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_update_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
;
546 unsigned i
, j
, level
;
548 /* Reset tiling flags for old surfaces to default values. */
549 for (i
= 0; i
< old_state
->nr_cbufs
; i
++) {
550 for (j
= 0; j
< new_state
->nr_cbufs
; j
++) {
551 if (old_state
->cbufs
[i
]->texture
== new_state
->cbufs
[j
]->texture
) {
555 /* If not binding the surface again... */
556 if (j
!= new_state
->nr_cbufs
) {
560 tex
= r300_texture(old_state
->cbufs
[i
]->texture
);
563 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
569 if (old_state
->zsbuf
&&
570 (!new_state
->zsbuf
||
571 old_state
->zsbuf
->texture
!= new_state
->zsbuf
->texture
)) {
572 tex
= r300_texture(old_state
->zsbuf
->texture
);
575 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
582 /* Set tiling flags for new surfaces. */
583 for (i
= 0; i
< new_state
->nr_cbufs
; i
++) {
584 tex
= r300_texture(new_state
->cbufs
[i
]->texture
);
585 level
= new_state
->cbufs
[i
]->level
;
587 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
590 tex
->mip_macrotile
[level
]);
592 if (new_state
->zsbuf
) {
593 tex
= r300_texture(new_state
->zsbuf
->texture
);
594 level
= new_state
->zsbuf
->level
;
596 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
599 tex
->mip_macrotile
[level
]);
604 r300_set_framebuffer_state(struct pipe_context
* pipe
,
605 const struct pipe_framebuffer_state
* state
)
607 struct r300_context
* r300
= r300_context(pipe
);
608 struct pipe_framebuffer_state
*old_state
= r300
->fb_state
.state
;
609 unsigned max_width
, max_height
;
610 uint32_t zbuffer_bpp
= 0;
612 if (state
->nr_cbufs
> 4) {
613 fprintf(stderr
, "r300: Implementation error: Too many MRTs in %s, "
614 "refusing to bind framebuffer state!\n", __FUNCTION__
);
618 if (r300
->screen
->caps
.is_r500
) {
619 max_width
= max_height
= 4096;
620 } else if (r300
->screen
->caps
.is_r400
) {
621 max_width
= max_height
= 4021;
623 max_width
= max_height
= 2560;
626 if (state
->width
> max_width
|| state
->height
> max_height
) {
627 fprintf(stderr
, "r300: Implementation error: Render targets are too "
628 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
633 draw_flush(r300
->draw
);
636 r300
->fb_state
.dirty
= TRUE
;
638 /* If nr_cbufs is changed from zero to non-zero or vice versa... */
639 if (!!old_state
->nr_cbufs
!= !!state
->nr_cbufs
) {
640 r300
->blend_state
.dirty
= TRUE
;
642 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
643 if (!!old_state
->zsbuf
!= !!state
->zsbuf
) {
644 r300
->dsa_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) + 11;
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
);
687 void r300_mark_fs_code_dirty(struct r300_context
*r300
)
689 struct r300_fragment_shader
* fs
= r300_fs(r300
);
691 r300
->fs
.dirty
= TRUE
;
692 r300
->fs_rc_constant_state
.dirty
= TRUE
;
693 r300
->fs_constants
.dirty
= TRUE
;
695 if (r300
->screen
->caps
.is_r500
) {
696 r300
->fs
.size
= r500_get_fs_atom_size(r300
);
697 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 7;
698 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 3;
700 r300
->fs
.size
= r300_get_fs_atom_size(r300
);
701 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 5;
702 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 1;
706 /* Bind fragment shader state. */
707 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
709 struct r300_context
* r300
= r300_context(pipe
);
710 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
713 r300
->fs
.state
= NULL
;
718 r300_pick_fragment_shader(r300
);
719 r300_mark_fs_code_dirty(r300
);
721 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
723 if (r300
->vs_state
.state
&& r300_vertex_shader_setup_wpos(r300
)) {
724 r300
->vap_output_state
.dirty
= TRUE
;
728 /* Delete fragment shader state. */
729 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
731 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
732 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
737 rc_constants_destroy(&tmp
->code
.constants
);
740 FREE((void*)fs
->state
.tokens
);
744 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
745 const struct pipe_poly_stipple
* state
)
747 /* XXX no idea how to set this up, but not terribly important */
750 /* Create a new rasterizer state based on the CSO rasterizer state.
