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_clip(r300
->blitter
, (struct pipe_clip_state
*)r300
->clip_state
.state
);
66 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 if (op
& R300_SAVE_FRAMEBUFFER
) {
71 util_blitter_save_framebuffer(r300
->blitter
, r300
->fb_state
.state
);
74 if (op
& R300_SAVE_TEXTURES
) {
75 struct r300_textures_state
* state
=
76 (struct r300_textures_state
*)r300
->textures_state
.state
;
78 util_blitter_save_fragment_sampler_states(
79 r300
->blitter
, state
->sampler_state_count
,
80 (void**)state
->sampler_states
);
82 util_blitter_save_fragment_sampler_views(
83 r300
->blitter
, state
->sampler_view_count
,
84 (struct pipe_sampler_view
**)state
->sampler_views
);
87 if (op
& R300_IGNORE_RENDER_COND
) {
89 r300
->blitter_saved_skip_rendering
= r300
->skip_rendering
+1;
90 r300
->skip_rendering
= FALSE
;
92 r300
->blitter_saved_skip_rendering
= 0;
96 static void r300_blitter_end(struct r300_context
*r300
)
98 if (r300
->blitter_saved_query
) {
99 r300_resume_query(r300
, r300
->blitter_saved_query
);
100 r300
->blitter_saved_query
= NULL
;
103 if (r300
->blitter_saved_skip_rendering
) {
104 /* Restore the flag. */
105 r300
->skip_rendering
= r300
->blitter_saved_skip_rendering
-1;
109 static uint32_t r300_depth_clear_cb_value(enum pipe_format format
,
113 util_pack_color(rgba
, format
, &uc
);
115 if (util_format_get_blocksizebits(format
) == 32)
118 return uc
.us
| (uc
.us
<< 16);
121 static boolean
r300_cbzb_clear_allowed(struct r300_context
*r300
,
122 unsigned clear_buffers
)
124 struct pipe_framebuffer_state
*fb
=
125 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
127 /* Only color clear allowed, and only one colorbuffer. */
128 if (clear_buffers
!= PIPE_CLEAR_COLOR
|| fb
->nr_cbufs
!= 1)
131 return r300_surface(fb
->cbufs
[0])->cbzb_allowed
;
134 static boolean
r300_fast_zclear_allowed(struct r300_context
*r300
)
136 struct pipe_framebuffer_state
*fb
=
137 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
139 return r300_resource(fb
->zsbuf
->texture
)->tex
.zmask_dwords
[fb
->zsbuf
->u
.tex
.level
] != 0;
142 static boolean
r300_hiz_clear_allowed(struct r300_context
*r300
)
144 struct pipe_framebuffer_state
*fb
=
145 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
147 return r300_resource(fb
->zsbuf
->texture
)->tex
.hiz_dwords
[fb
->zsbuf
->u
.tex
.level
] != 0;
150 static uint32_t r300_depth_clear_value(enum pipe_format format
,
151 double depth
, unsigned stencil
)
154 case PIPE_FORMAT_Z16_UNORM
:
155 case PIPE_FORMAT_X8Z24_UNORM
:
156 return util_pack_z(format
, depth
);
158 case PIPE_FORMAT_S8_USCALED_Z24_UNORM
:
159 return util_pack_z_stencil(format
, depth
, stencil
);
167 static uint32_t r300_hiz_clear_value(double depth
)
169 uint32_t r
= (uint32_t)(CLAMP(depth
, 0, 1) * 255.5);
171 return r
| (r
<< 8) | (r
<< 16) | (r
<< 24);
174 /* Clear currently bound buffers. */
175 static void r300_clear(struct pipe_context
* pipe
,
181 /* My notes about Zbuffer compression:
183 * 1) The zbuffer must be micro-tiled and whole microtiles must be
184 * written if compression is enabled. If microtiling is disabled,
187 * 2) There is ZMASK RAM which contains a compressed zbuffer.
188 * Each dword of the Z Mask contains compression information
189 * for 16 4x4 pixel tiles, that is 2 bits for each tile.
190 * On chips with 2 Z pipes, every other dword maps to a different
191 * pipe. On newer chipsets, there is a new compression mode
192 * with 8x8 pixel tiles per 2 bits.
194 * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
195 * it should look in the ZMASK RAM first before fetching from a real
198 * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
199 * during zbuffer reads instead of the value that is actually stored
200 * in the zbuffer memory. A pixel is in a cleared state when its ZMASK
201 * is equal to 0. Therefore, if you clear ZMASK with zeros, you may
202 * leave the zbuffer memory uninitialized, but then you must enable
203 * compression, so that the ZMASK RAM is actually used.
