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"
35 #include "r300_emit.h"
37 #include "r300_screen.h"
38 #include "r300_screen_buffer.h"
39 #include "r300_state.h"
40 #include "r300_state_inlines.h"
42 #include "r300_texture.h"
44 #include "r300_winsys.h"
46 /* r300_state: Functions used to intialize state context by translating
47 * Gallium state objects into semi-native r300 state objects. */
49 #define UPDATE_STATE(cso, atom) \
50 if (cso != atom.state) { \
55 static boolean
blend_discard_if_src_alpha_0(unsigned srcRGB
, unsigned srcA
,
56 unsigned dstRGB
, unsigned dstA
)
58 /* If the blend equation is ADD or REVERSE_SUBTRACT,
59 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
60 * will not be changed.
61 * Notice that the dst factors are the src factors inverted. */
62 return (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
63 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
64 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
65 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
66 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
67 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
68 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
69 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
70 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
71 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
72 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
73 dstA
== PIPE_BLENDFACTOR_ONE
);
76 static boolean
blend_discard_if_src_alpha_1(unsigned srcRGB
, unsigned srcA
,
77 unsigned dstRGB
, unsigned dstA
)
79 /* If the blend equation is ADD or REVERSE_SUBTRACT,
80 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
81 * will not be changed.
82 * Notice that the dst factors are the src factors inverted. */
83 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
84 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
85 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
86 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
87 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
88 (dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
89 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
90 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
91 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
92 dstA
== PIPE_BLENDFACTOR_ONE
);
95 static boolean
blend_discard_if_src_color_0(unsigned srcRGB
, unsigned srcA
,
96 unsigned dstRGB
, unsigned dstA
)
98 /* If the blend equation is ADD or REVERSE_SUBTRACT,
99 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
100 * will not be changed.
101 * Notice that the dst factors are the src factors inverted. */
102 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
103 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
104 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
105 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
106 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
107 (dstA
== PIPE_BLENDFACTOR_ONE
);
110 static boolean
blend_discard_if_src_color_1(unsigned srcRGB
, unsigned srcA
,
111 unsigned dstRGB
, unsigned dstA
)
113 /* If the blend equation is ADD or REVERSE_SUBTRACT,
114 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
115 * will not be changed.
116 * Notice that the dst factors are the src factors inverted. */
117 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
118 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
119 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
120 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
121 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
122 (dstA
== PIPE_BLENDFACTOR_ONE
);
125 static boolean
blend_discard_if_src_alpha_color_0(unsigned srcRGB
, unsigned srcA
,
126 unsigned dstRGB
, unsigned dstA
)
128 /* If the blend equation is ADD or REVERSE_SUBTRACT,
129 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
130 * the colorbuffer will not be changed.
131 * Notice that the dst factors are the src factors inverted. */
132 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
133 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
134 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
135 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
136 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
137 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
138 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
139 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
140 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
141 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
142 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
143 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
144 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
145 dstA
== PIPE_BLENDFACTOR_ONE
);
148 static boolean
blend_discard_if_src_alpha_color_1(unsigned srcRGB
, unsigned srcA
,
149 unsigned dstRGB
, unsigned dstA
)
151 /* If the blend equation is ADD or REVERSE_SUBTRACT,
152 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
153 * the colorbuffer will not be changed.
154 * Notice that the dst factors are the src factors inverted. */
155 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
156 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
157 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
158 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
159 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
160 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
161 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
162 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
163 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
164 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
165 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
166 dstA
== PIPE_BLENDFACTOR_ONE
);
169 static unsigned bgra_cmask(unsigned mask
)
171 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
173 return ((mask
& PIPE_MASK_R
) << 2) |
174 ((mask
& PIPE_MASK_B
) >> 2) |
175 (mask
& (PIPE_MASK_G
| PIPE_MASK_A
));
178 /* Create a new blend state based on the CSO blend state.
180 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
181 static void* r300_create_blend_state(struct pipe_context
* pipe
,
182 const struct pipe_blend_state
* state
)
184 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
185 struct r300_blend_state
* blend
= CALLOC_STRUCT(r300_blend_state
);
187 if (state
->rt
[0].blend_enable
)
189 unsigned eqRGB
= state
->rt
[0].rgb_func
;
190 unsigned srcRGB
= state
->rt
[0].rgb_src_factor
;
191 unsigned dstRGB
= state
->rt
[0].rgb_dst_factor
;
193 unsigned eqA
= state
->rt
[0].alpha_func
;
194 unsigned srcA
= state
->rt
[0].alpha_src_factor
;
195 unsigned dstA
= state
->rt
[0].alpha_dst_factor
;
197 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
198 * this is just the crappy D3D naming */
199 blend
->blend_control
= R300_ALPHA_BLEND_ENABLE
|
200 r300_translate_blend_function(eqRGB
) |
201 ( r300_translate_blend_factor(srcRGB
) << R300_SRC_BLEND_SHIFT
) |
202 ( r300_translate_blend_factor(dstRGB
) << R300_DST_BLEND_SHIFT
);
204 /* Optimization: some operations do not require the destination color.
