Merge branch 'master' of ssh://people.freedesktop.org/~jbarnes/mesa
[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.state = state;
321 r300->blend_state.dirty = TRUE;
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 struct r300_blend_color_state* state =
344 (struct r300_blend_color_state*)r300->blend_color_state.state;
345 union util_color uc;
346
347 util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
348 state->blend_color = uc.ui;
349
350 /* XXX if FP16 blending is enabled, we should use the FP16 format */
351 state->blend_color_red_alpha =
352 float_to_fixed10(color->color[0]) |
353 (float_to_fixed10(color->color[3]) << 16);
354 state->blend_color_green_blue =
355 float_to_fixed10(color->color[2]) |
356 (float_to_fixed10(color->color[1]) << 16);
357
358 r300->blend_color_state.dirty = TRUE;
359 }
360
361 static void r300_set_clip_state(struct pipe_context* pipe,
362 const struct pipe_clip_state* state)
363 {
364 struct r300_context* r300 = r300_context(pipe);
365
366 if (r300_screen(pipe->screen)->caps->has_tcl) {
367 memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
368 r300->clip_state.dirty = TRUE;
369 } else {
370 draw_flush(r300->draw);
371 draw_set_clip_state(r300->draw, state);
372 }
373 }
374
375 /* Create a new depth, stencil, and alpha state based on the CSO dsa state.
376 *
377 * This contains the depth buffer, stencil buffer, alpha test, and such.
378 * On the Radeon, depth and stencil buffer setup are intertwined, which is
379 * the reason for some of the strange-looking assignments across registers. */
380 static void*
381 r300_create_dsa_state(struct pipe_context* pipe,
382 const struct pipe_depth_stencil_alpha_state* state)
383 {
384 struct r300_capabilities *caps =
385 r300_screen(r300_context(pipe)->context.screen)->caps;
386 struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
387
388 /* Depth test setup. */
389 if (state->depth.enabled) {
390 dsa->z_buffer_control |= R300_Z_ENABLE;
391
392 if (state->depth.writemask) {
393 dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
394 }
395
396 dsa->z_stencil_control |=
397 (r300_translate_depth_stencil_function(state->depth.func) <<
398 R300_Z_FUNC_SHIFT);
399 }
400
401 /* Stencil buffer setup. */
402 if (state->stencil[0].enabled) {
403 dsa->z_buffer_control |= R300_STENCIL_ENABLE;
404 dsa->z_stencil_control |=
405 (r300_translate_depth_stencil_function(state->stencil[0].func) <<
406 R300_S_FRONT_FUNC_SHIFT) |
407 (r300_translate_stencil_op(state->stencil[0].fail_op) <<
408 R300_S_FRONT_SFAIL_OP_SHIFT) |
409 (r300_translate_stencil_op(state->stencil[0].zpass_op) <<
410 R300_S_FRONT_ZPASS_OP_SHIFT) |
411 (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
412 R300_S_FRONT_ZFAIL_OP_SHIFT);
413
414 dsa->stencil_ref_mask = (state->stencil[0].ref_value) |
415 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
416 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
417
418 if (state->stencil[1].enabled) {
419 dsa->z_buffer_control |= R300_STENCIL_FRONT_BACK;
420 dsa->z_stencil_control |=
421 (r300_translate_depth_stencil_function(state->stencil[1].func) <<
422 R300_S_BACK_FUNC_SHIFT) |
423 (r300_translate_stencil_op(state->stencil[1].fail_op) <<
424 R300_S_BACK_SFAIL_OP_SHIFT) |
425 (r300_translate_stencil_op(state->stencil[1].zpass_op) <<
426 R300_S_BACK_ZPASS_OP_SHIFT) |
427 (r300_translate_stencil_op(state->stencil[1].zfail_op) <<
428 R300_S_BACK_ZFAIL_OP_SHIFT);
