r300g: if no DS buffer is set, disable reading from and writing to it
[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 #include "pipe/internal/p_winsys_screen.h"
34
35 #include "r300_context.h"
36 #include "r300_reg.h"
37 #include "r300_screen.h"
38 #include "r300_state_inlines.h"
39 #include "r300_fs.h"
40 #include "r300_vs.h"
41
42 /* r300_state: Functions used to intialize state context by translating
43 * Gallium state objects into semi-native r300 state objects. */
44
45 static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
46 unsigned dstRGB, unsigned dstA)
47 {
48 /* If the blend equation is ADD or REVERSE_SUBTRACT,
49 * SRC_ALPHA == 0, and the following state is set, the colorbuffer
50 * will not be changed.
51 * Notice that the dst factors are the src factors inverted. */
52 return (srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
53 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
54 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
55 (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
56 srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
57 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
58 srcA == PIPE_BLENDFACTOR_ZERO) &&
59 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
60 dstRGB == PIPE_BLENDFACTOR_ONE) &&
61 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
62 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
63 dstA == PIPE_BLENDFACTOR_ONE);
64 }
65
66 static boolean blend_discard_if_src_alpha_1(unsigned srcRGB, unsigned srcA,
67 unsigned dstRGB, unsigned dstA)
68 {
69 /* If the blend equation is ADD or REVERSE_SUBTRACT,
70 * SRC_ALPHA == 1, and the following state is set, the colorbuffer
71 * will not be changed.
72 * Notice that the dst factors are the src factors inverted. */
73 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
74 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
75 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
76 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
77 srcA == PIPE_BLENDFACTOR_ZERO) &&
78 (dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
79 dstRGB == PIPE_BLENDFACTOR_ONE) &&
80 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
81 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
82 dstA == PIPE_BLENDFACTOR_ONE);
83 }
84
85 static boolean blend_discard_if_src_color_0(unsigned srcRGB, unsigned srcA,
86 unsigned dstRGB, unsigned dstA)
87 {
88 /* If the blend equation is ADD or REVERSE_SUBTRACT,
89 * SRC_COLOR == (0,0,0), and the following state is set, the colorbuffer
90 * will not be changed.
91 * Notice that the dst factors are the src factors inverted. */
92 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
93 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
94 (srcA == PIPE_BLENDFACTOR_ZERO) &&
95 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
96 dstRGB == PIPE_BLENDFACTOR_ONE) &&
97 (dstA == PIPE_BLENDFACTOR_ONE);
98 }
99
100 static boolean blend_discard_if_src_color_1(unsigned srcRGB, unsigned srcA,
101 unsigned dstRGB, unsigned dstA)
102 {
103 /* If the blend equation is ADD or REVERSE_SUBTRACT,
104 * SRC_COLOR == (1,1,1), and the following state is set, the colorbuffer
105 * will not be changed.
106 * Notice that the dst factors are the src factors inverted. */
107 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
108 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
109 (srcA == PIPE_BLENDFACTOR_ZERO) &&
110 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
111 dstRGB == PIPE_BLENDFACTOR_ONE) &&
112 (dstA == PIPE_BLENDFACTOR_ONE);
113 }
114
115 static boolean blend_discard_if_src_alpha_color_0(unsigned srcRGB, unsigned srcA,
116 unsigned dstRGB, unsigned dstA)
117 {
118 /* If the blend equation is ADD or REVERSE_SUBTRACT,
119 * SRC_ALPHA_COLOR == (0,0,0,0), and the following state is set,
120 * the colorbuffer will not be changed.
121 * Notice that the dst factors are the src factors inverted. */
122 return (srcRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
123 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
124 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
125 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
126 (srcA == PIPE_BLENDFACTOR_SRC_COLOR ||
127 srcA == PIPE_BLENDFACTOR_SRC_ALPHA ||
128 srcA == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE ||
129 srcA == PIPE_BLENDFACTOR_ZERO) &&
130 (dstRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
131 dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
132 dstRGB == PIPE_BLENDFACTOR_ONE) &&
133 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
134 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
135 dstA == PIPE_BLENDFACTOR_ONE);
136 }
137
138 static boolean blend_discard_if_src_alpha_color_1(unsigned srcRGB, unsigned srcA,
139 unsigned dstRGB, unsigned dstA)
140 {
141 /* If the blend equation is ADD or REVERSE_SUBTRACT,
142 * SRC_ALPHA_COLOR == (1,1,1,1), and the following state is set,
143 * the colorbuffer will not be changed.
