r300g: clear the ZB cache before clearing ZMASK or HIZ
[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_mm.h"
29
30 #include "r300_context.h"
31 #include "r300_cb.h"
32 #include "r300_cs.h"
33 #include "r300_emit.h"
34 #include "r300_fs.h"
35 #include "r300_screen.h"
36 #include "r300_screen_buffer.h"
37 #include "r300_vs.h"
38
39 void r300_emit_blend_state(struct r300_context* r300,
40 unsigned size, void* state)
41 {
42 struct r300_blend_state* blend = (struct r300_blend_state*)state;
43 struct pipe_framebuffer_state* fb =
44 (struct pipe_framebuffer_state*)r300->fb_state.state;
45 CS_LOCALS(r300);
46
47 if (fb->nr_cbufs) {
48 if (fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT) {
49 WRITE_CS_TABLE(blend->cb_noclamp, size);
50 } else {
51 unsigned swz = r300_surface(fb->cbufs[0])->colormask_swizzle;
52 WRITE_CS_TABLE(blend->cb_clamp[swz], size);
53 }
54 } else {
55 WRITE_CS_TABLE(blend->cb_no_readwrite, size);
56 }
57 }
58
59 void r300_emit_blend_color_state(struct r300_context* r300,
60 unsigned size, void* state)
61 {
62 struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state;
63 CS_LOCALS(r300);
64
65 WRITE_CS_TABLE(bc->cb, size);
66 }
67
68 void r300_emit_clip_state(struct r300_context* r300,
69 unsigned size, void* state)
70 {
71 struct r300_clip_state* clip = (struct r300_clip_state*)state;
72 CS_LOCALS(r300);
73
74 WRITE_CS_TABLE(clip->cb, size);
75 }
76
77 void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state)
78 {
79 struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
80 struct pipe_framebuffer_state* fb =
81 (struct pipe_framebuffer_state*)r300->fb_state.state;
82 CS_LOCALS(r300);
83
84 if (fb->zsbuf) {
85 if (fb->nr_cbufs && fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT)
86 WRITE_CS_TABLE(&dsa->cb_begin_fp16, size);
87 else
88 WRITE_CS_TABLE(&dsa->cb_begin, size);
89 } else {
90 if (fb->nr_cbufs && fb->cbufs[0]->format == PIPE_FORMAT_R16G16B16A16_FLOAT)
91 WRITE_CS_TABLE(dsa->cb_fp16_zb_no_readwrite, size);
92 else
93 WRITE_CS_TABLE(dsa->cb_zb_no_readwrite, size);
94 }
95 }
96
97 static void get_rc_constant_state(
98 float vec[4],
99 struct r300_context * r300,
100 struct rc_constant * constant)
101 {
102 struct r300_textures_state* texstate = r300->textures_state.state;
103 struct r300_resource *tex;
104
105 assert(constant->Type == RC_CONSTANT_STATE);
106
107 /* vec should either be (0, 0, 0, 1), which should be a relatively safe
108 * RGBA or STRQ value, or it could be one of the RC_CONSTANT_STATE
109 * state factors. */
110
111 switch (constant->u.State[0]) {
112 /* Factor for converting rectangle coords to
113 * normalized coords. Should only show up on non-r500. */
114 case RC_STATE_R300_TEXRECT_FACTOR:
115 tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture);
116 vec[0] = 1.0 / tex->tex.width0;
117 vec[1] = 1.0 / tex->tex.height0;
118 vec[2] = 0;
119 vec[3] = 1;
120 break;
121
122 case RC_STATE_R300_TEXSCALE_FACTOR:
123 tex = r300_resource(texstate->sampler_views[constant->u.State[1]]->base.texture);
124 /* Add a small number to the texture size to work around rounding errors in hw. */
125 vec[0] = tex->b.b.width0 / (tex->tex.width0 + 0.001f);
126 vec[1] = tex->b.b.height0 / (tex->tex.height0 + 0.001f);
127 vec[2] = tex->b.b.depth0 / (tex->tex.depth0 + 0.001f);
128 vec[3] = 1;
129 break;
130
131 case RC_STATE_R300_VIEWPORT_SCALE:
132 vec[0] = r300->viewport.scale[0];
133 vec[1] = r300->viewport.scale[1];
134 vec[2] = r300->viewport.scale[2];
135 vec[3] = 1;
136 break;
137
138 case RC_STATE_R300_VIEWPORT_OFFSET:
139 vec[0] = r300->viewport.translate[0];
140 vec[1] = r300->viewport.translate[1];
141 vec[2] = r300->viewport.translate[2];
142 vec[3] = 1;
143 break;
144
145 default:
146 fprintf(stderr, "r300: Implementation error: "
147 "Unknown RC_CONSTANT type %d\n", constant->u.State[0]);
148 vec[0] = 0;
149 vec[1] = 0;
150 vec[2] = 0;
151 vec[3] = 1;
152 }
153 }
154
155 /* Convert a normal single-precision float into the 7.16 format
156 * used by the R300 fragment shader.