752 * This is a very large chunk of state, and covers most of the graphics
753 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
755 * In a not entirely unironic sidenote, this state has nearly nothing to do
756 * with the actual block on the Radeon called the rasterizer (RS). */
757 static void* r300_create_rs_state(struct pipe_context
* pipe
,
758 const struct pipe_rasterizer_state
* state
)
760 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
763 /* Copy rasterizer state for Draw. */
766 #ifdef PIPE_ARCH_LITTLE_ENDIAN
767 rs
->vap_control_status
= R300_VC_NO_SWAP
;
769 rs
->vap_control_status
= R300_VC_32BIT_SWAP
;
772 /* If no TCL engine is present, turn off the HW TCL. */
773 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
774 rs
->vap_control_status
|= R300_VAP_TCL_BYPASS
;
777 rs
->point_size
= pack_float_16_6x(state
->point_size
) |
778 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
780 rs
->line_control
= pack_float_16_6x(state
->line_width
) |
781 R300_GA_LINE_CNTL_END_TYPE_COMP
;
783 /* Enable polygon mode */
784 if (state
->fill_cw
!= PIPE_POLYGON_MODE_FILL
||
785 state
->fill_ccw
!= PIPE_POLYGON_MODE_FILL
) {
786 rs
->polygon_mode
= R300_GA_POLY_MODE_DUAL
;
789 /* Radeons don't think in "CW/CCW", they think in "front/back". */
790 if (state
->front_winding
== PIPE_WINDING_CW
) {
791 rs
->cull_mode
= R300_FRONT_FACE_CW
;
794 if (state
->offset_cw
) {
795 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
797 if (state
->offset_ccw
) {
798 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
802 if (rs
->polygon_mode
) {
804 r300_translate_polygon_mode_front(state
->fill_cw
);
806 r300_translate_polygon_mode_back(state
->fill_ccw
);
809 rs
->cull_mode
= R300_FRONT_FACE_CCW
;
812 if (state
->offset_ccw
) {
813 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
815 if (state
->offset_cw
) {
816 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
820 if (rs
->polygon_mode
) {
822 r300_translate_polygon_mode_front(state
->fill_ccw
);
824 r300_translate_polygon_mode_back(state
->fill_cw
);
827 if (state
->front_winding
& state
->cull_mode
) {
828 rs
->cull_mode
|= R300_CULL_FRONT
;
830 if (~(state
->front_winding
) & state
->cull_mode
) {
831 rs
->cull_mode
|= R300_CULL_BACK
;
834 if (rs
->polygon_offset_enable
) {
835 rs
->depth_offset
= state
->offset_units
;
836 rs
->depth_scale
= state
->offset_scale
;
839 if (state
->line_stipple_enable
) {
840 rs
->line_stipple_config
=
841 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
842 (fui((float)state
->line_stipple_factor
) &
843 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
844 /* XXX this might need to be scaled up */
845 rs
->line_stipple_value
= state
->line_stipple_pattern
;
848 if (state
->flatshade
) {
849 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
851 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
854 rs
->clip_rule
= state
->scissor
? 0xAAAA : 0xFFFF;
856 /* XXX Disable point sprites until we know what's wrong with them. */
857 rs
->rs
.sprite_coord_enable
= 0;
860 if (rs
->rs
.sprite_coord_enable
) {
861 rs
->stuffing_enable
= R300_GB_POINT_STUFF_ENABLE
;
862 for (i
= 0; i
< 8; i
++) {
863 if (state
->sprite_coord_enable
& (1 << i
))
864 rs
->stuffing_enable
|=
865 R300_GB_TEX_STR
<< (R300_GB_TEX0_SOURCE_SHIFT
+ (i
*2));
868 rs
->point_texcoord_left
= 0.0f
;
869 rs
->point_texcoord_right
= 1.0f
;
871 switch (state
->sprite_coord_mode
) {
872 case PIPE_SPRITE_COORD_UPPER_LEFT
:
873 rs
->point_texcoord_top
= 0.0f
;
874 rs
->point_texcoord_bottom
= 1.0f
;
876 case PIPE_SPRITE_COORD_LOWER_LEFT
:
877 rs
->point_texcoord_top
= 1.0f
;
878 rs
->point_texcoord_bottom
= 0.0f
;
886 /* Bind rasterizer state. */
887 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
889 struct r300_context
* r300
= r300_context(pipe
);
890 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
