r300g: add fallback for unaligned/unsupported vertex stride/offset/format
[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 #include "r300_state_derived.h"
43
44 #include <limits.h>
45
46 static uint32_t r300_translate_primitive(unsigned prim)
47 {
48 switch (prim) {
49 case PIPE_PRIM_POINTS:
50 return R300_VAP_VF_CNTL__PRIM_POINTS;
51 case PIPE_PRIM_LINES:
52 return R300_VAP_VF_CNTL__PRIM_LINES;
53 case PIPE_PRIM_LINE_LOOP:
54 return R300_VAP_VF_CNTL__PRIM_LINE_LOOP;
55 case PIPE_PRIM_LINE_STRIP:
56 return R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
57 case PIPE_PRIM_TRIANGLES:
58 return R300_VAP_VF_CNTL__PRIM_TRIANGLES;
59 case PIPE_PRIM_TRIANGLE_STRIP:
60 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
61 case PIPE_PRIM_TRIANGLE_FAN:
62 return R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
63 case PIPE_PRIM_QUADS:
64 return R300_VAP_VF_CNTL__PRIM_QUADS;
65 case PIPE_PRIM_QUAD_STRIP:
66 return R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
67 case PIPE_PRIM_POLYGON:
68 return R300_VAP_VF_CNTL__PRIM_POLYGON;
69 default:
70 return 0;
71 }
72 }
73
74 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
75 unsigned mode)
76 {
77 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
78 uint32_t color_control = rs->color_control;
79
80 /* By default (see r300_state.c:r300_create_rs_state) color_control is
81 * initialized to provoking the first vertex.
82 *
83 * Triangle fans must be reduced to the second vertex, not the first, in
84 * Gallium flatshade-first mode, as per the GL spec.
85 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
86 *
87 * Quads never provoke correctly in flatshade-first mode. The first
88 * vertex is never considered as provoking, so only the second, third,
89 * and fourth vertices can be selected, and both "third" and "last" modes
90 * select the fourth vertex. This is probably due to D3D lacking quads.
91 *
92 * Similarly, polygons reduce to the first, not the last, vertex, when in
93 * "last" mode, and all other modes start from the second vertex.
94 *
95 * ~ C.
96 */
97
98 if (rs->rs.flatshade_first) {
99 switch (mode) {
100 case PIPE_PRIM_TRIANGLE_FAN:
101 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
102 break;
103 case PIPE_PRIM_QUADS:
104 case PIPE_PRIM_QUAD_STRIP:
105 case PIPE_PRIM_POLYGON:
106 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
107 break;
108 default:
109 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
110 break;
111 }
112 } else {
113 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
114 }
115
116 return color_control;
117 }
118
119 static boolean index_bias_supported(struct r300_context *r300)
120 {
121 return r300->screen->caps.is_r500 &&
122 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
123 }
124
125 static void r500_emit_index_bias(struct r300_context *r300, int index_bias)
126 {
127 CS_LOCALS(r300);
128
129 BEGIN_CS(2);
130 OUT_CS_REG(R500_VAP_INDEX_OFFSET,
131 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
132 END_CS;
133 }
134
135 /* This function splits the index bias value into two parts:
136 * - buffer_offset: the value that can be safely added to buffer offsets
137 * in r300_emit_aos (it must yield a positive offset when added to
138 * a vertex buffer offset)
139 * - index_offset: the value that must be manually subtracted from indices
140 * in an index buffer to achieve negative offsets. */
141 static void r300_split_index_bias(struct r300_context *r300, int index_bias,
142 int *buffer_offset, int *index_offset)
143 {
144 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer;
145 struct pipe_vertex_element *velem = r300->velems->velem;
146 unsigned i, size;
147 int max_neg_bias;
148
149 if (index_bias < 0) {
150 /* See how large index bias we may subtract. We must be careful
151 * here because negative buffer offsets are not allowed
152 * by the DRM API. */
153 max_neg_bias = INT_MAX;
154 for (i = 0; i < r300->velems->count; i++) {
155 vb = &vbufs[velem[i].vertex_buffer_index];
156 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride;
157 max_neg_bias = MIN2(max_neg_bias, size);
158 }
159
160 /* Now set the minimum allowed value. */
161 *buffer_offset = MAX2(-max_neg_bias, index_bias);
162 } else {
163 /* A positive index bias is OK. */
164 *buffer_offset = index_bias;
165 }
166
167 *index_offset = index_bias - *buffer_offset;
168 }
169
170 enum r300_prepare_flags {
171 PREP_FIRST_DRAW = (1 << 0), /* call emit_dirty_state and friends? */
172 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */
173 PREP_EMIT_AOS = (1 << 2), /* call emit_aos? */
174 PREP_EMIT_AOS_SWTCL = (1 << 3), /* call emit_aos_swtcl? */
175 PREP_INDEXED = (1 << 4) /* is this draw_elements? */
176 };
177
178 /**
179 * Check if the requested number of dwords is available in the CS and
180 * if not, flush. Then validate buffers and emit dirty state.