429
430 /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
431 if (caps->is_r500)
432 {
433 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
434 dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
435 (state->stencil[1].valuemask <<
436 R300_STENCILMASK_SHIFT) |
437 (state->stencil[1].writemask <<
438 R300_STENCILWRITEMASK_SHIFT);
439 }
440 }
441 }
442
443 /* Alpha test setup. */
444 if (state->alpha.enabled) {
445 dsa->alpha_function =
446 r300_translate_alpha_function(state->alpha.func) |
447 R300_FG_ALPHA_FUNC_ENABLE;
448
449 /* XXX figure out why emitting 10bit alpha ref causes CS to dump */
450 /* always use 8bit alpha ref */
451 dsa->alpha_function |= float_to_ubyte(state->alpha.ref_value);
452
453 if (caps->is_r500)
454 dsa->alpha_function |= R500_FG_ALPHA_FUNC_8BIT;
455 }
456
457 return (void*)dsa;
458 }
459
460 /* Bind DSA state. */
461 static void r300_bind_dsa_state(struct pipe_context* pipe,
462 void* state)
463 {
464 struct r300_context* r300 = r300_context(pipe);
465
466 r300->dsa_state.state = state;
467 r300->dsa_state.dirty = TRUE;
468 }
469
470 /* Free DSA state. */
471 static void r300_delete_dsa_state(struct pipe_context* pipe,
472 void* state)
473 {
474 FREE(state);
475 }
476
477 static void r300_set_scissor_regs(const struct pipe_scissor_state* state,
478 struct r300_scissor_regs *scissor,
479 boolean is_r500)
480 {
481 if (is_r500) {
482 scissor->top_left =
483 (state->minx << R300_SCISSORS_X_SHIFT) |
484 (state->miny << R300_SCISSORS_Y_SHIFT);
485 scissor->bottom_right =
486 ((state->maxx - 1) << R300_SCISSORS_X_SHIFT) |
487 ((state->maxy - 1) << R300_SCISSORS_Y_SHIFT);
488 } else {
489 /* Offset of 1440 in non-R500 chipsets. */
490 scissor->top_left =
491 ((state->minx + 1440) << R300_SCISSORS_X_SHIFT) |
492 ((state->miny + 1440) << R300_SCISSORS_Y_SHIFT);
493 scissor->bottom_right =
494 (((state->maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
495 (((state->maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
496 }
497 }
498
499 static void
500 r300_set_framebuffer_state(struct pipe_context* pipe,
501 const struct pipe_framebuffer_state* state)
502 {
503 struct r300_context* r300 = r300_context(pipe);
504 struct r300_scissor_state* scissor =
505 (struct r300_scissor_state*)r300->scissor_state.state;
506 struct pipe_scissor_state pscissor;
507
508 if (r300->draw) {
509 draw_flush(r300->draw);
510 }
511
512 r300->framebuffer_state = *state;
513
514 /* XXX Arg. This is silly. */
515 pscissor.minx = pscissor.miny = 0;
516 pscissor.maxx = state->width;
517 pscissor.maxy = state->height;
518 r300_set_scissor_regs(&pscissor, &scissor->framebuffer,
519 r300_screen(r300->context.screen)->caps->is_r500);
520
521 /* Don't rely on the order of states being set for the first time. */
522 r300->dirty_state |= R300_NEW_FRAMEBUFFERS;
523
524 r300->blend_state.dirty = TRUE;
525 r300->dsa_state.dirty = TRUE;
526 r300->scissor_state.dirty = TRUE;
527 }
528
529 /* Create fragment shader state. */
530 static void* r300_create_fs_state(struct pipe_context* pipe,
531 const struct pipe_shader_state* shader)
532 {
533 struct r300_fragment_shader* fs = NULL;
534
535 fs = (struct r300_fragment_shader*)CALLOC_STRUCT(r300_fragment_shader);
536
537 /* Copy state directly into shader. */
538 fs->state = *shader;
539 fs->state.tokens = tgsi_dup_tokens(shader->tokens);
540
541 tgsi_scan_shader(shader->tokens, &fs->info);