206 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
207 * otherwise blending gives incorrect results. It seems to be
209 if (eqRGB
== PIPE_BLEND_MIN
|| eqA
== PIPE_BLEND_MIN
||
210 eqRGB
== PIPE_BLEND_MAX
|| eqA
== PIPE_BLEND_MAX
||
211 dstRGB
!= PIPE_BLENDFACTOR_ZERO
||
212 dstA
!= PIPE_BLENDFACTOR_ZERO
||
213 srcRGB
== PIPE_BLENDFACTOR_DST_COLOR
||
214 srcRGB
== PIPE_BLENDFACTOR_DST_ALPHA
||
215 srcRGB
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
216 srcRGB
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
217 srcA
== PIPE_BLENDFACTOR_DST_COLOR
||
218 srcA
== PIPE_BLENDFACTOR_DST_ALPHA
||
219 srcA
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
220 srcA
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
221 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
) {
222 /* Enable reading from the colorbuffer. */
223 blend
->blend_control
|= R300_READ_ENABLE
;
225 if (r300screen
->caps
.is_r500
) {
226 /* Optimization: Depending on incoming pixels, we can
227 * conditionally disable the reading in hardware... */
228 if (eqRGB
!= PIPE_BLEND_MIN
&& eqA
!= PIPE_BLEND_MIN
&&
229 eqRGB
!= PIPE_BLEND_MAX
&& eqA
!= PIPE_BLEND_MAX
) {
230 /* Disable reading if SRC_ALPHA == 0. */
231 if ((dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
232 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
233 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
234 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
235 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
236 blend
->blend_control
|= R500_SRC_ALPHA_0_NO_READ
;
239 /* Disable reading if SRC_ALPHA == 1. */
240 if ((dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
241 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
242 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
243 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
244 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
245 blend
->blend_control
|= R500_SRC_ALPHA_1_NO_READ
;
251 /* Optimization: discard pixels which don't change the colorbuffer.
253 * The code below is non-trivial and some math is involved.
255 * Discarding pixels must be disabled when FP16 AA is enabled.
256 * This is a hardware bug. Also, this implementation wouldn't work
257 * with FP blending enabled and equation clamping disabled.
259 * Equations other than ADD are rarely used and therefore won't be
261 if ((eqRGB
== PIPE_BLEND_ADD
|| eqRGB
== PIPE_BLEND_REVERSE_SUBTRACT
) &&
262 (eqA
== PIPE_BLEND_ADD
|| eqA
== PIPE_BLEND_REVERSE_SUBTRACT
)) {
264 * REVERSE_SUBTRACT: Y-X
267 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
268 * then CB will not be changed.
270 * Given the srcFactor and dstFactor variables, we can derive
271 * what src and dst should be equal to and discard appropriate
274 if (blend_discard_if_src_alpha_0(srcRGB
, srcA
, dstRGB
, dstA
)) {
275 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0
;
276 } else if (blend_discard_if_src_alpha_1(srcRGB
, srcA
,
278 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1
;
279 } else if (blend_discard_if_src_color_0(srcRGB
, srcA
,
281 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0
;
282 } else if (blend_discard_if_src_color_1(srcRGB
, srcA
,
284 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1
;
285 } else if (blend_discard_if_src_alpha_color_0(srcRGB
, srcA
,
287 blend
->blend_control
|=
288 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0
;
289 } else if (blend_discard_if_src_alpha_color_1(srcRGB
, srcA
,
291 blend
->blend_control
|=
292 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1
;
297 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
298 blend
->blend_control
|= R300_SEPARATE_ALPHA_ENABLE
;
299 blend
->alpha_blend_control
=
300 r300_translate_blend_function(eqA
) |
301 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
302 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
306 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
307 if (state
->logicop_enable
) {
308 blend
->rop
= R300_RB3D_ROPCNTL_ROP_ENABLE
|
309 (state
->logicop_func
) << R300_RB3D_ROPCNTL_ROP_SHIFT
;
312 /* Color channel masks for all MRTs. */
313 blend
->color_channel_mask
= bgra_cmask(state
->rt
[0].colormask
);
314 if (r300screen
->caps
.is_r500
&& state
->independent_blend_enable
) {
315 if (state
->rt
[1].blend_enable
) {
316 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[1].colormask
) << 4;
318 if (state
->rt
[2].blend_enable
) {
319 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[2].colormask
) << 8;
321 if (state
->rt
[3].blend_enable
) {
322 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[3].colormask
) << 12;
326 /* Neither fglrx nor classic r300 ever set this, regardless of dithering
327 * state. Since it's an optional implementation detail, we can leave it
328 * out and never dither.
330 * This could be revisited if we ever get quality or conformance hints.
333 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
334 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
341 /* Bind blend state. */
342 static void r300_bind_blend_state(struct pipe_context
* pipe
,
345 struct r300_context
* r300
= r300_context(pipe
);
347 UPDATE_STATE(state
, r300
->blend_state
);
350 /* Free blend state. */
351 static void r300_delete_blend_state(struct pipe_context
* pipe
,
357 /* Convert float to 10bit integer */
358 static unsigned float_to_fixed10(float f
)
360 return CLAMP((unsigned)(f
* 1023.9f
), 0, 1023);
364 * Setup both R300 and R500 registers, figure out later which one to write. */
365 static void r300_set_blend_color(struct pipe_context
* pipe
,
366 const struct pipe_blend_color
* color
)
368 struct r300_context
* r300
= r300_context(pipe
);
369 struct r300_blend_color_state
* state
=
370 (struct r300_blend_color_state
*)r300
->blend_color_state
.state
;
373 util_pack_color(color
->color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
374 state
->blend_color
= uc
.ui
;
376 /* XXX if FP16 blending is enabled, we should use the FP16 format */
377 state
->blend_color_red_alpha
=
378 float_to_fixed10(color
->color
[0]) |
379 (float_to_fixed10(color
->color
[3]) << 16);
380 state
->blend_color_green_blue
=
381 float_to_fixed10(color
->color
[2]) |
382 (float_to_fixed10(color
->color
[1]) << 16);
384 r300
->blend_color_state
.size
= r300
->screen
->caps
.is_r500
? 3 : 2;
385 r300
->blend_color_state
.dirty
= TRUE
;
388 static void r300_set_clip_state(struct pipe_context
* pipe
,
389 const struct pipe_clip_state
* state
)
391 struct r300_context
* r300
= r300_context(pipe
);
395 if (r300
->screen
->caps
.has_tcl
) {
396 memcpy(r300
->clip_state
.state
, state
, sizeof(struct pipe_clip_state
));
397 r300
->clip_state
.size
= 29;
399 draw_flush(r300
->draw
);
400 draw_set_clip_state(r300
->draw
, state
);
401 r300
->clip_state
.size
= 2;
404 r300
->clip_state
.dirty
= TRUE
;
407 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
409 * This contains the depth buffer, stencil buffer, alpha test, and such.