157 */
158 uint32_t pack_float24(float f)
159 {
160 union {
161 float fl;
162 uint32_t u;
163 } u;
164 float mantissa;
165 int exponent;
166 uint32_t float24 = 0;
167
168 if (f == 0.0)
169 return 0;
170
171 u.fl = f;
172
173 mantissa = frexpf(f, &exponent);
174
175 /* Handle -ve */
176 if (mantissa < 0) {
177 float24 |= (1 << 23);
178 mantissa = mantissa * -1.0;
179 }
180 /* Handle exponent, bias of 63 */
181 exponent += 62;
182 float24 |= (exponent << 16);
183 /* Kill 7 LSB of mantissa */
184 float24 |= (u.u & 0x7FFFFF) >> 7;
185
186 return float24;
187 }
188
189 void r300_emit_fs(struct r300_context* r300, unsigned size, void *state)
190 {
191 struct r300_fragment_shader *fs = r300_fs(r300);
192 CS_LOCALS(r300);
193
194 WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
195 }
196
197 void r300_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
198 {
199 struct r300_fragment_shader *fs = r300_fs(r300);
200 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
201 unsigned count = fs->shader->externals_count;
202 unsigned i, j;
203 CS_LOCALS(r300);
204
205 if (count == 0)
206 return;
207
208 BEGIN_CS(size);
209 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4);
210 if (buf->remap_table){
211 for (i = 0; i < count; i++) {
212 float *data = (float*)&buf->ptr[buf->remap_table[i]*4];
213 for (j = 0; j < 4; j++)
214 OUT_CS(pack_float24(data[j]));
215 }
216 } else {
217 for (i = 0; i < count; i++)
218 for (j = 0; j < 4; j++)
219 OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j]));
220 }
221
222 END_CS;
223 }
224
225 void r300_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
226 {
227 struct r300_fragment_shader *fs = r300_fs(r300);
228 struct rc_constant_list *constants = &fs->shader->code.constants;
229 unsigned i;
230 unsigned count = fs->shader->rc_state_count;
231 unsigned first = fs->shader->externals_count;
232 unsigned end = constants->Count;
233 unsigned j;
234 CS_LOCALS(r300);
235
236 if (count == 0)
237 return;
238
239 BEGIN_CS(size);
240 for(i = first; i < end; ++i) {
241 if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
242 float data[4];
243
244 get_rc_constant_state(data, r300, &constants->Constants[i]);
245
246 OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X + i * 16, 4);
247 for (j = 0; j < 4; j++)
248 OUT_CS(pack_float24(data[j]));
249 }
250 }
251 END_CS;
252 }
253
254 void r500_emit_fs(struct r300_context* r300, unsigned size, void *state)
255 {
256 struct r300_fragment_shader *fs = r300_fs(r300);
257 CS_LOCALS(r300);
258
259 WRITE_CS_TABLE(fs->shader->cb_code, fs->shader->cb_code_size);
260 }
261
262 void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
263 {
264 struct r300_fragment_shader *fs = r300_fs(r300);
265 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
266 unsigned count = fs->shader->externals_count;
267 CS_LOCALS(r300);
268
269 if (count == 0)
270 return;
271
272 BEGIN_CS(size);
273 OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
274 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4);
275 if (buf->remap_table){
276 for (unsigned i = 0; i < count; i++) {
277 uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
278 OUT_CS_TABLE(data, 4);
279 }
280 } else {
281 OUT_CS_TABLE(buf->ptr, count * 4);
282 }
283 END_CS;
284 }
285
286 void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
287 {
288 struct r300_fragment_shader *fs = r300_fs(r300);
289 struct rc_constant_list *constants = &fs->shader->code.constants;
290 unsigned i;
291 unsigned count = fs->shader->rc_state_count;
292 unsigned first = fs->shader->externals_count;
293 unsigned end = constants->Count;
294 CS_LOCALS(r300);
295
296 if (count == 0)
297 return;
298
299 BEGIN_CS(size);
300 for(i = first; i < end; ++i) {
301 if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
302 float data[4];
303
304 get_rc_constant_state(data, r300, &constants->Constants[i]);
305
306 OUT_CS_REG(R500_GA_US_VECTOR_INDEX,
307 R500_GA_US_VECTOR_INDEX_TYPE_CONST |
308 (i & R500_GA_US_VECTOR_INDEX_MASK));
309 OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
310 OUT_CS_TABLE(data, 4);
311 }
312 }
313 END_CS;
314 }
315
316 void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state)
317 {
318 struct r300_gpu_flush *gpuflush = (struct r300_gpu_flush*)state;
319 struct pipe_framebuffer_state* fb =
320 (struct pipe_framebuffer_state*)r300->fb_state.state;
321 uint32_t height = fb->height;
322 uint32_t width = fb->width;
323 CS_LOCALS(r300);
324
325 if (r300->cbzb_clear) {
326 struct r300_surface *surf = r300_surface(fb->cbufs[0]);
327
328 height = surf->cbzb_height;
329 width = surf->cbzb_width;
330 }
331
332 DBG(r300, DBG_SCISSOR,
333 "r300: Scissor width: %i, height: %i, CBZB clear: %s\n",
334 width, height, r300->cbzb_clear ? "YES" : "NO");
335
336 BEGIN_CS(size);
337
338 /* Set up scissors.
