gallium/draw: use correct rasterization state for wide/AA points/lines
[mesa.git] / src / gallium / drivers / r300 / r300_state.c
1 /*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2009 Marek Olšák <maraeo@gmail.com>
4 *
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:
11 *
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
14 * Software.
15 *
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. */
23
24 #include "draw/draw_context.h"
25
26 #include "util/u_math.h"
27 #include "util/u_memory.h"
28 #include "util/u_pack_color.h"
29
30 #include "tgsi/tgsi_parse.h"
31
32 #include "pipe/p_config.h"
33
34 #include "r300_context.h"
35 #include "r300_reg.h"
36 #include "r300_screen.h"
37 #include "r300_state_inlines.h"
38 #include "r300_fs.h"
39 #include "r300_vs.h"
40
41 #include "radeon_winsys.h"
42
43 /* r300_state: Functions used to intialize state context by translating
44 * Gallium state objects into semi-native r300 state objects. */
45
46 #define UPDATE_STATE(cso, atom) \
47 if (cso != atom.state) { \
48 atom.state = cso; \
49 atom.dirty = TRUE; \
50 }
51
52 static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
53 unsigned dstRGB, unsigned dstA)
54 {
55 /* If the blend equation is ADD or REVERSE_SUBTRACT,
56 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
57 * will not be changed.
58 * Notice that the dst factors are the src factors inverted. */
59 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
60 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
61 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
62 (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
63 srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
64 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
65 srcA == PIPE_BLENDFACTOR_ZERO) &&
66 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
67 dstRGB == PIPE_BLENDFACTOR_ONE) &&
68 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
69 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
70 dstA == PIPE_BLENDFACTOR_ONE);
71 }
72
73 static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
74 unsigned dstRGB, unsigned dstA)
75 {
76 /* If the blend equation is ADD or REVERSE_SUBTRACT,
77 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
78 * will not be changed.
79 * Notice that the dst factors are the src factors inverted. */
80 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
81 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
82 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
83 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
84 srcA == PIPE_BLENDFACTOR_ZERO) &&
85 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
86 dstRGB == PIPE_BLENDFACTOR_ONE) &&
87 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
88 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
89 dstA == PIPE_BLENDFACTOR_ONE);
90 }
91
92 static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
93 unsigned dstRGB, unsigned dstA)
94 {
95 /* If the blend equation is ADD or REVERSE_SUBTRACT,
96 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
97 * will not be changed.
98 * Notice that the dst factors are the src factors inverted. */
99 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
100 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
101 (srcA == PIPE_BLENDFACTOR_ZERO) &&
102 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
103 dstRGB == PIPE_BLENDFACTOR_ONE) &&
104 (dstA == PIPE_BLENDFACTOR_ONE);
105 }
106
107 static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
108 unsigned dstRGB, unsigned dstA)
109 {
110 /* If the blend equation is ADD or REVERSE_SUBTRACT,
111 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
112 * will not be changed.
113 * Notice that the dst factors are the src factors inverted. */
114 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
115 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
116 (srcA == PIPE_BLENDFACTOR_ZERO) &&
117 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
118 dstRGB == PIPE_BLENDFACTOR_ONE) &&
119 (dstA == PIPE_BLENDFACTOR_ONE);
120 }
121
122 static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
123 unsigned dstRGB, unsigned dstA)
124 {
125 /* If the blend equation is ADD or REVERSE_SUBTRACT,
126 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
127 * the colorbuffer will not be changed.
128 * Notice that the dst factors are the src factors inverted. */
129 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
130 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
131 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
132 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
133 (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
134 srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
135 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
136 srcA == PIPE_BLENDFACTOR_ZERO) &&
137 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
138 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
139 dstRGB == PIPE_BLENDFACTOR_ONE) &&
140 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
141 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
142 dstA == PIPE_BLENDFACTOR_ONE);
143 }
144
145 static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
146 unsigned dstRGB, unsigned dstA)
147 {
148 /* If the blend equation is ADD or REVERSE_SUBTRACT,
149 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
150 * the colorbuffer will not be changed.
151 * Notice that the dst factors are the src factors inverted. */
152 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
153 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
154 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
155 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
156 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
157 srcA == PIPE_BLENDFACTOR_ZERO) &&
158 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
159 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
160 dstRGB == PIPE_BLENDFACTOR_ONE) &&
161 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
162 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
163 dstA == PIPE_BLENDFACTOR_ONE);
164 }
165
166 static unsigned bgra_cmask(unsigned mask)
167 {
168 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
169
170 return ((mask & PIPE_MASK_R) << 2) |
171 ((mask & PIPE_MASK_B) >> 2) |
172 (mask & (PIPE_MASK_G | PIPE_MASK_A));
173 }
174
175 /* Create a new blend state based on the CSO blend state.
