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_math.h"
27 #include "util/u_memory.h"
28 #include "util/u_pack_color.h"
30 #include "tgsi/tgsi_parse.h"
32 #include "pipe/p_config.h"
34 #include "r300_context.h"
36 #include "r300_screen.h"
37 #include "r300_screen_buffer.h"
38 #include "r300_state_inlines.h"
41 #include "r300_winsys.h"
43 /* r300_state: Functions used to intialize state context by translating
44 * Gallium state objects into semi-native r300 state objects. */
46 #define UPDATE_STATE(cso, atom) \
47 if (cso != atom.state) { \
52 static boolean
blend_discard_if_src_alpha_0(unsigned srcRGB
, unsigned srcA
,
53 unsigned dstRGB
, unsigned dstA
)
55 /* If the blend equation is ADD or REVERSE_SUBTRACT,
56 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
57 * will not be changed.
58 * Notice that the dst factors are the src factors inverted. */
59 return (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
60 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
61 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
62 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
63 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
64 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
65 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
66 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
67 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
68 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
69 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
70 dstA
== PIPE_BLENDFACTOR_ONE
);
73 static boolean
blend_discard_if_src_alpha_1(unsigned srcRGB
, unsigned srcA
,
74 unsigned dstRGB
, unsigned dstA
)
76 /* If the blend equation is ADD or REVERSE_SUBTRACT,
77 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
78 * will not be changed.
79 * Notice that the dst factors are the src factors inverted. */
80 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
81 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
82 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
83 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
84 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
85 (dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
86 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
87 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
88 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
89 dstA
== PIPE_BLENDFACTOR_ONE
);
92 static boolean
blend_discard_if_src_color_0(unsigned srcRGB
, unsigned srcA
,
93 unsigned dstRGB
, unsigned dstA
)
95 /* If the blend equation is ADD or REVERSE_SUBTRACT,
96 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
97 * will not be changed.
98 * Notice that the dst factors are the src factors inverted. */
99 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
100 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
101 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
102 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
103 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
104 (dstA
== PIPE_BLENDFACTOR_ONE
);
107 static boolean
blend_discard_if_src_color_1(unsigned srcRGB
, unsigned srcA
,
108 unsigned dstRGB
, unsigned dstA
)
110 /* If the blend equation is ADD or REVERSE_SUBTRACT,
111 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
112 * will not be changed.
113 * Notice that the dst factors are the src factors inverted. */
114 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
115 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
116 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
117 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
118 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
119 (dstA
== PIPE_BLENDFACTOR_ONE
);
122 static boolean
blend_discard_if_src_alpha_color_0(unsigned srcRGB
, unsigned srcA
,
123 unsigned dstRGB
, unsigned dstA
)
125 /* If the blend equation is ADD or REVERSE_SUBTRACT,
126 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
127 * the colorbuffer will not be changed.
128 * Notice that the dst factors are the src factors inverted. */
129 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
130 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
131 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
132 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
133 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
134 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
135 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
136 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
137 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
138 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
139 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
140 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
141 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
142 dstA
== PIPE_BLENDFACTOR_ONE
);
145 static boolean
blend_discard_if_src_alpha_color_1(unsigned srcRGB
, unsigned srcA
,
146 unsigned dstRGB
, unsigned dstA
)
148 /* If the blend equation is ADD or REVERSE_SUBTRACT,
149 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
150 * the colorbuffer will not be changed.
151 * Notice that the dst factors are the src factors inverted. */
152 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
153 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
154 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
155 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
156 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
157 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
158 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
159 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
160 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
161 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
162 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
163 dstA
== PIPE_BLENDFACTOR_ONE
);
166 static unsigned bgra_cmask(unsigned mask
)
168 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
170 return ((mask
& PIPE_MASK_R
) << 2) |
171 ((mask
& PIPE_MASK_B
) >> 2) |
172 (mask
& (PIPE_MASK_G
| PIPE_MASK_A
));
175 /* Create a new blend state based on the CSO blend state.
177 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
178 static void* r300_create_blend_state(struct pipe_context
* pipe
,
179 const struct pipe_blend_state
* state
)
181 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
182 struct r300_blend_state
* blend
= CALLOC_STRUCT(r300_blend_state
);
184 if (state
->rt
[0].blend_enable
)
186 unsigned eqRGB
= state
->rt
[0].rgb_func
;
187 unsigned srcRGB
= state
->rt
[0].rgb_src_factor
;
188 unsigned dstRGB
= state
->rt
[0].rgb_dst_factor
;
190 unsigned eqA
= state
->rt
[0].alpha_func
;
191 unsigned srcA
= state
->rt
[0].alpha_src_factor
;
192 unsigned dstA
= state
->rt
[0].alpha_dst_factor
;
194 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
195 * this is just the crappy D3D naming */
196 blend
->blend_control
= R300_ALPHA_BLEND_ENABLE
|
197 r300_translate_blend_function(eqRGB
) |
198 ( r300_translate_blend_factor(srcRGB
) << R300_SRC_BLEND_SHIFT
) |
199 ( r300_translate_blend_factor(dstRGB
) << R300_DST_BLEND_SHIFT
);
201 /* Optimization: some operations do not require the destination color.