410 * On the Radeon, depth and stencil buffer setup are intertwined, which is
411 * the reason for some of the strange-looking assignments across registers. */
413 r300_create_dsa_state(struct pipe_context
* pipe
,
414 const struct pipe_depth_stencil_alpha_state
* state
)
416 struct r300_capabilities
*caps
= &r300_screen(pipe
->screen
)->caps
;
417 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
419 /* Depth test setup. */
420 if (state
->depth
.enabled
) {
421 dsa
->z_buffer_control
|= R300_Z_ENABLE
;
423 if (state
->depth
.writemask
) {
424 dsa
->z_buffer_control
|= R300_Z_WRITE_ENABLE
;
427 dsa
->z_stencil_control
|=
428 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
432 /* Stencil buffer setup. */
433 if (state
->stencil
[0].enabled
) {
434 dsa
->z_buffer_control
|= R300_STENCIL_ENABLE
;
435 dsa
->z_stencil_control
|=
436 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
437 R300_S_FRONT_FUNC_SHIFT
) |
438 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
439 R300_S_FRONT_SFAIL_OP_SHIFT
) |
440 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
441 R300_S_FRONT_ZPASS_OP_SHIFT
) |
442 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
443 R300_S_FRONT_ZFAIL_OP_SHIFT
);
445 dsa
->stencil_ref_mask
=
446 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
447 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
449 if (state
->stencil
[1].enabled
) {
450 dsa
->two_sided
= TRUE
;
452 dsa
->z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
453 dsa
->z_stencil_control
|=
454 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
455 R300_S_BACK_FUNC_SHIFT
) |
456 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
457 R300_S_BACK_SFAIL_OP_SHIFT
) |
458 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
459 R300_S_BACK_ZPASS_OP_SHIFT
) |
460 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
461 R300_S_BACK_ZFAIL_OP_SHIFT
);
463 dsa
->stencil_ref_bf
=
464 (state
->stencil
[1].valuemask
<< R300_STENCILMASK_SHIFT
) |
465 (state
->stencil
[1].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
468 dsa
->z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
470 dsa
->stencil_ref_bf_fallback
=
471 (state
->stencil
[0].valuemask
!= state
->stencil
[1].valuemask
||
472 state
->stencil
[0].writemask
!= state
->stencil
[1].writemask
);
477 /* Alpha test setup. */
478 if (state
->alpha
.enabled
) {
479 dsa
->alpha_function
=
480 r300_translate_alpha_function(state
->alpha
.func
) |
481 R300_FG_ALPHA_FUNC_ENABLE
;
483 /* We could use 10bit alpha ref but who needs that? */
484 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
487 dsa
->alpha_function
|= R500_FG_ALPHA_FUNC_8BIT
;
493 static void r300_update_stencil_ref_fallback_status(struct r300_context
*r300
)
495 struct r300_dsa_state
*dsa
= (struct r300_dsa_state
*)r300
->dsa_state
.state
;
497 if (r300
->screen
->caps
.is_r500
) {
501 r300
->stencil_ref_bf_fallback
=
502 dsa
->stencil_ref_bf_fallback
||
504 r300
->stencil_ref
.ref_value
[0] != r300
->stencil_ref
.ref_value
[1]);
507 /* Bind DSA state. */
508 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
511 struct r300_context
* r300
= r300_context(pipe
);
517 UPDATE_STATE(state
, r300
->dsa_state
);
519 r300_update_stencil_ref_fallback_status(r300
);
522 /* Free DSA state. */
523 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
529 static void r300_set_stencil_ref(struct pipe_context
* pipe
,
530 const struct pipe_stencil_ref
* sr
)
532 struct r300_context
* r300
= r300_context(pipe
);
534 r300
->stencil_ref
= *sr
;
535 r300
->dsa_state
.dirty
= TRUE
;
537 r300_update_stencil_ref_fallback_status(r300
);
540 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
541 static void r300_fb_set_tiling_flags(struct r300_context
*r300
,
542 const struct pipe_framebuffer_state
*old_state
,
543 const struct pipe_framebuffer_state
*new_state
)
545 struct r300_texture
*tex
;
548 /* Set tiling flags for new surfaces. */
549 for (i
= 0; i
< new_state
->nr_cbufs
; i
++) {
550 tex
= r300_texture(new_state
->cbufs
[i
]->texture
);
551 level
= new_state
->cbufs
[i
]->level
;
553 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
556 tex
->mip_macrotile
[level
]);
558 if (new_state
->zsbuf
) {
559 tex
= r300_texture(new_state
->zsbuf
->texture
);
560 level
= new_state
->zsbuf
->level
;
562 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
565 tex
->mip_macrotile
[level
]);
570 r300_set_framebuffer_state(struct pipe_context
* pipe
,
571 const struct pipe_framebuffer_state
* state
)
573 struct r300_context
* r300
= r300_context(pipe
);
574 struct pipe_framebuffer_state
*old_state
= r300
->fb_state
.state
;
575 unsigned max_width
, max_height
;
576 uint32_t zbuffer_bpp
= 0;
578 if (state
->nr_cbufs
> 4) {
579 fprintf(stderr
, "r300: Implementation error: Too many MRTs in %s, "
580 "refusing to bind framebuffer state!\n", __FUNCTION__
);
584 if (r300
->screen
->caps
.is_r500
) {
585 max_width
= max_height
= 4096;
586 } else if (r300
->screen
->caps
.is_r400
) {
587 max_width
= max_height
= 4021;
589 max_width
= max_height
= 2560;
592 if (state
->width
> max_width
|| state
->height
> max_height
) {
593 fprintf(stderr
, "r300: Implementation error: Render targets are too "
594 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
