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_state_inlines.h"
41 /* r300_state: Functions used to intialize state context by translating
42 * Gallium state objects into semi-native r300 state objects. */
44 static boolean
blend_discard_if_src_alpha_0(unsigned srcRGB
, unsigned srcA
,
45 unsigned dstRGB
, unsigned dstA
)
47 /* If the blend equation is ADD or REVERSE_SUBTRACT,
48 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
49 * will not be changed.
50 * Notice that the dst factors are the src factors inverted. */
51 return (srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
52 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
53 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
54 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
55 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
56 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
57 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
58 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
59 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
60 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
61 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
62 dstA
== PIPE_BLENDFACTOR_ONE
);
65 static boolean
blend_discard_if_src_alpha_1(unsigned srcRGB
, unsigned srcA
,
66 unsigned dstRGB
, unsigned dstA
)
68 /* If the blend equation is ADD or REVERSE_SUBTRACT,
69 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
70 * will not be changed.
71 * Notice that the dst factors are the src factors inverted. */
72 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
73 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
74 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
75 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
76 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
77 (dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
78 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
79 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
80 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
81 dstA
== PIPE_BLENDFACTOR_ONE
);
84 static boolean
blend_discard_if_src_color_0(unsigned srcRGB
, unsigned srcA
,
85 unsigned dstRGB
, unsigned dstA
)
87 /* If the blend equation is ADD or REVERSE_SUBTRACT,
88 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
89 * will not be changed.
90 * Notice that the dst factors are the src factors inverted. */
91 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
92 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
93 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
94 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
95 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
96 (dstA
== PIPE_BLENDFACTOR_ONE
);
99 static boolean
blend_discard_if_src_color_1(unsigned srcRGB
, unsigned srcA
,
100 unsigned dstRGB
, unsigned dstA
)
102 /* If the blend equation is ADD or REVERSE_SUBTRACT,
103 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
104 * will not be changed.
105 * Notice that the dst factors are the src factors inverted. */
106 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
107 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
108 (srcA
== PIPE_BLENDFACTOR_ZERO
) &&
109 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
110 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
111 (dstA
== PIPE_BLENDFACTOR_ONE
);
114 static boolean
blend_discard_if_src_alpha_color_0(unsigned srcRGB
, unsigned srcA
,
115 unsigned dstRGB
, unsigned dstA
)
117 /* If the blend equation is ADD or REVERSE_SUBTRACT,
118 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
119 * the colorbuffer will not be changed.
120 * Notice that the dst factors are the src factors inverted. */
121 return (srcRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
122 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
123 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
124 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
125 (srcA
== PIPE_BLENDFACTOR_SRC_COLOR
||
126 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
127 srcA
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
||
128 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
129 (dstRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
130 dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
131 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
132 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
133 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
134 dstA
== PIPE_BLENDFACTOR_ONE
);
137 static boolean
blend_discard_if_src_alpha_color_1(unsigned srcRGB
, unsigned srcA
,
138 unsigned dstRGB
, unsigned dstA
)
140 /* If the blend equation is ADD or REVERSE_SUBTRACT,
141 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
142 * the colorbuffer will not be changed.
143 * Notice that the dst factors are the src factors inverted. */
144 return (srcRGB
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
145 srcRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
146 srcRGB
== PIPE_BLENDFACTOR_ZERO
) &&
147 (srcA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
148 srcA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
149 srcA
== PIPE_BLENDFACTOR_ZERO
) &&
150 (dstRGB
== PIPE_BLENDFACTOR_SRC_COLOR
||
151 dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
152 dstRGB
== PIPE_BLENDFACTOR_ONE
) &&
153 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
154 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
155 dstA
== PIPE_BLENDFACTOR_ONE
);
158 static unsigned bgra_cmask(unsigned mask
)
160 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
162 return ((mask
& PIPE_MASK_R
) << 2) |
163 ((mask
& PIPE_MASK_B
) >> 2) |
164 (mask
& (PIPE_MASK_G
| PIPE_MASK_A
));
167 /* Create a new blend state based on the CSO blend state.
169 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
170 static void* r300_create_blend_state(struct pipe_context
* pipe
,
171 const struct pipe_blend_state
* state
)
173 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
174 struct r300_blend_state
* blend
= CALLOC_STRUCT(r300_blend_state
);
176 if (state
->rt
[0].blend_enable
)
178 unsigned eqRGB
= state
->rt
[0].rgb_func
;
179 unsigned srcRGB
= state
->rt
[0].rgb_src_factor
;
180 unsigned dstRGB
= state
->rt
[0].rgb_dst_factor
;
182 unsigned eqA
= state
->rt
[0].alpha_func
;
183 unsigned srcA
= state
->rt
[0].alpha_src_factor
;
184 unsigned dstA
= state
->rt
[0].alpha_dst_factor
;
186 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
187 * this is just the crappy D3D naming */
188 blend
->blend_control
= R300_ALPHA_BLEND_ENABLE
|
189 r300_translate_blend_function(eqRGB
) |
190 ( r300_translate_blend_factor(srcRGB
) << R300_SRC_BLEND_SHIFT
) |
191 ( r300_translate_blend_factor(dstRGB
) << R300_DST_BLEND_SHIFT
);
193 /* Optimization: some operations do not require the destination color.