181 * \param r300 The context.
182 * \param flags See r300_prepare_flags.
183 * \param index_buffer The index buffer to validate. The parameter may be NULL.
184 * \param cs_dwords The number of dwords to reserve in CS.
185 * \param aos_offset The offset passed to emit_aos.
186 * \param index_bias The index bias to emit.
187 * \param end_cs_dwords The number of free dwords which must be available
188 * at the end of CS after drawing in case the CS space
189 * management is performed by a draw_* function manually.
190 * The parameter may be NULL.
191 */
192 static void r300_prepare_for_rendering(struct r300_context *r300,
193 enum r300_prepare_flags flags,
194 struct pipe_resource *index_buffer,
195 unsigned cs_dwords,
196 int aos_offset,
197 int index_bias,
198 unsigned *end_cs_dwords)
199 {
200 unsigned end_dwords = 0;
201 boolean flushed = FALSE;
202 boolean first_draw = flags & PREP_FIRST_DRAW;
203 boolean emit_aos = flags & PREP_EMIT_AOS;
204 boolean emit_aos_swtcl = flags & PREP_EMIT_AOS_SWTCL;
205 boolean indexed = flags & PREP_INDEXED;
206 boolean hw_index_bias = index_bias_supported(r300);
207
208 /* Add dirty state, index offset, and AOS. */
209 if (first_draw) {
210 cs_dwords += r300_get_num_dirty_dwords(r300);
211
212 if (hw_index_bias)
213 cs_dwords += 2; /* emit_index_offset */
214
215 if (emit_aos)
216 cs_dwords += 55; /* emit_aos */
217
218 if (emit_aos_swtcl)
219 cs_dwords += 7; /* emit_aos_swtcl */
220 }
221
222 /* Emitted in flush. */
223 end_dwords += 26; /* emit_query_end */
224
225 cs_dwords += end_dwords;
226
227 /* Reserve requested CS space. */
228 if (!r300_check_cs(r300, cs_dwords)) {
229 r300->context.flush(&r300->context, 0, NULL);
230 flushed = TRUE;
231 }
232
233 /* Validate buffers and emit dirty state if needed. */
234 if (first_draw || flushed) {
235 r300_emit_buffer_validate(r300, flags & PREP_VALIDATE_VBOS, index_buffer);
236 r300_emit_dirty_state(r300);
237 if (hw_index_bias) {
238 if (r300->screen->caps.has_tcl)
239 r500_emit_index_bias(r300, index_bias);
240 else
241 r500_emit_index_bias(r300, 0);
242 }
243
244 if (emit_aos)
245 r300_emit_aos(r300, aos_offset, indexed);
246
247 if (emit_aos_swtcl)
248 r300_emit_aos_swtcl(r300, indexed);
249 }
250
251 if (end_cs_dwords)
252 *end_cs_dwords = end_dwords;
253 }
254
255 static boolean immd_is_good_idea(struct r300_context *r300,
256 unsigned count)
257 {
258 struct pipe_vertex_element* velem;
259 struct pipe_vertex_buffer* vbuf;
260 boolean checked[PIPE_MAX_ATTRIBS] = {0};
261 unsigned vertex_element_count = r300->velems->count;
262 unsigned i, vbi;
263
264 if (DBG_ON(r300, DBG_NO_IMMD)) {
265 return FALSE;
266 }
267
268 if (r300->draw) {
269 return FALSE;
270 }
271
272 if (count > 10) {
273 return FALSE;
274 }
275
276 /* We shouldn't map buffers referenced by CS, busy buffers,
277 * and ones placed in VRAM. */
278 /* XXX Check for VRAM buffers. */
279 for (i = 0; i < vertex_element_count; i++) {
280 velem = &r300->velems->velem[i];
281 vbi = velem->vertex_buffer_index;
282
283 if (!checked[vbi]) {
284 vbuf = &r300->vertex_buffer[vbi];
285
286 if (r300_buffer_is_referenced(&r300->context,
287 vbuf->buffer,
288 R300_REF_CS | R300_REF_HW)) {
289 /* It's a very bad idea to map it... */
290 return FALSE;
291 }
292 checked[vbi] = TRUE;
293 }
294 }
295 return TRUE;
296 }
297
298 /*****************************************************************************
299 * The emission of draw packets for r500. Older GPUs may use these functions *
300 * after resolving fallback issues (e.g. stencil ref two-sided). *
301 ****************************************************************************/
