projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r300g: fix zero-area scissor
[mesa.git] / src / gallium / drivers / r300 / r300_emit.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 /* r300_emit: Functions for emitting state. */
25
26 #include "util/u_format.h"
27 #include "util/u_math.h"
28 #include "util/u_simple_list.h"
29
30 #include "r300_context.h"
31 #include "r300_cs.h"
32 #include "r300_emit.h"
33 #include "r300_fs.h"
34 #include "r300_screen.h"
35 #include "r300_state_derived.h"
36 #include "r300_state_inlines.h"
37 #include "r300_texture.h"
38 #include "r300_vs.h"
39
40 void r300_emit_blend_state(struct r300_context* r300, void* state)
41 {
42 struct r300_blend_state* blend = (struct r300_blend_state*)state;
43 CS_LOCALS(r300);
44
45 BEGIN_CS(8);
46 OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
47 OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3);
48 if (r300->framebuffer_state.nr_cbufs) {
49 OUT_CS(blend->blend_control);
50 OUT_CS(blend->alpha_blend_control);
51 OUT_CS(blend->color_channel_mask);
52 } else {
53 OUT_CS(0);
54 OUT_CS(0);
55 OUT_CS(0);
56 /* XXX also disable fastfill here once it's supported */
57 }
58 OUT_CS_REG(R300_RB3D_DITHER_CTL, blend->dither);
59 END_CS;
60 }
61
62 void r300_emit_blend_color_state(struct r300_context* r300, void* state)
63 {
64 struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state;
65 struct r300_screen* r300screen = r300_screen(r300->context.screen);
66 CS_LOCALS(r300);
67
68 if (r300screen->caps->is_r500) {
69 BEGIN_CS(3);
70 OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
71 OUT_CS(bc->blend_color_red_alpha);
72 OUT_CS(bc->blend_color_green_blue);
73 END_CS;
74 } else {
75 BEGIN_CS(2);
76 OUT_CS_REG(R300_RB3D_BLEND_COLOR, bc->blend_color);
77 END_CS;
78 }
79 }
80
81 void r300_emit_clip_state(struct r300_context* r300, void* state)
82 {
83 struct pipe_clip_state* clip = (struct pipe_clip_state*)state;
84 int i;
85 struct r300_screen* r300screen = r300_screen(r300->context.screen);
86 CS_LOCALS(r300);
87
88 if (r300screen->caps->has_tcl) {
89 BEGIN_CS(5 + (6 * 4));
90 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
91 (r300screen->caps->is_r500 ?
92 R500_PVS_UCP_START : R300_PVS_UCP_START));
93 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, 6 * 4);
94 for (i = 0; i < 6; i++) {
95 OUT_CS_32F(clip->ucp[i][0]);
96 OUT_CS_32F(clip->ucp[i][1]);
97 OUT_CS_32F(clip->ucp[i][2]);
98 OUT_CS_32F(clip->ucp[i][3]);
99 }
100 OUT_CS_REG(R300_VAP_CLIP_CNTL, ((1 << clip->nr) - 1) |
101 R300_PS_UCP_MODE_CLIP_AS_TRIFAN);
102 END_CS;
103 } else {
104 BEGIN_CS(2);
105 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
106 END_CS;
107 }
108
109 }
110
111 void r300_emit_dsa_state(struct r300_context* r300, void* state)
112 {
113 struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
114 struct r300_screen* r300screen = r300_screen(r300->context.screen);
115 CS_LOCALS(r300);
116
117 BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6);
118 OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
119
120 /* not needed since we use the 8bit alpha ref */
121 /*if (r300screen->caps->is_r500) {
122 OUT_CS_REG(R500_FG_ALPHA_VALUE, dsa->alpha_reference);
123 }*/
124
125 OUT_CS_REG_SEQ(R300_ZB_CNTL, 3);
126
127 if (r300->framebuffer_state.zsbuf) {
128 OUT_CS(dsa->z_buffer_control);
129 OUT_CS(dsa->z_stencil_control);
130 } else {
131 OUT_CS(0);
132 OUT_CS(0);
133 }
134
135 OUT_CS(dsa->stencil_ref_mask);
136
137 /* XXX it seems r3xx doesn't support STENCILREFMASK_BF */
138 if (r300screen->caps->is_r500) {
139 OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf);
140 }
141 END_CS;
142 }
143
144 static const float * get_shader_constant(
145 struct r300_context * r300,
146 struct rc_constant * constant,
147 struct r300_constant_buffer * externals)
148 {
149 struct r300_viewport_state* viewport =
150 (struct r300_viewport_state*)r300->viewport_state.state;
151 static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
152 struct pipe_texture *tex;
153
154 switch(constant->Type) {
155 case RC_CONSTANT_EXTERNAL:
156 return externals->constants[constant->u.External];
157
158 case RC_CONSTANT_IMMEDIATE:
159 return constant->u.Immediate;
160
161 case RC_CONSTANT_STATE:
162 switch (constant->u.State[0]) {
163 /* Factor for converting rectangle coords to
164 * normalized coords. Should only show up on non-r500. */
165 case RC_STATE_R300_TEXRECT_FACTOR:
166 tex = &r300->textures[constant->u.State[1]]->tex;
167 vec[0] = 1.0 / tex->width0;
168 vec[1] = 1.0 / tex->height0;
169 break;
170
171 /* Texture compare-fail value. */
172 /* XXX Since Gallium doesn't support GL_ARB_shadow_ambient,
173 * this is always (0,0,0,0), right? */
174 case RC_STATE_SHADOW_AMBIENT:
175 vec[3] = 0;
176 break;
177
178 case RC_STATE_R300_VIEWPORT_SCALE:
179 if (r300->tcl_bypass) {
180 vec[0] = 1;
181 vec[1] = 1;
182 vec[2] = 1;
183 } else {
184 vec[0] = viewport->xscale;
185 vec[1] = viewport->yscale;
186 vec[2] = viewport->zscale;
187 }
188 break;
189
190 case RC_STATE_R300_VIEWPORT_OFFSET:
191 if (!r300->tcl_bypass) {
192 vec[0] = viewport->xoffset;
193 vec[1] = viewport->yoffset;
194 vec[2] = viewport->zoffset;
195 }
196 break;
197
198 default:
199 debug_printf("r300: Implementation error: "
200 "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
201 }
202 break;
203
204 default:
205 debug_printf("r300: Implementation error: "
206 "Unhandled constant type %d\n", constant->Type);
207 }
208
209 /* This should either be (0, 0, 0, 1), which should be a relatively safe
210 * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
211 * state factors. */
212 return vec;
213 }
214
215 /* Convert a normal single-precision float into the 7.16 format
216 * used by the R300 fragment shader.
