2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * on the rights to use, copy, modify, merge, publish, distribute, sub
9 * license, and/or sell copies of the Software, and to permit persons to whom
10 * the Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
20 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
21 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
22 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
24 #include "draw/draw_context.h"
26 #include "util/u_math.h"
27 #include "util/u_memory.h"
28 #include "util/u_pack_color.h"
30 #include "tgsi/tgsi_parse.h"
32 #include "pipe/p_config.h"
34 #include "r300_context.h"
36 #include "r300_screen.h"
37 #include "r300_screen_buffer.h"
38 #include "r300_state_inlines.h"
40 #include "r300_texture.h"
42 #include "r300_winsys.h"
44 /* r300_state: Functions used to intialize state context by translating
45 * Gallium state objects into semi-native r300 state objects. */
47 #define UPDATE_STATE(cso, atom) \
48 if (cso != atom.state) { \
53 static boolean
blend_discard_if_src_alpha_0(unsigned srcRGB
, unsigned srcA
,
54 unsigned dstRGB
, unsigned dstA
)
56 /* If the blend equation is ADD or REVERSE_SUBTRACT,
57 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
58 * will not be changed.
59 * Notice that the dst factors are the src factors inverted. */
60 return (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
61 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
62 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
63 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
64 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
65 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
66 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
67 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
68 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
69 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
70 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
71 dstA
== PIPE_BLENDFACTOR_ONE
);
74 static boolean
blend_discard_if_src_alpha_1(unsigned srcRGB
, unsigned srcA
,
75 unsigned dstRGB
, unsigned dstA
)
77 /* If the blend equation is ADD or REVERSE_SUBTRACT,
78 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
79 * will not be changed.
80 * Notice that the dst factors are the src factors inverted. */
81 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
82 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
83 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
84 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
85 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
86 (dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
87 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
88 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
89 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
90 dstA
== PIPE_BLENDFACTOR_ONE
);
93 static boolean
blend_discard_if_src_color_0(unsigned srcRGB
, unsigned srcA
,
94 unsigned dstRGB
, unsigned dstA
)
96 /* If the blend equation is ADD or REVERSE_SUBTRACT,
97 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
98 * will not be changed.
99 * Notice that the dst factors are the src factors inverted. */
100 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
101 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
102 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
103 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
104 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
105 (dstA
== PIPE_BLENDFACTOR_ONE
);
108 static boolean
blend_discard_if_src_color_1(unsigned srcRGB
, unsigned srcA
,
109 unsigned dstRGB
, unsigned dstA
)
111 /* If the blend equation is ADD or REVERSE_SUBTRACT,
112 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
113 * will not be changed.
114 * Notice that the dst factors are the src factors inverted. */
115 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
116 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
117 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
118 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
119 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
120 (dstA
== PIPE_BLENDFACTOR_ONE
);
123 static boolean
blend_discard_if_src_alpha_color_0(unsigned srcRGB
, unsigned srcA
,
124 unsigned dstRGB
, unsigned dstA
)
126 /* If the blend equation is ADD or REVERSE_SUBTRACT,
127 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
128 * the colorbuffer will not be changed.
129 * Notice that the dst factors are the src factors inverted. */
130 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
131 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
132 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
133 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
134 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
135 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
136 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
137 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
138 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
139 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
140 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
141 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
142 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
143 dstA
== PIPE_BLENDFACTOR_ONE
);
146 static boolean
blend_discard_if_src_alpha_color_1(unsigned srcRGB
, unsigned srcA
,
147 unsigned dstRGB
, unsigned dstA
)
149 /* If the blend equation is ADD or REVERSE_SUBTRACT,
150 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
151 * the colorbuffer will not be changed.
152 * Notice that the dst factors are the src factors inverted. */
153 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
154 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
155 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
156 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
157 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
158 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
159 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
160 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
161 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
162 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
163 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
164 dstA
== PIPE_BLENDFACTOR_ONE
);
167 static unsigned bgra_cmask(unsigned mask
)
169 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
171 return ((mask
& PIPE_MASK_R
) << 2) |
172 ((mask
& PIPE_MASK_B
) >> 2) |
173 (mask
& (PIPE_MASK_G
| PIPE_MASK_A
));
176 /* Create a new blend state based on the CSO blend state.
178 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
179 static void* r300_create_blend_state(struct pipe_context
* pipe
,
180 const struct pipe_blend_state
* state
)
182 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
183 struct r300_blend_state
* blend
= CALLOC_STRUCT(r300_blend_state
);
185 if (state
->rt
[0].blend_enable
)
187 unsigned eqRGB
= state
->rt
[0].rgb_func
;
188 unsigned srcRGB
= state
->rt
[0].rgb_src_factor
;
189 unsigned dstRGB
= state
->rt
[0].rgb_dst_factor
;
191 unsigned eqA
= state
->rt
[0].alpha_func
;
192 unsigned srcA
= state
->rt
[0].alpha_src_factor
;
193 unsigned dstA
= state
->rt
[0].alpha_dst_factor
;
195 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
196 * this is just the crappy D3D naming */
197 blend
->blend_control
= R300_ALPHA_BLEND_ENABLE
|
198 r300_translate_blend_function(eqRGB
) |
199 ( r300_translate_blend_factor(srcRGB
) << R300_SRC_BLEND_SHIFT
) |
200 ( r300_translate_blend_factor(dstRGB
) << R300_DST_BLEND_SHIFT
);
202 /* Optimization: some operations do not require the destination color.