302
303 static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
304 unsigned mode,
305 unsigned start,
306 unsigned count)
307 {
308 struct pipe_vertex_element* velem;
309 struct pipe_vertex_buffer* vbuf;
310 unsigned vertex_element_count = r300->velems->count;
311 unsigned i, v, vbi, dw, elem_offset, dwords;
312
313 /* Size of the vertex, in dwords. */
314 unsigned vertex_size = 0;
315
316 /* Offsets of the attribute, in dwords, from the start of the vertex. */
317 unsigned offset[PIPE_MAX_ATTRIBS];
318
319 /* Size of the vertex element, in dwords. */
320 unsigned size[PIPE_MAX_ATTRIBS];
321
322 /* Stride to the same attrib in the next vertex in the vertex buffer,
323 * in dwords. */
324 unsigned stride[PIPE_MAX_ATTRIBS] = {0};
325
326 /* Mapped vertex buffers. */
327 uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
328 struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {NULL};
329
330 CS_LOCALS(r300);
331
332 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
333 for (i = 0; i < vertex_element_count; i++) {
334 velem = &r300->velems->velem[i];
335 offset[i] = velem->src_offset / 4;
336 size[i] = r300->velems->hw_format_size[i] / 4;
337 vertex_size += size[i];
338 vbi = velem->vertex_buffer_index;
339
340 /* Map the buffer. */
341 if (!map[vbi]) {
342 vbuf = &r300->vertex_buffer[vbi];
343 map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
344 vbuf->buffer,
345 PIPE_TRANSFER_READ,
346 &transfer[vbi]);
347 map[vbi] += vbuf->buffer_offset / 4;
348 stride[vbi] = vbuf->stride / 4;
349 }
350 }
351
352 dwords = 9 + count * vertex_size;
353
354 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL);
355
356 BEGIN_CS(dwords);
357 OUT_CS_REG(R300_GA_COLOR_CONTROL,
358 r300_provoking_vertex_fixes(r300, mode));
359 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
360 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
361 OUT_CS(count - 1);
362 OUT_CS(0);
363 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
364 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
365 r300_translate_primitive(mode));
366
367 /* Emit vertices. */
368 for (v = 0; v < count; v++) {
369 for (i = 0; i < vertex_element_count; i++) {
370 velem = &r300->velems->velem[i];
371 vbi = velem->vertex_buffer_index;
372 elem_offset = offset[i] + stride[vbi] * (v + start);
373
374 for (dw = 0; dw < size[i]; dw++) {
375 OUT_CS(map[vbi][elem_offset + dw]);
376 }
377 }
378 }
379 END_CS;
380
381 /* Unmap buffers. */
382 for (i = 0; i < vertex_element_count; i++) {
383 vbi = r300->velems->velem[i].vertex_buffer_index;
384
385 if (map[vbi]) {
386 vbuf = &r300->vertex_buffer[vbi];
387 pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]);
388 map[vbi] = NULL;
389 }
390 }
391 }
392
393 static void r300_emit_draw_arrays(struct r300_context *r300,
394 unsigned mode,
395 unsigned count)
396 {
397 boolean alt_num_verts = count > 65535;
398 CS_LOCALS(r300);
399
400 if (count >= (1 << 24)) {
401 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
402 "refusing to render.\n", count);
403 return;
404 }
405
406 BEGIN_CS(7 + (alt_num_verts ? 2 : 0));
407 if (alt_num_verts) {
408 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
409 }
410 OUT_CS_REG(R300_GA_COLOR_CONTROL,
411 r300_provoking_vertex_fixes(r300, mode));
412 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
413 OUT_CS(count - 1);
414 OUT_CS(0);
415 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
416 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
417 r300_translate_primitive(mode) |
418 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
419 END_CS;
420 }
421