217 */
218 static uint32_t pack_float24(float f)
219 {
220 union {
221 float fl;
222 uint32_t u;
223 } u;
224 float mantissa;
225 int exponent;
226 uint32_t float24 = 0;
227
228 if (f == 0.0)
229 return 0;
230
231 u.fl = f;
232
233 mantissa = frexpf(f, &exponent);
234
235 /* Handle -ve */
236 if (mantissa < 0) {
237 float24 |= (1 << 23);
238 mantissa = mantissa * -1.0;
239 }
240 /* Handle exponent, bias of 63 */
241 exponent += 62;
242 float24 |= (exponent << 16);
243 /* Kill 7 LSB of mantissa */
244 float24 |= (u.u & 0x7FFFFF) >> 7;
245
246 return float24;
247 }
248
249 void r300_emit_fragment_program_code(struct r300_context* r300,
250 struct rX00_fragment_program_code* generic_code)
251 {
252 struct r300_fragment_program_code * code = &generic_code->code.r300;
253 int i;
254 CS_LOCALS(r300);
255
256 BEGIN_CS(15 +
257 code->alu.length * 4 +
258 (code->tex.length ? (1 + code->tex.length) : 0));
259
260 OUT_CS_REG(R300_US_CONFIG, code->config);
261 OUT_CS_REG(R300_US_PIXSIZE, code->pixsize);
262 OUT_CS_REG(R300_US_CODE_OFFSET, code->code_offset);
263
264 OUT_CS_REG_SEQ(R300_US_CODE_ADDR_0, 4);
265 for(i = 0; i < 4; ++i)
266 OUT_CS(code->code_addr[i]);
267
268 OUT_CS_REG_SEQ(R300_US_ALU_RGB_INST_0, code->alu.length);
269 for (i = 0; i < code->alu.length; i++)
270 OUT_CS(code->alu.inst[i].rgb_inst);
271
272 OUT_CS_REG_SEQ(R300_US_ALU_RGB_ADDR_0, code->alu.length);
273 for (i = 0; i < code->alu.length; i++)
274 OUT_CS(code->alu.inst[i].rgb_addr);
275
276 OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_INST_0, code->alu.length);
277 for (i = 0; i < code->alu.length; i++)
278 OUT_CS(code->alu.inst[i].alpha_inst);
279
280 OUT_CS_REG_SEQ(R300_US_ALU_ALPHA_ADDR_0, code->alu.length);
281 for (i = 0; i < code->alu.length; i++)
282 OUT_CS(code->alu.inst[i].alpha_addr);
283
284 if (code->tex.length) {
285 OUT_CS_REG_SEQ(R300_US_TEX_INST_0, code->tex.length);
286 for(i = 0; i < code->tex.length; ++i)
287 OUT_CS(code->tex.inst[i]);
288 }
289
290 END_CS;
291 }
292
293 void r300_emit_fs_constant_buffer(struct r300_context* r300,
294 struct rc_constant_list* constants)
295 {
296 int i;
297 CS_LOCALS(r300);
298
299 if (constants->Count == 0)
300 return;
301
302 BEGIN_CS(constants->Count * 4 + 1);
303 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, constants->Count * 4);
304 for(i = 0; i < constants->Count; ++i) {
305 const float * data = get_shader_constant(r300,
306 &constants->Constants[i],
307 &r300->shader_constants[PIPE_SHADER_FRAGMENT]);
308 OUT_CS(pack_float24(data[0]));
309 OUT_CS(pack_float24(data[1]));
310 OUT_CS(pack_float24(data[2]));
311 OUT_CS(pack_float24(data[3]));
312 }
313 END_CS;
314 }
315
316 static void r300_emit_fragment_depth_config(struct r300_context* r300,
317 struct r300_fragment_shader* fs)
318 {
319 CS_LOCALS(r300);
320
321 BEGIN_CS(4);
322 if (r300_fragment_shader_writes_depth(fs)) {
323 OUT_CS_REG(R300_FG_DEPTH_SRC, R300_FG_DEPTH_SRC_SHADER);
324 OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W24 | R300_W_SRC_US);
325 } else {
326 OUT_CS_REG(R300_FG_DEPTH_SRC, R300_FG_DEPTH_SRC_SCAN);
327 OUT_CS_REG(R300_US_W_FMT, R300_W_FMT_W0 | R300_W_SRC_US);
328 }
329 END_CS;
330 }
331
332 void r500_emit_fragment_program_code(struct r300_context* r300,
333 struct rX00_fragment_program_code* generic_code)
334 {
335 struct r500_fragment_program_code * code = &generic_code->code.r500;
336 int i;
337 CS_LOCALS(r300);
338
339 BEGIN_CS(13 +
340 ((code->inst_end + 1) * 6));
341 OUT_CS_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
342 OUT_CS_REG(R500_US_PIXSIZE, code->max_temp_idx);
343 OUT_CS_REG(R500_US_CODE_RANGE,
344 R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
345 OUT_CS_REG(R500_US_CODE_OFFSET, 0);
346 OUT_CS_REG(R500_US_CODE_ADDR,
347 R500_US_CODE_START_ADDR(0) | R500_US_CODE_END_ADDR(code->inst_end));
348
349 OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_INSTR);
350 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, (code->inst_end + 1) * 6);
351 for (i = 0; i <= code->inst_end; i++) {
352 OUT_CS(code->inst[i].inst0);
353 OUT_CS(code->inst[i].inst1);
354 OUT_CS(code->inst[i].inst2);
355 OUT_CS(code->inst[i].inst3);
356 OUT_CS(code->inst[i].inst4);