339 * By writing to the SC registers, SC & US assert idle. */
340 OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
341 if (r300->screen->caps.is_r500) {
342 OUT_CS(0);
343 OUT_CS(((width - 1) << R300_SCISSORS_X_SHIFT) |
344 ((height - 1) << R300_SCISSORS_Y_SHIFT));
345 } else {
346 OUT_CS((1440 << R300_SCISSORS_X_SHIFT) |
347 (1440 << R300_SCISSORS_Y_SHIFT));
348 OUT_CS(((width + 1440-1) << R300_SCISSORS_X_SHIFT) |
349 ((height + 1440-1) << R300_SCISSORS_Y_SHIFT));
350 }
351
352 /* Flush CB & ZB caches and wait until the 3D engine is idle and clean. */
353 OUT_CS_TABLE(gpuflush->cb_flush_clean, 6);
354 END_CS;
355 }
356
357 void r300_emit_aa_state(struct r300_context *r300, unsigned size, void *state)
358 {
359 struct r300_aa_state *aa = (struct r300_aa_state*)state;
360 CS_LOCALS(r300);
361
362 BEGIN_CS(size);
363 OUT_CS_REG(R300_GB_AA_CONFIG, aa->aa_config);
364
365 if (aa->dest) {
366 OUT_CS_REG(R300_RB3D_AARESOLVE_OFFSET, aa->dest->offset);
367 OUT_CS_RELOC(aa->dest);
368 OUT_CS_REG(R300_RB3D_AARESOLVE_PITCH, aa->dest->pitch);
369 }
370
371 OUT_CS_REG(R300_RB3D_AARESOLVE_CTL, aa->aaresolve_ctl);
372 END_CS;
373 }
374
375 void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
376 {
377 struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
378 struct r300_surface* surf;
379 unsigned i;
380 uint32_t rb3d_cctl = 0;
381
382 CS_LOCALS(r300);
383
384 BEGIN_CS(size);
385
386 /* NUM_MULTIWRITES replicates COLOR[0] to all colorbuffers, which is not
387 * what we usually want. */
388 if (r300->screen->caps.is_r500) {
389 rb3d_cctl = R300_RB3D_CCTL_INDEPENDENT_COLORFORMAT_ENABLE_ENABLE;
390 }
391 if (fb->nr_cbufs && r300->fb_multiwrite) {
392 rb3d_cctl |= R300_RB3D_CCTL_NUM_MULTIWRITES(fb->nr_cbufs);
393 }
394
395 OUT_CS_REG(R300_RB3D_CCTL, rb3d_cctl);
396
397 /* Set up colorbuffers. */
398 for (i = 0; i < fb->nr_cbufs; i++) {
399 surf = r300_surface(fb->cbufs[i]);
400
401 OUT_CS_REG(R300_RB3D_COLOROFFSET0 + (4 * i), surf->offset);
402 OUT_CS_RELOC(surf);
403
404 OUT_CS_REG(R300_RB3D_COLORPITCH0 + (4 * i), surf->pitch);
405 OUT_CS_RELOC(surf);
406 }
407
408 /* Set up the ZB part of the CBZB clear. */
409 if (r300->cbzb_clear) {
410 surf = r300_surface(fb->cbufs[0]);
411
412 OUT_CS_REG(R300_ZB_FORMAT, surf->cbzb_format);
413
414 OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->cbzb_midpoint_offset);
415 OUT_CS_RELOC(surf);
416
417 OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->cbzb_pitch);
418 OUT_CS_RELOC(surf);
419
420 DBG(r300, DBG_CBZB,
421 "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
422 surf->cbzb_pitch);
423 }
424 /* Set up a zbuffer. */
425 else if (fb->zsbuf) {
426 surf = r300_surface(fb->zsbuf);
427
428 OUT_CS_REG(R300_ZB_FORMAT, surf->format);
429
430 OUT_CS_REG(R300_ZB_DEPTHOFFSET, surf->offset);
431 OUT_CS_RELOC(surf);
432
433 OUT_CS_REG(R300_ZB_DEPTHPITCH, surf->pitch);
434 OUT_CS_RELOC(surf);
435
436 if (r300->hyperz_enabled) {
437 /* HiZ RAM. */
438 OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
439 OUT_CS_REG(R300_ZB_HIZ_PITCH, surf->pitch_hiz);
440 /* Z Mask RAM. (compressed zbuffer) */
441 OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
442 OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf->pitch_zmask);
443 }
444 }
445
446 END_CS;
447 }
448
449 void r300_emit_hyperz_state(struct r300_context *r300,
450 unsigned size, void *state)
451 {
452 struct r300_hyperz_state *z = state;
453 CS_LOCALS(r300);
454
455 if (z->flush)
456 WRITE_CS_TABLE(&z->cb_flush_begin, size);
457 else
458 WRITE_CS_TABLE(&z->cb_begin, size - 2);
459 }
460
461 void r300_emit_hyperz_end(struct r300_context *r300)
462 {
463 struct r300_hyperz_state z =
464 *(struct r300_hyperz_state*)r300->hyperz_state.state;
465
466 z.flush = 1;
467 z.zb_bw_cntl = 0;
468 z.zb_depthclearvalue = 0;
469 z.sc_hyperz = R300_SC_HYPERZ_ADJ_2;
470 z.gb_z_peq_config = 0;
471
472 r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z);
473 }
474
475 void r300_emit_fb_state_pipelined(struct r300_context *r300,
476 unsigned size, void *state)
477 {
478 struct pipe_framebuffer_state* fb =
479 (struct pipe_framebuffer_state*)r300->fb_state.state;
480 unsigned i, num_cbufs = fb->nr_cbufs;
481 unsigned mspos0, mspos1;
482 CS_LOCALS(r300);
483
484 /* If we use the multiwrite feature, the colorbuffers 2,3,4 must be
485 * marked as UNUSED in the US block. */
486 if (r300->fb_multiwrite) {
487 num_cbufs = MIN2(num_cbufs, 1);
488 }
489
490 BEGIN_CS(size);
491
492 /* Colorbuffer format in the US block.
493 * (must be written after unpipelined regs) */
494 OUT_CS_REG_SEQ(R300_US_OUT_FMT_0, 4);
495 for (i = 0; i < num_cbufs; i++) {
496 OUT_CS(r300_surface(fb->cbufs[i])->format);
497 }
498 for (; i < 1; i++) {
499 OUT_CS(R300_US_OUT_FMT_C4_8 |
500 R300_C0_SEL_B | R300_C1_SEL_G |
501 R300_C2_SEL_R | R300_C3_SEL_A);
502 }
503 for (; i < 4; i++) {
504 OUT_CS(R300_US_OUT_FMT_UNUSED);
505 }
506
507 /* Multisampling. Depends on framebuffer sample count.