203 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
204 * otherwise blending gives incorrect results. It seems to be
206 if (eqRGB
== PIPE_BLEND_MIN
|| eqA
== PIPE_BLEND_MIN
||
207 eqRGB
== PIPE_BLEND_MAX
|| eqA
== PIPE_BLEND_MAX
||
208 dstRGB
!= PIPE_BLENDFACTOR_ZERO
||
209 dstA
!= PIPE_BLENDFACTOR_ZERO
||
210 srcRGB
== PIPE_BLENDFACTOR_DST_COLOR
||
211 srcRGB
== PIPE_BLENDFACTOR_DST_ALPHA
||
212 srcRGB
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
213 srcRGB
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
214 srcA
== PIPE_BLENDFACTOR_DST_COLOR
||
215 srcA
== PIPE_BLENDFACTOR_DST_ALPHA
||
216 srcA
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
217 srcA
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
218 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
) {
219 /* Enable reading from the colorbuffer. */
220 blend
->blend_control
|= R300_READ_ENABLE
;
222 if (r300screen
->caps
.is_r500
) {
223 /* Optimization: Depending on incoming pixels, we can
224 * conditionally disable the reading in hardware... */
225 if (eqRGB
!= PIPE_BLEND_MIN
&& eqA
!= PIPE_BLEND_MIN
&&
226 eqRGB
!= PIPE_BLEND_MAX
&& eqA
!= PIPE_BLEND_MAX
) {
227 /* Disable reading if SRC_ALPHA == 0. */
228 if ((dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
229 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
230 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
231 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
232 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
233 blend
->blend_control
|= R500_SRC_ALPHA_0_NO_READ
;
236 /* Disable reading if SRC_ALPHA == 1. */
237 if ((dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
238 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
239 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
240 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
241 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
242 blend
->blend_control
|= R500_SRC_ALPHA_1_NO_READ
;
248 /* Optimization: discard pixels which don't change the colorbuffer.
250 * The code below is non-trivial and some math is involved.
252 * Discarding pixels must be disabled when FP16 AA is enabled.
253 * This is a hardware bug. Also, this implementation wouldn't work
254 * with FP blending enabled and equation clamping disabled.
256 * Equations other than ADD are rarely used and therefore won't be
258 if ((eqRGB
== PIPE_BLEND_ADD
|| eqRGB
== PIPE_BLEND_REVERSE_SUBTRACT
) &&
259 (eqA
== PIPE_BLEND_ADD
|| eqA
== PIPE_BLEND_REVERSE_SUBTRACT
)) {
261 * REVERSE_SUBTRACT: Y-X
264 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
265 * then CB will not be changed.
267 * Given the srcFactor and dstFactor variables, we can derive
268 * what src and dst should be equal to and discard appropriate
271 if (blend_discard_if_src_alpha_0(srcRGB
, srcA
, dstRGB
, dstA
)) {
272 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0
;
273 } else if (blend_discard_if_src_alpha_1(srcRGB
, srcA
,
275 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1
;
276 } else if (blend_discard_if_src_color_0(srcRGB
, srcA
,
278 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0
;
279 } else if (blend_discard_if_src_color_1(srcRGB
, srcA
,
281 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1
;
282 } else if (blend_discard_if_src_alpha_color_0(srcRGB
, srcA
,
284 blend
->blend_control
|=
285 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0
;
286 } else if (blend_discard_if_src_alpha_color_1(srcRGB
, srcA
,
288 blend
->blend_control
|=
289 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1
;
294 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
295 blend
->blend_control
|= R300_SEPARATE_ALPHA_ENABLE
;
296 blend
->alpha_blend_control
=
297 r300_translate_blend_function(eqA
) |
298 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
299 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
303 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
304 if (state
->logicop_enable
) {
305 blend
->rop
= R300_RB3D_ROPCNTL_ROP_ENABLE
|
306 (state
->logicop_func
) << R300_RB3D_ROPCNTL_ROP_SHIFT
;
309 /* Color channel masks for all MRTs. */
310 blend
->color_channel_mask
= bgra_cmask(state
->rt
[0].colormask
);
311 if (r300screen
->caps
.is_r500
&& state
->independent_blend_enable
) {
312 if (state
->rt
[1].blend_enable
) {
313 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[1].colormask
) << 4;
315 if (state
->rt
[2].blend_enable
) {
316 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[2].colormask
) << 8;
318 if (state
->rt
[3].blend_enable
) {
319 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[3].colormask
) << 12;
323 /* Neither fglrx nor classic r300 ever set this, regardless of dithering
324 * state. Since it's an optional implementation detail, we can leave it
325 * out and never dither.
327 * This could be revisited if we ever get quality or conformance hints.