195 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
196 * otherwise blending gives incorrect results. It seems to be
198 if (eqRGB
== PIPE_BLEND_MIN
|| eqA
== PIPE_BLEND_MIN
||
199 eqRGB
== PIPE_BLEND_MAX
|| eqA
== PIPE_BLEND_MAX
||
200 dstRGB
!= PIPE_BLENDFACTOR_ZERO
||
201 dstA
!= PIPE_BLENDFACTOR_ZERO
||
202 srcRGB
== PIPE_BLENDFACTOR_DST_COLOR
||
203 srcRGB
== PIPE_BLENDFACTOR_DST_ALPHA
||
204 srcRGB
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
205 srcRGB
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
206 srcA
== PIPE_BLENDFACTOR_DST_COLOR
||
207 srcA
== PIPE_BLENDFACTOR_DST_ALPHA
||
208 srcA
== PIPE_BLENDFACTOR_INV_DST_COLOR
||
209 srcA
== PIPE_BLENDFACTOR_INV_DST_ALPHA
||
210 srcRGB
== PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE
) {
211 /* Enable reading from the colorbuffer. */
212 blend
->blend_control
|= R300_READ_ENABLE
;
214 if (r300_screen(r300_context(pipe
)->context
.screen
)->caps
->is_r500
) {
215 /* Optimization: Depending on incoming pixels, we can
216 * conditionally disable the reading in hardware... */
217 if (eqRGB
!= PIPE_BLEND_MIN
&& eqA
!= PIPE_BLEND_MIN
&&
218 eqRGB
!= PIPE_BLEND_MAX
&& eqA
!= PIPE_BLEND_MAX
) {
219 /* Disable reading if SRC_ALPHA == 0. */
220 if ((dstRGB
== PIPE_BLENDFACTOR_SRC_ALPHA
||
221 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
222 (dstA
== PIPE_BLENDFACTOR_SRC_COLOR
||
223 dstA
== PIPE_BLENDFACTOR_SRC_ALPHA
||
224 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
225 blend
->blend_control
|= R500_SRC_ALPHA_0_NO_READ
;
228 /* Disable reading if SRC_ALPHA == 1. */
229 if ((dstRGB
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
230 dstRGB
== PIPE_BLENDFACTOR_ZERO
) &&
231 (dstA
== PIPE_BLENDFACTOR_INV_SRC_COLOR
||
232 dstA
== PIPE_BLENDFACTOR_INV_SRC_ALPHA
||
233 dstA
== PIPE_BLENDFACTOR_ZERO
)) {
234 blend
->blend_control
|= R500_SRC_ALPHA_1_NO_READ
;
240 /* Optimization: discard pixels which don't change the colorbuffer.
242 * The code below is non-trivial and some math is involved.
244 * Discarding pixels must be disabled when FP16 AA is enabled.
245 * This is a hardware bug. Also, this implementation wouldn't work
246 * with FP blending enabled and equation clamping disabled.
248 * Equations other than ADD are rarely used and therefore won't be
250 if ((eqRGB
== PIPE_BLEND_ADD
|| eqRGB
== PIPE_BLEND_REVERSE_SUBTRACT
) &&
251 (eqA
== PIPE_BLEND_ADD
|| eqA
== PIPE_BLEND_REVERSE_SUBTRACT
)) {
253 * REVERSE_SUBTRACT: Y-X
256 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
257 * then CB will not be changed.