204 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
205 * otherwise blending gives incorrect results. It seems to be
207 if (eqRGB
== PIPE_BLEND_MIN
|| eqA
== PIPE_BLEND_MIN
||
208 eqRGB
== PIPE_BLEND_MAX
|| eqA
== PIPE_BLEND_MAX
||
209 dstRGB
!= PIPE_BLENDFACTOR_ZERO
||
210 dstA
!= PIPE_BLENDFACTOR_ZERO
||
211 srcRGB
== PIPE_BLENDFACTOR_DST_COLOR
||
212 srcRGB
== PIPE_BLENDFACTOR_DST_ALPHA
||
213 srcRGB
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
214 srcRGB
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
215 srcA
== PIPE_BLENDFACTOR_DST_COLOR
||
216 srcA
== PIPE_BLENDFACTOR_DST_ALPHA
||
217 srcA
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
218 srcA
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
219 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
) {
220 /* Enable reading from the colorbuffer. */
221 blend
->blend_control
|= R300_READ_ENABLE
;
223 if (r300screen
->caps
.is_r500
) {
224 /* Optimization: Depending on incoming pixels, we can
225 * conditionally disable the reading in hardware... */
226 if (eqRGB
!= PIPE_BLEND_MIN
&& eqA
!= PIPE_BLEND_MIN
&&
227 eqRGB
!= PIPE_BLEND_MAX
&& eqA
!= PIPE_BLEND_MAX
) {
228 /* Disable reading if SRC_ALPHA == 0. */
229 if ((dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
230 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
231 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
232 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
233 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
234 blend
->blend_control
|= R500_SRC_ALPHA_0_NO_READ
;
237 /* Disable reading if SRC_ALPHA == 1. */
238 if ((dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
239 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
240 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
241 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
242 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
243 blend
->blend_control
|= R500_SRC_ALPHA_1_NO_READ
;
249 /* Optimization: discard pixels which don't change the colorbuffer.
251 * The code below is non-trivial and some math is involved.
253 * Discarding pixels must be disabled when FP16 AA is enabled.
254 * This is a hardware bug. Also, this implementation wouldn't work
255 * with FP blending enabled and equation clamping disabled.
257 * Equations other than ADD are rarely used and therefore won't be
259 if ((eqRGB
== PIPE_BLEND_ADD
|| eqRGB
== PIPE_BLEND_REVERSE_SUBTRACT
) &&
260 (eqA
== PIPE_BLEND_ADD
|| eqA
== PIPE_BLEND_REVERSE_SUBTRACT
)) {
262 * REVERSE_SUBTRACT: Y-X
265 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
266 * then CB will not be changed.
268 * Given the srcFactor and dstFactor variables, we can derive
269 * what src and dst should be equal to and discard appropriate
272 if (blend_discard_if_src_alpha_0(srcRGB
, srcA
, dstRGB
, dstA
)) {
273 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0
;
274 } else if (blend_discard_if_src_alpha_1(srcRGB
, srcA
,
276 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1
;
277 } else if (blend_discard_if_src_color_0(srcRGB
, srcA
,
279 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0
;
280 } else if (blend_discard_if_src_color_1(srcRGB
, srcA
,
282 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1
;
283 } else if (blend_discard_if_src_alpha_color_0(srcRGB
, srcA
,
285 blend
->blend_control
|=
286 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0
;
287 } else if (blend_discard_if_src_alpha_color_1(srcRGB
, srcA
,
289 blend
->blend_control
|=
290 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1
;
295 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
296 blend
->blend_control
|= R300_SEPARATE_ALPHA_ENABLE
;
297 blend
->alpha_blend_control
=
298 r300_translate_blend_function(eqA
) |
299 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
300 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
304 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
305 if (state
->logicop_enable
) {
306 blend
->rop
= R300_RB3D_ROPCNTL_ROP_ENABLE
|
307 (state
->logicop_func
) << R300_RB3D_ROPCNTL_ROP_SHIFT
;
310 /* Color channel masks for all MRTs. */
311 blend
->color_channel_mask
= bgra_cmask(state
->rt
[0].colormask
);
312 if (r300screen
->caps
.is_r500
&& state
->independent_blend_enable
) {
313 if (state
->rt
[1].blend_enable
) {
314 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[1].colormask
) << 4;
316 if (state
->rt
[2].blend_enable
) {
317 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[2].colormask
) << 8;
319 if (state
->rt
[3].blend_enable
) {
320 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[3].colormask
) << 12;
324 /* Neither fglrx nor classic r300 ever set this, regardless of dithering
325 * state. Since it's an optional implementation detail, we can leave it
326 * out and never dither.
328 * This could be revisited if we ever get quality or conformance hints.
