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
;
408 r300_set_sample_mask(struct pipe_context
*pipe
,
409 unsigned sample_mask
)
414 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
416 * This contains the depth buffer, stencil buffer, alpha test, and such.
417 * On the Radeon, depth and stencil buffer setup are intertwined, which is
418 * the reason for some of the strange-looking assignments across registers. */
420 r300_create_dsa_state(struct pipe_context
* pipe
,
421 const struct pipe_depth_stencil_alpha_state
* state
)
423 struct r300_capabilities
*caps
= &r300_screen(pipe
->screen
)->caps
;
424 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
426 /* Depth test setup. */
427 if (state
->depth
.enabled
) {
428 dsa
->z_buffer_control
|= R300_Z_ENABLE
;
430 if (state
->depth
.writemask
) {
431 dsa
->z_buffer_control
|= R300_Z_WRITE_ENABLE
;
434 dsa
->z_stencil_control
|=
435 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
439 /* Stencil buffer setup. */
440 if (state
->stencil
[0].enabled
) {
441 dsa
->z_buffer_control
|= R300_STENCIL_ENABLE
;
442 dsa
->z_stencil_control
|=
443 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
444 R300_S_FRONT_FUNC_SHIFT
) |
445 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
446 R300_S_FRONT_SFAIL_OP_SHIFT
) |
447 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
448 R300_S_FRONT_ZPASS_OP_SHIFT
) |
449 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
450 R300_S_FRONT_ZFAIL_OP_SHIFT
);
452 dsa
->stencil_ref_mask
=
453 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
454 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
456 if (state
->stencil
[1].enabled
) {
457 dsa
->two_sided
= TRUE
;
459 dsa
->z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
460 dsa
->z_stencil_control
|=
461 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
462 R300_S_BACK_FUNC_SHIFT
) |
463 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
464 R300_S_BACK_SFAIL_OP_SHIFT
) |
465 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
466 R300_S_BACK_ZPASS_OP_SHIFT
) |
467 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
468 R300_S_BACK_ZFAIL_OP_SHIFT
);
470 dsa
->stencil_ref_bf
=
471 (state
->stencil
[1].valuemask
<< R300_STENCILMASK_SHIFT
) |
472 (state
->stencil
[1].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
475 dsa
->z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
477 dsa
->stencil_ref_bf_fallback
=
478 (state
->stencil
[0].valuemask
!= state
->stencil
[1].valuemask
||
479 state
->stencil
[0].writemask
!= state
->stencil
[1].writemask
);
484 /* Alpha test setup. */
485 if (state
->alpha
.enabled
) {
486 dsa
->alpha_function
=
487 r300_translate_alpha_function(state
->alpha
.func
) |
488 R300_FG_ALPHA_FUNC_ENABLE
;
490 /* We could use 10bit alpha ref but who needs that? */
491 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
494 dsa
->alpha_function
|= R500_FG_ALPHA_FUNC_8BIT
;
500 static void r300_update_stencil_ref_fallback_status(struct r300_context
*r300
)
502 struct r300_dsa_state
*dsa
= (struct r300_dsa_state
*)r300
->dsa_state
.state
;
504 if (r300
->screen
->caps
.is_r500
) {
508 r300
->stencil_ref_bf_fallback
=
509 dsa
->stencil_ref_bf_fallback
||
511 r300
->stencil_ref
.ref_value
[0] != r300
->stencil_ref
.ref_value
[1]);
514 /* Bind DSA state. */
515 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
518 struct r300_context
* r300
= r300_context(pipe
);
524 UPDATE_STATE(state
, r300
->dsa_state
);
526 r300_update_stencil_ref_fallback_status(r300
);
529 /* Free DSA state. */
530 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
536 static void r300_set_stencil_ref(struct pipe_context
* pipe
,
537 const struct pipe_stencil_ref
* sr
)
539 struct r300_context
* r300
= r300_context(pipe
);
541 r300
->stencil_ref
= *sr
;
542 r300
->dsa_state
.dirty
= TRUE
;
544 r300_update_stencil_ref_fallback_status(r300
);
547 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
548 static void r300_fb_set_tiling_flags(struct r300_context
*r300
,
549 const struct pipe_framebuffer_state
*old_state
,
550 const struct pipe_framebuffer_state
*new_state
)
552 struct r300_texture
*tex
;
555 /* Set tiling flags for new surfaces. */
556 for (i
= 0; i
< new_state
->nr_cbufs
; i
++) {
557 tex
