2 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
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:
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
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. */
23 #include "r300_context.h"
24 #include "r300_emit.h"
25 #include "r300_texture.h"
27 #include "util/u_format.h"
28 #include "util/u_pack_color.h"
29 #include "util/u_surface.h"
31 enum r300_blitter_op
/* bitmask */
34 R300_SAVE_TEXTURES
= 2,
35 R300_SAVE_FRAMEBUFFER
= 4,
36 R300_IGNORE_RENDER_COND
= 8,
38 R300_CLEAR
= R300_STOP_QUERY
,
40 R300_CLEAR_SURFACE
= R300_STOP_QUERY
| R300_SAVE_FRAMEBUFFER
,
42 R300_COPY
= R300_STOP_QUERY
| R300_SAVE_FRAMEBUFFER
|
43 R300_SAVE_TEXTURES
| R300_IGNORE_RENDER_COND
,
45 R300_DECOMPRESS
= R300_STOP_QUERY
| R300_IGNORE_RENDER_COND
,
48 static void r300_blitter_begin(struct r300_context
* r300
, enum r300_blitter_op op
)
50 if ((op
& R300_STOP_QUERY
) && r300
->query_current
) {
51 r300
->blitter_saved_query
= r300
->query_current
;
52 r300_stop_query(r300
);
55 /* Yeah we have to save all those states to ensure the blitter operation
56 * is really transparent. The states will be restored by the blitter once
58 util_blitter_save_blend(r300
->blitter
, r300
->blend_state
.state
);
59 util_blitter_save_depth_stencil_alpha(r300
->blitter
, r300
->dsa_state
.state
);
60 util_blitter_save_stencil_ref(r300
->blitter
, &(r300
->stencil_ref
));
61 util_blitter_save_rasterizer(r300
->blitter
, r300
->rs_state
.state
);
62 util_blitter_save_fragment_shader(r300
->blitter
, r300
->fs
.state
);
63 util_blitter_save_vertex_shader(r300
->blitter
, r300
->vs_state
.state
);
64 util_blitter_save_viewport(r300
->blitter
, &r300
->viewport
);
65 util_blitter_save_vertex_elements(r300
->blitter
, r300
->velems
);
67 util_blitter_save_vertex_buffers(r300
->blitter
, r300
->vbuf_mgr
->nr_vertex_buffers
,
68 r300
->vbuf_mgr
->vertex_buffer
);
70 util_blitter_save_vertex_buffers(r300
->blitter
, r300
->nr_vertex_buffers
,
74 if (op
& R300_SAVE_FRAMEBUFFER
) {
75 util_blitter_save_framebuffer(r300
->blitter
, r300
->fb_state
.state
);
78 if (op
& R300_SAVE_TEXTURES
) {
79 struct r300_textures_state
* state
=
80 (struct r300_textures_state
*)r300
->textures_state
.state
;
82 util_blitter_save_fragment_sampler_states(
83 r300
->blitter
, state
->sampler_state_count
,
84 (void**)state
->sampler_states
);
86 util_blitter_save_fragment_sampler_views(
87 r300
->blitter
, state
->sampler_view_count
,
88 (struct pipe_sampler_view
**)state
->sampler_views
);
91 if (op
& R300_IGNORE_RENDER_COND
) {
93 r300
->blitter_saved_skip_rendering
= r300
->skip_rendering
+1;
94 r300
->skip_rendering
= FALSE
;
96 r300
->blitter_saved_skip_rendering
= 0;
100 static void r300_blitter_end(struct r300_context
*r300
)
102 if (r300
->blitter_saved_query
) {
103 r300_resume_query(r300
, r300
->blitter_saved_query
);
104 r300
->blitter_saved_query
= NULL
;
107 if (r300
->blitter_saved_skip_rendering
) {
108 /* Restore the flag. */
109 r300
->skip_rendering
= r300
->blitter_saved_skip_rendering
-1;
113 static uint32_t r300_depth_clear_cb_value(enum pipe_format format
,
117 util_pack_color(rgba
, format
, &uc
);
119 if (util_format_get_blocksizebits(format
) == 32)
122 return uc
.us
| (uc
.us
<< 16);
125 static boolean
r300_cbzb_clear_allowed(struct r300_context
*r300
,
126 unsigned clear_buffers
)
128 struct pipe_framebuffer_state
*fb
=
129 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
131 /* Only color clear allowed, and only one colorbuffer. */