331 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
332 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
339 /* Bind blend state. */
340 static void r300_bind_blend_state(struct pipe_context
* pipe
,
343 struct r300_context
* r300
= r300_context(pipe
);
345 UPDATE_STATE(state
, r300
->blend_state
);
348 /* Free blend state. */
349 static void r300_delete_blend_state(struct pipe_context
* pipe
,
355 /* Convert float to 10bit integer */
356 static unsigned float_to_fixed10(float f
)
358 return CLAMP((unsigned)(f
* 1023.9f
), 0, 1023);
362 * Setup both R300 and R500 registers, figure out later which one to write. */
363 static void r300_set_blend_color(struct pipe_context
* pipe
,
364 const struct pipe_blend_color
* color
)
366 struct r300_context
* r300
= r300_context(pipe
);
367 struct r300_blend_color_state
* state
=
368 (struct r300_blend_color_state
*)r300
->blend_color_state
.state
;
371 util_pack_color(color
->color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
372 state
->blend_color
= uc
.ui
;
374 /* XXX if FP16 blending is enabled, we should use the FP16 format */
375 state
->blend_color_red_alpha
=
376 float_to_fixed10(color
->color
[0]) |
377 (float_to_fixed10(color
->color
[3]) << 16);
378 state
->blend_color_green_blue
=
379 float_to_fixed10(color
->color
[2]) |
380 (float_to_fixed10(color
->color
[1]) << 16);
382 r300
->blend_color_state
.size
= r300
->screen
->caps
.is_r500
? 3 : 2;
383 r300
->blend_color_state
.dirty
= TRUE
;
386 static void r300_set_clip_state(struct pipe_context
* pipe
,
387 const struct pipe_clip_state
* state
)
389 struct r300_context
* r300
= r300_context(pipe
);
393 if (r300
->screen
->caps
.has_tcl
) {
394 memcpy(r300
->clip_state
.state
, state
, sizeof(struct pipe_clip_state
));
395 r300
->clip_state
.size
= 29;
397 draw_flush(r300
->draw
);
398 draw_set_clip_state(r300
->draw
, state
);
399 r300
->clip_state
.size
= 2;
402 r300
->clip_state
.dirty
= TRUE
;
405 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
407 * This contains the depth buffer, stencil buffer, alpha test, and such.
408 * On the Radeon, depth and stencil buffer setup are intertwined, which is
409 * the reason for some of the strange-looking assignments across registers. */
411 r300_create_dsa_state(struct pipe_context
* pipe
,
412 const struct pipe_depth_stencil_alpha_state
* state
)
414 struct r300_capabilities
*caps
= &r300_screen(pipe
->screen
)->caps
;
415 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
417 /* Depth test setup. */
418 if (state
->depth
.enabled
) {
419 dsa
->z_buffer_control
|= R300_Z_ENABLE
;
421 if (state
->depth
.writemask
) {
422 dsa
->z_buffer_control
|= R300_Z_WRITE_ENABLE
;
425 dsa
->z_stencil_control
|=
426 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
430 /* Stencil buffer setup. */
431 if (state
->stencil
[0].enabled
) {
432 dsa
->z_buffer_control
|= R300_STENCIL_ENABLE
;
433 dsa
->z_stencil_control
|=
434 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
435 R300_S_FRONT_FUNC_SHIFT
) |
436 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
437 R300_S_FRONT_SFAIL_OP_SHIFT
) |
438 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
439 R300_S_FRONT_ZPASS_OP_SHIFT
) |
440 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
441 R300_S_FRONT_ZFAIL_OP_SHIFT
);
443 dsa
->stencil_ref_mask
=
444 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
445 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
447 if (state
->stencil
[1].enabled
) {
448 dsa
->two_sided
= TRUE
;
450 dsa
->z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
451 dsa
->z_stencil_control
|=
452 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
453 R300_S_BACK_FUNC_SHIFT
) |
454 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
455 R300_S_BACK_SFAIL_OP_SHIFT
) |
456 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
457 R300_S_BACK_ZPASS_OP_SHIFT
) |
458 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
459 R300_S_BACK_ZFAIL_OP_SHIFT
);
461 dsa
->stencil_ref_bf
=
462 (state
->stencil
[1].valuemask
<< R300_STENCILMASK_SHIFT
) |
463 (state
->stencil
[1].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
466 dsa
->z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
468 dsa
->stencil_ref_bf_fallback
=
469 (state
->stencil
[0].valuemask
!= state
->stencil
[1].valuemask
||
470 state
->stencil
[0].writemask
!= state
->stencil
[1].writemask
);
475 /* Alpha test setup. */
476 if (state
->alpha
.enabled
) {
477 dsa
->alpha_function
=
478 r300_translate_alpha_function(state
->alpha
.func
) |
479 R300_FG_ALPHA_FUNC_ENABLE
;
481 /* We could use 10bit alpha ref but who needs that? */
482 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
485 dsa
->alpha_function
|= R500_FG_ALPHA_FUNC_8BIT
;
491 static void r300_update_stencil_ref_fallback_status(struct r300_context
*r300
)
493 struct r300_dsa_state
*dsa
= (struct r300_dsa_state
*)r300
->dsa_state
.state
;
495 if (r300
->screen
->caps
.is_r500
) {
499 r300
->stencil_ref_bf_fallback
=
500 dsa
->stencil_ref_bf_fallback
||
502 r300
->stencil_ref
.ref_value
[0] != r300
->stencil_ref
.ref_value
[1]);
505 /* Bind DSA state. */
506 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
509 struct r300_context
* r300
= r300_context(pipe
);