144 * Notice that the dst factors are the src factors inverted. */
145 return (srcRGB == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
146 srcRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
147 srcRGB == PIPE_BLENDFACTOR_ZERO) &&
148 (srcA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
149 srcA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
150 srcA == PIPE_BLENDFACTOR_ZERO) &&
151 (dstRGB == PIPE_BLENDFACTOR_SRC_COLOR ||
152 dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
153 dstRGB == PIPE_BLENDFACTOR_ONE) &&
154 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
155 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
156 dstA == PIPE_BLENDFACTOR_ONE);
157 }
158
159 /* Create a new blend state based on the CSO blend state.
160 *
161 * This encompasses alpha blending, logic/raster ops, and blend dithering. */
162 static void* r300_create_blend_state(struct pipe_context* pipe,
163 const struct pipe_blend_state* state)
164 {
165 struct r300_blend_state* blend = CALLOC_STRUCT(r300_blend_state);
166
167 if (state->blend_enable)
168 {
169 unsigned eqRGB = state->rgb_func;
170 unsigned srcRGB = state->rgb_src_factor;
171 unsigned dstRGB = state->rgb_dst_factor;
172
173 unsigned eqA = state->alpha_func;
174 unsigned srcA = state->alpha_src_factor;
175 unsigned dstA = state->alpha_dst_factor;
176
177 /* despite the name, ALPHA_BLEND_ENABLE has nothing to do with alpha,
178 * this is just the crappy D3D naming */
179 blend->blend_control = R300_ALPHA_BLEND_ENABLE |
180 r300_translate_blend_function(eqRGB) |
181 ( r300_translate_blend_factor(srcRGB) << R300_SRC_BLEND_SHIFT) |
182 ( r300_translate_blend_factor(dstRGB) << R300_DST_BLEND_SHIFT);
183
184 /* Optimization: some operations do not require the destination color.
185 *
186 * When SRC_ALPHA_SATURATE is used, colorbuffer reads must be enabled,
187 * otherwise blending gives incorrect results. It seems to be
188 * a hardware bug. */
189 if (eqRGB == PIPE_BLEND_MIN || eqA == PIPE_BLEND_MIN ||
190 eqRGB == PIPE_BLEND_MAX || eqA == PIPE_BLEND_MAX ||
191 dstRGB != PIPE_BLENDFACTOR_ZERO ||
192 dstA != PIPE_BLENDFACTOR_ZERO ||
193 srcRGB == PIPE_BLENDFACTOR_DST_COLOR ||
194 srcRGB == PIPE_BLENDFACTOR_DST_ALPHA ||
195 srcRGB == PIPE_BLENDFACTOR_INV_DST_COLOR ||
196 srcRGB == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
197 srcA == PIPE_BLENDFACTOR_DST_COLOR ||
198 srcA == PIPE_BLENDFACTOR_DST_ALPHA ||
199 srcA == PIPE_BLENDFACTOR_INV_DST_COLOR ||
200 srcA == PIPE_BLENDFACTOR_INV_DST_ALPHA ||
201 srcRGB == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE) {
202 /* Enable reading from the colorbuffer. */
203 blend->blend_control |= R300_READ_ENABLE;
204
205 if (r300_screen(r300_context(pipe)->context.screen)->caps->is_r500) {
206 /* Optimization: Depending on incoming pixels, we can
207 * conditionally disable the reading in hardware... */
208 if (eqRGB != PIPE_BLEND_MIN && eqA != PIPE_BLEND_MIN &&
209 eqRGB != PIPE_BLEND_MAX && eqA != PIPE_BLEND_MAX) {
210 /* Disable reading if SRC_ALPHA == 0. */
211 if ((dstRGB == PIPE_BLENDFACTOR_SRC_ALPHA ||
212 dstRGB == PIPE_BLENDFACTOR_ZERO) &&
213 (dstA == PIPE_BLENDFACTOR_SRC_COLOR ||
214 dstA == PIPE_BLENDFACTOR_SRC_ALPHA ||
215 dstA == PIPE_BLENDFACTOR_ZERO)) {
216 blend->blend_control |= R500_SRC_ALPHA_0_NO_READ;
217 }
218
219 /* Disable reading if SRC_ALPHA == 1. */
220 if ((dstRGB == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
221 dstRGB == PIPE_BLENDFACTOR_ZERO) &&
222 (dstA == PIPE_BLENDFACTOR_INV_SRC_COLOR ||
223 dstA == PIPE_BLENDFACTOR_INV_SRC_ALPHA ||
224 dstA == PIPE_BLENDFACTOR_ZERO)) {
225 blend->blend_control |= R500_SRC_ALPHA_1_NO_READ;
226 }
227 }
228 }
229 }
230
231 /* Optimization: discard pixels which don't change the colorbuffer.
232 *
233 * The code below is non-trivial and some math is involved.
234 *
235 * Discarding pixels must be disabled when FP16 AA is enabled.
236 * This is a hardware bug. Also, this implementation wouldn't work
237 * with FP blending enabled and equation clamping disabled.