542 r300_shader_read_fs_inputs(&fs->info, &fs->inputs);
543
544 return (void*)fs;
545 }
546
547 /* Bind fragment shader state. */
548 static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
549 {
550 struct r300_context* r300 = r300_context(pipe);
551 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
552
553 if (fs == NULL) {
554 r300->fs = NULL;
555 return;
556 }
557
558 r300->fs = fs;
559 r300_pick_fragment_shader(r300);
560
561 if (r300->vs && r300_vertex_shader_setup_wpos(r300)) {
562 r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
563 }
564
565 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
566 }
567
568 /* Delete fragment shader state. */
569 static void r300_delete_fs_state(struct pipe_context* pipe, void* shader)
570 {
571 struct r300_fragment_shader* fs = (struct r300_fragment_shader*)shader;
572 struct r300_fragment_shader_code *tmp, *ptr = fs->first;
573
574 while (ptr) {
575 tmp = ptr;
576 ptr = ptr->next;
577 rc_constants_destroy(&tmp->code.constants);
578 FREE(tmp);
579 }
580 FREE((void*)fs->state.tokens);
581 FREE(shader);
582 }
583
584 static void r300_set_polygon_stipple(struct pipe_context* pipe,
585 const struct pipe_poly_stipple* state)
586 {
587 /* XXX no idea how to set this up, but not terribly important */
588 }
589
590 /* Create a new rasterizer state based on the CSO rasterizer state.
591 *
592 * This is a very large chunk of state, and covers most of the graphics
593 * backend (GB), geometry assembly (GA), and setup unit (SU) blocks.
594 *
595 * In a not entirely unironic sidenote, this state has nearly nothing to do
596 * with the actual block on the Radeon called the rasterizer (RS). */
597 static void* r300_create_rs_state(struct pipe_context* pipe,
598 const struct pipe_rasterizer_state* state)
599 {
600 struct r300_rs_state* rs = CALLOC_STRUCT(r300_rs_state);
601
602 /* Copy rasterizer state for Draw. */
603 rs->rs = *state;
604
605 #ifdef PIPE_ARCH_LITTLE_ENDIAN
606 rs->vap_control_status = R300_VC_NO_SWAP;
607 #else
608 rs->vap_control_status = R300_VC_32BIT_SWAP;
609 #endif
610
611 /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
612 * Else, enable HW TCL and force Draw's TCL off. */
613 if (state->bypass_vs_clip_and_viewport ||
614 !r300_screen(pipe->screen)->caps->has_tcl) {
615 rs->vap_control_status |= R300_VAP_TCL_BYPASS;
616 }
617
618 rs->point_size = pack_float_16_6x(state->point_size) |
619 (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
620
621 rs->point_minmax =
622 ((int)(state->point_size_min * 6.0) <<
623 R300_GA_POINT_MINMAX_MIN_SHIFT) |
624 ((int)(state->point_size_max * 6.0) <<
625 R300_GA_POINT_MINMAX_MAX_SHIFT);
626
627 rs->line_control = pack_float_16_6x(state->line_width) |
628 R300_GA_LINE_CNTL_END_TYPE_COMP;
629
630 /* XXX I think there is something wrong with the polygon mode,
631 * XXX re-test when r300g is in a better shape */
632
633 /* Enable polygon mode */
634 if (state->fill_cw != PIPE_POLYGON_MODE_FILL ||
635 state->fill_ccw != PIPE_POLYGON_MODE_FILL) {
636 rs->polygon_mode = R300_GA_POLY_MODE_DUAL;
637 }
638
639 /* Radeons don't think in "CW/CCW", they think in "front/back". */
640 if (state->front_winding == PIPE_WINDING_CW) {
641 rs->cull_mode = R300_FRONT_FACE_CW;
642
643 /* Polygon offset */
644 if (state->offset_cw) {
645 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
646 }
647 if (state->offset_ccw) {
648 rs->polygon_offset_enable |= R300_BACK_ENABLE;
649 }