891 int last_sprite_coord_enable
= r300
->sprite_coord_enable
;
894 draw_flush(r300
->draw
);
895 draw_set_rasterizer_state(r300
->draw
, &rs
->rs
, state
);
899 r300
->polygon_offset_enabled
= rs
->rs
.offset_cw
|| rs
->rs
.offset_ccw
;
900 r300
->sprite_coord_enable
= rs
->rs
.sprite_coord_enable
;
902 r300
->polygon_offset_enabled
= FALSE
;
903 r300
->sprite_coord_enable
= 0;
906 UPDATE_STATE(state
, r300
->rs_state
);
907 r300
->rs_state
.size
= 26 + (r300
->polygon_offset_enabled
? 5 : 0);
909 if (last_sprite_coord_enable
!= r300
->sprite_coord_enable
) {
910 r300
->rs_block_state
.dirty
= TRUE
;
914 /* Free rasterizer state. */
915 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
921 r300_create_sampler_state(struct pipe_context
* pipe
,
922 const struct pipe_sampler_state
* state
)
924 struct r300_context
* r300
= r300_context(pipe
);
925 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
926 boolean is_r500
= r300
->screen
->caps
.is_r500
;
930 sampler
->state
= *state
;
933 (r300_translate_wrap(state
->wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
934 (r300_translate_wrap(state
->wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
935 (r300_translate_wrap(state
->wrap_r
) << R300_TX_WRAP_R_SHIFT
);
937 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
938 state
->mag_img_filter
,
939 state
->min_mip_filter
,
940 state
->max_anisotropy
> 0);
942 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
944 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
945 /* We must pass these to the merge function to clamp them properly. */
946 sampler
->min_lod
= MAX2((unsigned)state
->min_lod
, 0);
947 sampler
->max_lod
= MAX2((unsigned)ceilf(state
->max_lod
), 0);
949 lod_bias
= CLAMP((int)(state
->lod_bias
* 32), -(1 << 9), (1 << 9) - 1);
951 sampler
->filter1
|= lod_bias
<< R300_LOD_BIAS_SHIFT
;
953 /* This is very high quality anisotropic filtering for R5xx.
954 * It's good for benchmarking the performance of texturing but
955 * in practice we don't want to slow down the driver because it's
956 * a pretty good performance killer. Feel free to play with it. */
957 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
958 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
961 util_pack_color(state
->border_color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
962 sampler
->border_color
= uc
.ui
;
964 /* R500-specific fixups and optimizations */
965 if (r300
->screen
->caps
.is_r500
) {
966 sampler
->filter1
|= R500_BORDER_FIX
;
969 return (void*)sampler
;
972 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
976 struct r300_context
* r300
= r300_context(pipe
);
977 struct r300_textures_state
* state
=
978 (struct r300_textures_state
*)r300
->textures_state
.state
;
979 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
981 if (count
> tex_units
) {
985 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
986 state
->sampler_state_count
= count
;
988 r300
->textures_state
.dirty
= TRUE
;
991 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
997 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
1002 static void r300_set_fragment_sampler_views(struct pipe_context
* pipe
,
1004 struct pipe_sampler_view
** views
)
1006 struct r300_context
* r300
= r300_context(pipe
);
1007 struct r300_textures_state
* state
=
1008 (struct r300_textures_state
*)r300
->textures_state
.state
;
1009 struct r300_texture
*texture
;
1011 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
1012 boolean dirty_tex
= FALSE
;
1014 if (count
> tex_units
) {
1018 for (i
= 0; i
< count
; i
++) {
1019 if (&state
->sampler_views
[i
]->base
!= views
[i
]) {
1020 pipe_sampler_view_reference(
1021 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1028 /* A new sampler view (= texture)... */
1031 /* Set the texrect factor in the fragment shader.