238 *
239 * Equations other than ADD are rarely used and therefore won't be
240 * optimized. */
241 if ((eqRGB == PIPE_BLEND_ADD || eqRGB == PIPE_BLEND_REVERSE_SUBTRACT) &&
242 (eqA == PIPE_BLEND_ADD || eqA == PIPE_BLEND_REVERSE_SUBTRACT)) {
243 /* ADD: X+Y
244 * REVERSE_SUBTRACT: Y-X
245 *
246 * The idea is:
247 * If X = src*srcFactor = 0 and Y = dst*dstFactor = 1,
248 * then CB will not be changed.
249 *
250 * Given the srcFactor and dstFactor variables, we can derive
251 * what src and dst should be equal to and discard appropriate
252 * pixels.
253 */
254 if (blend_discard_if_src_alpha_0(srcRGB, srcA, dstRGB, dstA)) {
255 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_0;
256 } else if (blend_discard_if_src_alpha_1(srcRGB, srcA,
257 dstRGB, dstA)) {
258 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_ALPHA_1;
259 } else if (blend_discard_if_src_color_0(srcRGB, srcA,
260 dstRGB, dstA)) {
261 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_0;
262 } else if (blend_discard_if_src_color_1(srcRGB, srcA,
263 dstRGB, dstA)) {
264 blend->blend_control |= R300_DISCARD_SRC_PIXELS_SRC_COLOR_1;
265 } else if (blend_discard_if_src_alpha_color_0(srcRGB, srcA,
266 dstRGB, dstA)) {
267 blend->blend_control |=
268 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_0;
269 } else if (blend_discard_if_src_alpha_color_1(srcRGB, srcA,
270 dstRGB, dstA)) {
271 blend->blend_control |=
272 R300_DISCARD_SRC_PIXELS_SRC_ALPHA_COLOR_1;
273 }
274 }
275
276 /* separate alpha */
277 if (srcA != srcRGB || dstA != dstRGB || eqA != eqRGB) {
278 blend->blend_control |= R300_SEPARATE_ALPHA_ENABLE;
279 blend->alpha_blend_control =
280 r300_translate_blend_function(eqA) |
281 (r300_translate_blend_factor(srcA) << R300_SRC_BLEND_SHIFT) |
282 (r300_translate_blend_factor(dstA) << R300_DST_BLEND_SHIFT);
283 }
284 }
285
286 /* PIPE_LOGICOP_* don't need to be translated, fortunately. */
287 if (state->logicop_enable) {
288 blend->rop = R300_RB3D_ROPCNTL_ROP_ENABLE |
289 (state->logicop_func) << R300_RB3D_ROPCNTL_ROP_SHIFT;
290 }
291
292 /* Color Channel Mask */
293 if (state->colormask & PIPE_MASK_R) {
294 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_RED_MASK0;
295 }
296 if (state->colormask & PIPE_MASK_G) {
297 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_GREEN_MASK0;
298 }
299 if (state->colormask & PIPE_MASK_B) {
300 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_BLUE_MASK0;
301 }
302 if (state->colormask & PIPE_MASK_A) {
303 blend->color_channel_mask |= RB3D_COLOR_CHANNEL_MASK_ALPHA_MASK0;
304 }
305
306 if (state->dither) {
307 blend->dither = R300_RB3D_DITHER_CTL_DITHER_MODE_LUT |
308 R300_RB3D_DITHER_CTL_ALPHA_DITHER_MODE_LUT;
309 }
310
311 return (void*)blend;
312 }
313
314 /* Bind blend state. */
315 static void r300_bind_blend_state(struct pipe_context* pipe,
316 void* state)
317 {
318 struct r300_context* r300 = r300_context(pipe);
319
320 r300->blend_state = (struct r300_blend_state*)state;
321 r300->dirty_state |= R300_NEW_BLEND;
322 }
323
324 /* Free blend state. */
325 static void r300_delete_blend_state(struct pipe_context* pipe,
326 void* state)
327 {
328 FREE(state);
329 }
330
331 /* Convert float to 10bit integer */
332 static unsigned float_to_fixed10(float f)
333 {
334 return CLAMP((unsigned)(f * 1023.9f), 0, 1023);
335 }
336
337 /* Set blend color.
