projects / mesa.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
history | raw | HEAD
r300g: put indices in CS if there's just a few of them and are in user memory
[mesa.git] / src / gallium / drivers / r300 / r300_render.c
1 /*
2 * Copyright 2009 Corbin Simpson <MostAwesomeDude@gmail.com>
3 * Copyright 2010 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_render: Vertex and index buffer primitive emission. Contains both
25 * HW TCL fastpath rendering, and SW TCL Draw-assisted rendering. */
26
27 #include "draw/draw_context.h"
28 #include "draw/draw_vbuf.h"
29
30 #include "util/u_inlines.h"
31
32 #include "util/u_format.h"
33 #include "util/u_memory.h"
34 #include "util/u_upload_mgr.h"
35 #include "util/u_prim.h"
36
37 #include "r300_cs.h"
38 #include "r300_context.h"
39 #include "r300_screen_buffer.h"
40 #include "r300_emit.h"
41 #include "r300_reg.h"
42
43 #include <limits.h>
44
45 #define IMMD_DWORDS 32
46
47 static uint32_t r300_translate_primitive(unsigned prim)
48 {
49 switch (prim) {
50 case PIPE_PRIM_POINTS:
51 return R300_VAP_VF_CNTL__PRIM_POINTS;
52 case PIPE_PRIM_LINES:
53 return R300_VAP_VF_CNTL__PRIM_LINES;
54 case PIPE_PRIM_LINE_LOOP:
55 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
56 case PIPE_PRIM_LINE_STRIP:
57 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
58 case PIPE_PRIM_TRIANGLES:
59 return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
60 case PIPE_PRIM_TRIANGLE_STRIP:
61 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
62 case PIPE_PRIM_TRIANGLE_FAN:
63 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
64 case PIPE_PRIM_QUADS:
65 return R300_VAP_VF_CNTL__PRIM_QUADS;
66 case PIPE_PRIM_QUAD_STRIP:
67 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
68 case PIPE_PRIM_POLYGON:
69 return R300_VAP_VF_CNTL__PRIM_POLYGON;
70 default:
71 return 0;
72 }
73 }
74
75 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
76 unsigned mode)
77 {
78 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
79 uint32_t color_control = rs->color_control;
80
81 /* By default (see r300_state.c:r300_create_rs_state) color_control is
82 * initialized to provoking the first vertex.
83 *
84 * Triangle fans must be reduced to the second vertex, not the first, in
85 * Gallium flatshade-first mode, as per the GL spec.
86 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
87 *
88 * Quads never provoke correctly in flatshade-first mode. The first
89 * vertex is never considered as provoking, so only the second, third,
90 * and fourth vertices can be selected, and both "third" and "last" modes
91 * select the fourth vertex. This is probably due to D3D lacking quads.
92 *
93 * Similarly, polygons reduce to the first, not the last, vertex, when in
94 * "last" mode, and all other modes start from the second vertex.
95 *
96 * ~ C.
97 */
98
99 if (rs->rs.flatshade_first) {
100 switch (mode) {
101 case PIPE_PRIM_TRIANGLE_FAN:
102 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
103 break;
104 case PIPE_PRIM_QUADS:
105 case PIPE_PRIM_QUAD_STRIP:
106 case PIPE_PRIM_POLYGON:
107 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
108 break;
109 default:
110 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
111 break;
112 }
113 } else {
114 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
115 }
116
117 return color_control;
118 }
119
120 void r500_emit_index_bias(struct r300_context *r300, int index_bias)
121 {
122 CS_LOCALS(r300);
123
124 BEGIN_CS(2);
125 OUT_CS_REG(R500_VAP_INDEX_OFFSET,
126 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
127 END_CS;
128 }
129
130 static void r300_emit_draw_init(struct r300_context *r300, unsigned mode,
131 unsigned min_index, unsigned max_index)
132 {
133 CS_LOCALS(r300);
134
135 BEGIN_CS(5);
136 OUT_CS_REG(R300_GA_COLOR_CONTROL,
137 r300_provoking_vertex_fixes(r300, mode));
138 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
139 OUT_CS(max_index);
140 OUT_CS(min_index);
141 END_CS;
142 }
143
144 /* This function splits the index bias value into two parts:
145 * - buffer_offset: the value that can be safely added to buffer offsets
146 * in r300_emit_vertex_arrays (it must yield a positive offset when added to
147 * a vertex buffer offset)
148 * - index_offset: the value that must be manually subtracted from indices
149 * in an index buffer to achieve negative offsets. */
150 static void r300_split_index_bias(struct r300_context *r300, int index_bias,
151 int *buffer_offset, int *index_offset)
152 {
153 struct pipe_vertex_buffer *vb, *vbufs = r300->vbuf_mgr->vertex_buffer;
154 struct pipe_vertex_element *velem = r300->velems->velem;
155 unsigned i, size;
156 int max_neg_bias;
157
158 if (index_bias < 0) {
159 /* See how large index bias we may subtract. We must be careful
160 * here because negative buffer offsets are not allowed
161 * by the DRM API. */
162 max_neg_bias = INT_MAX;
163 for (i = 0; i < r300->velems->count; i++) {
164 vb = &vbufs[velem[i].vertex_buffer_index];
165 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride;
166 max_neg_bias = MIN2(max_neg_bias, size);
167 }
168
169 /* Now set the minimum allowed value. */
170 *buffer_offset = MAX2(-max_neg_bias, index_bias);
171 } else {
172 /* A positive index bias is OK. */
173 *buffer_offset = index_bias;
174 }
175
176 *index_offset = index_bias - *buffer_offset;
177 }
178
179 enum r300_prepare_flags {
180 PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */
181 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */
182 PREP_EMIT_AOS = (1 << 2), /* call emit_vertex_arrays? */
183 PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_vertex_arrays_swtcl? */
184 PREP_INDEXED = (1 << 4) /* is this draw_elements? */
185 };
186
187 /**
188 * Check if the requested number of dwords is available in the CS and
189 * if not, flush.
