2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
24 #include "draw/draw_context.h"
26 #include "util/u_framebuffer.h"
27 #include "util/u_half.h"
28 #include "util/u_helpers.h"
29 #include "util/u_math.h"
30 #include "util/u_mm.h"
31 #include "util/u_memory.h"
32 #include "util/u_pack_color.h"
33 #include "util/u_transfer.h"
35 #include "tgsi/tgsi_parse.h"
37 #include "pipe/p_config.h"
40 #include "r300_context.h"
41 #include "r300_emit.h"
43 #include "r300_screen.h"
44 #include "r300_screen_buffer.h"
45 #include "r300_state_inlines.h"
47 #include "r300_texture.h"
50 /* r300_state: Functions used to intialize state context by translating
51 * Gallium state objects into semi-native r300 state objects. */
53 #define UPDATE_STATE(cso, atom) \
54 if (cso != atom.state) { \
56 r300_mark_atom_dirty(r300, &(atom)); \
59 static boolean
blend_discard_if_src_alpha_0(unsigned srcRGB
, unsigned srcA
,
60 unsigned dstRGB
, unsigned dstA
)
62 /* If the blend equation is ADD or REVERSE_SUBTRACT,
63 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
64 * will not be changed.
65 * Notice that the dst factors are the src factors inverted. */
66 return (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
67 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
68 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
69 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
70 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
71 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
72 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
73 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
74 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
75 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
76 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
77 dstA
== PIPE_BLENDFACTOR_ONE
);
80 static boolean
blend_discard_if_src_alpha_1(unsigned srcRGB
, unsigned srcA
,
81 unsigned dstRGB
, unsigned dstA
)
83 /* If the blend equation is ADD or REVERSE_SUBTRACT,
84 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
85 * will not be changed.
86 * Notice that the dst factors are the src factors inverted. */
87 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
88 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
89 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
90 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
91 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
92 (dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
93 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
94 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
95 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
96 dstA
== PIPE_BLENDFACTOR_ONE
);
99 static boolean
blend_discard_if_src_color_0(unsigned srcRGB
, unsigned srcA
,
100 unsigned dstRGB
, unsigned dstA
)
102 /* If the blend equation is ADD or REVERSE_SUBTRACT,
103 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
104 * will not be changed.
105 * Notice that the dst factors are the src factors inverted. */
106 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
107 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
108 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
109 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
110 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
111 (dstA
== PIPE_BLENDFACTOR_ONE
);
114 static boolean
blend_discard_if_src_color_1(unsigned srcRGB
, unsigned srcA
,
115 unsigned dstRGB
, unsigned dstA
)
117 /* If the blend equation is ADD or REVERSE_SUBTRACT,
118 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
119 * will not be changed.
120 * Notice that the dst factors are the src factors inverted. */
121 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
122 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
123 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
124 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
125 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
126 (dstA
== PIPE_BLENDFACTOR_ONE
);
129 static boolean
blend_discard_if_src_alpha_color_0(unsigned srcRGB
, unsigned srcA
,
130 unsigned dstRGB
, unsigned dstA
)
132 /* If the blend equation is ADD or REVERSE_SUBTRACT,
133 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
134 * the colorbuffer will not be changed.
135 * Notice that the dst factors are the src factors inverted. */
136 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
137 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
138 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
139 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
140 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
141 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
142 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
143 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
144 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
145 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
146 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
147 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
148 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
149 dstA
== PIPE_BLENDFACTOR_ONE
);
152 static boolean
blend_discard_if_src_alpha_color_1(unsigned srcRGB
, unsigned srcA
,
153 unsigned dstRGB
, unsigned dstA
)
155 /* If the blend equation is ADD or REVERSE_SUBTRACT,
156 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
157 * the colorbuffer will not be changed.
158 * Notice that the dst factors are the src factors inverted. */
159 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
160 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
161 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
162 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
163 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
164 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
165 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
166 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
167 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
168 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
169 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
170 dstA
== PIPE_BLENDFACTOR_ONE
);
173 static unsigned blend_discard_conditionally(unsigned eqRGB
, unsigned eqA
,
174 unsigned dstRGB
, unsigned dstA
,
175 unsigned srcRGB
, unsigned srcA
)
177 unsigned blend_control
= 0;
179 /* Optimization: discard pixels which don't change the colorbuffer.
181 * The code below is non-trivial and some math is involved.
183 * Discarding pixels must be disabled when FP16 AA is enabled.
184 * This is a hardware bug. Also, this implementation wouldn't work
185 * with FP blending enabled and equation clamping disabled.
187 * Equations other than ADD are rarely used and therefore won't be
189 if ((eqRGB
== PIPE_BLEND_ADD
|| eqRGB
== PIPE_BLEND_REVERSE_SUBTRACT
) &&
190 (eqA
== PIPE_BLEND_ADD
|| eqA
== PIPE_BLEND_REVERSE_SUBTRACT
)) {
192 * REVERSE_SUBTRACT: Y-X
195 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
196 * then CB will not be changed.
198 * Given the srcFactor and dstFactor variables, we can derive
199 * what src and dst should be equal to and discard appropriate
202 if (blend_discard_if_src_alpha_0(srcRGB
, srcA
, dstRGB
, dstA
)) {
203 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0
;
204 } else if (blend_discard_if_src_alpha_1(srcRGB
, srcA
,
206 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1
;
207 } else if (blend_discard_if_src_color_0(srcRGB
, srcA
,
209 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0
;
210 } else if (blend_discard_if_src_color_1(srcRGB
, srcA
,
212 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1
;
213 } else if (blend_discard_if_src_alpha_color_0(srcRGB
, srcA
,
216 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0
;
217 } else if (blend_discard_if_src_alpha_color_1(srcRGB
, srcA
,
220 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1
;
223 return blend_control
;
226 /* The hardware colormask is clunky a must be swizzled depending on the format.
227 * This was figured out by trial-and-error. */
228 static unsigned bgra_cmask(unsigned mask
)
230 return ((mask
& PIPE_MASK_R
) << 2) |
231 ((mask
& PIPE_MASK_B
) >> 2) |
232 (mask
& (PIPE_MASK_G
| PIPE_MASK_A
));
235 static unsigned rgba_cmask(unsigned mask
)
237 return mask
& PIPE_MASK_RGBA
;
240 static unsigned rrrr_cmask(unsigned mask
)
242 return (mask
& PIPE_MASK_R
) |
243 ((mask
& PIPE_MASK_R
) << 1) |
244 ((mask
& PIPE_MASK_R
) << 2) |
245 ((mask
& PIPE_MASK_R
) << 3);
248 static unsigned aaaa_cmask(unsigned mask
)
250 return ((mask
& PIPE_MASK_A
) >> 3) |
251 ((mask
& PIPE_MASK_A
) >> 2) |
252 ((mask
& PIPE_MASK_A
) >> 1) |
253 (mask
& PIPE_MASK_A
);
256 static unsigned grrg_cmask(unsigned mask
)
258 return ((mask
& PIPE_MASK_R
) << 1) |
259 ((mask
& PIPE_MASK_R
) << 2) |
260 ((mask
& PIPE_MASK_G
) >> 1) |
261 ((mask
& PIPE_MASK_G
) << 2);
264 static unsigned arra_cmask(unsigned mask
)
266 return ((mask
& PIPE_MASK_R
) << 1) |
267 ((mask
& PIPE_MASK_R
) << 2) |
268 ((mask
& PIPE_MASK_A
) >> 3) |
269 (mask
& PIPE_MASK_A
);
272 static unsigned blend_read_enable(unsigned eqRGB
, unsigned eqA
,
273 unsigned dstRGB
, unsigned dstA
,
274 unsigned srcRGB
, unsigned srcA
,
275 boolean src_alpha_optz
)
277 unsigned blend_control
= 0;
279 /* Optimization: some operations do not require the destination color.
281 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
282 * otherwise blending gives incorrect results. It seems to be
284 if (eqRGB
== PIPE_BLEND_MIN
|| eqA
== PIPE_BLEND_MIN
||
285 eqRGB
== PIPE_BLEND_MAX
|| eqA
== PIPE_BLEND_MAX
||
286 dstRGB
!= PIPE_BLENDFACTOR_ZERO
||
287 dstA
!= PIPE_BLENDFACTOR_ZERO
||
288 srcRGB
== PIPE_BLENDFACTOR_DST_COLOR
||
289 srcRGB
== PIPE_BLENDFACTOR_DST_ALPHA
||
290 srcRGB
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
291 srcRGB
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
292 srcA
== PIPE_BLENDFACTOR_DST_COLOR
||
293 srcA
== PIPE_BLENDFACTOR_DST_ALPHA
||
294 srcA
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
295 srcA
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
296 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
) {
297 /* Enable reading from the colorbuffer. */
298 blend_control
|= R300_READ_ENABLE
;
300 if (src_alpha_optz
) {
301 /* Optimization: Depending on incoming pixels, we can
302 * conditionally disable the reading in hardware... */
303 if (eqRGB
!= PIPE_BLEND_MIN
&& eqA
!= PIPE_BLEND_MIN
&&
304 eqRGB
!= PIPE_BLEND_MAX
&& eqA
!= PIPE_BLEND_MAX
) {
305 /* Disable reading if SRC_ALPHA == 0. */
306 if ((dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
307 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
308 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
309 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
310 dstA
== PIPE_BLENDFACTOR_ZERO
) &&
311 (srcRGB
!= PIPE_BLENDFACTOR_DST_COLOR
&&
312 srcRGB
!= PIPE_BLENDFACTOR_DST_ALPHA
&&
313 srcRGB
!= PIPE_BLENDFACTOR_INV_DST_COLOR
&&
314 srcRGB
!= PIPE_BLENDFACTOR_INV_DST_ALPHA
)) {
315 blend_control
|= R500_SRC_ALPHA_0_NO_READ
;
318 /* Disable reading if SRC_ALPHA == 1. */
319 if ((dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
320 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
321 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
322 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
323 dstA
== PIPE_BLENDFACTOR_ZERO
) &&
324 (srcRGB
!= PIPE_BLENDFACTOR_DST_COLOR
&&
325 srcRGB
!= PIPE_BLENDFACTOR_DST_ALPHA
&&
326 srcRGB
!= PIPE_BLENDFACTOR_INV_DST_COLOR
&&
327 srcRGB
!= PIPE_BLENDFACTOR_INV_DST_ALPHA
)) {
328 blend_control
|= R500_SRC_ALPHA_1_NO_READ
;
333 return blend_control
;
336 /* Create a new blend state based on the CSO blend state.