422 static void r300_emit_draw_elements(struct r300_context *r300,
423 struct pipe_resource* indexBuffer,
424 unsigned indexSize,
425 unsigned minIndex,
426 unsigned maxIndex,
427 unsigned mode,
428 unsigned start,
429 unsigned count)
430 {
431 uint32_t count_dwords;
432 uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
433 boolean alt_num_verts = count > 65535;
434 CS_LOCALS(r300);
435
436 if (count >= (1 << 24)) {
437 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
438 "refusing to render.\n", count);
439 return;
440 }
441
442 maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
443
444 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
445 count, minIndex, maxIndex);
446
447 BEGIN_CS(13 + (alt_num_verts ? 2 : 0));
448 if (alt_num_verts) {
449 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
450 }
451 OUT_CS_REG(R300_GA_COLOR_CONTROL,
452 r300_provoking_vertex_fixes(r300, mode));
453 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
454 OUT_CS(maxIndex);
455 OUT_CS(minIndex);
456 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
457 if (indexSize == 4) {
458 count_dwords = count;
459 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
460 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
461 r300_translate_primitive(mode) |
462 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
463 } else {
464 count_dwords = (count + 1) / 2;
465 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
466 r300_translate_primitive(mode) |
467 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
468 }
469
470 /* INDX_BUFFER is a truly special packet3.
471 * Unlike most other packet3, where the offset is after the count,
472 * the order is reversed, so the relocation ends up carrying the
473 * size of the indexbuf instead of the offset.
474 */
475 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
476 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
477 (0 << R300_INDX_BUFFER_SKIP_SHIFT));
478 OUT_CS(offset_dwords << 2);
479 OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
480 r300_buffer(indexBuffer)->domain, 0, 0);
481
482 END_CS;
483 }
484
485 static void r300_shorten_ubyte_elts(struct r300_context* r300,
486 struct pipe_resource** elts,
487 int index_bias,
488 unsigned start,
489 unsigned count)
490 {
491 struct pipe_context* context = &r300->context;
492 struct pipe_screen* screen = r300->context.screen;
493 struct pipe_resource* new_elts;
494 unsigned char *in_map;
495 unsigned short *out_map;
496 struct pipe_transfer *src_transfer, *dst_transfer;
497 unsigned i;
498
499 new_elts = pipe_buffer_create(screen,
500 PIPE_BIND_INDEX_BUFFER,
501 2 * count);
502
503 in_map = pipe_buffer_map(context, *elts, PIPE_TRANSFER_READ, &src_transfer);
504 out_map = pipe_buffer_map(context, new_elts, PIPE_TRANSFER_WRITE, &dst_transfer);
505
506 in_map += start;
507
508 for (i = 0; i < count; i++) {
509 *out_map = (unsigned short)(*in_map + index_bias);
510 in_map++;
511 out_map++;
512 }
513
514 pipe_buffer_unmap(context, *elts, src_transfer);
515 pipe_buffer_unmap(context, new_elts, dst_transfer);
516
517 *elts = new_elts;
518 }
519
520 static void r300_rebuild_ushort_elts(struct r300_context *r300,
521 struct pipe_resource **elts,
522 int index_bias,
523 unsigned start, unsigned count)
524 {
525 struct pipe_context *context = &r300->context;
526 struct pipe_transfer *in_transfer = NULL;
527 struct pipe_transfer *out_transfer = NULL;
528 struct pipe_resource *new_elts;
529 unsigned short *in_map;
530 unsigned short *out_map;
531 unsigned i;
532
533 new_elts = pipe_buffer_create(context->screen,
534 PIPE_BIND_INDEX_BUFFER,
535 2 * count);
536