190 * \param r300 The context.
191 * \param flags See r300_prepare_flags.
192 * \param cs_dwords The number of dwords to reserve in CS.
193 * \return TRUE if the CS was flushed
194 */
195 static boolean r300_reserve_cs_dwords(struct r300_context *r300,
196 enum r300_prepare_flags flags,
197 unsigned cs_dwords)
198 {
199 boolean flushed = FALSE;
200 boolean first_draw = flags & PREP_FIRST_DRAW;
201 boolean emit_vertex_arrays = flags & PREP_EMIT_AOS;
202 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_AOS_SWTCL;
203
204 /* Add dirty state, index offset, and AOS. */
205 if (first_draw) {
206 cs_dwords += r300_get_num_dirty_dwords(r300);
207
208 if (r300->screen->caps.index_bias_supported)
209 cs_dwords += 2; /* emit_index_offset */
210
211 if (emit_vertex_arrays)
212 cs_dwords += 55; /* emit_vertex_arrays */
213
214 if (emit_vertex_arrays_swtcl)
215 cs_dwords += 7; /* emit_vertex_arrays_swtcl */
216 }
217
218 cs_dwords += r300_get_num_cs_end_dwords(r300);
219
220 /* Reserve requested CS space. */
221 if (cs_dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
222 r300->context.flush(&r300->context, 0, NULL);
223 flushed = TRUE;
224 }
225
226 return flushed;
227 }
228
229 /**
230 * Validate buffers and emit dirty state.
231 * \param r300 The context.
232 * \param flags See r300_prepare_flags.
233 * \param index_buffer The index buffer to validate. The parameter may be NULL.
234 * \param buffer_offset The offset passed to emit_vertex_arrays.
235 * \param index_bias The index bias to emit.
236 * \return TRUE if rendering should be skipped
237 */
238 static boolean r300_emit_states(struct r300_context *r300,
239 enum r300_prepare_flags flags,
240 struct pipe_resource *index_buffer,
241 int buffer_offset,
242 int index_bias,
243 boolean user_buffers)
244 {
245 boolean first_draw = flags & PREP_FIRST_DRAW;
246 boolean emit_vertex_arrays = flags & PREP_EMIT_AOS;
247 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_AOS_SWTCL;
248 boolean indexed = flags & PREP_INDEXED;
249 boolean validate_vbos = flags & PREP_VALIDATE_VBOS;
250
251 /* Validate buffers and emit dirty state if needed. */
252 if (first_draw) {
253 if (r300->validate_buffers) {
254 if (!r300_emit_buffer_validate(r300, validate_vbos,
255 index_buffer)) {
256 fprintf(stderr, "r300: CS space validation failed. "
257 "(not enough memory?) Skipping rendering.\n");
258 return FALSE;
259 }
260
261 /* Consider the validation done only if everything was validated. */
262 if (validate_vbos) {
263 r300->validate_buffers = FALSE;
264 if (user_buffers)
265 r300->upload_vb_validated = TRUE;
266 if (r300->index_buffer.buffer &&
267 r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
268 r300->upload_ib_validated = TRUE;
269 }
270 }
271 }
272
273 r300_emit_dirty_state(r300);
274 if (r300->screen->caps.index_bias_supported) {
275 if (r300->screen->caps.has_tcl)
276 r500_emit_index_bias(r300, index_bias);
277 else
278 r500_emit_index_bias(r300, 0);
279 }
280
281 if (emit_vertex_arrays &&
282 (r300->vertex_arrays_dirty ||
283 r300->vertex_arrays_indexed != indexed ||
284 r300->vertex_arrays_offset != buffer_offset)) {
285 r300_emit_vertex_arrays(r300, buffer_offset, indexed);
286
287 r300->vertex_arrays_dirty = FALSE;
288 r300->vertex_arrays_indexed = indexed;
289 r300->vertex_arrays_offset = buffer_offset;
290 }
291
292 if (emit_vertex_arrays_swtcl)
293 r300_emit_vertex_arrays_swtcl(r300, indexed);
294 }
295
296 return TRUE;
297 }
298
299 /**
300 * Check if the requested number of dwords is available in the CS and
301 * if not, flush. Then validate buffers and emit dirty state.
302 * \param r300 The context.
303 * \param flags See r300_prepare_flags.
304 * \param index_buffer The index buffer to validate. The parameter may be NULL.
305 * \param cs_dwords The number of dwords to reserve in CS.
306 * \param buffer_offset The offset passed to emit_vertex_arrays.
307 * \param index_bias The index bias to emit.
