r300g: atomize texture and sampler states
[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 static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
47 unsigned dstRGB, unsigned dstA)
48 {
49 /* If the blend equation is ADD or REVERSE_SUBTRACT,
50 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
51 * will not be changed.
52 * Notice that the dst factors are the src factors inverted. */
53 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
54 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
55 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
56 (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
57 srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
58 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
59 srcA == PIPE_BLENDFACTOR_ZERO) &&
60 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
61 dstRGB == PIPE_BLENDFACTOR_ONE) &&
62 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
63 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
64 dstA == PIPE_BLENDFACTOR_ONE);
65 }
66
67 static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
68 unsigned dstRGB, unsigned dstA)
69 {
70 /* If the blend equation is ADD or REVERSE_SUBTRACT,
71 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
72 * will not be changed.
73 * Notice that the dst factors are the src factors inverted. */
74 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
75 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
76 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
77 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
78 srcA == PIPE_BLENDFACTOR_ZERO) &&
79 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
80 dstRGB == PIPE_BLENDFACTOR_ONE) &&
81 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
82 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
83 dstA == PIPE_BLENDFACTOR_ONE);
84 }
85
86 static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
87 unsigned dstRGB, unsigned dstA)
88 {
89 /* If the blend equation is ADD or REVERSE_SUBTRACT,
90 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
91 * will not be changed.
92 * Notice that the dst factors are the src factors inverted. */
93 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
94 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
95 (srcA == PIPE_BLENDFACTOR_ZERO) &&
96 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
97 dstRGB == PIPE_BLENDFACTOR_ONE) &&
98 (dstA == PIPE_BLENDFACTOR_ONE);
99 }
100
101 static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
102 unsigned dstRGB, unsigned dstA)
103 {
104 /* If the blend equation is ADD or REVERSE_SUBTRACT,
105 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
106 * will not be changed.
107 * Notice that the dst factors are the src factors inverted. */
108 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
109 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
110 (srcA == PIPE_BLENDFACTOR_ZERO) &&
111 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
112 dstRGB == PIPE_BLENDFACTOR_ONE) &&
113 (dstA == PIPE_BLENDFACTOR_ONE);
114 }
115
116 static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
117 unsigned dstRGB, unsigned dstA)
118 {
119 /* If the blend equation is ADD or REVERSE_SUBTRACT,
120 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
121 * the colorbuffer will not be changed.
122 * Notice that the dst factors are the src factors inverted. */
123 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
124 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
125 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
126 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
127 (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
128 srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
129 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
130 srcA == PIPE_BLENDFACTOR_ZERO) &&
131 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
132 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
133 dstRGB == PIPE_BLENDFACTOR_ONE) &&
134 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
135 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
136 dstA == PIPE_BLENDFACTOR_ONE);
137 }
138
139 static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
140 unsigned dstRGB, unsigned dstA)
141 {
142 /* If the blend equation is ADD or REVERSE_SUBTRACT,
143 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
144 * the colorbuffer will not be changed.
145 * Notice that the dst factors are the src factors inverted. */
146 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
147 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
148 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
149 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
150 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
151 srcA == PIPE_BLENDFACTOR_ZERO) &&
152 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
153 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
154 dstRGB == PIPE_BLENDFACTOR_ONE) &&
155 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
156 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
157 dstA == PIPE_BLENDFACTOR_ONE);
158 }
159
160 static unsigned bgra_cmask(unsigned mask)
161 {
162 /* Gallium uses RGBA color ordering while R300 expects BGRA. */
163
164 return ((mask & PIPE_MASK_R) << 2) |
165 ((mask & PIPE_MASK_B) >> 2) |
166 (mask & (PIPE_MASK_G | PIPE_MASK_A));
167 }
168
169 /* Create a new blend state based on the CSO blend state.
170 *
171 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
172 static void* r300_create_blend_state(struct pipe_context* pipe,
173 const struct pipe_blend_state* state)
174 {
175 struct r300_screen* r300screen = r300_screen(pipe->screen);
176 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
177
178 if (state->rt[0].blend_enable)
179 {
180 unsigned eqRGB = state->rt[0].rgb_func;
181 unsigned srcRGB = state->rt[0].rgb_src_factor;
182 unsigned dstRGB = state->rt[0].rgb_dst_factor;
183
184 unsigned eqA = state->rt[0].alpha_func;
185 unsigned srcA = state->rt[0].alpha_src_factor;
186 unsigned dstA = state->rt[0].alpha_dst_factor;
187
188 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
189 * this is just the crappy D3D naming */
190 blend->blend_control = R300_ALPHA_BLEND_ENABLE |
191 r300_translate_blend_function(eqRGB) |
192 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
193 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
194
195 /* Optimization: some operations do not require the destination color.