357 OUT_CS(code->inst[i].inst5);
358 }
359
360 END_CS;
361 }
362
363 void r500_emit_fs_constant_buffer(struct r300_context* r300,
364 struct rc_constant_list* constants)
365 {
366 int i;
367 CS_LOCALS(r300);
368
369 if (constants->Count == 0)
370 return;
371
372 BEGIN_CS(constants->Count * 4 + 3);
373 OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
374 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, constants->Count * 4);
375 for (i = 0; i < constants->Count; i++) {
376 const float * data = get_shader_constant(r300,
377 &constants->Constants[i],
378 &r300->shader_constants[PIPE_SHADER_FRAGMENT]);
379 OUT_CS_32F(data[0]);
380 OUT_CS_32F(data[1]);
381 OUT_CS_32F(data[2]);
382 OUT_CS_32F(data[3]);
383 }
384 END_CS;
385 }
386
387 void r300_emit_fb_state(struct r300_context* r300,
388 struct pipe_framebuffer_state* fb)
389 {
390 struct r300_texture* tex;
391 struct pipe_surface* surf;
392 int i;
393 CS_LOCALS(r300);
394
395 /* Shouldn't fail unless there is a bug in the state tracker. */
396 assert(fb->nr_cbufs <= 4);
397
398 BEGIN_CS((10 * fb->nr_cbufs) + (2 * (4 - fb->nr_cbufs)) +
399 (fb->zsbuf ? 10 : 0) + 6);
400
401 /* Flush and free renderbuffer caches. */
402 OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT,
403 R300_RB3D_DSTCACHE_CTLSTAT_DC_FREE_FREE_3D_TAGS |
404 R300_RB3D_DSTCACHE_CTLSTAT_DC_FLUSH_FLUSH_DIRTY_3D);
405 OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT,
406 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
407 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
408
409 /* Set the number of colorbuffers. */
410 OUT_CS_REG(R300_RB3D_CCTL, R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs));
411
412 /* Set up colorbuffers. */
413 for (i = 0; i < fb->nr_cbufs; i++) {
414 surf = fb->cbufs[i];
415 tex = (struct r300_texture*)surf->texture;
416 assert(tex && tex->buffer && "cbuf is marked, but NULL!");
417
418 OUT_CS_REG_SEQ(R300_RB3D_COLOROFFSET0 + (4 * i), 1);
419 OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
420
421 OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1);
422 OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
423 r300_translate_colorformat(tex->tex.format) |
424 R300_COLOR_TILE(tex->macrotile) |
425 R300_COLOR_MICROTILE(tex->microtile),
426 0, RADEON_GEM_DOMAIN_VRAM, 0);
427
428 OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i),
429 r300_translate_out_fmt(surf->format));
430 }
431
432 /* Disable unused colorbuffers. */
433 for (; i < 4; i++) {
434 OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED);
435 }
436
437 /* Set up a zbuffer. */
438 if (fb->zsbuf) {
439 surf = fb->zsbuf;
440 tex = (struct r300_texture*)surf->texture;
441 assert(tex && tex->buffer && "zsbuf is marked, but NULL!");
442
443 OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1);
444 OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
445
446 OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format));
447
448 OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
449 OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
450 R300_DEPTHMACROTILE(tex->macrotile) |
451 R300_DEPTHMICROTILE(tex->microtile),
452 0, RADEON_GEM_DOMAIN_VRAM, 0);
453 }
454
455 END_CS;
456 }
457
458 static void r300_emit_query_start(struct r300_context *r300)
459 {
460 struct r300_capabilities *caps = r300_screen(r300->context.screen)->caps;
461 struct r300_query *query = r300->query_current;
462 CS_LOCALS(r300);
463
464 if (!query)
465 return;
466
467 BEGIN_CS(4);
468 if (caps->family == CHIP_FAMILY_RV530) {
469 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
470 } else {
471 OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL);
472 }
473 OUT_CS_REG(R300_ZB_ZPASS_DATA, 0);
474 END_CS;
475 query->begin_emitted = TRUE;
476 }
477
478
479 static void r300_emit_query_finish(struct r300_context *r300,
480 struct r300_query *query)
481 {
482 struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
483 CS_LOCALS(r300);
484
485 assert(caps->num_frag_pipes);
486
487 BEGIN_CS(6 * caps->num_frag_pipes + 2);
488 /* I'm not so sure I like this switch, but it's hard to be elegant
489 * when there's so many special cases...