515 UPDATE_STATE(state
, r300
->dsa_state
);
517 r300_update_stencil_ref_fallback_status(r300
);
520 /* Free DSA state. */
521 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
527 static void r300_set_stencil_ref(struct pipe_context
* pipe
,
528 const struct pipe_stencil_ref
* sr
)
530 struct r300_context
* r300
= r300_context(pipe
);
532 r300
->stencil_ref
= *sr
;
533 r300
->dsa_state
.dirty
= TRUE
;
535 r300_update_stencil_ref_fallback_status(r300
);
538 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
539 static void r300_fb_update_tiling_flags(struct r300_context
*r300
,
540 const struct pipe_framebuffer_state
*old_state
,
541 const struct pipe_framebuffer_state
*new_state
)
543 struct r300_texture
*tex
;
544 unsigned i
, j
, level
;
546 /* Reset tiling flags for old surfaces to default values. */
547 for (i
= 0; i
< old_state
->nr_cbufs
; i
++) {
548 for (j
= 0; j
< new_state
->nr_cbufs
; j
++) {
549 if (old_state
->cbufs
[i
]->texture
== new_state
->cbufs
[j
]->texture
) {
553 /* If not binding the surface again... */
554 if (j
!= new_state
->nr_cbufs
) {
558 tex
= r300_texture(old_state
->cbufs
[i
]->texture
);
561 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
567 if (old_state
->zsbuf
&&
568 (!new_state
->zsbuf
||
569 old_state
->zsbuf
->texture
!= new_state
->zsbuf
->texture
)) {
570 tex
= r300_texture(old_state
->zsbuf
->texture
);
573 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
580 /* Set tiling flags for new surfaces. */
581 for (i
= 0; i
< new_state
->nr_cbufs
; i
++) {
582 tex
= r300_texture(new_state
->cbufs
[i
]->texture
);
583 level
= new_state
->cbufs
[i
]->level
;
585 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
588 tex
->mip_macrotile
[level
]);
590 if (new_state
->zsbuf
) {
591 tex
= r300_texture(new_state
->zsbuf
->texture
);
592 level
= new_state
->zsbuf
->level
;
594 r300
->rws
->buffer_set_tiling(r300
->rws
, tex
->buffer
,
597 tex
->mip_macrotile
[level
]);
602 r300_set_framebuffer_state(struct pipe_context
* pipe
,
603 const struct pipe_framebuffer_state
* state
)
605 struct r300_context
* r300
= r300_context(pipe
);
606 struct pipe_framebuffer_state
*old_state
= r300
->fb_state
.state
;
607 unsigned max_width
, max_height
;
608 uint32_t zbuffer_bpp
= 0;
610 if (state
->nr_cbufs
> 4) {
611 fprintf(stderr
, "r300: Implementation error: Too many MRTs in %s, "
612 "refusing to bind framebuffer state!\n", __FUNCTION__
);
616 if (r300
->screen
->caps
.is_r500
) {
617 max_width
= max_height
= 4096;
618 } else if (r300
->screen
->caps
.is_r400
) {
619 max_width
= max_height
= 4021;
621 max_width
= max_height
= 2560;
624 if (state
->width
> max_width
|| state
->height
> max_height
) {
625 fprintf(stderr
, "r300: Implementation error: Render targets are too "
626 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__
);
631 draw_flush(r300
->draw
);
634 r300
->fb_state
.dirty
= TRUE
;
636 /* If nr_cbufs is changed from zero to non-zero or vice versa... */
637 if (!!old_state
->nr_cbufs
!= !!state
->nr_cbufs
) {
638 r300
->blend_state
.dirty
= TRUE
;
640 /* If zsbuf is set from NULL to non-NULL or vice versa.. */
641 if (!!old_state
->zsbuf
!= !!state
->zsbuf
) {
642 r300
->dsa_state
.dirty
= TRUE
;
645 r300_fb_update_tiling_flags(r300
, r300
->fb_state
.state
, state
);
647 memcpy(r300
->fb_state
.state
, state
, sizeof(struct pipe_framebuffer_state
));
649 r300
->fb_state
.size
= (10 * state
->nr_cbufs
) + (2 * (4 - state
->nr_cbufs
)) +
650 (state
->zsbuf
? 10 : 0) + 11;
652 /* Polygon offset depends on the zbuffer bit depth. */
653 if (state
->zsbuf
&& r300
->polygon_offset_enabled
) {
654 switch (util_format_get_blocksize(state
->zsbuf
->texture
->format
)) {
663 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
664 r300
->zbuffer_bpp
= zbuffer_bpp
;
665 r300
->rs_state
.dirty
= TRUE
;
670 /* Create fragment shader state. */
671 static void* r300_create_fs_state(struct pipe_context
* pipe
,
672 const struct pipe_shader_state
* shader
)
674 struct r300_fragment_shader
* fs
= NULL
;
676 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
678 /* Copy state directly into shader. */
680 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
685 /* Bind fragment shader state. */
686 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
688 struct r300_context
* r300
= r300_context(pipe
);
689 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
697 r300_pick_fragment_shader(r300
);
699 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
701 if (r300
->vs_state
.state
&& r300_vertex_shader_setup_wpos(r300
)) {
702 r300
->vap_output_state
.dirty
= TRUE
;
705 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER
| R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
708 /* Delete fragment shader state. */
709 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
711 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
712 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
717 rc_constants_destroy(&tmp
->code
.constants
);
720 FREE((void*)fs
->state
.tokens
);
724 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
725 const struct pipe_poly_stipple
* state
)
727 /* XXX no idea how to set this up, but not terribly important */
730 /* Create a new rasterizer state based on the CSO rasterizer state.