308 * \return TRUE if rendering should be skipped
309 */
310 static boolean r300_prepare_for_rendering(struct r300_context *r300,
311 enum r300_prepare_flags flags,
312 struct pipe_resource *index_buffer,
313 unsigned cs_dwords,
314 int buffer_offset,
315 int index_bias,
316 boolean user_buffers)
317 {
318 if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
319 flags |= PREP_FIRST_DRAW;
320
321 return r300_emit_states(r300, flags, index_buffer, buffer_offset,
322 index_bias, user_buffers);
323 }
324
325 static boolean immd_is_good_idea(struct r300_context *r300,
326 unsigned count)
327 {
328 struct pipe_vertex_element* velem;
329 struct pipe_resource *buf;
330 boolean checked[PIPE_MAX_ATTRIBS] = {0};
331 unsigned vertex_element_count = r300->velems->count;
332 unsigned i, vbi;
333
334 if (DBG_ON(r300, DBG_NO_IMMD)) {
335 return FALSE;
336 }
337
338 if (r300->draw) {
339 return FALSE;
340 }
341
342 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
343 return FALSE;
344 }
345
346 /* We shouldn't map buffers referenced by CS, busy buffers,
347 * and ones placed in VRAM. */
348 for (i = 0; i < vertex_element_count; i++) {
349 velem = &r300->velems->velem[i];
350 vbi = velem->vertex_buffer_index;
351
352 if (!checked[vbi]) {
353 buf = r300->vbuf_mgr->real_vertex_buffer[vbi];
354
355 if ((r300_resource(buf)->domain != R300_DOMAIN_GTT)) {
356 return FALSE;
357 }
358
359 checked[vbi] = TRUE;
360 }
361 }
362 return TRUE;
363 }
364
365 /*****************************************************************************
366 * The HWTCL draw functions. *
367 ****************************************************************************/
368
369 static void r300_draw_arrays_immediate(struct r300_context *r300,
370 unsigned mode, unsigned start,
371 unsigned count)
372 {
373 struct pipe_vertex_element* velem;
374 struct pipe_vertex_buffer* vbuf;
375 unsigned vertex_element_count = r300->velems->count;
376 unsigned i, v, vbi;
377
378 /* Size of the vertex, in dwords. */
379 unsigned vertex_size = r300->velems->vertex_size_dwords;
380
381 /* The number of dwords for this draw operation. */
382 unsigned dwords = 4 + count * vertex_size;
383
384 /* Size of the vertex element, in dwords. */
385 unsigned size[PIPE_MAX_ATTRIBS];
386
387 /* Stride to the same attrib in the next vertex in the vertex buffer,
388 * in dwords. */
389 unsigned stride[PIPE_MAX_ATTRIBS];
390
391 /* Mapped vertex buffers. */
392 uint32_t* map[PIPE_MAX_ATTRIBS];
393 uint32_t* mapelem[PIPE_MAX_ATTRIBS];
394 struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0};
395
396 CS_LOCALS(r300);
397
398 if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0,
399 FALSE))
400 return;
401
402 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
403 for (i = 0; i < vertex_element_count; i++) {
404 velem = &r300->velems->velem[i];
405 size[i] = r300->velems->format_size[i] / 4;
406 vbi = velem->vertex_buffer_index;
407 vbuf = &r300->vbuf_mgr->vertex_buffer[vbi];
408 stride[i] = vbuf->stride / 4;
409
410 /* Map the buffer. */
411 if (!transfer[vbi]) {
412 map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
413 r300->vbuf_mgr->real_vertex_buffer[vbi],
414 PIPE_TRANSFER_READ |
415 PIPE_TRANSFER_UNSYNCHRONIZED,
416 &transfer[vbi]);
417 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start;
418 }
419 mapelem[i] = map[vbi] + (velem->src_offset / 4);
420 }
421
422 r300_emit_draw_init(r300, mode, 0, count-1);
423
424 BEGIN_CS(dwords);
425 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
426 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
427 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
428 r300_translate_primitive(mode));
429
430 /* Emit vertices. */
431 for (v = 0; v < count; v++) {
432 for (i = 0; i < vertex_element_count; i++) {
433 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
434 }
435 }
436 END_CS;
437
438 /* Unmap buffers. */
439 for (i = 0; i < vertex_element_count; i++) {
440 vbi = r300->velems->velem[i].vertex_buffer_index;
441
442 if (transfer[vbi]) {
443 pipe_buffer_unmap(&r300->context, transfer[vbi]);
444 transfer[vbi] = NULL;
445 }
446 }
447 }
448
449 static void r300_emit_draw_arrays(struct r300_context *r300,
450 unsigned mode,
451 unsigned count)
452 {
453 boolean alt_num_verts = count > 65535;
454 CS_LOCALS(r300);
455
456 if (count >= (1 << 24)) {
457 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
458 "refusing to render.\n", count);
459 return;
460 }
461
462 r300_emit_draw_init(r300, mode, 0, count-1);
463
464 BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
465 if (alt_num_verts) {
466 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
467 }
468 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
469 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
470 r300_translate_primitive(mode) |
471 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
472 END_CS;
473 }
474
475 static void r300_emit_draw_elements(struct r300_context *r300,
476 struct pipe_resource* indexBuffer,
477 unsigned indexSize,
478 unsigned minIndex,
479 unsigned maxIndex,
480 unsigned mode,
481 unsigned start,
482 unsigned count,
483 uint16_t *imm_indices3)
484 {
485 uint32_t count_dwords, offset_dwords;
486 boolean alt_num_verts = count > 65535;
487 CS_LOCALS(r300);
488
489 if (count >= (1 << 24) || maxIndex >= (1 << 24)) {
490 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
491 "refusing to render (maxIndex: %i).\n", count, maxIndex);
492 return;
493 }
494
495 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
496 count, minIndex, maxIndex);