490 *
491 * So here's the basic idea. For each pipe, enable writes to it only,
492 * then put out the relocation for ZPASS_ADDR, taking into account a
493 * 4-byte offset for each pipe. RV380 and older are special; they have
494 * only two pipes, and the second pipe's enable is on bit 3, not bit 1,
495 * so there's a chipset cap for that. */
496 switch (caps->num_frag_pipes) {
497 case 4:
498 /* pipe 3 only */
499 OUT_CS_REG(R300_SU_REG_DEST, 1 << 3);
500 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
501 OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 3),
502 0, RADEON_GEM_DOMAIN_GTT, 0);
503 case 3:
504 /* pipe 2 only */
505 OUT_CS_REG(R300_SU_REG_DEST, 1 << 2);
506 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
507 OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 2),
508 0, RADEON_GEM_DOMAIN_GTT, 0);
509 case 2:
510 /* pipe 1 only */
511 /* As mentioned above, accomodate RV380 and older. */
512 OUT_CS_REG(R300_SU_REG_DEST,
513 1 << (caps->high_second_pipe ? 3 : 1));
514 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
515 OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 1),
516 0, RADEON_GEM_DOMAIN_GTT, 0);
517 case 1:
518 /* pipe 0 only */
519 OUT_CS_REG(R300_SU_REG_DEST, 1 << 0);
520 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
521 OUT_CS_RELOC(r300->oqbo, query->offset + (sizeof(uint32_t) * 0),
522 0, RADEON_GEM_DOMAIN_GTT, 0);
523 break;
524 default:
525 debug_printf("r300: Implementation error: Chipset reports %d"
526 " pixel pipes!\n", caps->num_frag_pipes);
527 assert(0);
528 }
529
530 /* And, finally, reset it to normal... */
531 OUT_CS_REG(R300_SU_REG_DEST, 0xF);
532 END_CS;
533 }
534
535 static void rv530_emit_query_single(struct r300_context *r300,
536 struct r300_query *query)
537 {
538 CS_LOCALS(r300);
539
540 BEGIN_CS(8);
541 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
542 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
543 OUT_CS_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0);
544 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
545 END_CS;
546 }
547
548 static void rv530_emit_query_double(struct r300_context *r300,
549 struct r300_query *query)
550 {
551 CS_LOCALS(r300);
552
553 BEGIN_CS(14);
554 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
555 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
556 OUT_CS_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0);
557 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1);
558 OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
559 OUT_CS_RELOC(r300->oqbo, query->offset + sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0);
560 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
561 END_CS;
562 }
563
564 void r300_emit_query_end(struct r300_context* r300)
565 {
566 struct r300_capabilities *caps = r300_screen(r300->context.screen)->caps;
567 struct r300_query *query = r300->query_current;
568
569 if (!query)
570 return;
571
572 if (query->begin_emitted == FALSE)
573 return;
574
575 if (caps->family == CHIP_FAMILY_RV530) {
576 if (caps->num_z_pipes == 2)
577 rv530_emit_query_double(r300, query);
578 else
579 rv530_emit_query_single(r300, query);
580 } else
581 r300_emit_query_finish(r300, query);
582 }
583
584 void r300_emit_rs_state(struct r300_context* r300, void* state)
585 {
586 struct r300_rs_state* rs = (struct r300_rs_state*)state;
587 float scale, offset;
588 CS_LOCALS(r300);
589
590 BEGIN_CS(20 + (rs->polygon_offset_enable ? 5 : 0));
591 OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status);
592
593 OUT_CS_REG(R300_GB_AA_CONFIG, rs->antialiasing_config);
594
595 OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size);
596 OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX, 2);
597 OUT_CS(rs->point_minmax);
598 OUT_CS(rs->line_control);
599
600 if (rs->polygon_offset_enable) {
601 scale = rs->depth_scale * 12;
602 offset = rs->depth_offset;
603
604 switch (r300->zbuffer_bpp) {
605 case 16:
606 offset *= 4;
607 break;
608 case 24:
609 offset *= 2;
610 break;
611 }
612
