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_CLEAR_SURFACE
= 2,
38 static void r300_blitter_begin(struct r300_context
* r300
, enum r300_blitter_op op
)
40 if (r300
->query_current
) {
41 r300
->blitter_saved_query
= r300
->query_current
;
42 r300_stop_query(r300
);
45 /* Yeah we have to save all those states to ensure the blitter operation
46 * is really transparent. The states will be restored by the blitter once
48 util_blitter_save_blend(r300
->blitter
, r300
->blend_state
.state
);
49 util_blitter_save_depth_stencil_alpha(r300
->blitter
, r300
->dsa_state
.state
);
50 util_blitter_save_stencil_ref(r300
->blitter
, &(r300
->stencil_ref
));
51 util_blitter_save_rasterizer(r300
->blitter
, r300
->rs_state
.state
);
52 util_blitter_save_fragment_shader(r300
->blitter
, r300
->fs
.state
);
53 util_blitter_save_vertex_shader(r300
->blitter
, r300
->vs_state
.state
);
54 util_blitter_save_viewport(r300
->blitter
, &r300
->viewport
);
55 util_blitter_save_clip(r300
->blitter
, (struct pipe_clip_state
*)r300
->clip_state
.state
);
56 util_blitter_save_vertex_elements(r300
->blitter
, r300
->velems
);
57 util_blitter_save_vertex_buffers(r300
->blitter
, r300
->vbuf_mgr
->nr_vertex_buffers
,
58 r300
->vbuf_mgr
->vertex_buffer
);
60 if (op
& (R300_CLEAR_SURFACE
| R300_COPY
)) {
61 util_blitter_save_framebuffer(r300
->blitter
, r300
->fb_state
.state
);
65 struct r300_textures_state
* state
=
66 (struct r300_textures_state
*)r300
->textures_state
.state
;
68 util_blitter_save_fragment_sampler_states(
69 r300
->blitter
, state
->sampler_state_count
,
70 (void**)state
->sampler_states
);
72 util_blitter_save_fragment_sampler_views(
73 r300
->blitter
, state
->sampler_view_count
,
74 (struct pipe_sampler_view
**)state
->sampler_views
);
78 static void r300_blitter_end(struct r300_context
*r300
)
80 if (r300
->blitter_saved_query
) {
81 r300_resume_query(r300
, r300
->blitter_saved_query
);
82 r300
->blitter_saved_query
= NULL
;
86 static uint32_t r300_depth_clear_cb_value(enum pipe_format format
,
90 util_pack_color(rgba
, format
, &uc
);
92 if (util_format_get_blocksizebits(format
) == 32)
95 return uc
.us
| (uc
.us
<< 16);
98 static boolean
r300_cbzb_clear_allowed(struct r300_context
*r300
,
99 unsigned clear_buffers
)
101 struct pipe_framebuffer_state
*fb
=
102 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
104 /* Only color clear allowed, and only one colorbuffer. */
105 if (clear_buffers
!= PIPE_CLEAR_COLOR
|| fb
->nr_cbufs
!= 1)
108 return r300_surface(fb
->cbufs
[0])->cbzb_allowed
;
111 static boolean
r300_fast_zclear_allowed(struct r300_context
*r300
)
113 struct pipe_framebuffer_state
*fb
=
114 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
116 return r300_resource(fb
->zsbuf
->texture
)->tex
.zmask_dwords
[fb
->zsbuf
->u
.tex
.level
];
119 static boolean
r300_hiz_clear_allowed(struct r300_context
*r300
)
121 struct pipe_framebuffer_state
*fb
=
122 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
124 return r300_resource(fb
->zsbuf
->texture
)->tex
.hiz_dwords
[fb
->zsbuf
->u
.tex
.level
];
127 static uint32_t r300_depth_clear_value(enum pipe_format format
,
128 double depth
, unsigned stencil
)
131 case PIPE_FORMAT_Z16_UNORM
:
132 case PIPE_FORMAT_X8Z24_UNORM
:
133 return util_pack_z(format
, depth
);
135 case PIPE_FORMAT_S8_USCALED_Z24_UNORM
:
136 return util_pack_z_stencil(format
, depth
, stencil
);
144 static uint32_t r300_hiz_clear_value(double depth
)
146 uint32_t r
= (uint32_t)(CLAMP(depth
, 0, 1) * 255.5);
148 return r
| (r
<< 8) | (r
<< 16) | (r
<< 24);
151 /* Clear currently bound buffers. */
152 static void r300_clear(struct pipe_context
* pipe
,
158 /* My notes about Zbuffer compression:
160 * 1) The zbuffer must be micro-tiled and whole microtiles must be
161 * written if compression is enabled. If microtiling is disabled,
164 * 2) There is ZMASK RAM which contains a compressed zbuffer.