= r300_texture(new_state
->cbufs
[i
]->texture
);
558 level
= new_state
->cbufs
[i
]->level
;
560 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
563 tex
->mip_macrotile
[level
]);
565 if (new_state
->zsbuf
) {
566 tex
= r300_texture(new_state
->zsbuf
->texture
);
567 level
= new_state
->zsbuf
->level
;
569 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
572 tex
->mip_macrotile
[level
]);
577 r300_set_framebuffer_state(struct pipe_context
* pipe
,
578 const struct pipe_framebuffer_state
* state
)
580 struct r300_context
* r300
= r300_context(pipe
);
581 struct pipe_framebuffer_state
*old_state
= r300
->fb_state
.state
;
582 unsigned max_width
, max_height
;
583 uint32_t zbuffer_bpp
= 0;
585 if (state
->nr_cbufs
> 4) {
586 fprintf(stderr
, "r300: Implementation error: Too many MRTs in %s, "
587 "refusing to bind framebuffer state!\n", __FUNCTION__
);
591 if (r300
->screen
->caps
.is_r500
) {
592 max_width
= max_height
= 4096;
593 } else if (r300
->screen
->caps
.is_r400
) {
594 max_width
= max_height
= 4021;
596 max_width
= max_height
= 2560;
599 if (state
->width
> max_width
|| state
->height
> max_height
) {
600 fprintf(stderr
, "r300: Implementation error: Render targets are too "
601 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
606 draw_flush(r300
->draw
);
609 r300
->fb_state
.dirty
= TRUE
;
611 /* If nr_cbufs is changed from zero to non-zero or vice versa... */
612 if (!!old_state
->nr_cbufs
!= !!state
->nr_cbufs
) {
613 r300
->blend_state
.dirty
= TRUE
;
615 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
616 if (!!old_state
->zsbuf
!= !!state
->zsbuf
) {
617 r300
->dsa_state
.dirty
= TRUE
;
620 /* The tiling flags are dependent on the surface miplevel, unfortunately. */
621 r300_fb_set_tiling_flags(r300
, r300
->fb_state
.state
, state
);
623 memcpy(r300
->fb_state
.state
, state
, sizeof(struct pipe_framebuffer_state
));
625 r300
->fb_state
.size
= (10 * state
->nr_cbufs
) + (2 * (4 - state
->nr_cbufs
)) +
626 (state
->zsbuf
? 10 : 0) + 11;
628 /* Polygon offset depends on the zbuffer bit depth. */
629 if (state
->zsbuf
&& r300
->polygon_offset_enabled
) {
630 switch (util_format_get_blocksize(state
->zsbuf
->texture
->format
)) {
639 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
640 r300
->zbuffer_bpp
= zbuffer_bpp
;
641 r300
->rs_state
.dirty
= TRUE
;
646 /* Create fragment shader state. */
647 static void* r300_create_fs_state(struct pipe_context
* pipe
,
648 const struct pipe_shader_state
* shader
)
650 struct r300_fragment_shader
* fs
= NULL
;
652 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
654 /* Copy state directly into shader. */
656 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
661 void r300_mark_fs_code_dirty(struct r300_context
*r300
)
663 struct r300_fragment_shader
* fs
= r300_fs(r300
);
665 r300
->fs
.dirty
= TRUE
;
666 r300
->fs_rc_constant_state
.dirty
= TRUE
;
667 r300
->fs_constants
.dirty
= TRUE
;
669 if (r300
->screen
->caps
.is_r500
) {
670 r300
->fs
.size
= r500_get_fs_atom_size(r300
);
671 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 7;
672 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 3;
674 r300
->fs
.size
= r300_get_fs_atom_size(r300
);
675 r300
->fs_rc_constant_state
.size
= fs
->shader
->rc_state_count
* 5;
676 r300
->fs_constants
.size
= fs
->shader
->externals_count
* 4 + 1;
680 /* Bind fragment shader state. */
681 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
683 struct r300_context
* r300
= r300_context(pipe
);
684 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
687 r300
->fs
.state
= NULL
;
692 r300_pick_fragment_shader(r300
);
693 r300_mark_fs_code_dirty(r300
);
695 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
698 /* Delete fragment shader state. */
699 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
701 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
702 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
707 rc_constants_destroy(&tmp
->code
.constants
);
710 FREE((void*)fs
->state
.tokens
);
714 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
715 const struct pipe_poly_stipple
* state
)
717 /* XXX no idea how to set this up, but not terribly important */
720 /* Create a new rasterizer state based on the CSO rasterizer state.