330 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
331 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
338 /* Bind blend state. */
339 static void r300_bind_blend_state(struct pipe_context
* pipe
,
342 struct r300_context
* r300
= r300_context(pipe
);
344 UPDATE_STATE(state
, r300
->blend_state
);
347 /* Free blend state. */
348 static void r300_delete_blend_state(struct pipe_context
* pipe
,
354 /* Convert float to 10bit integer */
355 static unsigned float_to_fixed10(float f
)
357 return CLAMP((unsigned)(f
* 1023.9f
), 0, 1023);
361 * Setup both R300 and R500 registers, figure out later which one to write. */
362 static void r300_set_blend_color(struct pipe_context
* pipe
,
363 const struct pipe_blend_color
* color
)
365 struct r300_context
* r300
= r300_context(pipe
);
366 struct r300_blend_color_state
* state
=
367 (struct r300_blend_color_state
*)r300
->blend_color_state
.state
;
370 util_pack_color(color
->color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
371 state
->blend_color
= uc
.ui
;
373 /* XXX if FP16 blending is enabled, we should use the FP16 format */
374 state
->blend_color_red_alpha
=
375 float_to_fixed10(color
->color
[0]) |
376 (float_to_fixed10(color
->color
[3]) << 16);
377 state
->blend_color_green_blue
=
378 float_to_fixed10(color
->color
[2]) |
379 (float_to_fixed10(color
->color
[1]) << 16);
381 r300
->blend_color_state
.size
= r300
->screen
->caps
.is_r500
? 3 : 2;
382 r300
->blend_color_state
.dirty
= TRUE
;
385 static void r300_set_clip_state(struct pipe_context
* pipe
,
386 const struct pipe_clip_state
* state
)
388 struct r300_context
* r300
= r300_context(pipe
);
392 if (r300
->screen
->caps
.has_tcl
) {
393 memcpy(r300
->clip_state
.state
, state
, sizeof(struct pipe_clip_state
));
394 r300
->clip_state
.size
= 29;
396 draw_flush(r300
->draw
);
397 draw_set_clip_state(r300
->draw
, state
);
398 r300
->clip_state
.size
= 2;
401 r300
->clip_state
.dirty
= TRUE
;
404 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
406 * This contains the depth buffer, stencil buffer, alpha test, and such.
407 * On the Radeon, depth and stencil buffer setup are intertwined, which is
408 * the reason for some of the strange-looking assignments across registers. */
410 r300_create_dsa_state(struct pipe_context
* pipe
,
411 const struct pipe_depth_stencil_alpha_state
* state
)
413 struct r300_capabilities
*caps
= &r300_screen(pipe
->screen
)->caps
;
414 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
416 /* Depth test setup. */
417 if (state
->depth
.enabled
) {
418 dsa
->z_buffer_control
|= R300_Z_ENABLE
;
420 if (state
->depth
.writemask
) {
421 dsa
->z_buffer_control
|= R300_Z_WRITE_ENABLE
;
424 dsa
->z_stencil_control
|=
425 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
429 /* Stencil buffer setup. */
430 if (state
->stencil
[0].enabled
) {
431 dsa
->z_buffer_control
|= R300_STENCIL_ENABLE
;
432 dsa
->z_stencil_control
|=
433 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
434 R300_S_FRONT_FUNC_SHIFT
) |
435 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
436 R300_S_FRONT_SFAIL_OP_SHIFT
) |
437 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
438 R300_S_FRONT_ZPASS_OP_SHIFT
) |
439 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
440 R300_S_FRONT_ZFAIL_OP_SHIFT
);
442 dsa
->stencil_ref_mask
=
443 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
444 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
446 if (state
->stencil
[1].enabled
) {
447 dsa
->two_sided
= TRUE
;
449 dsa
->z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
450 dsa
->z_stencil_control
|=
451 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
452 R300_S_BACK_FUNC_SHIFT
) |
453 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
454 R300_S_BACK_SFAIL_OP_SHIFT
) |
455 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
456 R300_S_BACK_ZPASS_OP_SHIFT
) |
457 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
458 R300_S_BACK_ZFAIL_OP_SHIFT
);
460 dsa
->stencil_ref_bf
=
461 (state
->stencil
[1].valuemask
<< R300_STENCILMASK_SHIFT
) |
462 (state
->stencil
[1].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
465 dsa
->z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
467 dsa
->stencil_ref_bf_fallback
=
468 (state
->stencil
[0].valuemask
!= state
->stencil
[1].valuemask
||
469 state
->stencil
[0].writemask
!= state
->stencil
[1].writemask
);
474 /* Alpha test setup. */
475 if (state
->alpha
.enabled
) {
476 dsa
->alpha_function
=
477 r300_translate_alpha_function(state
->alpha
.func
) |
478 R300_FG_ALPHA_FUNC_ENABLE
;
480 /* We could use 10bit alpha ref but who needs that? */
481 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
484 dsa
->alpha_function
|= R500_FG_ALPHA_FUNC_8BIT
;
490 static void r300_update_stencil_ref_fallback_status(struct r300_context
*r300
)
492 struct r300_dsa_state
*dsa
= (struct r300_dsa_state
*)r300
->dsa_state
.state
;
494 if (r300
->screen
->caps
.is_r500
) {
498 r300
->stencil_ref_bf_fallback
=
499 dsa
->stencil_ref_bf_fallback
||
501 r300
->stencil_ref
.ref_value
[0] != r300
->stencil_ref
.ref_value
[1]);
504 /* Bind DSA state. */
505 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
508 struct r300_context
* r300
= r300_context(pipe
);
514 UPDATE_STATE(state
, r300
->dsa_state
);
516 r300_update_stencil_ref_fallback_status(r300
);
519 /* Free DSA state. */
