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_math.h"
29 #include "util/u_mm.h"
30 #include "util/u_memory.h"
31 #include "util/u_pack_color.h"
32 #include "util/u_transfer.h"
34 #include "tgsi/tgsi_parse.h"
36 #include "pipe/p_config.h"
39 #include "r300_context.h"
40 #include "r300_emit.h"
42 #include "r300_screen.h"
43 #include "r300_screen_buffer.h"
44 #include "r300_state_inlines.h"
46 #include "r300_texture.h"
49 /* r300_state: Functions used to intialize state context by translating
50 * Gallium state objects into semi-native r300 state objects. */
52 #define UPDATE_STATE(cso, atom) \
53 if (cso != atom.state) { \
55 r300_mark_atom_dirty(r300, &(atom)); \
58 static boolean
blend_discard_if_src_alpha_0(unsigned srcRGB
, unsigned srcA
,
59 unsigned dstRGB
, unsigned dstA
)
61 /* If the blend equation is ADD or REVERSE_SUBTRACT,
62 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
63 * will not be changed.
64 * Notice that the dst factors are the src factors inverted. */
65 return (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
66 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
67 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
68 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
69 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
70 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
71 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
72 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
73 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
74 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
75 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
76 dstA
== PIPE_BLENDFACTOR_ONE
);
79 static boolean
blend_discard_if_src_alpha_1(unsigned srcRGB
, unsigned srcA
,
80 unsigned dstRGB
, unsigned dstA
)
82 /* If the blend equation is ADD or REVERSE_SUBTRACT,
83 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
84 * will not be changed.
85 * Notice that the dst factors are the src factors inverted. */
86 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
87 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
88 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
89 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
90 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
91 (dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
92 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
93 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
94 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
95 dstA
== PIPE_BLENDFACTOR_ONE
);
98 static boolean
blend_discard_if_src_color_0(unsigned srcRGB
, unsigned srcA
,
99 unsigned dstRGB
, unsigned dstA
)
101 /* If the blend equation is ADD or REVERSE_SUBTRACT,
102 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
103 * will not be changed.
104 * Notice that the dst factors are the src factors inverted. */
105 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
106 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
107 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
108 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
109 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
110 (dstA
== PIPE_BLENDFACTOR_ONE
);
113 static boolean
blend_discard_if_src_color_1(unsigned srcRGB
, unsigned srcA
,
114 unsigned dstRGB
, unsigned dstA
)
116 /* If the blend equation is ADD or REVERSE_SUBTRACT,
117 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
118 * will not be changed.
119 * Notice that the dst factors are the src factors inverted. */
120 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
121 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
122 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
123 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
124 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
125 (dstA
== PIPE_BLENDFACTOR_ONE
);
128 static boolean
blend_discard_if_src_alpha_color_0(unsigned srcRGB
, unsigned srcA
,
129 unsigned dstRGB
, unsigned dstA
)
131 /* If the blend equation is ADD or REVERSE_SUBTRACT,
132 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
133 * the colorbuffer will not be changed.
134 * Notice that the dst factors are the src factors inverted. */
135 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
136 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
137 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
138 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
139 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
140 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
141 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
142 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
143 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
144 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
145 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
146 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
147 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
148 dstA
== PIPE_BLENDFACTOR_ONE
);
151 static boolean
blend_discard_if_src_alpha_color_1(unsigned srcRGB
, unsigned srcA
,
152 unsigned dstRGB
, unsigned dstA
)
154 /* If the blend equation is ADD or REVERSE_SUBTRACT,
155 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
156 * the colorbuffer will not be changed.
157 * Notice that the dst factors are the src factors inverted. */
158 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
159 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
160 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
161 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
162 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
163 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
164 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
165 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
166 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
167 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
168 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
169 dstA
== PIPE_BLENDFACTOR_ONE
);
172 static unsigned bgra_cmask(unsigned mask
)
174 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
176 return ((mask
& PIPE_MASK_R
) << 2) |
177 ((mask
& PIPE_MASK_B
) >> 2) |
178 (mask
& (PIPE_MASK_G
| PIPE_MASK_A
));
181 /* Create a new blend state based on the CSO blend state.
183 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
184 static void* r300_create_blend_state(struct pipe_context
* pipe
,
185 const struct pipe_blend_state
* state
)
187 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
188 struct r300_blend_state
* blend
= CALLOC_STRUCT(r300_blend_state
);
189 uint32_t blend_control
= 0; /* R300_RB3D_CBLEND: 0x4e04 */
190 uint32_t blend_control_noclamp
= 0; /* R300_RB3D_CBLEND: 0x4e04 */
191 uint32_t alpha_blend_control
= 0; /* R300_RB3D_ABLEND: 0x4e08 */
192 uint32_t alpha_blend_control_noclamp
= 0; /* R300_RB3D_ABLEND: 0x4e08 */
193 uint32_t color_channel_mask
= 0; /* R300_RB3D_COLOR_CHANNEL_MASK: 0x4e0c */
194 uint32_t rop
= 0; /* R300_RB3D_ROPCNTL: 0x4e18 */
195 uint32_t dither
= 0; /* R300_RB3D_DITHER_CTL: 0x4e50 */
198 blend
->state
= *state
;
200 if (state
->rt
[0].blend_enable
)
202 unsigned eqRGB
= state
->rt
[0].rgb_func
;
203 unsigned srcRGB
= state
->rt
[0].rgb_src_factor
;
204 unsigned dstRGB
= state
->rt
[0].rgb_dst_factor
;
206 unsigned eqA
= state
->rt
[0].alpha_func
;
207 unsigned srcA
= state
->rt
[0].alpha_src_factor
;
208 unsigned dstA
= state
->rt
[0].alpha_dst_factor
;
210 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
211 * this is just the crappy D3D naming */
212 blend_control
= blend_control_noclamp
=
213 R300_ALPHA_BLEND_ENABLE
|
214 ( r300_translate_blend_factor(srcRGB
) << R300_SRC_BLEND_SHIFT
) |
215 ( r300_translate_blend_factor(dstRGB
) << R300_DST_BLEND_SHIFT
);
217 r300_translate_blend_function(eqRGB
, TRUE
);
218 blend_control_noclamp
|=
219 r300_translate_blend_function(eqRGB
, FALSE
);
221 /* Optimization: some operations do not require the destination color.
223 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
224 * otherwise blending gives incorrect results. It seems to be
226 if (eqRGB
== PIPE_BLEND_MIN
|| eqA
== PIPE_BLEND_MIN
||
227 eqRGB
== PIPE_BLEND_MAX
|| eqA
== PIPE_BLEND_MAX
||
228 dstRGB
!= PIPE_BLENDFACTOR_ZERO
||
229 dstA
!= PIPE_BLENDFACTOR_ZERO
||
230 srcRGB
== PIPE_BLENDFACTOR_DST_COLOR
||
231 srcRGB
== PIPE_BLENDFACTOR_DST_ALPHA
||
232 srcRGB
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
233 srcRGB
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
234 srcA
== PIPE_BLENDFACTOR_DST_COLOR
||
235 srcA
== PIPE_BLENDFACTOR_DST_ALPHA
||
236 srcA
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
237 srcA
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
238 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
) {
239 /* Enable reading from the colorbuffer. */
240 blend_control
|= R300_READ_ENABLE
;
241 blend_control_noclamp
|= R300_READ_ENABLE
;
243 if (r300screen
->caps
.is_r500
) {
244 /* Optimization: Depending on incoming pixels, we can
245 * conditionally disable the reading in hardware... */
246 if (eqRGB
!= PIPE_BLEND_MIN
&& eqA
!= PIPE_BLEND_MIN
&&
247 eqRGB
!= PIPE_BLEND_MAX
&& eqA
!= PIPE_BLEND_MAX
) {
248 /* Disable reading if SRC_ALPHA == 0. */
249 if ((dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
250 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
251 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
252 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
253 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
254 blend_control
|= R500_SRC_ALPHA_0_NO_READ
;
257 /* Disable reading if SRC_ALPHA == 1. */
258 if ((dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
259 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
260 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
261 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
262 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
263 blend_control
|= R500_SRC_ALPHA_1_NO_READ
;
269 /* Optimization: discard pixels which don't change the colorbuffer.