613 OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_FRONT_SCALE, 4);
614 OUT_CS_32F(scale);
615 OUT_CS_32F(offset);
616 OUT_CS_32F(scale);
617 OUT_CS_32F(offset);
618 }
619
620 OUT_CS_REG_SEQ(R300_SU_POLY_OFFSET_ENABLE, 2);
621 OUT_CS(rs->polygon_offset_enable);
622 OUT_CS(rs->cull_mode);
623 OUT_CS_REG(R300_GA_LINE_STIPPLE_CONFIG, rs->line_stipple_config);
624 OUT_CS_REG(R300_GA_LINE_STIPPLE_VALUE, rs->line_stipple_value);
625 OUT_CS_REG(R300_GA_COLOR_CONTROL, rs->color_control);
626 OUT_CS_REG(R300_GA_POLY_MODE, rs->polygon_mode);
627 END_CS;
628 }
629
630 void r300_emit_rs_block_state(struct r300_context* r300,
631 struct r300_rs_block* rs)
632 {
633 int i;
634 struct r300_screen* r300screen = r300_screen(r300->context.screen);
635 CS_LOCALS(r300);
636
637 DBG(r300, DBG_DRAW, "r300: RS emit:\n");
638
639 BEGIN_CS(21);
640 if (r300screen->caps->is_r500) {
641 OUT_CS_REG_SEQ(R500_RS_IP_0, 8);
642 } else {
643 OUT_CS_REG_SEQ(R300_RS_IP_0, 8);
644 }
645 for (i = 0; i < 8; i++) {
646 OUT_CS(rs->ip[i]);
647 DBG(r300, DBG_DRAW, " : ip %d: 0x%08x\n", i, rs->ip[i]);
648 }
649
650 OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
651 OUT_CS(rs->count);
652 OUT_CS(rs->inst_count);
653
654 if (r300screen->caps->is_r500) {
655 OUT_CS_REG_SEQ(R500_RS_INST_0, 8);
656 } else {
657 OUT_CS_REG_SEQ(R300_RS_INST_0, 8);
658 }
659 for (i = 0; i < 8; i++) {
660 OUT_CS(rs->inst[i]);
661 DBG(r300, DBG_DRAW, " : inst %d: 0x%08x\n", i, rs->inst[i]);
662 }
663
664 DBG(r300, DBG_DRAW, " : count: 0x%08x inst_count: 0x%08x\n",
665 rs->count, rs->inst_count);
666
667 END_CS;
668 }
669
670 void r300_emit_scissor_state(struct r300_context* r300, void* state)
671 {
672 unsigned minx, miny, maxx, maxy;
673 uint32_t top_left, bottom_right;
674 struct r300_screen* r300screen = r300_screen(r300->context.screen);
675 struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state;
676 CS_LOCALS(r300);
677
678 minx = miny = 0;
679 maxx = r300->framebuffer_state.width;
680 maxy = r300->framebuffer_state.height;
681
682 if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) {
683 minx = MAX2(minx, scissor->minx);
684 miny = MAX2(miny, scissor->miny);
685 maxx = MIN2(maxx, scissor->maxx);
686 maxy = MIN2(maxy, scissor->maxy);
687 }
688
689 /* Special case for zero-area scissor.
690 *
691 * We can't allow the variables maxx and maxy to be zero because they are
692 * subtracted from later in the code, which would cause emitting ~0 and
693 * making the kernel checker angry.
694 *
695 * Let's consider we change maxx and maxy to 1, which is effectively
696 * a one-pixel area. We must then change minx and miny to a number which is
697 * greater than 1 to get the zero area back. */
698 if (!maxx || !maxy) {
699 minx = 2;
700 miny = 2;
701 maxx = 1;
702 maxy = 1;
703 }
704
705 if (r300screen->caps->is_r500) {
706 top_left =
707 (minx << R300_SCISSORS_X_SHIFT) |
708 (miny << R300_SCISSORS_Y_SHIFT);
709 bottom_right =
710 ((maxx - 1) << R300_SCISSORS_X_SHIFT) |
711 ((maxy - 1) << R300_SCISSORS_Y_SHIFT);
712 } else {
713 /* Offset of 1440 in non-R500 chipsets. */
714 top_left =
715 ((minx + 1440) << R300_SCISSORS_X_SHIFT) |
716 ((miny + 1440) << R300_SCISSORS_Y_SHIFT);
717 bottom_right =
718 (((maxx - 1) + 1440) << R300_SCISSORS_X_SHIFT) |
719 (((maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
720 }
721
722 BEGIN_CS(3);
723 OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
724 OUT_CS(top_left);
725 OUT_CS(bottom_right);
726 END_CS;
727 }
728
729 void r300_emit_texture(struct r300_context* r300,
730 struct r300_sampler_state* sampler,
731 struct r300_texture* tex,
732 unsigned offset)
733 {
734 uint32_t filter0 = sampler->filter0;
735 uint32_t format0 = tex->state.format0;
736 unsigned min_level, max_level;
737 CS_LOCALS(r300);
738
739 /* to emulate 1D textures through 2D ones correctly */
740 if (tex->tex.target == PIPE_TEXTURE_1D) {
741 filter0 &= ~R300_TX_WRAP_T_MASK;
742 filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
743 }
744
745 if (tex->is_npot) {
746 /* NPOT textures don't support mip filter, unfortunately.