205 * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
206 * during zbuffer updates. A special decompressing operation should be
207 * used to fully decompress a zbuffer, which basically just stores all
208 * compressed tiles in ZMASK to the zbuffer memory.
210 * 6) For a 16-bit zbuffer, compression causes a hung with one or
211 * two samples and should not be used.
213 * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
214 * to avoid needless decompression.
216 * 8) Fastfill must not be used if reading of compressed Z data is disabled
217 * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
218 * i.e. it cannot be used to compress the zbuffer.
220 * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
225 struct r300_context
* r300
= r300_context(pipe
);
226 struct pipe_framebuffer_state
*fb
=
227 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
228 struct r300_hyperz_state
*hyperz
=
229 (struct r300_hyperz_state
*)r300
->hyperz_state
.state
;
230 uint32_t width
= fb
->width
;
231 uint32_t height
= fb
->height
;
232 uint32_t hyperz_dcv
= hyperz
->zb_depthclearvalue
;
234 /* Enable fast Z clear.
235 * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
236 if (buffers
& PIPE_CLEAR_DEPTHSTENCIL
) {
237 boolean zmask_clear
, hiz_clear
;
239 zmask_clear
= r300_fast_zclear_allowed(r300
);
240 hiz_clear
= r300_hiz_clear_allowed(r300
);
242 /* If we need Hyper-Z. */
243 if (zmask_clear
|| hiz_clear
) {
244 r300
->num_z_clears
++;
246 /* Try to obtain the access to Hyper-Z buffers if we don't have one. */
247 if (!r300
->hyperz_enabled
) {
248 r300
->hyperz_enabled
=
249 r300
->rws
->cs_request_feature(r300
->cs
,
250 RADEON_FID_HYPERZ_RAM_ACCESS
,
252 if (r300
->hyperz_enabled
) {
253 /* Need to emit HyperZ buffer regs for the first time. */
254 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
258 /* Setup Hyper-Z clears. */
259 if (r300
->hyperz_enabled
) {
260 DBG(r300
, DBG_HYPERZ
, "r300: Clear memory: %s%s\n",
261 zmask_clear
? "ZMASK " : "", hiz_clear
? "HIZ" : "");
264 hyperz_dcv
= hyperz
->zb_depthclearvalue
=
265 r300_depth_clear_value(fb
->zsbuf
->format
, depth
, stencil
);
267 r300_mark_atom_dirty(r300
, &r300
->zmask_clear
);
268 buffers
&= ~PIPE_CLEAR_DEPTHSTENCIL
;
272 r300
->hiz_clear_value
= r300_hiz_clear_value(depth
);
273 r300_mark_atom_dirty(r300
, &r300
->hiz_clear
);
279 /* Enable CBZB clear. */
280 if (r300_cbzb_clear_allowed(r300
, buffers
)) {
281 struct r300_surface
*surf
= r300_surface(fb
->cbufs
[0]);
283 hyperz
->zb_depthclearvalue
=
284 r300_depth_clear_cb_value(surf
->base
.format
, rgba
);
286 width
= surf
->cbzb_width
;
287 height
= surf
->cbzb_height
;
289 r300
->cbzb_clear
= TRUE
;
290 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
295 /* Clear using the blitter. */
296 r300_blitter_begin(r300
, R300_CLEAR
);
297 util_blitter_clear(r300
->blitter
,
301 buffers
, rgba
, depth
, stencil
);
302 r300_blitter_end(r300
);
303 } else if (r300
->zmask_clear
.dirty
|| r300
->hiz_clear
.dirty
) {
304 /* Just clear zmask and hiz now, this does not use the standard draw
306 /* Calculate zmask_clear and hiz_clear atom sizes. */
308 (r300
->zmask_clear
.dirty
? r300
->zmask_clear
.size
: 0) +
309 (r300
->hiz_clear
.dirty
? r300
->hiz_clear
.size
: 0) +
310 r300_get_num_cs_end_dwords(r300
);
312 /* Reserve CS space. */
313 if (dwords
> (RADEON_MAX_CMDBUF_DWORDS
- r300
->cs
->cdw
)) {
314 r300_flush(&r300
->context
, RADEON_FLUSH_ASYNC
, NULL
);
317 /* Emit clear packets. */
318 if (r300
->zmask_clear
.dirty
) {
319 r300_emit_zmask_clear(r300
, r300
->zmask_clear
.size
,
320 r300
->zmask_clear
.state
);
321 r300
->zmask_clear
.dirty
= FALSE
;
323 if (r300
->hiz_clear
.dirty
) {
324 r300_emit_hiz_clear(r300
, r300
->hiz_clear
.size
,
325 r300
->hiz_clear
.state
);
326 r300
->hiz_clear
.dirty
= FALSE
;
332 /* Disable CBZB clear. */
333 if (r300
->cbzb_clear
) {
334 r300
->cbzb_clear
= FALSE
;
335 hyperz
->zb_depthclearvalue
= hyperz_dcv
;
336 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
339 /* Enable fastfill and/or hiz.