176 *
177 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
178 static void* r300_create_blend_state(struct pipe_context* pipe,
179 const struct pipe_blend_state* state)
180 {
181 struct r300_screen* r300screen = r300_screen(pipe->screen);
182 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
183
184 if (state->rt[0].blend_enable)
185 {
186 unsigned eqRGB = state->rt[0].rgb_func;
187 unsigned srcRGB = state->rt[0].rgb_src_factor;
188 unsigned dstRGB = state->rt[0].rgb_dst_factor;
189
190 unsigned eqA = state->rt[0].alpha_func;
191 unsigned srcA = state->rt[0].alpha_src_factor;
192 unsigned dstA = state->rt[0].alpha_dst_factor;
193
194 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
195 * this is just the crappy D3D naming */
196 blend->blend_control = R300_ALPHA_BLEND_ENABLE |
197 r300_translate_blend_function(eqRGB) |
198 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
199 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
200
201 /* Optimization: some operations do not require the destination color.
202 *
203 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
204 * otherwise blending gives incorrect results. It seems to be
205 * a hardware bug. */
206 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
207 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
208 dstRGB != PIPE_BLENDFACTOR_ZERO ||
209 dstA != PIPE_BLENDFACTOR_ZERO ||
210 srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
211 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
212 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
213 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
214 srcA == PIPE_BLENDFACTOR_DST_COLOR ||
215 srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
216 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
217 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
218 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
219 /* Enable reading from the colorbuffer. */
220 blend->blend_control |= R300_READ_ENABLE;
221
222 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) {
223 /* Optimization: Depending on incoming pixels, we can
224 * conditionally disable the reading in hardware... */
225 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
226 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
227 /* Disable reading if SRC_ALPHA == 0. */
228 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
229 dstRGB == PIPE_BLENDFACTOR_ZERO) &&
230 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
231 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
232 dstA == PIPE_BLENDFACTOR_ZERO)) {
233 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ;
234 }
235
236 /* Disable reading if SRC_ALPHA == 1. */
237 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
238 dstRGB == PIPE_BLENDFACTOR_ZERO) &&
239 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
240 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
241 dstA == PIPE_BLENDFACTOR_ZERO)) {
242 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ;
243 }
244 }
245 }
246 }
247
248 /* Optimization: discard pixels which don't change the colorbuffer.
249 *
250 * The code below is non-trivial and some math is involved.
251 *
252 * Discarding pixels must be disabled when FP16 AA is enabled.
253 * This is a hardware bug. Also, this implementation wouldn't work
254 * with FP blending enabled and equation clamping disabled.
255 *
256 * Equations other than ADD are rarely used and therefore won't be
257 * optimized. */
258 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
259 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
260 /* ADD: X+Y
261 * REVERSE_SUBTRACT: Y-X
262 *
263 * The idea is:
264 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
265 * then CB will not be changed.
266 *
267 * Given the srcFactor and dstFactor variables, we can derive
268 * what src and dst should be equal to and discard appropriate
269 * pixels.
270 */
271 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
272 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
273 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
274 dstRGB, dstA)) {
275 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
276 } else if (blend_discard_if_src_color_0(srcRGB, srcA,
277 dstRGB, dstA)) {
278 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
279 } else if (blend_discard_if_src_color_1(srcRGB, srcA,
280 dstRGB, dstA)) {
281 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
282 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
283 dstRGB, dstA)) {
284 blend->blend_control |=
285 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
286 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
287 dstRGB, dstA)) {
288 blend->blend_control |=
289 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
290 }
291 }
292
293 /* separate alpha */
294 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
295 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE;
296 blend->alpha_blend_control =
297 r300_translate_blend_function(eqA) |
298 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
299 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
300 }
301 }
302
303 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
304 if (state->logicop_enable) {
305 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
306 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
307 }
308
309 /* Color channel masks for all MRTs. */
310 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask);
311 if (r300screen->caps->is_r500 && state->independent_blend_enable) {
312 if (state->rt[1].blend_enable) {
313 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4;
314 }
315 if (state->rt[2].blend_enable) {
316 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8;
317 }
318 if (state->rt[3].blend_enable) {
319 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12;
320 }
321 }
322
323 if (state->dither) {
324 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
325 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
326 }
327
328 return (void*)blend;
329 }
330
331 /* Bind blend state. */
332 static void r300_bind_blend_state(struct pipe_context* pipe,
333 void* state)
334 {
335 struct r300_context* r300 = r300_context(pipe);
336
337 UPDATE_STATE(state, r300->blend_state);
338 }
339
340 /* Free blend state. */
341 static void r300_delete_blend_state(struct pipe_context* pipe,
342 void* state)
343 {
344 FREE(state);
345 }
346
347 /* Convert float to 10bit integer */
348 static unsigned float_to_fixed10(float f)
349 {
350 return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
351 }
352
353 /* Set blend color.