722 * This is a very large chunk of state, and covers most of the graphics
723 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
725 * In a not entirely unironic sidenote, this state has nearly nothing to do
726 * with the actual block on the Radeon called the rasterizer (RS). */
727 static void* r300_create_rs_state(struct pipe_context
* pipe
,
728 const struct pipe_rasterizer_state
* state
)
730 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
733 /* Copy rasterizer state for Draw. */
736 #ifdef PIPE_ARCH_LITTLE_ENDIAN
737 rs
->vap_control_status
= R300_VC_NO_SWAP
;
739 rs
->vap_control_status
= R300_VC_32BIT_SWAP
;
742 /* If no TCL engine is present, turn off the HW TCL. */
743 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
744 rs
->vap_control_status
|= R300_VAP_TCL_BYPASS
;
747 rs
->point_size
= pack_float_16_6x(state
->point_size
) |
748 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
750 rs
->line_control
= pack_float_16_6x(state
->line_width
) |
751 R300_GA_LINE_CNTL_END_TYPE_COMP
;
753 /* Enable polygon mode */
754 if (state
->fill_cw
!= PIPE_POLYGON_MODE_FILL
||
755 state
->fill_ccw
!= PIPE_POLYGON_MODE_FILL
) {
756 rs
->polygon_mode
= R300_GA_POLY_MODE_DUAL
;
759 /* Radeons don't think in "CW/CCW", they think in "front/back". */
760 if (state
->front_winding
== PIPE_WINDING_CW
) {
761 rs
->cull_mode
= R300_FRONT_FACE_CW
;
764 if (state
->offset_cw
) {
765 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
767 if (state
->offset_ccw
) {
768 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
772 if (rs
->polygon_mode
) {
774 r300_translate_polygon_mode_front(state
->fill_cw
);
776 r300_translate_polygon_mode_back(state
->fill_ccw
);
779 rs
->cull_mode
= R300_FRONT_FACE_CCW
;
782 if (state
->offset_ccw
) {
783 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
785 if (state
->offset_cw
) {
786 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
790 if (rs
->polygon_mode
) {
792 r300_translate_polygon_mode_front(state
->fill_ccw
);
794 r300_translate_polygon_mode_back(state
->fill_cw
);
797 if (state
->front_winding
& state
->cull_mode
) {
798 rs
->cull_mode
|= R300_CULL_FRONT
;
800 if (~(state
->front_winding
) & state
->cull_mode
) {
801 rs
->cull_mode
|= R300_CULL_BACK
;
804 if (rs
->polygon_offset_enable
) {
805 rs
->depth_offset
= state
->offset_units
;
806 rs
->depth_scale
= state
->offset_scale
;
809 if (state
->line_stipple_enable
) {
810 rs
->line_stipple_config
=
811 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
812 (fui((float)state
->line_stipple_factor
) &
813 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
814 /* XXX this might need to be scaled up */
815 rs
->line_stipple_value
= state
->line_stipple_pattern
;
818 if (state
->flatshade
) {
819 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
821 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
824 rs
->clip_rule
= state
->scissor
? 0xAAAA : 0xFFFF;
827 if (state
->sprite_coord_enable
) {
828 rs
->stuffing_enable
= R300_GB_POINT_STUFF_ENABLE
;
829 for (i
= 0; i
< 8; i
++) {
830 if (state
->sprite_coord_enable
& (1 << i
))
831 rs
->stuffing_enable
|=
832 R300_GB_TEX_STR
<< (R300_GB_TEX0_SOURCE_SHIFT
+ (i
*2));
835 rs
->point_texcoord_left
= 0.0f
;
836 rs
->point_texcoord_right
= 1.0f
;
838 switch (state
->sprite_coord_mode
) {
839 case PIPE_SPRITE_COORD_UPPER_LEFT
:
840 rs
->point_texcoord_top
= 0.0f
;
841 rs
->point_texcoord_bottom
= 1.0f
;
843 case PIPE_SPRITE_COORD_LOWER_LEFT
:
844 rs
->point_texcoord_top
= 1.0f
;
845 rs
->point_texcoord_bottom
= 0.0f
;
853 /* Bind rasterizer state. */
854 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
856 struct r300_context
* r300
= r300_context(pipe
);
857 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
858 int last_sprite_coord_enable