132 if (clear_buffers
!= PIPE_CLEAR_COLOR
|| fb
->nr_cbufs
!= 1)
135 return r300_surface(fb
->cbufs
[0])->cbzb_allowed
;
138 static boolean
r300_fast_zclear_allowed(struct r300_context
*r300
)
140 struct pipe_framebuffer_state
*fb
=
141 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
143 return r300_resource(fb
->zsbuf
->texture
)->tex
.zmask_dwords
[fb
->zsbuf
->u
.tex
.level
] != 0;
146 static boolean
r300_hiz_clear_allowed(struct r300_context
*r300
)
148 struct pipe_framebuffer_state
*fb
=
149 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
151 return r300_resource(fb
->zsbuf
->texture
)->tex
.hiz_dwords
[fb
->zsbuf
->u
.tex
.level
] != 0;
154 static uint32_t r300_depth_clear_value(enum pipe_format format
,
155 double depth
, unsigned stencil
)
158 case PIPE_FORMAT_Z16_UNORM
:
159 case PIPE_FORMAT_X8Z24_UNORM
:
160 return util_pack_z(format
, depth
);
162 case PIPE_FORMAT_S8_UINT_Z24_UNORM
:
163 return util_pack_z_stencil(format
, depth
, stencil
);
171 static uint32_t r300_hiz_clear_value(double depth
)
173 uint32_t r
= (uint32_t)(CLAMP(depth
, 0, 1) * 255.5);
175 return r
| (r
<< 8) | (r
<< 16) | (r
<< 24);
178 /* Clear currently bound buffers. */
179 static void r300_clear(struct pipe_context
* pipe
,
181 const union pipe_color_union
*color
,
185 /* My notes about Zbuffer compression:
187 * 1) The zbuffer must be micro-tiled and whole microtiles must be
188 * written if compression is enabled. If microtiling is disabled,
191 * 2) There is ZMASK RAM which contains a compressed zbuffer.
192 * Each dword of the Z Mask contains compression information
193 * for 16 4x4 pixel tiles, that is 2 bits for each tile.
194 * On chips with 2 Z pipes, every other dword maps to a different
195 * pipe. On newer chipsets, there is a new compression mode
196 * with 8x8 pixel tiles per 2 bits.
198 * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
199 * it should look in the ZMASK RAM first before fetching from a real
202 * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
203 * during zbuffer reads instead of the value that is actually stored
204 * in the zbuffer memory. A pixel is in a cleared state when its ZMASK
205 * is equal to 0. Therefore, if you clear ZMASK with zeros, you may
206 * leave the zbuffer memory uninitialized, but then you must enable
207 * compression, so that the ZMASK RAM is actually used.
209 * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
210 * during zbuffer updates. A special decompressing operation should be
211 * used to fully decompress a zbuffer, which basically just stores all
212 * compressed tiles in ZMASK to the zbuffer memory.
214 * 6) For a 16-bit zbuffer, compression causes a hung with one or
215 * two samples and should not be used.
217 * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
218 * to avoid needless decompression.
220 * 8) Fastfill must not be used if reading of compressed Z data is disabled
221 * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
222 * i.e. it cannot be used to compress the zbuffer.
224 * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
229 struct r300_context
* r300
= r300_context(pipe
);
230 struct pipe_framebuffer_state
*fb
=
231 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
232 struct r300_hyperz_state
*hyperz
=
233 (struct r300_hyperz_state
*)r300
->hyperz_state
.state
;
234 uint32_t width
= fb
->width
;
235 uint32_t height
= fb
->height
;
236 uint32_t hyperz_dcv
= hyperz
->zb_depthclearvalue
;
238 /* Enable fast Z clear.