497
498 r300_emit_draw_init(r300, mode, minIndex, maxIndex);
499
500 /* If start is odd, render the first triangle with indices embedded
501 * in the command stream. This will increase start by 3 and make it
502 * even. We can then proceed without a fallback. */
503 if (indexSize == 2 && (start & 1) &&
504 mode == PIPE_PRIM_TRIANGLES) {
505 BEGIN_CS(4);
506 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
507 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
508 R300_VAP_VF_CNTL__PRIM_TRIANGLES);
509 OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
510 OUT_CS(imm_indices3[2]);
511 END_CS;
512
513 start += 3;
514 count -= 3;
515 if (!count)
516 return;
517 }
518
519 offset_dwords = indexSize * start / sizeof(uint32_t);
520
521 BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
522 if (alt_num_verts) {
523 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
524 }
525 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
526 if (indexSize == 4) {
527 count_dwords = count;
528 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
529 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
530 r300_translate_primitive(mode) |
531 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
532 } else {
533 count_dwords = (count + 1) / 2;
534 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
535 r300_translate_primitive(mode) |
536 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
537 }
538
539 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
540 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
541 (0 << R300_INDX_BUFFER_SKIP_SHIFT));
542 OUT_CS(offset_dwords << 2);
543 OUT_CS(count_dwords);
544 OUT_CS_RELOC(r300_resource(indexBuffer));
545 END_CS;
546 }
547
548 static void r300_draw_elements_immediate(struct r300_context *r300,
549 int indexBias, unsigned minIndex,
550 unsigned maxIndex, unsigned mode,
551 unsigned start, unsigned count,
552 boolean user_buffers)
553 {
554 uint8_t *ptr1;
555 uint16_t *ptr2;
556 uint32_t *ptr4;
557 unsigned index_size = r300->index_buffer.index_size;
558 unsigned i, count_dwords = index_size == 4 ? count : (count + 1) / 2;
559 CS_LOCALS(r300);
560
561 /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
562 if (!r300_prepare_for_rendering(r300,
563 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS |
564 PREP_INDEXED, NULL, 2+count_dwords, 0, indexBias,
565 user_buffers))
566 return;
567
568 r300_emit_draw_init(r300, mode, minIndex, maxIndex);
569
570 BEGIN_CS(2 + count_dwords);
571 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
572
573 switch (index_size) {
574 case 1:
575 ptr1 = r300_resource(r300->index_buffer.buffer)->b.user_ptr;
576 ptr1 += start;
577
578 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
579 r300_translate_primitive(mode));
580
581 if (indexBias && !r300->screen->caps.index_bias_supported) {
582 for (i = 0; i < count-1; i += 2)
583 OUT_CS(((ptr1[i+1] + indexBias) << 16) |
584 (ptr1[i] + indexBias));
585
586 if (count & 1)
587 OUT_CS(ptr1[i] + indexBias);
588 } else {
589 for (i = 0; i < count-1; i += 2)
590 OUT_CS(((ptr1[i+1]) << 16) |
591 (ptr1[i] ));
592
593 if (count & 1)
594 OUT_CS(ptr1[i]);
595 }
596 break;
597
598 case 2:
599 ptr2 = (uint16_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
600 ptr2 += start;
601
602 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
603 r300_translate_primitive(mode));
604
605 if (indexBias && !r300->screen->caps.index_bias_supported) {
606 for (i = 0; i < count-1; i += 2)
607 OUT_CS(((ptr2[i+1] + indexBias) << 16) |
608 (ptr2[i] + indexBias));
609
610 if (count & 1)
611 OUT_CS(ptr2[i] + indexBias);
612 } else {
613 OUT_CS_TABLE(ptr2, count_dwords);
614 }
615 break;
616
617 case 4:
618 ptr4 = (uint32_t*)r300_resource(r300->index_buffer.buffer)->b.user_ptr;
619 ptr4 += start;
620
621 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
622 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
623 r300_translate_primitive(mode));
624
625 if (indexBias && !r300->screen->caps.index_bias_supported) {
626 for (i = 0; i < count; i++)
627 OUT_CS(ptr4[i] + indexBias);
628 } else {
629 OUT_CS_TABLE(ptr4, count_dwords);
630 }
631 break;
632 }
633 END_CS;
634 }
635
636 static void r300_draw_elements(struct r300_context *r300, int indexBias,
637 unsigned minIndex, unsigned maxIndex,
638 unsigned mode, unsigned start, unsigned count,
639 boolean user_buffers)
640 {
641 struct pipe_resource *indexBuffer = r300->index_buffer.buffer;
642 unsigned indexSize = r300->index_buffer.index_size;
643 struct pipe_resource* orgIndexBuffer = indexBuffer;
644 boolean alt_num_verts = r300->screen->caps.is_r500 &&
645 count > 65536 &&
646 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
647 unsigned short_count;
648 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
649 uint16_t indices3[3];
650
651 if (indexBias && !r300->screen->caps.index_bias_supported) {
652 r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset);
653 }
654
655 r300_translate_index_buffer(r300, &indexBuffer, &indexSize, index_offset,
656 &start, count);
657
658 /* Fallback for misaligned ushort indices. */
659 if (indexSize == 2 && (start & 1) &&
660 !r300_resource(indexBuffer)->b.user_ptr) {
661 struct pipe_transfer *transfer;
662
663 uint16_t *ptr = pipe_buffer_map(&r300->context, indexBuffer,
664 PIPE_TRANSFER_READ |
665 PIPE_TRANSFER_UNSYNCHRONIZED,
666 &transfer);
667
668 if (mode == PIPE_PRIM_TRIANGLES) {
669 memcpy(indices3, ptr + start, 6);
670 } else {
671 /* Copy the mapped index buffer directly to the upload buffer.