259 * Given the srcFactor and dstFactor variables, we can derive
260 * what src and dst should be equal to and discard appropriate
263 if (blend_discard_if_src_alpha_0(srcRGB
, srcA
, dstRGB
, dstA
)) {
264 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0
;
265 } else if (blend_discard_if_src_alpha_1(srcRGB
, srcA
,
267 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1
;
268 } else if (blend_discard_if_src_color_0(srcRGB
, srcA
,
270 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0
;
271 } else if (blend_discard_if_src_color_1(srcRGB
, srcA
,
273 blend
->blend_control
|= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1
;
274 } else if (blend_discard_if_src_alpha_color_0(srcRGB
, srcA
,
276 blend
->blend_control
|=
277 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0
;
278 } else if (blend_discard_if_src_alpha_color_1(srcRGB
, srcA
,
280 blend
->blend_control
|=
281 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1
;
286 if (srcA
!= srcRGB
|| dstA
!= dstRGB
|| eqA
!= eqRGB
) {
287 blend
->blend_control
|= R300_SEPARATE_ALPHA_ENABLE
;
288 blend
->alpha_blend_control
=
289 r300_translate_blend_function(eqA
) |
290 (r300_translate_blend_factor(srcA
) << R300_SRC_BLEND_SHIFT
) |
291 (r300_translate_blend_factor(dstA
) << R300_DST_BLEND_SHIFT
);
295 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
296 if (state
->logicop_enable
) {
297 blend
->rop
= R300_RB3D_ROPCNTL_ROP_ENABLE
|
298 (state
->logicop_func
) << R300_RB3D_ROPCNTL_ROP_SHIFT
;
301 /* Color channel masks for all MRTs. */
302 blend
->color_channel_mask
= bgra_cmask(state
->rt
[0].colormask
);
303 if (r300screen
->caps
->is_r500
&& state
->independent_blend_enable
) {
304 if (state
->rt
[1].blend_enable
) {
305 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[1].colormask
) << 4;
307 if (state
->rt
[2].blend_enable
) {
308 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[2].colormask
) << 8;
310 if (state
->rt
[3].blend_enable
) {
311 blend
->color_channel_mask
|= bgra_cmask(state
->rt
[3].colormask
) << 12;
316 blend
->dither
= R300_RB3D_DITHER_CTL_DITHER_MODE_LUT
|
317 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT
;
323 /* Bind blend state. */
324 static void r300_bind_blend_state(struct pipe_context
* pipe
,
327 struct r300_context
* r300
= r300_context(pipe
);
329 r300
->blend_state
.state
= state
;
330 r300
->blend_state
.dirty
= TRUE
;
333 /* Free blend state. */
334 static void r300_delete_blend_state(struct pipe_context
* pipe
,
340 /* Convert float to 10bit integer */
341 static unsigned float_to_fixed10(float f
)
343 return CLAMP((unsigned)(f
* 1023.9f
), 0, 1023);
347 * Setup both R300 and R500 registers, figure out later which one to write. */
348 static void r300_set_blend_color(struct pipe_context
* pipe
,
349 const struct pipe_blend_color
* color
)
351 struct r300_context
* r300
= r300_context(pipe
);
352 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
353 struct r300_blend_color_state
* state
=
354 (struct r300_blend_color_state
*)r300
->blend_color_state
.state
;
357 util_pack_color(color
->color
, PIPE_FORMAT_A8R8G8B8_UNORM
, &uc
);
358 state
->blend_color
= uc
.ui
;
360 /* XXX if FP16 blending is enabled, we should use the FP16 format */
361 state
->blend_color_red_alpha
=
362 float_to_fixed10(color
->color
[0]) |
363 (float_to_fixed10(color
->color
[3]) << 16);
364 state
->blend_color_green_blue
=
365 float_to_fixed10(color
->color
[2]) |
366 (float_to_fixed10(color
->color
[1]) << 16);
368 r300
->blend_color_state
.size
= r300screen
->caps
->is_r500
? 3 : 2;
369 r300
->blend_color_state
.dirty
= TRUE
;
372 static void r300_set_clip_state(struct pipe_context
* pipe
,
373 const struct pipe_clip_state
* state
)
375 struct r300_context
* r300
= r300_context(pipe
);
377 if (r300_screen(pipe
->screen
)->caps
->has_tcl
) {
378 memcpy(r300
->clip_state
.state
, state
, sizeof(struct pipe_clip_state
));
379 r300
->clip_state
.size
= 29;
381 draw_flush(r300
->draw
);
382 draw_set_clip_state(r300
->draw
, state
);
383 r300
->clip_state
.size
= 2;
386 r300
->clip_state
.dirty
= TRUE
;
389 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
391 * This contains the depth buffer, stencil buffer, alpha test, and such.
392 * On the Radeon, depth and stencil buffer setup are intertwined, which is
393 * the reason for some of the strange-looking assignments across registers. */
395 r300_create_dsa_state(struct pipe_context
* pipe
,
396 const struct pipe_depth_stencil_alpha_state
* state
)
398 struct r300_capabilities
*caps
=
399 r300_screen(r300_context(pipe
)->context
.screen
)->caps
;
400 struct r300_dsa_state
* dsa
= CALLOC_STRUCT(r300_dsa_state
);
402 /* Depth test setup. */
403 if (state
->depth
.enabled
) {
404 dsa
->z_buffer_control
|= R300_Z_ENABLE
;
406 if (state
->depth
.writemask
) {
407 dsa
->z_buffer_control
|= R300_Z_WRITE_ENABLE
;
410 dsa
->z_stencil_control
|=
411 (r300_translate_depth_stencil_function(state
->depth
.func
) <<
415 /* Stencil buffer setup. */
416 if (state
->stencil
[0].enabled
) {