354 * Setup both R300 and R500 registers, figure out later which one to write. */
355 static void r300_set_blend_color(struct pipe_context* pipe,
356 const struct pipe_blend_color* color)
357 {
358 struct r300_context* r300 = r300_context(pipe);
359 struct r300_screen* r300screen = r300_screen(pipe->screen);
360 struct r300_blend_color_state* state =
361 (struct r300_blend_color_state*)r300->blend_color_state.state;
362 union util_color uc;
363
364 util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
365 state->blend_color = uc.ui;
366
367 /* XXX if FP16 blending is enabled, we should use the FP16 format */
368 state->blend_color_red_alpha =
369 float_to_fixed10(color->color[0]) |
370 (float_to_fixed10(color->color[3]) << 16);
371 state->blend_color_green_blue =
372 float_to_fixed10(color->color[2]) |
373 (float_to_fixed10(color->color[1]) << 16);
374
375 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2;
376 r300->blend_color_state.dirty = TRUE;
377 }
378
379 static void r300_set_clip_state(struct pipe_context* pipe,
380 const struct pipe_clip_state* state)
381 {
382 struct r300_context* r300 = r300_context(pipe);
383
384 r300->clip = *state;
385
386 if (r300_screen(pipe->screen)->caps->has_tcl) {
387 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
388 r300->clip_state.size = 29;
389 } else {
390 draw_flush(r300->draw);
391 draw_set_clip_state(r300->draw, state);
392 r300->clip_state.size = 2;
393 }
394
395 r300->clip_state.dirty = TRUE;
396 }
397
398 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
399 *
400 * This contains the depth buffer, stencil buffer, alpha test, and such.
401 * On the Radeon, depth and stencil buffer setup are intertwined, which is
402 * the reason for some of the strange-looking assignments across registers. */
403 static void*
404 r300_create_dsa_state(struct pipe_context* pipe,
405 const struct pipe_depth_stencil_alpha_state* state)
406 {
407 struct r300_capabilities *caps =
408 r300_screen(r300_context(pipe)->context.screen)->caps;
409 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
410
411 /* Depth test setup. */
412 if (state->depth.enabled) {
413 dsa->z_buffer_control |= R300_Z_ENABLE;
414
415 if (state->depth.writemask) {
416 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
417 }
418
419 dsa->z_stencil_control |=
420 (r300_translate_depth_stencil_function(state->depth.func) <<
421 R300_Z_FUNC_SHIFT);
422 }
423
424 /* Stencil buffer setup. */
425 if (state->stencil[0].enabled) {
426 dsa->z_buffer_control |= R300_STENCIL_ENABLE;
427 dsa->z_stencil_control |=
428 (r300_translate_depth_stencil_function(state->stencil[0].func) <<
429 R300_S_FRONT_FUNC_SHIFT) |
430 (r300_translate_stencil_op(state->stencil[0].fail_op) <<
431 R300_S_FRONT_SFAIL_OP_SHIFT) |
432 (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
433 R300_S_FRONT_ZPASS_OP_SHIFT) |
434 (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
435 R300_S_FRONT_ZFAIL_OP_SHIFT);
436
437 dsa->stencil_ref_mask =
438 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
439 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
440
441 if (state->stencil[1].enabled) {
442 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
443 dsa->z_stencil_control |=
444 (r300_translate_depth_stencil_function(state->stencil[1].func) <<
445 R300_S_BACK_FUNC_SHIFT) |
446 (r300_translate_stencil_op(state->stencil[1].fail_op) <<
447 R300_S_BACK_SFAIL_OP_SHIFT) |
448 (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
449 R300_S_BACK_ZPASS_OP_SHIFT) |
450 (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
451 R300_S_BACK_ZFAIL_OP_SHIFT);
452
453 if (caps->is_r500)
454 {
455 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
456 dsa->stencil_ref_bf =
457 (state->stencil[1].valuemask <<
458 R300_STENCILMASK_SHIFT) |
459 (state->stencil[1].writemask <<
460 R300_STENCILWRITEMASK_SHIFT);
461 }
462 }
463 }
464
465 /* Alpha test setup. */
466 if (state->alpha.enabled) {
467 dsa->alpha_function =
468 r300_translate_alpha_function(state->alpha.func) |
469 R300_FG_ALPHA_FUNC_ENABLE;
470
471 /* We could use 10bit alpha ref but who needs that? */
472 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
473
474 if (caps->is_r500)
475 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
476 }
477
478 return (void*)dsa;
479 }
480
481 /* Bind DSA state. */
482 static void r300_bind_dsa_state(struct pipe_context* pipe,
483 void* state)
484 {
485 struct r300_context* r300 = r300_context(pipe);
486
487 UPDATE_STATE(state, r300->dsa_state);
488 }
489
490 /* Free DSA state. */
491 static void r300_delete_dsa_state(struct pipe_context* pipe,
492 void* state)
493 {
494 FREE(state);
495 }
496
497 static void r300_set_stencil_ref(struct pipe_context* pipe,
498 const struct pipe_stencil_ref* sr)
499 {
500 struct r300_context* r300 = r300_context(pipe);
501 r300->stencil_ref = *sr;
502 r300->dsa_state.dirty = TRUE;
503 }
504
505 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
506 static void r300_fb_update_tiling_flags(struct r300_context *r300,
507 const struct pipe_framebuffer_state *old_state,
508 const struct pipe_framebuffer_state *new_state)
509 {
510 struct r300_texture *tex;
511 unsigned i, j, level;