271 * The code below is non-trivial and some math is involved.
273 * Discarding pixels must be disabled when FP16 AA is enabled.
274 * This is a hardware bug. Also, this implementation wouldn't work
275 * with FP blending enabled and equation clamping disabled.
277 * Equations other than ADD are rarely used and therefore won't be
279 if ((eqRGB
== PIPE_BLEND_ADD
|| eqRGB
== PIPE_BLEND_REVERSE_SUBTRACT
) &&
280 (eqA
== PIPE_BLEND_ADD
|| eqA
== PIPE_BLEND_REVERSE_SUBTRACT
)) {
282 * REVERSE_SUBTRACT: Y-X
285 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
286 * then CB will not be changed.
288 * Given the srcFactor and dstFactor variables, we can derive
289 * what src and dst should be equal to and discard appropriate
292 if (blend_discard_if_src_alpha_0(srcRGB
, srcA
, dstRGB
, dstA
)) {
293 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0
;
294 } else if (blend_discard_if_src_alpha_1(srcRGB
, srcA
,
296 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1
;
297 } else if (blend_discard_if_src_color_0(srcRGB
, srcA
,
299 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0
;
300 } else if (blend_discard_if_src_color_1(srcRGB
, srcA
,
302 blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1
;
303 } else if (blend_discard_if_src_alpha_color_0(srcRGB
, srcA
,
306 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0
;
307 } else if (blend_discard_if_src_alpha_color_1(srcRGB
, srcA
,
310 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1
;
315 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
316 blend_control
|= R300_SEPARATE_ALPHA_ENABLE
;
317 blend_control_noclamp
|= R300_SEPARATE_ALPHA_ENABLE
;
318 alpha_blend_control
= alpha_blend_control_noclamp
=
319 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
320 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
321 alpha_blend_control
|=
322 r300_translate_blend_function(eqA
, TRUE
);
323 alpha_blend_control_noclamp
|=
324 r300_translate_blend_function(eqA
, FALSE
);
328 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
329 if (state
->logicop_enable
) {
330 rop
= R300_RB3D_ROPCNTL_ROP_ENABLE
|
331 (state
->logicop_func
) << R300_RB3D_ROPCNTL_ROP_SHIFT
;
334 /* Color channel masks for all MRTs. */
335 color_channel_mask
= bgra_cmask(state
->rt
[0].colormask
);
336 if (r300screen
->caps
.is_r500
&& state
->independent_blend_enable
) {
337 if (state
->rt
[1].blend_enable
) {
338 color_channel_mask
|= bgra_cmask(state
->rt
[1].colormask
) << 4;
340 if (state
->rt
[2].blend_enable
) {
341 color_channel_mask
|= bgra_cmask(state
->rt
[2].colormask
) << 8;
343 if (state
->rt
[3].blend_enable
) {
344 color_channel_mask
|= bgra_cmask(state
->rt
[3].colormask
) << 12;
348 /* Neither fglrx nor classic r300 ever set this, regardless of dithering
349 * state. Since it's an optional implementation detail, we can leave it
350 * out and never dither.
352 * This could be revisited if we ever get quality or conformance hints.
355 dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
356 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
360 /* Build a command buffer. */
361 BEGIN_CB(blend
->cb_clamp
, 8);
362 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
363 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
364 OUT_CB(blend_control
);
365 OUT_CB(alpha_blend_control
);
366 OUT_CB(color_channel_mask
);
367 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
370 /* Build a command buffer. */
371 BEGIN_CB(blend
->cb_noclamp
, 8);
372 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
373 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
374 OUT_CB(blend_control_noclamp
);
375 OUT_CB(alpha_blend_control_noclamp
);
376 OUT_CB(color_channel_mask
);
377 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
380 /* The same as above, but with no colorbuffer reads and writes. */
381 BEGIN_CB(blend
->cb_no_readwrite
, 8);
382 OUT_CB_REG(R300_RB3D_ROPCNTL
, rop
);
383 OUT_CB_REG_SEQ(R300_RB3D_CBLEND
, 3);
387 OUT_CB_REG(R300_RB3D_DITHER_CTL
, dither
);
393 /* Bind blend state. */
394 static void r300_bind_blend_state(struct pipe_context
* pipe
,
397 struct r300_context
* r300
= r300_context(pipe
);
399 UPDATE_STATE(state
, r300
->blend_state
);
402 /* Free blend state. */
403 static void r300_delete_blend_state(struct pipe_context
* pipe
,
409 /* Convert float to 10bit integer */
410 static unsigned float_to_fixed10(float f
)
412 return CLAMP((unsigned)(f
* 1023.9f
), 0, 1023);
416 * Setup both R300 and R500 registers, figure out later which one to write. */
417 static void r300_set_blend_color(struct pipe_context
* pipe
,
418 const struct pipe_blend_color
* color
)
420 struct r300_context
* r300
= r300_context(pipe
);
421 struct pipe_framebuffer_state
*fb
= r300
->fb_state
.state
;
422 struct r300_blend_color_state
*state
=
423 (struct r300_blend_color_state
*)r300
->blend_color_state
.state
;
424 struct pipe_blend_color c
;
425 enum pipe_format format
= fb
->nr_cbufs
? fb
->cbufs
[0]->format
: 0;
428 state
->state
= *color
; /* Save it, so that we can reuse it in set_fb_state */
431 /* The blend color is dependent on the colorbuffer format. */
434 case PIPE_FORMAT_R8_UNORM
:
435 case PIPE_FORMAT_L8_UNORM
:
436 case PIPE_FORMAT_I8_UNORM
:
437 c
.color
[1] = c
.color
[0];
440 case PIPE_FORMAT_A8_UNORM
:
441 c
.color
[1] = c
.color
[3];
444 case PIPE_FORMAT_R8G8_UNORM
:
445 c
.color
[2] = c
.color
[1];
448 case PIPE_FORMAT_L8A8_UNORM
:
449 c
.color
[2] = c
.color
[3];
456 if (r300
->screen
->caps
.is_r500
) {
457 BEGIN_CB(state
->cb
, 3);
458 OUT_CB_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR
, 2);
461 case PIPE_FORMAT_R16G16B16A16_FLOAT
:
462 OUT_CB(util_float_to_half(c
.color
[2]) |
463 (util_float_to_half(c
.color
[3]) << 16));
464 OUT_CB(util_float_to_half(c
.color
[0]) |
465 (util_float_to_half(c
.color
[1]) << 16));
469 OUT_CB(float_to_fixed10(c
.color
[0]) |
470 (float_to_fixed10(c
.color
[3]) << 16));
471 OUT_CB(float_to_fixed10(c
.color
[2]) |
472 (float_to_fixed10(c
.color
[1]) << 16));
478 util_pack_color(c
.color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
480 BEGIN_CB(state
->cb
, 2);
481 OUT_CB_REG(R300_RB3D_BLEND_COLOR
, uc
.ui
);
485 r300_mark_atom_dirty(r300
, &r300
->blend_color_state
);
488 static void r300_set_clip_state(struct pipe_context
* pipe
,
489 const struct pipe_clip_state
* state
)
491 struct r300_context
* r300
= r300_context(pipe
);
492 struct r300_clip_state
*clip
=
493 (struct r300_clip_state
*)r300
->clip_state
.state
;
498 if (r300
->screen
->caps
.has_tcl
) {
499 r300
->clip_state
.size
= 2 + !!state
->nr
* 3 + state
->nr
* 4;
501 BEGIN_CB(clip
->cb
, r300
->clip_state
.size
);
503 OUT_CB_REG(R300_VAP_PVS_VECTOR_INDX_REG
,
504 (r300
->screen
->caps
.is_r500
?