196 *
197 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
198 * otherwise blending gives incorrect results. It seems to be
199 * a hardware bug. */
200 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
201 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
202 dstRGB != PIPE_BLENDFACTOR_ZERO ||
203 dstA != PIPE_BLENDFACTOR_ZERO ||
204 srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
205 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
206 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
207 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
208 srcA == PIPE_BLENDFACTOR_DST_COLOR ||
209 srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
210 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
211 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
212 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
213 /* Enable reading from the colorbuffer. */
214 blend->blend_control |= R300_READ_ENABLE;
215
216 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) {
217 /* Optimization: Depending on incoming pixels, we can
218 * conditionally disable the reading in hardware... */
219 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
220 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
221 /* Disable reading if SRC_ALPHA == 0. */
222 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
223 dstRGB == PIPE_BLENDFACTOR_ZERO) &&
224 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
225 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
226 dstA == PIPE_BLENDFACTOR_ZERO)) {
227 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ;
228 }
229
230 /* Disable reading if SRC_ALPHA == 1. */
231 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
232 dstRGB == PIPE_BLENDFACTOR_ZERO) &&
233 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
234 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
235 dstA == PIPE_BLENDFACTOR_ZERO)) {
236 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ;
237 }
238 }
239 }
240 }
241
242 /* Optimization: discard pixels which don't change the colorbuffer.
243 *
244 * The code below is non-trivial and some math is involved.
245 *
246 * Discarding pixels must be disabled when FP16 AA is enabled.
247 * This is a hardware bug. Also, this implementation wouldn't work
248 * with FP blending enabled and equation clamping disabled.
249 *
250 * Equations other than ADD are rarely used and therefore won't be
251 * optimized. */
252 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
253 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
254 /* ADD: X+Y
255 * REVERSE_SUBTRACT: Y-X
256 *
257 * The idea is:
258 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
259 * then CB will not be changed.
260 *
261 * Given the srcFactor and dstFactor variables, we can derive
262 * what src and dst should be equal to and discard appropriate
263 * pixels.
264 */
265 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
266 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
267 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
268 dstRGB, dstA)) {
269 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
270 } else if (blend_discard_if_src_color_0(srcRGB, srcA,
271 dstRGB, dstA)) {
272 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
273 } else if (blend_discard_if_src_color_1(srcRGB, srcA,
274 dstRGB, dstA)) {
275 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
276 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
277 dstRGB, dstA)) {
278 blend->blend_control |=
279 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
280 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
281 dstRGB, dstA)) {
282 blend->blend_control |=
283 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
284 }
285 }
286
287 /* separate alpha */
288 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
289 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE;
290 blend->alpha_blend_control =
291 r300_translate_blend_function(eqA) |
292 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
293 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
294 }
295 }
296
297 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
298 if (state->logicop_enable) {
299 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
300 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
301 }
302
303 /* Color channel masks for all MRTs. */
304 blend->color_channel_mask = bgra_cmask(state->rt[0].colormask);
305 if (r300screen->caps->is_r500 && state->independent_blend_enable) {
306 if (state->rt[1].blend_enable) {
307 blend->color_channel_mask |= bgra_cmask(state->rt[1].colormask) << 4;
308 }
309 if (state->rt[2].blend_enable) {
310 blend->color_channel_mask |= bgra_cmask(state->rt[2].colormask) << 8;
311 }
312 if (state->rt[3].blend_enable) {
313 blend->color_channel_mask |= bgra_cmask(state->rt[3].colormask) << 12;
314 }
315 }
316
317 if (state->dither) {
318 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
319 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
320 }
321
322 return (void*)blend;
323 }
324
325 /* Bind blend state. */
326 static void r300_bind_blend_state(struct pipe_context* pipe,
327 void* state)
328 {
329 struct r300_context* r300 = r300_context(pipe);
330
331 r300->blend_state.state = state;
332 r300->blend_state.dirty = TRUE;
333 }
334
335 /* Free blend state. */
336 static void r300_delete_blend_state(struct pipe_context* pipe,
337 void* state)
338 {
339 FREE(state);
340 }
341
342 /* Convert float to 10bit integer */
343 static unsigned float_to_fixed10(float f)
344 {
345 return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
346 }
347
348 /* Set blend color.