= r300
->sprite_coord_enable
;
861 draw_flush(r300
->draw
);
862 draw_set_rasterizer_state(r300
->draw
, &rs
->rs
, state
);
866 r300
->polygon_offset_enabled
= rs
->rs
.offset_cw
|| rs
->rs
.offset_ccw
;
867 r300
->sprite_coord_enable
= rs
->rs
.sprite_coord_enable
;
869 r300
->polygon_offset_enabled
= FALSE
;
870 r300
->sprite_coord_enable
= 0;
873 UPDATE_STATE(state
, r300
->rs_state
);
874 r300
->rs_state
.size
= 26 + (r300
->polygon_offset_enabled
? 5 : 0);
876 if (last_sprite_coord_enable
!= r300
->sprite_coord_enable
) {
877 r300
->rs_block_state
.dirty
= TRUE
;
881 /* Free rasterizer state. */
882 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
888 r300_create_sampler_state(struct pipe_context
* pipe
,
889 const struct pipe_sampler_state
* state
)
891 struct r300_context
* r300
= r300_context(pipe
);
892 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
893 boolean is_r500
= r300
->screen
->caps
.is_r500
;
897 sampler
->state
= *state
;
900 (r300_translate_wrap(state
->wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
901 (r300_translate_wrap(state
->wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
902 (r300_translate_wrap(state
->wrap_r
) << R300_TX_WRAP_R_SHIFT
);
904 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
905 state
->mag_img_filter
,
906 state
->min_mip_filter
,
907 state
->max_anisotropy
> 0);
909 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
911 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
912 /* We must pass these to the merge function to clamp them properly. */
913 sampler
->min_lod
= MAX2((unsigned)state
->min_lod
, 0);
914 sampler
->max_lod
= MAX2((unsigned)ceilf(state
->max_lod
), 0);
916 lod_bias
= CLAMP((int)(state
->lod_bias
* 32), -(1 << 9), (1 << 9) - 1);
918 sampler
->filter1
|= lod_bias
<< R300_LOD_BIAS_SHIFT
;
920 /* This is very high quality anisotropic filtering for R5xx.
921 * It's good for benchmarking the performance of texturing but
922 * in practice we don't want to slow down the driver because it's
923 * a pretty good performance killer. Feel free to play with it. */
924 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
925 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
928 util_pack_color(state
->border_color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
929 sampler
->border_color
= uc
.ui
;
931 /* R500-specific fixups and optimizations */
932 if (r300
->screen
->caps
.is_r500
) {
933 sampler
->filter1
|= R500_BORDER_FIX
;
936 return (void*)sampler
;
939 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
943 struct r300_context
* r300
= r300_context(pipe
);
944 struct r300_textures_state
* state
=
945 (struct r300_textures_state
*)r300
->textures_state
.state
;
946 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
948 if (count
> tex_units
) {
952 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
953 state
->sampler_state_count
= count
;
955 r300
->textures_state
.dirty
= TRUE
;
958 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
964 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
969 static void r300_set_fragment_sampler_views(struct pipe_context
* pipe
,
971 struct pipe_sampler_view
** views
)
973 struct r300_context
* r300
= r300_context(pipe
);
974 struct r300_textures_state
* state
=
975 (struct r300_textures_state
*)r300
->textures_state
.state
;
976 struct r300_texture
*texture
;
978 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
979 boolean dirty_tex
= FALSE
;
981 if (count
> tex_units
) {
985 for (i
= 0; i
< count
; i
++) {
986 if (&state
->sampler_views
[i
]->base
!= views
[i
]) {
987 pipe_sampler_view_reference(
988 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
995 /* A new sampler view (= texture)... */
998 /* Set the texrect factor in the fragment shader.