672 * The start index will be aligned simply from the fact that
673 * every sub-buffer in the upload buffer is aligned. */
674 r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
675 count, (uint8_t*)ptr);
676 }
677 pipe_buffer_unmap(&r300->context, transfer);
678 } else {
679 if (r300_resource(indexBuffer)->b.user_ptr)
680 r300_upload_index_buffer(r300, &indexBuffer, indexSize,
681 &start, count,
682 r300_resource(indexBuffer)->b.user_ptr);
683 }
684
685 /* 19 dwords for emit_draw_elements. Give up if the function fails. */
686 if (!r300_prepare_for_rendering(r300,
687 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS |
688 PREP_INDEXED, indexBuffer, 19, buffer_offset, indexBias,
689 user_buffers))
690 goto done;
691
692 if (alt_num_verts || count <= 65535) {
693 r300_emit_draw_elements(r300, indexBuffer, indexSize,
694 minIndex, maxIndex, mode, start, count, indices3);
695 } else {
696 do {
697 if (indexSize == 2 && (start & 1))
698 short_count = MIN2(count, 65535);
699 else
700 short_count = MIN2(count, 65534);
701
702 r300_emit_draw_elements(r300, indexBuffer, indexSize,
703 minIndex, maxIndex,
704 mode, start, short_count, indices3);
705
706 start += short_count;
707 count -= short_count;
708
709 /* 15 dwords for emit_draw_elements */
710 if (count) {
711 if (!r300_prepare_for_rendering(r300,
712 PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
713 indexBuffer, 19, buffer_offset, indexBias, user_buffers))
714 goto done;
715 }
716 } while (count);
717 }
718
719 done:
720 if (indexBuffer != orgIndexBuffer) {
721 pipe_resource_reference( &indexBuffer, NULL );
722 }
723 }
724
725 static void r300_draw_arrays(struct r300_context *r300, unsigned mode,
726 unsigned start, unsigned count,
727 boolean user_buffers)
728 {
729 boolean alt_num_verts = r300->screen->caps.is_r500 &&
730 count > 65536 &&
731 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
732 unsigned short_count;
733
734 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
735 if (!r300_prepare_for_rendering(r300,
736 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
737 NULL, 9, start, 0, user_buffers))
738 return;
739
740 if (alt_num_verts || count <= 65535) {
741 r300_emit_draw_arrays(r300, mode, count);
742 } else {
743 do {
744 short_count = MIN2(count, 65535);
745 r300_emit_draw_arrays(r300, mode, short_count);
746
747 start += short_count;
748 count -= short_count;
749
750 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
751 if (count) {
752 if (!r300_prepare_for_rendering(r300,
753 PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
754 start, 0, user_buffers))
755 return;
756 }
757 } while (count);
758 }
759 }
760
761 static void r300_draw_vbo(struct pipe_context* pipe,
762 const struct pipe_draw_info *info)
763 {
764 struct r300_context* r300 = r300_context(pipe);
765 unsigned count = info->count;
766 boolean buffers_updated, uploader_flushed;
767 boolean indexed = info->indexed && r300->index_buffer.buffer;
768 unsigned start_indexed = info->start + r300->index_buffer.offset;
769 int max_index = MIN2(r300->vbuf_mgr->max_index, info->max_index);
770
771 if (r300->skip_rendering ||
772 !u_trim_pipe_prim(info->mode, &count)) {
773 return;
774 }
775
776 /* Start the vbuf manager and update buffers if needed. */
777 u_vbuf_mgr_draw_begin(r300->vbuf_mgr, info,
778 &buffers_updated, &uploader_flushed);
779 if (buffers_updated) {
780 r300->vertex_arrays_dirty = TRUE;
781
782 if (uploader_flushed || !r300->upload_vb_validated) {
783 r300->upload_vb_validated = FALSE;
784 r300->validate_buffers = TRUE;
785 }
786 } else {
787 r300->upload_vb_validated = FALSE;
788 }
789
790 /* Draw. */
791 r300_update_derived_state(r300);
792
793 if (indexed) {
794 if (count <= 8 &&
795 r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
796 r300_draw_elements_immediate(r300, info->index_bias,
797 info->min_index, max_index,
798 info->mode, start_indexed, count,
799 buffers_updated);
800 } else {
801 r300_draw_elements(r300, info->index_bias, info->min_index,
802 max_index, info->mode, start_indexed, count,
803 buffers_updated);
804 }
805 } else {
806 if (immd_is_good_idea(r300, count)) {
807 r300_draw_arrays_immediate(r300, info->mode, info->start, count);
808 } else {
809 r300_draw_arrays(r300, info->mode, info->start, count,
810 buffers_updated);
811 }
812 }
813
814 u_vbuf_mgr_draw_end(r300->vbuf_mgr);
815 }
816
817 /****************************************************************************
818 * The rest of this file is for SW TCL rendering only. Please be polite and *
819 * keep these functions separated so that they are easier to locate. ~C. *
820 ***************************************************************************/
821
822 /* SW TCL elements, using Draw. */
823 static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
824 const struct pipe_draw_info *info)
825 {
826 struct r300_context* r300 = r300_context(pipe);
827 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
828 struct pipe_transfer *ib_transfer = NULL;
829 unsigned count = info->count;
830 int i;
831 void *indices = NULL;