732 * This is a very large chunk of state, and covers most of the graphics
733 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
735 * In a not entirely unironic sidenote, this state has nearly nothing to do
736 * with the actual block on the Radeon called the rasterizer (RS). */
737 static void* r300_create_rs_state(struct pipe_context
* pipe
,
738 const struct pipe_rasterizer_state
* state
)
740 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
742 /* Copy rasterizer state for Draw. */
745 #ifdef PIPE_ARCH_LITTLE_ENDIAN
746 rs
->vap_control_status
= R300_VC_NO_SWAP
;
748 rs
->vap_control_status
= R300_VC_32BIT_SWAP
;
751 /* If no TCL engine is present, turn off the HW TCL. */
752 if (!r300_screen(pipe
->screen
)->caps
.has_tcl
) {
753 rs
->vap_control_status
|= R300_VAP_TCL_BYPASS
;
756 rs
->point_size
= pack_float_16_6x(state
->point_size
) |
757 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
759 rs
->line_control
= pack_float_16_6x(state
->line_width
) |
760 R300_GA_LINE_CNTL_END_TYPE_COMP
;
762 /* Enable polygon mode */
763 if (state
->fill_cw
!= PIPE_POLYGON_MODE_FILL
||
764 state
->fill_ccw
!= PIPE_POLYGON_MODE_FILL
) {
765 rs
->polygon_mode
= R300_GA_POLY_MODE_DUAL
;
768 /* Radeons don't think in "CW/CCW", they think in "front/back". */
769 if (state
->front_winding
== PIPE_WINDING_CW
) {
770 rs
->cull_mode
= R300_FRONT_FACE_CW
;
773 if (state
->offset_cw
) {
774 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
776 if (state
->offset_ccw
) {
777 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
781 if (rs
->polygon_mode
) {
783 r300_translate_polygon_mode_front(state
->fill_cw
);
785 r300_translate_polygon_mode_back(state
->fill_ccw
);
788 rs
->cull_mode
= R300_FRONT_FACE_CCW
;
791 if (state
->offset_ccw
) {
792 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
794 if (state
->offset_cw
) {
795 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
799 if (rs
->polygon_mode
) {
801 r300_translate_polygon_mode_front(state
->fill_ccw
);
803 r300_translate_polygon_mode_back(state
->fill_cw
);
806 if (state
->front_winding
& state
->cull_mode
) {
807 rs
->cull_mode
|= R300_CULL_FRONT
;
809 if (~(state
->front_winding
) & state
->cull_mode
) {
810 rs
->cull_mode
|= R300_CULL_BACK
;
813 if (rs
->polygon_offset_enable
) {
814 rs
->depth_offset
= state
->offset_units
;
815 rs
->depth_scale
= state
->offset_scale
;
818 if (state
->line_stipple_enable
) {
819 rs
->line_stipple_config
=
820 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
821 (fui((float)state
->line_stipple_factor
) &
822 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
823 /* XXX this might need to be scaled up */
824 rs
->line_stipple_value
= state
->line_stipple_pattern
;
827 if (state
->flatshade
) {
828 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
830 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
833 rs
->clip_rule
= state
->scissor
? 0xAAAA : 0xFFFF;
838 /* Bind rasterizer state. */
839 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
841 struct r300_context
* r300
= r300_context(pipe
);
842 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
845 draw_flush(r300
->draw
);
846 draw_set_rasterizer_state(r300
->draw
, &rs
->rs
);
850 r300
->polygon_offset_enabled
= rs
->rs
.offset_cw
|| rs
->rs
.offset_ccw
;
852 r300
->polygon_offset_enabled
= FALSE
;
855 UPDATE_STATE(state
, r300
->rs_state
);
856 r300
->rs_state
.size
= 19 + (r300
->polygon_offset_enabled
? 5 : 0);
859 /* Free rasterizer state. */
860 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
866 r300_create_sampler_state(struct pipe_context
* pipe
,
867 const struct pipe_sampler_state
* state
)
869 struct r300_context
* r300
= r300_context(pipe
);
870 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
871 boolean is_r500
= r300
->screen
->caps
.is_r500
;
875 sampler
->state
= *state
;
878 (r300_translate_wrap(state
->wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
879 (r300_translate_wrap(state
->wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
880 (r300_translate_wrap(state
->wrap_r
) << R300_TX_WRAP_R_SHIFT
);
882 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
883 state
->mag_img_filter
,
884 state
->min_mip_filter
,
885 state
->max_anisotropy
> 0);
887 sampler
->filter0
|= r300_anisotropy(state
->max_anisotropy
);
889 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
890 /* We must pass these to the merge function to clamp them properly. */
891 sampler
->min_lod
= MAX2((unsigned)state
->min_lod
, 0);
892 sampler
->max_lod
= MAX2((unsigned)ceilf(state
->max_lod
), 0);
894 lod_bias
= CLAMP((int)(state
->lod_bias
* 32), -(1 << 9), (1 << 9) - 1);
896 sampler
->filter1
|= lod_bias
<< R300_LOD_BIAS_SHIFT
;
898 /* This is very high quality anisotropic filtering for R5xx.