599 draw_flush(r300
->draw
);
602 r300
->fb_state
.dirty
= TRUE
;
604 /* If nr_cbufs is changed from zero to non-zero or vice versa... */
605 if (!!old_state
->nr_cbufs
!= !!state
->nr_cbufs
) {
606 r300
->blend_state
.dirty
= TRUE
;
608 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
609 if (!!old_state
->zsbuf
!= !!state
->zsbuf
) {
610 r300
->dsa_state
.dirty
= TRUE
;
613 /* The tiling flags are dependent on the surface miplevel, unfortunately. */
614 r300_fb_set_tiling_flags(r300
, r300
->fb_state
.state
, state
);
616 memcpy(r300
->fb_state
.state
, state
, sizeof(struct pipe_framebuffer_state
));
618 r300
->fb_state
.size
= (10 * state
->nr_cbufs
) + (2 * (4 - state
->nr_cbufs
)) +
619 (state
->zsbuf
? 10 : 0) + 11;
621 /* Polygon offset depends on the zbuffer bit depth. */
622 if (state
->zsbuf
&& r300
->polygon_offset_enabled
) {
623 switch (util_format_get_blocksize(state
->zsbuf
->texture
->format
)) {
632 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
633 r300
->zbuffer_bpp
= zbuffer_bpp
;
634 r300
->rs_state
.dirty
= TRUE
;
639 /* Create fragment shader state. */
640 static void* r300_create_fs_state(struct pipe_context
* pipe
,
641 const struct pipe_shader_state
* shader
)
643 struct r300_fragment_shader
* fs
= NULL
;
645 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
647 /* Copy state directly into shader. */
649 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
654 void r300_mark_fs_code_dirty(struct r300_context
*r300
)
656 struct r300_fragment_shader
* fs
= r300_fs(r300
);
658 r300
->fs
.dirty
= TRUE
;
659 r300
->fs_rc_constant_state
.dirty
= TRUE
;
660 r300
->fs_constants
.dirty
= TRUE
;
662 if (r300
->screen
->caps
.is_r500
) {
663 r300
->fs
.size
= r500_get_fs_atom_size(r300
);
664 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 7;
665 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 3;
667 r300
->fs
.size
= r300_get_fs_atom_size(r300
);
668 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 5;
669 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 1;
673 /* Bind fragment shader state. */
674 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
676 struct r300_context
* r300
= r300_context(pipe
);
677 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
680 r300
->fs
.state
= NULL
;
685 r300_pick_fragment_shader(r300
);
686 r300_mark_fs_code_dirty(r300
);
688 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
691 /* Delete fragment shader state. */
692 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
694 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
695 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
700 rc_constants_destroy(&tmp
->code
.constants
);
703 FREE((void*)fs
->state
.tokens
);
707 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
708 const struct pipe_poly_stipple
* state
)
710 /* XXX no idea how to set this up, but not terribly important */
713 /* Create a new rasterizer state based on the CSO rasterizer state.
715 * This is a very large chunk of state, and covers most of the graphics
716 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
718 * In a not entirely unironic sidenote, this state has nearly nothing to do
719 * with the actual block on the Radeon called the rasterizer (RS). */
720 static void* r300_create_rs_state(struct pipe_context
* pipe
,
721 const struct pipe_rasterizer_state
* state
)
723 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
726 /* Copy rasterizer state for Draw. */
729 #ifdef PIPE_ARCH_LITTLE_ENDIAN
730 rs
->vap_control_status
= R300_VC_NO_SWAP
;
732 rs
->vap_control_status
= R300_VC_32BIT_SWAP
;
735 /* If no TCL engine is present, turn off the HW TCL. */
736 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
737 rs
->vap_control_status
|= R300_VAP_TCL_BYPASS
;
740 rs
->point_size
= pack_float_16_6x(state
->point_size
) |
741 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
743 rs
->line_control
= pack_float_16_6x(state
->line_width
) |
744 R300_GA_LINE_CNTL_END_TYPE_COMP
;
746 /* Enable polygon mode */
747 if (state
->fill_cw
!= PIPE_POLYGON_MODE_FILL
||
748 state
->fill_ccw
!= PIPE_POLYGON_MODE_FILL
) {
749 rs
->polygon_mode
= R300_GA_POLY_MODE_DUAL
;
752 /* Radeons don't think in "CW/CCW", they think in "front/back". */
753 if (state
->front_winding
== PIPE_WINDING_CW
) {
754 rs
->cull_mode
= R300_FRONT_FACE_CW
;
757 if (state
->offset_cw
) {
758 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
760 if (state
->offset_ccw
) {
761 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
765 if (rs
->polygon_mode
) {
767 r300_translate_polygon_mode_front(state
->fill_cw
);
769 r300_translate_polygon_mode_back(state
->fill_ccw
);
772 rs
->cull_mode
= R300_FRONT_FACE_CCW
;
775 if (state
->offset_ccw
) {
776 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
778 if (state
->offset_cw
) {
779 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
783 if (rs
->polygon_mode
) {
785 r300_translate_polygon_mode_front(state
->fill_ccw
);
787 r300_translate_polygon_mode_back(state
->fill_cw
);
790 if (state
->front_winding
& state
->cull_mode
) {
791 rs
->cull_mode
|= R300_CULL_FRONT
;
793 if (~(state
->front_winding
) & state
->cull_mode
) {
794 rs
->cull_mode
|= R300_CULL_BACK
;
797 if (rs
->polygon_offset_enable
) {
798 rs
