = PIPE_MASK_B; /* stencil only */
865 }
866 }
867 }
868
869 /* Decompress ZMASK. */
870 if (r300->zmask_in_use && !r300->locked_zbuffer) {
871 if (fb->zsbuf->texture == info.src.resource ||
872 fb->zsbuf->texture == info.dst.resource) {
873 r300_decompress_zmask(r300);
874 }
875 }
876
877 r300_blitter_begin(r300, R300_BLIT |
878 (info.render_condition_enable ? 0 : R300_IGNORE_RENDER_COND));
879 util_blitter_blit(r300->blitter, &info);
880 r300_blitter_end(r300);
881 }
882
r300_flush_resource(struct pipe_context * ctx,struct pipe_resource * resource)883 static void r300_flush_resource(struct pipe_context *ctx,
884 struct pipe_resource *resource)
885 {
886 }
887
r300_init_blit_functions(struct r300_context * r300)888 void r300_init_blit_functions(struct r300_context *r300)
889 {
890 r300->context.clear = r300_clear;
891 r300->context.clear_render_target = r300_clear_render_target;
892 r300->context.clear_depth_stencil = r300_clear_depth_stencil;
893 r300->context.resource_copy_region = r300_resource_copy_region;
894 r300->context.blit = r300_blit;
895 r300->context.flush_resource = r300_flush_resource;
896 }
897