505 R500_PVS_UCP_START
: R300_PVS_UCP_START
));
506 OUT_CB_ONE_REG(R300_VAP_PVS_UPLOAD_DATA
, state
->nr
* 4);
507 OUT_CB_TABLE(state
->ucp
, state
->nr
* 4);
509 OUT_CB_REG(R300_VAP_CLIP_CNTL
, ((1 << state
->nr
) - 1) |
510 R300_PS_UCP_MODE_CLIP_AS_TRIFAN
);
513 r300_mark_atom_dirty(r300
, &r300
->clip_state
);
515 draw_set_clip_state(r300
->draw
, state
);
520 r300_set_sample_mask(struct pipe_context
*pipe
,
521 unsigned sample_mask
)
526 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
528 * This contains the depth buffer, stencil buffer, alpha test, and such.
529 * On the Radeon, depth and stencil buffer setup are intertwined, which is
530 * the reason for some of the strange-looking assignments across registers. */
532 r300_create_dsa_state(struct pipe_context
* pipe
,
533 const struct pipe_depth_stencil_alpha_state
* state
)
535 struct r300_capabilities
*caps
= &r300_screen(pipe
->screen
)->caps
;
536 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
541 /* Depth test setup. - separate write mask depth for decomp flush */
542 if (state
->depth
.writemask
) {
543 dsa
->z_buffer_control
|= R300_Z_WRITE_ENABLE
;
546 if (state
->depth
.enabled
) {
547 dsa
->z_buffer_control
|= R300_Z_ENABLE
;
549 dsa
->z_stencil_control
|=
550 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
554 /* Stencil buffer setup. */
555 if (state
->stencil
[0].enabled
) {
556 dsa
->z_buffer_control
|= R300_STENCIL_ENABLE
;
557 dsa
->z_stencil_control
|=
558 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
559 R300_S_FRONT_FUNC_SHIFT
) |
560 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
561 R300_S_FRONT_SFAIL_OP_SHIFT
) |
562 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
563 R300_S_FRONT_ZPASS_OP_SHIFT
) |
564 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
565 R300_S_FRONT_ZFAIL_OP_SHIFT
);
567 dsa
->stencil_ref_mask
=
568 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
569 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
571 if (state
->stencil
[1].enabled
) {
572 dsa
->two_sided
= TRUE
;
574 dsa
->z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
575 dsa
->z_stencil_control
|=
576 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
577 R300_S_BACK_FUNC_SHIFT
) |
578 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
579 R300_S_BACK_SFAIL_OP_SHIFT
) |
580 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
581 R300_S_BACK_ZPASS_OP_SHIFT
) |
582 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
583 R300_S_BACK_ZFAIL_OP_SHIFT
);
585 dsa
->stencil_ref_bf
=
586 (state
->stencil
[1].valuemask
<< R300_STENCILMASK_SHIFT
) |
587 (state
->stencil
[1].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
590 dsa
->z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
592 dsa
->two_sided_stencil_ref
=
593 (state
->stencil
[0].valuemask
!= state
->stencil
[1].valuemask
||
594 state
->stencil
[0].writemask
!= state
->stencil
[1].writemask
);
599 /* Alpha test setup. */
600 if (state
->alpha
.enabled
) {
601 dsa
->alpha_function
=
602 r300_translate_alpha_function(state
->alpha
.func
) |
603 R300_FG_ALPHA_FUNC_ENABLE
;
605 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
606 dsa
->alpha_value
= util_float_to_half(state
->alpha
.ref_value
);
609 dsa
->alpha_function_fp16
= dsa
->alpha_function
|
610 R500_FG_ALPHA_FUNC_FP16_ENABLE
;
611 dsa
->alpha_function
|= R500_FG_ALPHA_FUNC_8BIT
;
615 BEGIN_CB(&dsa
->cb_begin
, 10);
616 OUT_CB_REG(R300_FG_ALPHA_FUNC
, dsa
->alpha_function
);
617 OUT_CB_REG_SEQ(R300_ZB_CNTL
, 3);
618 OUT_CB(dsa
->z_buffer_control
);
619 OUT_CB(dsa
->z_stencil_control
);
620 OUT_CB(dsa
->stencil_ref_mask
);
621 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF
, dsa
->stencil_ref_bf
);
622 OUT_CB_REG(R500_FG_ALPHA_VALUE
, dsa
->alpha_value
);
625 BEGIN_CB(&dsa
->cb_begin_fp16
, 10);
626 OUT_CB_REG(R300_FG_ALPHA_FUNC
, dsa
->alpha_function_fp16
);
627 OUT_CB_REG_SEQ(R300_ZB_CNTL
, 3);
628 OUT_CB(dsa
->z_buffer_control
);
629 OUT_CB(dsa
->z_stencil_control
);
630 OUT_CB(dsa
->stencil_ref_mask
);
631 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF
, dsa
->stencil_ref_bf
);
632 OUT_CB_REG(R500_FG_ALPHA_VALUE
, dsa
->alpha_value
);
635 BEGIN_CB(dsa
->cb_zb_no_readwrite
, 10);
636 OUT_CB_REG(R300_FG_ALPHA_FUNC
, dsa
->alpha_function
);
637 OUT_CB_REG_SEQ(R300_ZB_CNTL
, 3);
641 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF
, 0);
642 OUT_CB_REG(R500_FG_ALPHA_VALUE
, dsa
->alpha_value
);
645 BEGIN_CB(dsa
->cb_fp16_zb_no_readwrite
, 10);
646 OUT_CB_REG(R300_FG_ALPHA_FUNC
, dsa
->alpha_function_fp16
);
647 OUT_CB_REG_SEQ(R300_ZB_CNTL
, 3);
651 OUT_CB_REG(R500_ZB_STENCILREFMASK_BF
, 0);
652 OUT_CB_REG(R500_FG_ALPHA_VALUE
, dsa
->alpha_value
);
658 static void r300_dsa_inject_stencilref(struct r300_context
*r300
)
660 struct r300_dsa_state
*dsa
=
661 (struct r300_dsa_state
*)r300
->dsa_state
.state
;
666 dsa
->stencil_ref_mask
=
667 (dsa
->stencil_ref_mask
& ~R300_STENCILREF_MASK
) |
668 r300
->stencil_ref
.ref_value
[0];
669 dsa
->stencil_ref_bf
=
670 (dsa
->stencil_ref_bf
& ~R300_STENCILREF_MASK
) |
671 r300
->stencil_ref
.ref_value
[1];
674 /* Bind DSA state. */
675 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
678 struct r300_context
* r300
= r300_context(pipe
);
684 UPDATE_STATE(state
, r300
->dsa_state
);
686 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
); /* Will be updated before the emission. */
687 r300_dsa_inject_stencilref(r300
);
690 /* Free DSA state. */
691 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
697 static void r300_set_stencil_ref(struct pipe_context
* pipe
,
698 const struct pipe_stencil_ref
* sr
)
700 struct r300_context
* r300
= r300_context(pipe
);
702 r300
->stencil_ref
= *sr
;
704 r300_dsa_inject_stencilref(r300
);
705 r300_mark_atom_dirty(r300
, &r300
->dsa_state
);
708 static void r300_tex_set_tiling_flags(struct r300_context
*r300
,
709 struct r300_resource
*tex
,
712 /* Check if the macrotile flag needs to be changed.