->depth_offset
= state
->offset_units
;
799 rs
->depth_scale
= state
->offset_scale
;
802 if (state
->line_stipple_enable
) {
803 rs
->line_stipple_config
=
804 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
805 (fui((float)state
->line_stipple_factor
) &
806 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
807 /* XXX this might need to be scaled up */
808 rs
->line_stipple_value
= state
->line_stipple_pattern
;
811 if (state
->flatshade
) {
812 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
814 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
817 rs
->clip_rule
= state
->scissor
? 0xAAAA : 0xFFFF;
820 if (state
->sprite_coord_enable
) {
821 rs
->stuffing_enable
= R300_GB_POINT_STUFF_ENABLE
;
822 for (i
= 0; i
< 8; i
++) {
823 if (state
->sprite_coord_enable
& (1 << i
))
824 rs
->stuffing_enable
|=
825 R300_GB_TEX_STR
<< (R300_GB_TEX0_SOURCE_SHIFT
+ (i
*2));
828 rs
->point_texcoord_left
= 0.0f
;
829 rs
->point_texcoord_right
= 1.0f
;
831 switch (state
->sprite_coord_mode
) {
832 case PIPE_SPRITE_COORD_UPPER_LEFT
:
833 rs
->point_texcoord_top
= 0.0f
;
834 rs
->point_texcoord_bottom
= 1.0f
;
836 case PIPE_SPRITE_COORD_LOWER_LEFT
:
837 rs
->point_texcoord_top
= 1.0f
;
838 rs
->point_texcoord_bottom
= 0.0f
;
846 /* Bind rasterizer state. */
847 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
849 struct r300_context
* r300
= r300_context(pipe
);
850 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
851 int last_sprite_coord_enable
= r300
->sprite_coord_enable
;
854 draw_flush(r300
->draw
);
855 draw_set_rasterizer_state(r300
->draw
, &rs
->rs
, state
);
859 r300
->polygon_offset_enabled
= rs
->rs
.offset_cw
|| rs
->rs
.offset_ccw
;
860 r300
->sprite_coord_enable
= rs
->rs
.sprite_coord_enable
;
862 r300
->polygon_offset_enabled
= FALSE
;
863 r300
->sprite_coord_enable
= 0;
866 UPDATE_STATE(state
, r300
->rs_state
);
867 r300
->rs_state
.size
= 26 + (r300
->polygon_offset_enabled
? 5 : 0);
869 if (last_sprite_coord_enable
!= r300
->sprite_coord_enable
) {
870 r300
->rs_block_state
.dirty
= TRUE
;
874 /* Free rasterizer state. */
875 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
881 r300_create_sampler_state(struct pipe_context
* pipe
,
882 const struct pipe_sampler_state
* state
)
884 struct r300_context
* r300
= r300_context(pipe
);
885 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
886 boolean is_r500
= r300
->screen
->caps
.is_r500
;
890 sampler
->state
= *state
;
893 (r300_translate_wrap(state
->wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
894 (r300_translate_wrap(state
->wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
895 (r300_translate_wrap(state
->wrap_r
) << R300_TX_WRAP_R_SHIFT
);
897 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
898 state
->mag_img_filter
,
899 state
->min_mip_filter
,
900 state
->max_anisotropy
> 0);
902 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
904 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
905 /* We must pass these to the merge function to clamp them properly. */
906 sampler
->min_lod
= MAX2((unsigned)state
->min_lod
, 0);
907 sampler
->max_lod
= MAX2((unsigned)ceilf(state
->max_lod
), 0);
909 lod_bias
= CLAMP((int)(state
->lod_bias
* 32 + 1), -(1 << 9), (1 << 9) - 1);
911 sampler
->filter1
|= lod_bias
<< R300_LOD_BIAS_SHIFT
;
913 /* This is very high quality anisotropic filtering for R5xx.
914 * It's good for benchmarking the performance of texturing but
915 * in practice we don't want to slow down the driver because it's
916 * a pretty good performance killer. Feel free to play with it. */
917 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
918 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
921 util_pack_color(state
->border_color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
922 sampler
->border_color
= uc
.ui
;
924 /* R500-specific fixups and optimizations */
925 if (r300
->screen
->caps
.is_r500
) {
926 sampler
->filter1
|= R500_BORDER_FIX
;
929 return (void*)sampler
;
932 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
936 struct r300_context
* r300
= r300_context(pipe
);
937 struct r300_textures_state
* state
=
938 (struct r300_textures_state
*)r300
->textures_state
.state
;
939 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
941 if (count
> tex_units
) {
945 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
946 state
->sampler_state_count
= count
;
948 r300
->textures_state
.dirty
= TRUE
;
951 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
957 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
962 static void r300_set_fragment_sampler_views(struct pipe_context
* pipe
,
964 struct pipe_sampler_view
** views
)
966 struct r300_context
* r300
= r300_context(pipe
);
967 struct r300_textures_state
* state
=
968 (struct r300_textures_state
*)r300
->textures_state
.state
;
969 struct r300_texture
*texture
;
971 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
972 boolean dirty_tex
= FALSE
;
974 if (count
> tex_units
) {
978 for (i
= 0; i
< count
; i
++) {
979 if (&state
->sampler_views
[i
]->base
!= views
[i
]) {
980 pipe_sampler_view_reference(
981 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
988 /* A new sampler view (= texture)... */
991 /* Set the texrect factor in the fragment shader.