520 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
526 static void r300_set_stencil_ref(struct pipe_context
* pipe
,
527 const struct pipe_stencil_ref
* sr
)
529 struct r300_context
* r300
= r300_context(pipe
);
531 r300
->stencil_ref
= *sr
;
532 r300
->dsa_state
.dirty
= TRUE
;
534 r300_update_stencil_ref_fallback_status(r300
);
537 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
538 static void r300_fb_update_tiling_flags(struct r300_context
*r300
,
539 const struct pipe_framebuffer_state
*old_state
,
540 const struct pipe_framebuffer_state
*new_state
)
542 struct r300_texture
*tex
;
543 unsigned i
, j
, level
;
545 /* Reset tiling flags for old surfaces to default values. */
546 for (i
= 0; i
< old_state
->nr_cbufs
; i
++) {
547 for (j
= 0; j
< new_state
->nr_cbufs
; j
++) {
548 if (old_state
->cbufs
[i
]->texture
== new_state
->cbufs
[j
]->texture
) {
552 /* If not binding the surface again... */
553 if (j
!= new_state
->nr_cbufs
) {
557 tex
= (struct r300_texture
*)old_state
->cbufs
[i
]->texture
;
560 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
566 if (old_state
->zsbuf
&&
567 (!new_state
->zsbuf
||
568 old_state
->zsbuf
->texture
!= new_state
->zsbuf
->texture
)) {
569 tex
= (struct r300_texture
*)old_state
->zsbuf
->texture
;
572 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
579 /* Set tiling flags for new surfaces. */
580 for (i
= 0; i
< new_state
->nr_cbufs
; i
++) {
581 tex
= (struct r300_texture
*)new_state
->cbufs
[i
]->texture
;
582 level
= new_state
->cbufs
[i
]->level
;
584 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
587 tex
->mip_macrotile
[level
]);
589 if (new_state
->zsbuf
) {
590 tex
= (struct r300_texture
*)new_state
->zsbuf
->texture
;
591 level
= new_state
->zsbuf
->level
;
593 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
596 tex
->mip_macrotile
[level
]);
601 r300_set_framebuffer_state(struct pipe_context
* pipe
,
602 const struct pipe_framebuffer_state
* state
)
604 struct r300_context
* r300
= r300_context(pipe
);
605 struct pipe_framebuffer_state
*old_state
= r300
->fb_state
.state
;
606 unsigned max_width
, max_height
;
607 uint32_t zbuffer_bpp
= 0;
609 if (state
->nr_cbufs
> 4) {
610 fprintf(stderr
, "r300: Implementation error: Too many MRTs in %s, "
611 "refusing to bind framebuffer state!\n", __FUNCTION__
);
615 if (r300
->screen
->caps
.is_r500
) {
616 max_width
= max_height
= 4096;
617 } else if (r300
->screen
->caps
.is_r400
) {
618 max_width
= max_height
= 4021;
620 max_width
= max_height
= 2560;
623 if (state
->width
> max_width
|| state
->height
> max_height
) {
624 fprintf(stderr
, "r300: Implementation error: Render targets are too "
625 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
630 draw_flush(r300
->draw
);
633 r300
->fb_state
.dirty
= TRUE
;
635 /* If nr_cbufs is changed from zero to non-zero or vice versa... */
636 if (!!old_state
->nr_cbufs
!= !!state
->nr_cbufs
) {
637 r300
->blend_state
.dirty
= TRUE
;
639 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
640 if (!!old_state
->zsbuf
!= !!state
->zsbuf
) {
641 r300
->dsa_state
.dirty
= TRUE
;
643 if (!r300
->scissor_enabled
) {
644 r300
->scissor_state
.dirty
= TRUE
;
647 r300_fb_update_tiling_flags(r300
, r300
->fb_state
.state
, state
);
649 memcpy(r300
->fb_state
.state
, state
, sizeof(struct pipe_framebuffer_state
));
651 r300
->fb_state
.size
= (10 * state
->nr_cbufs
) + (2 * (4 - state
->nr_cbufs
)) +
652 (state
->zsbuf
? 10 : 0) + 8;
654 /* Polygon offset depends on the zbuffer bit depth. */
655 if (state
->zsbuf
&& r300
->polygon_offset_enabled
) {
656 switch (util_format_get_blocksize(state
->zsbuf
->texture
->format
)) {
665 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
666 r300
->zbuffer_bpp
= zbuffer_bpp
;
667 r300
->rs_state
.dirty
= TRUE
;
672 /* Create fragment shader state. */
673 static void* r300_create_fs_state(struct pipe_context
* pipe
,
674 const struct pipe_shader_state
* shader
)
676 struct r300_fragment_shader
* fs
= NULL
;
678 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
680 /* Copy state directly into shader. */
682 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
684 tgsi_scan_shader(shader
->tokens
, &fs
->info
);
685 r300_shader_read_fs_inputs(&fs
->info
, &fs
->inputs
);
690 /* Bind fragment shader state. */
691 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
693 struct r300_context
* r300
= r300_context(pipe
);
694 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
702 r300_pick_fragment_shader(r300
);
704 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
706 if (r300
->vs_state
.state
&& r300_vertex_shader_setup_wpos(r300
)) {
707 r300
->vap_output_state
.dirty
= TRUE
;
710 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER
| R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
713 /* Delete fragment shader state. */
714 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
716 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
717 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
722 rc_constants_destroy(&tmp
->code
.constants
);
725 FREE((void*)fs
->state
.tokens
);
729 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
730 const struct pipe_poly_stipple
* state
)
732 /* XXX no idea how to set this up, but not terribly important */
735 /* Create a new rasterizer state based on the CSO rasterizer state.