832 boolean indexed = info->indexed && r300->index_buffer.buffer;
833
834 if (r300->skip_rendering) {
835 return;
836 }
837
838 if (!u_trim_pipe_prim(info->mode, &count)) {
839 return;
840 }
841
842 r300_update_derived_state(r300);
843
844 r300_reserve_cs_dwords(r300,
845 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL |
846 (indexed ? PREP_INDEXED : 0),
847 indexed ? 256 : 6);
848
849 for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
850 if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
851 void *buf = pipe_buffer_map(pipe,
852 r300->vbuf_mgr->vertex_buffer[i].buffer,
853 PIPE_TRANSFER_READ |
854 PIPE_TRANSFER_UNSYNCHRONIZED,
855 &vb_transfer[i]);
856 draw_set_mapped_vertex_buffer(r300->draw, i, buf);
857 }
858 }
859
860 if (indexed) {
861 indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
862 PIPE_TRANSFER_READ |
863 PIPE_TRANSFER_UNSYNCHRONIZED, &ib_transfer);
864 }
865
866 draw_set_mapped_index_buffer(r300->draw, indices);
867
868 r300->draw_vbo_locked = TRUE;
869 r300->draw_first_emitted = FALSE;
870 draw_vbo(r300->draw, info);
871 draw_flush(r300->draw);
872 r300->draw_vbo_locked = FALSE;
873
874 for (i = 0; i < r300->vbuf_mgr->nr_vertex_buffers; i++) {
875 if (r300->vbuf_mgr->vertex_buffer[i].buffer) {
876 pipe_buffer_unmap(pipe, vb_transfer[i]);
877 draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
878 }
879 }
880
881 if (indexed) {
882 pipe_buffer_unmap(pipe, ib_transfer);
883 draw_set_mapped_index_buffer(r300->draw, NULL);
884 }
885 }
886
887 /* Object for rendering using Draw. */
888 struct r300_render {
889 /* Parent class */
890 struct vbuf_render base;
891
892 /* Pipe context */
893 struct r300_context* r300;
894
895 /* Vertex information */
896 size_t vertex_size;
897 unsigned prim;
898 unsigned hwprim;
899
900 /* VBO */
901 size_t vbo_max_used;
902 void * vbo_ptr;
903
904 struct pipe_transfer *vbo_transfer;
905 };
906
907 static INLINE struct r300_render*
908 r300_render(struct vbuf_render* render)
909 {
910 return (struct r300_render*)render;
911 }
912
913 static const struct vertex_info*
914 r300_render_get_vertex_info(struct vbuf_render* render)
915 {
916 struct r300_render* r300render = r300_render(render);
917 struct r300_context* r300 = r300render->r300;
918
919 return &r300->vertex_info;
920 }
921
922 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
923 ushort vertex_size,
924 ushort count)
925 {
926 struct r300_render* r300render = r300_render(render);
927 struct r300_context* r300 = r300render->r300;
928 struct pipe_screen* screen = r300->context.screen;
929 size_t size = (size_t)vertex_size * (size_t)count;
930
931 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
932
933 if (size + r300->draw_vbo_offset > r300->draw_vbo_size)
934 {
935 pipe_resource_reference(&r300->vbo, NULL);
936 r300->vbo = pipe_buffer_create(screen,
937 PIPE_BIND_VERTEX_BUFFER,
938 R300_MAX_DRAW_VBO_SIZE);
939 r300->draw_vbo_offset = 0;
940 r300->draw_vbo_size = R300_MAX_DRAW_VBO_SIZE;
941 r300->validate_buffers = TRUE;
942 }
943
944 r300render->vertex_size = vertex_size;
945
946 return (r300->vbo) ? TRUE : FALSE;
947 }
948
949 static void* r300_render_map_vertices(struct vbuf_render* render)
950 {
951 struct r300_render* r300render = r300_render(render);
952 struct r300_context* r300 = r300render->r300;
953
954 assert(!r300render->vbo_transfer);
955
956 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
957
958 r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
959 r300->vbo,
960 PIPE_TRANSFER_WRITE |
961 PIPE_TRANSFER_UNSYNCHRONIZED,
962 &r300render->vbo_transfer);
963
964 assert(r300render->vbo_ptr);
965
966 return ((uint8_t*)r300render->vbo_ptr + r300->draw_vbo_offset);
967 }
968
969 static void r300_render_unmap_vertices(struct vbuf_render* render,
970 ushort min,
971 ushort max)
972 {
973 struct r300_render* r300render = r300_render(render);
974 struct pipe_context* context = &r300render->r300->context;
975 struct r300_context* r300 = r300render->r300;
976
977 assert(r300render->vbo_transfer);
978
979 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
980
981 r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
982 r300render->vertex_size * (max + 1));
983 pipe_buffer_unmap(context, r300render->vbo_transfer);
984
985 r300render->vbo_transfer = NULL;
986 }
987
988 static void r300_render_release_vertices(struct vbuf_render* render)
989 {
990 struct r300_render* r300render = r300_render(render);
991 struct r300_context* r300 = r300render->r300;
992
993 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
994
995 r300->draw_vbo_offset += r300render->vbo_max_used;
996 r300render->vbo_max_used = 0;
997 }
998
999 static boolean r300_render_set_primitive(struct vbuf_render* render,
1000 unsigned prim)
1001 {
1002 struct r300_render* r300render = r300_render(render);
1003
1004 r300render->prim = prim;
1005 r300render->hwprim = r300_translate_primitive(prim);
1006
1007 return TRUE;
1008 }
1009
1010 static void r300_render_draw_arrays(struct vbuf_render* render,
1011 unsigned start,
1012 unsigned count)
1013 {
1014 struct r300_render* r300render = r300_render(render);