239 * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
240 if (buffers
& PIPE_CLEAR_DEPTHSTENCIL
) {
241 boolean zmask_clear
, hiz_clear
;
243 zmask_clear
= r300_fast_zclear_allowed(r300
);
244 hiz_clear
= r300_hiz_clear_allowed(r300
);
246 /* If we need Hyper-Z. */
247 if (zmask_clear
|| hiz_clear
) {
248 r300
->num_z_clears
++;
250 /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
251 if (!r300
->hyperz_enabled
) {
252 r300
->hyperz_enabled
=
253 r300
->rws
->cs_request_feature(r300
->cs
,
254 RADEON_FID_R300_HYPERZ_ACCESS
,
256 if (r300
->hyperz_enabled
) {
257 /* Need to emit HyperZ buffer regs for the first time. */
258 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
262 /* Setup Hyper-Z clears. */
263 if (r300
->hyperz_enabled
) {
264 DBG(r300
, DBG_HYPERZ
, "r300: Clear memory: %s%s\n",
265 zmask_clear
? "ZMASK " : "", hiz_clear
? "HIZ" : "");
268 hyperz_dcv
= hyperz
->zb_depthclearvalue
=
269 r300_depth_clear_value(fb
->zsbuf
->format
, depth
, stencil
);
271 r300_mark_atom_dirty(r300
, &r300
->zmask_clear
);
272 buffers
&= ~PIPE_CLEAR_DEPTHSTENCIL
;
276 r300
->hiz_clear_value
= r300_hiz_clear_value(depth
);
277 r300_mark_atom_dirty(r300
, &r300
->hiz_clear
);
283 /* Enable CBZB clear. */
284 if (r300_cbzb_clear_allowed(r300
, buffers
)) {
285 struct r300_surface
*surf
= r300_surface(fb
->cbufs
[0]);
287 hyperz
->zb_depthclearvalue
=
288 r300_depth_clear_cb_value(surf
->base
.format
, color
->f
);
290 width
= surf
->cbzb_width
;
291 height
= surf
->cbzb_height
;
293 r300
->cbzb_clear
= TRUE
;
294 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
299 enum pipe_format cformat
= fb
->nr_cbufs
? fb
->cbufs
[0]->format
: PIPE_FORMAT_NONE
;
300 /* Clear using the blitter. */
301 r300_blitter_begin(r300
, R300_CLEAR
);
302 util_blitter_clear(r300
->blitter
,
306 buffers
, cformat
, color
, depth
, stencil
);
307 r300_blitter_end(r300
);
308 } else if (r300
->zmask_clear
.dirty
|| r300
->hiz_clear
.dirty
) {
309 /* Just clear zmask and hiz now, this does not use the standard draw
311 /* Calculate zmask_clear and hiz_clear atom sizes. */
313 (r300
->zmask_clear
.dirty
? r300
->zmask_clear
.size
: 0) +
314 (r300
->hiz_clear
.dirty
? r300
->hiz_clear
.size
: 0) +
315 r300_get_num_cs_end_dwords(r300
);
317 /* Reserve CS space. */
318 if (dwords
> (RADEON_MAX_CMDBUF_DWORDS
- r300
->cs
->cdw
)) {
319 r300_flush(&r300
->context
, RADEON_FLUSH_ASYNC
, NULL
);
322 /* Emit clear packets. */
323 if (r300
->zmask_clear
.dirty
) {
324 r300_emit_zmask_clear(r300
, r300
->zmask_clear
.size
,
325 r300
->zmask_clear
.state
);
326 r300
->zmask_clear
.dirty
= FALSE
;
328 if (r300
->hiz_clear
.dirty
) {
329 r300_emit_hiz_clear(r300
, r300
->hiz_clear
.size
,
330 r300
->hiz_clear
.state
);
331 r300
->hiz_clear
.dirty
= FALSE
;
337 /* Disable CBZB clear. */
338 if (r300
->cbzb_clear
) {
339 r300
->cbzb_clear
= FALSE
;
340 hyperz
->zb_depthclearvalue
= hyperz_dcv
;
341 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
344 /* Enable fastfill and/or hiz.