338 * Setup both R300 and R500 registers, figure out later which one to write. */
339 static void r300_set_blend_color(struct pipe_context* pipe,
340 const struct pipe_blend_color* color)
341 {
342 struct r300_context* r300 = r300_context(pipe);
343 union util_color uc;
344
345 util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
346 r300->blend_color_state->blend_color = uc.ui;
347
348 /* XXX if FP16 blending is enabled, we should use the FP16 format */
349 r300->blend_color_state->blend_color_red_alpha =
350 float_to_fixed10(color->color[0]) |
351 (float_to_fixed10(color->color[3]) << 16);
352 r300->blend_color_state->blend_color_green_blue =
353 float_to_fixed10(color->color[2]) |
354 (float_to_fixed10(color->color[1]) << 16);
355
356 r300->dirty_state |= R300_NEW_BLEND_COLOR;
357 }
358
359 static void r300_set_clip_state(struct pipe_context* pipe,
360 const struct pipe_clip_state* state)
361 {
362 struct r300_context* r300 = r300_context(pipe);
363
364 if (r300_screen(pipe->screen)->caps->has_tcl) {
365 r300->clip_state = *state;
366 r300->dirty_state |= R300_NEW_CLIP;
367 } else {
368 draw_flush(r300->draw);
369 draw_set_clip_state(r300->draw, state);
370 }
371 }
372
373 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
374 *
375 * This contains the depth buffer, stencil buffer, alpha test, and such.
376 * On the Radeon, depth and stencil buffer setup are intertwined, which is
377 * the reason for some of the strange-looking assignments across registers. */
378 static void*
379 r300_create_dsa_state(struct pipe_context* pipe,
380 const struct pipe_depth_stencil_alpha_state* state)
381 {
382 struct r300_capabilities *caps =
383 r300_screen(r300_context(pipe)->context.screen)->caps;
384 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
385
386 /* Depth test setup. */
387 if (state->depth.enabled) {
388 dsa->z_buffer_control |= R300_Z_ENABLE;
389
390 if (state->depth.writemask) {
391 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
392 }
393
394 dsa->z_stencil_control |=
395 (r300_translate_depth_stencil_function(state->depth.func) <<
396 R300_Z_FUNC_SHIFT);
397 }
398
399 /* Stencil buffer setup. */
400 if (state->stencil[0].enabled) {
401 dsa->z_buffer_control |= R300_STENCIL_ENABLE;
402 dsa->z_stencil_control |=
403 (r300_translate_depth_stencil_function(state->stencil[0].func) <<
404 R300_S_FRONT_FUNC_SHIFT) |
405 (r300_translate_stencil_op(state->stencil[0].fail_op) <<
406 R300_S_FRONT_SFAIL_OP_SHIFT) |
407 (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
408 R300_S_FRONT_ZPASS_OP_SHIFT) |
409 (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
410 R300_S_FRONT_ZFAIL_OP_SHIFT);
411
412 dsa->stencil_ref_mask = (state->stencil[0].ref_value) |
413 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
414 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
415
416 if (state->stencil[1].enabled) {
417 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
418 dsa->z_stencil_control |=
419 (r300_translate_depth_stencil_function(state->stencil[1].func) <<
420 R300_S_BACK_FUNC_SHIFT) |
421 (r300_translate_stencil_op(state->stencil[1].fail_op) <<
422 R300_S_BACK_SFAIL_OP_SHIFT) |
423 (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
424 R300_S_BACK_ZPASS_OP_SHIFT) |
425 (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
426 R300_S_BACK_ZFAIL_OP_SHIFT);
427
428 /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
429 if (caps->is_r500)
430 {
431 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
432 dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
433 (state->stencil[1].valuemask <<
434 R300_STENCILMASK_SHIFT) |
435 (state->stencil[1].writemask <<
436 R300_STENCILWRITEMASK_SHIFT);
437 }
438 }
439 }
440
441 /* Alpha test setup. */
442 if (state->alpha.enabled) {
443 dsa->alpha_function =
444 r300_translate_alpha_function(state->alpha.func) |
445 R300_FG_ALPHA_FUNC_ENABLE;
446
447 /* XXX figure out why emitting 10bit alpha ref causes CS to dump */
448 /* always use 8bit alpha ref */
449 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
450
451 if (caps->is_r500)
452 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
453 }
454
455 return (void*)dsa;
456 }
457
458 /* Bind DSA state. */
459 static void r300_bind_dsa_state(struct pipe_context* pipe,
460 void* state)
461 {
462 struct r300_context* r300 = r300_context(pipe);
463
464 r300->dsa_state = (struct r300_dsa_state*)state;
465 r300->dirty_state |= R300_NEW_DSA;
466 }
467
468 /* Free DSA state. */
469 static void r300_delete_dsa_state(struct pipe_context* pipe,
470 void* state)
471 {
472 FREE(state);
473 }
474
475 static void r300_set_scissor_regs(const struct pipe_scissor_state* state,
476 struct r300_scissor_regs *scissor,
477 boolean is_r500)