999 * Needed for RECT and NPOT fallback. */
1000 texture
= r300_texture(views
[i
]->texture
);
1001 if (texture
->uses_pitch
) {
1002 r300
->fs_rc_constant_state
.dirty
= TRUE
;
1007 for (i
= count
; i
< tex_units
; i
++) {
1008 if (state
->sampler_views
[i
]) {
1009 pipe_sampler_view_reference(
1010 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1015 state
->sampler_view_count
= count
;
1017 r300
->textures_state
.dirty
= TRUE
;
1020 r300
->texture_cache_inval
.dirty
= TRUE
;
1024 static struct pipe_sampler_view
*
1025 r300_create_sampler_view(struct pipe_context
*pipe
,
1026 struct pipe_resource
*texture
,
1027 const struct pipe_sampler_view
*templ
)
1029 struct r300_sampler_view
*view
= CALLOC_STRUCT(r300_sampler_view
);
1030 struct r300_texture
*tex
= r300_texture(texture
);
1031 unsigned char swizzle
[4];
1034 view
->base
= *templ
;
1035 view
->base
.reference
.count
= 1;
1036 view
->base
.context
= pipe
;
1037 view
->base
.texture
= NULL
;
1038 pipe_resource_reference(&view
->base
.texture
, texture
);
1040 swizzle
[0] = templ
->swizzle_r
;
1041 swizzle
[1] = templ
->swizzle_g
;
1042 swizzle
[2] = templ
->swizzle_b
;
1043 swizzle
[3] = templ
->swizzle_a
;
1045 view
->format
= tex
->tx_format
;
1046 view
->format
.format1
|= r300_translate_texformat(templ
->format
,
1048 if (r300_screen(pipe
->screen
)->caps
.is_r500
) {
1049 view
->format
.format2
|= r500_tx_format_msb_bit(templ
->format
);
1053 return (struct pipe_sampler_view
*)view
;
1057 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1058 struct pipe_sampler_view
*view
)
1060 pipe_resource_reference(&view
->texture
, NULL
);
1064 static void r300_set_scissor_state(struct pipe_context
* pipe
,
1065 const struct pipe_scissor_state
* state
)
1067 struct r300_context
* r300
= r300_context(pipe
);
1069 memcpy(r300
->scissor_state
.state
, state
,
1070 sizeof(struct pipe_scissor_state
));
1072 r300
->scissor_state
.dirty
= TRUE
;
1075 static void r300_set_viewport_state(struct pipe_context
* pipe
,
1076 const struct pipe_viewport_state
* state
)
1078 struct r300_context
* r300
= r300_context(pipe
);
1079 struct r300_viewport_state
* viewport
=
1080 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1082 r300
->viewport
= *state
;
1084 /* Do the transform in HW. */
1085 viewport
->vte_control
= R300_VTX_W0_FMT
;
1087 if (state
->scale
[0] != 1.0f
) {
1088 viewport
->xscale
= state
->scale
[0];
1089 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1091 if (state
->scale
[1] != 1.0f
) {
1092 viewport
->yscale
= state
->scale
[1];
1093 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1095 if (state
->scale
[2] != 1.0f
) {
1096 viewport
->zscale
= state
->scale
[2];
1097 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1099 if (state
->translate
[0] != 0.0f
) {
1100 viewport
->xoffset
= state
->translate
[0];
1101 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1103 if (state
->translate
[1] != 0.0f
) {
1104 viewport
->yoffset
= state
->translate
[1];
1105 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1107 if (state
->translate
[2] != 0.0f
) {
1108 viewport
->zoffset
= state
->translate
[2];
1109 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1112 r300
->viewport_state
.dirty
= TRUE
;
1113 if (r300
->fs
.state
&& r300_fs(r300
)->shader
->inputs
.wpos
!= ATTR_UNUSED
) {
1114 r300
->fs_rc_constant_state
.dirty
= TRUE
;
1118 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
1120 const struct pipe_vertex_buffer
* buffers
)
1122 struct r300_context
* r300
= r300_context(pipe
);
1123 struct pipe_vertex_buffer
*vbo
;
1124 unsigned i
, max_index
= (1 << 24) - 1;
1125 boolean any_user_buffer
= FALSE
;
1127 if (count
== r300
->vertex_buffer_count
&&
1128 memcmp(r300
->vertex_buffer
, buffers
,
1129 sizeof(struct pipe_vertex_buffer
) * count
) == 0) {
1133 /* Check if the stride is aligned to the size of DWORD. */
1134 for (i
= 0; i
< count
; i
++) {
1135 if (buffers
[i
].buffer
) {
1136 if (buffers
[i
].stride
% 4 != 0) {
1137 // XXX Shouldn't we align the buffer?