1032 * Needed for RECT and NPOT fallback. */
1033 texture
= r300_texture(views
[i
]->texture
);
1034 if (texture
->uses_pitch
) {
1035 r300
->fs_rc_constant_state
.dirty
= TRUE
;
1040 for (i
= count
; i
< tex_units
; i
++) {
1041 if (state
->sampler_views
[i
]) {
1042 pipe_sampler_view_reference(
1043 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1048 state
->sampler_view_count
= count
;
1050 r300
->textures_state
.dirty
= TRUE
;
1053 r300
->texture_cache_inval
.dirty
= TRUE
;
1057 static struct pipe_sampler_view
*
1058 r300_create_sampler_view(struct pipe_context
*pipe
,
1059 struct pipe_resource
*texture
,
1060 const struct pipe_sampler_view
*templ
)
1062 struct r300_sampler_view
*view
= CALLOC_STRUCT(r300_sampler_view
);
1063 struct r300_texture
*tex
= r300_texture(texture
);
1064 unsigned char swizzle
[4];
1067 view
->base
= *templ
;
1068 view
->base
.reference
.count
= 1;
1069 view
->base
.context
= pipe
;
1070 view
->base
.texture
= NULL
;
1071 pipe_resource_reference(&view
->base
.texture
, texture
);
1073 swizzle
[0] = templ
->swizzle_r
;
1074 swizzle
[1] = templ
->swizzle_g
;
1075 swizzle
[2] = templ
->swizzle_b
;
1076 swizzle
[3] = templ
->swizzle_a
;
1078 /* XXX Enable swizzles when they become supported. Now we get RGBA
1079 * everywhere. And do testing! */
1080 view
->format
= tex
->tx_format
;
1081 view
->format
.format1
|= r300_translate_texformat(templ
->format
,
1083 if (r300_screen(pipe
->screen
)->caps
.is_r500
) {
1084 view
->format
.format2
|= r500_tx_format_msb_bit(templ
->format
);
1088 return (struct pipe_sampler_view
*)view
;
1092 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1093 struct pipe_sampler_view
*view
)
1095 pipe_resource_reference(&view
->texture
, NULL
);
1099 static void r300_set_scissor_state(struct pipe_context
* pipe
,
1100 const struct pipe_scissor_state
* state
)
1102 struct r300_context
* r300
= r300_context(pipe
);
1104 memcpy(r300
->scissor_state
.state
, state
,
1105 sizeof(struct pipe_scissor_state
));
1107 r300
->scissor_state
.dirty
= TRUE
;
1110 static void r300_set_viewport_state(struct pipe_context
* pipe
,
1111 const struct pipe_viewport_state
* state
)
1113 struct r300_context
* r300
= r300_context(pipe
);
1114 struct r300_viewport_state
* viewport
=
1115 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1117 r300
->viewport
= *state
;
1119 /* Do the transform in HW. */
1120 viewport
->vte_control
= R300_VTX_W0_FMT
;
1122 if (state
->scale
[0] != 1.0f
) {
1123 viewport
->xscale
= state
->scale
[0];
1124 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1126 if (state
->scale
[1] != 1.0f
) {
1127 viewport
->yscale
= state
->scale
[1];
1128 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1130 if (state
->scale
[2] != 1.0f
) {
1131 viewport
->zscale
= state
->scale
[2];
1132 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1134 if (state
->translate
[0] != 0.0f
) {
1135 viewport
->xoffset
= state
->translate
[0];
1136 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1138 if (state
->translate
[1] != 0.0f
) {
1139 viewport
->yoffset
= state
->translate
[1];
1140 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1142 if (state
->translate
[2] != 0.0f
) {
1143 viewport
->zoffset
= state
->translate
[2];
1144 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1147 r300
->viewport_state
.dirty
= TRUE
;
1148 if (r300
->fs
.state
&& r300_fs(r300
)->shader
->inputs
.wpos
!= ATTR_UNUSED
) {
1149 r300
->fs_rc_constant_state
.dirty
= TRUE
;
1153 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
1155 const struct pipe_vertex_buffer
* buffers
)
1157 struct r300_context