349 * Setup both R300 and R500 registers, figure out later which one to write. */
350 static void r300_set_blend_color(struct pipe_context* pipe,
351 const struct pipe_blend_color* color)
352 {
353 struct r300_context* r300 = r300_context(pipe);
354 struct r300_screen* r300screen = r300_screen(pipe->screen);
355 struct r300_blend_color_state* state =
356 (struct r300_blend_color_state*)r300->blend_color_state.state;
357 union util_color uc;
358
359 util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
360 state->blend_color = uc.ui;
361
362 /* XXX if FP16 blending is enabled, we should use the FP16 format */
363 state->blend_color_red_alpha =
364 float_to_fixed10(color->color[0]) |
365 (float_to_fixed10(color->color[3]) << 16);
366 state->blend_color_green_blue =
367 float_to_fixed10(color->color[2]) |
368 (float_to_fixed10(color->color[1]) << 16);
369
370 r300->blend_color_state.size = r300screen->caps->is_r500 ? 3 : 2;
371 r300->blend_color_state.dirty = TRUE;
372 }
373
374 static void r300_set_clip_state(struct pipe_context* pipe,
375 const struct pipe_clip_state* state)
376 {
377 struct r300_context* r300 = r300_context(pipe);
378
379 if (r300_screen(pipe->screen)->caps->has_tcl) {
380 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
381 r300->clip_state.size = 29;
382 } else {
383 draw_flush(r300->draw);
384 draw_set_clip_state(r300->draw, state);
385 r300->clip_state.size = 2;
386 }
387
388 r300->clip_state.dirty = TRUE;
389 }
390
391 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
392 *
393 * This contains the depth buffer, stencil buffer, alpha test, and such.
394 * On the Radeon, depth and stencil buffer setup are intertwined, which is
395 * the reason for some of the strange-looking assignments across registers. */
396 static void*
397 r300_create_dsa_state(struct pipe_context* pipe,
398 const struct pipe_depth_stencil_alpha_state* state)
399 {
400 struct r300_capabilities *caps =
401 r300_screen(r300_context(pipe)->context.screen)->caps;
402 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
403
404 /* Depth test setup. */
405 if (state->depth.enabled) {
406 dsa->z_buffer_control |= R300_Z_ENABLE;
407
408 if (state->depth.writemask) {
409 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
410 }
411
412 dsa->z_stencil_control |=
413 (r300_translate_depth_stencil_function(state->depth.func) <<
414 R300_Z_FUNC_SHIFT);
415 }
416
417 /* Stencil buffer setup. */
418 if (state->stencil[0].enabled) {
419 dsa->z_buffer_control |= R300_STENCIL_ENABLE;
420 dsa->z_stencil_control |=
421 (r300_translate_depth_stencil_function(state->stencil[0].func) <<
422 R300_S_FRONT_FUNC_SHIFT) |
423 (r300_translate_stencil_op(state->stencil[0].fail_op) <<
424 R300_S_FRONT_SFAIL_OP_SHIFT) |
425 (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
426 R300_S_FRONT_ZPASS_OP_SHIFT) |
427 (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
428 R300_S_FRONT_ZFAIL_OP_SHIFT);
429
430 dsa->stencil_ref_mask =
431 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
432 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
433
434 if (state->stencil[1].enabled) {
435 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
436 dsa->z_stencil_control |=
437 (r300_translate_depth_stencil_function(state->stencil[1].func) <<
438 R300_S_BACK_FUNC_SHIFT) |
439 (r300_translate_stencil_op(state->stencil[1].fail_op) <<
440 R300_S_BACK_SFAIL_OP_SHIFT) |
441 (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
442 R300_S_BACK_ZPASS_OP_SHIFT) |
443 (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
444 R300_S_BACK_ZFAIL_OP_SHIFT);
445
446 if (caps->is_r500)
447 {
448 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
449 dsa->stencil_ref_bf =
450 (state->stencil[1].valuemask <<
451 R300_STENCILMASK_SHIFT) |
452 (state->stencil[1].writemask <<
453 R300_STENCILWRITEMASK_SHIFT);
454 }
455 }
456 }
457
458 /* Alpha test setup. */
459 if (state->alpha.enabled) {
460 dsa->alpha_function =
461 r300_translate_alpha_function(state->alpha.func) |
462 R300_FG_ALPHA_FUNC_ENABLE;
463
464 /* We could use 10bit alpha ref but who needs that? */
465 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
466
467 if (caps->is_r500)
468 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
469 }
470
471 return (void*)dsa;
472 }
473
474 /* Bind DSA state. */
475 static void r300_bind_dsa_state(struct pipe_context* pipe,
476 void* state)
477 {
478 struct r300_context* r300 = r300_context(pipe);
479 struct r300_screen* r300screen = r300_screen(pipe->screen);
480
481 r300->dsa_state.state = state;
482 r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6;
483 r300->dsa_state.dirty = TRUE;
484 }
485
486 /* Free DSA state. */
487 static void r300_delete_dsa_state(struct pipe_context* pipe,
488 void* state)
489 {
490 FREE(state);
491 }
492
493 static void r300_set_stencil_ref(struct pipe_context* pipe,
494 const struct pipe_stencil_ref* sr)
495 {
496 struct r300_context* r300 = r300_context(pipe);
497 r300->stencil_ref = *sr;
498 r300->dsa_state.dirty = TRUE;
499 }
500
501 /* This switcheroo is needed just because of goddamned MACRO_SWITCH. */
502 static void r300_fb_update_tiling_flags(struct r300_context *r300,
503 const struct pipe_framebuffer_state *old_state,
504 const struct pipe_framebuffer_state *new_state)