338 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
339 static void* r300_create_blend_state(struct pipe_context
* pipe
,
340 const struct pipe_blend_state
* state
)
342 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
343 struct r300_blend_state
* blend
= CALLOC_STRUCT(r300_blend_state
);
344 uint32_t blend_control
= 0; /* R300_RB3D_CBLEND: 0x4e04 */
345 uint32_t blend_control_noclamp
= 0; /* R300_RB3D_CBLEND: 0x4e04 */
346 uint32_t blend_control_noalpha
= 0; /* R300_RB3D_CBLEND: 0x4e04 */
347 uint32_t blend_control_noalpha_noclamp
= 0; /* R300_RB3D_CBLEND: 0x4e04 */
348 uint32_t alpha_blend_control
= 0; /* R300_RB3D_ABLEND: 0x4e08 */
349 uint32_t alpha_blend_control_noclamp
= 0; /* R300_RB3D_ABLEND: 0x4e08 */
350 uint32_t alpha_blend_control_noalpha
= 0; /* R300_RB3D_ABLEND: 0x4e08 */
351 uint32_t alpha_blend_control_noalpha_noclamp
= 0; /* R300_RB3D_ABLEND: 0x4e08 */
352 uint32_t rop
= 0; /* R300_RB3D_ROPCNTL: 0x4e18 */
353 uint32_t dither
= 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */
356 const unsigned eqRGB
= state
->rt
[0].rgb_func
;
357 const unsigned srcRGB
= state
->rt
[0].rgb_src_factor
;
358 const unsigned dstRGB
= state
->rt
[0].rgb_dst_factor
;
360 const unsigned eqA
= state
->rt
[0].alpha_func
;
361 const unsigned srcA
= state
->rt
[0].alpha_src_factor
;
362 const unsigned dstA
= state
->rt
[0].alpha_dst_factor
;
364 unsigned srcRGBX
= srcRGB
;
365 unsigned dstRGBX
= dstRGB
;
368 blend
->state
= *state
;
370 /* force DST_ALPHA to ONE where we can */
372 case PIPE_BLENDFACTOR_DST_ALPHA
:
373 srcRGBX
= PIPE_BLENDFACTOR_ONE
;
375 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
376 srcRGBX
= PIPE_BLENDFACTOR_ZERO
;
381 case PIPE_BLENDFACTOR_DST_ALPHA
:
382 dstRGBX
= PIPE_BLENDFACTOR_ONE
;
384 case PIPE_BLENDFACTOR_INV_DST_ALPHA
:
385 dstRGBX
= PIPE_BLENDFACTOR_ZERO
;
389 /* Get blending register values. */
390 if (state
->rt
[0].blend_enable
) {
391 unsigned blend_eq
, blend_eq_noclamp
;
393 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
394 * this is just the crappy D3D naming */
395 blend_control
= blend_control_noclamp
=
396 R300_ALPHA_BLEND_ENABLE
|
397 ( r300_translate_blend_factor(srcRGB
) << R300_SRC_BLEND_SHIFT
) |
398 ( r300_translate_blend_factor(dstRGB
) << R300_DST_BLEND_SHIFT
);
400 blend_control_noalpha
= blend_control_noalpha_noclamp
=
401 R300_ALPHA_BLEND_ENABLE
|
402 ( r300_translate_blend_factor(srcRGBX
) << R300_SRC_BLEND_SHIFT
) |
403 ( r300_translate_blend_factor(dstRGBX
) << R300_DST_BLEND_SHIFT
);
405 blend_eq
= r300_translate_blend_function(eqRGB
, TRUE
);
406 blend_eq_noclamp
= r300_translate_blend_function(eqRGB
, FALSE
);
408 blend_control
|= blend_eq
;
409 blend_control_noalpha
|= blend_eq
;
410 blend_control_noclamp
|= blend_eq_noclamp
;
411 blend_control_noalpha_noclamp
|= blend_eq_noclamp
;
413 /* Optimization: some operations do not require the destination color. */
414 blend_control
|= blend_read_enable(eqRGB
, eqA
, dstRGB
, dstA
,
415 srcRGB
, srcA
, r300screen
->caps
.is_r500
);
416 blend_control_noclamp
|= blend_read_enable(eqRGB
, eqA
, dstRGB
, dstA
,
417 srcRGB
, srcA
, FALSE
);
418 blend_control_noalpha
|= blend_read_enable(eqRGB
, eqA
, dstRGBX
, dstA
,
419 srcRGBX
, srcA
, r300screen
->caps
.is_r500
);
420 blend_control_noalpha_noclamp
|= blend_read_enable(eqRGB
, eqA
, dstRGBX
, dstA
,
421 srcRGBX
, srcA
, FALSE
);
423 /* Optimization: discard pixels which don't change the colorbuffer.
424 * It cannot be used with FP16 AA. */
425 blend_control
|= blend_discard_conditionally(eqRGB
, eqA
, dstRGB
, dstA
,
427 blend_control_noalpha
|= blend_discard_conditionally(eqRGB
, eqA
, dstRGBX
, dstA
,
431 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
432 blend_control
|= R300_SEPARATE_ALPHA_ENABLE
;
433 blend_control_noclamp
|= R300_SEPARATE_ALPHA_ENABLE
;
435 alpha_blend_control
= alpha_blend_control_noclamp
=
436 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
437 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
438 alpha_blend_control
|= r300_translate_blend_function(eqA
, TRUE
);
439 alpha_blend_control_noclamp
|= r300_translate_blend_function(eqA
, FALSE
);
441 if (srcA
!= srcRGBX
|| dstA
!= dstRGBX
|| eqA
!= eqRGB
) {
442 blend_control_noalpha
|= R300_SEPARATE_ALPHA_ENABLE
;
443 blend_control_noalpha_noclamp
|= R300_SEPARATE_ALPHA_ENABLE
;
445 alpha_blend_control_noalpha
= alpha_blend_control_noalpha_noclamp
=
446 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
447 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
448 alpha_blend_control_noalpha
|= r300_translate_blend_function(eqA
, TRUE
);
449 alpha_blend_control_noalpha_noclamp
|= r300_translate_blend_function(eqA
, FALSE
);
453 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
454 if (state
->logicop_enable
) {
455 rop
= R300_RB3D_ROPCNTL_ROP_ENABLE
|
456 (state
->logicop_func
) << R300_RB3D_ROPCNTL_ROP_SHIFT
;
459 /* Neither fglrx nor classic r300 ever set this, regardless of dithering
460 * state. Since it's an optional implementation detail, we can leave it
461 * out and never dither.
463 * This could be revisited if we ever get quality or conformance hints.