165 * Each dword of the Z Mask contains compression information
166 * for 16 4x4 pixel tiles, that is 2 bits for each tile.
167 * On chips with 2 Z pipes, every other dword maps to a different
168 * pipe. On newer chipsets, there is a new compression mode
169 * with 8x8 pixel tiles per 2 bits.
171 * 3) The FASTFILL bit has nothing to do with filling. It only tells hw
172 * it should look in the ZMASK RAM first before fetching from a real
175 * 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
176 * during zbuffer reads instead of the value that is actually stored
177 * in the zbuffer memory. A pixel is in a cleared state when its ZMASK
178 * is equal to 0. Therefore, if you clear ZMASK with zeros, you may
179 * leave the zbuffer memory uninitialized, but then you must enable
180 * compression, so that the ZMASK RAM is actually used.
182 * 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
183 * during zbuffer updates. A special decompressing operation should be
184 * used to fully decompress a zbuffer, which basically just stores all
185 * compressed tiles in ZMASK to the zbuffer memory.
187 * 6) For a 16-bit zbuffer, compression causes a hung with one or
188 * two samples and should not be used.
190 * 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
191 * to avoid needless decompression.
193 * 8) Fastfill must not be used if reading of compressed Z data is disabled
194 * and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
195 * i.e. it cannot be used to compress the zbuffer.
197 * 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
202 struct r300_context
* r300
= r300_context(pipe
);
203 struct pipe_framebuffer_state
*fb
=
204 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
205 struct r300_hyperz_state
*hyperz
=
206 (struct r300_hyperz_state
*)r300
->hyperz_state
.state
;
207 uint32_t width
= fb
->width
;
208 uint32_t height
= fb
->height
;
209 boolean can_hyperz
= r300
->rws
->get_value(r300
->rws
, RADEON_VID_CAN_HYPERZ
);
210 uint32_t hyperz_dcv
= hyperz
->zb_depthclearvalue
;
212 /* Enable fast Z clear.
213 * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
214 if ((buffers
& PIPE_CLEAR_DEPTHSTENCIL
) && can_hyperz
) {
215 if (r300_fast_zclear_allowed(r300
)) {
216 hyperz_dcv
= hyperz
->zb_depthclearvalue
=
217 r300_depth_clear_value(fb
->zsbuf
->format
, depth
, stencil
);
219 r300_mark_atom_dirty(r300
, &r300
->zmask_clear
);
220 buffers
&= ~PIPE_CLEAR_DEPTHSTENCIL
;
223 if (r300_hiz_clear_allowed(r300
)) {
224 r300
->hiz_clear_value
= r300_hiz_clear_value(depth
);
225 r300_mark_atom_dirty(r300
, &r300
->hiz_clear
);
229 /* Enable CBZB clear. */
230 if (r300_cbzb_clear_allowed(r300
, buffers
)) {
231 struct r300_surface
*surf
= r300_surface(fb
->cbufs
[0]);
233 hyperz
->zb_depthclearvalue
=
234 r300_depth_clear_cb_value(surf
->base
.format
, rgba
);
236 width
= surf
->cbzb_width
;
237 height
= surf
->cbzb_height
;
239 r300
->cbzb_clear
= TRUE
;
240 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
245 /* Clear using the blitter. */
246 r300_blitter_begin(r300
, R300_CLEAR
);
247 util_blitter_clear(r300
->blitter
,
251 buffers
, rgba
, depth
, stencil
);
252 r300_blitter_end(r300
);
253 } else if (r300
->zmask_clear
.dirty
|| r300
->hiz_clear
.dirty
) {
254 /* Just clear zmask and hiz now, this does not use the standard draw
256 /* Calculate zmask_clear and hiz_clear atom sizes. */
258 (r300
->zmask_clear
.dirty
? r300
->zmask_clear
.size
: 0) +
259 (r300
->hiz_clear
.dirty
? r300
->hiz_clear
.size
: 0) +
260 r300_get_num_cs_end_dwords(r300
);
262 /* Reserve CS space. */
263 if (dwords
> (RADEON_MAX_CMDBUF_DWORDS
- r300
->cs
->cdw
)) {
264 r300_flush(&r300
->context
, RADEON_FLUSH_ASYNC
, NULL
);
267 /* Emit clear packets. */
268 if (r300
->zmask_clear
.dirty
) {
269 r300_emit_zmask_clear(r300
, r300
->zmask_clear
.size
,
270 r300
->zmask_clear
.state
);
271 r300
->zmask_clear
.dirty
= FALSE
;
273 if (r300
->hiz_clear
.dirty
) {
274 r300_emit_hiz_clear(r300
, r300
->hiz_clear
.size
,
275 r300
->hiz_clear
.state
);
276 r300
->hiz_clear
.dirty
= FALSE
;
282 /* Disable CBZB clear. */
283 if (r300
->cbzb_clear
) {
284 r300
->cbzb_clear
= FALSE
;
285 hyperz
->zb_depthclearvalue
= hyperz_dcv
;
286 r300_mark_fb_state_dirty(r300
, R300_CHANGED_HYPERZ_FLAG
);
289 /* Enable fastfill and/or hiz.