650
651 /* Polygon mode */
652 if (rs->polygon_mode) {
653 rs->polygon_mode |=
654 r300_translate_polygon_mode_front(state->fill_cw);
655 rs->polygon_mode |=
656 r300_translate_polygon_mode_back(state->fill_ccw);
657 }
658 } else {
659 rs->cull_mode = R300_FRONT_FACE_CCW;
660
661 /* Polygon offset */
662 if (state->offset_ccw) {
663 rs->polygon_offset_enable |= R300_FRONT_ENABLE;
664 }
665 if (state->offset_cw) {
666 rs->polygon_offset_enable |= R300_BACK_ENABLE;
667 }
668
669 /* Polygon mode */
670 if (rs->polygon_mode) {
671 rs->polygon_mode |=
672 r300_translate_polygon_mode_front(state->fill_ccw);
673 rs->polygon_mode |=
674 r300_translate_polygon_mode_back(state->fill_cw);
675 }
676 }
677 if (state->front_winding & state->cull_mode) {
678 rs->cull_mode |= R300_CULL_FRONT;
679 }
680 if (~(state->front_winding) & state->cull_mode) {
681 rs->cull_mode |= R300_CULL_BACK;
682 }
683
684 if (rs->polygon_offset_enable) {
685 rs->depth_offset_front = rs->depth_offset_back =
686 fui(state->offset_units);
687 rs->depth_scale_front = rs->depth_scale_back =
688 fui(state->offset_scale);
689 }
690
691 if (state->line_stipple_enable) {
692 rs->line_stipple_config =
693 R300_GA_LINE_STIPPLE_CONFIG_LINE_RESET_LINE |
694 (fui((float)state->line_stipple_factor) &
695 R300_GA_LINE_STIPPLE_CONFIG_STIPPLE_SCALE_MASK);
696 /* XXX this might need to be scaled up */
697 rs->line_stipple_value = state->line_stipple_pattern;
698 }
699
700 if (state->flatshade) {
701 rs->color_control = R300_SHADE_MODEL_FLAT;
702 } else {
703 rs->color_control = R300_SHADE_MODEL_SMOOTH;
704 }
705
706 return (void*)rs;
707 }
708
709 /* Bind rasterizer state. */
710 static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
711 {
712 struct r300_context* r300 = r300_context(pipe);
713 struct r300_rs_state* rs = (struct r300_rs_state*)state;
714
715 if (r300->draw) {
716 draw_flush(r300->draw);
717 draw_set_rasterizer_state(r300->draw, &rs->rs);
718 }
719
720 r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport;
721
722 r300->rs_state.state = rs;
723 r300->rs_state.dirty = TRUE;
724 /* XXX Why is this still needed, dammit!? */
725 r300->scissor_state.dirty = TRUE;
726 r300->viewport_state.dirty = TRUE;
727
728 /* XXX Clean these up when we move to atom emits */
729 r300->dirty_state |= R300_NEW_RS_BLOCK;
730 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
731 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
732 }
733 }
734
735 /* Free rasterizer state. */
736 static void r300_delete_rs_state(struct pipe_context* pipe, void* state)
737 {
738 FREE(state);
739 }
740
741 static void*
742 r300_create_sampler_state(struct pipe_context* pipe,
743 const struct pipe_sampler_state* state)
744 {
745 struct r300_context* r300 = r300_context(pipe);
746 struct r300_sampler_state* sampler = CALLOC_STRUCT(r300_sampler_state);
747 int lod_bias;
748 union util_color uc;
749
750 sampler->state = *state;
751
752 sampler->filter0 |=
753 (r300_translate_wrap(state->wrap_s) << R300_TX_WRAP_S_SHIFT) |
754 (r300_translate_wrap(state->wrap_t) << R300_TX_WRAP_T_SHIFT) |
755 (r300_translate_wrap(state->wrap_r) << R300_TX_WRAP_R_SHIFT);
756
757 sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
758 state->mag_img_filter,
759 state->min_mip_filter,
760 state->max_anisotropy > 1.0);
761
762 /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
763 /* We must pass these to the emit function to clamp them properly. */