737 * This is a very large chunk of state, and covers most of the graphics
738 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
740 * In a not entirely unironic sidenote, this state has nearly nothing to do
741 * with the actual block on the Radeon called the rasterizer (RS). */
742 static void* r300_create_rs_state(struct pipe_context
* pipe
,
743 const struct pipe_rasterizer_state
* state
)
745 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
747 /* Copy rasterizer state for Draw. */
750 #ifdef PIPE_ARCH_LITTLE_ENDIAN
751 rs
->vap_control_status
= R300_VC_NO_SWAP
;
753 rs
->vap_control_status
= R300_VC_32BIT_SWAP
;
756 /* If no TCL engine is present, turn off the HW TCL. */
757 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
758 rs
->vap_control_status
|= R300_VAP_TCL_BYPASS
;
761 rs
->point_size
= pack_float_16_6x(state
->point_size
) |
762 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
764 rs
->line_control
= pack_float_16_6x(state
->line_width
) |
765 R300_GA_LINE_CNTL_END_TYPE_COMP
;
767 /* Enable polygon mode */
768 if (state
->fill_cw
!= PIPE_POLYGON_MODE_FILL
||
769 state
->fill_ccw
!= PIPE_POLYGON_MODE_FILL
) {
770 rs
->polygon_mode
= R300_GA_POLY_MODE_DUAL
;
773 /* Radeons don't think in "CW/CCW", they think in "front/back". */
774 if (state
->front_winding
== PIPE_WINDING_CW
) {
775 rs
->cull_mode
= R300_FRONT_FACE_CW
;
778 if (state
->offset_cw
) {
779 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
781 if (state
->offset_ccw
) {
782 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
786 if (rs
->polygon_mode
) {
788 r300_translate_polygon_mode_front(state
->fill_cw
);
790 r300_translate_polygon_mode_back(state
->fill_ccw
);
793 rs
->cull_mode
= R300_FRONT_FACE_CCW
;
796 if (state
->offset_ccw
) {
797 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
799 if (state
->offset_cw
) {
800 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
804 if (rs
->polygon_mode
) {
806 r300_translate_polygon_mode_front(state
->fill_ccw
);
808 r300_translate_polygon_mode_back(state
->fill_cw
);
811 if (state
->front_winding
& state
->cull_mode
) {
812 rs
->cull_mode
|= R300_CULL_FRONT
;
814 if (~(state
->front_winding
) & state
->cull_mode
) {
815 rs
->cull_mode
|= R300_CULL_BACK
;
818 if (rs
->polygon_offset_enable
) {
819 rs
->depth_offset
= state
->offset_units
;
820 rs
->depth_scale
= state
->offset_scale
;
823 if (state
->line_stipple_enable
) {
824 rs
->line_stipple_config
=
825 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
826 (fui((float)state
->line_stipple_factor
) &
827 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
828 /* XXX this might need to be scaled up */
829 rs
->line_stipple_value
= state
->line_stipple_pattern
;
832 if (state
->flatshade
) {
833 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
835 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
841 /* Bind rasterizer state. */
842 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
844 struct r300_context
* r300
= r300_context(pipe
);
845 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
846 boolean scissor_was_enabled
= r300
->scissor_enabled
;
849 draw_flush(r300
->draw
);
850 draw_set_rasterizer_state(r300
->draw
, &rs
->rs
);
854 r300
->polygon_offset_enabled
= rs
->rs
.offset_cw
|| rs
->rs
.offset_ccw
;
855 r300
->scissor_enabled
= rs
->rs
.scissor
;
857 r300
->polygon_offset_enabled
= FALSE
;
858 r300
->scissor_enabled
= FALSE
;
861 UPDATE_STATE(state
, r300
->rs_state
);
862 r300
->rs_state
.size
= 17 + (r300
->polygon_offset_enabled
? 5 : 0);
864 if (scissor_was_enabled
!= r300
->scissor_enabled
) {
865 r300
->scissor_state
.dirty
= TRUE
;
869 /* Free rasterizer state. */
870 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
876 r300_create_sampler_state(struct pipe_context
* pipe
,
877 const struct pipe_sampler_state
* state
)
879 struct r300_context
* r300
= r300_context(pipe
);
880 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
881 boolean is_r500
= r300
->screen
->caps
.is_r500
;
885 sampler
->state
= *state
;
888 (r300_translate_wrap(state
->wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
889 (r300_translate_wrap(state
->wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
890 (r300_translate_wrap(state
->wrap_r
) << R300_TX_WRAP_R_SHIFT
);
892 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
893 state
->mag_img_filter
,
894 state
->min_mip_filter
,
895 state
->max_anisotropy
> 0);
897 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
899 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
900 /* We must pass these to the merge function to clamp them properly. */
901 sampler
->min_lod
= MAX2((unsigned)state
->min_lod
, 0);
902 sampler
->max_lod
= MAX2((unsigned)ceilf(state
->max_lod
), 0);
904 lod_bias
= CLAMP((int)(state
->lod_bias
* 32), -(1 << 9), (1 << 9) - 1);
906 sampler
->filter1
|= lod_bias
<< R300_LOD_BIAS_SHIFT
;
908 /* This is very high quality anisotropic filtering for R5xx.