747 * This prevents incorrect rendering. */
748 filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK;
749 } else {
750 /* determine min/max levels */
751 /* the MAX_MIP level is the largest (finest) one */
752 max_level = MIN2(sampler->max_lod, tex->tex.last_level);
753 min_level = MIN2(sampler->min_lod, max_level);
754 format0 |= R300_TX_NUM_LEVELS(max_level);
755 filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
756 }
757
758 BEGIN_CS(16);
759 OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 |
760 (offset << 28));
761 OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1);
762 OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color);
763
764 OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0);
765 OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1);
766 OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2);
767 OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1);
768 OUT_CS_RELOC(tex->buffer,
769 R300_TXO_MACRO_TILE(tex->macrotile) |
770 R300_TXO_MICRO_TILE(tex->microtile),
771 RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0);
772 END_CS;
773 }
774
775 static boolean r300_validate_aos(struct r300_context *r300)
776 {
777 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
778 struct pipe_vertex_element *velem = r300->vertex_element;
779 int i;
780
781 /* Check if formats and strides are aligned to the size of DWORD. */
782 for (i = 0; i < r300->vertex_element_count; i++) {
783 if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
784 util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
785 return FALSE;
786 }
787 }
788 return TRUE;
789 }
790
791 void r300_emit_aos(struct r300_context* r300, unsigned offset)
792 {
793 struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
794 struct pipe_vertex_element *velem = r300->vertex_element;
795 int i;
796 unsigned size1, size2, aos_count = r300->vertex_element_count;
797 unsigned packet_size = (aos_count * 3 + 1) / 2;
798 CS_LOCALS(r300);
799
800 /* XXX Move this checking to a more approriate place. */
801 if (!r300_validate_aos(r300)) {
802 /* XXX We should fallback using Draw. */
803 assert(0);
804 }
805
806 BEGIN_CS(2 + packet_size + aos_count * 2);
807 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
808 OUT_CS(aos_count);
809
810 for (i = 0; i < aos_count - 1; i += 2) {
811 vb1 = &vbuf[velem[i].vertex_buffer_index];
812 vb2 = &vbuf[velem[i+1].vertex_buffer_index];
813 size1 = util_format_get_blocksize(velem[i].src_format);
814 size2 = util_format_get_blocksize(velem[i+1].src_format);
815
816 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
817 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
818 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
819 OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
820 }
821
822 if (aos_count & 1) {
823 vb1 = &vbuf[velem[i].vertex_buffer_index];
824 size1 = util_format_get_blocksize(velem[i].src_format);
825
826 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
827 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
828 }
829
830 for (i = 0; i < aos_count; i++) {
831 OUT_CS_RELOC_NO_OFFSET(vbuf[velem[i].vertex_buffer_index].buffer,
832 RADEON_GEM_DOMAIN_GTT, 0, 0);
833 }
834 END_CS;
835 }
836
837 void r300_emit_vertex_format_state(struct r300_context* r300)
838 {
839 int i;
840 CS_LOCALS(r300);
841
842 DBG(r300, DBG_DRAW, "r300: VAP/PSC emit:\n");
843
844 BEGIN_CS(26);
845 OUT_CS_REG(R300_VAP_VTX_SIZE, r300->vertex_info->vinfo.size);
846
847 OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
848 OUT_CS(r300->vertex_info->vinfo.hwfmt[0]);
849 OUT_CS(r300->vertex_info->vinfo.hwfmt[1]);
850 OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
851 OUT_CS(r300->vertex_info->vinfo.hwfmt[2]);
852 OUT_CS(r300->vertex_info->vinfo.hwfmt[3]);
853 for (i = 0; i < 4; i++) {
854 DBG(r300, DBG_DRAW, " : hwfmt%d: 0x%08x\n", i,
855 r300->vertex_info->vinfo.hwfmt[i]);
856 }
857
858 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8);
859 for (i = 0; i < 8; i++) {
860 OUT_CS(r300->vertex_info->vap_prog_stream_cntl[i]);
861 DBG(r300, DBG_DRAW, " : prog_stream_cntl%d: 0x%08x\n", i,
862 r300->vertex_info->vap_prog_stream_cntl[i]);
863 }
864 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8);
865 for (i = 0; i < 8; i++) {
866 OUT_CS(r300->vertex_info->vap_prog_stream_cntl_ext[i]);
867 DBG(r300, DBG_DRAW, " : prog_stream_cntl_ext%d: 0x%08x\n", i,
868 r300->vertex_info->vap_prog_stream_cntl_ext[i]);
869 }
870 END_CS;
871 }
872
873
874 void r300_emit_vertex_program_code(struct r300_context* r300,
875 struct r300_vertex_program_code* code)
876 {
877 int i;
878 struct r300_screen* r300screen = r300_screen(r300->context.screen);
879 unsigned instruction_count = code->length / 4;
880
881 int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72;
882 int input_count = MAX2(util_bitcount(code->InputsRead), 1);
883 int output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
884 int temp_count = MAX2(code->num_temporaries, 1);
885 int pvs_num_slots = MIN3(vtx_mem_size / input_count,
886 vtx_mem_size / output_count, 10);
887 int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6);
888
889 CS_LOCALS(r300);
890
891 if (!r300screen->caps->has_tcl) {
892 debug_printf("r300: Implementation error: emit_vertex_shader called,"
893 " but has_tcl is FALSE!\n");
894 return;
895 }
896
897 BEGIN_CS(9 + code->length);
898 /* R300_VAP_PVS_CODE_CNTL_0
899 * R300_VAP_PVS_CONST_CNTL
900 * R300_VAP_PVS_CODE_CNTL_1
901 * See the r5xx docs for instructions on how to use these. */
902 OUT_CS_REG_SEQ(R300_VAP_PVS_CODE_CNTL_0, 3);
903 OUT_CS(R300_PVS_FIRST_INST(0) |
904 R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
905 R300_PVS_LAST_INST(instruction_count - 1));
906 OUT_CS(R300_PVS_MAX_CONST_ADDR(code->constants.Count - 1));
907 OUT_CS(instruction_count - 1);
908
909 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
910 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
911 for (i = 0; i < code->length; i++)
912 OUT_CS(code->body.d[i]);
913
914 OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
915 R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
916 R300_PVS_NUM_FPUS(r300screen->caps->num_vert_fpus) |
917 R300_PVS_VF_MAX_VTX_NUM(12) |
918 (r300screen->caps->is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
919 END_CS;
920 }
921
922 void r300_emit_vertex_shader(struct r300_context* r300,
923 struct r300_vertex_shader* vs)
924 {
925 r300_emit_vertex_program_code(r300, &vs->code);
926 }
927
928 void r300_emit_vs_constant_buffer(struct r300_context* r300,
929 struct rc_constant_list* constants)
930 {
931 int i;
932 struct r300_screen* r300screen = r300_screen(r300->context.screen);
933 CS_LOCALS(r300);
934
935 if (!r300screen->caps->has_tcl) {
936 debug_printf("r300: Implementation error: emit_vertex_shader called,"
937 " but has_tcl is FALSE!\n");
938 return;
939 }
940
941 if (constants->Count == 0)
942 return;
943
944 BEGIN_CS(constants->Count * 4 + 3);
945 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
946 (r300screen->caps->is_r500 ?