537 in_map = pipe_buffer_map(context, *elts,
538 PIPE_TRANSFER_READ, &in_transfer);
539 out_map = pipe_buffer_map(context, new_elts,
540 PIPE_TRANSFER_WRITE, &out_transfer);
541
542 in_map += start;
543 for (i = 0; i < count; i++) {
544 *out_map = (unsigned short)(*in_map + index_bias);
545 in_map++;
546 out_map++;
547 }
548
549 pipe_buffer_unmap(context, *elts, in_transfer);
550 pipe_buffer_unmap(context, new_elts, out_transfer);
551
552 *elts = new_elts;
553 }
554
555 static void r300_rebuild_uint_elts(struct r300_context *r300,
556 struct pipe_resource **elts,
557 int index_bias,
558 unsigned start, unsigned count)
559 {
560 struct pipe_context *context = &r300->context;
561 struct pipe_transfer *in_transfer = NULL;
562 struct pipe_transfer *out_transfer = NULL;
563 struct pipe_resource *new_elts;
564 unsigned int *in_map;
565 unsigned int *out_map;
566 unsigned i;
567
568 new_elts = pipe_buffer_create(context->screen,
569 PIPE_BIND_INDEX_BUFFER,
570 2 * count);
571
572 in_map = pipe_buffer_map(context, *elts,
573 PIPE_TRANSFER_READ, &in_transfer);
574 out_map = pipe_buffer_map(context, new_elts,
575 PIPE_TRANSFER_WRITE, &out_transfer);
576
577 in_map += start;
578 for (i = 0; i < count; i++) {
579 *out_map = (unsigned int)(*in_map + index_bias);
580 in_map++;
581 out_map++;
582 }
583
584 pipe_buffer_unmap(context, *elts, in_transfer);
585 pipe_buffer_unmap(context, new_elts, out_transfer);
586
587 *elts = new_elts;
588 }
589
590 /* This is the fast-path drawing & emission for HW TCL. */
591 static void r300_draw_range_elements(struct pipe_context* pipe,
592 struct pipe_resource* indexBuffer,
593 unsigned indexSize,
594 int indexBias,
595 unsigned minIndex,
596 unsigned maxIndex,
597 unsigned mode,
598 unsigned start,
599 unsigned count)
600 {
601 struct r300_context* r300 = r300_context(pipe);
602 struct pipe_resource* orgIndexBuffer = indexBuffer;
603 boolean alt_num_verts = r300->screen->caps.is_r500 &&
604 count > 65536 &&
605 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
606 unsigned short_count;
607 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
608 boolean translate = FALSE;
609
610 if (r300->skip_rendering) {
611 return;
612 }
613
614 if (!u_trim_pipe_prim(mode, &count)) {
615 return;
616 }
617
618 /* Set up fallback for incompatible vertex layout if needed. */
619 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
620 r300_begin_vertex_translate(r300);
621 translate = TRUE;
622 }
623
624 if (indexBias && !index_bias_supported(r300)) {
625 r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset);
626 }
627
628 /* Rebuild the index buffer if needed. */
629 switch (indexSize) {
630 case 1:
631 r300_shorten_ubyte_elts(r300, &indexBuffer, index_offset, start, count);
632 indexSize = 2;
633 start = 0;
634 break;
635
636 case 2:
637 if (start % 2 != 0 || index_offset) {
638 r300_rebuild_ushort_elts(r300, &indexBuffer, index_offset, start, count);
639 start = 0;
640 }
641 break;
642
643 case 4:
644 if (index_offset) {
645 r300_rebuild_uint_elts(r300, &indexBuffer, index_offset, start, count);
646 start = 0;
647 }
648 break;
649 }
650
651 r300_update_derived_state(r300);
652 r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
653
654 /* 15 dwords for emit_draw_elements */
655 r300_prepare_for_rendering(r300,
656 PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
657 indexBuffer, 15, buffer_offset, indexBias, NULL);
658
659 u_upload_flush(r300->upload_vb);
660 u_upload_flush(r300->upload_ib);
661 if (alt_num_verts || count <= 65535) {
662 r300_emit_draw_elements(r300, indexBuffer, indexSize,
663 minIndex, maxIndex, mode, start, count);