508 * These are pipelined regs and as such cannot be moved
509 * to the AA state. */
510 mspos0 = 0x66666666;
511 mspos1 = 0x6666666;
512
513 if (fb->nr_cbufs && fb->cbufs[0]->texture->nr_samples > 1) {
514 /* Subsample placement. These may not be optimal. */
515 switch (fb->cbufs[0]->texture->nr_samples) {
516 case 2:
517 mspos0 = 0x33996633;
518 mspos1 = 0x6666663;
519 break;
520 case 3:
521 mspos0 = 0x33936933;
522 mspos1 = 0x6666663;
523 break;
524 case 4:
525 mspos0 = 0x33939933;
526 mspos1 = 0x3966663;
527 break;
528 case 6:
529 mspos0 = 0x22a2aa22;
530 mspos1 = 0x2a65672;
531 break;
532 default:
533 debug_printf("r300: Bad number of multisamples!\n");
534 }
535 }
536
537 OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2);
538 OUT_CS(mspos0);
539 OUT_CS(mspos1);
540 END_CS;
541 }
542
543 void r300_emit_query_start(struct r300_context *r300, unsigned size, void*state)
544 {
545 struct r300_query *query = r300->query_current;
546 CS_LOCALS(r300);
547
548 if (!query)
549 return;
550
551 BEGIN_CS(size);
552 if (r300->screen->caps.family == CHIP_FAMILY_RV530) {
553 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
554 } else {
555 OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL);
556 }
557 OUT_CS_REG(R300_ZB_ZPASS_DATA, 0);
558 END_CS;
559 query->begin_emitted = TRUE;
560 }
561
562 static void r300_emit_query_end_frag_pipes(struct r300_context *r300,
563 struct r300_query *query)
564 {
565 struct r300_capabilities* caps = &r300->screen->caps;
566 uint32_t gb_pipes = r300->screen->info.r300_num_gb_pipes;
567 CS_LOCALS(r300);
568
569 assert(gb_pipes);
570
571 BEGIN_CS(6 * gb_pipes + 2);
572 /* I'm not so sure I like this switch, but it's hard to be elegant
573 * when there's so many special cases...
574 *
575 * So here's the basic idea. For each pipe, enable writes to it only,
576 * then put out the relocation for ZPASS_ADDR, taking into account a
577 * 4-byte offset for each pipe. RV380 and older are special; they have
578 * only two pipes, and the second pipe's enable is on bit 3, not bit 1,
579 * so there's a chipset cap for that. */
580 switch (gb_pipes) {
581 case 4:
582 /* pipe 3 only */
583 OUT_CS_REG(R300_SU_REG_DEST, 1 << 3);
584 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 3) * 4);
585 OUT_CS_RELOC(r300->query_current);
586 case 3:
587 /* pipe 2 only */
588 OUT_CS_REG(R300_SU_REG_DEST, 1 << 2);
589 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 2) * 4);
590 OUT_CS_RELOC(r300->query_current);
591 case 2:
592 /* pipe 1 only */
593 /* As mentioned above, accomodate RV380 and older. */
594 OUT_CS_REG(R300_SU_REG_DEST,
595 1 << (caps->high_second_pipe ? 3 : 1));
596 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4);
597 OUT_CS_RELOC(r300->query_current);
598 case 1:
599 /* pipe 0 only */
600 OUT_CS_REG(R300_SU_REG_DEST, 1 << 0);
601 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4);
602 OUT_CS_RELOC(r300->query_current);
603 break;
604 default:
605 fprintf(stderr, "r300: Implementation error: Chipset reports %d"
606 " pixel pipes!\n", gb_pipes);
607 abort();
608 }
609
610 /* And, finally, reset it to normal... */
611 OUT_CS_REG(R300_SU_REG_DEST, 0xF);
612 END_CS;
613 }
614
615 static void rv530_emit_query_end_single_z(struct r300_context *r300,
616 struct r300_query *query)
617 {
618 CS_LOCALS(r300);
619
620 BEGIN_CS(8);
621 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
622 OUT_CS_REG(R300_ZB_ZPASS_ADDR, query->num_results * 4);
623 OUT_CS_RELOC(r300->query_current);
624 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
625 END_CS;
626 }
627
628 static void rv530_emit_query_end_double_z(struct r300_context *r300,
629 struct r300_query *query)
630 {
631 CS_LOCALS(r300);
632
633 BEGIN_CS(14);
634 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
635 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 0) * 4);
636 OUT_CS_RELOC(r300->query_current);
637 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1);
638 OUT_CS_REG(R300_ZB_ZPASS_ADDR, (query->num_results + 1) * 4);
639 OUT_CS_RELOC(r300->query_current);
640 OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
641 END_CS;
642 }
643
644 void r300_emit_query_end(struct r300_context* r300)
645 {
646 struct r300_capabilities *caps = &r300->screen->caps;
647 struct r300_query *query = r300->query_current;
648
649 if (!query)
650 return;
651
652 if (query->begin_emitted == FALSE)
653 return;
654
655 if (caps->family == CHIP_FAMILY_RV530) {
656 if (r300->screen->info.r300_num_z_pipes == 2)
657 rv530_emit_query_end_double_z(r300, query);
658 else
659 rv530_emit_query_end_single_z(r300, query);
660 } else
661 r300_emit_query_end_frag_pipes(r300, query);
662
663 query->begin_emitted = FALSE;
664 query->num_results += query->num_pipes;
665
666 /* XXX grab all the results and reset the counter. */
667 if (query->num_results >= query->buf->size / 4 - 4) {
668 query->num_results = (query->buf->size / 4) / 2;
669 fprintf(stderr, "r300: Rewinding OQBO...\n");
670 }
671 }
672
673 void r300_emit_invariant_state(struct r300_context *r300,
674 unsigned size, void *state)