512
513 /* Reset tiling flags for old surfaces to default values. */
514 for (i = 0; i < old_state->nr_cbufs; i++) {
515 for (j = 0; j < new_state->nr_cbufs; j++) {
516 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) {
517 break;
518 }
519 }
520 /* If not binding the surface again... */
521 if (j != new_state->nr_cbufs) {
522 continue;
523 }
524
525 tex = (struct r300_texture*)old_state->cbufs[i]->texture;
526
527 if (tex) {
528 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
529 tex->pitch[0],
530 tex->microtile != 0,
531 tex->macrotile != 0);
532 }
533 }
534 if (old_state->zsbuf &&
535 (!new_state->zsbuf ||
536 old_state->zsbuf->texture != new_state->zsbuf->texture)) {
537 tex = (struct r300_texture*)old_state->zsbuf->texture;
538
539 if (tex) {
540 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
541 tex->pitch[0],
542 tex->microtile != 0,
543 tex->macrotile != 0);
544 }
545 }
546
547 /* Set tiling flags for new surfaces. */
548 for (i = 0; i < new_state->nr_cbufs; i++) {
549 tex = (struct r300_texture*)new_state->cbufs[i]->texture;
550 level = new_state->cbufs[i]->level;
551
552 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
553 tex->pitch[level],
554 tex->microtile != 0,
555 tex->mip_macrotile[level] != 0);
556 }
557 if (new_state->zsbuf) {
558 tex = (struct r300_texture*)new_state->zsbuf->texture;
559 level = new_state->zsbuf->level;
560
561 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
562 tex->pitch[level],
563 tex->microtile != 0,
564 tex->mip_macrotile[level] != 0);
565 }
566 }
567
568 static void
569 r300_set_framebuffer_state(struct pipe_context* pipe,
570 const struct pipe_framebuffer_state* state)
571 {
572 struct r300_context* r300 = r300_context(pipe);
573 struct r300_screen* r300screen = r300_screen(pipe->screen);
574 unsigned max_width, max_height;
575 uint32_t zbuffer_bpp = 0;
576
577
578 if (state->nr_cbufs > 4) {
579 debug_printf("r300: Implementation error: Too many MRTs in %s, "
580 "refusing to bind framebuffer state!\n", __FUNCTION__);
581 return;
582 }
583
584 if (r300screen->caps->is_r500) {
585 max_width = max_height = 4096;
586 } else if (r300screen->caps->is_r400) {
587 max_width = max_height = 4021;
588 } else {
589 max_width = max_height = 2560;
590 }
591
592 if (state->width > max_width || state->height > max_height) {
593 debug_printf("r300: Implementation error: Render targets are too "
594 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__);
595 return;
596 }
597
598
599 if (r300->draw) {
600 draw_flush(r300->draw);
601 }
602
603 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
604
605 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) +
606 (state->zsbuf ? 10 : 0) + 8;
607
608 r300_fb_update_tiling_flags(r300, r300->fb_state.state, state);
609
610 /* XXX wait what */
611 r300->blend_state.dirty = TRUE;
612 r300->dsa_state.dirty = TRUE;
613 r300->fb_state.dirty = TRUE;
614 r300->scissor_state.dirty = TRUE;
615
616 /* Polygon offset depends on the zbuffer bit depth. */
617 if (state->zsbuf && r300->polygon_offset_enabled) {
618 switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
619 case 2:
620 zbuffer_bpp = 16;
621 break;
622 case 4:
623 zbuffer_bpp = 24;
624 break;
625 }
626
627 if (r300->zbuffer_bpp != zbuffer_bpp) {
628 r300->zbuffer_bpp = zbuffer_bpp;
629 r300->rs_state.dirty = TRUE;
630 }
631 }
632 }
633
634 /* Create fragment shader state. */
635 static void* r300_create_fs_state(struct pipe_context* pipe,
636 const struct pipe_shader_state* shader)
637 {
638 struct r300_fragment_shader* fs = NULL;
639
640 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
641
642 /* Copy state directly into shader. */
643 fs->state = *shader;
644 fs->state.tokens = tgsi_dup_tokens(shader->tokens);
645
646 tgsi_scan_shader(shader->tokens, &fs->info);
647 r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
648
649 return (void*)fs;
650 }
651
652 /* Bind fragment shader state. */
653 static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
654 {
655 struct r300_context* r300 = r300_context(pipe);
656 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
657
658 if (fs == NULL) {
659 r300->fs = NULL;
660 return;
661 }
662
663 r300->fs = fs;
664 r300_pick_fragment_shader(r300);
665
666 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
667
668 if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) {
669 r300->vap_output_state.dirty = TRUE;
670 }
671
672 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
673 }
674
675 /* Delete fragment shader state. */
676 static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
677 {
678 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
679 struct r300_fragment_shader_code *tmp, *ptr = fs->first;
680
681 while (ptr) {
682 tmp = ptr;
683 ptr = ptr->next;
684 rc_constants_destroy(&tmp->code.constants);
685 FREE(tmp);
686 }
687 FREE((void*)fs->state.tokens);
688 FREE(shader);
689 }
690
691 static void r300_set_polygon_stipple(struct pipe_context* pipe,
692 const struct pipe_poly_stipple* state)
693 {
694 /* XXX no idea how to set this up, but not terribly important */
695 }
696
697 /* Create a new rasterizer state based on the CSO rasterizer state.