417 dsa
->z_buffer_control
|= R300_STENCIL_ENABLE
;
418 dsa
->z_stencil_control
|=
419 (r300_translate_depth_stencil_function(state
->stencil
[0].func
) <<
420 R300_S_FRONT_FUNC_SHIFT
) |
421 (r300_translate_stencil_op(state
->stencil
[0].fail_op
) <<
422 R300_S_FRONT_SFAIL_OP_SHIFT
) |
423 (r300_translate_stencil_op(state
->stencil
[0].zpass_op
) <<
424 R300_S_FRONT_ZPASS_OP_SHIFT
) |
425 (r300_translate_stencil_op(state
->stencil
[0].zfail_op
) <<
426 R300_S_FRONT_ZFAIL_OP_SHIFT
);
428 dsa
->stencil_ref_mask
= (state
->stencil
[0].ref_value
) |
429 (state
->stencil
[0].valuemask
<< R300_STENCILMASK_SHIFT
) |
430 (state
->stencil
[0].writemask
<< R300_STENCILWRITEMASK_SHIFT
);
432 if (state
->stencil
[1].enabled
) {
433 dsa
->z_buffer_control
|= R300_STENCIL_FRONT_BACK
;
434 dsa
->z_stencil_control
|=
435 (r300_translate_depth_stencil_function(state
->stencil
[1].func
) <<
436 R300_S_BACK_FUNC_SHIFT
) |
437 (r300_translate_stencil_op(state
->stencil
[1].fail_op
) <<
438 R300_S_BACK_SFAIL_OP_SHIFT
) |
439 (r300_translate_stencil_op(state
->stencil
[1].zpass_op
) <<
440 R300_S_BACK_ZPASS_OP_SHIFT
) |
441 (r300_translate_stencil_op(state
->stencil
[1].zfail_op
) <<
442 R300_S_BACK_ZFAIL_OP_SHIFT
);
446 dsa
->z_buffer_control
|= R500_STENCIL_REFMASK_FRONT_BACK
;
447 dsa
->stencil_ref_bf
= (state
->stencil
[1].ref_value
) |
448 (state
->stencil
[1].valuemask
<<
449 R300_STENCILMASK_SHIFT
) |
450 (state
->stencil
[1].writemask
<<
451 R300_STENCILWRITEMASK_SHIFT
);
456 /* Alpha test setup. */
457 if (state
->alpha
.enabled
) {
458 dsa
->alpha_function
=
459 r300_translate_alpha_function(state
->alpha
.func
) |
460 R300_FG_ALPHA_FUNC_ENABLE
;
462 /* We could use 10bit alpha ref but who needs that? */
463 dsa
->alpha_function
|= float_to_ubyte(state
->alpha
.ref_value
);
466 dsa
->alpha_function
|= R500_FG_ALPHA_FUNC_8BIT
;
472 /* Bind DSA state. */
473 static void r300_bind_dsa_state(struct pipe_context
* pipe
,
476 struct r300_context
* r300
= r300_context(pipe
);
477 struct r300_screen
* r300screen
= r300_screen(pipe
->screen
);
479 r300
->dsa_state
.state
= state
;
480 r300
->dsa_state
.size
= r300screen
->caps
->is_r500
? 8 : 6;
481 r300
->dsa_state
.dirty
= TRUE
;
484 /* Free DSA state. */
485 static void r300_delete_dsa_state(struct pipe_context
* pipe
,
492 r300_set_framebuffer_state(struct pipe_context
* pipe
,
493 const struct pipe_framebuffer_state
* state
)
495 struct r300_context
* r300
= r300_context(pipe
);
496 uint32_t zbuffer_bpp
= 0;
498 r300
->fb_state
.size
= (10 * state
->nr_cbufs
) +
499 (2 * (4 - state
->nr_cbufs
)) +
500 (state
->zsbuf
? 10 : 0) + 6;
502 if (state
->nr_cbufs
> 4) {
503 debug_printf("r300: Implementation error: Too many MRTs in %s, "
504 "refusing to bind framebuffer state!\n", __FUNCTION__
);
509 draw_flush(r300
->draw
);
512 memcpy(r300
->fb_state
.state
, state
, sizeof(struct pipe_framebuffer_state
));
514 /* Don't rely on the order of states being set for the first time. */
516 r300
->blend_state
.dirty
= TRUE
;
517 r300
->dsa_state
.dirty
= TRUE
;
518 r300
->fb_state
.dirty
= TRUE
;
519 r300
->scissor_state
.dirty
= TRUE
;
521 /* Polygon offset depends on the zbuffer bit depth. */
522 if (state
->zsbuf
&& r300
->polygon_offset_enabled
) {
523 switch (util_format_get_blocksize(state
->zsbuf
->texture
->format
)) {
532 if (r300
->zbuffer_bpp
!= zbuffer_bpp
) {
533 r300
->zbuffer_bpp
= zbuffer_bpp
;
534 r300
->rs_state
.dirty
= TRUE
;
539 /* Create fragment shader state. */
540 static void* r300_create_fs_state(struct pipe_context
* pipe
,
541 const struct pipe_shader_state
* shader
)
543 struct r300_fragment_shader
* fs
= NULL
;
545 fs
= (struct r300_fragment_shader
*)CALLOC_STRUCT(r300_fragment_shader
);
547 /* Copy state directly into shader. */
549 fs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
551 tgsi_scan_shader(shader
->tokens
, &fs
->info
);
552 r300_shader_read_fs_inputs(&fs
->info
, &fs
->inputs
);
557 /* Bind fragment shader state. */
558 static void r300_bind_fs_state(struct pipe_context
* pipe
, void* shader
)
560 struct r300_context
* r300
= r300_context(pipe
);
561 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
569 r300_pick_fragment_shader(r300
);
571 if (r300
->vs
&& r300_vertex_shader_setup_wpos(r300
)) {
572 r300
->vertex_format_state
.dirty
= TRUE
;
575 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER
| R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
578 /* Delete fragment shader state. */
579 static void r300_delete_fs_state(struct pipe_context
* pipe
, void* shader
)
581 struct r300_fragment_shader
* fs
= (struct r300_fragment_shader
*)shader
;
582 struct r300_fragment_shader_code
*tmp
, *ptr
= fs
->first
;
587 rc_constants_destroy(&tmp
->code
.constants
);
590 FREE((void*)fs
->state
.tokens
);
594 static void r300_set_polygon_stipple(struct pipe_context
* pipe
,
595 const struct pipe_poly_stipple
* state
)
597 /* XXX no idea how to set this up, but not terribly important */
600 /* Create a new rasterizer state based on the CSO rasterizer state.