675 {
676 CS_LOCALS(r300);
677 WRITE_CS_TABLE(state, size);
678 }
679
680 void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state)
681 {
682 struct r300_rs_state* rs = state;
683 CS_LOCALS(r300);
684
685 BEGIN_CS(size);
686 OUT_CS_TABLE(rs->cb_main, RS_STATE_MAIN_SIZE);
687 if (rs->polygon_offset_enable) {
688 if (r300->zbuffer_bpp == 16) {
689 OUT_CS_TABLE(rs->cb_poly_offset_zb16, 5);
690 } else {
691 OUT_CS_TABLE(rs->cb_poly_offset_zb24, 5);
692 }
693 }
694 END_CS;
695 }
696
697 void r300_emit_rs_block_state(struct r300_context* r300,
698 unsigned size, void* state)
699 {
700 struct r300_rs_block* rs = (struct r300_rs_block*)state;
701 unsigned i;
702 /* It's the same for both INST and IP tables */
703 unsigned count = (rs->inst_count & R300_RS_INST_COUNT_MASK) + 1;
704 CS_LOCALS(r300);
705
706 if (DBG_ON(r300, DBG_RS_BLOCK)) {
707 r500_dump_rs_block(rs);
708
709 fprintf(stderr, "r300: RS emit:\n");
710
711 for (i = 0; i < count; i++)
712 fprintf(stderr, " : ip %d: 0x%08x\n", i, rs->ip[i]);
713
714 for (i = 0; i < count; i++)
715 fprintf(stderr, " : inst %d: 0x%08x\n", i, rs->inst[i]);
716
717 fprintf(stderr, " : count: 0x%08x inst_count: 0x%08x\n",
718 rs->count, rs->inst_count);
719 }
720
721 BEGIN_CS(size);
722 OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
723 OUT_CS(rs->vap_vtx_state_cntl);
724 OUT_CS(rs->vap_vsm_vtx_assm);
725 OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
726 OUT_CS(rs->vap_out_vtx_fmt[0]);
727 OUT_CS(rs->vap_out_vtx_fmt[1]);
728 OUT_CS_REG_SEQ(R300_GB_ENABLE, 1);
729 OUT_CS(rs->gb_enable);
730
731 if (r300->screen->caps.is_r500) {
732 OUT_CS_REG_SEQ(R500_RS_IP_0, count);
733 } else {
734 OUT_CS_REG_SEQ(R300_RS_IP_0, count);
735 }
736 OUT_CS_TABLE(rs->ip, count);
737
738 OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
739 OUT_CS(rs->count);
740 OUT_CS(rs->inst_count);
741
742 if (r300->screen->caps.is_r500) {
743 OUT_CS_REG_SEQ(R500_RS_INST_0, count);
744 } else {
745 OUT_CS_REG_SEQ(R300_RS_INST_0, count);
746 }
747 OUT_CS_TABLE(rs->inst, count);
748 END_CS;
749 }
750
751 void r300_emit_scissor_state(struct r300_context* r300,
752 unsigned size, void* state)
753 {
754 struct pipe_scissor_state* scissor = (struct pipe_scissor_state*)state;
755 CS_LOCALS(r300);
756
757 BEGIN_CS(size);
758 OUT_CS_REG_SEQ(R300_SC_CLIPRECT_TL_0, 2);
759 if (r300->screen->caps.is_r500) {
760 OUT_CS((scissor->minx << R300_CLIPRECT_X_SHIFT) |
761 (scissor->miny << R300_CLIPRECT_Y_SHIFT));
762 OUT_CS(((scissor->maxx - 1) << R300_CLIPRECT_X_SHIFT) |
763 ((scissor->maxy - 1) << R300_CLIPRECT_Y_SHIFT));
764 } else {
765 OUT_CS(((scissor->minx + 1440) << R300_CLIPRECT_X_SHIFT) |
766 ((scissor->miny + 1440) << R300_CLIPRECT_Y_SHIFT));
767 OUT_CS(((scissor->maxx + 1440-1) << R300_CLIPRECT_X_SHIFT) |
768 ((scissor->maxy + 1440-1) << R300_CLIPRECT_Y_SHIFT));
769 }
770 END_CS;
771 }
772
773 void r300_emit_textures_state(struct r300_context *r300,
774 unsigned size, void *state)
775 {
776 struct r300_textures_state *allstate = (struct r300_textures_state*)state;
777 struct r300_texture_sampler_state *texstate;
778 struct r300_resource *tex;
779 unsigned i;
780 boolean has_us_format = r300->screen->caps.has_us_format;
781 CS_LOCALS(r300);
782
783 BEGIN_CS(size);
784 OUT_CS_REG(R300_TX_ENABLE, allstate->tx_enable);
785
786 for (i = 0; i < allstate->count; i++) {
787 if ((1 << i) & allstate->tx_enable) {
788 texstate = &allstate->regs[i];
789 tex = r300_resource(allstate->sampler_views[i]->base.texture);
790
791 OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter0);
792 OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter1);
793 OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (i * 4),
794 texstate->border_color);
795
796 OUT_CS_REG(R300_TX_FORMAT0_0 + (i * 4), texstate->format.format0);
797 OUT_CS_REG(R300_TX_FORMAT1_0 + (i * 4), texstate->format.format1);
798 OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format.format2);
799
800 OUT_CS_REG(R300_TX_OFFSET_0 + (i * 4), texstate->format.tile_config);
801 OUT_CS_RELOC(tex);
802
803 if (has_us_format) {
804 OUT_CS_REG(R500_US_FORMAT0_0 + (i * 4),
805 texstate->format.us_format0);
806 }
807 }
808 }
809 END_CS;
810 }
811
812 void r300_emit_vertex_arrays(struct r300_context* r300, int offset,
813 boolean indexed, int instance_id)
814 {
815 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
816 struct pipe_vertex_element *velem = r300->velems->velem;
817 struct r300_resource *buf;
818 int i;
819 unsigned vertex_array_count = r300->velems->count;
820 unsigned packet_size = (vertex_array_count * 3 + 1) / 2;
821 struct pipe_vertex_buffer *vb1, *vb2;
822 unsigned *hw_format_size = r300->velems->format_size;
823 unsigned size1, size2, offset1, offset2, stride1, stride2;
824 CS_LOCALS(r300);
825
826 BEGIN_CS(2 + packet_size + vertex_array_count * 2);
827 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
828 OUT_CS(vertex_array_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
829
830 if (instance_id == -1) {
831 /* Non-instanced arrays. This ignores instance_divisor and instance_id. */