478 {
479 if (is_r500) {
480 scissor->top_left =
481 (state->minx << R300_SCISSORS_X_SHIFT) |
482 (state->miny << R300_SCISSORS_Y_SHIFT);
483 scissor->bottom_right =
484 ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) |
485 ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT);
486 } else {
487 /* Offset of 1440 in non-R500 chipsets. */
488 scissor->top_left =
489 ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) |
490 ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT);
491 scissor->bottom_right =
492 (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
493 (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
494 }
495
496 scissor->empty_area = state->minx >= state->maxx ||
497 state->miny >= state->maxy;
498 }
499
500 static void
501 r300_set_framebuffer_state(struct pipe_context* pipe,
502 const struct pipe_framebuffer_state* state)
503 {
504 struct r300_context* r300 = r300_context(pipe);
505 struct pipe_scissor_state scissor;
506
507 if (r300->draw) {
508 draw_flush(r300->draw);
509 }
510
511 r300->framebuffer_state = *state;
512
513 scissor.minx = scissor.miny = 0;
514 scissor.maxx = state->width;
515 scissor.maxy = state->height;
516 r300_set_scissor_regs(&scissor, &r300->scissor_state->framebuffer,
517 r300_screen(r300->context.screen)->caps->is_r500);
518
519 /* Don't rely on the order of states being set for the first time. */
520 if (!r300->rs_state || !r300->rs_state->rs.scissor) {
521 r300->dirty_state |= R300_NEW_SCISSOR;
522 }
523 r300->dirty_state |= R300_NEW_FRAMEBUFFERS;
524 r300->dirty_state |= R300_NEW_BLEND;
525 r300->dirty_state |= R300_NEW_DSA;
526 }
527
528 /* Create fragment shader state. */
529 static void* r300_create_fs_state(struct pipe_context* pipe,
530 const struct pipe_shader_state* shader)
531 {
532 struct r300_fragment_shader* fs = NULL;
533
534 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
535
536 /* Copy state directly into shader. */
537 fs->state = *shader;
538 fs->state.tokens = tgsi_dup_tokens(shader->tokens);
539
540 tgsi_scan_shader(shader->tokens, &fs->info);
541 r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
542
543 return (void*)fs;
544 }
545
546 /* Bind fragment shader state. */
547 static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
548 {
549 struct r300_context* r300 = r300_context(pipe);
550 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
551
552 if (fs == NULL) {
553 r300->fs = NULL;
554 return;
555 }
556
557 r300->fs = fs;
558 r300_pick_fragment_shader(r300);
559
560 if (r300->vs && r300_vertex_shader_setup_wpos(r300)) {
561 r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
562 }
563
564 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
565 }
566
567 /* Delete fragment shader state. */
568 static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
569 {
570 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
571 struct r300_fragment_shader_code *tmp, *ptr = fs->first;
572
573 while (ptr) {
574 tmp = ptr;
575 ptr = ptr->next;
576 rc_constants_destroy(&tmp->code.constants);
577 FREE(tmp);
578 }
579 FREE((void*)fs->state.tokens);
580 FREE(shader);
581 }
582
583 static void r300_set_polygon_stipple(struct pipe_context* pipe,
584 const struct pipe_poly_stipple* state)
585 {
586 /* XXX no idea how to set this up, but not terribly important */
587 }
588
589 /* Create a new rasterizer state based on the CSO rasterizer state.
590 *
591 * This is a very large chunk of state, and covers most of the graphics
592 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
593 *
594 * In a not entirely unironic sidenote, this state has nearly nothing to do
595 * with the actual block on the Radeon called the rasterizer (RS). */
596 static void* r300_create_rs_state(struct pipe_context* pipe,
597 const struct pipe_rasterizer_state* state)
598 {
599 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
600
601 /* Copy rasterizer state for Draw. */
602 rs->rs = *state;
603
604 rs->enable_vte = !state->bypass_vs_clip_and_viewport;
605
606 #ifdef PIPE_ARCH_LITTLE_ENDIAN
607 rs->vap_control_status = R300_VC_NO_SWAP;
608 #else
609 rs->vap_control_status = R300_VC_32BIT_SWAP;
610 #endif
611
612 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
613 * Else, enable HW TCL and force Draw's TCL off. */
614 if (state->bypass_vs_clip_and_viewport ||
615 !r300_screen(pipe->screen)->caps->has_tcl) {
616 rs->vap_control_status |= R300_VAP_TCL_BYPASS;
617 }
618
619 rs->point_size = pack_float_16_6x(state->point_size) |
620 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
621
622 rs->point_minmax =
623 ((int)(state->point_size_min * 6.0) <<
624 R300_GA_POINT_MINMAX_MIN_SHIFT) |
625 ((int)(state->point_size_max * 6.0) <<
626 R300_GA_POINT_MINMAX_MAX_SHIFT);
627