1015 struct r300_context* r300 = r300render->r300;
1016 uint8_t* ptr;
1017 unsigned i;
1018 unsigned dwords = 6;
1019
1020 CS_LOCALS(r300);
1021 (void) i; (void) ptr;
1022
1023 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1024
1025 if (r300->draw_first_emitted) {
1026 if (!r300_prepare_for_rendering(r300,
1027 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
1028 NULL, dwords, 0, 0, FALSE))
1029 return;
1030 } else {
1031 if (!r300_emit_states(r300,
1032 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
1033 NULL, 0, 0, FALSE))
1034 return;
1035 }
1036
1037 BEGIN_CS(dwords);
1038 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1039 r300_provoking_vertex_fixes(r300, r300render->prim));
1040 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1041 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1042 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1043 r300render->hwprim);
1044 END_CS;
1045
1046 r300->draw_first_emitted = TRUE;
1047 }
1048
1049 static void r300_render_draw_elements(struct vbuf_render* render,
1050 const ushort* indices,
1051 uint count)
1052 {
1053 struct r300_render* r300render = r300_render(render);
1054 struct r300_context* r300 = r300render->r300;
1055 int i;
1056 unsigned end_cs_dwords;
1057 unsigned max_index = (r300->draw_vbo_size - r300->draw_vbo_offset) /
1058 (r300render->r300->vertex_info.size * 4) - 1;
1059 unsigned short_count;
1060 unsigned free_dwords;
1061
1062 CS_LOCALS(r300);
1063 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1064
1065 if (r300->draw_first_emitted) {
1066 if (!r300_prepare_for_rendering(r300,
1067 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
1068 NULL, 256, 0, 0, FALSE))
1069 return;
1070 } else {
1071 if (!r300_emit_states(r300,
1072 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
1073 NULL, 0, 0, FALSE))
1074 return;
1075 }
1076
1077 /* Below we manage the CS space manually because there may be more
1078 * indices than it can fit in CS. */
1079
1080 end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1081
1082 while (count) {
1083 free_dwords = R300_MAX_CMDBUF_DWORDS - r300->cs->cdw;
1084
1085 short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
1086
1087 BEGIN_CS(6 + (short_count+1)/2);
1088 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1089 r300_provoking_vertex_fixes(r300, r300render->prim));
1090 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1091 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
1092 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
1093 r300render->hwprim);
1094 for (i = 0; i < short_count-1; i += 2) {
1095 OUT_CS(indices[i+1] << 16 | indices[i]);
1096 }
1097 if (short_count % 2) {
1098 OUT_CS(indices[short_count-1]);
1099 }
1100 END_CS;
1101
1102 /* OK now subtract the emitted indices and see if we need to emit
1103 * another draw packet. */
1104 indices += short_count;
1105 count -= short_count;
1106
1107 if (count) {
1108 if (!r300_prepare_for_rendering(r300,
1109 PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
1110 NULL, 256, 0, 0, FALSE))
1111 return;
1112
1113 end_cs_dwords = r300_get_num_cs_end_dwords(r300);
1114 }
1115 }
1116
1117 r300->draw_first_emitted = TRUE;
1118 }
1119
1120 static void r300_render_destroy(struct vbuf_render* render)
1121 {
1122 FREE(render);
1123 }
1124
1125 static struct vbuf_render* r300_render_create(struct r300_context* r300)
1126 {
1127 struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1128
1129 r300render->r300 = r300;
1130
1131 r300render->base.max_vertex_buffer_bytes = 1024 * 1024;
1132 r300render->base.max_indices = 16 * 1024;
1133
1134 r300render->base.get_vertex_info = r300_render_get_vertex_info;
1135 r300render->base.allocate_vertices = r300_render_allocate_vertices;
1136 r300render->base.map_vertices = r300_render_map_vertices;
1137 r300render->base.unmap_vertices = r300_render_unmap_vertices;
1138 r300render->base.set_primitive = r300_render_set_primitive;
1139 r300render->base.draw_elements = r300_render_draw_elements;
1140 r300render->base.draw_arrays = r300_render_draw_arrays;
1141 r300render->base.release_vertices = r300_render_release_vertices;
1142 r300render->base.destroy = r300_render_destroy;
1143
1144 return &r300render->base;
1145 }
1146
1147 struct draw_stage* r300_draw_stage(struct r300_context* r300)
1148 {
1149 struct vbuf_render* render;
1150 struct draw_stage* stage;
1151
1152 render = r300_render_create(r300);
1153
1154 if (!render) {
1155 return NULL;
1156 }
1157
1158 stage = draw_vbuf_stage(r300->draw, render);
1159
1160 if (!stage) {
1161 render->destroy(render);
1162 return NULL;
1163 }
1164
1165 draw_set_render(r300->draw, render);
1166
1167 return stage;
1168 }
1169
1170 void r300_draw_flush_vbuf(struct r300_context *r300)
1171 {
1172 pipe_resource_reference(&r300->vbo, NULL);
1173 r300->draw_vbo_size = 0;
1174 }
1175
1176 /****************************************************************************
1177 * End of SW TCL functions *
1178 ***************************************************************************/
1179
1180 /* This functions is used to draw a rectangle for the blitter module.