* r300
= r300_context(pipe
);
1158 struct pipe_vertex_buffer
*vbo
;
1159 unsigned i
, max_index
= (1 << 24) - 1;
1160 boolean any_user_buffer
= FALSE
;
1162 if (count
== r300
->vertex_buffer_count
&&
1163 memcmp(r300
->vertex_buffer
, buffers
,
1164 sizeof(struct pipe_vertex_buffer
) * count
) == 0) {
1168 /* Check if the stride is aligned to the size of DWORD. */
1169 for (i
= 0; i
< count
; i
++) {
1170 if (buffers
[i
].buffer
) {
1171 if (buffers
[i
].stride
% 4 != 0) {
1172 // XXX Shouldn't we align the buffer?
1173 fprintf(stderr
, "r300: set_vertex_buffers: "
1174 "Unaligned buffer stride %i isn't supported.\n",
1181 for (i
= 0; i
< count
; i
++) {
1182 /* Why, yes, I AM casting away constness. How did you know? */
1183 vbo
= (struct pipe_vertex_buffer
*)&buffers
[i
];
1185 /* Reference our buffer. */
1186 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, vbo
->buffer
);
1188 /* Skip NULL buffers */
1189 if (!buffers
[i
].buffer
) {
1193 if (r300_buffer_is_user_buffer(vbo
->buffer
)) {
1194 any_user_buffer
= TRUE
;
1197 if (vbo
->max_index
== ~0) {
1198 /* if no VBO stride then only one vertex value so max index is 1 */
1199 /* should think about converting to VS constants like svga does */
1204 (vbo
->buffer
->width0
- vbo
->buffer_offset
) / vbo
->stride
;
1207 max_index
= MIN2(vbo
->max_index
, max_index
);
1210 for (; i
< r300
->vertex_buffer_count
; i
++) {
1211 /* Dereference any old buffers. */
1212 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, NULL
);
1215 memcpy(r300
->vertex_buffer
, buffers
,
1216 sizeof(struct pipe_vertex_buffer
) * count
);
1218 r300
->vertex_buffer_count
= count
;
1219 r300
->vertex_buffer_max_index
= max_index
;
1220 r300
->any_user_vbs
= any_user_buffer
;
1223 draw_flush(r300
->draw
);
1224 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
1228 /* Update the PSC tables. */
1229 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1231 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1232 uint16_t type
, swizzle
;
1233 enum pipe_format format
;
1236 if (velems
->count
> 16) {
1237 fprintf(stderr
, "r300: More than 16 vertex elements are not supported,"
1238 " requested %i, using 16.\n", velems
->count
);
1242 /* Vertex shaders have no semantics on their inputs,
1243 * so PSC should just route stuff based on the vertex elements,
1244 * and not on attrib information. */
1245 for (i
= 0; i
< velems
->count
; i
++) {
1246 format
= velems
->velem
[i
].src_format
;
1248 type
= r300_translate_vertex_data_type(format
) |
1249 (i
<< R300_DST_VEC_LOC_SHIFT
);
1250 swizzle
= r300_translate_vertex_data_swizzle(format
);
1253 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1254 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1256 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1257 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1261 /* Set the last vector in the PSC. */
1265 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1266 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1268 vstream
->count
= (i
>> 1) + 1;
1271 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1273 const struct pipe_vertex_element
* attribs
)
1275 struct r300_vertex_element_state
*velems
;
1277 enum pipe_format
*format
;
1279 assert(count
<= PIPE_MAX_ATTRIBS
);
1280 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1281 if (velems
!= NULL
) {
1282 velems
->count
= count
;
1283 memcpy(velems
->velem
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
1285 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1286 r300_vertex_psc(velems
);
1288 /* Check if the format is aligned to the size of DWORD.