628 rs->line_control = pack_float_16_6x(state->line_width) |
629 R300_GA_LINE_CNTL_END_TYPE_COMP;
630
631 /* XXX I think there is something wrong with the polygon mode,
632 * XXX re-test when r300g is in a better shape */
633
634 /* Enable polygon mode */
635 if (state->fill_cw != PIPE_POLYGON_MODE_FILL ||
636 state->fill_ccw != PIPE_POLYGON_MODE_FILL) {
637 rs->polygon_mode = R300_GA_POLY_MODE_DUAL;
638 }
639
640 /* Radeons don't think in "CW/CCW", they think in "front/back". */
641 if (state->front_winding == PIPE_WINDING_CW) {
642 rs->cull_mode = R300_FRONT_FACE_CW;
643
644 /* Polygon offset */
645 if (state->offset_cw) {
646 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
647 }
648 if (state->offset_ccw) {
649 rs->polygon_offset_enable |= R300_BACK_ENABLE;
650 }
651
652 /* Polygon mode */
653 if (rs->polygon_mode) {
654 rs->polygon_mode |=
655 r300_translate_polygon_mode_front(state->fill_cw);
656 rs->polygon_mode |=
657 r300_translate_polygon_mode_back(state->fill_ccw);
658 }
659 } else {
660 rs->cull_mode = R300_FRONT_FACE_CCW;
661
662 /* Polygon offset */
663 if (state->offset_ccw) {
664 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
665 }
666 if (state->offset_cw) {
667 rs->polygon_offset_enable |= R300_BACK_ENABLE;
668 }
669
670 /* Polygon mode */
671 if (rs->polygon_mode) {
672 rs->polygon_mode |=
673 r300_translate_polygon_mode_front(state->fill_ccw);
674 rs->polygon_mode |=
675 r300_translate_polygon_mode_back(state->fill_cw);
676 }
677 }
678 if (state->front_winding & state->cull_mode) {
679 rs->cull_mode |= R300_CULL_FRONT;
680 }
681 if (~(state->front_winding) & state->cull_mode) {
682 rs->cull_mode |= R300_CULL_BACK;
683 }
684
685 if (rs->polygon_offset_enable) {
686 rs->depth_offset_front = rs->depth_offset_back =
687 fui(state->offset_units);
688 rs->depth_scale_front = rs->depth_scale_back =
689 fui(state->offset_scale);
690 }
691
692 if (state->line_stipple_enable) {
693 rs->line_stipple_config =
694 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
695 (fui((float)state->line_stipple_factor) &
696 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
697 /* XXX this might need to be scaled up */
698 rs->line_stipple_value = state->line_stipple_pattern;
699 }
700
701 if (state->flatshade) {
702 rs->color_control = R300_SHADE_MODEL_FLAT;
703 } else {
704 rs->color_control = R300_SHADE_MODEL_SMOOTH;
705 }
706
707 return (void*)rs;
708 }
709
710 /* Bind rasterizer state. */
711 static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
712 {
713 struct r300_context* r300 = r300_context(pipe);
714 struct r300_rs_state* rs = (struct r300_rs_state*)state;
715
716 if (r300->draw) {
717 draw_flush(r300->draw);
718 draw_set_rasterizer_state(r300->draw, &rs->rs);
719 }
720
721 r300->rs_state = rs;
722 /* XXX Clean these up when we move to atom emits */
723 r300->dirty_state |= R300_NEW_RASTERIZER;
724 r300->dirty_state |= R300_NEW_RS_BLOCK;
725 r300->dirty_state |= R300_NEW_SCISSOR;
726 r300->dirty_state |= R300_NEW_VIEWPORT;
727 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
728 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
729 }
730 }
731
732 /* Free rasterizer state. */
733 static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
734 {
735 FREE(state);
736 }
737
738 static void*
739 r300_create_sampler_state(struct pipe_context* pipe,
740 const struct pipe_sampler_state* state)
741 {
742 struct r300_context* r300 = r300_context(pipe);
743 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
744 int lod_bias;
745 union util_color uc;
746
747 sampler->state = *state;
748
749 sampler->filter0 |=
750 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
751 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
752 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
753
754 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
755 state->mag_img_filter,
756 state->min_mip_filter,
757 state->max_anisotropy > 1.0);
758
759 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
760 /* We must pass these to the emit function to clamp them properly. */
761 sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
762 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
763
764 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
765
766 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
767
768 sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
769
770 util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
771 sampler->border_color = uc.ui;
772
773 /* R500-specific fixups and optimizations */
774 if (r300_screen(r300->context.screen)->caps->is_r500) {
775 sampler->filter1 |= R500_BORDER_FIX;
776 }
777
778 return (void*)sampler;
779 }
780
781 static void r300_bind_sampler_states(struct pipe_context* pipe,