346 * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
347 * looks if zmask/hiz is in use and programs hardware accordingly. */
348 if (r300
->zmask_in_use
|| r300
->hiz_in_use
) {
349 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
353 /* Clear a region of a color surface to a constant value. */
354 static void r300_clear_render_target(struct pipe_context
*pipe
,
355 struct pipe_surface
*dst
,
356 const union pipe_color_union
*color
,
357 unsigned dstx
, unsigned dsty
,
358 unsigned width
, unsigned height
)
360 struct r300_context
*r300
= r300_context(pipe
);
362 r300_blitter_begin(r300
, R300_CLEAR_SURFACE
);
363 util_blitter_clear_render_target(r300
->blitter
, dst
, color
,
364 dstx
, dsty
, width
, height
);
365 r300_blitter_end(r300
);
368 /* Clear a region of a depth stencil surface. */
369 static void r300_clear_depth_stencil(struct pipe_context
*pipe
,
370 struct pipe_surface
*dst
,
371 unsigned clear_flags
,
374 unsigned dstx
, unsigned dsty
,
375 unsigned width
, unsigned height
)
377 struct r300_context
*r300
= r300_context(pipe
);
378 struct pipe_framebuffer_state
*fb
=
379 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
381 if (r300
->zmask_in_use
&& !r300
->locked_zbuffer
) {
382 if (fb
->zsbuf
->texture
== dst
->texture
) {
383 r300_decompress_zmask(r300
);
387 /* XXX Do not decompress ZMask of the currently-set zbuffer. */
388 r300_blitter_begin(r300
, R300_CLEAR_SURFACE
);
389 util_blitter_clear_depth_stencil(r300
->blitter
, dst
, clear_flags
, depth
, stencil
,
390 dstx
, dsty
, width
, height
);
391 r300_blitter_end(r300
);
394 void r300_decompress_zmask(struct r300_context
*r300
)
396 struct pipe_framebuffer_state
*fb
=
397 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
399 if (!r300
->zmask_in_use
|| r300
->locked_zbuffer
)
402 r300
->zmask_decompress
= TRUE
;
403 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
405 r300_blitter_begin(r300
, R300_DECOMPRESS
);
406 util_blitter_clear_depth_custom(r300
->blitter
, fb
->width
, fb
->height
, 0,
407 r300
->dsa_decompress_zmask
);
408 r300_blitter_end(r300
);
410 r300
->zmask_decompress
= FALSE
;
411 r300
->zmask_in_use
= FALSE
;
412 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
415 void r300_decompress_zmask_locked_unsafe(struct r300_context
*r300
)
417 struct pipe_framebuffer_state fb
;
419 memset(&fb
, 0, sizeof(fb
));
420 fb
.width
= r300
->locked_zbuffer
->width
;
421 fb
.height
= r300
->locked_zbuffer
->height
;
422 fb
.zsbuf
= r300
->locked_zbuffer
;
424 r300
->context
.set_framebuffer_state(&r300
->context
, &fb
);
425 r300_decompress_zmask(r300
);
428 void r300_decompress_zmask_locked(struct r300_context
*r300
)
430 struct pipe_framebuffer_state saved_fb
;
432 memset(&saved_fb
, 0, sizeof(saved_fb
));
433 util_copy_framebuffer_state(&saved_fb
, r300
->fb_state
.state
);
434 r300_decompress_zmask_locked_unsafe(r300
);
435 r300
->context
.set_framebuffer_state(&r300
->context
, &saved_fb
);
436 util_unreference_framebuffer_state(&saved_fb
);
438 pipe_surface_reference(&r300
->locked_zbuffer
, NULL
);
441 bool r300_is_blit_supported(enum pipe_format format
)
443 const struct util_format_description
*desc
=
444 util_format_description(format
);
446 return desc
->layout
== UTIL_FORMAT_LAYOUT_PLAIN
||
447 desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
||
448 desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
;
451 /* Copy a block of pixels from one surface to another. */
452 static void r300_resource_copy_region(struct pipe_context
*pipe
,
453 struct pipe_resource
*dst
,
455 unsigned dstx
, unsigned dsty
, unsigned dstz
,
456 struct pipe_resource
*src
,
458 const struct pipe_box
*src_box
)
460 struct pipe_screen
*screen
= pipe
->screen
;
461 struct r300_context
*r300
= r300_context(pipe
);
462 struct pipe_framebuffer_state
*fb
=
463 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
464 unsigned src_width0
= r300_resource(src
)->tex
.width0
;
465 unsigned src_height0
= r300_resource(src
)->tex
.height0
;
466 unsigned dst_width0
= r300_resource(dst
)->tex
.width0
;
467 unsigned dst_height0