505 {
506 struct r300_texture *tex;
507 unsigned i, j, level;
508
509 /* Reset tiling flags for old surfaces to default values. */
510 for (i = 0; i < old_state->nr_cbufs; i++) {
511 for (j = 0; j < new_state->nr_cbufs; j++) {
512 if (old_state->cbufs[i]->texture == new_state->cbufs[j]->texture) {
513 break;
514 }
515 }
516 /* If not binding the surface again... */
517 if (j != new_state->nr_cbufs) {
518 continue;
519 }
520
521 tex = (struct r300_texture*)old_state->cbufs[i]->texture;
522
523 if (tex) {
524 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
525 tex->pitch[0],
526 tex->microtile != 0,
527 tex->macrotile != 0);
528 }
529 }
530 if (old_state->zsbuf &&
531 (!new_state->zsbuf ||
532 old_state->zsbuf->texture != new_state->zsbuf->texture)) {
533 tex = (struct r300_texture*)old_state->zsbuf->texture;
534
535 if (tex) {
536 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
537 tex->pitch[0],
538 tex->microtile != 0,
539 tex->macrotile != 0);
540 }
541 }
542
543 /* Set tiling flags for new surfaces. */
544 for (i = 0; i < new_state->nr_cbufs; i++) {
545 tex = (struct r300_texture*)new_state->cbufs[i]->texture;
546 level = new_state->cbufs[i]->level;
547
548 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
549 tex->pitch[level],
550 tex->microtile != 0,
551 tex->mip_macrotile[level] != 0);
552 }
553 if (new_state->zsbuf) {
554 tex = (struct r300_texture*)new_state->zsbuf->texture;
555 level = new_state->zsbuf->level;
556
557 r300->winsys->buffer_set_tiling(r300->winsys, tex->buffer,
558 tex->pitch[level],
559 tex->microtile != 0,
560 tex->mip_macrotile[level] != 0);
561 }
562 }
563
564 static void
565 r300_set_framebuffer_state(struct pipe_context* pipe,
566 const struct pipe_framebuffer_state* state)
567 {
568 struct r300_context* r300 = r300_context(pipe);
569 struct r300_screen* r300screen = r300_screen(pipe->screen);
570 unsigned max_width, max_height;
571 uint32_t zbuffer_bpp = 0;
572
573
574 if (state->nr_cbufs > 4) {
575 debug_printf("r300: Implementation error: Too many MRTs in %s, "
576 "refusing to bind framebuffer state!\n", __FUNCTION__);
577 return;
578 }
579
580 if (r300screen->caps->is_r500) {
581 max_width = max_height = 4096;
582 } else if (r300screen->caps->is_r400) {
583 max_width = max_height = 4021;
584 } else {
585 max_width = max_height = 2560;
586 }
587
588 if (state->width > max_width || state->height > max_height) {
589 debug_printf("r300: Implementation error: Render targets are too "
590 "big in %s, refusing to bind framebuffer state!\n", __FUNCTION__);
591 return;
592 }
593
594
595 if (r300->draw) {
596 draw_flush(r300->draw);
597 }
598
599 memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
600
601 r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) +
602 (state->zsbuf ? 10 : 0) + 8;
603
604 r300_fb_update_tiling_flags(r300, r300->fb_state.state, state);
605
606 /* XXX wait what */
607 r300->blend_state.dirty = TRUE;
608 r300->dsa_state.dirty = TRUE;
609 r300->fb_state.dirty = TRUE;
610 r300->scissor_state.dirty = TRUE;
611
612 /* Polygon offset depends on the zbuffer bit depth. */
613 if (state->zsbuf && r300->polygon_offset_enabled) {
614 switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
615 case 2:
616 zbuffer_bpp = 16;
617 break;
618 case 4:
619 zbuffer_bpp = 24;
620 break;
621 }
622
623 if (r300->zbuffer_bpp != zbuffer_bpp) {
624 r300->zbuffer_bpp = zbuffer_bpp;
625 r300->rs_state.dirty = TRUE;
626 }
627 }
628 }
629
630 /* Create fragment shader state. */
631 static void* r300_create_fs_state(struct pipe_context* pipe,
632 const struct pipe_shader_state* shader)
633 {
634 struct r300_fragment_shader* fs = NULL;
635
636 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
637
638 /* Copy state directly into shader. */
639 fs->state = *shader;
640 fs->state.tokens = tgsi_dup_tokens(shader->tokens);
641
642 tgsi_scan_shader(shader->tokens, &fs->info);
643 r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
644
645 return (void*)fs;
646 }
647
648 /* Bind fragment shader state. */
649 static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
650 {
651 struct r300_context* r300 = r300_context(pipe);
652 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
653
654 if (fs == NULL) {
655 r300->fs = NULL;
656 return;
657 }
658
659 r300->fs = fs;
660 r300_pick_fragment_shader(r300);
661
662 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
663
664 if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) {
665 r300->vap_output_state.dirty = TRUE;
666 }
667
668 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
669 }
670
671 /* Delete fragment shader state. */
672 static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
673 {
674 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
675 struct r300_fragment_shader_code *tmp, *ptr = fs->first;
676
677 while (ptr) {
678 tmp = ptr;
679 ptr = ptr->next;
680 rc_constants_destroy(&tmp->code.constants);
681 FREE(tmp);
682 }
683 FREE((void*)fs->state.tokens);
684 FREE(shader);
685 }
686
687 static void r300_set_polygon_stipple(struct pipe_context* pipe,
688 const struct pipe_poly_stipple* state)
689 {
690 /* XXX no idea how to set this up, but not terribly important */
691 }
692
693 /* Create a new rasterizer state based on the CSO rasterizer state.