341 * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
342 * looks if zmask/hiz is in use and programs hardware accordingly. */
343 if (r300
->zmask_in_use
|| r300
->hiz_in_use
) {
344 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
348 /* Clear a region of a color surface to a constant value. */
349 static void r300_clear_render_target(struct pipe_context
*pipe
,
350 struct pipe_surface
*dst
,
352 unsigned dstx
, unsigned dsty
,
353 unsigned width
, unsigned height
)
355 struct r300_context
*r300
= r300_context(pipe
);
357 r300_blitter_begin(r300
, R300_CLEAR_SURFACE
);
358 util_blitter_clear_render_target(r300
->blitter
, dst
, rgba
,
359 dstx
, dsty
, width
, height
);
360 r300_blitter_end(r300
);
363 /* Clear a region of a depth stencil surface. */
364 static void r300_clear_depth_stencil(struct pipe_context
*pipe
,
365 struct pipe_surface
*dst
,
366 unsigned clear_flags
,
369 unsigned dstx
, unsigned dsty
,
370 unsigned width
, unsigned height
)
372 struct r300_context
*r300
= r300_context(pipe
);
373 struct pipe_framebuffer_state
*fb
=
374 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
376 if (r300
->zmask_in_use
&& !r300
->locked_zbuffer
) {
377 if (fb
->zsbuf
->texture
== dst
->texture
) {
378 r300_decompress_zmask(r300
);
382 /* XXX Do not decompress ZMask of the currently-set zbuffer. */
383 r300_blitter_begin(r300
, R300_CLEAR_SURFACE
);
384 util_blitter_clear_depth_stencil(r300
->blitter
, dst
, clear_flags
, depth
, stencil
,
385 dstx
, dsty
, width
, height
);
386 r300_blitter_end(r300
);
389 void r300_decompress_zmask(struct r300_context
*r300
)
391 struct pipe_framebuffer_state
*fb
=
392 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
394 if (!r300
->zmask_in_use
|| r300
->locked_zbuffer
)
397 r300
->zmask_decompress
= TRUE
;
398 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
400 r300_blitter_begin(r300
, R300_DECOMPRESS
);
401 util_blitter_clear_depth_custom(r300
->blitter
, fb
->width
, fb
->height
, 0,
402 r300
->dsa_decompress_zmask
);
403 r300_blitter_end(r300
);
405 r300
->zmask_decompress
= FALSE
;
406 r300
->zmask_in_use
= FALSE
;
407 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
410 void r300_decompress_zmask_locked_unsafe(struct r300_context
*r300
)
412 struct pipe_framebuffer_state fb
= {0};
413 fb
.width
= r300
->locked_zbuffer
->width
;
414 fb
.height
= r300
->locked_zbuffer
->height
;
416 fb
.zsbuf
= r300
->locked_zbuffer
;
418 r300
->context
.set_framebuffer_state(&r300
->context
, &fb
);
419 r300_decompress_zmask(r300
);
422 void r300_decompress_zmask_locked(struct r300_context
*r300
)
424 struct pipe_framebuffer_state saved_fb
= {0};
426 util_copy_framebuffer_state(&saved_fb
, r300
->fb_state
.state
);
427 r300_decompress_zmask_locked_unsafe(r300
);
428 r300
->context
.set_framebuffer_state(&r300
->context
, &saved_fb
);
429 util_unreference_framebuffer_state(&saved_fb
);
431 pipe_surface_reference(&r300
->locked_zbuffer
, NULL
);
434 /* Copy a block of pixels from one surface to another using HW. */
435 static void r300_hw_copy_region(struct pipe_context
* pipe
,
436 struct pipe_resource
*dst
,
438 unsigned dstx
, unsigned dsty
, unsigned dstz
,
439 struct pipe_resource
*src
,
441 const struct pipe_box
*src_box
)
443 struct r300_context
* r300
= r300_context(pipe
);
445 r300_blitter_begin(r300
, R300_COPY
);
446 util_blitter_copy_region(r300
->blitter
, dst
, dst_level
, dstx
, dsty