698 *
699 * This is a very large chunk of state, and covers most of the graphics
700 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
701 *
702 * In a not entirely unironic sidenote, this state has nearly nothing to do
703 * with the actual block on the Radeon called the rasterizer (RS). */
704 static void* r300_create_rs_state(struct pipe_context* pipe,
705 const struct pipe_rasterizer_state* state)
706 {
707 struct r300_screen* r300screen = r300_screen(pipe->screen);
708 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
709
710 /* Copy rasterizer state for Draw. */
711 rs->rs = *state;
712
713 #ifdef PIPE_ARCH_LITTLE_ENDIAN
714 rs->vap_control_status = R300_VC_NO_SWAP;
715 #else
716 rs->vap_control_status = R300_VC_32BIT_SWAP;
717 #endif
718
719 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
720 * Else, enable HW TCL and force Draw's TCL off. */
721 if (!r300screen->caps->has_tcl) {
722 rs->vap_control_status |= R300_VAP_TCL_BYPASS;
723 }
724
725 rs->point_size = pack_float_16_6x(state->point_size) |
726 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
727
728 rs->line_control = pack_float_16_6x(state->line_width) |
729 R300_GA_LINE_CNTL_END_TYPE_COMP;
730
731 /* Enable polygon mode */
732 if (state->fill_cw != PIPE_POLYGON_MODE_FILL ||
733 state->fill_ccw != PIPE_POLYGON_MODE_FILL) {
734 rs->polygon_mode = R300_GA_POLY_MODE_DUAL;
735 }
736
737 /* Radeons don't think in "CW/CCW", they think in "front/back". */
738 if (state->front_winding == PIPE_WINDING_CW) {
739 rs->cull_mode = R300_FRONT_FACE_CW;
740
741 /* Polygon offset */
742 if (state->offset_cw) {
743 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
744 }
745 if (state->offset_ccw) {
746 rs->polygon_offset_enable |= R300_BACK_ENABLE;
747 }
748
749 /* Polygon mode */
750 if (rs->polygon_mode) {
751 rs->polygon_mode |=
752 r300_translate_polygon_mode_front(state->fill_cw);
753 rs->polygon_mode |=
754 r300_translate_polygon_mode_back(state->fill_ccw);
755 }
756 } else {
757 rs->cull_mode = R300_FRONT_FACE_CCW;
758
759 /* Polygon offset */
760 if (state->offset_ccw) {
761 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
762 }
763 if (state->offset_cw) {
764 rs->polygon_offset_enable |= R300_BACK_ENABLE;
765 }
766
767 /* Polygon mode */
768 if (rs->polygon_mode) {
769 rs->polygon_mode |=
770 r300_translate_polygon_mode_front(state->fill_ccw);
771 rs->polygon_mode |=
772 r300_translate_polygon_mode_back(state->fill_cw);
773 }
774 }
775 if (state->front_winding & state->cull_mode) {
776 rs->cull_mode |= R300_CULL_FRONT;
777 }
778 if (~(state->front_winding) & state->cull_mode) {
779 rs->cull_mode |= R300_CULL_BACK;
780 }
781
782 if (rs->polygon_offset_enable) {
783 rs->depth_offset = state->offset_units;
784 rs->depth_scale = state->offset_scale;
785 }
786
787 if (state->line_stipple_enable) {
788 rs->line_stipple_config =
789 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
790 (fui((float)state->line_stipple_factor) &
791 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
792 /* XXX this might need to be scaled up */
793 rs->line_stipple_value = state->line_stipple_pattern;
794 }
795
796 if (state->flatshade) {
797 rs->color_control = R300_SHADE_MODEL_FLAT;
798 } else {
799 rs->color_control = R300_SHADE_MODEL_SMOOTH;
800 }
801
802 return (void*)rs;
803 }
804
805 /* Bind rasterizer state. */
806 static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
807 {
808 struct r300_context* r300 = r300_context(pipe);
809 struct r300_rs_state* rs = (struct r300_rs_state*)state;
810
811 if (r300->draw) {
812 draw_flush(r300->draw);
813 draw_set_rasterizer_state(r300->draw, &rs->rs, state);
814 }
815
816 if (rs) {
817 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw;
818 } else {
819 r300->polygon_offset_enabled = FALSE;
820 }
821
822 UPDATE_STATE(state, r300->rs_state);
823 r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0);
824
825 /* XXX Why is this still needed, dammit!? */
826 r300->scissor_state.dirty = TRUE;
827 r300->viewport_state.dirty = TRUE;
828
829 /* XXX Clean these up when we move to atom emits */
830 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
831 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
832 }
833 }
834
835 /* Free rasterizer state. */
836 static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
837 {