992 * Needed for RECT and NPOT fallback. */
993 texture
= r300_texture(views
[i
]->texture
);
994 if (texture
->uses_pitch
) {
995 r300
->fs_rc_constant_state
.dirty
= TRUE
;
1000 for (i
= count
; i
< tex_units
; i
++) {
1001 if (state
->sampler_views
[i
]) {
1002 pipe_sampler_view_reference(
1003 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1008 state
->sampler_view_count
= count
;
1010 r300
->textures_state
.dirty
= TRUE
;
1013 r300
->texture_cache_inval
.dirty
= TRUE
;
1017 static struct pipe_sampler_view
*
1018 r300_create_sampler_view(struct pipe_context
*pipe
,
1019 struct pipe_resource
*texture
,
1020 const struct pipe_sampler_view
*templ
)
1022 struct r300_sampler_view
*view
= CALLOC_STRUCT(r300_sampler_view
);
1023 struct r300_texture
*tex
= r300_texture(texture
);
1026 view
->base
= *templ
;
1027 view
->base
.reference
.count
= 1;
1028 view
->base
.context
= pipe
;
1029 view
->base
.texture
= NULL
;
1030 pipe_resource_reference(&view
->base
.texture
, texture
);
1032 view
->swizzle
[0] = templ
->swizzle_r
;
1033 view
->swizzle
[1] = templ
->swizzle_g
;
1034 view
->swizzle
[2] = templ
->swizzle_b
;
1035 view
->swizzle
[3] = templ
->swizzle_a
;
1037 view
->format
= tex
->tx_format
;
1038 view
->format
.format1
|= r300_translate_texformat(templ
->format
,
1040 if (r300_screen(pipe
->screen
)->caps
.is_r500
) {
1041 view
->format
.format2
|= r500_tx_format_msb_bit(templ
->format
);
1045 return (struct pipe_sampler_view
*)view
;
1049 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1050 struct pipe_sampler_view
*view
)
1052 pipe_resource_reference(&view
->texture
, NULL
);
1056 static void r300_set_scissor_state(struct pipe_context
* pipe
,
1057 const struct pipe_scissor_state
* state
)
1059 struct r300_context
* r300
= r300_context(pipe
);
1061 memcpy(r300
->scissor_state
.state
, state
,
1062 sizeof(struct pipe_scissor_state
));
1064 r300
->scissor_state
.dirty
= TRUE
;
1067 static void r300_set_viewport_state(struct pipe_context
* pipe
,
1068 const struct pipe_viewport_state
* state
)
1070 struct r300_context
* r300
= r300_context(pipe
);
1071 struct r300_viewport_state
* viewport
=
1072 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1074 r300
->viewport
= *state
;
1076 /* Do the transform in HW. */
1077 viewport
->vte_control
= R300_VTX_W0_FMT
;
1079 if (state
->scale
[0] != 1.0f
) {
1080 viewport
->xscale
= state
->scale
[0];
1081 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1083 if (state
->scale
[1] != 1.0f
) {
1084 viewport
->yscale
= state
->scale
[1];
1085 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1087 if (state
->scale
[2] != 1.0f
) {
1088 viewport
->zscale
= state
->scale
[2];
1089 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1091 if (state
->translate
[0] != 0.0f
) {
1092 viewport
->xoffset
= state
->translate
[0];
1093 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1095 if (state
->translate
[1] != 0.0f
) {
1096 viewport
->yoffset
= state
->translate
[1];
1097 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1099 if (state
->translate
[2] != 0.0f
) {
1100 viewport
->zoffset
= state
->translate
[2];
1101 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1104 r300
->viewport_state
.dirty
= TRUE
;
1105 if (r300
->fs
.state
&& r300_fs(r300
)->shader
->inputs
.wpos
!= ATTR_UNUSED
) {
1106 r300
->fs_rc_constant_state
.dirty
= TRUE
;
1110 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
1112 const struct pipe_vertex_buffer
* buffers
)
1114 struct r300_context
* r300
= r300_context(pipe
);
1115 struct pipe_vertex_buffer
*vbo
;
1116 unsigned i
, max_index
= (1 << 24) - 1;
1117 boolean any_user_buffer
= FALSE
;
1119 if (count
== r300
->vertex_buffer_count
&&
1120 memcmp(r300
->vertex_buffer
, buffers
,
1121 sizeof(struct pipe_vertex_buffer
) * count
) == 0) {
1125 /* Check if the stride is aligned to the size of DWORD. */
1126 for (i
= 0; i
< count
; i
++) {
1127 if (buffers
[i
].buffer
) {
1128 if (buffers
[i
].stride
% 4 != 0) {
1129 // XXX Shouldn't we align the buffer?