602 * This is a very large chunk of state, and covers most of the graphics
603 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
605 * In a not entirely unironic sidenote, this state has nearly nothing to do
606 * with the actual block on the Radeon called the rasterizer (RS). */
607 static void* r300_create_rs_state(struct pipe_context
* pipe
,
608 const struct pipe_rasterizer_state
* state
)
610 struct r300_rs_state
* rs
= CALLOC_STRUCT(r300_rs_state
);
612 /* Copy rasterizer state for Draw. */
615 #ifdef PIPE_ARCH_LITTLE_ENDIAN
616 rs
->vap_control_status
= R300_VC_NO_SWAP
;
618 rs
->vap_control_status
= R300_VC_32BIT_SWAP
;
621 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
622 * Else, enable HW TCL and force Draw's TCL off. */
623 if (state
->bypass_vs_clip_and_viewport
||
624 !r300_screen(pipe
->screen
)->caps
->has_tcl
) {
625 rs
->vap_control_status
|= R300_VAP_TCL_BYPASS
;
628 rs
->point_size
= pack_float_16_6x(state
->point_size
) |
629 (pack_float_16_6x(state
->point_size
) << R300_POINTSIZE_X_SHIFT
);
632 ((int)(state
->point_size_min
* 6.0) <<
633 R300_GA_POINT_MINMAX_MIN_SHIFT
) |
634 ((int)(state
->point_size_max
* 6.0) <<
635 R300_GA_POINT_MINMAX_MAX_SHIFT
);
637 rs
->line_control
= pack_float_16_6x(state
->line_width
) |
638 R300_GA_LINE_CNTL_END_TYPE_COMP
;
640 /* Enable polygon mode */
641 if (state
->fill_cw
!= PIPE_POLYGON_MODE_FILL
||
642 state
->fill_ccw
!= PIPE_POLYGON_MODE_FILL
) {
643 rs
->polygon_mode
= R300_GA_POLY_MODE_DUAL
;
646 /* Radeons don't think in "CW/CCW", they think in "front/back". */
647 if (state
->front_winding
== PIPE_WINDING_CW
) {
648 rs
->cull_mode
= R300_FRONT_FACE_CW
;
651 if (state
->offset_cw
) {
652 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
654 if (state
->offset_ccw
) {
655 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
659 if (rs
->polygon_mode
) {
661 r300_translate_polygon_mode_front(state
->fill_cw
);
663 r300_translate_polygon_mode_back(state
->fill_ccw
);
666 rs
->cull_mode
= R300_FRONT_FACE_CCW
;
669 if (state
->offset_ccw
) {
670 rs
->polygon_offset_enable
|= R300_FRONT_ENABLE
;
672 if (state
->offset_cw
) {
673 rs
->polygon_offset_enable
|= R300_BACK_ENABLE
;
677 if (rs
->polygon_mode
) {
679 r300_translate_polygon_mode_front(state
->fill_ccw
);
681 r300_translate_polygon_mode_back(state
->fill_cw
);
684 if (state
->front_winding
& state
->cull_mode
) {
685 rs
->cull_mode
|= R300_CULL_FRONT
;
687 if (~(state
->front_winding
) & state
->cull_mode
) {
688 rs
->cull_mode
|= R300_CULL_BACK
;
691 if (rs
->polygon_offset_enable
) {
692 rs
->depth_offset
= state
->offset_units
;
693 rs
->depth_scale
= state
->offset_scale
;
696 if (state
->line_stipple_enable
) {
697 rs
->line_stipple_config
=
698 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE
|
699 (fui((float)state
->line_stipple_factor
) &
700 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK
);
701 /* XXX this might need to be scaled up */
702 rs
->line_stipple_value
= state
->line_stipple_pattern
;
705 if (state
->flatshade
) {
706 rs
->color_control
= R300_SHADE_MODEL_FLAT
;
708 rs
->color_control
= R300_SHADE_MODEL_SMOOTH
;
714 /* Bind rasterizer state. */
715 static void r300_bind_rs_state(struct pipe_context
* pipe
, void* state
)
717 struct r300_context
* r300
= r300_context(pipe
);
718 struct r300_rs_state
* rs
= (struct r300_rs_state
*)state
;
721 draw_flush(r300
->draw
);
722 draw_set_rasterizer_state(r300
->draw
, &rs
->rs
);
726 r300
->tcl_bypass
= rs
->rs
.bypass_vs_clip_and_viewport
;
727 r300
->polygon_offset_enabled
= rs
->rs
.offset_cw
|| rs
->rs
.offset_ccw
;
729 r300
->tcl_bypass
= FALSE
;
730 r300
->polygon_offset_enabled
= FALSE
;
733 r300
->rs_state
.state
= rs
;
734 r300
->rs_state
.dirty