947 R500_PVS_CONST_START : R300_PVS_CONST_START));
948 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, constants->Count * 4);
949 for (i = 0; i < constants->Count; i++) {
950 const float * data = get_shader_constant(r300,
951 &constants->Constants[i],
952 &r300->shader_constants[PIPE_SHADER_VERTEX]);
953 OUT_CS_32F(data[0]);
954 OUT_CS_32F(data[1]);
955 OUT_CS_32F(data[2]);
956 OUT_CS_32F(data[3]);
957 }
958 END_CS;
959 }
960
961 void r300_emit_viewport_state(struct r300_context* r300, void* state)
962 {
963 struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
964 CS_LOCALS(r300);
965
966 if (r300->tcl_bypass) {
967 BEGIN_CS(2);
968 OUT_CS_REG(R300_VAP_VTE_CNTL, 0);
969 END_CS;
970 } else {
971 BEGIN_CS(9);
972 OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
973 OUT_CS_32F(viewport->xscale);
974 OUT_CS_32F(viewport->xoffset);
975 OUT_CS_32F(viewport->yscale);
976 OUT_CS_32F(viewport->yoffset);
977 OUT_CS_32F(viewport->zscale);
978 OUT_CS_32F(viewport->zoffset);
979 OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
980 END_CS;
981 }
982 }
983
984 void r300_emit_texture_count(struct r300_context* r300)
985 {
986 uint32_t tx_enable = 0;
987 int i;
988 CS_LOCALS(r300);
989
990 /* Notice that texture_count and sampler_count are just sizes
991 * of the respective arrays. We still have to check for the individual
992 * elements. */
993 for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
994 if (r300->textures[i]) {
995 tx_enable |= 1 << i;
996 }
997 }
998
999 BEGIN_CS(2);
1000 OUT_CS_REG(R300_TX_ENABLE, tx_enable);
1001 END_CS;
1002
1003 }
1004
1005 void r300_emit_ztop_state(struct r300_context* r300, void* state)
1006 {
1007 struct r300_ztop_state* ztop = (struct r300_ztop_state*)state;
1008 CS_LOCALS(r300);
1009
1010 BEGIN_CS(2);
1011 OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top);
1012 END_CS;
1013 }
1014
1015 void r300_flush_textures(struct r300_context* r300)
1016 {
1017 CS_LOCALS(r300);
1018
1019 BEGIN_CS(2);
1020 OUT_CS_REG(R300_TX_INVALTAGS, 0);
1021 END_CS;
1022 }
1023
1024 static void r300_flush_pvs(struct r300_context* r300)
1025 {
1026 CS_LOCALS(r300);
1027
1028 BEGIN_CS(2);
1029 OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
1030 END_CS;
1031 }
1032
1033 void r300_emit_buffer_validate(struct r300_context *r300)
1034 {
1035 struct r300_texture* tex;
1036 unsigned i;
1037 boolean invalid = FALSE;
1038
1039 /* Clean out BOs. */
1040 r300->winsys->reset_bos(r300->winsys);
1041
1042 validate:
1043 /* Color buffers... */
1044 for (i = 0; i < r300->framebuffer_state.nr_cbufs; i++) {
1045 tex = (struct r300_texture*)r300->framebuffer_state.cbufs[i]->texture;
1046 assert(tex && tex->buffer && "cbuf is marked, but NULL!");
1047 if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
1048 0, RADEON_GEM_DOMAIN_VRAM)) {
1049 r300->context.flush(&r300->context, 0, NULL);
1050 goto validate;
1051 }
1052 }
1053 /* ...depth buffer... */
1054 if (r300->framebuffer_state.zsbuf) {
1055 tex = (struct r300_texture*)r300->framebuffer_state.zsbuf->texture;
1056 assert(tex && tex->buffer && "zsbuf is marked, but NULL!");
1057 if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
1058 0, RADEON_GEM_DOMAIN_VRAM)) {
1059 r300->context.flush(&r300->context, 0, NULL);
1060 goto validate;
1061 }
1062 }
1063 /* ...textures... */
1064 for (i = 0; i < r300->texture_count; i++) {
1065 tex = r300->textures[i];
1066 if (!tex)
1067 continue;
1068 if (!r300->winsys->add_buffer(r300->winsys, tex->buffer,
1069 RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0)) {
1070 r300->context.flush(&r300->context, 0, NULL);
1071 goto validate;
1072 }
1073 }
1074 /* ...occlusion query buffer... */
1075 if (r300->dirty_state & R300_NEW_QUERY) {
1076 if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo,
1077 0, RADEON_GEM_DOMAIN_GTT)) {
1078 r300->context.flush(&r300->context, 0, NULL);
1079 goto validate;
1080 }
1081 }
1082 /* ...and vertex buffer. */
1083 if (r300->vbo) {
1084 if (!r300->winsys->add_buffer(r300->winsys, r300->vbo,
1085 RADEON_GEM_DOMAIN_GTT, 0)) {
1086 r300->context.flush(&r300->context, 0, NULL);
1087 goto validate;
1088 }
1089 } else {