466 dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
467 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
471 /* Build a command buffer. */
473 unsigned (*func
[COLORMASK_NUM_SWIZZLES
])(unsigned) = {
484 for (i
= 0; i
< COLORMASK_NUM_SWIZZLES
; i
++) {
485 boolean has_alpha
= i
!= COLORMASK_RGBX
&& i
!= COLORMASK_BGRX
;
487 BEGIN_CB(blend
->cb_clamp
[i
], 8);
488 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
489 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
490 OUT_CB(has_alpha
? blend_control
: blend_control_noalpha
);
491 OUT_CB(has_alpha
? alpha_blend_control
: alpha_blend_control_noalpha
);
492 OUT_CB(func
[i
](state
->rt
[0].colormask
));
493 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
498 /* Build a command buffer (for RGBA16F). */
499 BEGIN_CB(blend
->cb_noclamp
, 8);
500 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
501 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
502 OUT_CB(blend_control_noclamp
);
503 OUT_CB(alpha_blend_control_noclamp
);
504 OUT_CB(rgba_cmask(state
->rt
[0].colormask
));
505 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
508 /* Build a command buffer (for RGB16F). */
509 BEGIN_CB(blend
->cb_noclamp_noalpha
, 8);
510 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
511 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
512 OUT_CB(blend_control_noalpha_noclamp
);
513 OUT_CB(alpha_blend_control_noalpha_noclamp
);
514 OUT_CB(rgba_cmask(state
->rt
[0].colormask
));
515 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
518 /* The same as above, but with no colorbuffer reads and writes. */
519 BEGIN_CB(blend
->cb_no_readwrite
, 8);
520 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
521 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
525 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
531 /* Bind blend state. */
532 static void r300_bind_blend_state(struct pipe_context
* pipe
,
535 struct r300_context
* r300
= r300_context(pipe
);
536 struct r300_blend_state
*blend
= (struct r300_blend_state
*)state
;
537 boolean last_alpha_to_one
= r300
->alpha_to_one
;
538 boolean last_alpha_to_coverage
= r300
->alpha_to_coverage
;
540 UPDATE_STATE(state
, r300
->blend_state
);
545 r300
->alpha_to_one
= blend
->state
.alpha_to_one
;
546 r300
->alpha_to_coverage
= blend
->state
.alpha_to_coverage
;
548 if (r300
->alpha_to_one
!= last_alpha_to_one
&& r300
->msaa_enable
&&
549 r300
->fs_status
== FRAGMENT_SHADER_VALID
) {
550 r300
->fs_status
= FRAGMENT_SHADER_MAYBE_DIRTY
;
553 if (r300
->alpha_to_coverage
!= last_alpha_to_coverage
&&
555 r300_mark_atom_dirty(r300
, &r300
->dsa_state
);
559 /* Free blend state. */
560 static void r300_delete_blend_state(struct pipe_context
* pipe
,
566 /* Convert float to 10bit integer */
567 static unsigned float_to_fixed10(float f
)
569 return CLAMP((unsigned)(f
* 1023.9f
), 0, 1023);
573 * Setup both R300 and R500 registers, figure out later which one to write. */
574 static void r300_set_blend_color(struct pipe_context
* pipe
,
575 const struct pipe_blend_color
* color
)
577 struct r300_context
* r300
= r300_context(pipe
);
578 struct pipe_framebuffer_state
*fb
= r300
->fb_state
.state
;
579 struct r300_blend_color_state
*state
=
580 (struct r300_blend_color_state
*)r300
->blend_color_state
.state
;
581 struct pipe_blend_color c
;
582 struct pipe_surface
*cb
;
586 state
->state
= *color
; /* Save it, so that we can reuse it in set_fb_state */
588 cb
= fb
->nr_cbufs
? r300_get_nonnull_cb(fb
, 0) : NULL
;
590 /* The blend color is dependent on the colorbuffer format. */
592 switch (cb
->format
) {
593 case PIPE_FORMAT_R8_UNORM
:
594 case PIPE_FORMAT_L8_UNORM
:
595 case PIPE_FORMAT_I8_UNORM
:
596 c
.color
[1] = c
.color
[0];
599 case PIPE_FORMAT_A8_UNORM
:
600 c
.color
[1] = c
.color
[3];
603 case PIPE_FORMAT_R8G8_UNORM
:
604 c
.color
[2] = c
.color
[1];
607 case PIPE_FORMAT_L8A8_UNORM
:
608 case PIPE_FORMAT_R8A8_UNORM
:
609 c
.color
[2] = c
.color
[3];
612 case PIPE_FORMAT_R8G8B8A8_UNORM
:
613 case PIPE_FORMAT_R8G8B8X8_UNORM
:
615 c
.color
[0] = c
.color
[2];
623 if (r300
->screen
->caps
.is_r500
) {
624 BEGIN_CB(state
->cb
, 3);
625 OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR
, 2);
627 switch (cb
? cb
->format
: 0) {
628 case PIPE_FORMAT_R16G16B16A16_FLOAT
:
629 case PIPE_FORMAT_R16G16B16X16_FLOAT
:
630 OUT_CB(util_float_to_half(c
.color
[2]) |
631 (util_float_to_half(c
.color
[3]) << 16));
632 OUT_CB(util_float_to_half(c
.color
[0]) |
633 (util_float_to_half(c
.color
[1]) << 16));
637 OUT_CB(float_to_fixed10(c
.color
[0]) |
638 (float_to_fixed10(c
.color
[3]) << 16));
639 OUT_CB(float_to_fixed10(c
.color
[2]) |
640 (float_to_fixed10(c
.color
[1]) << 16));
646 util_pack_color(c
.color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
648 BEGIN_CB(state
->cb
, 2);
649 OUT_CB_REG(R300_RB3D_BLEND_COLOR
, uc
.ui
[0]);
653 r300_mark_atom_dirty(r300
, &r300
->blend_color_state
);
656 static void r300_set_clip_state(struct pipe_context
* pipe
,
657 const struct pipe_clip_state
* state
)
659 struct r300_context
* r300
= r300_context(pipe
);
660 struct r300_clip_state
*clip
=
661 (struct r300_clip_state
*)r300
->clip_state
.state
;
664 if (r300
->screen
->caps
.has_tcl
) {
665 BEGIN_CB(clip
->cb
, r300
->clip_state
.size
);
666 OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG
,
667 (r300
->screen
->caps
.is_r500
?
668 R500_PVS_UCP_START
: R300_PVS_UCP_START
));
669 OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, 6 * 4);
670 OUT_CB_TABLE(state
->ucp
, 6 * 4);
673 r300_mark_atom_dirty(r300
, &r300
->clip_state
);
675 draw_set_clip_state(r300
->draw
, state
);
679 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
681 * This contains the depth buffer, stencil buffer, alpha test, and such.
682 * On the Radeon, depth and stencil buffer setup are intertwined, which is
683 * the reason for some of the strange-looking assignments across registers. */
684 static void* r300_create_dsa_state(struct pipe_context
* pipe
,
685 const struct pipe_depth_stencil_alpha_state
* state
)
687 boolean is_r500
= r300_screen(pipe
->screen
)->caps
.is_r500
;
688 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
690 uint32_t alpha_value_fp16
= 0;
691 uint32_t z_buffer_control
= 0;
692 uint32_t z_stencil_control
= 0;
693 uint32_t stencil_ref_mask
= 0;
694 uint32_t stencil_ref_bf
= 0;
698 /* Depth test setup. - separate write mask depth for decomp flush */
699 if (state
->depth
.writemask
) {
700 z_buffer_control
|= R300_Z_WRITE_ENABLE
;
703 if (state
->depth
.enabled
) {
704 z_buffer_control
|= R300_Z_ENABLE
;
707 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
711 /* Stencil buffer setup. */
712 if (state
->stencil
[0].enabled
) {
713 z_buffer_control
|= R300_STENCIL_ENABLE
;
715 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
716 R300_S_FRONT_FUNC_SHIFT
) |
717 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
718 R300_S_FRONT_SFAIL_OP_SHIFT
) |
719 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
720 R300_S_FRONT_ZPASS_OP_SHIFT
) |
721 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
722 R300_S_FRONT_ZFAIL_OP_SHIFT
);
725 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
726 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
728 if (state
->stencil
[1].enabled
) {
729 dsa
->two_sided
= TRUE
;
731 z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
733 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
734 R300_S_BACK_FUNC_SHIFT
) |
735 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
736 R300_S_BACK_SFAIL_OP_SHIFT
) |
737 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
738 R300_S_BACK_ZPASS_OP_SHIFT
) |
739 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
740 R300_S_BACK_ZFAIL_OP_SHIFT
);
743 (state
->stencil
[1].valuemask
<< R300_STENCILMASK_SHIFT
) |
744 (state
->stencil
[1].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
747 z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
749 dsa
->two_sided_stencil_ref
=
750 (state
->stencil
[0].valuemask
!= state
->stencil
[1].valuemask
||
751 state
->stencil
[0].writemask
!= state
->stencil
[1].writemask
);
756 /* Alpha test setup. */
757 if (state
->alpha
.enabled
) {
758 dsa
->alpha_function
=
759 r300_translate_alpha_function(state
->alpha
.func
) |
760 R300_FG_ALPHA_FUNC_ENABLE
;
762 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
763 alpha_value_fp16
= util_float_to_half(state
->alpha
.ref_value
);
766 BEGIN_CB(&dsa
->cb_begin
, 8);
767 OUT_CB_REG_SEQ(R300_ZB_CNTL
, 3);
768 OUT_CB(z_buffer_control
);
769 OUT_CB(z_stencil_control
);
770 OUT_CB(stencil_ref_mask
);
771 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF
, stencil_ref_bf
);
772 OUT_CB_REG(R500_FG_ALPHA_VALUE
, alpha_value_fp16
);
775 BEGIN_CB(dsa
->cb_zb_no_readwrite
, 8);
776 OUT_CB_REG_SEQ(R300_ZB_CNTL
, 3);
780 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF
, 0);
781 OUT_CB_REG(R500_FG_ALPHA_VALUE
, alpha_value_fp16
);
787 static void r300_dsa_inject_stencilref(struct r300_context
*r300
)
789 struct r300_dsa_state
*dsa
=
790 (struct r300_dsa_state
*)r300
->dsa_state
.state
;
795 dsa
->stencil_ref_mask
=
796 (dsa
->stencil_ref_mask
& ~R300_STENCILREF_MASK
) |
797 r300
->stencil_ref
.ref_value
[0];
798 dsa
->stencil_ref_bf
=
799 (dsa
->stencil_ref_bf
& ~R300_STENCILREF_MASK
) |
800 r300
->stencil_ref
.ref_value
[1];
803 /* Bind DSA state. */
804 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
807 struct r300_context
* r300
= r300_context(pipe
);
813 UPDATE_STATE(state
, r300
->dsa_state
);
815 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
); /* Will be updated before the emission. */
816 r300_dsa_inject_stencilref(r300
);
819 /* Free DSA state. */
820 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
826 static void r300_set_stencil_ref(struct pipe_context
* pipe
,
827 const struct pipe_stencil_ref
* sr
)
829 struct r300_context
* r300
= r300_context(pipe
);
831 r300
->stencil_ref
= *sr
;
833 r300_dsa_inject_stencilref(r300
);
834 r300_mark_atom_dirty(r300
, &r300
->dsa_state
);
837 static void r300_tex_set_tiling_flags(struct r300_context
*r300
,
838 struct r300_resource
*tex
,
841 /* Check if the macrotile flag needs to be changed.