= TRUE
;
735 /* XXX Why is this still needed, dammit!? */
736 r300
->scissor_state
.dirty
= TRUE
;
737 r300
->viewport_state
.dirty
= TRUE
;
739 /* XXX Clean these up when we move to atom emits */
740 if (r300
->fs
&& r300
->fs
->inputs
.wpos
!= ATTR_UNUSED
) {
741 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
745 /* Free rasterizer state. */
746 static void r300_delete_rs_state(struct pipe_context
* pipe
, void* state
)
752 r300_create_sampler_state(struct pipe_context
* pipe
,
753 const struct pipe_sampler_state
* state
)
755 struct r300_context
* r300
= r300_context(pipe
);
756 struct r300_sampler_state
* sampler
= CALLOC_STRUCT(r300_sampler_state
);
760 sampler
->state
= *state
;
763 (r300_translate_wrap(state
->wrap_s
) << R300_TX_WRAP_S_SHIFT
) |
764 (r300_translate_wrap(state
->wrap_t
) << R300_TX_WRAP_T_SHIFT
) |
765 (r300_translate_wrap(state
->wrap_r
) << R300_TX_WRAP_R_SHIFT
);
767 sampler
->filter0
|= r300_translate_tex_filters(state
->min_img_filter
,
768 state
->mag_img_filter
,
769 state
->min_mip_filter
,
770 state
->max_anisotropy
> 1.0);
772 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
773 /* We must pass these to the emit function to clamp them properly. */
774 sampler
->min_lod
= MAX2((unsigned)state
->min_lod
, 0);
775 sampler
->max_lod
= MAX2((unsigned)ceilf(state
->max_lod
), 0);
777 lod_bias
= CLAMP((int)(state
->lod_bias
* 32), -(1 << 9), (1 << 9) - 1);
779 sampler
->filter1
|= lod_bias
<< R300_LOD_BIAS_SHIFT
;
781 sampler
->filter1
|= r300_anisotropy(state
->max_anisotropy
);
783 util_pack_color(state
->border_color
, PIPE_FORMAT_A8R8G8B8_UNORM
, &uc
);
784 sampler
->border_color
= uc
.ui
;
786 /* R500-specific fixups and optimizations */
787 if (r300_screen(r300
->context
.screen
)->caps
->is_r500
) {
788 sampler
->filter1
|= R500_BORDER_FIX
;
791 return (void*)sampler
;
794 static void r300_bind_sampler_states(struct pipe_context
* pipe
,
798 struct r300_context
* r300
= r300_context(pipe
);
805 for (i
= 0; i
< count
; i
++) {
806 if (r300
->sampler_states
[i
] != states
[i
]) {
807 r300
->sampler_states
[i
] = (struct r300_sampler_state
*)states
[i
];
808 r300
->dirty_state
|= (R300_NEW_SAMPLER
<< i
);
812 r300
->sampler_count
= count
;
814 /* Pick a fragment shader based on the texture compare state. */
815 if (r300
->fs
&& (r300
->dirty_state
& R300_ANY_NEW_SAMPLERS
)) {
816 if (r300_pick_fragment_shader(r300
)) {
817 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER
|
818 R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
823 static void r300_lacks_vertex_textures(struct pipe_context
* pipe
,
829 static void r300_delete_sampler_state(struct pipe_context
* pipe
, void* state
)
834 static void r300_set_sampler_textures(struct pipe_context
* pipe
,
836 struct pipe_texture
** texture
)
838 struct r300_context
* r300
= r300_context(pipe
);
839 boolean is_r500
= r300_screen(r300
->context
.screen
)->caps
->is_r500
;
847 for (i
= 0; i
< count
; i
++) {
848 if (r300
->textures
[i
] != (struct r300_texture
*)texture
[i
]) {
849 pipe_texture_reference((struct pipe_texture
**)&r300
->textures
[i
],
851 r300
->dirty_state
|= (R300_NEW_TEXTURE
<< i
);
853 /* R300-specific - set the texrect factor in a fragment shader */
854 if (!is_r500
&& r300
->textures
[i
]->is_npot
) {
855 /* XXX It would be nice to re-emit just 1 constant,
856 * XXX not all of them */
857 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
862 for (i
= count
; i
< 8; i
++) {
863 if (r300
->textures
[i
]) {
864 pipe_texture_reference((struct pipe_texture
**)&r300
->textures
[i
],
866 r300
->dirty_state
|= (R300_NEW_TEXTURE
<< i
);
870 r300
->texture_count
= count
;
873 static void r300_set_scissor_state(struct pipe_context
* pipe
,