1138 fprintf(stderr
, "r300: set_vertex_buffers: "
1139 "Unaligned buffer stride %i isn't supported.\n",
1146 for (i
= 0; i
< count
; i
++) {
1147 /* Why, yes, I AM casting away constness. How did you know? */
1148 vbo
= (struct pipe_vertex_buffer
*)&buffers
[i
];
1150 /* Reference our buffer. */
1151 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, vbo
->buffer
);
1153 /* Skip NULL buffers */
1154 if (!buffers
[i
].buffer
) {
1158 if (r300_buffer_is_user_buffer(vbo
->buffer
)) {
1159 any_user_buffer
= TRUE
;
1162 if (vbo
->max_index
== ~0) {
1163 /* if no VBO stride then only one vertex value so max index is 1 */
1164 /* should think about converting to VS constants like svga does */
1169 (vbo
->buffer
->width0
- vbo
->buffer_offset
) / vbo
->stride
;
1172 max_index
= MIN2(vbo
->max_index
, max_index
);
1175 for (; i
< r300
->vertex_buffer_count
; i
++) {
1176 /* Dereference any old buffers. */
1177 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, NULL
);
1180 memcpy(r300
->vertex_buffer
, buffers
,
1181 sizeof(struct pipe_vertex_buffer
) * count
);
1183 r300
->vertex_buffer_count
= count
;
1184 r300
->vertex_buffer_max_index
= max_index
;
1185 r300
->any_user_vbs
= any_user_buffer
;
1188 draw_flush(r300
->draw
);
1189 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
1193 /* Update the PSC tables. */
1194 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1196 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1197 uint16_t type
, swizzle
;
1198 enum pipe_format format
;
1201 if (velems
->count
> 16) {
1202 fprintf(stderr
, "r300: More than 16 vertex elements are not supported,"
1203 " requested %i, using 16.\n", velems
->count
);
1207 /* Vertex shaders have no semantics on their inputs,
1208 * so PSC should just route stuff based on the vertex elements,
1209 * and not on attrib information. */
1210 for (i
= 0; i
< velems
->count
; i
++) {
1211 format
= velems
->velem
[i
].src_format
;
1213 type
= r300_translate_vertex_data_type(format
) |
1214 (i
<< R300_DST_VEC_LOC_SHIFT
);
1215 swizzle
= r300_translate_vertex_data_swizzle(format
);
1218 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1219 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1221 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1222 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1226 /* Set the last vector in the PSC. */
1230 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1231 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1233 vstream
->count
= (i
>> 1) + 1;
1236 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1238 const struct pipe_vertex_element
* attribs
)
1240 struct r300_vertex_element_state
*velems
;
1242 enum pipe_format
*format
;
1244 assert(count
<= PIPE_MAX_ATTRIBS
);
1245 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1246 if (velems
!= NULL
) {
1247 velems
->count
= count
;
1248 memcpy(velems
->velem
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
1250 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1251 r300_vertex_psc(velems
);
1253 /* Check if the format is aligned to the size of DWORD.