782 unsigned count,
783 void** states)
784 {
785 struct r300_context* r300 = r300_context(pipe);
786 int i;
787
788 if (count > 8) {
789 return;
790 }
791
792 for (i = 0; i < count; i++) {
793 if (r300->sampler_states[i] != states[i]) {
794 r300->sampler_states[i] = (struct r300_sampler_state*)states[i];
795 r300->dirty_state |= (R300_NEW_SAMPLER << i);
796 }
797 }
798
799 r300->sampler_count = count;
800
801 /* Pick a fragment shader based on the texture compare state. */
802 if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) {
803 if (r300_pick_fragment_shader(r300)) {
804 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
805 R300_NEW_FRAGMENT_SHADER_CONSTANTS;
806 }
807 }
808 }
809
810 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
811 unsigned count,
812 void** states)
813 {
814 }
815
816 static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
817 {
818 FREE(state);
819 }
820
821 static void r300_set_sampler_textures(struct pipe_context* pipe,
822 unsigned count,
823 struct pipe_texture** texture)
824 {
825 struct r300_context* r300 = r300_context(pipe);
826 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
827 int i;
828
829 /* XXX magic num */
830 if (count > 8) {
831 return;
832 }
833
834 for (i = 0; i < count; i++) {
835 if (r300->textures[i] != (struct r300_texture*)texture[i]) {
836 pipe_texture_reference((struct pipe_texture**)&r300->textures[i],
837 texture[i]);
838 r300->dirty_state |= (R300_NEW_TEXTURE << i);
839
840 /* R300-specific - set the texrect factor in a fragment shader */
841 if (!is_r500 && r300->textures[i]->is_npot) {
842 /* XXX It would be nice to re-emit just 1 constant,
843 * XXX not all of them */
844 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
845 }
846 }
847 }
848
849 for (i = count; i < 8; i++) {
850 if (r300->textures[i]) {
851 pipe_texture_reference((struct pipe_texture**)&r300->textures[i],
852 NULL);
853 r300->dirty_state |= (R300_NEW_TEXTURE << i);
854 }
855 }
856
857 r300->texture_count = count;
858 }
859
860 static void r300_set_scissor_state(struct pipe_context* pipe,
861 const struct pipe_scissor_state* state)
862 {
863 struct r300_context* r300 = r300_context(pipe);
864
865 r300_set_scissor_regs(state, &r300->scissor_state->scissor,
866 r300_screen(r300->context.screen)->caps->is_r500);
867
868 /* Don't rely on the order of states being set for the first time. */
869 if (!r300->rs_state || r300->rs_state->rs.scissor) {
870 r300->dirty_state |= R300_NEW_SCISSOR;
871 }
872 }
873
874 static void r300_set_viewport_state(struct pipe_context* pipe,
875 const struct pipe_viewport_state* state)
876 {
877 struct r300_context* r300 = r300_context(pipe);
878
879 /* Do the transform in HW. */
880 r300->viewport_state->vte_control = R300_VTX_W0_FMT;
881
882 if (state->scale[0] != 1.0f) {
883 r300->viewport_state->xscale = state->scale[0];
884 r300->viewport_state->vte_control |= R300_VPORT_X_SCALE_ENA;
885 }
886 if (state->scale[1] != 1.0f) {
887 r300->viewport_state->yscale = state->scale[1];
888 r300->viewport_state->vte_control |= R300_VPORT_Y_SCALE_ENA;
889 }
890 if (state->scale[2] != 1.0f) {
891 r300->viewport_state->zscale = state->scale[2];
892 r300->viewport_state->vte_control |= R300_VPORT_Z_SCALE_ENA;
893 }
894 if (state->translate[0] != 0.0f) {
895 r300->viewport_state->xoffset = state->translate[0];
896 r300->viewport_state->vte_control |= R300_VPORT_X_OFFSET_ENA;
897 }
898 if (state->translate[1] != 0.0f) {
899 r300->viewport_state->yoffset = state->translate[1];
900 r300->viewport_state->vte_control |= R300_VPORT_Y_OFFSET_ENA;
901 }
902 if (state->translate[2] != 0.0f) {
903 r300->viewport_state->zoffset = state->translate[2];
904 r300->viewport_state->vte_control |= R300_VPORT_Z_OFFSET_ENA;
905 }
906
907 r300->dirty_state |= R300_NEW_VIEWPORT;
908 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
909 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
910 }
911 }
912
913 static void r300_set_vertex_buffers(struct pipe_context* pipe,
914 unsigned count,
915 const struct pipe_vertex_buffer* buffers)
916 {
917 struct r300_context* r300 = r300_context(pipe);
918
919 memcpy(r300->vertex_buffer, buffers,
920 sizeof(struct pipe_vertex_buffer) * count);
921 r300->vertex_buffer_count = count;
922
923 if (r300->draw) {
924 draw_flush(r300->draw);
925 draw_set_vertex_buffers(r300->draw, count, buffers);
926 }
927
928 r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
929 }
930
931 static void r300_set_vertex_elements(struct pipe_context* pipe,
932 unsigned count,
933 const struct pipe_vertex_element* elements)
934 {
935 struct r300_context* r300 = r300_context(pipe);
936
937 memcpy(r300->vertex_element,
938 elements,
939 sizeof(struct pipe_vertex_element) * count);