832 for (i = 0; i < vertex_array_count - 1; i += 2) {
833 vb1 = &vbuf[velem[i].vertex_buffer_index];
834 vb2 = &vbuf[velem[i+1].vertex_buffer_index];
835 size1 = hw_format_size[i];
836 size2 = hw_format_size[i+1];
837
838 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
839 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
840 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
841 OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
842 }
843
844 if (vertex_array_count & 1) {
845 vb1 = &vbuf[velem[i].vertex_buffer_index];
846 size1 = hw_format_size[i];
847
848 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
849 OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
850 }
851
852 for (i = 0; i < vertex_array_count; i++) {
853 buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer);
854 OUT_CS_RELOC(buf);
855 }
856 } else {
857 /* Instanced arrays. */
858 for (i = 0; i < vertex_array_count - 1; i += 2) {
859 vb1 = &vbuf[velem[i].vertex_buffer_index];
860 vb2 = &vbuf[velem[i+1].vertex_buffer_index];
861 size1 = hw_format_size[i];
862 size2 = hw_format_size[i+1];
863
864 if (velem[i].instance_divisor) {
865 stride1 = 0;
866 offset1 = vb1->buffer_offset + velem[i].src_offset +
867 (instance_id / velem[i].instance_divisor) * vb1->stride;
868 } else {
869 stride1 = vb1->stride;
870 offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
871 }
872 if (velem[i+1].instance_divisor) {
873 stride2 = 0;
874 offset2 = vb2->buffer_offset + velem[i+1].src_offset +
875 (instance_id / velem[i+1].instance_divisor) * vb2->stride;
876 } else {
877 stride2 = vb2->stride;
878 offset2 = vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride;
879 }
880
881 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1) |
882 R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(stride2));
883 OUT_CS(offset1);
884 OUT_CS(offset2);
885 }
886
887 if (vertex_array_count & 1) {
888 vb1 = &vbuf[velem[i].vertex_buffer_index];
889 size1 = hw_format_size[i];
890
891 if (velem[i].instance_divisor) {
892 stride1 = 0;
893 offset1 = vb1->buffer_offset + velem[i].src_offset +
894 (instance_id / velem[i].instance_divisor) * vb1->stride;
895 } else {
896 stride1 = vb1->stride;
897 offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
898 }
899
900 OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1));
901 OUT_CS(offset1);
902 }
903
904 for (i = 0; i < vertex_array_count; i++) {
905 buf = r300_resource(vbuf[velem[i].vertex_buffer_index].buffer);
906 OUT_CS_RELOC(buf);
907 }
908 }
909 END_CS;
910 }
911
912 void r300_emit_vertex_arrays_swtcl(struct r300_context *r300, boolean indexed)
913 {
914 CS_LOCALS(r300);
915
916 DBG(r300, DBG_SWTCL, "r300: Preparing vertex buffer %p for render, "
917 "vertex size %d\n", r300->vbo,
918 r300->vertex_info.size);
919 /* Set the pointer to our vertex buffer. The emitted values are this:
920 * PACKET3 [3D_LOAD_VBPNTR]
921 * COUNT [1]
922 * FORMAT [size | stride << 8]
923 * OFFSET [offset into BO]
924 * VBPNTR [relocated BO]
925 */
926 BEGIN_CS(7);
927 OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
928 OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
929 OUT_CS(r300->vertex_info.size |
930 (r300->vertex_info.size << 8));
931 OUT_CS(r300->draw_vbo_offset);
932 OUT_CS(0);
933 OUT_CS_RELOC(r300_resource(r300->vbo));
934 END_CS;
935 }
936
937 void r300_emit_vertex_stream_state(struct r300_context* r300,
938 unsigned size, void* state)
939 {
940 struct r300_vertex_stream_state *streams =
941 (struct r300_vertex_stream_state*)state;
942 unsigned i;
943 CS_LOCALS(r300);
944
945 if (DBG_ON(r300, DBG_PSC)) {
946 fprintf(stderr, "r300: PSC emit:\n");
947
948 for (i = 0; i < streams->count; i++) {
949 fprintf(stderr, " : prog_stream_cntl%d: 0x%08x\n", i,
950 streams->vap_prog_stream_cntl[i]);
951 }
952
953 for (i = 0; i < streams->count; i++) {
954 fprintf(stderr, " : prog_stream_cntl_ext%d: 0x%08x\n", i,
955 streams->vap_prog_stream_cntl_ext[i]);
956 }
957 }
958
959 BEGIN_CS(size);
960 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count);
961 OUT_CS_TABLE(streams->vap_prog_stream_cntl, streams->count);
962 OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count);
963 OUT_CS_TABLE(streams->vap_prog_stream_cntl_ext, streams->count);
964 END_CS;
965 }
966
967 void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state)
968 {
969 CS_LOCALS(r300);
970
971 BEGIN_CS(size);
972 OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
973 END_CS;
974 }
975
976 void r300_emit_vap_invariant_state(struct r300_context *r300,
977 unsigned size, void *state)
978 {
979 CS_LOCALS(r300);
980 WRITE_CS_TABLE(state, size);
981 }
982
983 void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
984 {
985 struct r300_vertex_shader* vs = (struct r300_vertex_shader*)state;
986 struct r300_vertex_program_code* code = &vs->code;
987 struct r300_screen* r300screen = r300->screen;
988 unsigned instruction_count = code->length / 4;
989
990 unsigned vtx_mem_size = r300screen->caps.is_r500 ? 128 : 72;