838 FREE(state);
839 }
840
841 static void*
842 r300_create_sampler_state(struct pipe_context* pipe,
843 const struct pipe_sampler_state* state)
844 {
845 struct r300_context* r300 = r300_context(pipe);
846 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
847 int lod_bias;
848 union util_color uc;
849
850 sampler->state = *state;
851
852 sampler->filter0 |=
853 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
854 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
855 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
856
857 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
858 state->mag_img_filter,
859 state->min_mip_filter,
860 state->max_anisotropy > 0);
861
862 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
863 /* We must pass these to the merge function to clamp them properly. */
864 sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
865 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
866
867 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
868
869 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
870
871 sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
872
873 util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
874 sampler->border_color = uc.ui;
875
876 /* R500-specific fixups and optimizations */
877 if (r300_screen(r300->context.screen)->caps->is_r500) {
878 sampler->filter1 |= R500_BORDER_FIX;
879 }
880
881 return (void*)sampler;
882 }
883
884 static void r300_bind_sampler_states(struct pipe_context* pipe,
885 unsigned count,
886 void** states)
887 {
888 struct r300_context* r300 = r300_context(pipe);
889 struct r300_textures_state* state =
890 (struct r300_textures_state*)r300->textures_state.state;
891
892 if (count > 8) {
893 return;
894 }
895
896 memcpy(state->sampler_states, states, sizeof(void*) * count);
897 state->sampler_count = count;
898
899 r300->textures_state.dirty = TRUE;
900
901 /* Pick a fragment shader based on the texture compare state. */
902 if (r300->fs && count) {
903 if (r300_pick_fragment_shader(r300)) {
904 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
905 R300_NEW_FRAGMENT_SHADER_CONSTANTS;
906 }
907 }
908 }
909
910 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
911 unsigned count,
912 void** states)
913 {
914 }
915
916 static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
917 {
918 FREE(state);
919 }
920
921 static void r300_set_sampler_textures(struct pipe_context* pipe,
922 unsigned count,
923 struct pipe_texture** texture)
924 {
925 struct r300_context* r300 = r300_context(pipe);
926 struct r300_textures_state* state =
927 (struct r300_textures_state*)r300->textures_state.state;
928 unsigned i;
929 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
930 boolean dirty_tex = FALSE;
931
932 /* XXX magic num */
933 if (count > 8) {
934 return;
935 }
936
937 for (i = 0; i < count; i++) {
938 if (state->textures[i] != (struct r300_texture*)texture[i]) {
939 pipe_texture_reference((struct pipe_texture**)&state->textures[i],
940 texture[i]);
941 dirty_tex = TRUE;
942
943 /* R300-specific - set the texrect factor in the fragment shader */
944 if (!is_r500 && state->textures[i]->is_npot) {
945 /* XXX It would be nice to re-emit just 1 constant,
946 * XXX not all of them */
947 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
948 }
949 }
950 }
951
952 for (i = count; i < 8; i++) {
953 if (state->textures[i]) {
954 pipe_texture_reference((struct pipe_texture**)&state->textures[i],
955 NULL);
956 }
957 }
958
959 state->texture_count = count;
960
961 r300->textures_state.dirty = TRUE;
962
963 if (dirty_tex) {
964 r300->texture_cache_inval.dirty = TRUE;
965 }
966 }
967
968 static void r300_set_scissor_state(struct pipe_context* pipe,
969 const struct pipe_scissor_state* state)
970 {
971 struct r300_context* r300 = r300_context(pipe);
972
973 memcpy(r300->scissor_state.state, state,
974 sizeof(struct pipe_scissor_state));
975
976 r300->scissor_state.dirty = TRUE;
977 }
978
979 static void r300_set_viewport_state(struct pipe_context* pipe,
980 const struct pipe_viewport_state* state)
981 {
982 struct r300_context* r300 = r300_context(pipe);
983 struct r300_viewport_state* viewport =
984 (struct r300_viewport_state*)r300->viewport_state.state;
985
986 r300->viewport = *state;
987
988 /* Do the transform in HW. */
989 viewport->vte_control = R300_VTX_W0_FMT;