1254 * We only care about the blocksizes of the formats since
1255 * swizzles are already set up. */
1256 for (i
= 0; i
< count
; i
++) {
1257 format
= &velems
->velem
[i
].src_format
;
1259 /* Replace some formats with their aligned counterparts,
1260 * this is OK because we check for aligned strides too. */
1262 /* Align to RGBA8. */
1263 case PIPE_FORMAT_R8_UNORM
:
1264 case PIPE_FORMAT_R8G8_UNORM
:
1265 case PIPE_FORMAT_R8G8B8_UNORM
:
1266 *format
= PIPE_FORMAT_R8G8B8A8_UNORM
;
1268 case PIPE_FORMAT_R8_SNORM
:
1269 case PIPE_FORMAT_R8G8_SNORM
:
1270 case PIPE_FORMAT_R8G8B8_SNORM
:
1271 *format
= PIPE_FORMAT_R8G8B8A8_SNORM
;
1273 case PIPE_FORMAT_R8_USCALED
:
1274 case PIPE_FORMAT_R8G8_USCALED
:
1275 case PIPE_FORMAT_R8G8B8_USCALED
:
1276 *format
= PIPE_FORMAT_R8G8B8A8_USCALED
;
1278 case PIPE_FORMAT_R8_SSCALED
:
1279 case PIPE_FORMAT_R8G8_SSCALED
:
1280 case PIPE_FORMAT_R8G8B8_SSCALED
:
1281 *format
= PIPE_FORMAT_R8G8B8A8_SSCALED
;
1284 /* Align to RG16. */
1285 case PIPE_FORMAT_R16_UNORM
:
1286 *format
= PIPE_FORMAT_R16G16_UNORM
;
1288 case PIPE_FORMAT_R16_SNORM
:
1289 *format
= PIPE_FORMAT_R16G16_SNORM
;
1291 case PIPE_FORMAT_R16_USCALED
:
1292 *format
= PIPE_FORMAT_R16G16_USCALED
;
1294 case PIPE_FORMAT_R16_SSCALED
:
1295 *format
= PIPE_FORMAT_R16G16_SSCALED
;
1297 case PIPE_FORMAT_R16_FLOAT
:
1298 *format
= PIPE_FORMAT_R16G16_FLOAT
;
1301 /* Align to RGBA16. */
1302 case PIPE_FORMAT_R16G16B16_UNORM
:
1303 *format
= PIPE_FORMAT_R16G16B16A16_UNORM
;
1305 case PIPE_FORMAT_R16G16B16_SNORM
:
1306 *format
= PIPE_FORMAT_R16G16B16A16_SNORM
;
1308 case PIPE_FORMAT_R16G16B16_USCALED
:
1309 *format
= PIPE_FORMAT_R16G16B16A16_USCALED
;
1311 case PIPE_FORMAT_R16G16B16_SSCALED
:
1312 *format
= PIPE_FORMAT_R16G16B16A16_SSCALED
;
1314 case PIPE_FORMAT_R16G16B16_FLOAT
:
1315 *format
= PIPE_FORMAT_R16G16B16A16_FLOAT
;
1321 size
= util_format_get_blocksize(*format
);
1323 if (size
% 4 != 0) {
1324 /* XXX Shouldn't we align the format? */
1325 fprintf(stderr
, "r300_create_vertex_elements_state: "
1326 "Unaligned format %s:%i isn't supported\n",
1327 util_format_short_name(*format
), size
);
1338 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1341 struct r300_context
*r300
= r300_context(pipe
);
1342 struct r300_vertex_element_state
*velems
= state
;
1344 if (velems
== NULL
) {
1348 r300
->velems
= velems
;
1351 draw_flush(r300
->draw
);
1352 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1355 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1356 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1359 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1364 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1365 const struct pipe_shader_state
* shader
)
1367 struct r300_context
* r300
= r300_context(pipe
);
1369 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1371 /* Copy state directly into shader. */
1372 vs
->state
= *shader
;
1373 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
1375 if (r300
->screen
->caps
.has_tcl
) {
1376 r300_translate_vertex_shader(r300
, vs
, vs
->state
.tokens
);
1378 vs
->draw_vs
= draw_create_vertex_shader(r300
->draw
, shader
);
1384 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1386 struct r300_context
* r300
= r300_context(pipe
);
1387 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1390 r300
->vs_state
.state
= NULL
;
1393 if (vs
== r300
->vs_state
.state
) {
1396 r300
->vs_state
.state
= vs
;
1398 /* The majority of the RS block bits is dependent on the vertex shader. */
1399 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
1401 if (r300
->screen
->caps
.has_tcl
) {
1402 r300
->vs_state
.dirty
= TRUE
;
1403 r300
->vs_state
.size
=
1404 vs
->code
.length
+ 9 +
1405 (vs
->immediates_count
? vs
->immediates_count
* 4 + 3 : 0);
1407 if (vs
->externals_count
) {
1408 r300
->vs_constants
.dirty
= TRUE
;
1409 r300
->vs_constants
.size
= vs
->externals_count
* 4 + 3;
1411 r300
->vs_constants
.size
= 0;
1414 r300
->pvs_flush
.dirty
= TRUE
;
1416 draw_flush(r300
->draw
);
1417 draw_bind_vertex_shader(r300
->draw
,
1418 (struct draw_vertex_shader