764 sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
765 sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
766
767 lod_bias = CLAMP((int)(state->lod_bias * 32), -(1 << 9), (1 << 9) - 1);
768
769 sampler->filter1 |= lod_bias << R300_LOD_BIAS_SHIFT;
770
771 sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
772
773 util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
774 sampler->border_color = uc.ui;
775
776 /* R500-specific fixups and optimizations */
777 if (r300_screen(r300->context.screen)->caps->is_r500) {
778 sampler->filter1 |= R500_BORDER_FIX;
779 }
780
781 return (void*)sampler;
782 }
783
784 static void r300_bind_sampler_states(struct pipe_context* pipe,
785 unsigned count,
786 void** states)
787 {
788 struct r300_context* r300 = r300_context(pipe);
789 int i;
790
791 if (count > 8) {
792 return;
793 }
794
795 for (i = 0; i < count; i++) {
796 if (r300->sampler_states[i] != states[i]) {
797 r300->sampler_states[i] = (struct r300_sampler_state*)states[i];
798 r300->dirty_state |= (R300_NEW_SAMPLER << i);
799 }
800 }
801
802 r300->sampler_count = count;
803
804 /* Pick a fragment shader based on the texture compare state. */
805 if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) {
806 if (r300_pick_fragment_shader(r300)) {
807 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
808 R300_NEW_FRAGMENT_SHADER_CONSTANTS;
809 }
810 }
811 }
812
813 static void r300_lacks_vertex_textures(struct pipe_context* pipe,
814 unsigned count,
815 void** states)
816 {
817 }
818
819 static void r300_delete_sampler_state(struct pipe_context* pipe, void* state)
820 {
821 FREE(state);
822 }
823
824 static void r300_set_sampler_textures(struct pipe_context* pipe,
825 unsigned count,
826 struct pipe_texture** texture)
827 {
828 struct r300_context* r300 = r300_context(pipe);
829 boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
830 int i;
831
832 /* XXX magic num */
833 if (count > 8) {
834 return;
835 }
836
837 for (i = 0; i < count; i++) {
838 if (r300->textures[i] != (struct r300_texture*)texture[i]) {
839 pipe_texture_reference((struct pipe_texture**)&r300->textures[i],
840 texture[i]);
841 r300->dirty_state |= (R300_NEW_TEXTURE << i);
842
843 /* R300-specific - set the texrect factor in a fragment shader */
844 if (!is_r500 && r300->textures[i]->is_npot) {
845 /* XXX It would be nice to re-emit just 1 constant,
846 * XXX not all of them */
847 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
848 }
849 }
850 }
851
852 for (i = count; i < 8; i++) {
853 if (r300->textures[i]) {
854 pipe_texture_reference((struct pipe_texture**)&r300->textures[i],
855 NULL);
856 r300->dirty_state |= (R300_NEW_TEXTURE << i);
857 }
858 }
859
860 r300->texture_count = count;
861 }
862
863 static void r300_set_scissor_state(struct pipe_context* pipe,
864 const struct pipe_scissor_state* state)
865 {
866 struct r300_context* r300 = r300_context(pipe);
867 struct r300_scissor_state* scissor =
868 (struct r300_scissor_state*)r300->scissor_state.state;
869
870 r300_set_scissor_regs(state, &scissor->scissor,
871 r300_screen(r300->context.screen)->caps->is_r500);
872
873 r300->scissor_state.dirty = TRUE;
874 }
875
876 static void r300_set_viewport_state(struct pipe_context* pipe,
877 const struct pipe_viewport_state* state)
878 {
879 struct r300_context* r300 = r300_context(pipe);
880 struct r300_viewport_state* viewport =