874 const struct pipe_scissor_state
* state
)
876 struct r300_context
* r300
= r300_context(pipe
);
878 memcpy(r300
->scissor_state
.state
, state
,
879 sizeof(struct pipe_scissor_state
));
881 r300
->scissor_state
.dirty
= TRUE
;
884 static void r300_set_viewport_state(struct pipe_context
* pipe
,
885 const struct pipe_viewport_state
* state
)
887 struct r300_context
* r300
= r300_context(pipe
);
888 struct r300_viewport_state
* viewport
=
889 (struct r300_viewport_state
*)r300
->viewport_state
.state
;
891 /* Do the transform in HW. */
892 viewport
->vte_control
= R300_VTX_W0_FMT
;
894 if (state
->scale
[0] != 1.0f
) {
895 viewport
->xscale
= state
->scale
[0];
896 viewport
->vte_control
|= R300_VPORT_X_SCALE_ENA
;
898 if (state
->scale
[1] != 1.0f
) {
899 viewport
->yscale
= state
->scale
[1];
900 viewport
->vte_control
|= R300_VPORT_Y_SCALE_ENA
;
902 if (state
->scale
[2] != 1.0f
) {
903 viewport
->zscale
= state
->scale
[2];
904 viewport
->vte_control
|= R300_VPORT_Z_SCALE_ENA
;
906 if (state
->translate
[0] != 0.0f
) {
907 viewport
->xoffset
= state
->translate
[0];
908 viewport
->vte_control
|= R300_VPORT_X_OFFSET_ENA
;
910 if (state
->translate
[1] != 0.0f
) {
911 viewport
->yoffset
= state
->translate
[1];
912 viewport
->vte_control
|= R300_VPORT_Y_OFFSET_ENA
;
914 if (state
->translate
[2] != 0.0f
) {
915 viewport
->zoffset
= state
->translate
[2];
916 viewport
->vte_control
|= R300_VPORT_Z_OFFSET_ENA
;
919 r300
->viewport_state
.dirty
= TRUE
;
920 if (r300
->fs
&& r300
->fs
->inputs
.wpos
!= ATTR_UNUSED
) {
921 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
925 static void r300_set_vertex_buffers(struct pipe_context
* pipe
,
927 const struct pipe_vertex_buffer
* buffers
)
929 struct r300_context
* r300
= r300_context(pipe
);
931 memcpy(r300
->vertex_buffer
, buffers
,
932 sizeof(struct pipe_vertex_buffer
) * count
);
933 r300
->vertex_buffer_count
= count
;
936 draw_flush(r300
->draw
);
937 draw_set_vertex_buffers(r300
->draw
, count
, buffers
);
940 r300
->vertex_format_state
.dirty
= TRUE
;
943 static boolean
r300_validate_aos(struct r300_context
*r300
)
945 struct pipe_vertex_buffer
*vbuf
= r300
->vertex_buffer
;
946 struct pipe_vertex_element
*velem
= r300
->vertex_element
;
949 /* Check if formats and strides are aligned to the size of DWORD. */
950 for (i
= 0; i
< r300
->vertex_element_count
; i
++) {
951 if (vbuf
[velem
[i
].vertex_buffer_index
].stride
% 4 != 0 ||
952 util_format_get_blocksize(velem
[i
].src_format
) % 4 != 0) {
959 static void r300_set_vertex_elements(struct pipe_context
* pipe
,
961 const struct pipe_vertex_element
* elements
)
963 struct r300_context
* r300
= r300_context(pipe
);
965 memcpy(r300
->vertex_element
,
967 sizeof(struct pipe_vertex_element
) * count
);
968 r300
->vertex_element_count
= count
;
971 draw_flush(r300
->draw
);
972 draw_set_vertex_elements(r300
->draw
, count
, elements
);
975 if (!r300_validate_aos(r300
)) {
976 /* XXX We should fallback using draw. */
982 static void* r300_create_vs_state(struct pipe_context
* pipe
,
983 const struct pipe_shader_state
* shader
)
985 struct r300_context
* r300
= r300_context(pipe
);
987 if (r300_screen(pipe
->screen
)->caps
->has_tcl
) {
988 struct r300_vertex_shader
* vs
= CALLOC_STRUCT(r300_vertex_shader
);
989 /* Copy state directly into shader. */
991 vs
->state
.tokens
= tgsi_dup_tokens(shader
->tokens
);
993 tgsi_scan_shader(shader
->tokens
, &vs
->info
);
997 return draw_create_vertex_shader(r300
->draw
, shader
);
1001 static void r300_bind_vs_state(struct pipe_context
* pipe
, void* shader
)
1003 struct r300_context
* r300
= r300_context(pipe
);
1005 if (r300_screen(pipe
->screen
)->caps
->has_tcl
) {
1006 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1011 } else if (!vs