= r300_resource(dst
)->tex
.height0
;
470 struct pipe_sampler_view src_templ
, *src_view
;
471 struct pipe_surface dst_templ
, *dst_view
;
473 /* Fallback for buffers. */
474 if ((dst
->target
== PIPE_BUFFER
&& src
->target
== PIPE_BUFFER
) ||
475 !r300_is_blit_supported(dst
->format
)) {
476 util_resource_copy_region(pipe
, dst
, dst_level
, dstx
, dsty
, dstz
,
477 src
, src_level
, src_box
);
481 /* The code below changes the texture format so that the copy can be done
482 * on hardware. E.g. depth-stencil surfaces are copied as RGBA
485 util_blitter_default_dst_texture(&dst_templ
, dst
, dst_level
, dstz
, src_box
);
486 util_blitter_default_src_texture(&src_templ
, src
, src_level
);
488 layout
= util_format_description(dst_templ
.format
)->layout
;
490 /* Handle non-renderable plain formats. */
491 if (layout
== UTIL_FORMAT_LAYOUT_PLAIN
&&
492 (!screen
->is_format_supported(screen
, src_templ
.format
, src
->target
,
494 PIPE_BIND_SAMPLER_VIEW
) ||
495 !screen
->is_format_supported(screen
, dst_templ
.format
, dst
->target
,
497 PIPE_BIND_RENDER_TARGET
))) {
498 switch (util_format_get_blocksize(dst_templ
.format
)) {
500 dst_templ
.format
= PIPE_FORMAT_I8_UNORM
;
503 dst_templ
.format
= PIPE_FORMAT_B4G4R4A4_UNORM
;
506 dst_templ
.format
= PIPE_FORMAT_B8G8R8A8_UNORM
;
509 dst_templ
.format
= PIPE_FORMAT_R16G16B16A16_UNORM
;
512 debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
513 "r300: copy_region: Software fallback doesn't work for tiled textures.\n",
514 util_format_short_name(dst_templ
.format
));
516 src_templ
.format
= dst_templ
.format
;
519 /* Handle compressed formats. */
520 if (layout
== UTIL_FORMAT_LAYOUT_S3TC
||
521 layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
522 assert(src_templ
.format
== dst_templ
.format
);
527 dst_width0
= align(dst_width0
, 4);
528 dst_height0
= align(dst_height0
, 4);
529 src_width0
= align(src_width0
, 4);
530 src_height0
= align(src_height0
, 4);
531 box
.width
= align(box
.width
, 4);
532 box
.height
= align(box
.height
, 4);
534 switch (util_format_get_blocksize(dst_templ
.format
)) {
536 /* one 4x4 pixel block has 8 bytes.
537 * we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
538 dst_templ
.format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
539 dst_width0
= dst_width0
/ 2;
540 src_width0
= src_width0
/ 2;
546 /* one 4x4 pixel block has 16 bytes.
547 * we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
548 dst_templ
.format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
551 src_templ
.format
= dst_templ
.format
;
553 dst_height0
= dst_height0
/ 4;
554 src_height0
= src_height0
/ 4;
560 /* Fallback for textures. */
561 if (!screen
->is_format_supported(screen
, dst_templ
.format
,
562 dst
->target
, dst
->nr_samples
,
563 PIPE_BIND_RENDER_TARGET
) ||
564 !screen
->is_format_supported(screen
, src_templ
.format
,
565 src
->target
, src
->nr_samples
,
566 PIPE_BIND_SAMPLER_VIEW
)) {
567 assert(0 && "this shouldn't happen, update r300_is_blit_supported");
568 util_resource_copy_region(pipe
, dst
, dst_level
, dstx
, dsty
, dstz
,
569 src
, src_level
, src_box
);
573 /* Decompress ZMASK. */
574 if (r300
->zmask_in_use
&& !r300
->locked_zbuffer
) {
575 if (fb
->zsbuf
->texture
== src
||
576 fb
->zsbuf
->texture
== dst
) {
577 r300_decompress_zmask(r300
);
581 dst_view
= r300_create_surface_custom(pipe
, dst
, &dst_templ
, dst_width0
, dst_height0
);
582 src_view
= r300_create_sampler_view_custom(pipe
, src
, &src_templ
, src_width0
, src_height0
);
584 r300_blitter_begin(r300
, R300_COPY
);
585 util_blitter_copy_texture_view(r300
->blitter
, dst_view
, dstx
, dsty
,
587 src_width0
, src_height0
);
588 r300_blitter_end(r300
);
590 pipe_surface_reference(&dst_view
, NULL
);
591 pipe_sampler_view_reference(&src_view
, NULL
);
594 void r300_init_blit_functions(struct r300_context
*r300
)
596 r300
->context
.clear
= r300_clear
;
597 r300
->context
.clear_render_target
= r300_clear_render_target
;
598 r300
->context
.clear_depth_stencil
= r300_clear_depth_stencil
;
599 r300
->context
.resource_copy_region
= r300_resource_copy_region
;