1130 fprintf(stderr
, "r300: set_vertex_buffers: "
1131 "Unaligned buffer stride %i isn't supported.\n",
1138 for (i
= 0; i
< count
; i
++) {
1139 /* Why, yes, I AM casting away constness. How did you know? */
1140 vbo
= (struct pipe_vertex_buffer
*)&buffers
[i
];
1142 /* Reference our buffer. */
1143 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, vbo
->buffer
);
1145 /* Skip NULL buffers */
1146 if (!buffers
[i
].buffer
) {
1150 if (r300_buffer_is_user_buffer(vbo
->buffer
)) {
1151 any_user_buffer
= TRUE
;
1154 if (vbo
->max_index
== ~0) {
1155 /* if no VBO stride then only one vertex value so max index is 1 */
1156 /* should think about converting to VS constants like svga does */
1161 (vbo
->buffer
->width0
- vbo
->buffer_offset
) / vbo
->stride
;
1164 max_index
= MIN2(vbo
->max_index
, max_index
);
1167 for (; i
< r300
->vertex_buffer_count
; i
++) {
1168 /* Dereference any old buffers. */
1169 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, NULL
);
1172 memcpy(r300
->vertex_buffer
, buffers
,
1173 sizeof(struct pipe_vertex_buffer
) * count
);
1175 r300
->vertex_buffer_count
= count
;
1176 r300
->vertex_buffer_max_index
= max_index
;
1177 r300
->any_user_vbs
= any_user_buffer
;
1180 draw_flush(r300
->draw
);
1181 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
1185 /* Update the PSC tables. */
1186 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1188 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1189 uint16_t type
, swizzle
;
1190 enum pipe_format format
;
1193 if (velems
->count
> 16) {
1194 fprintf(stderr
, "r300: More than 16 vertex elements are not supported,"
1195 " requested %i, using 16.\n", velems
->count
);
1199 /* Vertex shaders have no semantics on their inputs,
1200 * so PSC should just route stuff based on the vertex elements,
1201 * and not on attrib information. */
1202 for (i
= 0; i
< velems
->count
; i
++) {
1203 format
= velems
->velem
[i
].src_format
;
1205 type
= r300_translate_vertex_data_type(format
) |
1206 (i
<< R300_DST_VEC_LOC_SHIFT
);
1207 swizzle
= r300_translate_vertex_data_swizzle(format
);
1210 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1211 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1213 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1214 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1218 /* Set the last vector in the PSC. */
1222 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1223 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1225 vstream
->count
= (i
>> 1) + 1;
1228 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1230 const struct pipe_vertex_element
* attribs
)
1232 struct r300_vertex_element_state
*velems
;
1234 enum pipe_format
*format
;
1236 assert(count
<= PIPE_MAX_ATTRIBS
);
1237 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1238 if (velems
!= NULL
) {
1239 velems
->count
= count
;
1240 memcpy(velems
->velem
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
1242 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1243 r300_vertex_psc(velems
);
1245 /* Check if the format is aligned to the size of DWORD.
1246 * We only care about the blocksizes of the formats since
1247 * swizzles are already set up. */
1248 for (i
= 0; i
< count
; i
++) {
1249 format
= &velems
->velem
[i
].src_format
;
1251 /* Replace some formats with their aligned counterparts,
1252 * this is OK because we check for aligned strides too. */
1254 /* Align to RGBA8. */
1255 case PIPE_FORMAT_R8_UNORM
:
1256 case PIPE_FORMAT_R8G8_UNORM
:
1257 case PIPE_FORMAT_R8G8B8_UNORM
:
1258 *format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
1260 case PIPE_FORMAT_R8_SNORM
:
1261 case PIPE_FORMAT_R8G8_SNORM
:
1262 case PIPE_FORMAT_R8G8B8_SNORM
:
1263 *format
= PIPE_FORMAT_R8G8B8A8_SNORM
;
1265 case PIPE_FORMAT_R8_USCALED
:
1266 case PIPE_FORMAT_R8G8_USCALED
:
1267 case PIPE_FORMAT_R8G8B8_USCALED
:
1268 *format
= PIPE_FORMAT_R8G8B8A8_USCALED
;
1270 case PIPE_FORMAT_R8_SSCALED
:
1271 case PIPE_FORMAT_R8G8_SSCALED
:
1272 case PIPE_FORMAT_R8G8B8_SSCALED
:
1273 *format
= PIPE_FORMAT_R8G8B8A8_SSCALED
;
1276 /* Align to RG16. */
1277 case PIPE_FORMAT_R16_UNORM
:
1278 *format
= PIPE_FORMAT_R16G16_UNORM
;
1280 case PIPE_FORMAT_R16_SNORM
:
1281 *format
= PIPE_FORMAT_R16G16_SNORM
;
1283 case PIPE_FORMAT_R16_USCALED
:
1284 *format
= PIPE_FORMAT_R16G16_USCALED
;
1286 case PIPE_FORMAT_R16_SSCALED
:
1287 *format
= PIPE_FORMAT_R16G16_SSCALED
;
1289 case PIPE_FORMAT_R16_FLOAT
:
1290 *format
= PIPE_FORMAT_R16G16_FLOAT
;
1293 /* Align to RGBA16. */
1294 case PIPE_FORMAT_R16G16B16_UNORM
:
1295 *format
= PIPE_FORMAT_R16G16B16A16_UNORM
;
1297 case PIPE_FORMAT_R16G16B16_SNORM
:
1298 *format
= PIPE_FORMAT_R16G16B16A16_SNORM
;
1300 case PIPE_FORMAT_R16G16B16_USCALED
:
1301 *format
= PIPE_FORMAT_R16G16B16A16_USCALED
;
1303 case PIPE_FORMAT_R16G16B16_SSCALED
:
1304 *format
= PIPE_FORMAT_R16G16B16A16_SSCALED
;
1306 case PIPE_FORMAT_R16G16B16_FLOAT
:
1307 *format
= PIPE_FORMAT_R16G16B16A16_FLOAT
;
1313 size
= util_format_get_blocksize(*format
);
1315 if (size
% 4 != 0) {
1316 /* XXX Shouldn't we align the format? */
1317 fprintf(stderr
, "r300_create_vertex_elements_state: "
1318 "Unaligned format %s:%i isn't supported\n",
1319 util_format_short_name(*format
), size
);
1330 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1333 struct r300_context
*r300
= r300_context(pipe
);
1334 struct r300_vertex_element_state
*velems
= state
;
1336 if (velems
== NULL
) {
1340 r300
->velems
= velems
;
1343 draw_flush(r300
->draw
);