909 * It's good for benchmarking the performance of texturing but
910 * in practice we don't want to slow down the driver because it's
911 * a pretty good performance killer. Feel free to play with it. */
912 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
913 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
916 util_pack_color(state
->border_color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
917 sampler
->border_color
= uc
.ui
;
919 /* R500-specific fixups and optimizations */
920 if (r300
->screen
->caps
.is_r500
) {
921 sampler
->filter1
|= R500_BORDER_FIX
;
924 return (void*)sampler
;
927 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
931 struct r300_context
* r300
= r300_context(pipe
);
932 struct r300_textures_state
* state
=
933 (struct r300_textures_state
*)r300
->textures_state
.state
;
934 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
936 if (count
> tex_units
) {
940 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
941 state
->sampler_count
= count
;
943 r300
->textures_state
.dirty
= TRUE
;
945 /* Pick a fragment shader based on the texture compare state. */
946 if (r300
->fs
&& count
) {
947 if (r300_pick_fragment_shader(r300
)) {
948 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER
|
949 R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
954 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
960 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
965 static void r300_set_fragment_sampler_views(struct pipe_context
* pipe
,
967 struct pipe_sampler_view
** views
)
969 struct r300_context
* r300
= r300_context(pipe
);
970 struct r300_textures_state
* state
=
971 (struct r300_textures_state
*)r300
->textures_state
.state
;
972 struct r300_texture
*texture
;
974 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
975 boolean is_r500
= r300
->screen
->caps
.is_r500
;
976 boolean dirty_tex
= FALSE
;
978 if (count
> tex_units
) {
982 for (i
= 0; i
< count
; i
++) {
983 if (state
->fragment_sampler_views
[i
] != views
[i
]) {
984 pipe_sampler_view_reference(&state
->fragment_sampler_views
[i
],
991 /* A new sampler view (= texture)... */
994 /* R300-specific - set the texrect factor in the fragment shader */
995 texture
= (struct r300_texture
*)views
[i
]->texture
;
996 if (!is_r500
&& texture
->uses_pitch
) {
997 /* XXX It would be nice to re-emit just 1 constant,
998 * XXX not all of them */
999 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1004 for (i
= count
; i
< tex_units
; i
++) {
1005 if (state
->fragment_sampler_views
[i
]) {
1006 pipe_sampler_view_reference(&state
->fragment_sampler_views
[i
],
1011 state
->texture_count
= count
;
1013 r300
->textures_state
.dirty
= TRUE
;
1016 r300
->texture_cache_inval
.dirty
= TRUE
;
1020 static struct pipe_sampler_view
*
1021 r300_create_sampler_view(struct pipe_context
*pipe
,
1022 struct pipe_texture
*texture
,
1023 const struct pipe_sampler_view
*templ
)
1025 struct pipe_sampler_view
*view
= CALLOC_STRUCT(pipe_sampler_view
);
1029 view
->reference
.count
= 1;
1030 view
->texture
= NULL
;
1031 pipe_texture_reference(&view
->texture
, texture
);
1032 view
->context
= pipe
;
1039 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1040 struct pipe_sampler_view
*view
)
1042 pipe_texture_reference(&view
->texture
, NULL
);
1046 static void r300_set_scissor_state(struct pipe_context
* pipe
,
1047 const struct pipe_scissor_state
* state
)
1049 struct r300_context
* r300
= r300_context(pipe
);
1051 memcpy(r300
->scissor_state
.state
, state
,
1052 sizeof(struct pipe_scissor_state
));
1054 if (r300
->scissor_enabled
) {
1055 r300
->scissor_state
.dirty
= TRUE
;
1059 static void r300_set_viewport_state(struct pipe_context
* pipe
,
1060 const struct pipe_viewport_state
* state
)
1062 struct r300_context
* r300
= r300_context(pipe
);
1063 struct r300_viewport_state
* viewport
=
1064 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1066 r300
->viewport
= *state
;
1068 /* Do the transform in HW. */
1069 viewport
->vte_control
= R300_VTX_W0_FMT
;
1071 if (state
->scale
[0] != 1.0f
) {
1072 viewport
->xscale
= state
->scale
[0];
1073 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1075 if (state
->scale
[1] != 1.0f
) {
1076 viewport
->yscale
= state
->scale
[1];
1077 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1079 if (state
->scale
[2] != 1.0f
) {
1080 viewport
->zscale
= state
->scale
[2];
1081 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1083 if (state
->translate
[0] != 0.0f
) {
1084 viewport
->xoffset
= state
->translate
[0];
1085 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1087 if (state
->translate
[1] != 0.0f
) {
1088 viewport
->yoffset
= state
->translate
[1];
1089 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1091 if (state
->translate
[2] != 0.0f
) {
1092 viewport
->zoffset
= state
->translate
[2];
1093 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1096 r300
->viewport_state
.dirty
= TRUE
;
1097 if (r300
->fs
&& r300
->fs