842 * Skip changing the flags otherwise. */
843 if (tex
->tex
.macrotile
[tex
->surface_level
] !=
844 tex
->tex
.macrotile
[level
]) {
845 r300
->rws
->buffer_set_tiling(tex
->buf
, r300
->cs
,
846 tex
->tex
.microtile
, tex
->tex
.macrotile
[level
],
848 tex
->tex
.stride_in_bytes
[0], false);
850 tex
->surface_level
= level
;
854 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
855 static void r300_fb_set_tiling_flags(struct r300_context
*r300
,
856 const struct pipe_framebuffer_state
*state
)
860 /* Set tiling flags for new surfaces. */
861 for (i
= 0; i
< state
->nr_cbufs
; i
++) {
862 if (!state
->cbufs
[i
])
865 r300_tex_set_tiling_flags(r300
,
866 r300_resource(state
->cbufs
[i
]->texture
),
867 state
->cbufs
[i
]->u
.tex
.level
);
870 r300_tex_set_tiling_flags(r300
,
871 r300_resource(state
->zsbuf
->texture
),
872 state
->zsbuf
->u
.tex
.level
);
876 static void r300_print_fb_surf_info(struct pipe_surface
*surf
, unsigned index
,
879 struct pipe_resource
*tex
= surf
->texture
;
880 struct r300_resource
*rtex
= r300_resource(tex
);
883 "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, "
884 "Lastlayer: %i, Level: %i, Format: %s\n"
886 "r300: TEX: Macro: %s, Micro: %s, "
887 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n",
889 binding
, index
, surf
->width
, surf
->height
,
890 surf
->u
.tex
.first_layer
, surf
->u
.tex
.last_layer
, surf
->u
.tex
.level
,
891 util_format_short_name(surf
->format
),
893 rtex
->tex
.macrotile
[0] ? "YES" : " NO",
894 rtex
->tex
.microtile
? "YES" : " NO",
895 tex
->width0
, tex
->height0
, tex
->depth0
,
896 tex
->last_level
, util_format_short_name(surf
->format
));
899 void r300_mark_fb_state_dirty(struct r300_context
*r300
,
900 enum r300_fb_state_change change
)
902 struct pipe_framebuffer_state
*state
= r300
->fb_state
.state
;
904 r300_mark_atom_dirty(r300
, &r300
->gpu_flush
);
905 r300_mark_atom_dirty(r300
, &r300
->fb_state
);
907 /* What is marked as dirty depends on the enum r300_fb_state_change. */
908 if (change
== R300_CHANGED_FB_STATE
) {
909 r300_mark_atom_dirty(r300
, &r300
->aa_state
);
910 r300_mark_atom_dirty(r300
, &r300
->dsa_state
); /* for AlphaRef */
911 r300_set_blend_color(&r300
->context
, r300
->blend_color_state
.state
);
914 if (change
== R300_CHANGED_FB_STATE
||
915 change
== R300_CHANGED_HYPERZ_FLAG
) {
916 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
919 if (change
== R300_CHANGED_FB_STATE
||
920 change
== R300_CHANGED_MULTIWRITE
) {
921 r300_mark_atom_dirty(r300
, &r300
->fb_state_pipelined
);
924 /* Now compute the fb_state atom size. */
925 r300
->fb_state
.size
= 2 + (8 * state
->nr_cbufs
);
927 if (r300
->cbzb_clear
)
928 r300
->fb_state
.size
+= 10;
929 else if (state
->zsbuf
) {
930 r300
->fb_state
.size
+= 10;
931 if (r300
->hyperz_enabled
)
932 r300
->fb_state
.size
+= 8;
935 if (r300
->cmask_in_use
) {
936 r300
->fb_state
.size
+= 6;
937 if (r300
->screen
->caps
.is_r500
&& r300
->screen
->info
.drm_minor
>= 29) {
938 r300
->fb_state
.size
+= 3;
942 /* The size of the rest of atoms stays the same. */
945 static unsigned r300_get_num_samples(struct r300_context
*r300
)
947 struct pipe_framebuffer_state
* fb
=
948 (struct pipe_framebuffer_state
*)r300
->fb_state
.state
;
949 unsigned i
, num_samples
;
951 if (!fb
->nr_cbufs
&& !fb
->zsbuf
)
956 for (i
= 0; i
< fb
->nr_cbufs
; i
++)
958 num_samples
= MIN2(num_samples
, fb
->cbufs
[i
]->texture
->nr_samples
);
961 num_samples
= MIN2(num_samples
, fb
->zsbuf
->texture
->nr_samples
);
970 r300_set_framebuffer_state(struct pipe_context
* pipe
,
971 const struct pipe_framebuffer_state
* state
)
973 struct r300_context
* r300
= r300_context(pipe
);
974 struct r300_aa_state
*aa
= (struct r300_aa_state
*)r300
->aa_state
.state
;
975 struct pipe_framebuffer_state
*current_state
= r300
->fb_state
.state
;
976 unsigned max_width
, max_height
, i
;
977 uint32_t zbuffer_bpp
= 0;
978 boolean unlock_zbuffer
= FALSE
;
980 if (r300
->screen
->caps
.is_r500
) {
981 max_width
= max_height
= 4096;
982 } else if (r300
->screen
->caps
.is_r400
) {
983 max_width
= max_height
= 4021;
985 max_width
= max_height
= 2560;
988 if (state
->width
> max_width
|| state
->height
> max_height
) {
989 fprintf(stderr
, "r300: Implementation error: Render targets are too "
990 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
994 if (current_state
->zsbuf
&& r300
->zmask_in_use
&& !r300
->locked_zbuffer
) {
995 /* There is a zmask in use, what are we gonna do? */
997 if (!pipe_surface_equal(current_state
->zsbuf
, state
->zsbuf
)) {
998 /* Decompress the currently bound zbuffer before we bind another one. */
999 r300_decompress_zmask(r300
);
1000 r300
->hiz_in_use
= FALSE
;
1003 /* We don't bind another zbuffer, so lock the current one. */
1004 pipe_surface_reference(&r300
->locked_zbuffer
, current_state
->zsbuf
);
1006 } else if (r300
->locked_zbuffer
) {
1007 /* We have a locked zbuffer now, what are we gonna do? */
1009 if (!pipe_surface_equal(r300
->locked_zbuffer
, state
->zsbuf
)) {
1010 /* We are binding some other zbuffer, so decompress the locked one,
1011 * it gets unlocked automatically. */
1012 r300_decompress_zmask_locked_unsafe(r300
);
1013 r300
->hiz_in_use
= FALSE
;
1015 /* We are binding the locked zbuffer again, so unlock it. */
1016 unlock_zbuffer
= TRUE
;
1020 assert(state
->zsbuf
|| (r300
->locked_zbuffer
&& !unlock_zbuffer
) || !r300
->zmask_in_use
);
1022 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
1023 if (!!current_state
->zsbuf
!= !!state
->zsbuf
) {
1024 r300_mark_atom_dirty(r300
, &r300
->dsa_state
);
1027 util_copy_framebuffer_state(r300
->fb_state
.state
, state
);
1029 /* Remove trailing NULL colorbuffers. */
1030 while (current_state
->nr_cbufs
&& !current_state
->cbufs
[current_state
->nr_cbufs
-1])
1031 current_state
->nr_cbufs
--;
1033 /* Set whether CMASK can be used. */
1034 r300
->cmask_in_use
=
1035 state
->nr_cbufs
== 1 && state
->cbufs
[0] &&
1036 r300
->screen
->cmask_resource
== state
->cbufs
[0]->texture
;
1038 /* Need to reset clamping or colormask. */
1039 r300_mark_atom_dirty(r300
, &r300
->blend_state
);
1041 /* Re-swizzle the blend color. */
1042 r300_set_blend_color(pipe
, &((struct r300_blend_color_state
*)r300
->blend_color_state
.state
)->state
);
1044 if (r300
->screen
->info
.drm_minor
< 12) {
1045 /* The tiling flags are dependent on the surface miplevel, unfortunately.