1090 /* debug_printf("No VBO while emitting dirty state!\n"); */
1091 }
1092 if (!r300->winsys->validate(r300->winsys)) {
1093 r300->context.flush(&r300->context, 0, NULL);
1094 if (invalid) {
1095 /* Well, hell. */
1096 debug_printf("r300: Stuck in validation loop, gonna quit now.");
1097 exit(1);
1098 }
1099 invalid = TRUE;
1100 goto validate;
1101 }
1102 }
1103
1104 /* Emit all dirty state. */
1105 void r300_emit_dirty_state(struct r300_context* r300)
1106 {
1107 struct r300_screen* r300screen = r300_screen(r300->context.screen);
1108 struct r300_atom* atom;
1109 unsigned i, dwords = 1024;
1110 int dirty_tex = 0;
1111
1112 /* Check the required number of dwords against the space remaining in the
1113 * current CS object. If we need more, then flush. */
1114
1115 foreach(atom, &r300->atom_list) {
1116 if (atom->dirty || atom->always_dirty) {
1117 dwords += atom->size;
1118 }
1119 }
1120
1121 /* Make sure we have at least 2*1024 spare dwords. */
1122 /* XXX It would be nice to know the number of dwords we really need to
1123 * XXX emit. */
1124 while (!r300->winsys->check_cs(r300->winsys, dwords)) {
1125 r300->context.flush(&r300->context, 0, NULL);
1126 }
1127
1128 if (r300->dirty_state & R300_NEW_QUERY) {
1129 r300_emit_query_start(r300);
1130 r300->dirty_state &= ~R300_NEW_QUERY;
1131 }
1132
1133 foreach(atom, &r300->atom_list) {
1134 if (atom->dirty || atom->always_dirty) {
1135 atom->emit(r300, atom->state);
1136 atom->dirty = FALSE;
1137 }
1138 }
1139
1140 if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) {
1141 r300_emit_fragment_depth_config(r300, r300->fs);
1142 if (r300screen->caps->is_r500) {
1143 r500_emit_fragment_program_code(r300, &r300->fs->shader->code);
1144 } else {
1145 r300_emit_fragment_program_code(r300, &r300->fs->shader->code);
1146 }
1147 r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER;
1148 }
1149
1150 if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER_CONSTANTS) {
1151 if (r300screen->caps->is_r500) {
1152 r500_emit_fs_constant_buffer(r300,
1153 &r300->fs->shader->code.constants);
1154 } else {
1155 r300_emit_fs_constant_buffer(r300,
1156 &r300->fs->shader->code.constants);
1157 }
1158 r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS;
1159 }
1160
1161 if (r300->dirty_state & R300_NEW_FRAMEBUFFERS) {
1162 r300_emit_fb_state(r300, &r300->framebuffer_state);
1163 r300->dirty_state &= ~R300_NEW_FRAMEBUFFERS;
1164 }
1165
1166 if (r300->dirty_state & R300_NEW_RS_BLOCK) {
1167 r300_emit_rs_block_state(r300, r300->rs_block);
1168 r300->dirty_state &= ~R300_NEW_RS_BLOCK;
1169 }
1170
1171 /* Samplers and textures are tracked separately but emitted together. */
1172 if (r300->dirty_state &
1173 (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
1174 r300_emit_texture_count(r300);
1175
1176 for (i = 0; i < MIN2(r300->sampler_count, r300->texture_count); i++) {
1177 if (r300->dirty_state &
1178 ((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) {
1179 if (r300->textures[i])
1180 r300_emit_texture(r300,
1181 r300->sampler_states[i],
1182 r300->textures[i],
1183 i);
1184 r300->dirty_state &=
1185 ~((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i));
1186 dirty_tex++;
1187 }
1188 }
1189 r300->dirty_state &= ~(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES);
1190 }
1191
1192 if (dirty_tex) {
1193 r300_flush_textures(r300);
1194 }
1195
1196 if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) {
1197 r300_emit_vertex_format_state(r300);
1198 r300->dirty_state &= ~R300_NEW_VERTEX_FORMAT;
1199 }
1200
1201 if (r300->dirty_state & (R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS)) {
1202 r300_flush_pvs(r300);
1203 }
1204
1205 if (r300->dirty_state & R300_NEW_VERTEX_SHADER) {
1206 r300_emit_vertex_shader(r300, r300->vs);
1207 r300->dirty_state &= ~R300_NEW_VERTEX_SHADER;
1208 }
1209
1210 if (r300->dirty_state & R300_NEW_VERTEX_SHADER_CONSTANTS) {
1211 r300_emit_vs_constant_buffer(r300, &r300->vs->code.constants);
1212 r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS;
1213 }
1214
1215 /* XXX
1216 assert(r300->dirty_state == 0);
1217 */
1218
1219 /* Finally, emit the VBO. */
1220 /* r300_emit_vertex_buffer(r300); */
1221
1222 r300->dirty_hw++;
1223 }