1344 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1347 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1348 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1351 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1356 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1357 const struct pipe_shader_state
* shader
)
1359 struct r300_context
* r300
= r300_context(pipe
);
1361 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1363 /* Copy state directly into shader. */
1364 vs
->state
= *shader
;
1365 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
1367 if (r300
->screen
->caps
.has_tcl
) {
1368 r300_translate_vertex_shader(r300
, vs
, vs
->state
.tokens
);
1370 vs
->draw_vs
= draw_create_vertex_shader(r300
->draw
, shader
);
1376 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1378 struct r300_context
* r300
= r300_context(pipe
);
1379 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1382 r300
->vs_state
.state
= NULL
;
1385 if (vs
== r300
->vs_state
.state
) {
1388 r300
->vs_state
.state
= vs
;
1390 /* The majority of the RS block bits is dependent on the vertex shader. */
1391 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
1393 if (r300
->screen
->caps
.has_tcl
) {
1394 r300
->vs_state
.dirty
= TRUE
;
1395 r300
->vs_state
.size
=
1396 vs
->code
.length
+ 9 +
1397 (vs
->immediates_count
? vs
->immediates_count
* 4 + 3 : 0);
1399 if (vs
->externals_count
) {
1400 r300
->vs_constants
.dirty
= TRUE
;
1401 r300
->vs_constants
.size
= vs
->externals_count
* 4 + 3;
1403 r300
->vs_constants
.size
= 0;
1406 r300
->pvs_flush
.dirty
= TRUE
;
1408 draw_flush(r300
->draw
);
1409 draw_bind_vertex_shader(r300
->draw
,
1410 (struct draw_vertex_shader
*)vs
->draw_vs
);
1414 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1416 struct r300_context
* r300
= r300_context(pipe
);
1417 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1419 if (r300
->screen
->caps
.has_tcl
) {
1420 rc_constants_destroy(&vs
->code
.constants
);
1422 draw_delete_vertex_shader(r300
->draw
,
1423 (struct draw_vertex_shader
*)vs
->draw_vs
);
1426 FREE((void*)vs
->state
.tokens
);
1430 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1431 uint shader
, uint index
,
1432 struct pipe_resource
*buf
)
1434 struct r300_context
* r300
= r300_context(pipe
);
1435 struct r300_constant_buffer
*cbuf
;
1436 struct pipe_transfer
*tr
;
1441 case PIPE_SHADER_VERTEX
:
1442 cbuf
= (struct r300_constant_buffer
*)r300
->vs_constants
.state
;
1445 case PIPE_SHADER_FRAGMENT
:
1446 cbuf
= (struct r300_constant_buffer
*)r300
->fs_constants
.state
;
1447 if (r300
->screen
->caps
.is_r500
) {
1458 if (buf
== NULL
|| buf
->width0
== 0 ||
1459 (mapped
= pipe_buffer_map(pipe
, buf
, PIPE_TRANSFER_READ
, &tr
)) == NULL
)
1465 assert((buf
->width0
% 4 * sizeof(float)) == 0);
1467 /* Check the size of the constant buffer. */
1468 /* XXX Subtract immediates and RC_STATE_* variables. */
1469 if (buf
->width0
> (sizeof(float) * 4 * max_size
)) {
1470 fprintf(stderr
, "r300: Max size of the constant buffer is "
1471 "%i*4 floats.\n", max_size
);
1475 memcpy(cbuf
->constants
, mapped
, buf
->width0
);
1476 cbuf
->count
= buf
->width0
/ (4 * sizeof(float));
1477 pipe_buffer_unmap(pipe
, buf
, tr
);
1479 if (shader
== PIPE_SHADER_VERTEX
) {
1480 if (r300
->screen
->caps
.has_tcl
) {
1481 if (r300
->vs_constants
.size
) {
1482 r300
->vs_constants
.dirty
= TRUE
;
1484 r300
->pvs_flush
.dirty
= TRUE
;
1485 } else if (r300
->draw
) {
1486 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
1490 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
1491 r300
->fs_constants
.dirty
= TRUE
;
1495 void r300_init_state_functions(struct r300_context
* r300
)
1497 r300
->context
.create_blend_state
= r300_create_blend_state
;
1498 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1499 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1501 r300
->context
.set_blend_color
= r300_set_blend_color
;
1503 r300
->context
.set_clip_state
= r300_set_clip_state
;
1505 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1507 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1508 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1509 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1511 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
1513 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1515 r300
->context
.create_fs_state
= r300_create_fs_state
;
1516 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1517 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1519 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1521 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1522 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1523 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1525 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1526 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1527 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1528 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1530 r300
->context
.set_fragment_sampler_views
= r300_set_fragment_sampler_views
;
1531 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
1532 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
1534 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1536 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1538 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1540 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
1541 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
1542 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
1544 r300
->context
.create_vs_state
= r300_create_vs_state
;
1545 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1546 r300
->context
.delete_vs_state
= r300_delete_vs_state
;