1046 * This workarounds a bad design decision in old kernels which were
1047 * rewriting tile fields in registers. */
1048 r300_fb_set_tiling_flags(r300
, state
);
1051 if (unlock_zbuffer
) {
1052 pipe_surface_reference(&r300
->locked_zbuffer
, NULL
);
1055 r300_mark_fb_state_dirty(r300
, R300_CHANGED_FB_STATE
);
1058 switch (util_format_get_blocksize(state
->zsbuf
->format
)) {
1067 /* Polygon offset depends on the zbuffer bit depth. */
1068 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
1069 r300
->zbuffer_bpp
= zbuffer_bpp
;
1071 if (r300
->polygon_offset_enabled
)
1072 r300_mark_atom_dirty(r300
, &r300
->rs_state
);
1076 r300
->num_samples
= r300_get_num_samples(r300
);
1078 /* Set up AA config. */
1079 if (r300
->num_samples
> 1) {
1080 switch (r300
->num_samples
) {
1082 aa
->aa_config
= R300_GB_AA_CONFIG_AA_ENABLE
|
1083 R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2
;
1086 aa
->aa_config
= R300_GB_AA_CONFIG_AA_ENABLE
|
1087 R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4
;
1090 aa
->aa_config
= R300_GB_AA_CONFIG_AA_ENABLE
|
1091 R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6
;
1098 if (DBG_ON(r300
, DBG_FB
)) {
1099 fprintf(stderr
, "r300: set_framebuffer_state:\n");
1100 for (i
= 0; i
< state
->nr_cbufs
; i
++) {
1101 if (state
->cbufs
[i
])
1102 r300_print_fb_surf_info(state
->cbufs
[i
], i
, "CB");
1105 r300_print_fb_surf_info(state
->zsbuf
, 0, "ZB");
1110 /* Create fragment shader state. */
1111 static void* r300_create_fs_state(struct pipe_context
* pipe
,
1112 const struct pipe_shader_state
* shader
)
1114 struct r300_fragment_shader
* fs
= NULL
;
1116 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
1118 /* Copy state directly into shader. */
1119 fs
->state
= *shader
;
1120 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
1125 void r300_mark_fs_code_dirty(struct r300_context
*r300
)
1127 struct r300_fragment_shader
* fs
= r300_fs(r300
);
1129 r300_mark_atom_dirty(r300
, &r300
->fs
);
1130 r300_mark_atom_dirty(r300
, &r300
->fs_rc_constant_state
);
1131 r300_mark_atom_dirty(r300
, &r300
->fs_constants
);
1132 r300
->fs
.size
= fs
->shader
->cb_code_size
;
1134 if (r300
->screen
->caps
.is_r500
) {
1135 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 7;
1136 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 3;
1138 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 5;
1139 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 1;
1142 ((struct r300_constant_buffer
*)r300
->fs_constants
.state
)->remap_table
=
1143 fs
->shader
->code
.constants_remap_table
;
1146 /* Bind fragment shader state. */
1147 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
1149 struct r300_context
* r300
= r300_context(pipe
);
1150 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
1153 r300
->fs
.state
= NULL
;
1157 r300
->fs
.state
= fs
;
1158 r300
->fs_status
= FRAGMENT_SHADER_DIRTY
;
1160 r300_mark_atom_dirty(r300
, &r300
->rs_block_state
); /* Will be updated before the emission. */
1163 /* Delete fragment shader state. */
1164 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
1166 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
1167 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
1172 rc_constants_destroy(&tmp
->code
.constants
);
1176 FREE((void*)fs
->state
.tokens
);
1180 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
1181 const struct pipe_poly_stipple
* state
)
1183 /* XXX no idea how to set this up, but not terribly important */
1186 /* Create a new rasterizer state based on the CSO rasterizer state.
1188 * This is a very large chunk of state, and covers most of the graphics
1189 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
1191 * In a not entirely unironic sidenote, this state has nearly nothing to do
1192 * with the actual block on the Radeon called the rasterizer (RS). */
1193 static void* r300_create_rs_state(struct pipe_context
* pipe
,
1194 const struct pipe_rasterizer_state
* state
)
1196 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
1197 uint32_t vap_control_status
; /* R300_VAP_CNTL_STATUS: 0x2140 */
1198 uint32_t vap_clip_cntl
; /* R300_VAP_CLIP_CNTL: 0x221C */
1199 uint32_t point_size
; /* R300_GA_POINT_SIZE: 0x421c */
1200 uint32_t point_minmax
; /* R300_GA_POINT_MINMAX: 0x4230 */
1201 uint32_t line_control
; /* R300_GA_LINE_CNTL: 0x4234 */
1202 uint32_t polygon_offset_enable
; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */
1203 uint32_t cull_mode
; /* R300_SU_CULL_MODE: 0x42b8 */
1204 uint32_t line_stipple_config
; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
1205 uint32_t line_stipple_value
; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
1206 uint32_t polygon_mode
; /* R300_GA_POLY_MODE: 0x4288 */
1207 uint32_t clip_rule
; /* R300_SC_CLIP_RULE: 0x43D0 */
1208 uint32_t round_mode
; /* R300_GA_ROUND_MODE: 0x428c */
1210 /* Point sprites texture coordinates, 0: lower left, 1: upper right */
1211 float point_texcoord_left
= 0; /* R300_GA_POINT_S0: 0x4200 */
1212 float point_texcoord_bottom
= 0;/* R300_GA_POINT_T0: 0x4204 */
1213 float point_texcoord_right
= 1; /* R300_GA_POINT_S1: 0x4208 */
1214 float point_texcoord_top
= 0; /* R300_GA_POINT_T1: 0x420c */
1215 boolean vclamp
= !r300_context(pipe
)->screen
->caps
.is_r500
;
1218 /* Copy rasterizer state. */
1220 rs
->rs_draw
= *state
;
1222 rs
->rs
.sprite_coord_enable
= state
->point_quad_rasterization
*
1223 state
->sprite_coord_enable
;
1225 /* Override some states for Draw. */
1226 rs
->rs_draw
.sprite_coord_enable
= 0; /* We can do this in HW. */
1227 rs
->rs_draw
.offset_point
= 0;
1228 rs
->rs_draw
.offset_line
= 0;
1229 rs
->rs_draw
.offset_tri
= 0;
1230 rs
->rs_draw
.offset_clamp
= 0;
1232 #ifdef PIPE_ARCH_LITTLE_ENDIAN
1233 vap_control_status
= R300_VC_NO_SWAP
;
1235 vap_control_status
= R300_VC_32BIT_SWAP
;
1238 /* If no TCL engine is present, turn off the HW TCL. */
1239 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1240 vap_control_status
|= R300_VAP_TCL_BYPASS
;
1243 /* Point size width and height. */
1245 pack_float_16_6x(state
->point_size
) |
1246 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
1248 /* Point size clamping. */
1249 if (state
->point_size_per_vertex
) {
1250 /* Per-vertex point size.
1251 * Clamp to [0, max FB size] */
1252 float min_psiz
= util_get_min_point_size(state
);
1253 float max_psiz
= pipe
->screen
->get_paramf(pipe
->screen
,
1254 PIPE_CAPF_MAX_POINT_WIDTH
);
1256 (pack_float_16_6x(min_psiz
) << R300_GA_POINT_MINMAX_MIN_SHIFT
) |
1257 (pack_float_16_6x(max_psiz
) << R300_GA_POINT_MINMAX_MAX_SHIFT
);
1259 /* We cannot disable the point-size vertex output,
1261 float psiz
= state
->point_size
;
1263 (pack_float_16_6x(psiz
) << R300_GA_POINT_MINMAX_MIN_SHIFT
) |
1264 (pack_float_16_6x(psiz
) << R300_GA_POINT_MINMAX_MAX_SHIFT
);
1268 line_control
= pack_float_16_6x(state
->line_width
) |
1269 R300_GA_LINE_CNTL_END_TYPE_COMP
;
1271 /* Enable polygon mode */
1273 if (state
->fill_front
!= PIPE_POLYGON_MODE_FILL
||
1274 state
->fill_back
!= PIPE_POLYGON_MODE_FILL
) {
1275 polygon_mode
= R300_GA_POLY_MODE_DUAL
;
1279 if (state
->front_ccw
)
1280 cull_mode
= R300_FRONT_FACE_CCW
;
1282 cull_mode
= R300_FRONT_FACE_CW
;
1284 /* Polygon offset */
1285 polygon_offset_enable
= 0;
1286 if (util_get_offset(state
, state
->fill_front
)) {
1287 polygon_offset_enable
|= R300_FRONT_ENABLE
;
1289 if (util_get_offset(state
, state
->fill_back
)) {
1290 polygon_offset_enable
|= R300_BACK_ENABLE
;
1293 rs
->polygon_offset_enable
= polygon_offset_enable
!= 0;
1298 r300_translate_polygon_mode_front(state
->fill_front
);
1300 r300_translate_polygon_mode_back(state
->fill_back
);
1303 if (state
->cull_face
& PIPE_FACE_FRONT
) {
1304 cull_mode
|= R300_CULL_FRONT
;
1306 if (state
->cull_face
& PIPE_FACE_BACK
) {
1307 cull_mode
|= R300_CULL_BACK
;
1310 if (state
->line_stipple_enable
) {
1311 line_stipple_config
=
1312 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
1313 (fui((float)state
->line_stipple_factor
) &
1314 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
1315 /* XXX this might need to be scaled up */
1316 line_stipple_value
= state
->line_stipple_pattern
;
1318 line_stipple_config
= 0;
1319 line_stipple_value
= 0;
1322 if (state
->flatshade
) {
1323 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
1325 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