881 (struct r300_viewport_state*)r300->viewport_state.state;
882
883 /* Do the transform in HW. */
884 viewport->vte_control = R300_VTX_W0_FMT;
885
886 if (state->scale[0] != 1.0f) {
887 viewport->xscale = state->scale[0];
888 viewport->vte_control |= R300_VPORT_X_SCALE_ENA;
889 }
890 if (state->scale[1] != 1.0f) {
891 viewport->yscale = state->scale[1];
892 viewport->vte_control |= R300_VPORT_Y_SCALE_ENA;
893 }
894 if (state->scale[2] != 1.0f) {
895 viewport->zscale = state->scale[2];
896 viewport->vte_control |= R300_VPORT_Z_SCALE_ENA;
897 }
898 if (state->translate[0] != 0.0f) {
899 viewport->xoffset = state->translate[0];
900 viewport->vte_control |= R300_VPORT_X_OFFSET_ENA;
901 }
902 if (state->translate[1] != 0.0f) {
903 viewport->yoffset = state->translate[1];
904 viewport->vte_control |= R300_VPORT_Y_OFFSET_ENA;
905 }
906 if (state->translate[2] != 0.0f) {
907 viewport->zoffset = state->translate[2];
908 viewport->vte_control |= R300_VPORT_Z_OFFSET_ENA;
909 }
910
911 r300->viewport_state.dirty = TRUE;
912 if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
913 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
914 }
915 }
916
917 static void r300_set_vertex_buffers(struct pipe_context* pipe,
918 unsigned count,
919 const struct pipe_vertex_buffer* buffers)
920 {
921 struct r300_context* r300 = r300_context(pipe);
922
923 memcpy(r300->vertex_buffer, buffers,
924 sizeof(struct pipe_vertex_buffer) * count);
925 r300->vertex_buffer_count = count;
926
927 if (r300->draw) {
928 draw_flush(r300->draw);
929 draw_set_vertex_buffers(r300->draw, count, buffers);
930 }
931
932 r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
933 }
934
935 static void r300_set_vertex_elements(struct pipe_context* pipe,
936 unsigned count,
937 const struct pipe_vertex_element* elements)
938 {
939 struct r300_context* r300 = r300_context(pipe);
940
941 memcpy(r300->vertex_element,
942 elements,
943 sizeof(struct pipe_vertex_element) * count);
944 r300->vertex_element_count = count;
945
946 if (r300->draw) {
947 draw_flush(r300->draw);
948 draw_set_vertex_elements(r300->draw, count, elements);
949 }
950 }
951
952 static void* r300_create_vs_state(struct pipe_context* pipe,
953 const struct pipe_shader_state* shader)
954 {
955 struct r300_context* r300 = r300_context(pipe);
956
957 if (r300_screen(pipe->screen)->caps->has_tcl) {
958 struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
959 /* Copy state directly into shader. */
960 vs->state = *shader;
961 vs->state.tokens = tgsi_dup_tokens(shader->tokens);
962
963 tgsi_scan_shader(shader->tokens, &vs->info);
964
965 return (void*)vs;
966 } else {
967 return draw_create_vertex_shader(r300->draw, shader);
968 }
969 }
970
971 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
972 {
973 struct r300_context* r300 = r300_context(pipe);
974
975 if (r300_screen(pipe->screen)->caps->has_tcl) {
976 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
977
978 if (vs == NULL) {
979 r300->vs = NULL;
980 return;
981 } else if (!vs->translated) {
982 r300_translate_vertex_shader(r300, vs);
983 }
984
985 r300->vs = vs;
986 if (r300->fs) {
987 r300_vertex_shader_setup_wpos(r300);
988 }
989
990 r300->dirty_state |=
991 R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS |
992 R300_NEW_VERTEX_FORMAT;
993 } else {
994 draw_flush(r300->draw);
995 draw_bind_vertex_shader(r300->draw,
996 (struct draw_vertex_shader*)shader);
997 }
998 }
999