->translated
) {
1012 r300_translate_vertex_shader(r300
, vs
);
1017 r300_vertex_shader_setup_wpos(r300
);
1020 r300
->vertex_format_state
.dirty
= TRUE
;
1022 r300
->dirty_state
|=
1023 R300_NEW_VERTEX_SHADER
| R300_NEW_VERTEX_SHADER_CONSTANTS
;
1025 draw_flush(r300
->draw
);
1026 draw_bind_vertex_shader(r300
->draw
,
1027 (struct draw_vertex_shader
*)shader
);
1031 static void r300_delete_vs_state(struct pipe_context
* pipe
, void* shader
)
1033 struct r300_context
* r300
= r300_context(pipe
);
1035 if (r300_screen(pipe
->screen
)->caps
->has_tcl
) {
1036 struct r300_vertex_shader
* vs
= (struct r300_vertex_shader
*)shader
;
1038 rc_constants_destroy(&vs
->code
.constants
);
1039 FREE((void*)vs
->state
.tokens
);
1042 draw_delete_vertex_shader(r300
->draw
,
1043 (struct draw_vertex_shader
*)shader
);
1047 static void r300_set_constant_buffer(struct pipe_context
*pipe
,
1048 uint shader
, uint index
,
1049 struct pipe_buffer
*buf
)
1051 struct r300_context
* r300
= r300_context(pipe
);
1054 if (buf
== NULL
|| buf
->size
== 0 ||
1055 (mapped
= pipe_buffer_map(pipe
->screen
, buf
, PIPE_BUFFER_USAGE_CPU_READ
)) == NULL
)
1057 r300
->shader_constants
[shader
].count
= 0;
1061 assert((buf
->size
% 4 * sizeof(float)) == 0);
1062 memcpy(r300
->shader_constants
[shader
].constants
, mapped
, buf
->size
);
1063 r300
->shader_constants
[shader
].count
= buf
->size
/ (4 * sizeof(float));
1064 pipe_buffer_unmap(pipe
->screen
, buf
);
1066 if (shader
== PIPE_SHADER_VERTEX
)
1067 r300
->dirty_state
|= R300_NEW_VERTEX_SHADER_CONSTANTS
;
1068 else if (shader
== PIPE_SHADER_FRAGMENT
)
1069 r300
->dirty_state
|= R300_NEW_FRAGMENT_SHADER_CONSTANTS
;
1072 void r300_init_state_functions(struct r300_context
* r300
)
1074 r300
->context
.create_blend_state
= r300_create_blend_state
;
1075 r300
->context
.bind_blend_state
= r300_bind_blend_state
;
1076 r300
->context
.delete_blend_state
= r300_delete_blend_state
;
1078 r300
->context
.set_blend_color
= r300_set_blend_color
;
1080 r300
->context
.set_clip_state
= r300_set_clip_state
;
1082 r300
->context
.set_constant_buffer
= r300_set_constant_buffer
;
1084 r300
->context
.create_depth_stencil_alpha_state
= r300_create_dsa_state
;
1085 r300
->context
.bind_depth_stencil_alpha_state
= r300_bind_dsa_state
;
1086 r300
->context
.delete_depth_stencil_alpha_state
= r300_delete_dsa_state
;
1088 r300
->context
.set_framebuffer_state
= r300_set_framebuffer_state
;
1090 r300
->context
.create_fs_state
= r300_create_fs_state
;
1091 r300
->context
.bind_fs_state
= r300_bind_fs_state
;
1092 r300
->context
.delete_fs_state
= r300_delete_fs_state
;
1094 r300
->context
.set_polygon_stipple
= r300_set_polygon_stipple
;
1096 r300
->context
.create_rasterizer_state
= r300_create_rs_state
;
1097 r300
->context
.bind_rasterizer_state
= r300_bind_rs_state
;
1098 r300
->context
.delete_rasterizer_state
= r300_delete_rs_state
;
1100 r300
->context
.create_sampler_state
= r300_create_sampler_state
;
1101 r300
->context
.bind_fragment_sampler_states
= r300_bind_sampler_states
;
1102 r300
->context
.bind_vertex_sampler_states
= r300_lacks_vertex_textures
;
1103 r300
->context
.delete_sampler_state
= r300_delete_sampler_state
;
1105 r300
->context
.set_fragment_sampler_textures
= r300_set_sampler_textures
;
1107 r300
->context
.set_scissor_state
= r300_set_scissor_state
;
1109 r300
->context
.set_viewport_state
= r300_set_viewport_state
;
1111 r300
->context
.set_vertex_buffers
= r300_set_vertex_buffers
;
1112 r300
->context
.set_vertex_elements
= r300_set_vertex_elements
;
1114 r300
->context
.create_vs_state
= r300_create_vs_state
;
1115 r300
->context
.bind_vs_state
= r300_bind_vs_state
;
1116 r300
->context
.delete_vs_state
= r300_delete_vs_state
;