1328 clip_rule
= state
->scissor
? 0xAAAA : 0xFFFF;
1330 /* Point sprites coord mode */
1331 if (rs
->rs
.sprite_coord_enable
) {
1332 switch (state
->sprite_coord_mode
) {
1333 case PIPE_SPRITE_COORD_UPPER_LEFT
:
1334 point_texcoord_top
= 0.0f
;
1335 point_texcoord_bottom
= 1.0f
;
1337 case PIPE_SPRITE_COORD_LOWER_LEFT
:
1338 point_texcoord_top
= 1.0f
;
1339 point_texcoord_bottom
= 0.0f
;
1344 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1345 vap_clip_cntl
= (state
->clip_plane_enable
& 63) |
1346 R300_PS_UCP_MODE_CLIP_AS_TRIFAN
;
1348 vap_clip_cntl
= R300_CLIP_DISABLE
;
1351 /* Vertex color clamping. FP20 means no clamping. */
1353 R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST
|
1354 (!vclamp
? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20
|
1355 R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20
) : 0);
1357 /* Build the main command buffer. */
1358 BEGIN_CB(rs
->cb_main
, RS_STATE_MAIN_SIZE
);
1359 OUT_CB_REG(R300_VAP_CNTL_STATUS
, vap_control_status
);
1360 OUT_CB_REG(R300_VAP_CLIP_CNTL
, vap_clip_cntl
);
1361 OUT_CB_REG(R300_GA_POINT_SIZE
, point_size
);
1362 OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX
, 2);
1363 OUT_CB(point_minmax
);
1364 OUT_CB(line_control
);
1365 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE
, 2);
1366 OUT_CB(polygon_offset_enable
);
1367 rs
->cull_mode_index
= 11;
1369 OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG
, line_stipple_config
);
1370 OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE
, line_stipple_value
);
1371 OUT_CB_REG(R300_GA_POLY_MODE
, polygon_mode
);
1372 OUT_CB_REG(R300_GA_ROUND_MODE
, round_mode
);
1373 OUT_CB_REG(R300_SC_CLIP_RULE
, clip_rule
);
1374 OUT_CB_REG_SEQ(R300_GA_POINT_S0
, 4);
1375 OUT_CB_32F(point_texcoord_left
);
1376 OUT_CB_32F(point_texcoord_bottom
);
1377 OUT_CB_32F(point_texcoord_right
);
1378 OUT_CB_32F(point_texcoord_top
);
1381 /* Build the two command buffers for polygon offset setup. */
1382 if (polygon_offset_enable
) {
1383 float scale
= state
->offset_scale
* 12;
1384 float offset
= state
->offset_units
* 4;
1386 BEGIN_CB(rs
->cb_poly_offset_zb16
, 5);
1387 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE
, 4);
1394 offset
= state
->offset_units
* 2;
1396 BEGIN_CB(rs
->cb_poly_offset_zb24
, 5);
1397 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE
, 4);
1408 /* Bind rasterizer state. */
1409 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
1411 struct r300_context
* r300
= r300_context(pipe
);
1412 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
1413 int last_sprite_coord_enable
= r300
->sprite_coord_enable
;
1414 boolean last_two_sided_color
= r300
->two_sided_color
;
1415 boolean last_msaa_enable
= r300
->msaa_enable
;
1416 boolean last_flatshade
= r300
->flatshade
;
1418 if (r300
->draw
&& rs
) {
1419 draw_set_rasterizer_state(r300
->draw
, &rs
->rs_draw
, state
);
1423 r300
->polygon_offset_enabled
= rs
->polygon_offset_enable
;
1424 r300
->sprite_coord_enable
= rs
->rs
.sprite_coord_enable
;
1425 r300
->two_sided_color
= rs
->rs
.light_twoside
;
1426 r300
->msaa_enable
= rs
->rs
.multisample
;
1427 r300
->flatshade
= rs
->rs
.flatshade
;
1429 r300
->polygon_offset_enabled
= FALSE
;
1430 r300
->sprite_coord_enable
= 0;
1431 r300
->two_sided_color
= FALSE
;
1432 r300
->msaa_enable
= FALSE
;
1433 r300
->flatshade
= FALSE
;
1436 UPDATE_STATE(state
, r300
->rs_state
);
1437 r300
->rs_state
.size
= RS_STATE_MAIN_SIZE
+ (r300
->polygon_offset_enabled
? 5 : 0);
1439 if (last_sprite_coord_enable
!= r300
->sprite_coord_enable
||
1440 last_two_sided_color
!= r300
->two_sided_color
||
1441 last_flatshade
!= r300
->flatshade
) {
1442 r300_mark_atom_dirty(r300
, &r300
->rs_block_state
);
1445 if (last_msaa_enable
!= r300
->msaa_enable
) {
1446 if (r300
->alpha_to_coverage
) {
1447 r300_mark_atom_dirty(r300
, &r300
->dsa_state
);
1450 if (r300
->alpha_to_one
&&
1451 r300
->fs_status
== FRAGMENT_SHADER_VALID
) {
1452 r300
->fs_status
= FRAGMENT_SHADER_MAYBE_DIRTY
;
1457 /* Free rasterizer state. */
1458 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
1464 r300_create_sampler_state(struct pipe_context
* pipe
,
1465 const struct pipe_sampler_state
* state
)
1467 struct r300_context
* r300
= r300_context(pipe
);
1468 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
1469 boolean is_r500
= r300
->screen
->caps
.is_r500
;
1472 sampler
->state
= *state
;
1474 /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG
1475 * or MIN filter is NEAREST. Since texwrap produces same results
1476 * for CLAMP and CLAMP_TO_EDGE, we use them instead. */
1477 if (sampler
->state
.min_img_filter
== PIPE_TEX_FILTER_NEAREST
||
1478 sampler
->state
.mag_img_filter
== PIPE_TEX_FILTER_NEAREST
) {
1480 if (sampler
->state
.wrap_s
== PIPE_TEX_WRAP_CLAMP
)
1481 sampler
->state
.wrap_s
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1482 else if (sampler
->state
.wrap_s
== PIPE_TEX_WRAP_MIRROR_CLAMP
)
1483 sampler
->state
.wrap_s
= PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
;
1486 if (sampler
->state
.wrap_t
== PIPE_TEX_WRAP_CLAMP
)
1487 sampler
->state
.wrap_t
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1488 else if (sampler
->state
.wrap_t
== PIPE_TEX_WRAP_MIRROR_CLAMP
)
1489 sampler
->state
.wrap_t
= PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
;
1492 if (sampler
->state
.wrap_r
== PIPE_TEX_WRAP_CLAMP
)
1493 sampler
->state
.wrap_r
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1494 else if (sampler
->state
.wrap_r
== PIPE_TEX_WRAP_MIRROR_CLAMP
)
1495 sampler
->state
.wrap_r
= PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
;
1499 (r300_translate_wrap(sampler
->state
.wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
1500 (r300_translate_wrap(sampler
->state
.wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
1501 (r300_translate_wrap(sampler
->state
.wrap_r
) << R300_TX_WRAP_R_SHIFT
);
1503 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
1504 state
->mag_img_filter
,
1505 state
->min_mip_filter
,
1506 state
->max_anisotropy
> 1);
1508 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
1510 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
1511 /* We must pass these to the merge function to clamp them properly. */
1512 sampler
->min_lod
= (unsigned)MAX2(state
->min_lod
, 0);
1513 sampler
->max_lod
= (unsigned)MAX2(ceilf(state
->max_lod
), 0);
1515 lod_bias
= CLAMP((int)(state
->lod_bias
* 32 + 1), -(1 << 9), (1 << 9) - 1);
1517 sampler
->filter1
|= (lod_bias
<< R300_LOD_BIAS_SHIFT
) & R300_LOD_BIAS_MASK
;
1519 /* This is very high quality anisotropic filtering for R5xx.
1520 * It's good for benchmarking the performance of texturing but
1521 * in practice we don't want to slow down the driver because it's
1522 * a pretty good performance killer. Feel free to play with it. */
1523 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
1524 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
1527 /* R500-specific fixups and optimizations */
1528 if (r300
->screen
->caps
.is_r500
) {
1529 sampler
->filter1
|= R500_BORDER_FIX
;
1532 return (void*)sampler
;
1535 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
1537 unsigned start
, unsigned count
,
1540 struct r300_context
* r300
= r300_context(pipe
);
1541 struct r300_textures_state
* state
=
1542 (struct r300_textures_state
*)r300
->textures_state
.state
;
1543 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
1547 if (shader
!= PIPE_SHADER_FRAGMENT
)
1550 if (count
> tex_units
)
1553 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
1554 state
->sampler_state_count
= count
;
1556 r300_mark_atom_dirty(r300
, &r300
->textures_state
);
1559 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
1564 static uint32_t r300_assign_texture_cache_region(unsigned index
, unsigned num
)
1566 /* This looks like a hack, but I believe it's suppose to work like
1567 * that. To illustrate how this works, let's assume you have 5 textures.
1568 * From docs, 5 and the successive numbers are:
1576 * First 3 textures will get 3/4 of size of the cache, divived evenly
1577 * between them. The last 1/4 of the cache must be divided between
1578 * the last 2 textures, each will therefore get 1/8 of the cache.
1579 * Why not just to use "5 + texture_index" ?
1581 * This simple trick works for all "num" <= 16.