713 * Skip changing the flags otherwise. */
714 if (tex
->tex
.macrotile
[tex
->surface_level
] !=
715 tex
->tex
.macrotile
[level
]) {
716 r300
->rws
->buffer_set_tiling(tex
->buf
, r300
->cs
,
717 tex
->tex
.microtile
, tex
->tex
.macrotile
[level
],
718 tex
->tex
.stride_in_bytes
[0]);
720 tex
->surface_level
= level
;
724 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
725 static void r300_fb_set_tiling_flags(struct r300_context
*r300
,
726 const struct pipe_framebuffer_state
*state
)
730 /* Set tiling flags for new surfaces. */
731 for (i
= 0; i
< state
->nr_cbufs
; i
++) {
732 r300_tex_set_tiling_flags(r300
,
733 r300_resource(state
->cbufs
[i
]->texture
),
734 state
->cbufs
[i
]->u
.tex
.level
);
737 r300_tex_set_tiling_flags(r300
,
738 r300_resource(state
->zsbuf
->texture
),
739 state
->zsbuf
->u
.tex
.level
);
743 static void r300_print_fb_surf_info(struct pipe_surface
*surf
, unsigned index
,
746 struct pipe_resource
*tex
= surf
->texture
;
747 struct r300_resource
*rtex
= r300_resource(tex
);
750 "r300: %s[%i] Dim: %ix%i, Firstlayer: %i, "
751 "Lastlayer: %i, Level: %i, Format: %s\n"
753 "r300: TEX: Macro: %s, Micro: %s, Pitch: %i, "
754 "Dim: %ix%ix%i, LastLevel: %i, Format: %s\n",
756 binding
, index
, surf
->width
, surf
->height
,
757 surf
->u
.tex
.first_layer
, surf
->u
.tex
.last_layer
, surf
->u
.tex
.level
,
758 util_format_short_name(surf
->format
),
760 rtex
->tex
.macrotile
[0] ? "YES" : " NO",
761 rtex
->tex
.microtile
? "YES" : " NO",
762 rtex
->tex
.stride_in_pixels
[0],
763 tex
->width0
, tex
->height0
, tex
->depth0
,
764 tex
->last_level
, util_format_short_name(tex
->format
));
767 void r300_mark_fb_state_dirty(struct r300_context
*r300
,
768 enum r300_fb_state_change change
)
770 struct pipe_framebuffer_state
*state
= r300
->fb_state
.state
;
771 boolean can_hyperz
= r300
->rws
->get_value(r300
->rws
, RADEON_VID_CAN_HYPERZ
);
773 r300_mark_atom_dirty(r300
, &r300
->gpu_flush
);
774 r300_mark_atom_dirty(r300
, &r300
->fb_state
);
776 /* What is marked as dirty depends on the enum r300_fb_state_change. */
777 if (change
== R300_CHANGED_FB_STATE
) {
778 r300_mark_atom_dirty(r300
, &r300
->aa_state
);
779 r300_mark_atom_dirty(r300
, &r300
->dsa_state
); /* for AlphaRef */
780 r300_set_blend_color(&r300
->context
, r300
->blend_color_state
.state
);
783 if (change
== R300_CHANGED_FB_STATE
||
784 change
== R300_CHANGED_HYPERZ_FLAG
) {
785 r300_mark_atom_dirty(r300
, &r300
->hyperz_state
);
788 if (change
== R300_CHANGED_FB_STATE
||
789 change
== R300_CHANGED_MULTIWRITE
) {
790 r300_mark_atom_dirty(r300
, &r300
->fb_state_pipelined
);
793 /* Now compute the fb_state atom size. */
794 r300
->fb_state
.size
= 2 + (8 * state
->nr_cbufs
);
796 if (r300
->cbzb_clear
)
797 r300
->fb_state
.size
+= 10;
798 else if (state
->zsbuf
) {
799 r300
->fb_state
.size
+= 10;
801 r300
->fb_state
.size
+= 8;
804 /* The size of the rest of atoms stays the same. */
808 r300_set_framebuffer_state(struct pipe_context
* pipe
,
809 const struct pipe_framebuffer_state
* state
)
811 struct r300_context
* r300
= r300_context(pipe
);
812 struct r300_aa_state
*aa
= (struct r300_aa_state
*)r300
->aa_state
.state
;
813 struct pipe_framebuffer_state
*old_state
= r300
->fb_state
.state
;
814 unsigned max_width
, max_height
, i
;
815 uint32_t zbuffer_bpp
= 0;
817 if (r300
->screen
->caps
.is_r500
) {
818 max_width
= max_height
= 4096;
819 } else if (r300
->screen
->caps
.is_r400
) {
820 max_width
= max_height
= 4021;
822 max_width
= max_height
= 2560;
825 if (state
->width
> max_width
|| state
->height
> max_height
) {
826 fprintf(stderr
, "r300: Implementation error: Render targets are too "
827 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
831 if (old_state
->zsbuf
&& r300
->zmask_in_use
&& !r300
->hyperz_locked
) {
832 /* There is a zmask in use, what are we gonna do? */
834 if (!pipe_surface_equal(old_state
->zsbuf
, state
->zsbuf
)) {
835 /* Decompress the currently bound zbuffer before we bind another one. */
836 r300_decompress_zmask(r300
);
837 r300
->hiz_in_use
= FALSE
;
840 /* We don't bind another zbuffer, so lock the current one. */
841 r300
->hyperz_locked
= TRUE
;
842 pipe_surface_reference(&r300
->locked_zbuffer
, old_state
->zsbuf
);
844 } else if (r300
->hyperz_locked
&& r300
->locked_zbuffer
) {
845 /* We have a locked zbuffer now, what are we gonna do? */
847 if (!pipe_surface_equal(r300
->locked_zbuffer
, state
->zsbuf
)) {
848 /* We are binding some other zbuffer, so decompress the locked one,
849 * it gets unlocked automatically. */
850 r300_decompress_zmask_locked_unsafe(r300
);
851 r300
->hiz_in_use
= FALSE
;
853 /* We are binding the locked zbuffer again, so unlock it. */
854 r300
->hyperz_locked
= FALSE
;
858 assert(state
->zsbuf
|| r300
->hyperz_locked
|| !r300
->zmask_in_use
);
860 /* Need to reset clamping or colormask. */
861 r300_mark_atom_dirty(r300
, &r300
->blend_state
);
863 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
864 if (!!old_state
->zsbuf
!= !!state
->zsbuf
) {
865 r300_mark_atom_dirty(r300
, &r300
->dsa_state
);
868 /* The tiling flags are dependent on the surface miplevel, unfortunately. */
869 r300_fb_set_tiling_flags(r300
, state
);
871 util_copy_framebuffer_state(r300
->fb_state
.state
, state
);
873 if (!r300
->hyperz_locked
) {
874 pipe_surface_reference(&r300
->locked_zbuffer
, NULL
);
877 r300_mark_fb_state_dirty(r300
, R300_CHANGED_FB_STATE
);
880 switch (util_format_get_blocksize(state
->zsbuf
->texture
->format
)) {
889 /* Polygon offset depends on the zbuffer bit depth. */
890 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
891 r300
->zbuffer_bpp
= zbuffer_bpp
;
893 if (r300
->polygon_offset_enabled
)
894 r300_mark_atom_dirty(r300
, &r300
->rs_state
);
898 /* Set up AA config. */
899 if (state
->nr_cbufs
&& state
->cbufs
[0]->texture
->nr_samples
> 1) {
900 aa
->aa_config
= R300_GB_AA_CONFIG_AA_ENABLE
;
902 switch (state
->cbufs
[0]->texture
->nr_samples
) {
904 aa
->aa_config
|= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2
;
907 aa
->aa_config
|= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3
;
910 aa
->aa_config
|= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4
;
913 aa
->aa_config
|= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6
;
920 if (DBG_ON(r300
, DBG_FB
)) {
921 fprintf(stderr
, "r300: set_framebuffer_state:\n");
922 for (i
= 0; i
< state
->nr_cbufs
; i
++) {
923 r300_print_fb_surf_info(state
->cbufs
[i
], i
, "CB");
926 r300_print_fb_surf_info(state
->zsbuf
, 0, "ZB");
931 /* Create fragment shader state. */
932 static void* r300_create_fs_state(struct pipe_context
* pipe
,
933 const struct pipe_shader_state
* shader
)
935 struct r300_fragment_shader
* fs
= NULL
;
937 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
939 /* Copy state directly into shader. */
941 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
946 void r300_mark_fs_code_dirty(struct r300_context
*r300
)
948 struct r300_fragment_shader
* fs
= r300_fs(r300
);
950 r300_mark_atom_dirty(r300
, &r300
->fs
);
951 r300_mark_atom_dirty(r300
, &r300
->fs_rc_constant_state
);
952 r300_mark_atom_dirty(r300
, &r300
->fs_constants
);
953 r300
->fs
.size
= fs
->shader
->cb_code_size
;
955 if (r300
->screen
->caps
.is_r500
) {
956 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 7;
957 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 3;
959 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 5;
960 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 1;
963 ((struct r300_constant_buffer
*)r300
->fs_constants
.state
)->remap_table
=
964 fs
->shader
->code
.constants_remap_table
;
967 /* Bind fragment shader state. */
968 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
970 struct r300_context
* r300
= r300_context(pipe
);
971 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
974 r300
->fs
.state
= NULL
;
979 r300
->fs_status
= FRAGMENT_SHADER_DIRTY
;
981 r300_mark_atom_dirty(r300
, &r300
->rs_block_state
); /* Will be updated before the emission. */
984 /* Delete fragment shader state. */
985 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
987 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
988 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
993 rc_constants_destroy(&tmp
->code
.constants
);
997 FREE((void*)fs
->state
.tokens
);
1001 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
1002 const struct pipe_poly_stipple
* state
)
1004 /* XXX no idea how to set this up, but not terribly important */
1007 /* Create a new rasterizer state based on the CSO rasterizer state.