->inputs
.wpos
!= ATTR_UNUSED
) {
1098 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1102 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
1104 const struct pipe_vertex_buffer
* buffers
)
1106 struct r300_context
* r300
= r300_context(pipe
);
1107 struct pipe_vertex_buffer
*vbo
;
1108 unsigned i
, max_index
= (1 << 24) - 1;
1109 boolean any_user_buffer
= FALSE
;
1111 if (count
== r300
->vertex_buffer_count
&&
1112 memcmp(r300
->vertex_buffer
, buffers
,
1113 sizeof(struct pipe_vertex_buffer
) * count
) == 0) {
1117 /* Check if the stride is aligned to the size of DWORD. */
1118 for (i
= 0; i
< count
; i
++) {
1119 if (buffers
[i
].buffer
) {
1120 if (buffers
[i
].stride
% 4 != 0) {
1121 // XXX Shouldn't we align the buffer?
1122 fprintf(stderr
, "r300_set_vertex_buffers: "
1123 "Unaligned buffer stride %i isn't supported.\n",
1131 for (i
= 0; i
< count
; i
++) {
1132 /* Why, yes, I AM casting away constness. How did you know? */
1133 vbo
= (struct pipe_vertex_buffer
*)&buffers
[i
];
1135 /* Reference our buffer. */
1136 pipe_buffer_reference(&r300
->vertex_buffer
[i
].buffer
, vbo
->buffer
);
1138 /* Skip NULL buffers */
1139 if (!buffers
[i
].buffer
) {
1143 if (r300_buffer_is_user_buffer(vbo
->buffer
)) {
1144 any_user_buffer
= TRUE
;
1147 if (vbo
->max_index
== ~0) {
1148 /* Bogus value from broken state tracker; hax it. */
1150 (vbo
->buffer
->size
- vbo
->buffer_offset
) / vbo
->stride
;
1153 max_index
= MIN2(vbo
->max_index
, max_index
);
1156 for (; i
< r300
->vertex_buffer_count
; i
++) {
1157 /* Dereference any old buffers. */
1158 pipe_buffer_reference(&r300
->vertex_buffer
[i
].buffer
, NULL
);
1161 memcpy(r300
->vertex_buffer
, buffers
,
1162 sizeof(struct pipe_vertex_buffer
) * count
);
1164 r300
->vertex_buffer_count
= count
;
1165 r300
->vertex_buffer_max_index
= max_index
;
1166 r300
->any_user_vbs
= any_user_buffer
;
1169 draw_flush(r300
->draw
);
1170 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
1174 /* Update the PSC tables. */
1175 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1177 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1178 uint16_t type
, swizzle
;
1179 enum pipe_format format
;
1182 assert(velems
->count
<= 16);
1184 /* Vertex shaders have no semantics on their inputs,
1185 * so PSC should just route stuff based on the vertex elements,
1186 * and not on attrib information. */
1187 for (i
= 0; i
< velems
->count
; i
++) {
1188 format
= velems
->velem
[i
].src_format
;
1190 type
= r300_translate_vertex_data_type(format
) |
1191 (i
<< R300_DST_VEC_LOC_SHIFT
);
1192 swizzle
= r300_translate_vertex_data_swizzle(format
);
1195 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1196 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1198 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1199 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1203 /* Set the last vector in the PSC. */
1207 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1208 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1210 vstream
->count
= (i
>> 1) + 1;
1213 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1215 const struct pipe_vertex_element
* attribs
)
1217 struct r300_vertex_element_state
*velems
;
1220 assert(count
<= PIPE_MAX_ATTRIBS
);
1221 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1222 if (velems
!= NULL
) {
1223 velems
->count
= count
;
1224 memcpy(velems
->velem
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
1226 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1227 /* Check if the format is aligned to the size of DWORD. */
1228 for (i
= 0; i
< count
; i
++) {
1229 size
= util_format_get_blocksize(attribs
[i
].src_format
);
1231 if (size
% 4 != 0) {
1232 /* XXX Shouldn't we align the format? */
1233 fprintf(stderr
, "r300_create_vertex_elements_state: "
1234 "Unaligned format %s:%i isn't supported\n",
1235 util_format_name(attribs
[i
].src_format
), size
);
1241 r300_vertex_psc(velems
);
1247 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1250 struct r300_context
*r300
= r300_context(pipe
);
1251 struct r300_vertex_element_state
*velems
= state
;
1253 if (velems
== NULL
) {
1257 r300
->velems
= velems
;
1260 draw_flush(r300
->draw
);
1261 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1264 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1265 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1268 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1273 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1274 const struct pipe_shader_state
* shader
)
1276 struct r300_context
* r300
= r300_context(pipe
);
1278 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1279 r300_vertex_shader_common_init(vs
, shader
);
1281 if (r300
->screen
->caps
.has_tcl
) {
1282 r300_translate_vertex_shader(r300
, vs
);
1284 vs
->draw_vs
= draw_create_vertex_shader(r300
->draw
, shader
);
1290 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1292 struct r300_context
* r300
= r300_context(pipe
);