1584 return R300_TX_CACHE(R300_TX_CACHE_WHOLE
);
1586 return R300_TX_CACHE(num
+ index
);
1589 static void r300_set_sampler_views(struct pipe_context
* pipe
, unsigned shader
,
1590 unsigned start
, unsigned count
,
1591 struct pipe_sampler_view
** views
)
1593 struct r300_context
* r300
= r300_context(pipe
);
1594 struct r300_textures_state
* state
=
1595 (struct r300_textures_state
*)r300
->textures_state
.state
;
1596 struct r300_resource
*texture
;
1597 unsigned i
, real_num_views
= 0, view_index
= 0;
1598 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
1599 boolean dirty_tex
= FALSE
;
1601 if (shader
!= PIPE_SHADER_FRAGMENT
)
1604 assert(start
== 0); /* non-zero not handled yet */
1606 if (count
> tex_units
) {
1610 /* Calculate the real number of views. */
1611 for (i
= 0; i
< count
; i
++) {
1616 for (i
= 0; i
< count
; i
++) {
1617 pipe_sampler_view_reference(
1618 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1625 /* A new sampler view (= texture)... */
1628 /* Set the texrect factor in the fragment shader.
1629 * Needed for RECT and NPOT fallback. */
1630 texture
= r300_resource(views
[i
]->texture
);
1631 if (texture
->tex
.is_npot
) {
1632 r300_mark_atom_dirty(r300
, &r300
->fs_rc_constant_state
);
1635 state
->sampler_views
[i
]->texcache_region
=
1636 r300_assign_texture_cache_region(view_index
, real_num_views
);
1640 for (i
= count
; i
< tex_units
; i
++) {
1641 if (state
->sampler_views
[i
]) {
1642 pipe_sampler_view_reference(
1643 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1648 state
->sampler_view_count
= count
;
1650 r300_mark_atom_dirty(r300
, &r300
->textures_state
);
1653 r300_mark_atom_dirty(r300
, &r300
->texture_cache_inval
);
1657 struct pipe_sampler_view
*
1658 r300_create_sampler_view_custom(struct pipe_context
*pipe
,
1659 struct pipe_resource
*texture
,
1660 const struct pipe_sampler_view
*templ
,
1661 unsigned width0_override
,
1662 unsigned height0_override
)
1664 struct r300_sampler_view
*view
= CALLOC_STRUCT(r300_sampler_view
);
1665 struct r300_resource
*tex
= r300_resource(texture
);
1666 boolean is_r500
= r300_screen(pipe
->screen
)->caps
.is_r500
;
1667 boolean dxtc_swizzle
= r300_screen(pipe
->screen
)->caps
.dxtc_swizzle
;
1672 view
->base
= *templ
;
1673 view
->base
.reference
.count
= 1;
1674 view
->base
.context
= pipe
;
1675 view
->base
.texture
= NULL
;
1676 pipe_resource_reference(&view
->base
.texture
, texture
);
1678 view
->width0_override
= width0_override
;
1679 view
->height0_override
= height0_override
;
1680 view
->swizzle
[0] = templ
->swizzle_r
;
1681 view
->swizzle
[1] = templ
->swizzle_g
;
1682 view
->swizzle
[2] = templ
->swizzle_b
;
1683 view
->swizzle
[3] = templ
->swizzle_a
;
1685 hwformat
= r300_translate_texformat(templ
->format
,
1690 if (hwformat
== ~0) {
1691 fprintf(stderr
, "r300: Ooops. Got unsupported format %s in %s.\n",
1692 util_format_short_name(templ
->format
), __func__
);
1694 assert(hwformat
!= ~0);
1696 r300_texture_setup_format_state(r300_screen(pipe
->screen
), tex
,
1698 width0_override
, height0_override
,
1700 view
->format
.format1
|= hwformat
;
1702 view
->format
.format2
|= r500_tx_format_msb_bit(templ
->format
);
1706 return (struct pipe_sampler_view
*)view
;
1709 static struct pipe_sampler_view
*
1710 r300_create_sampler_view(struct pipe_context
*pipe
,
1711 struct pipe_resource
*texture
,
1712 const struct pipe_sampler_view
*templ
)
1714 return r300_create_sampler_view_custom(pipe
, texture
, templ
,
1715 r300_resource(texture
)->tex
.width0
,
1716 r300_resource(texture
)->tex
.height0
);
1721 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1722 struct pipe_sampler_view
*view
)
1724 pipe_resource_reference(&view
->texture
, NULL
);
1728 static void r300_set_sample_mask(struct pipe_context
*pipe
,
1731 struct r300_context
* r300
= r300_context(pipe
);
1733 *((unsigned*)r300
->sample_mask
.state
) = mask
;
1735 r300_mark_atom_dirty(r300
, &r300
->sample_mask
);
1738 static void r300_set_scissor_states(struct pipe_context
* pipe
,
1739 unsigned start_slot
,
1740 unsigned num_scissors
,
1741 const struct pipe_scissor_state
* state
)
1743 struct r300_context
* r300
= r300_context(pipe
);
1745 memcpy(r300
->scissor_state
.state
, state
,
1746 sizeof(struct pipe_scissor_state
));
1748 r300_mark_atom_dirty(r300
, &r300
->scissor_state
);
1751 static void r300_set_viewport_states(struct pipe_context
* pipe
,
1752 unsigned start_slot
,
1753 unsigned num_viewports
,
1754 const struct pipe_viewport_state
* state
)
1756 struct r300_context
* r300
= r300_context(pipe
);
1757 struct r300_viewport_state
* viewport
=
1758 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1760 r300
->viewport
= *state
;
1763 draw_set_viewport_states(r300
->draw
, start_slot
, num_viewports
, state
);
1764 viewport
->vte_control
= R300_VTX_XY_FMT
| R300_VTX_Z_FMT
;
1768 /* Do the transform in HW. */
1769 viewport
->vte_control
= R300_VTX_W0_FMT
;
1771 if (state
->scale
[0] != 1.0f
) {
1772 viewport
->xscale
= state
->scale
[0];
1773 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1775 if (state
->scale
[1] != 1.0f
) {
1776 viewport
->yscale
= state
->scale
[1];
1777 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1779 if (state
->scale
[2] != 1.0f
) {
1780 viewport
->zscale
= state
->scale
[2];
1781 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1783 if (state
->translate
[0] != 0.0f
) {
1784 viewport
->xoffset
= state
->translate
[0];
1785 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1787 if (state
->translate
[1] != 0.0f
) {
1788 viewport
->yoffset
= state
->translate
[1];
1789 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1791 if (state
->translate
[2] != 0.0f
) {
1792 viewport
->zoffset
= state
->translate
[2];
1793 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1796 r300_mark_atom_dirty(r300
, &r300
->viewport_state
);
1797 if (r300
->fs
.state
&& r300_fs(r300
)->shader
&&
1798 r300_fs(r300
)->shader
->inputs
.wpos
!= ATTR_UNUSED
) {
1799 r300_mark_atom_dirty(r300
, &r300
->fs_rc_constant_state
);
1803 static void r300_set_vertex_buffers_hwtcl(struct pipe_context
* pipe
,
1804 unsigned start_slot
, unsigned count
,
1805 const struct pipe_vertex_buffer
* buffers
)
1807 struct r300_context
* r300
= r300_context(pipe
);
1809 util_set_vertex_buffers_count(r300
->vertex_buffer
,
1810 &r300
->nr_vertex_buffers
,
1811 buffers
, start_slot
, count
);
1813 /* There must be at least one vertex buffer set, otherwise it locks up. */
1814 if (!r300
->nr_vertex_buffers
) {
1815 util_set_vertex_buffers_count(r300
->vertex_buffer
,
1816 &r300
->nr_vertex_buffers
,
1817 &r300
->dummy_vb
, 0, 1);
1820 r300
->vertex_arrays_dirty
= TRUE
;
1823 static void r300_set_vertex_buffers_swtcl(struct pipe_context
* pipe
,
1824 unsigned start_slot
, unsigned count
,
1825 const struct pipe_vertex_buffer
* buffers
)
1827 struct r300_context
* r300
= r300_context(pipe
);
1830 util_set_vertex_buffers_count(r300
->vertex_buffer
,
1831 &r300
->nr_vertex_buffers
,
1832 buffers
, start_slot
, count
);
1833 draw_set_vertex_buffers(r300
->draw
, start_slot
, count
, buffers
);
1838 for (i
= 0; i
< count
; i
++) {
1839 if (buffers
[i
].user_buffer
) {
1840 draw_set_mapped_vertex_buffer(r300
->draw
, start_slot
+ i
,
1841 buffers
[i
].user_buffer
, ~0);
1842 } else if (buffers
[i
].buffer
) {
1843 draw_set_mapped_vertex_buffer(r300
->draw
, start_slot
+ i
,
1844 r300_resource(buffers
[i
].buffer
)->malloced_buffer
, ~0);
1849 static void r300_set_index_buffer_hwtcl(struct pipe_context
* pipe
,
1850 const struct pipe_index_buffer
*ib
)
1852 struct r300_context
* r300
= r300_context(pipe
);
1855 pipe_resource_reference(&r300
->index_buffer
.buffer
, ib
->buffer
);
1856 memcpy(&r300
->index_buffer
, ib
, sizeof(*ib
));
1858 pipe_resource_reference(&r300
->index_buffer
.buffer
, NULL
);
1862 static void r300_set_index_buffer_swtcl(struct pipe_context
* pipe
,
1863 const struct pipe_index_buffer
*ib
)
1865 struct r300_context
* r300
= r300_context(pipe
);
1868 const void *buf
= NULL
;
1869 if (ib
->user_buffer
) {
1870 buf
= ib
->user_buffer
;
1871 } else if (ib
->buffer
) {
1872 buf
= r300_resource(ib
->buffer
)->malloced_buffer
;
1874 draw_set_indexes(r300
->draw
,
1875 (const ubyte
*) buf
+ ib
->offset
,
1876 ib
->index_size
, ~0);
1880 /* Initialize the PSC tables. */
1881 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1883 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1884 uint16_t type
, swizzle
;
1885 enum pipe_format format
;
1888 /* Vertex shaders have no semantics on their inputs,
1889 * so PSC should just route stuff based on the vertex elements,
1890 * and not on attrib information. */
1891 for (i
= 0; i
< velems
->count
; i
++) {
1892 format
= velems
->velem
[i
].src_format
;
1894 type
= r300_translate_vertex_data_type(format
);
1895 if (type
== R300_INVALID_FORMAT
) {
1896 fprintf(stderr