1009 * This is a very large chunk of state, and covers most of the graphics
1010 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
1012 * In a not entirely unironic sidenote, this state has nearly nothing to do
1013 * with the actual block on the Radeon called the rasterizer (RS). */
1014 static void* r300_create_rs_state(struct pipe_context
* pipe
,
1015 const struct pipe_rasterizer_state
* state
)
1017 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
1019 uint32_t vap_control_status
; /* R300_VAP_CNTL_STATUS: 0x2140 */
1020 uint32_t point_size
; /* R300_GA_POINT_SIZE: 0x421c */
1021 uint32_t point_minmax
; /* R300_GA_POINT_MINMAX: 0x4230 */
1022 uint32_t line_control
; /* R300_GA_LINE_CNTL: 0x4234 */
1023 uint32_t polygon_offset_enable
; /* R300_SU_POLY_OFFSET_ENABLE: 0x42b4 */
1024 uint32_t cull_mode
; /* R300_SU_CULL_MODE: 0x42b8 */
1025 uint32_t line_stipple_config
; /* R300_GA_LINE_STIPPLE_CONFIG: 0x4328 */
1026 uint32_t line_stipple_value
; /* R300_GA_LINE_STIPPLE_VALUE: 0x4260 */
1027 uint32_t polygon_mode
; /* R300_GA_POLY_MODE: 0x4288 */
1028 uint32_t clip_rule
; /* R300_SC_CLIP_RULE: 0x43D0 */
1029 uint32_t round_mode
; /* R300_GA_ROUND_MODE: 0x428c */
1031 /* Point sprites texture coordinates, 0: lower left, 1: upper right */
1032 float point_texcoord_left
= 0; /* R300_GA_POINT_S0: 0x4200 */
1033 float point_texcoord_bottom
= 0;/* R300_GA_POINT_T0: 0x4204 */
1034 float point_texcoord_right
= 1; /* R300_GA_POINT_S1: 0x4208 */
1035 float point_texcoord_top
= 0; /* R300_GA_POINT_T1: 0x420c */
1036 boolean vclamp
= state
->clamp_vertex_color
;
1039 /* Copy rasterizer state. */
1041 rs
->rs_draw
= *state
;
1043 rs
->rs
.sprite_coord_enable
= state
->point_quad_rasterization
*
1044 state
->sprite_coord_enable
;
1046 /* Override some states for Draw. */
1047 rs
->rs_draw
.sprite_coord_enable
= 0; /* We can do this in HW. */
1049 #ifdef PIPE_ARCH_LITTLE_ENDIAN
1050 vap_control_status
= R300_VC_NO_SWAP
;
1052 vap_control_status
= R300_VC_32BIT_SWAP
;
1055 /* If no TCL engine is present, turn off the HW TCL. */
1056 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1057 vap_control_status
|= R300_VAP_TCL_BYPASS
;
1060 /* Point size width and height. */
1062 pack_float_16_6x(state
->point_size
) |
1063 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
1065 /* Point size clamping. */
1066 if (state
->point_size_per_vertex
) {
1067 /* Per-vertex point size.
1068 * Clamp to [0, max FB size] */
1069 psiz
= pipe
->screen
->get_paramf(pipe
->screen
,
1070 PIPE_CAP_MAX_POINT_WIDTH
);
1072 pack_float_16_6x(psiz
) << R300_GA_POINT_MINMAX_MAX_SHIFT
;
1074 /* We cannot disable the point-size vertex output,
1076 psiz
= state
->point_size
;
1078 (pack_float_16_6x(psiz
) << R300_GA_POINT_MINMAX_MIN_SHIFT
) |
1079 (pack_float_16_6x(psiz
) << R300_GA_POINT_MINMAX_MAX_SHIFT
);
1083 line_control
= pack_float_16_6x(state
->line_width
) |
1084 R300_GA_LINE_CNTL_END_TYPE_COMP
;
1086 /* Enable polygon mode */
1088 if (state
->fill_front
!= PIPE_POLYGON_MODE_FILL
||
1089 state
->fill_back
!= PIPE_POLYGON_MODE_FILL
) {
1090 polygon_mode
= R300_GA_POLY_MODE_DUAL
;
1094 if (state
->front_ccw
)
1095 cull_mode
= R300_FRONT_FACE_CCW
;
1097 cull_mode
= R300_FRONT_FACE_CW
;
1099 /* Polygon offset */
1100 polygon_offset_enable
= 0;
1101 if (util_get_offset(state
, state
->fill_front
)) {
1102 polygon_offset_enable
|= R300_FRONT_ENABLE
;
1104 if (util_get_offset(state
, state
->fill_back
)) {
1105 polygon_offset_enable
|= R300_BACK_ENABLE
;
1108 rs
->polygon_offset_enable
= polygon_offset_enable
!= 0;
1113 r300_translate_polygon_mode_front(state
->fill_front
);
1115 r300_translate_polygon_mode_back(state
->fill_back
);
1118 if (state
->cull_face
& PIPE_FACE_FRONT
) {
1119 cull_mode
|= R300_CULL_FRONT
;
1121 if (state
->cull_face
& PIPE_FACE_BACK
) {
1122 cull_mode
|= R300_CULL_BACK
;
1125 if (state
->line_stipple_enable
) {
1126 line_stipple_config
=
1127 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
1128 (fui((float)state
->line_stipple_factor
) &
1129 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
1130 /* XXX this might need to be scaled up */
1131 line_stipple_value
= state
->line_stipple_pattern
;
1133 line_stipple_config
= 0;
1134 line_stipple_value
= 0;
1137 if (state
->flatshade
) {
1138 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
1140 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
1143 clip_rule
= state
->scissor
? 0xAAAA : 0xFFFF;
1145 /* Point sprites coord mode */
1146 if (rs
->rs
.sprite_coord_enable
) {
1147 switch (state
->sprite_coord_mode
) {
1148 case PIPE_SPRITE_COORD_UPPER_LEFT
:
1149 point_texcoord_top
= 0.0f
;
1150 point_texcoord_bottom
= 1.0f
;
1152 case PIPE_SPRITE_COORD_LOWER_LEFT
:
1153 point_texcoord_top
= 1.0f
;
1154 point_texcoord_bottom
= 0.0f
;
1159 /* Vertex color clamping. FP20 means no clamping. */
1161 R300_GA_ROUND_MODE_GEOMETRY_ROUND_NEAREST
|
1162 (!vclamp
? (R300_GA_ROUND_MODE_RGB_CLAMP_FP20
|
1163 R300_GA_ROUND_MODE_ALPHA_CLAMP_FP20
) : 0);
1165 /* Build the main command buffer. */
1166 BEGIN_CB(rs
->cb_main
, RS_STATE_MAIN_SIZE
);
1167 OUT_CB_REG(R300_VAP_CNTL_STATUS
, vap_control_status
);
1168 OUT_CB_REG(R300_GA_POINT_SIZE
, point_size
);
1169 OUT_CB_REG_SEQ(R300_GA_POINT_MINMAX
, 2);
1170 OUT_CB(point_minmax
);
1171 OUT_CB(line_control
);
1172 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE
, 2);
1173 OUT_CB(polygon_offset_enable
);
1174 rs
->cull_mode_index
= 9;
1176 OUT_CB_REG(R300_GA_LINE_STIPPLE_CONFIG
, line_stipple_config
);
1177 OUT_CB_REG(R300_GA_LINE_STIPPLE_VALUE
, line_stipple_value
);
1178 OUT_CB_REG(R300_GA_POLY_MODE
, polygon_mode
);
1179 OUT_CB_REG(R300_GA_ROUND_MODE
, round_mode
);
1180 OUT_CB_REG(R300_SC_CLIP_RULE
, clip_rule
);
1181 OUT_CB_REG_SEQ(R300_GA_POINT_S0
, 4);
1182 OUT_CB_32F(point_texcoord_left
);
1183 OUT_CB_32F(point_texcoord_bottom
);
1184 OUT_CB_32F(point_texcoord_right
);
1185 OUT_CB_32F(point_texcoord_top
);
1188 /* Build the two command buffers for polygon offset setup. */
1189 if (polygon_offset_enable