991 unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1);
992 unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
993 unsigned temp_count = MAX2(code->num_temporaries, 1);
994
995 unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count,
996 vtx_mem_size / output_count, 10);
997 unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5);
998
999 CS_LOCALS(r300);
1000
1001 BEGIN_CS(size);
1002
1003 /* R300_VAP_PVS_CODE_CNTL_0
1004 * R300_VAP_PVS_CONST_CNTL
1005 * R300_VAP_PVS_CODE_CNTL_1
1006 * See the r5xx docs for instructions on how to use these. */
1007 OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) |
1008 R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
1009 R300_PVS_LAST_INST(instruction_count - 1));
1010 OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, instruction_count - 1);
1011
1012 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
1013 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
1014 OUT_CS_TABLE(code->body.d, code->length);
1015
1016 OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
1017 R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
1018 R300_PVS_NUM_FPUS(r300screen->caps.num_vert_fpus) |
1019 R300_PVS_VF_MAX_VTX_NUM(12) |
1020 (r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
1021
1022 /* Emit flow control instructions. Even if there are no fc instructions,
1023 * we still need to write the registers to make sure they are cleared. */
1024 OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
1025 if (r300screen->caps.is_r500) {
1026 OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, R300_VS_MAX_FC_OPS * 2);
1027 OUT_CS_TABLE(code->fc_op_addrs.r500, R300_VS_MAX_FC_OPS * 2);
1028 } else {
1029 OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, R300_VS_MAX_FC_OPS);
1030 OUT_CS_TABLE(code->fc_op_addrs.r300, R300_VS_MAX_FC_OPS);
1031 }
1032 OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, R300_VS_MAX_FC_OPS);
1033 OUT_CS_TABLE(code->fc_loop_index, R300_VS_MAX_FC_OPS);
1034
1035 END_CS;
1036 }
1037
1038 void r300_emit_vs_constants(struct r300_context* r300,
1039 unsigned size, void *state)
1040 {
1041 unsigned count =
1042 ((struct r300_vertex_shader*)r300->vs_state.state)->externals_count;
1043 struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
1044 struct r300_vertex_shader *vs = (struct r300_vertex_shader*)r300->vs_state.state;
1045 unsigned i;
1046 int imm_first = vs->externals_count;
1047 int imm_end = vs->code.constants.Count;
1048 int imm_count = vs->immediates_count;
1049 CS_LOCALS(r300);
1050
1051 BEGIN_CS(size);
1052 OUT_CS_REG(R300_VAP_PVS_CONST_CNTL,
1053 R300_PVS_CONST_BASE_OFFSET(buf->buffer_base) |
1054 R300_PVS_MAX_CONST_ADDR(MAX2(imm_end - 1, 0)));
1055 if (vs->externals_count) {
1056 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
1057 (r300->screen->caps.is_r500 ?
1058 R500_PVS_CONST_START : R300_PVS_CONST_START) + buf->buffer_base);
1059 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
1060 if (buf->remap_table){
1061 for (i = 0; i < count; i++) {
1062 uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
1063 OUT_CS_TABLE(data, 4);
1064 }
1065 } else {
1066 OUT_CS_TABLE(buf->ptr, count * 4);
1067 }
1068 }
1069
1070 /* Emit immediates. */
1071 if (imm_count) {
1072 OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
1073 (r300->screen->caps.is_r500 ?
1074 R500_PVS_CONST_START : R300_PVS_CONST_START) +
1075 buf->buffer_base + imm_first);
1076 OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4);
1077 for (i = imm_first; i < imm_end; i++) {
1078 const float *data = vs->code.constants.Constants[i].u.Immediate;
1079 OUT_CS_TABLE(data, 4);
1080 }
1081 }
1082 END_CS;
1083 }
1084
1085 void r300_emit_viewport_state(struct r300_context* r300,
1086 unsigned size, void* state)
1087 {
1088 struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
1089 CS_LOCALS(r300);
1090
1091 BEGIN_CS(size);
1092 OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
1093 OUT_CS_TABLE(&viewport->xscale, 6);
1094 OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
1095 END_CS;
1096 }
1097
1098 void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
1099 {
1100 struct pipe_framebuffer_state *fb =
1101 (struct pipe_framebuffer_state*)r300->fb_state.state;
1102 struct r300_resource* tex;
1103 CS_LOCALS(r300);
1104
1105 tex = r300_resource(fb->zsbuf->texture);
1106
1107 BEGIN_CS(size);
1108 OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT,
1109 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
1110 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
1111 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
1112 OUT_CS(0);
1113 OUT_CS(tex->tex.hiz_dwords[fb->zsbuf->u.tex.level]);
1114 OUT_CS(r300->hiz_clear_value);
1115 END_CS;
1116
1117 /* Mark the current zbuffer's hiz ram as in use. */
1118 r300->hiz_in_use = TRUE;
1119 r300->hiz_func = HIZ_FUNC_NONE;
1120 r300_mark_atom_dirty(r300, &r300->hyperz_state);
1121 }
1122
1123 void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
1124 {
1125 struct pipe_framebuffer_state *fb =
1126 (struct pipe_framebuffer_state*)r300->fb_state.state;