1293 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1296 r300
->vs_state
.state
= NULL
;
1299 if (vs
== r300
->vs_state
.state
) {
1302 r300
->vs_state
.state
= vs
;
1304 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
1306 r300_vertex_shader_setup_wpos(r300
);
1308 memcpy(r300
->vap_output_state
.state
, &vs
->vap_out
,
1309 sizeof(struct r300_vap_output_state
));
1310 r300
->vap_output_state
.dirty
= TRUE
;
1312 /* The majority of the RS block bits is dependent on the vertex shader. */
1313 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
1315 if (r300
->screen
->caps
.has_tcl
) {
1316 r300
->vs_state
.dirty
= TRUE
;
1317 r300
->vs_state
.size
= vs
->code
.length
+ 9;
1319 r300
->pvs_flush
.dirty
= TRUE
;
1321 r300
->dirty_state
|= R300_NEW_VERTEX_SHADER_CONSTANTS
;
1323 draw_flush(r300
->draw
);
1324 draw_bind_vertex_shader(r300
->draw
,
1325 (struct draw_vertex_shader
*)vs
->draw_vs
);
1329 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1331 struct r300_context
* r300
= r300_context(pipe
);
1332 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1334 if (r300
->screen
->caps
.has_tcl
) {
1335 rc_constants_destroy(&vs
->code
.constants
);
1337 draw_delete_vertex_shader(r300
->draw
,
1338 (struct draw_vertex_shader
*)vs
->draw_vs
);
1341 FREE((void*)vs
->state
.tokens
);
1345 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1346 uint shader
, uint index
,
1347 struct pipe_buffer
*buf
)
1349 struct r300_context
* r300
= r300_context(pipe
);
1353 if (buf
== NULL
|| buf
->size
== 0 ||
1354 (mapped
= pipe_buffer_map(pipe
->screen
, buf
, PIPE_BUFFER_USAGE_CPU_READ
)) == NULL
)
1356 r300
->shader_constants
[shader
].count
= 0;
1360 assert((buf
->size
% 4 * sizeof(float)) == 0);
1362 /* Check the size of the constant buffer. */
1364 case PIPE_SHADER_VERTEX
:
1367 case PIPE_SHADER_FRAGMENT
:
1368 if (r300
->screen
->caps
.is_r500
) {
1370 /* XXX Implement emission of r400's extended constant buffer. */
1371 /*} else if (r300->screen->caps.is_r400) {
1381 /* XXX Subtract immediates and RC_STATE_* variables. */
1382 if (buf
->size
> (sizeof(float) * 4 * max_size
)) {
1383 fprintf(stderr
, "r300: Max size of the constant buffer is "
1384 "%i*4 floats.\n", max_size
);
1388 memcpy(r300
->shader_constants
[shader
].constants
, mapped
, buf
->size
);
1389 r300
->shader_constants
[shader
].count
= buf
->size
/ (4 * sizeof(float));
1390 pipe_buffer_unmap(pipe
->screen
, buf
);
1392 if (shader
== PIPE_SHADER_VERTEX
) {
1393 if (r300
->screen
->caps
.has_tcl
) {
1394 r300
->dirty_state
|= R300_NEW_VERTEX_SHADER_CONSTANTS
;
1395 r300
->pvs_flush
.dirty
= TRUE
;
1396 } else if (r300
->draw
) {
1397 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
1398 0, r300
->shader_constants
[PIPE_SHADER_VERTEX
].constants
,
1401 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
1402 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1406 void r300_init_state_functions(struct r300_context
* r300
)
1408 r300
->context
.create_blend_state
= r300_create_blend_state
;
1409 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1410 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1412 r300
->context
.set_blend_color
= r300_set_blend_color
;
1414 r300
->context
.set_clip_state
= r300_set_clip_state
;
1416 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1418 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1419 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1420 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1422 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
1424 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1426 r300
->context
.create_fs_state
= r300_create_fs_state
;
1427 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1428 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1430 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1432 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1433 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1434 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1436 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1437 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1438 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1439 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1441 r300
->context
.set_fragment_sampler_views
= r300_set_fragment_sampler_views
;
1442 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
1443 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
1445 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1447 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1449 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1451 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
1452 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
1453 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
1455 r300
->context
.create_vs_state
= r300_create_vs_state
;
1456 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1457 r300
->context
.delete_vs_state
= r300_delete_vs_state
;