) {
1190 float scale
= state
->offset_scale
* 12;
1191 float offset
= state
->offset_units
* 4;
1193 BEGIN_CB(rs
->cb_poly_offset_zb16
, 5);
1194 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE
, 4);
1201 offset
= state
->offset_units
* 2;
1203 BEGIN_CB(rs
->cb_poly_offset_zb24
, 5);
1204 OUT_CB_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE
, 4);
1215 /* Bind rasterizer state. */
1216 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
1218 struct r300_context
* r300
= r300_context(pipe
);
1219 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
1220 int last_sprite_coord_enable
= r300
->sprite_coord_enable
;
1221 boolean last_two_sided_color
= r300
->two_sided_color
;
1222 boolean last_frag_clamp
= r300
->frag_clamp
;
1224 if (r300
->draw
&& rs
) {
1225 draw_set_rasterizer_state(r300
->draw
, &rs
->rs_draw
, state
);
1229 r300
->polygon_offset_enabled
= rs
->polygon_offset_enable
;
1230 r300
->sprite_coord_enable
= rs
->rs
.sprite_coord_enable
;
1231 r300
->two_sided_color
= rs
->rs
.light_twoside
;
1232 r300
->frag_clamp
= rs
->rs
.clamp_fragment_color
;
1234 r300
->polygon_offset_enabled
= FALSE
;
1235 r300
->sprite_coord_enable
= 0;
1236 r300
->two_sided_color
= FALSE
;
1237 r300
->frag_clamp
= FALSE
;
1240 UPDATE_STATE(state
, r300
->rs_state
);
1241 r300
->rs_state
.size
= RS_STATE_MAIN_SIZE
+ (r300
->polygon_offset_enabled
? 5 : 0);
1243 if (last_sprite_coord_enable
!= r300
->sprite_coord_enable
||
1244 last_two_sided_color
!= r300
->two_sided_color
) {
1245 r300_mark_atom_dirty(r300
, &r300
->rs_block_state
);
1248 if (last_frag_clamp
!= r300
->frag_clamp
&&
1249 r300
->fs_status
== FRAGMENT_SHADER_VALID
) {
1250 r300
->fs_status
= FRAGMENT_SHADER_MAYBE_DIRTY
;
1254 /* Free rasterizer state. */
1255 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
1261 r300_create_sampler_state(struct pipe_context
* pipe
,
1262 const struct pipe_sampler_state
* state
)
1264 struct r300_context
* r300
= r300_context(pipe
);
1265 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
1266 boolean is_r500
= r300
->screen
->caps
.is_r500
;
1269 sampler
->state
= *state
;
1271 /* r300 doesn't handle CLAMP and MIRROR_CLAMP correctly when either MAG
1272 * or MIN filter is NEAREST. Since texwrap produces same results
1273 * for CLAMP and CLAMP_TO_EDGE, we use them instead. */
1274 if (sampler
->state
.min_img_filter
== PIPE_TEX_FILTER_NEAREST
||
1275 sampler
->state
.mag_img_filter
== PIPE_TEX_FILTER_NEAREST
) {
1277 if (sampler
->state
.wrap_s
== PIPE_TEX_WRAP_CLAMP
)
1278 sampler
->state
.wrap_s
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1279 else if (sampler
->state
.wrap_s
== PIPE_TEX_WRAP_MIRROR_CLAMP
)
1280 sampler
->state
.wrap_s
= PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
;
1283 if (sampler
->state
.wrap_t
== PIPE_TEX_WRAP_CLAMP
)
1284 sampler
->state
.wrap_t
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1285 else if (sampler
->state
.wrap_t
== PIPE_TEX_WRAP_MIRROR_CLAMP
)
1286 sampler
->state
.wrap_t
= PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
;
1289 if (sampler
->state
.wrap_r
== PIPE_TEX_WRAP_CLAMP
)
1290 sampler
->state
.wrap_r
= PIPE_TEX_WRAP_CLAMP_TO_EDGE
;
1291 else if (sampler
->state
.wrap_r
== PIPE_TEX_WRAP_MIRROR_CLAMP
)
1292 sampler
->state
.wrap_r
= PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE
;
1296 (r300_translate_wrap(sampler
->state
.wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
1297 (r300_translate_wrap(sampler
->state
.wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
1298 (r300_translate_wrap(sampler
->state
.wrap_r
) << R300_TX_WRAP_R_SHIFT
);
1300 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
1301 state
->mag_img_filter
,
1302 state
->min_mip_filter
,
1303 state
->max_anisotropy
> 0);
1305 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
1307 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
1308 /* We must pass these to the merge function to clamp them properly. */
1309 sampler
->min_lod
= (unsigned)MAX2(state
->min_lod
, 0);
1310 sampler
->max_lod
= (unsigned)MAX2(ceilf(state
->max_lod
), 0);
1312 lod_bias
= CLAMP((int)(state
->lod_bias
* 32 + 1), -(1 << 9), (1 << 9) - 1);
1314 sampler
->filter1
|= (lod_bias
<< R300_LOD_BIAS_SHIFT
) & R300_LOD_BIAS_MASK
;
1316 /* This is very high quality anisotropic filtering for R5xx.
1317 * It's good for benchmarking the performance of texturing but
1318 * in practice we don't want to slow down the driver because it's
1319 * a pretty good performance killer. Feel free to play with it. */
1320 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
1321 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
1324 /* R500-specific fixups and optimizations */
1325 if (r300
->screen
->caps
.is_r500
) {
1326 sampler
->filter1
|= R500_BORDER_FIX
;
1329 return (void*)sampler
;
1332 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
1336 struct r300_context
* r300
= r300_context(pipe
);
1337 struct r300_textures_state
* state
=
1338 (struct r300_textures_state
*)r300
->textures_state
.state
;
1339 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
1341 if (count
> tex_units
) {
1345 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
1346 state
->sampler_state_count
= count
;
1348 r300_mark_atom_dirty(r300
, &r300
->textures_state
);
1351 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
1357 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
1362 static uint32_t r300_assign_texture_cache_region(unsigned index
, unsigned num
)
1364 /* This looks like a hack, but I believe it's suppose to work like
1365 * that. To illustrate how this works, let's assume you have 5 textures.
1366 * From docs, 5 and the successive numbers are:
1374 * First 3 textures will get 3/4 of size of the cache, divived evenly
1375 * between them. The last 1/4 of the cache must be divided between
1376 * the last 2 textures, each will therefore get 1/8 of the cache.
1377 * Why not just to use "5 + texture_index" ?
1379 * This simple trick works for all "num" <= 16.