1000 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
1001 {
1002 struct r300_context* r300 = r300_context(pipe);
1003
1004 if (r300_screen(pipe->screen)->caps->has_tcl) {
1005 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
1006
1007 rc_constants_destroy(&vs->code.constants);
1008 FREE((void*)vs->state.tokens);
1009 FREE(shader);
1010 } else {
1011 draw_delete_vertex_shader(r300->draw,
1012 (struct draw_vertex_shader*)shader);
1013 }
1014 }
1015
1016 static void r300_set_constant_buffer(struct pipe_context *pipe,
1017 uint shader, uint index,
1018 const struct pipe_constant_buffer *buf)
1019 {
1020 struct r300_context* r300 = r300_context(pipe);
1021 void *mapped;
1022
1023 if (buf == NULL || buf->buffer->size == 0 ||
1024 (mapped = pipe_buffer_map(pipe->screen, buf->buffer, PIPE_BUFFER_USAGE_CPU_READ)) == NULL)
1025 {
1026 r300->shader_constants[shader].count = 0;
1027 return;
1028 }
1029
1030 assert((buf->buffer->size % 4 * sizeof(float)) == 0);
1031 memcpy(r300->shader_constants[shader].constants, mapped, buf->buffer->size);
1032 r300->shader_constants[shader].count = buf->buffer->size / (4 * sizeof(float));
1033 pipe_buffer_unmap(pipe->screen, buf->buffer);
1034
1035 if (shader == PIPE_SHADER_VERTEX)
1036 r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
1037 else if (shader == PIPE_SHADER_FRAGMENT)
1038 r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1039 }
1040
1041 void r300_init_state_functions(struct r300_context* r300)
1042 {
1043 r300->context.create_blend_state = r300_create_blend_state;
1044 r300->context.bind_blend_state = r300_bind_blend_state;
1045 r300->context.delete_blend_state = r300_delete_blend_state;
1046
1047 r300->context.set_blend_color = r300_set_blend_color;
1048
1049 r300->context.set_clip_state = r300_set_clip_state;
1050
1051 r300->context.set_constant_buffer = r300_set_constant_buffer;
1052
1053 r300->context.create_depth_stencil_alpha_state = r300_create_dsa_state;
1054 r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
1055 r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
1056
1057 r300->context.set_framebuffer_state = r300_set_framebuffer_state;
1058
1059 r300->context.create_fs_state = r300_create_fs_state;
1060 r300->context.bind_fs_state = r300_bind_fs_state;
1061 r300->context.delete_fs_state = r300_delete_fs_state;
1062
1063 r300->context.set_polygon_stipple = r300_set_polygon_stipple;
1064
1065 r300->context.create_rasterizer_state = r300_create_rs_state;
1066 r300->context.bind_rasterizer_state = r300_bind_rs_state;
1067 r300->context.delete_rasterizer_state = r300_delete_rs_state;
1068
1069 r300->context.create_sampler_state = r300_create_sampler_state;
1070 r300->context.bind_fragment_sampler_states = r300_bind_sampler_states;
1071 r300->context.bind_vertex_sampler_states = r300_lacks_vertex_textures;
1072 r300->context.delete_sampler_state = r300_delete_sampler_state;
1073
1074 r300->context.set_fragment_sampler_textures = r300_set_sampler_textures;
1075
1076 r300->context.set_scissor_state = r300_set_scissor_state;
1077
1078 r300->context.set_viewport_state = r300_set_viewport_state;
1079
1080 r300->context.set_vertex_buffers = r300_set_vertex_buffers;
1081 r300->context.set_vertex_elements = r300_set_vertex_elements;
1082
1083 r300->context.create_vs_state = r300_create_vs_state;
1084 r300->context.bind_vs_state = r300_bind_vs_state;
1085 r300->context.delete_vs_state = r300_delete_vs_state;
1086 }