, dstz
,
447 src
, src_level
, src_box
, TRUE
);
448 r300_blitter_end(r300
);
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 r300_context
*r300
= r300_context(pipe
);
461 struct pipe_framebuffer_state
*fb
=
462 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
463 struct pipe_resource old_src
= *src
;
464 struct pipe_resource old_dst
= *dst
;
465 struct pipe_resource new_src
= old_src
;
466 struct pipe_resource new_dst
= old_dst
;
467 const struct util_format_description
*desc
=
468 util_format_description(dst
->format
);
471 /* Fallback for buffers. */
472 if (dst
->target
== PIPE_BUFFER
&& src
->target
== PIPE_BUFFER
) {
473 util_resource_copy_region(pipe
, dst
, dst_level
, dstx
, dsty
, dstz
,
474 src
, src_level
, src_box
);
478 if (r300
->zmask_in_use
&& !r300
->locked_zbuffer
) {
479 if (fb
->zsbuf
->texture
== src
||
480 fb
->zsbuf
->texture
== dst
) {
481 r300_decompress_zmask(r300
);
485 /* Handle non-renderable plain formats. */
486 if (desc
->layout
== UTIL_FORMAT_LAYOUT_PLAIN
&&
487 (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
||
488 !pipe
->screen
->is_format_supported(pipe
->screen
,
489 src
->format
, src
->target
,
491 PIPE_BIND_SAMPLER_VIEW
) ||
492 !pipe
->screen
->is_format_supported(pipe
->screen
,
493 dst
->format
, dst
->target
,
495 PIPE_BIND_RENDER_TARGET
))) {
496 switch (util_format_get_blocksize(old_dst
.format
)) {
498 new_dst
.format
= PIPE_FORMAT_I8_UNORM
;
501 new_dst
.format
= PIPE_FORMAT_B4G4R4A4_UNORM
;
504 new_dst
.format
= PIPE_FORMAT_B8G8R8A8_UNORM
;
507 new_dst
.format
= PIPE_FORMAT_R16G16B16A16_UNORM
;
510 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
511 "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
512 util_format_short_name(dst
->format
));
514 new_src
.format
= new_dst
.format
;
517 /* Handle compressed formats. */
518 if (desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
||
519 desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
520 switch (util_format_get_blocksize(old_dst
.format
)) {
523 * we set 1 pixel = 2 bytes ===> 4 times larger pixels. */
524 new_dst
.format
= PIPE_FORMAT_B4G4R4A4_UNORM
;
528 * we set 1 pixel = 4 bytes ===> 4 times larger pixels. */
529 new_dst
.format
= PIPE_FORMAT_B8G8R8A8_UNORM
;
533 /* Since the pixels are 4 times larger, we must decrease
534 * the image size and the coordinates 4 times. */
535 new_src
.format
= new_dst
.format
;
536 new_dst
.height0
= (new_dst
.height0
+ 3) / 4;
537 new_src
.height0
= (new_src
.height0
+ 3) / 4;
541 box
.height
= (box
.height
+ 3) / 4;
545 if (old_src
.format
!= new_src
.format
)
546 r300_resource_set_properties(pipe
->screen
, src
, 0, &new_src
);
547 if (old_dst
.format
!= new_dst
.format
)
548 r300_resource_set_properties(pipe
->screen
, dst
, 0, &new_dst
);
550 r300_hw_copy_region(pipe
, dst
, dst_level
, dstx
, dsty
, dstz
,
551 src
, src_level
, src_box
);
553 if (old_src
.format
!= new_src
.format
)
554 r300_resource_set_properties(pipe
->screen
, src
, 0, &old_src
);
555 if (old_dst
.format
!= new_dst
.format
)
556 r300_resource_set_properties(pipe
->screen
, dst
, 0, &old_dst
);
559 void r300_init_blit_functions(struct r300_context
*r300
)
561 r300
->context
.clear
= r300_clear
;
562 r300
->context
.clear_render_target
= r300_clear_render_target
;
563 r300
->context
.clear_depth_stencil
= r300_clear_depth_stencil
;
564 r300
->context
.resource_copy_region
= r300_resource_copy_region
;