1382 return R300_TX_CACHE(R300_TX_CACHE_WHOLE
);
1384 return R300_TX_CACHE(num
+ index
);
1387 static void r300_set_fragment_sampler_views(struct pipe_context
* pipe
,
1389 struct pipe_sampler_view
** views
)
1391 struct r300_context
* r300
= r300_context(pipe
);
1392 struct r300_textures_state
* state
=
1393 (struct r300_textures_state
*)r300
->textures_state
.state
;
1394 struct r300_resource
*texture
;
1395 unsigned i
, real_num_views
= 0, view_index
= 0;
1396 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
1397 boolean dirty_tex
= FALSE
;
1399 if (count
> tex_units
) {
1403 /* Calculate the real number of views. */
1404 for (i
= 0; i
< count
; i
++) {
1409 for (i
= 0; i
< count
; i
++) {
1410 pipe_sampler_view_reference(
1411 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1418 /* A new sampler view (= texture)... */
1421 /* Set the texrect factor in the fragment shader.
1422 * Needed for RECT and NPOT fallback. */
1423 texture
= r300_resource(views
[i
]->texture
);
1424 if (texture
->tex
.is_npot
) {
1425 r300_mark_atom_dirty(r300
, &r300
->fs_rc_constant_state
);
1428 state
->sampler_views
[i
]->texcache_region
=
1429 r300_assign_texture_cache_region(view_index
, real_num_views
);
1433 for (i
= count
; i
< tex_units
; i
++) {
1434 if (state
->sampler_views
[i
]) {
1435 pipe_sampler_view_reference(
1436 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1441 state
->sampler_view_count
= count
;
1443 r300_mark_atom_dirty(r300
, &r300
->textures_state
);
1446 r300_mark_atom_dirty(r300
, &r300
->texture_cache_inval
);
1450 static struct pipe_sampler_view
*
1451 r300_create_sampler_view(struct pipe_context
*pipe
,
1452 struct pipe_resource
*texture
,
1453 const struct pipe_sampler_view
*templ
)
1455 struct r300_sampler_view
*view
= CALLOC_STRUCT(r300_sampler_view
);
1456 struct r300_resource
*tex
= r300_resource(texture
);
1457 boolean is_r500
= r300_screen(pipe
->screen
)->caps
.is_r500
;
1458 boolean dxtc_swizzle
= r300_screen(pipe
->screen
)->caps
.dxtc_swizzle
;
1461 view
->base
= *templ
;
1462 view
->base
.reference
.count
= 1;
1463 view
->base
.context
= pipe
;
1464 view
->base
.texture
= NULL
;
1465 pipe_resource_reference(&view
->base
.texture
, texture
);
1467 view
->swizzle
[0] = templ
->swizzle_r
;
1468 view
->swizzle
[1] = templ
->swizzle_g
;
1469 view
->swizzle
[2] = templ
->swizzle_b
;
1470 view
->swizzle
[3] = templ
->swizzle_a
;
1472 view
->format
= tex
->tx_format
;
1473 view
->format
.format1
|= r300_translate_texformat(templ
->format
,
1478 view
->format
.format2
|= r500_tx_format_msb_bit(templ
->format
);
1482 return (struct pipe_sampler_view
*)view
;
1486 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1487 struct pipe_sampler_view
*view
)
1489 pipe_resource_reference(&view
->texture
, NULL
);
1493 static void r300_set_scissor_state(struct pipe_context
* pipe
,
1494 const struct pipe_scissor_state
* state
)
1496 struct r300_context
* r300
= r300_context(pipe
);
1498 memcpy(r300
->scissor_state
.state
, state
,
1499 sizeof(struct pipe_scissor_state
));
1501 r300_mark_atom_dirty(r300
, &r300
->scissor_state
);
1504 static void r300_set_viewport_state(struct pipe_context
* pipe
,
1505 const struct pipe_viewport_state
* state
)
1507 struct r300_context
* r300
= r300_context(pipe
);
1508 struct r300_viewport_state
* viewport
=
1509 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1511 r300
->viewport
= *state
;
1514 draw_set_viewport_state(r300
->draw
, state
);
1515 viewport
->vte_control
= R300_VTX_XY_FMT
| R300_VTX_Z_FMT
;
1519 /* Do the transform in HW. */
1520 viewport
->vte_control
= R300_VTX_W0_FMT
;
1522 if (state
->scale
[0] != 1.0f
) {
1523 viewport
->xscale
= state
->scale
[0];
1524 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1526 if (state
->scale
[1] != 1.0f
) {
1527 viewport
->yscale
= state
->scale
[1];
1528 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1530 if (state
->scale
[2] != 1.0f
) {
1531 viewport
->zscale
= state
->scale
[2];
1532 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1534 if (state
->translate
[0] != 0.0f
) {
1535 viewport
->xoffset
= state
->translate
[0];
1536 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1538 if (state
->translate
[1] != 0.0f
) {
1539 viewport
->yoffset
= state
->translate
[1];
1540 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1542 if (state
->translate
[2] != 0.0f
) {
1543 viewport
->zoffset
= state
->translate
[2];
1544 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1547 r300_mark_atom_dirty(r300
, &r300
->viewport_state
);
1548 if (r300
->fs
.state
&& r300_fs(r300
)->shader
&&
1549 r300_fs(r300
)->shader
->inputs
.wpos
!= ATTR_UNUSED
) {
1550 r300_mark_atom_dirty(r300
, &r300
->fs_rc_constant_state
);
1554 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
1556 const struct pipe_vertex_buffer
* buffers
)
1558 struct r300_context
* r300
= r300_context(pipe
);
1560 struct pipe_vertex_buffer dummy_vb
= {0};
1562 /* There must be at least one vertex buffer set, otherwise it locks up. */
1564 dummy_vb
.buffer
= r300
->dummy_vb
;
1565 buffers
= &dummy_vb
;
1569 u_vbuf_mgr_set_vertex_buffers(r300
->vbuf_mgr
, count
, buffers
);
1571 if (r300
->screen
->caps
.has_tcl
) {
1573 for (i
= 0; i
< count
; i
++) {
1574 if (buffers
[i
].buffer
&&
1575 !r300_resource(buffers
[i
].buffer
)->b
.user_ptr
) {
1578 r300
->vertex_arrays_dirty
= TRUE
;
1581 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
1585 static void r300_set_index_buffer(struct pipe_context
* pipe
,
1586 const struct pipe_index_buffer
*ib
)
1588 struct r300_context
* r300
= r300_context(pipe
);
1590 if (ib
&& ib
->buffer
) {
1591 assert(ib
->offset
% ib
->index_size
== 0);
1593 pipe_resource_reference(&r300
->index_buffer
.buffer
, ib
->buffer
);
1594 memcpy(&r300
->index_buffer
, ib
, sizeof(r300
->index_buffer
));
1595 r300
->index_buffer
.offset
/= r300
->index_buffer
.index_size
;
1598 pipe_resource_reference(&r300
->index_buffer
.buffer
, NULL
);
1599 memset(&r300
->index_buffer
, 0, sizeof(r300
->index_buffer
));
1602 if (!r300
->screen
->caps
.has_tcl
) {
1603 draw_set_index_buffer(r300
->draw
, ib
);
1607 /* Initialize the PSC tables. */
1608 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1610 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1611 uint16_t type
, swizzle
;
1612 enum pipe_format format
;
1615 /* Vertex shaders have no semantics on their inputs,
1616 * so PSC should just route stuff based on the vertex elements,
1617 * and not on attrib information. */
1618 for (i
= 0; i
< velems
->count
; i
++) {
1619 format
= velems
->velem
[i
].src_format
;
1621 type
= r300_translate_vertex_data_type(format
);
1622 if (type
== R300_INVALID_FORMAT
) {
1623 fprintf(stderr
, "r300: Bad vertex format %s.\n",
1624 util_format_short_name(format
));
1629 type
|= i
<< R300_DST_VEC_LOC_SHIFT
;
1630 swizzle
= r300_translate_vertex_data_swizzle(format
);
1633 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1634 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1636 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1637 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1641 /* Set the last vector in the PSC. */
1645 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1646 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1648 vstream
->count
= (i
>> 1) + 1;
1651 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1653 const struct pipe_vertex_element
* attribs
)
1655 struct r300_context
*r300
= r300_context(pipe
);
1656 struct r300_vertex_element_state
*velems
;
1658 struct pipe_vertex_element dummy_attrib
= {0};
1660 /* R300 Programmable Stream Control (PSC) doesn't support 0 vertex elements. */
1662 dummy_attrib
.src_format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
1663 attribs
= &dummy_attrib
;
1665 } else if (count
> 16) {
1666 fprintf(stderr
, "r300: More than 16 vertex elements are not supported,"