, "r300: Bad vertex format %s.\n",
1897 util_format_short_name(format
));
1902 type
|= i
<< R300_DST_VEC_LOC_SHIFT
;
1903 swizzle
= r300_translate_vertex_data_swizzle(format
);
1906 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1907 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1909 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1910 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1914 /* Set the last vector in the PSC. */
1918 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1919 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1921 vstream
->count
= (i
>> 1) + 1;
1924 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1926 const struct pipe_vertex_element
* attribs
)
1928 struct r300_vertex_element_state
*velems
;
1930 struct pipe_vertex_element dummy_attrib
= {0};
1932 /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */
1934 dummy_attrib
.src_format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
1935 attribs
= &dummy_attrib
;
1937 } else if (count
> 16) {
1938 fprintf(stderr
, "r300: More than 16 vertex elements are not supported,"
1939 " requested %i, using 16.\n", count
);
1943 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1947 velems
->count
= count
;
1948 memcpy(velems
->velem
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
1950 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1952 * The unused components will be replaced by (..., 0, 1). */
1953 r300_vertex_psc(velems
);
1955 for (i
= 0; i
< count
; i
++) {
1956 velems
->format_size
[i
] =
1957 align(util_format_get_blocksize(velems
->velem
[i
].src_format
), 4);
1958 velems
->vertex_size_dwords
+= velems
->format_size
[i
] / 4;
1965 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1968 struct r300_context
*r300
= r300_context(pipe
);
1969 struct r300_vertex_element_state
*velems
= state
;
1971 if (velems
== NULL
) {
1975 r300
->velems
= velems
;
1978 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1982 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1983 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1984 r300
->vertex_arrays_dirty
= TRUE
;
1987 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1992 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1993 const struct pipe_shader_state
* shader
)
1995 struct r300_context
* r300
= r300_context(pipe
);
1996 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1998 /* Copy state directly into shader. */
1999 vs
->state
= *shader
;
2000 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
2002 if (r300
->screen
->caps
.has_tcl
) {
2003 r300_init_vs_outputs(r300
, vs
);
2004 r300_translate_vertex_shader(r300
, vs
);
2006 r300_draw_init_vertex_shader(r300
, vs
);
2012 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
2014 struct r300_context
* r300
= r300_context(pipe
);
2015 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
2018 r300
->vs_state
.state
= NULL
;
2021 if (vs
== r300
->vs_state
.state
) {
2024 r300
->vs_state
.state
= vs
;
2026 /* The majority of the RS block bits is dependent on the vertex shader. */
2027 r300_mark_atom_dirty(r300
, &r300
->rs_block_state
); /* Will be updated before the emission. */
2029 if (r300
->screen
->caps
.has_tcl
) {
2030 unsigned fc_op_dwords
= r300
->screen
->caps
.is_r500
? 3 : 2;
2031 r300_mark_atom_dirty(r300
, &r300
->vs_state
);
2032 r300
->vs_state
.size
= vs
->code
.length
+ 9 +
2033 (R300_VS_MAX_FC_OPS
* fc_op_dwords
+ 4);
2035 r300_mark_atom_dirty(r300
, &r300
->vs_constants
);
2036 r300
->vs_constants
.size
=
2038 (vs
->externals_count
? vs
->externals_count
* 4 + 3 : 0) +
2039 (vs
->immediates_count
? vs
->immediates_count
* 4 + 3 : 0);
2041 ((struct r300_constant_buffer
*)r300
->vs_constants
.state
)->remap_table
=
2042 vs
->code
.constants_remap_table
;
2044 r300_mark_atom_dirty(r300
, &r300
->pvs_flush
);
2046 draw_bind_vertex_shader(r300
->draw
,
2047 (struct draw_vertex_shader
*)vs
->draw_vs
);
2051 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
2053 struct r300_context
* r300
= r300_context(pipe
);
2054 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
2056 if (r300
->screen
->caps
.has_tcl
) {
2057 rc_constants_destroy(&vs
->code
.constants
);
2058 FREE(vs
->code
.constants_remap_table
);
2060 draw_delete_vertex_shader(r300
->draw
,
2061 (struct draw_vertex_shader
*)vs
->draw_vs
);
2064 FREE((void*)vs
->state
.tokens
);
2068 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
2069 uint shader
, uint index
,
2070 struct pipe_constant_buffer
*cb
)
2072 struct r300_context
* r300
= r300_context(pipe
);
2073 struct r300_constant_buffer
*cbuf
;
2076 if (!cb
|| (!cb
->buffer
&& !cb
->user_buffer
))
2080 case PIPE_SHADER_VERTEX
:
2081 cbuf
= (struct r300_constant_buffer
*)r300
->vs_constants
.state
;
2083 case PIPE_SHADER_FRAGMENT
:
2084 cbuf
= (struct r300_constant_buffer
*)r300
->fs_constants
.state
;
2091 if (cb
->user_buffer
)
2092 mapped
= (uint32_t*)cb
->user_buffer
;
2094 struct r300_resource
*rbuf
= r300_resource(cb
->buffer
);
2096 if (rbuf
&& rbuf
->malloced_buffer
)
2097 mapped
= (uint32_t*)rbuf
->malloced_buffer
;
2102 if (shader
== PIPE_SHADER_FRAGMENT
||
2103 (shader
== PIPE_SHADER_VERTEX
&& r300
->screen
->caps
.has_tcl
)) {
2107 if (shader
== PIPE_SHADER_VERTEX
) {
2108 if (r300
->screen
->caps
.has_tcl
) {
2109 struct r300_vertex_shader
*vs
=
2110 (struct r300_vertex_shader
*)r300
->vs_state
.state
;
2113 cbuf
->buffer_base
= 0;
2117 cbuf
->buffer_base
= r300
->vs_const_base
;
2118 r300
->vs_const_base
+= vs
->code
.constants
.Count
;
2119 if (r300
->vs_const_base
> R500_MAX_PVS_CONST_VECS
) {
2120 r300
->vs_const_base
= vs
->code
.constants
.Count
;
2121 cbuf
->buffer_base
= 0;
2122 r300_mark_atom_dirty(r300
, &r300
->pvs_flush
);
2124 r300_mark_atom_dirty(r300
, &r300
->vs_constants
);
2125 } else if (r300
->draw
) {
2126 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
2127 0, mapped
, cb
->buffer_size
);
2129 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
2130 r300_mark_atom_dirty(r300
, &r300
->fs_constants
);
2134 static void r300_texture_barrier(struct pipe_context
*pipe
)
2136 struct r300_context
*r300
= r300_context(pipe
);
2138 r300_mark_atom_dirty(r300
, &r300
->gpu_flush
);
2139 r300_mark_atom_dirty(r300
, &r300
->texture_cache_inval
);
2142 static void r300_memory_barrier(struct pipe_context
*pipe
, unsigned flags
)
2146 void r300_init_state_functions(struct r300_context
* r300
)
2148 r300
->context
.create_blend_state
= r300_create_blend_state
;
2149 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
2150 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
2152 r300
->context
.set_blend_color
= r300_set_blend_color
;
2154 r300
->context
.set_clip_state
= r300_set_clip_state
;
2155 r300
->context
.set_sample_mask
= r300_set_sample_mask
;
2157 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
2159 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
2160 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
2161 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
2163 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
2165 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
2167 r300
->context
.create_fs_state
= r300_create_fs_state
;
2168 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
2169 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
2171 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
2173 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
2174 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
2175 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
2177 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
2178 r300
->context
.bind_sampler_states
= r300_bind_sampler_states
;
2179 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
2181 r300
->context
.set_sampler_views
= r300_set_sampler_views
;
2182 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
2183 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
2185 r300
->context
.set_scissor_states
= r300_set_scissor_states
;
2187 r300
->context
.set_viewport_states
= r300_set_viewport_states
;
2189 if (r300
->screen
->caps
.has_tcl
) {
2190 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers_hwtcl
;
2191 r300
->context
.set_index_buffer
= r300_set_index_buffer_hwtcl
;
2193 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers_swtcl
;
2194 r300
->context
.set_index_buffer
= r300_set_index_buffer_swtcl
;
2197 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
2198 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
2199 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
2201 r300
->context
.create_vs_state
= r300_create_vs_state
;
2202 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
2203 r300
->context
.delete_vs_state
= r300_delete_vs_state
;
2205 r300
->context
.texture_barrier
= r300_texture_barrier
;
2206 r300
->context
.memory_barrier
= r300_memory_barrier
;