694 *
695 * This is a very large chunk of state, and covers most of the graphics
696 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
697 *
698 * In a not entirely unironic sidenote, this state has nearly nothing to do
699 * with the actual block on the Radeon called the rasterizer (RS). */
700 static void* r300_create_rs_state(struct pipe_context* pipe,
701 const struct pipe_rasterizer_state* state)
702 {
703 struct r300_screen* r300screen = r300_screen(pipe->screen);
704 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
705
706 /* Copy rasterizer state for Draw. */
707 rs->rs = *state;
708
709 #ifdef PIPE_ARCH_LITTLE_ENDIAN
710 rs->vap_control_status = R300_VC_NO_SWAP;
711 #else
712 rs->vap_control_status = R300_VC_32BIT_SWAP;
713 #endif
714
715 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
716 * Else, enable HW TCL and force Draw's TCL off. */
717 if (state->bypass_vs_clip_and_viewport ||
718 !r300screen->caps->has_tcl) {
719 rs->vap_control_status |= R300_VAP_TCL_BYPASS;
720 }
721
722 rs->point_size = pack_float_16_6x(state->point_size) |
723 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
724
725 rs->line_control = pack_float_16_6x(state->line_width) |
726 R300_GA_LINE_CNTL_END_TYPE_COMP;
727
728 /* Enable polygon mode */
729 if (state->fill_cw != PIPE_POLYGON_MODE_FILL ||
730 state->fill_ccw != PIPE_POLYGON_MODE_FILL) {
731 rs->polygon_mode = R300_GA_POLY_MODE_DUAL;
732 }
733
734 /* Radeons don't think in "CW/CCW", they think in "front/back". */
735 if (state->front_winding == PIPE_WINDING_CW) {
736 rs->cull_mode = R300_FRONT_FACE_CW;
737
738 /* Polygon offset */
739 if (state->offset_cw) {
740 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
741 }
742 if (state->offset_ccw) {
743 rs->polygon_offset_enable |= R300_BACK_ENABLE;
744 }
745
746 /* Polygon mode */
747 if (rs->polygon_mode) {
748 rs->polygon_mode |=
749 r300_translate_polygon_mode_front(state->fill_cw);
750 rs->polygon_mode |=
751 r300_translate_polygon_mode_back(state->fill_ccw);
752 }
753 } else {
754 rs->cull_mode = R300_FRONT_FACE_CCW;
755
756 /* Polygon offset */
757 if (state->offset_ccw) {
758 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
759 }
760 if (state->offset_cw) {
761 rs->polygon_offset_enable |= R300_BACK_ENABLE;
762 }
763
764 /* Polygon mode */
765 if (rs->polygon_mode) {
766 rs->polygon_mode |=
767 r300_translate_polygon_mode_front(state->fill_ccw);
768 rs->polygon_mode |=
769 r300_translate_polygon_mode_back(state->fill_cw);
770 }
771 }
772 if (state->front_winding & state->cull_mode) {
773 rs->cull_mode |= R300_CULL_FRONT;
774 }
775 if (~(state->front_winding) & state->cull_mode) {
776 rs->cull_mode |= R300_CULL_BACK;
777 }
778
779 if (rs->polygon_offset_enable) {
780 rs->depth_offset = state->offset_units;
781 rs->depth_scale = state->offset_scale;
782 }
783
784 if (state->line_stipple_enable) {
785 rs->line_stipple_config =
786 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
787 (fui((float)state->line_stipple_factor) &
788 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
789 /* XXX this might need to be scaled up */
790 rs->line_stipple_value = state->line_stipple_pattern;
791 }
792
793 if (state->flatshade) {
794 rs->color_control = R300_SHADE_MODEL_FLAT;
795 } else {
796 rs->color_control = R300_SHADE_MODEL_SMOOTH;
797 }
798
799 return (void*)rs;
800 }
801
802 /* Bind rasterizer state. */
803 static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
804 {
805 struct r300_context* r300 = r300_context(pipe);
806 struct r300_rs_state* rs = (struct r300_rs_state*)state;
807
808 if (r300->draw) {
809 draw_flush(r300->draw);
810 draw_set_rasterizer_state(r300->draw, &rs->rs);
811 }
812
813 if (rs) {
814 r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport;
815 r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw;
816 r300->rs_state.dirty = TRUE;
817 } else {
818 r300->tcl_bypass = FALSE;
819 r300->polygon_offset_enabled = FALSE;
820 }
821
822 r300->rs_state.state = rs;
823 r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0);
824 /* XXX Why is this still needed, dammit!? */
825 r300->scissor_state.dirty = TRUE;
826 r300->viewport_state.dirty = TRUE;
827
828 /* XXX Clean these up when we move to atom emits */
829 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
830 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
831 }
832 }
833