899 * It's good for benchmarking the performance of texturing but
900 * in practice we don't want to slow down the driver because it's
901 * a pretty good performance killer. Feel free to play with it. */
902 if (DBG_ON(r300
, DBG_ANISOHQ
) && is_r500
) {
903 sampler
->filter1
|= r500_anisotropy(state
->max_anisotropy
);
906 util_pack_color(state
->border_color
, PIPE_FORMAT_B8G8R8A8_UNORM
, &uc
);
907 sampler
->border_color
= uc
.ui
;
909 /* R500-specific fixups and optimizations */
910 if (r300
->screen
->caps
.is_r500
) {
911 sampler
->filter1
|= R500_BORDER_FIX
;
914 return (void*)sampler
;
917 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
921 struct r300_context
* r300
= r300_context(pipe
);
922 struct r300_textures_state
* state
=
923 (struct r300_textures_state
*)r300
->textures_state
.state
;
924 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
926 if (count
> tex_units
) {
930 memcpy(state
->sampler_states
, states
, sizeof(void*) * count
);
931 state
->sampler_state_count
= count
;
933 r300
->textures_state
.dirty
= TRUE
;
935 /* Pick a fragment shader based on the texture compare state. */
936 if (r300
->fs
&& count
) {
937 if (r300_pick_fragment_shader(r300
)) {
938 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER
|
939 R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
944 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
950 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
955 static void r300_set_fragment_sampler_views(struct pipe_context
* pipe
,
957 struct pipe_sampler_view
** views
)
959 struct r300_context
* r300
= r300_context(pipe
);
960 struct r300_textures_state
* state
=
961 (struct r300_textures_state
*)r300
->textures_state
.state
;
962 struct r300_texture
*texture
;
964 unsigned tex_units
= r300
->screen
->caps
.num_tex_units
;
965 boolean is_r500
= r300
->screen
->caps
.is_r500
;
966 boolean dirty_tex
= FALSE
;
968 if (count
> tex_units
) {
972 for (i
= 0; i
< count
; i
++) {
973 if (&state
->sampler_views
[i
]->base
!= views
[i
]) {
974 pipe_sampler_view_reference(
975 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
982 /* A new sampler view (= texture)... */
985 /* R300-specific - set the texrect factor in the fragment shader */
986 texture
= r300_texture(views
[i
]->texture
);
987 if (!is_r500
&& texture
->uses_pitch
) {
988 /* XXX It would be nice to re-emit just 1 constant,
989 * XXX not all of them */
990 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
995 for (i
= count
; i
< tex_units
; i
++) {
996 if (state
->sampler_views
[i
]) {
997 pipe_sampler_view_reference(
998 (struct pipe_sampler_view
**)&state
->sampler_views
[i
],
1003 state
->sampler_view_count
= count
;
1005 r300
->textures_state
.dirty
= TRUE
;
1008 r300
->texture_cache_inval
.dirty
= TRUE
;
1012 static struct pipe_sampler_view
*
1013 r300_create_sampler_view(struct pipe_context
*pipe
,
1014 struct pipe_resource
*texture
,
1015 const struct pipe_sampler_view
*templ
)
1017 struct r300_sampler_view
*view
= CALLOC_STRUCT(r300_sampler_view
);
1018 struct r300_texture
*tex
= r300_texture(texture
);
1019 unsigned char swizzle
[4];
1022 view
->base
= *templ
;
1023 view
->base
.reference
.count
= 1;
1024 view
->base
.context
= pipe
;
1025 view
->base
.texture
= NULL
;
1026 pipe_resource_reference(&view
->base
.texture
, texture
);
1028 swizzle
[0] = templ
->swizzle_r
;
1029 swizzle
[1] = templ
->swizzle_g
;
1030 swizzle
[2] = templ
->swizzle_b
;
1031 swizzle
[3] = templ
->swizzle_a
;
1033 /* XXX Enable swizzles when they become supported. Now we get RGBA
1034 * everywhere. And do testing! */
1035 view
->format
= tex
->tx_format
;
1036 view
->format
.format1
|= r300_translate_texformat(templ
->format
,
1038 if (r300_screen(pipe
->screen
)->caps
.is_r500
) {
1039 view
->format
.format2
|= r500_tx_format_msb_bit(templ
->format
);
1043 return (struct pipe_sampler_view
*)view
;
1047 r300_sampler_view_destroy(struct pipe_context
*pipe
,
1048 struct pipe_sampler_view
*view
)
1050 pipe_resource_reference(&view
->texture
, NULL
);
1054 static void r300_set_scissor_state(struct pipe_context
* pipe
,
1055 const struct pipe_scissor_state
* state
)
1057 struct r300_context
* r300
= r300_context(pipe
);
1059 memcpy(r300
->scissor_state
.state
, state
,
1060 sizeof(struct pipe_scissor_state
));
1062 r300
->scissor_state
.dirty
= TRUE
;
1065 static void r300_set_viewport_state(struct pipe_context
* pipe
,
1066 const struct pipe_viewport_state
* state
)
1068 struct r300_context
* r300
= r300_context(pipe
);
1069 struct r300_viewport_state
* viewport
=
1070 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
1072 r300
->viewport
= *state
;
1074 /* Do the transform in HW. */
1075 viewport
->vte_control
= R300_VTX_W0_FMT
;
1077 if (state
->scale
[0] != 1.0f
) {
1078 viewport
->xscale
= state
->scale
[0];
1079 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
1081 if (state
->scale
[1] != 1.0f
) {
1082 viewport
->yscale
= state
->scale
[1];
1083 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
1085 if (state
->scale
[2] != 1.0f
) {
1086 viewport
->zscale
= state
->scale
[2];
1087 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
1089 if (state
->translate
[0] != 0.0f
) {
1090 viewport
->xoffset
= state
->translate
[0];
1091 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
1093 if (state
->translate
[1] != 0.0f
) {
1094 viewport
->yoffset
= state
->translate
[1];
1095 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
1097 if (state
->translate
[2] != 0.0f
) {
1098 viewport
->zoffset
= state
->translate
[2];
1099 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
1102 r300
->viewport_state
.dirty
= TRUE
;
1103 if (r300
->fs
&& r300
->fs
->shader
->inputs
.wpos
!= ATTR_UNUSED
) {
1104 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1108 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
1110 const struct pipe_vertex_buffer
* buffers
)
1112 struct r300_context
* r300
= r300_context(pipe
);
1113 struct pipe_vertex_buffer
*vbo
;
1114 unsigned i
, max_index
= (1 << 24) - 1;
1115 boolean any_user_buffer
= FALSE
;
1117 if (count
== r300
->vertex_buffer_count
&&
1118 memcmp(r300
->vertex_buffer
, buffers
,
1119 sizeof(struct pipe_vertex_buffer
) * count
) == 0) {
1123 /* Check if the stride is aligned to the size of DWORD. */
1124 for (i
= 0; i
< count
; i
++) {
1125 if (buffers
[i
].buffer
) {
1126 if (buffers
[i
].stride
% 4 != 0) {
1127 // XXX Shouldn't we align the buffer?