940 r300->vertex_element_count = count;
941
942 if (r300->draw) {
943 draw_flush(r300->draw);
944 draw_set_vertex_elements(r300->draw, count, elements);
945 }
946 }
947
948 static void* r300_create_vs_state(struct pipe_context* pipe,
949 const struct pipe_shader_state* shader)
950 {
951 struct r300_context* r300 = r300_context(pipe);
952
953 if (r300_screen(pipe->screen)->caps->has_tcl) {
954 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
955 /* Copy state directly into shader. */
956 vs->state = *shader;
957 vs->state.tokens = tgsi_dup_tokens(shader->tokens);
958
959 tgsi_scan_shader(shader->tokens, &vs->info);
960
961 return (void*)vs;
962 } else {
963 return draw_create_vertex_shader(r300->draw, shader);
964 }
965 }
966
967 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
968 {
969 struct r300_context* r300 = r300_context(pipe);
970
971 if (r300_screen(pipe->screen)->caps->has_tcl) {
972 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
973
974 if (vs == NULL) {
975 r300->vs = NULL;
976 return;
977 } else if (!vs->translated) {
978 r300_translate_vertex_shader(r300, vs);
979 }
980
981 r300->vs = vs;
982 if (r300->fs) {
983 r300_vertex_shader_setup_wpos(r300);
984 }
985
986 r300->dirty_state |=
987 R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS |
988 R300_NEW_VERTEX_FORMAT;
989 } else {
990 draw_flush(r300->draw);
991 draw_bind_vertex_shader(r300->draw,
992 (struct draw_vertex_shader*)shader);
993 }
994 }
995
996 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
997 {
998 struct r300_context* r300 = r300_context(pipe);
999
1000 if (r300_screen(pipe->screen)->caps->has_tcl) {
1001 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1002
1003 rc_constants_destroy(&vs->code.constants);
1004 FREE((void*)vs->state.tokens);
1005 FREE(shader);
1006 } else {
1007 draw_delete_vertex_shader(r300->draw,
1008 (struct draw_vertex_shader*)shader);
1009 }
1010 }
1011
1012 static void r300_set_constant_buffer(struct pipe_context *pipe,
1013 uint shader, uint index,
1014 const struct pipe_constant_buffer *buf)
1015 {
1016 struct r300_context* r300 = r300_context(pipe);
1017 void *mapped;
1018
1019 if (buf == NULL || buf->buffer->size == 0 ||
1020 (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
1021 {
1022 r300->shader_constants[shader].count = 0;
1023 return;
1024 }
1025
1026 assert((buf->buffer->size % 4 * sizeof(float)) == 0);
1027 memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size);
1028 r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float));
1029 pipe_buffer_unmap(pipe->screen, buf->buffer);
1030
1031 if (shader == PIPE_SHADER_VERTEX)
1032 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
1033 else if (shader == PIPE_SHADER_FRAGMENT)
1034 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1035 }
1036
1037 void r300_init_state_functions(struct r300_context* r300)
1038 {
1039 r300->context.create_blend_state = r300_create_blend_state;
1040 r300->context.bind_blend_state = r300_bind_blend_state;
1041 r300->context.delete_blend_state = r300_delete_blend_state;
1042
1043 r300->context.set_blend_color = r300_set_blend_color;
1044
1045 r300->context.set_clip_state = r300_set_clip_state;
1046
1047 r300->context.set_constant_buffer = r300_set_constant_buffer;
1048
1049 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1050 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1051 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1052
1053 r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1054
1055 r300->context.create_fs_state = r300_create_fs_state;
1056 r300->context.bind_fs_state = r300_bind_fs_state;
1057 r300->context.delete_fs_state = r300_delete_fs_state;
1058
1059 r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1060
1061 r300->context.create_rasterizer_state = r300_create_rs_state;
1062 r300->context.bind_rasterizer_state = r300_bind_rs_state;
1063 r300->context.delete_rasterizer_state = r300_delete_rs_state;
1064
1065 r300->context.create_sampler_state = r300_create_sampler_state;
1066 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1067 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1068 r300->context.delete_sampler_state = r300_delete_sampler_state;
1069
1070 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures;
1071
1072 r300->context.set_scissor_state = r300_set_scissor_state;
1073
1074 r300->context.set_viewport_state = r300_set_viewport_state;
1075
1076 r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1077 r300->context.set_vertex_elements = r300_set_vertex_elements;
1078
1079 r300->context.create_vs_state = r300_create_vs_state;
1080 r300->context.bind_vs_state = r300_bind_vs_state;
1081 r300->context.delete_vs_state = r300_delete_vs_state;
1082 }