291 * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
292 * looks if zmask/hiz is in use and programs hardware accordingly. */
293 if (r300
->zmask_in_use
|| r300
->hiz_in_use
) {
294 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
298 /* Clear a region of a color surface to a constant value. */
299 static void r300_clear_render_target(struct pipe_context
*pipe
,
300 struct pipe_surface
*dst
,
302 unsigned dstx
, unsigned dsty
,
303 unsigned width
, unsigned height
)
305 struct r300_context
*r300
= r300_context(pipe
);
307 r300_blitter_begin(r300
, R300_CLEAR_SURFACE
);
308 util_blitter_clear_render_target(r300
->blitter
, dst
, rgba
,
309 dstx
, dsty
, width
, height
);
310 r300_blitter_end(r300
);
313 /* Clear a region of a depth stencil surface. */
314 static void r300_clear_depth_stencil(struct pipe_context
*pipe
,
315 struct pipe_surface
*dst
,
316 unsigned clear_flags
,
319 unsigned dstx
, unsigned dsty
,
320 unsigned width
, unsigned height
)
322 struct r300_context
*r300
= r300_context(pipe
);
323 struct pipe_framebuffer_state
*fb
=
324 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
326 if (r300
->zmask_in_use
&& !r300
->hyperz_locked
) {
327 if (fb
->zsbuf
->texture
== dst
->texture
) {
328 r300_decompress_zmask(r300
);
332 /* XXX Do not decompress ZMask of the currently-set zbuffer. */
333 r300_blitter_begin(r300
, R300_CLEAR_SURFACE
);
334 util_blitter_clear_depth_stencil(r300
->blitter
, dst
, clear_flags
, depth
, stencil
,
335 dstx
, dsty
, width
, height
);
336 r300_blitter_end(r300
);
339 void r300_decompress_zmask(struct r300_context
*r300
)
341 struct pipe_framebuffer_state
*fb
=
342 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
344 if (!r300
->zmask_in_use
|| r300
->hyperz_locked
)
347 r300
->zmask_decompress
= TRUE
;
348 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
350 r300_blitter_begin(r300
, R300_CLEAR
);
351 util_blitter_clear_depth_custom(r300
->blitter
, fb
->width
, fb
->height
, 0,
352 r300
->dsa_decompress_zmask
);
353 r300_blitter_end(r300
);
355 r300
->zmask_decompress
= FALSE
;
356 r300
->zmask_in_use
= FALSE
;
357 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
360 void r300_decompress_zmask_locked_unsafe(struct r300_context
*r300
)
362 struct pipe_framebuffer_state fb
= {0};
363 fb
.width
= r300
->locked_zbuffer
->width
;
364 fb
.height
= r300
->locked_zbuffer
->height
;
366 fb
.zsbuf
= r300
->locked_zbuffer
;
368 r300
->context
.set_framebuffer_state(&r300
->context
, &fb
);
369 r300_decompress_zmask(r300
);
372 void r300_decompress_zmask_locked(struct r300_context
*r300
)
374 struct pipe_framebuffer_state saved_fb
= {0};
376 util_copy_framebuffer_state(&saved_fb
, r300
->fb_state
.state
);
377 r300_decompress_zmask_locked_unsafe(r300
);
378 r300
->context
.set_framebuffer_state(&r300
->context
, &saved_fb
);
379 util_unreference_framebuffer_state(&saved_fb
);
382 /* Copy a block of pixels from one surface to another using HW. */
383 static void r300_hw_copy_region(struct pipe_context
* pipe
,
384 struct pipe_resource
*dst
,
386 unsigned dstx
, unsigned dsty
, unsigned dstz
,
387 struct pipe_resource
*src
,
389 const struct pipe_box
*src_box
)
391 struct r300_context
* r300
= r300_context(pipe
);
393 r300_blitter_begin(r300
, R300_COPY
);
394 util_blitter_copy_region(r300
->blitter
, dst
, dst_level
, dstx
, dsty
, dstz
,
395 src
, src_level
, src_box
, TRUE
);
396 r300_blitter_end(r300
);
399 /* Copy a block of pixels from one surface to another. */