1128 fprintf(stderr
, "r300: set_vertex_buffers: "
1129 "Unaligned buffer stride %i isn't supported.\n",
1136 for (i
= 0; i
< count
; i
++) {
1137 /* Why, yes, I AM casting away constness. How did you know? */
1138 vbo
= (struct pipe_vertex_buffer
*)&buffers
[i
];
1140 /* Reference our buffer. */
1141 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, vbo
->buffer
);
1143 /* Skip NULL buffers */
1144 if (!buffers
[i
].buffer
) {
1148 if (r300_buffer_is_user_buffer(vbo
->buffer
)) {
1149 any_user_buffer
= TRUE
;
1152 if (vbo
->max_index
== ~0) {
1153 /* Bogus value from broken state tracker; hax it. */
1154 /* TODO - more hax - fixes doom3 from almos on irc */
1156 fprintf(stderr
, "r300: got a VBO with stride 0 fixing up to stide 4\n");
1160 (vbo
->buffer
->width0
- vbo
->buffer_offset
) / vbo
->stride
;
1163 max_index
= MIN2(vbo
->max_index
, max_index
);
1166 for (; i
< r300
->vertex_buffer_count
; i
++) {
1167 /* Dereference any old buffers. */
1168 pipe_resource_reference(&r300
->vertex_buffer
[i
].buffer
, NULL
);
1171 memcpy(r300
->vertex_buffer
, buffers
,
1172 sizeof(struct pipe_vertex_buffer
) * count
);
1174 r300
->vertex_buffer_count
= count
;
1175 r300
->vertex_buffer_max_index
= max_index
;
1176 r300
->any_user_vbs
= any_user_buffer
;
1179 draw_flush(r300
->draw
);
1180 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
1184 /* Update the PSC tables. */
1185 static void r300_vertex_psc(struct r300_vertex_element_state
*velems
)
1187 struct r300_vertex_stream_state
*vstream
= &velems
->vertex_stream
;
1188 uint16_t type
, swizzle
;
1189 enum pipe_format format
;
1192 if (velems
->count
> 16) {
1193 fprintf(stderr
, "r300: More than 16 vertex elements are not supported,"
1194 " requested %i, using 16.\n", velems
->count
);
1198 /* Vertex shaders have no semantics on their inputs,
1199 * so PSC should just route stuff based on the vertex elements,
1200 * and not on attrib information. */
1201 for (i
= 0; i
< velems
->count
; i
++) {
1202 format
= velems
->velem
[i
].src_format
;
1204 type
= r300_translate_vertex_data_type(format
) |
1205 (i
<< R300_DST_VEC_LOC_SHIFT
);
1206 swizzle
= r300_translate_vertex_data_swizzle(format
);
1209 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
<< 16;
1210 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
<< 16;
1212 vstream
->vap_prog_stream_cntl
[i
>> 1] |= type
;
1213 vstream
->vap_prog_stream_cntl_ext
[i
>> 1] |= swizzle
;
1217 /* Set the last vector in the PSC. */
1221 vstream
->vap_prog_stream_cntl
[i
>> 1] |=
1222 (R300_LAST_VEC
<< (i
& 1 ? 16 : 0));
1224 vstream
->count
= (i
>> 1) + 1;
1227 static void* r300_create_vertex_elements_state(struct pipe_context
* pipe
,
1229 const struct pipe_vertex_element
* attribs
)
1231 struct r300_vertex_element_state
*velems
;
1234 assert(count
<= PIPE_MAX_ATTRIBS
);
1235 velems
= CALLOC_STRUCT(r300_vertex_element_state
);
1236 if (velems
!= NULL
) {
1237 velems
->count
= count
;
1238 memcpy(velems
->velem
, attribs
, sizeof(struct pipe_vertex_element
) * count
);
1240 if (r300_screen(pipe
->screen
)->caps
.has_tcl
) {
1241 /* Check if the format is aligned to the size of DWORD. */
1242 for (i
= 0; i
< count
; i
++) {
1243 size
= util_format_get_blocksize(attribs
[i
].src_format
);
1245 if (size
% 4 != 0) {
1246 /* XXX Shouldn't we align the format? */
1247 fprintf(stderr
, "r300_create_vertex_elements_state: "
1248 "Unaligned format %s:%i isn't supported\n",
1249 util_format_name(attribs
[i
].src_format
), size
);
1255 r300_vertex_psc(velems
);
1261 static void r300_bind_vertex_elements_state(struct pipe_context
*pipe
,
1264 struct r300_context
*r300
= r300_context(pipe
);
1265 struct r300_vertex_element_state
*velems
= state
;
1267 if (velems
== NULL
) {
1271 r300
->velems
= velems
;
1274 draw_flush(r300
->draw
);
1275 draw_set_vertex_elements(r300
->draw
, velems
->count
, velems
->velem
);
1278 UPDATE_STATE(&velems
->vertex_stream
, r300
->vertex_stream_state
);
1279 r300
->vertex_stream_state
.size
= (1 + velems
->vertex_stream
.count
) * 2;
1282 static void r300_delete_vertex_elements_state(struct pipe_context
*pipe
, void *state
)
1287 static void* r300_create_vs_state(struct pipe_context
* pipe
,
1288 const struct pipe_shader_state
* shader
)
1290 struct r300_context
* r300
= r300_context(pipe
);
1292 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
1294 /* Copy state directly into shader. */
1295 vs
->state
= *shader
;
1296 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
1298 if (r300
->screen
->caps
.has_tcl
) {
1299 r300_translate_vertex_shader(r300
, vs
, vs
->state
.tokens
);
1301 vs
->draw_vs
= draw_create_vertex_shader(r300
->draw
, shader
);
1307 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1309 struct r300_context
* r300
= r300_context(pipe
);
1310 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1313 r300
->vs_state
.state
= NULL
;
1316 if (vs
== r300
->vs_state
.state
) {
1319 r300
->vs_state
.state
= vs
;