1289 * We only care about the blocksizes of the formats since
1290 * swizzles are already set up. */
1291 for (i
= 0; i
< count
; i
++) {
1292 format
= &velems
->velem
[i
].src_format
;
1294 /* Replace some formats with their aligned counterparts,
1295 * this is OK because we check for aligned strides too. */
1297 case PIPE_FORMAT_R8G8B8_UNORM
:
1298 *format
= PIPE_FORMAT_R8G8B8X8_UNORM
;
1300 case PIPE_FORMAT_R8G8B8_SNORM
:
1301 *format
= PIPE_FORMAT_R8G8B8A8_SNORM
;
1303 case PIPE_FORMAT_R8G8B8_USCALED
:
1304 *format
= PIPE_FORMAT_R8G8B8A8_USCALED
;
1306 case PIPE_FORMAT_R8G8B8_SSCALED
:
1307 *format
= PIPE_FORMAT_R8G8B8A8_SSCALED
;
1309 case PIPE_FORMAT_R16G16B16_UNORM
:
1310 *format
= PIPE_FORMAT_R16G16B16A16_UNORM
;
1312 case PIPE_FORMAT_R16G16B16_SNORM
:
1313 *format
= PIPE_FORMAT_R16G16B16A16_SNORM
;
1315 case PIPE_FORMAT_R16G16B16_USCALED
:
1316 *format
= PIPE_FORMAT_R16G16B16A16_USCALED
;
1318 case PIPE_FORMAT_R16G16B16_SSCALED
:
1319 *format
= PIPE_FORMAT_R16G16B16A16_SSCALED
;
1324 size
= util_format_get_blocksize(*format
);
1326 if (size
% 4 != 0) {
1327 /* XXX Shouldn't we align the format? */
1328 fprintf(stderr
, "r300_create_vertex_elements_state: "
1329 "Unaligned format %s:%i isn't supported\n",
1330 util_format_name(*format
), size
);
1341 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1344 struct r300_context
*r300
= r300_context(pipe
);
1345 struct r300_vertex_element_state
*velems
= state
;
1347 if (velems
== NULL
) {
1351 r300
->velems
= velems
;
1354 draw_flush(r300
->draw
);
1355 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1358 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1359 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1362 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1367 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1368 const struct pipe_shader_state
* shader
)
1370 struct r300_context
* r300
= r300_context(pipe
);
1372 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1374 /* Copy state directly into shader. */
1375 vs
->state
= *shader
;
1376 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
1378 if (r300
->screen
->caps
.has_tcl
) {
1379 r300_translate_vertex_shader(r300
, vs
, vs
->state
.tokens
);
1381 vs
->draw_vs
= draw_create_vertex_shader(r300
->draw
, shader
);
1387 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1389 struct r300_context
* r300
= r300_context(pipe
);
1390 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1393 r300
->vs_state
.state
= NULL
;
1396 if (vs
== r300
->vs_state
.state
) {
1399 r300
->vs_state
.state
= vs
;
1401 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
1402 if (r300
->fs
.state
) {
1403 r300_vertex_shader_setup_wpos(r300
);
1405 memcpy(r300
->vap_output_state
.state
, &vs
->vap_out
,
1406 sizeof(struct r300_vap_output_state
));
1407 r300
->vap_output_state
.dirty
= TRUE
;
1409 /* The majority of the RS block bits is dependent on the vertex shader. */
1410 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
1412 if (r300
->screen
->caps
.has_tcl
) {
1413 r300
->vs_state
.dirty
= TRUE
;
1414 r300
->vs_state
.size
=
1415 vs
->code
.length
+ 9 +
1416 (vs
->immediates_count
? vs
->immediates_count
* 4 + 3 : 0);
1418 if (vs
->externals_count
) {
1419 r300
->vs_constants
.dirty
= TRUE
;
1420 r300
->vs_constants
.size
= vs
->externals_count
* 4 + 3;
1422 r300
->vs_constants
.size
= 0;
1425 r300
->pvs_flush
.dirty
= TRUE
;
1427 draw_flush(r300
->draw
);
1428 draw_bind_vertex_shader(r300
->draw
,
1429 (struct draw_vertex_shader
*)vs
->draw_vs
);
1433 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1435 struct r300_context
* r300
= r300_context(pipe
);
1436 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1438 if (r300
->screen
->caps
.has_tcl
) {