1127 struct r300_resource *tex;
1128 CS_LOCALS(r300);
1129
1130 tex = r300_resource(fb->zsbuf->texture);
1131
1132 BEGIN_CS(size);
1133 OUT_CS_REG(R300_ZB_ZCACHE_CTLSTAT,
1134 R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE |
1135 R300_ZB_ZCACHE_CTLSTAT_ZC_FREE_FREE);
1136 OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2);
1137 OUT_CS(0);
1138 OUT_CS(tex->tex.zmask_dwords[fb->zsbuf->u.tex.level]);
1139 OUT_CS(0);
1140 END_CS;
1141
1142 /* Mark the current zbuffer's zmask as in use. */
1143 r300->zmask_in_use = TRUE;
1144 r300_mark_atom_dirty(r300, &r300->hyperz_state);
1145 }
1146
1147 void r300_emit_ztop_state(struct r300_context* r300,
1148 unsigned size, void* state)
1149 {
1150 struct r300_ztop_state* ztop = (struct r300_ztop_state*)state;
1151 CS_LOCALS(r300);
1152
1153 BEGIN_CS(size);
1154 OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top);
1155 END_CS;
1156 }
1157
1158 void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state)
1159 {
1160 CS_LOCALS(r300);
1161
1162 BEGIN_CS(size);
1163 OUT_CS_REG(R300_TX_INVALTAGS, 0);
1164 END_CS;
1165 }
1166
1167 boolean r300_emit_buffer_validate(struct r300_context *r300,
1168 boolean do_validate_vertex_buffers,
1169 struct pipe_resource *index_buffer)
1170 {
1171 struct pipe_framebuffer_state *fb =
1172 (struct pipe_framebuffer_state*)r300->fb_state.state;
1173 struct r300_textures_state *texstate =
1174 (struct r300_textures_state*)r300->textures_state.state;
1175 struct r300_resource *tex;
1176 unsigned i;
1177 boolean flushed = FALSE;
1178
1179 validate:
1180 if (r300->fb_state.dirty) {
1181 /* Color buffers... */
1182 for (i = 0; i < fb->nr_cbufs; i++) {
1183 tex = r300_resource(fb->cbufs[i]->texture);
1184 assert(tex && tex->buf && "cbuf is marked, but NULL!");
1185 r300->rws->cs_add_reloc(r300->cs, tex->cs_buf,
1186 RADEON_USAGE_READWRITE,
1187 r300_surface(fb->cbufs[i])->domain);
1188 }
1189 /* ...depth buffer... */
1190 if (fb->zsbuf) {
1191 tex = r300_resource(fb->zsbuf->texture);
1192 assert(tex && tex->buf && "zsbuf is marked, but NULL!");
1193 r300->rws->cs_add_reloc(r300->cs, tex->cs_buf,
1194 RADEON_USAGE_READWRITE,
1195 r300_surface(fb->zsbuf)->domain);
1196 }
1197 }
1198 if (r300->textures_state.dirty) {
1199 /* ...textures... */
1200 for (i = 0; i < texstate->count; i++) {
1201 if (!(texstate->tx_enable & (1 << i))) {
1202 continue;
1203 }
1204
1205 tex = r300_resource(texstate->sampler_views[i]->base.texture);
1206 r300->rws->cs_add_reloc(r300->cs, tex->cs_buf, RADEON_USAGE_READ,
1207 tex->domain);
1208 }
1209 }
1210 /* ...occlusion query buffer... */
1211 if (r300->query_current)
1212 r300->rws->cs_add_reloc(r300->cs, r300->query_current->cs_buf,
1213 RADEON_USAGE_WRITE, RADEON_DOMAIN_GTT);
1214 /* ...vertex buffer for SWTCL path... */
1215 if (r300->vbo)
1216 r300->rws->cs_add_reloc(r300->cs, r300_resource(r300->vbo)->cs_buf,
1217 RADEON_USAGE_READ,
1218 r300_resource(r300->vbo)->domain);
1219 /* ...vertex buffers for HWTCL path... */
1220 if (do_validate_vertex_buffers && r300->vertex_arrays_dirty) {
1221 struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
1222 struct pipe_vertex_buffer *last = r300->vertex_buffer +
1223 r300->nr_vertex_buffers;
1224 struct pipe_resource *buf;
1225
1226 for (; vbuf != last; vbuf++) {
1227 buf = vbuf->buffer;
1228 if (!buf)
1229 continue;
1230
1231 r300->rws->cs_add_reloc(r300->cs, r300_resource(buf)->cs_buf,
1232 RADEON_USAGE_READ,
1233 r300_resource(buf)->domain);
1234 }
1235 }
1236 /* ...and index buffer for HWTCL path. */
1237 if (index_buffer)
1238 r300->rws->cs_add_reloc(r300->cs, r300_resource(index_buffer)->cs_buf,
1239 RADEON_USAGE_READ,
1240 r300_resource(index_buffer)->domain);
1241
1242 /* Now do the validation (flush is called inside cs_validate on failure). */
1243 if (!r300->rws->cs_validate(r300->cs)) {
1244 /* Ooops, an infinite loop, give up. */
1245 if (flushed)
1246 return FALSE;
1247
1248 flushed = TRUE;
1249 goto validate;
1250 }
1251
1252 return TRUE;
1253 }
1254
1255 unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
1256 {
1257 struct r300_atom* atom;
1258 unsigned dwords = 0;
1259
1260 foreach_dirty_atom(r300, atom) {
1261 if (atom->dirty) {
1262 dwords += atom->size;
1263 }
1264 }
1265
1266 /* let's reserve some more, just in case */
1267 dwords += 32;
1268
1269 return dwords;
1270 }
1271
1272 unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
1273 {
1274 unsigned dwords = 0;
1275
1276 /* Emitted in flush. */
1277 dwords += 26; /* emit_query_end */
1278 dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
1279 if (r300->screen->caps.is_r500)
1280 dwords += 2;
1281
1282 return dwords;
1283 }
1284
1285 /* Emit all dirty state. */
1286 void r300_emit_dirty_state(struct r300_context* r300)
1287 {
1288 struct r300_atom *atom;
1289
1290 foreach_dirty_atom(r300, atom) {
1291 if (atom->dirty) {
1292 atom->emit(r300, atom->size, atom->state);
1293 atom->dirty = FALSE;
1294 }
1295 }
1296
1297 r300->first_dirty = NULL;
1298 r300->last_dirty = NULL;
1299 r300->dirty_hw++;
1300 }