834 /* Free rasterizer state. */
835 static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
836 {
837 FREE(state);
838 }
839
840 static void*
841 r300_create_sampler_state(struct pipe_context* pipe,
842 const struct pipe_sampler_state* state)
843 {
844 struct r300_context* r300 = r300_context(pipe);
845 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
846 int lod_bias;
847 union util_color uc;
848
849 sampler->state = *state;
850
851 sampler->filter0 |=
852 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
853 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
854 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
855
856 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
857 state->mag_img_filter,
858 state->min_mip_filter,
859 state->max_anisotropy > 0);
860
861 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
862 /* We must pass these to the merge function to clamp them properly. */
863 sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
864 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
865
866 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
867
868 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
869
870 sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
871
872 util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
873 sampler->border_color = uc.ui;
874
875 /* R500-specific fixups and optimizations */
876 if (r300_screen(r300->context.screen)->caps->is_r500) {
877 sampler->filter1 |= R500_BORDER_FIX;
878 }
879
880 return (void*)sampler;
881 }
882
883 static void r300_bind_sampler_states(struct pipe_context* pipe,
884 unsigned count,
885 void** states)
886 {
887 struct r300_context* r300 = r300_context(pipe);
888 struct r300_textures_state* state =
889 (struct r300_textures_state*)r300->textures_state.state;
890
891 if (count > 8) {
892 return;
893 }
894
895 memcpy(state->sampler_states, states, sizeof(void*) * count);
896 state->sampler_count = count;
897
898 r300->textures_state.dirty = TRUE;
899
900 /* Pick a fragment shader based on the texture compare state. */
901 if (r300->fs && count) {
902 if (r300_pick_fragment_shader(r300)) {
903 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
904 R300_NEW_FRAGMENT_SHADER_CONSTANTS;
905 }
906 }
907 }
908
909 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
910 unsigned count,
911 void** states)
912 {
913 }
914
915 static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
916 {
917 FREE(state);
918 }
919
920 static void r300_set_sampler_textures(struct pipe_context* pipe,
921 unsigned count,
922 struct pipe_texture** texture)
923 {
924 struct r300_context* r300 = r300_context(pipe);
925 struct r300_textures_state* state =
926 (struct r300_textures_state*)r300->textures_state.state;
927 unsigned i;
928 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
929 boolean dirty_tex = FALSE;
930
931 /* XXX magic num */
932 if (count > 8) {
933 return;
934 }
935
936 for (i = 0; i < count; i++) {
937 if (state->textures[i] != (struct r300_texture*)texture[i]) {
938 pipe_texture_reference((struct pipe_texture**)&state->textures[i],
939 texture[i]);
940 dirty_tex = TRUE;
941
942 /* R300-specific - set the texrect factor in the fragment shader */
943 if (!is_r500 && state->textures[i]->is_npot) {
944 /* XXX It would be nice to re-emit just 1 constant,
945 * XXX not all of them */
946 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
947 }
948 }
949 }
950
951 for (i = count; i < 8; i++) {
952 if (state->textures[i]) {
953 pipe_texture_reference((struct pipe_texture**)&state->textures[i],
954 NULL);
955 }
956 }
957
958 state->texture_count = count;
959
960 r300->textures_state.dirty = TRUE;
961
962 if (dirty_tex) {
963 r300->texture_cache_inval.dirty = TRUE;
964 }
965 }
966
967 static void r300_set_scissor_state(struct pipe_context* pipe,
968 const struct pipe_scissor_state* state)
969 {
970 struct r300_context* r300 = r300_context(pipe);
971
972 memcpy(r300->scissor_state.state, state,
973 sizeof(struct pipe_scissor_state));
974
975 r300->scissor_state.dirty = TRUE;
976 }
977
978 static void r300_set_viewport_state(struct pipe_context* pipe,
979 const struct pipe_viewport_state* state)
980 {
981 struct r300_context* r300 = r300_context(pipe);
982 struct r300_viewport_state* viewport =
983 (struct r300_viewport_state*)r300->viewport_state.state;
984
985 /* Do the transform in HW. */