990
991 if (state->scale[0] != 1.0f) {
992 viewport->xscale = state->scale[0];
993 viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
994 }
995 if (state->scale[1] != 1.0f) {
996 viewport->yscale = state->scale[1];
997 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
998 }
999 if (state->scale[2] != 1.0f) {
1000 viewport->zscale = state->scale[2];
1001 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
1002 }
1003 if (state->translate[0] != 0.0f) {
1004 viewport->xoffset = state->translate[0];
1005 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
1006 }
1007 if (state->translate[1] != 0.0f) {
1008 viewport->yoffset = state->translate[1];
1009 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
1010 }
1011 if (state->translate[2] != 0.0f) {
1012 viewport->zoffset = state->translate[2];
1013 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
1014 }
1015
1016 r300->viewport_state.dirty = TRUE;
1017 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
1018 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1019 }
1020 }
1021
1022 static void r300_set_vertex_buffers(struct pipe_context* pipe,
1023 unsigned count,
1024 const struct pipe_vertex_buffer* buffers)
1025 {
1026 struct r300_context* r300 = r300_context(pipe);
1027 unsigned i, max_index = ~0;
1028
1029 memcpy(r300->vertex_buffer, buffers,
1030 sizeof(struct pipe_vertex_buffer) * count);
1031
1032 for (i = 0; i < count; i++) {
1033 max_index = MIN2(buffers[i].max_index, max_index);
1034 }
1035
1036 r300->vertex_buffer_count = count;
1037 r300->vertex_buffer_max_index = max_index;
1038
1039 if (r300->draw) {
1040 draw_flush(r300->draw);
1041 draw_set_vertex_buffers(r300->draw, count, buffers);
1042 } else {
1043 r300->vertex_stream_state.dirty = TRUE;
1044 }
1045 }
1046
1047 static boolean r300_validate_aos(struct r300_context *r300)
1048 {
1049 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
1050 struct pipe_vertex_element *velem = r300->vertex_element;
1051 int i;
1052
1053 /* Check if formats and strides are aligned to the size of DWORD. */
1054 for (i = 0; i < r300->vertex_element_count; i++) {
1055 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
1056 util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
1057 return FALSE;
1058 }
1059 }
1060 return TRUE;
1061 }
1062
1063 static void r300_set_vertex_elements(struct pipe_context* pipe,
1064 unsigned count,
1065 const struct pipe_vertex_element* elements)
1066 {
1067 struct r300_context* r300 = r300_context(pipe);
1068
1069 memcpy(r300->vertex_element,
1070 elements,
1071 sizeof(struct pipe_vertex_element) * count);
1072 r300->vertex_element_count = count;
1073
1074 if (r300->draw) {
1075 draw_flush(r300->draw);
1076 draw_set_vertex_elements(r300->draw, count, elements);
1077 }
1078
1079 if (!r300_validate_aos(r300)) {
1080 /* XXX We should fallback using draw. */
1081 assert(0);
1082 abort();
1083 }
1084 }
1085
1086 static void* r300_create_vs_state(struct pipe_context* pipe,
1087 const struct pipe_shader_state* shader)
1088 {
1089 struct r300_context* r300 = r300_context(pipe);
1090
1091 if (r300_screen(pipe->screen)->caps->has_tcl) {
1092 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
1093 /* Copy state directly into shader. */
1094 vs->state = *shader;
1095 vs->state.tokens = tgsi_dup_tokens(shader->tokens);
1096
1097 tgsi_scan_shader(shader->tokens, &vs->info);
1098
1099 return (void*)vs;
1100 } else {
1101 return draw_create_vertex_shader(r300->draw, shader);
1102 }
1103 }
1104
1105 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
1106 {
1107 struct r300_context* r300 = r300_context(pipe);
1108
1109 if (r300_screen(pipe->screen)->caps->has_tcl) {
1110 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1111
1112 if (vs == NULL) {
1113 r300->vs_state.state = NULL;
1114 return;
1115 } else if (!vs->translated) {
1116 r300_translate_vertex_shader(r300, vs);
1117 }
1118
1119 UPDATE_STATE(shader, r300->vs_state);
1120 r300->vs_state.size = vs->code.length + 9;
1121
1122 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
1123 r300->vap_output_state.dirty = TRUE;
1124 r300->vertex_stream_state.dirty = TRUE; /* XXX needed for TCL bypass */
1125 r300->pvs_flush.dirty = TRUE;
1126
1127 if (r300->fs) {
1128 r300_vertex_shader_setup_wpos(r300);
1129 }
1130