1181 *
1182 * If we rendered a quad, the pixels on the main diagonal
1183 * would be computed and stored twice, which makes the clear/copy codepaths
1184 * somewhat inefficient. Instead we use a rectangular point sprite. */
1185 static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1186 unsigned x1, unsigned y1,
1187 unsigned x2, unsigned y2,
1188 float depth,
1189 enum blitter_attrib_type type,
1190 const float attrib[4])
1191 {
1192 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1193 unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1194 unsigned width = x2 - x1;
1195 unsigned height = y2 - y1;
1196 unsigned vertex_size =
1197 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1198 unsigned dwords = 13 + vertex_size +
1199 (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
1200 const float zeros[4] = {0, 0, 0, 0};
1201 CS_LOCALS(r300);
1202
1203 r300->context.set_vertex_buffers(&r300->context, 0, NULL);
1204
1205 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
1206 r300->sprite_coord_enable = 1;
1207
1208 r300_update_derived_state(r300);
1209
1210 /* Mark some states we don't care about as non-dirty. */
1211 r300->clip_state.dirty = FALSE;
1212 r300->viewport_state.dirty = FALSE;
1213
1214 if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0,
1215 FALSE))
1216 goto done;
1217
1218 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1219
1220 BEGIN_CS(dwords);
1221 /* Set up GA. */
1222 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1223
1224 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
1225 /* Set up the GA to generate texcoords. */
1226 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1227 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1228 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1229 OUT_CS_32F(attrib[0]);
1230 OUT_CS_32F(attrib[3]);
1231 OUT_CS_32F(attrib[2]);
1232 OUT_CS_32F(attrib[1]);
1233 }
1234
1235 /* Set up VAP controls. */
1236 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1237 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1238 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1239 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1240 OUT_CS(1);
1241 OUT_CS(0);
1242
1243 /* Draw. */
1244 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1245 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1246 R300_VAP_VF_CNTL__PRIM_POINTS);
1247
1248 OUT_CS_32F(x1 + width * 0.5f);
1249 OUT_CS_32F(y1 + height * 0.5f);
1250 OUT_CS_32F(depth);
1251 OUT_CS_32F(1);
1252
1253 if (vertex_size == 8) {
1254 if (!attrib)
1255 attrib = zeros;
1256 OUT_CS_TABLE(attrib, 4);
1257 }
1258 END_CS;
1259
1260 done:
1261 /* Restore the state. */
1262 r300_mark_atom_dirty(r300, &r300->clip_state);
1263 r300_mark_atom_dirty(r300, &r300->rs_state);
1264 r300_mark_atom_dirty(r300, &r300->viewport_state);
1265
1266 r300->sprite_coord_enable = last_sprite_coord_enable;
1267 }
1268
1269 static void r300_resource_resolve(struct pipe_context* pipe,
1270 struct pipe_resource* dest,
1271 unsigned dst_layer,
1272 struct pipe_resource* src,
1273 unsigned src_layer)
1274 {
1275 struct r300_context* r300 = r300_context(pipe);
1276 struct pipe_surface* srcsurf, surf_tmpl;
1277 struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
1278 float color[] = {0, 0, 0, 0};
1279
1280 memset(&surf_tmpl, 0, sizeof(surf_tmpl));
1281 surf_tmpl.format = src->format;
1282 surf_tmpl.usage = 0; /* not really a surface hence no bind flags */
1283 surf_tmpl.u.tex.level = 0; /* msaa resources cannot have mipmaps */
1284 surf_tmpl.u.tex.first_layer = src_layer;
1285 surf_tmpl.u.tex.last_layer = src_layer;
1286 srcsurf = pipe->create_surface(pipe, src, &surf_tmpl);
1287 surf_tmpl.format = dest->format;
1288 surf_tmpl.u.tex.first_layer = dst_layer;
1289 surf_tmpl.u.tex.last_layer = dst_layer;
1290
1291 DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
1292
1293 /* Enable AA resolve. */
1294 aa->dest = r300_surface(pipe->create_surface(pipe, dest, &surf_tmpl));
1295
1296 aa->aaresolve_ctl =
1297 R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
1298 R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
1299 r300->aa_state.size = 10;
1300 r300_mark_atom_dirty(r300, &r300->aa_state);
1301
1302 /* Resolve the surface. */
1303 r300->context.clear_render_target(pipe,
1304 srcsurf, color, 0, 0, src->width0, src->height0);
1305
1306 /* Disable AA resolve. */
1307 aa->aaresolve_ctl = 0;
1308 r300->aa_state.size = 4;
1309 r300_mark_atom_dirty(r300, &r300->aa_state);
1310
1311 pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
1312 pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL);
1313 }
1314
1315 void r300_init_render_functions(struct r300_context *r300)
1316 {
1317 /* Set draw functions based on presence of HW TCL. */
1318 if (r300->screen->caps.has_tcl) {
1319 r300->context.draw_vbo = r300_draw_vbo;
1320 } else {
1321 r300->context.draw_vbo = r300_swtcl_draw_vbo;
1322 }
1323
1324 r300->context.resource_resolve = r300_resource_resolve;
1325 r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
1326
1327 /* Plug in the two-sided stencil reference value fallback if needed. */
1328 if (!r300->screen->caps.is_r500)
1329 r300_plug_in_stencil_ref_fallback(r300);
1330 }