*)vs
->draw_vs
);
1422 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1424 struct r300_context
* r300
= r300_context(pipe
);
1425 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1427 if (r300
->screen
->caps
.has_tcl
) {
1428 rc_constants_destroy(&vs
->code
.constants
);
1430 draw_delete_vertex_shader(r300
->draw
,
1431 (struct draw_vertex_shader
*)vs
->draw_vs
);
1434 FREE((void*)vs
->state
.tokens
);
1438 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1439 uint shader
, uint index
,
1440 struct pipe_resource
*buf
)
1442 struct r300_context
* r300
= r300_context(pipe
);
1443 struct r300_constant_buffer
*cbuf
;
1444 struct pipe_transfer
*tr
;
1449 case PIPE_SHADER_VERTEX
:
1450 cbuf
= (struct r300_constant_buffer
*)r300
->vs_constants
.state
;
1453 case PIPE_SHADER_FRAGMENT
:
1454 cbuf
= (struct r300_constant_buffer
*)r300
->fs_constants
.state
;
1455 if (r300
->screen
->caps
.is_r500
) {
1466 if (buf
== NULL
|| buf
->width0
== 0 ||
1467 (mapped
= pipe_buffer_map(pipe
, buf
, PIPE_TRANSFER_READ
, &tr
)) == NULL
)
1473 assert((buf
->width0
% 4 * sizeof(float)) == 0);
1475 /* Check the size of the constant buffer. */
1476 /* XXX Subtract immediates and RC_STATE_* variables. */
1477 if (buf
->width0
> (sizeof(float) * 4 * max_size
)) {
1478 fprintf(stderr
, "r300: Max size of the constant buffer is "
1479 "%i*4 floats.\n", max_size
);
1483 memcpy(cbuf
->constants
, mapped
, buf
->width0
);
1484 cbuf
->count
= buf
->width0
/ (4 * sizeof(float));
1485 pipe_buffer_unmap(pipe
, buf
, tr
);
1487 if (shader
== PIPE_SHADER_VERTEX
) {
1488 if (r300
->screen
->caps
.has_tcl
) {
1489 if (r300
->vs_constants
.size
) {
1490 r300
->vs_constants
.dirty
= TRUE
;
1492 r300
->pvs_flush
.dirty
= TRUE
;
1493 } else if (r300
->draw
) {
1494 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
1498 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
1499 r300
->fs_constants
.dirty
= TRUE
;
1503 void r300_init_state_functions(struct r300_context
* r300
)
1505 r300
->context
.create_blend_state
= r300_create_blend_state
;
1506 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1507 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1509 r300
->context
.set_blend_color
= r300_set_blend_color
;
1511 r300
->context
.set_clip_state
= r300_set_clip_state
;
1512 r300
->context
.set_sample_mask
= r300_set_sample_mask
;
1514 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1516 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1517 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1518 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1520 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
1522 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1524 r300
->context
.create_fs_state
= r300_create_fs_state
;
1525 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1526 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1528 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1530 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1531 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1532 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1534 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1535 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1536 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1537 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1539 r300
->context
.set_fragment_sampler_views
= r300_set_fragment_sampler_views
;
1540 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
1541 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
1543 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1545 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1547 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1549 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
1550 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
1551 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
1553 r300
->context
.create_vs_state
= r300_create_vs_state
;
1554 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1555 r300
->context
.delete_vs_state
= r300_delete_vs_state
;