1439 rc_constants_destroy(&vs
->code
.constants
);
1441 draw_delete_vertex_shader(r300
->draw
,
1442 (struct draw_vertex_shader
*)vs
->draw_vs
);
1445 FREE((void*)vs
->state
.tokens
);
1449 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1450 uint shader
, uint index
,
1451 struct pipe_resource
*buf
)
1453 struct r300_context
* r300
= r300_context(pipe
);
1454 struct r300_constant_buffer
*cbuf
;
1455 struct pipe_transfer
*tr
;
1460 case PIPE_SHADER_VERTEX
:
1461 cbuf
= (struct r300_constant_buffer
*)r300
->vs_constants
.state
;
1464 case PIPE_SHADER_FRAGMENT
:
1465 cbuf
= (struct r300_constant_buffer
*)r300
->fs_constants
.state
;
1466 if (r300
->screen
->caps
.is_r500
) {
1477 if (buf
== NULL
|| buf
->width0
== 0 ||
1478 (mapped
= pipe_buffer_map(pipe
, buf
, PIPE_TRANSFER_READ
, &tr
)) == NULL
)
1484 assert((buf
->width0
% 4 * sizeof(float)) == 0);
1486 /* Check the size of the constant buffer. */
1487 /* XXX Subtract immediates and RC_STATE_* variables. */
1488 if (buf
->width0
> (sizeof(float) * 4 * max_size
)) {
1489 fprintf(stderr
, "r300: Max size of the constant buffer is "
1490 "%i*4 floats.\n", max_size
);
1494 memcpy(cbuf
->constants
, mapped
, buf
->width0
);
1495 cbuf
->count
= buf
->width0
/ (4 * sizeof(float));
1496 pipe_buffer_unmap(pipe
, buf
, tr
);
1498 if (shader
== PIPE_SHADER_VERTEX
) {
1499 if (r300
->screen
->caps
.has_tcl
) {
1500 if (r300
->vs_constants
.size
) {
1501 r300
->vs_constants
.dirty
= TRUE
;
1503 r300
->pvs_flush
.dirty
= TRUE
;
1504 } else if (r300
->draw
) {
1505 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
1509 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
1510 r300
->fs_constants
.dirty
= TRUE
;
1514 void r300_init_state_functions(struct r300_context
* r300
)
1516 r300
->context
.create_blend_state
= r300_create_blend_state
;
1517 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1518 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1520 r300
->context
.set_blend_color
= r300_set_blend_color
;
1522 r300
->context
.set_clip_state
= r300_set_clip_state
;
1524 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1526 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1527 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1528 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1530 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
1532 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1534 r300
->context
.create_fs_state
= r300_create_fs_state
;
1535 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1536 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1538 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1540 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1541 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1542 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1544 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1545 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1546 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1547 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1549 r300
->context
.set_fragment_sampler_views
= r300_set_fragment_sampler_views
;
1550 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
1551 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
1553 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1555 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1557 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1559 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
1560 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
1561 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
1563 r300
->context
.create_vs_state
= r300_create_vs_state
;
1564 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1565 r300
->context
.delete_vs_state
= r300_delete_vs_state
;