986 viewport->vte_control = R300_VTX_W0_FMT;
987
988 if (state->scale[0] != 1.0f) {
989 viewport->xscale = state->scale[0];
990 viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
991 }
992 if (state->scale[1] != 1.0f) {
993 viewport->yscale = state->scale[1];
994 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
995 }
996 if (state->scale[2] != 1.0f) {
997 viewport->zscale = state->scale[2];
998 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
999 }
1000 if (state->translate[0] != 0.0f) {
1001 viewport->xoffset = state->translate[0];
1002 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
1003 }
1004 if (state->translate[1] != 0.0f) {
1005 viewport->yoffset = state->translate[1];
1006 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
1007 }
1008 if (state->translate[2] != 0.0f) {
1009 viewport->zoffset = state->translate[2];
1010 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
1011 }
1012
1013 r300->viewport_state.dirty = TRUE;
1014 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
1015 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1016 }
1017 }
1018
1019 static void r300_set_vertex_buffers(struct pipe_context* pipe,
1020 unsigned count,
1021 const struct pipe_vertex_buffer* buffers)
1022 {
1023 struct r300_context* r300 = r300_context(pipe);
1024 unsigned i, max_index = ~0;
1025
1026 memcpy(r300->vertex_buffer, buffers,
1027 sizeof(struct pipe_vertex_buffer) * count);
1028
1029 for (i = 0; i < count; i++) {
1030 max_index = MIN2(buffers[i].max_index, max_index);
1031 }
1032
1033 r300->vertex_buffer_count = count;
1034 r300->vertex_buffer_max_index = max_index;
1035
1036 if (r300->draw) {
1037 draw_flush(r300->draw);
1038 draw_set_vertex_buffers(r300->draw, count, buffers);
1039 } else {
1040 r300->vertex_stream_state.dirty = TRUE;
1041 }
1042 }
1043
1044 static boolean r300_validate_aos(struct r300_context *r300)
1045 {
1046 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
1047 struct pipe_vertex_element *velem = r300->vertex_element;
1048 int i;
1049
1050 /* Check if formats and strides are aligned to the size of DWORD. */
1051 for (i = 0; i < r300->vertex_element_count; i++) {
1052 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
1053 util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
1054 return FALSE;
1055 }
1056 }
1057 return TRUE;
1058 }
1059
1060 static void r300_set_vertex_elements(struct pipe_context* pipe,
1061 unsigned count,
1062 const struct pipe_vertex_element* elements)
1063 {
1064 struct r300_context* r300 = r300_context(pipe);
1065
1066 memcpy(r300->vertex_element,
1067 elements,
1068 sizeof(struct pipe_vertex_element) * count);
1069 r300->vertex_element_count = count;
1070
1071 if (r300->draw) {
1072 draw_flush(r300->draw);
1073 draw_set_vertex_elements(r300->draw, count, elements);
1074 }
1075
1076 if (!r300_validate_aos(r300)) {
1077 /* XXX We should fallback using draw. */
1078 assert(0);
1079 abort();
1080 }
1081 }
1082
1083 static void* r300_create_vs_state(struct pipe_context* pipe,
1084 const struct pipe_shader_state* shader)
1085 {
1086 struct r300_context* r300 = r300_context(pipe);
1087
1088 if (r300_screen(pipe->screen)->caps->has_tcl) {
1089 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
1090 /* Copy state directly into shader. */
1091 vs->state = *shader;
1092 vs->state.tokens = tgsi_dup_tokens(shader->tokens);
1093
1094 tgsi_scan_shader(shader->tokens, &vs->info);
1095
1096 return (void*)vs;
1097 } else {
1098 return draw_create_vertex_shader(r300->draw, shader);
1099 }
1100 }
1101
1102 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
1103 {
1104 struct r300_context* r300 = r300_context(pipe);
1105
1106 if (r300_screen(pipe->screen)->caps->has_tcl) {
1107 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1108
1109 if (vs == NULL) {
1110 r300->vs_state.state = NULL;
1111 return;
1112 } else if (!vs->translated) {
1113 r300_translate_vertex_shader(r300, vs);
1114 }
1115
1116 r300->vs_state.state = vs;
1117 r300->vs_state.size = vs->code.length + 9;
1118 r300->vs_state.dirty = TRUE;
1119
1120 r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
1121 r300->vap_output_state.dirty = TRUE;
1122 r300->vertex_stream_state.dirty = TRUE; /* XXX needed for TCL bypass */
1123 r300->pvs_flush.dirty = TRUE;
1124
1125 if (r300->fs) {
1126 r300_vertex_shader_setup_wpos(r300);
1127 }
1128