1131 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
1132 } else {
1133 draw_flush(r300->draw);
1134 draw_bind_vertex_shader(r300->draw,
1135 (struct draw_vertex_shader*)shader);
1136 }
1137 }
1138
1139 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1140 {
1141 struct r300_context* r300 = r300_context(pipe);
1142
1143 if (r300_screen(pipe->screen)->caps->has_tcl) {
1144 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1145
1146 rc_constants_destroy(&vs->code.constants);
1147 FREE((void*)vs->state.tokens);
1148 FREE(shader);
1149 } else {
1150 draw_delete_vertex_shader(r300->draw,
1151 (struct draw_vertex_shader*)shader);
1152 }
1153 }
1154
1155 static void r300_set_constant_buffer(struct pipe_context *pipe,
1156 uint shader, uint index,
1157 struct pipe_buffer *buf)
1158 {
1159 struct r300_context* r300 = r300_context(pipe);
1160 struct r300_screen *r300screen = r300_screen(pipe->screen);
1161 void *mapped;
1162 int max_size = 0;
1163
1164 if (buf == NULL || buf->size == 0 ||
1165 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
1166 {
1167 r300->shader_constants[shader].count = 0;
1168 return;
1169 }
1170
1171 assert((buf->size % 4 * sizeof(float)) == 0);
1172
1173 /* Check the size of the constant buffer. */
1174 switch (shader) {
1175 case PIPE_SHADER_VERTEX:
1176 max_size = 256;
1177 break;
1178 case PIPE_SHADER_FRAGMENT:
1179 if (r300screen->caps->is_r500) {
1180 max_size = 256;
1181 /* XXX Implement emission of r400's extended constant buffer. */
1182 /*} else if (r300screen->caps->is_r400) {
1183 max_size = 64;*/
1184 } else {
1185 max_size = 32;
1186 }
1187 break;
1188 default:
1189 assert(0);
1190 }
1191
1192 /* XXX Subtract immediates and RC_STATE_* variables. */
1193 if (buf->size > (sizeof(float) * 4 * max_size)) {
1194 debug_printf("r300: Max size of the constant buffer is "
1195 "%i*4 floats.\n", max_size);
1196 abort();
1197 }
1198
1199 memcpy(r300->shader_constants[shader].constants, mapped, buf->size);
1200 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float));
1201 pipe_buffer_unmap(pipe->screen, buf);
1202
1203 if (shader == PIPE_SHADER_VERTEX) {
1204 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
1205 r300->pvs_flush.dirty = TRUE;
1206 }
1207 else if (shader == PIPE_SHADER_FRAGMENT)
1208 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1209 }
1210
1211 void r300_init_state_functions(struct r300_context* r300)
1212 {
1213 r300->context.create_blend_state = r300_create_blend_state;
1214 r300->context.bind_blend_state = r300_bind_blend_state;
1215 r300->context.delete_blend_state = r300_delete_blend_state;
1216
1217 r300->context.set_blend_color = r300_set_blend_color;
1218
1219 r300->context.set_clip_state = r300_set_clip_state;
1220
1221 r300->context.set_constant_buffer = r300_set_constant_buffer;
1222
1223 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1224 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1225 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1226
1227 r300->context.set_stencil_ref = r300_set_stencil_ref;
1228
1229 r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1230
1231 r300->context.create_fs_state = r300_create_fs_state;
1232 r300->context.bind_fs_state = r300_bind_fs_state;
1233 r300->context.delete_fs_state = r300_delete_fs_state;
1234
1235 r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1236
1237 r300->context.create_rasterizer_state = r300_create_rs_state;
1238 r300->context.bind_rasterizer_state = r300_bind_rs_state;
1239 r300->context.delete_rasterizer_state = r300_delete_rs_state;
1240
1241 r300->context.create_sampler_state = r300_create_sampler_state;
1242 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1243 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1244 r300->context.delete_sampler_state = r300_delete_sampler_state;
1245
1246 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures;
1247
1248 r300->context.set_scissor_state = r300_set_scissor_state;
1249
1250 r300->context.set_viewport_state = r300_set_viewport_state;
1251
1252 r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1253 r300->context.set_vertex_elements = r300_set_vertex_elements;
1254
1255 r300->context.create_vs_state = r300_create_vs_state;
1256 r300->context.bind_vs_state = r300_bind_vs_state;
1257 r300->context.delete_vs_state = r300_delete_vs_state;
1258 }