1321 // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
1323 r300_vertex_shader_setup_wpos(r300
);
1325 memcpy(r300
->vap_output_state
.state
, &vs
->vap_out
,
1326 sizeof(struct r300_vap_output_state
));
1327 r300
->vap_output_state
.dirty
= TRUE
;
1329 /* The majority of the RS block bits is dependent on the vertex shader. */
1330 r300
->rs_block_state
.dirty
= TRUE
; /* Will be updated before the emission. */
1332 if (r300
->screen
->caps
.has_tcl
) {
1333 r300
->vs_state
.dirty
= TRUE
;
1334 r300
->vs_state
.size
= vs
->code
.length
+ 9;
1336 r300
->pvs_flush
.dirty
= TRUE
;
1338 r300
->dirty_state
|= R300_NEW_VERTEX_SHADER_CONSTANTS
;
1340 draw_flush(r300
->draw
);
1341 draw_bind_vertex_shader(r300
->draw
,
1342 (struct draw_vertex_shader
*)vs
->draw_vs
);
1346 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1348 struct r300_context
* r300
= r300_context(pipe
);
1349 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1351 if (r300
->screen
->caps
.has_tcl
) {
1352 rc_constants_destroy(&vs
->code
.constants
);
1354 draw_delete_vertex_shader(r300
->draw
,
1355 (struct draw_vertex_shader
*)vs
->draw_vs
);
1358 FREE((void*)vs
->state
.tokens
);
1362 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1363 uint shader
, uint index
,
1364 struct pipe_resource
*buf
)
1366 struct r300_context
* r300
= r300_context(pipe
);
1367 struct pipe_transfer
*tr
;
1371 if (buf
== NULL
|| buf
->width0
== 0 ||
1372 (mapped
= pipe_buffer_map(pipe
, buf
, PIPE_TRANSFER_READ
, &tr
)) == NULL
)
1374 r300
->shader_constants
[shader
].count
= 0;
1378 assert((buf
->width0
% 4 * sizeof(float)) == 0);
1380 /* Check the size of the constant buffer. */
1382 case PIPE_SHADER_VERTEX
:
1385 case PIPE_SHADER_FRAGMENT
:
1386 if (r300
->screen
->caps
.is_r500
) {
1388 /* XXX Implement emission of r400's extended constant buffer. */
1389 /*} else if (r300->screen->caps.is_r400) {
1399 /* XXX Subtract immediates and RC_STATE_* variables. */
1400 if (buf
->width0
> (sizeof(float) * 4 * max_size
)) {
1401 fprintf(stderr
, "r300: Max size of the constant buffer is "
1402 "%i*4 floats.\n", max_size
);
1406 memcpy(r300
->shader_constants
[shader
].constants
, mapped
, buf
->width0
);
1407 r300
->shader_constants
[shader
].count
= buf
->width0
/ (4 * sizeof(float));
1408 pipe_buffer_unmap(pipe
, buf
, tr
);
1410 if (shader
== PIPE_SHADER_VERTEX
) {
1411 if (r300
->screen
->caps
.has_tcl
) {
1412 r300
->dirty_state
|= R300_NEW_VERTEX_SHADER_CONSTANTS
;
1413 r300
->pvs_flush
.dirty
= TRUE
;
1414 } else if (r300
->draw
) {
1415 draw_set_mapped_constant_buffer(r300
->draw
, PIPE_SHADER_VERTEX
,
1416 0, r300
->shader_constants
[PIPE_SHADER_VERTEX
].constants
,
1419 } else if (shader
== PIPE_SHADER_FRAGMENT
) {
1420 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1424 void r300_init_state_functions(struct r300_context
* r300
)
1426 r300
->context
.create_blend_state
= r300_create_blend_state
;
1427 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1428 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1430 r300
->context
.set_blend_color
= r300_set_blend_color
;
1432 r300
->context
.set_clip_state
= r300_set_clip_state
;
1434 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1436 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1437 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1438 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1440 r300
->context
.set_stencil_ref
= r300_set_stencil_ref
;
1442 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1444 r300
->context
.create_fs_state
= r300_create_fs_state
;
1445 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1446 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1448 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1450 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1451 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1452 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1454 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1455 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1456 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1457 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1459 r300
->context
.set_fragment_sampler_views
= r300_set_fragment_sampler_views
;
1460 r300
->context
.create_sampler_view
= r300_create_sampler_view
;
1461 r300
->context
.sampler_view_destroy
= r300_sampler_view_destroy
;
1463 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1465 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1467 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1469 r300
->context
.create_vertex_elements_state
= r300_create_vertex_elements_state
;
1470 r300
->context
.bind_vertex_elements_state
= r300_bind_vertex_elements_state
;
1471 r300
->context
.delete_vertex_elements_state
= r300_delete_vertex_elements_state
;
1473 r300
->context
.create_vs_state
= r300_create_vs_state
;
1474 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1475 r300
->context
.delete_vs_state
= r300_delete_vs_state
;