1129 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
1130 } else {
1131 draw_flush(r300->draw);
1132 draw_bind_vertex_shader(r300->draw,
1133 (struct draw_vertex_shader*)shader);
1134 }
1135 }
1136
1137 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1138 {
1139 struct r300_context* r300 = r300_context(pipe);
1140
1141 if (r300_screen(pipe->screen)->caps->has_tcl) {
1142 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1143
1144 rc_constants_destroy(&vs->code.constants);
1145 FREE((void*)vs->state.tokens);
1146 FREE(shader);
1147 } else {
1148 draw_delete_vertex_shader(r300->draw,
1149 (struct draw_vertex_shader*)shader);
1150 }
1151 }
1152
1153 static void r300_set_constant_buffer(struct pipe_context *pipe,
1154 uint shader, uint index,
1155 struct pipe_buffer *buf)
1156 {
1157 struct r300_context* r300 = r300_context(pipe);
1158 struct r300_screen *r300screen = r300_screen(pipe->screen);
1159 void *mapped;
1160 int max_size = 0;
1161
1162 if (buf == NULL || buf->size == 0 ||
1163 (mapped = pipe_buffer_map(pipe->screen, buf, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
1164 {
1165 r300->shader_constants[shader].count = 0;
1166 return;
1167 }
1168
1169 assert((buf->size % 4 * sizeof(float)) == 0);
1170
1171 /* Check the size of the constant buffer. */
1172 switch (shader) {
1173 case PIPE_SHADER_VERTEX:
1174 max_size = 256;
1175 break;
1176 case PIPE_SHADER_FRAGMENT:
1177 if (r300screen->caps->is_r500) {
1178 max_size = 256;
1179 /* XXX Implement emission of r400's extended constant buffer. */
1180 /*} else if (r300screen->caps->is_r400) {
1181 max_size = 64;*/
1182 } else {
1183 max_size = 32;
1184 }
1185 break;
1186 default:
1187 assert(0);
1188 }
1189
1190 /* XXX Subtract immediates and RC_STATE_* variables. */
1191 if (buf->size > (sizeof(float) * 4 * max_size)) {
1192 debug_printf("r300: Max size of the constant buffer is "
1193 "%i*4 floats.\n", max_size);
1194 abort();
1195 }
1196
1197 memcpy(r300->shader_constants[shader].constants, mapped, buf->size);
1198 r300->shader_constants[shader].count = buf->size / (4 * sizeof(float));
1199 pipe_buffer_unmap(pipe->screen, buf);
1200
1201 if (shader == PIPE_SHADER_VERTEX) {
1202 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
1203 r300->pvs_flush.dirty = TRUE;
1204 }
1205 else if (shader == PIPE_SHADER_FRAGMENT)
1206 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1207 }
1208
1209 void r300_init_state_functions(struct r300_context* r300)
1210 {
1211 r300->context.create_blend_state = r300_create_blend_state;
1212 r300->context.bind_blend_state = r300_bind_blend_state;
1213 r300->context.delete_blend_state = r300_delete_blend_state;
1214
1215 r300->context.set_blend_color = r300_set_blend_color;
1216
1217 r300->context.set_clip_state = r300_set_clip_state;
1218
1219 r300->context.set_constant_buffer = r300_set_constant_buffer;
1220
1221 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1222 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1223 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1224
1225 r300->context.set_stencil_ref = r300_set_stencil_ref;
1226
1227 r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1228
1229 r300->context.create_fs_state = r300_create_fs_state;
1230 r300->context.bind_fs_state = r300_bind_fs_state;
1231 r300->context.delete_fs_state = r300_delete_fs_state;
1232
1233 r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1234
1235 r300->context.create_rasterizer_state = r300_create_rs_state;
1236 r300->context.bind_rasterizer_state = r300_bind_rs_state;
1237 r300->context.delete_rasterizer_state = r300_delete_rs_state;
1238
1239 r300->context.create_sampler_state = r300_create_sampler_state;
1240 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1241 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1242 r300->context.delete_sampler_state = r300_delete_sampler_state;
1243
1244 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures;
1245
1246 r300->context.set_scissor_state = r300_set_scissor_state;
1247
1248 r300->context.set_viewport_state = r300_set_viewport_state;
1249
1250 r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1251 r300->context.set_vertex_elements = r300_set_vertex_elements;
1252
1253 r300->context.create_vs_state = r300_create_vs_state;
1254 r300->context.bind_vs_state = r300_bind_vs_state;
1255 r300->context.delete_vs_state = r300_delete_vs_state;
1256 }