400 static void r300_resource_copy_region(struct pipe_context
*pipe
,
401 struct pipe_resource
*dst
,
403 unsigned dstx
, unsigned dsty
, unsigned dstz
,
404 struct pipe_resource
*src
,
406 const struct pipe_box
*src_box
)
408 struct r300_context
*r300
= r300_context(pipe
);
409 struct pipe_framebuffer_state
*fb
=
410 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
411 struct pipe_resource old_src
= *src
;
412 struct pipe_resource old_dst
= *dst
;
413 struct pipe_resource new_src
= old_src
;
414 struct pipe_resource new_dst
= old_dst
;
415 const struct util_format_description
*desc
=
416 util_format_description(dst
->format
);
419 /* Fallback for buffers. */
420 if (dst
->target
== PIPE_BUFFER
&& src
->target
== PIPE_BUFFER
) {
421 util_resource_copy_region(pipe
, dst
, dst_level
, dstx
, dsty
, dstz
,
422 src
, src_level
, src_box
);
426 if (r300
->zmask_in_use
&& !r300
->hyperz_locked
) {
427 if (fb
->zsbuf
->texture
== src
||
428 fb
->zsbuf
->texture
== dst
) {
429 r300_decompress_zmask(r300
);
433 /* Handle non-renderable plain formats. */
434 if (desc
->layout
== UTIL_FORMAT_LAYOUT_PLAIN
&&
435 (desc
->colorspace
== UTIL_FORMAT_COLORSPACE_SRGB
||
436 !pipe
->screen
->is_format_supported(pipe
->screen
,
437 src
->format
, src
->target
,
439 PIPE_BIND_SAMPLER_VIEW
) ||
440 !pipe
->screen
->is_format_supported(pipe
->screen
,
441 dst
->format
, dst
->target
,
443 PIPE_BIND_RENDER_TARGET
))) {
444 switch (util_format_get_blocksize(old_dst
.format
)) {
446 new_dst
.format
= PIPE_FORMAT_I8_UNORM
;
449 new_dst
.format
= PIPE_FORMAT_B4G4R4A4_UNORM
;
452 new_dst
.format
= PIPE_FORMAT_B8G8R8A8_UNORM
;
455 new_dst
.format
= PIPE_FORMAT_R16G16B16A16_UNORM
;
458 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
459 "r300: surface_copy: Software fallback doesn't work for tiled textures.\n",
460 util_format_short_name(dst
->format
));
462 new_src
.format
= new_dst
.format
;
465 /* Handle compressed formats. */
466 if (desc
->layout
== UTIL_FORMAT_LAYOUT_S3TC
||
467 desc
->layout
== UTIL_FORMAT_LAYOUT_RGTC
) {
468 switch (util_format_get_blocksize(old_dst
.format
)) {
471 * we set 1 pixel = 2 bytes ===> 4 times larger pixels. */
472 new_dst
.format
= PIPE_FORMAT_B4G4R4A4_UNORM
;
476 * we set 1 pixel = 4 bytes ===> 4 times larger pixels. */
477 new_dst
.format
= PIPE_FORMAT_B8G8R8A8_UNORM
;
481 /* Since the pixels are 4 times larger, we must decrease
482 * the image size and the coordinates 4 times. */
483 new_src
.format
= new_dst
.format
;
484 new_dst
.height0
= (new_dst
.height0
+ 3) / 4;
485 new_src
.height0
= (new_src
.height0
+ 3) / 4;
489 box
.height
= (box
.height
+ 3) / 4;
493 if (old_src
.format
!= new_src
.format
)
494 r300_resource_set_properties(pipe
->screen
, src
, 0, &new_src
);
495 if (old_dst
.format
!= new_dst
.format
)
496 r300_resource_set_properties(pipe
->screen
, dst
, 0, &new_dst
);
498 r300_hw_copy_region(pipe
, dst
, dst_level
, dstx
, dsty
, dstz
,
499 src
, src_level
, src_box
);
501 if (old_src
.format
!= new_src
.format
)
502 r300_resource_set_properties(pipe
->screen
, src
, 0, &old_src
);
503 if (old_dst
.format
!= new_dst
.format
)
504 r300_resource_set_properties(pipe
->screen
, dst
, 0, &old_dst
);
507 void r300_init_blit_functions(struct r300_context
*r300
)
509 r300
->context
.clear
= r300_clear
;
510 r300
->context
.clear_render_target
= r300_clear_render_target
;
511 r300
->context
.clear_depth_stencil
= r300_clear_depth_stencil
;
512 r300
->context
.resource_copy_region
= r300_resource_copy_region
;