1667 " requested %i, using 16.\n", count
);
1671 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1675 velems
->count
= count
;
1676 velems
->vmgr_elements
=
1677 u_vbuf_mgr_create_vertex_elements(r300
->vbuf_mgr
, count
, attribs
,
1680 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1682 * The unused components will be replaced by (..., 0, 1). */
1683 r300_vertex_psc(velems
);
1685 for (i
= 0; i
< count
; i
++) {
1686 velems
->format_size
[i
] =
1687 align(util_format_get_blocksize(velems
->velem
[i
].src_format
), 4);
1688 velems
->vertex_size_dwords
+= velems
->format_size
[i
] / 4;
1695 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1698 struct r300_context
*r300
= r300_context(pipe
);
1699 struct r300_vertex_element_state
*velems
= state
;
1701 if (velems
== NULL
) {
1705 r300
->velems
= velems
;
1707 u_vbuf_mgr_bind_vertex_elements(r300
->vbuf_mgr
, state
, velems
->vmgr_elements
);
1710 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1714 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1715 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1716 r300
->vertex_arrays_dirty
= TRUE
;
1719 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1721 struct r300_context
*r300
= r300_context(pipe
);
1722 struct r300_vertex_element_state
*velems
= state
;
1724 u_vbuf_mgr_destroy_vertex_elements(r300
->vbuf_mgr
, velems
->vmgr_elements
);
1728 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1729 const struct pipe_shader_state
* shader
)
1731 struct r300_context
* r300
= r300_context(pipe
);
1732 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1734 /* Copy state directly into shader. */
1735 vs
->state
= *shader
;
1736 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
1738 if (r300
->screen
->caps
.has_tcl
) {
1739 r300_init_vs_outputs(vs
);
1740 r300_translate_vertex_shader(r300
, vs
);
1742 r300_draw_init_vertex_shader(r300
->draw
, vs
);
1748 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1750 struct r300_context
* r300
= r300_context(pipe
);
1751 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1754 r300
->vs_state
.state
= NULL
;
1757 if (vs
== r300
->vs_state
.state
) {
1760 r300
->vs_state
.state
= vs
;
1762 /* The majority of the RS block bits is dependent on the vertex shader. */
1763 r300_mark_atom_dirty(r300
, &r300
->rs_block_state
); /* Will be updated before the emission. */
1765 if (r300
->screen
->caps
.has_tcl
) {
1766 unsigned fc_op_dwords
= r300
->screen
->caps
.is_r500
? 3 : 2;
1767 r300_mark_atom_dirty(r300
, &r300
->vs_state
);
1768 r300
->vs_state
.size
=
1769 vs
->code
.length
+ 9 +
1770 (vs
->code
.num_fc_ops
? vs
->code
.num_fc_ops
* fc_op_dwords
+ 4 : 0);
1772 r300_mark_atom_dirty(r300
, &r300
->vs_constants
);
1773 r300
->vs_constants
.size
=
1775 (vs
->externals_count
? vs
->externals_count
* 4 + 3 : 0) +
1776 (vs
->immediates_count
? vs
->immediates_count
* 4 + 3 : 0);
1778 ((struct r300_constant_buffer
*)r300
->vs_constants
.state
)->remap_table
=
1779 vs
->code
.constants_remap_table
;
1781 r300_mark_atom_dirty(r300
, &r300
->pvs_flush
);
1783 draw_bind_vertex_shader(r300
->draw
,
1784 (struct draw_vertex_shader
*)vs
->draw_vs
);
1788 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1790 struct r300_context
* r300
= r300_context(pipe
);
1791 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1793 if (r300
->screen
->caps
.has_tcl
) {
1794 rc_constants_destroy(&vs
->code
.constants
);
1795 if (vs
->code
.constants_remap_table
)
1796 FREE(vs
->code
.constants_remap_table
);
1798 draw_delete_vertex_shader(r300
->draw
,
1799 (struct draw_vertex_shader
*)vs
->draw_vs
);
1802 FREE((void*)vs
->state
.tokens
);
1806 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1807 uint shader
, uint index
,
1808 struct pipe_resource
*buf
)
1810 struct r300_context
* r300
= r300_context(pipe
);
1811 struct r300_constant_buffer
*cbuf
;
1812 struct r300_resource
*rbuf
= r300_resource(buf
);
1816 case PIPE_SHADER_VERTEX
:
1817 cbuf
= (struct r300_constant_buffer
*)r300
->vs_constants
.state
;
1819 case PIPE_SHADER_FRAGMENT
:
1820 cbuf
= (struct r300_constant_buffer
*)r300
->fs_constants
.state
;
1826 if (buf
== NULL
|| buf
->width0
== 0)
1829 if (rbuf
->b
.user_ptr
)
1830 mapped
= (uint32_t*)rbuf
->b
.user_ptr
;
1831 else if (rbuf
->constant_buffer
)
1832 mapped
= (uint32_t*)rbuf
->constant_buffer
;
1836 if (shader
== PIPE_SHADER_FRAGMENT
||
1837 (shader
== PIPE_SHADER_VERTEX
&& r300
->screen
->caps
.has_tcl
)) {
1841 if (shader
== PIPE_SHADER_VERTEX
) {
1842 if (r300
->screen
->caps
.has_tcl
) {
1843 struct r300_vertex_shader
*vs
=
1844 (struct r300_vertex_shader
*)r300
->vs_state
.state
;
1847 cbuf
->buffer_base
= 0;
1851 cbuf
->buffer_base
= r300
->vs_const_base
;
1852 r300
->vs_const_base
+= vs
->code
.constants
.Count
;
1853 if (r300
->vs_const_base
> R500_MAX_PVS_CONST_VECS
) {
1854 r300
->vs_const_base
= vs
->code
.constants
.Count
;
1855 cbuf
->buffer_base
= 0;
1856 r300_mark_atom_dirty(r300
, &r300
->pvs_flush
);
1858 r300_mark_atom_dirty(r300
, &r300
->vs_constants
);
1859 } else if (r300
->draw
) {
1860 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
1861 0, mapped
, buf
->width0
);
1863 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
1864 r300_mark_atom_dirty(r300
, &r300
->fs_constants
);
1868 static void r300_texture_barrier(struct pipe_context
*pipe
)
1870 struct r300_context
*r300
= r300_context(pipe
);
1872 r300_mark_atom_dirty(r300
, &r300
->gpu_flush
);
1873 r300_mark_atom_dirty(r300
, &r300
->texture_cache_inval
);
1876 void r300_init_state_functions(struct r300_context
* r300
)
1878 r300
->context
.create_blend_state
= r300_create_blend_state
;
1879 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1880 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1882 r300
->context
.set_blend_color
= r300_set_blend_color
;
1884 r300
->context
.set_clip_state
= r300_set_clip_state
;
1885 r300
->context
.set_sample_mask
= r300_set_sample_mask
;
1887 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1889 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1890 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1891 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1893 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
1895 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1897 r300
->context
.create_fs_state
= r300_create_fs_state
;
1898 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1899 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1901 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1903 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1904 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1905 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1907 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1908 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1909 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1910 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1912 r300
->context
.set_fragment_sampler_views
= r300_set_fragment_sampler_views
;
1913 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
1914 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
1916 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1918 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1920 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1921 r300
->context
.set_index_buffer
= r300_set_index_buffer
;
1922 r300
->context
.redefine_user_buffer
= u_default_redefine_user_buffer
;
1924 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
1925 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
1926 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
1928 r300
->context
.create_vs_state
= r300_create_vs_state
;
1929 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1930 r300
->context
.delete_vs_state
= r300_delete_vs_state
;
1932 r300
->context
.texture_barrier
= r300_texture_barrier
;