664 } else {
665 do {
666 short_count = MIN2(count, 65534);
667 r300_emit_draw_elements(r300, indexBuffer, indexSize,
668 minIndex, maxIndex,
669 mode, start, short_count);
670
671 start += short_count;
672 count -= short_count;
673
674 /* 15 dwords for emit_draw_elements */
675 if (count) {
676 r300_prepare_for_rendering(r300,
677 PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
678 indexBuffer, 15, buffer_offset, indexBias, NULL);
679 }
680 } while (count);
681 }
682
683 if (indexBuffer != orgIndexBuffer) {
684 pipe_resource_reference( &indexBuffer, NULL );
685 }
686
687 if (translate) {
688 r300_end_vertex_translate(r300);
689 }
690 }
691
692 /* Simple helpers for context setup. Should probably be moved to util. */
693 static void r300_draw_elements(struct pipe_context* pipe,
694 struct pipe_resource* indexBuffer,
695 unsigned indexSize, int indexBias, unsigned mode,
696 unsigned start, unsigned count)
697 {
698 struct r300_context *r300 = r300_context(pipe);
699
700 pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
701 0, r300->vertex_buffer_max_index,
702 mode, start, count);
703 }
704
705 static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
706 unsigned start, unsigned count)
707 {
708 struct r300_context* r300 = r300_context(pipe);
709 boolean alt_num_verts = r300->screen->caps.is_r500 &&
710 count > 65536 &&
711 r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
712 unsigned short_count;
713 boolean translate = FALSE;
714
715 if (r300->skip_rendering) {
716 return;
717 }
718
719 if (!u_trim_pipe_prim(mode, &count)) {
720 return;
721 }
722
723 /* Set up fallback for incompatible vertex layout if needed. */
724 if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
725 r300_begin_vertex_translate(r300);
726 translate = TRUE;
727 }
728
729 r300_update_derived_state(r300);
730
731 if (immd_is_good_idea(r300, count)) {
732 r300_emit_draw_arrays_immediate(r300, mode, start, count);
733 } else {
734 /* 9 spare dwords for emit_draw_arrays. */
735 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
736 NULL, 9, start, 0, NULL);
737
738 if (alt_num_verts || count <= 65535) {
739 r300_emit_draw_arrays(r300, mode, count);
740 } else {
741 do {
742 short_count = MIN2(count, 65535);
743 r300_emit_draw_arrays(r300, mode, short_count);
744
745 start += short_count;
746 count -= short_count;
747
748 /* 9 spare dwords for emit_draw_arrays. */
749 if (count) {
750 r300_prepare_for_rendering(r300,
751 PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
752 start, 0, NULL);
753 }
754 } while (count);
755 }
756 u_upload_flush(r300->upload_vb);
757 }
758
759 if (translate) {
760 r300_end_vertex_translate(r300);
761 }
762 }
763
764 /****************************************************************************
765 * The rest of this file is for SW TCL rendering only. Please be polite and *
766 * keep these functions separated so that they are easier to locate. ~C. *
767 ***************************************************************************/
768
769 /* SW TCL arrays, using Draw. */
770 static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
771 unsigned mode,
772 unsigned start,
773 unsigned count)
774 {
775 struct r300_context* r300 = r300_context(pipe);
776 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
777 int i;
778
779 if (r300->skip_rendering) {
780 return;
781 }
782
783 if (!u_trim_pipe_prim(mode, &count)) {
784 return;
785 }
786
787 r300_update_derived_state(r300);
788
789 for (i = 0; i < r300->vertex_buffer_count; i++) {
790 void* buf = pipe_buffer_map(pipe,
791 r300->vertex_buffer[i].buffer,
792 PIPE_TRANSFER_READ,
793 &vb_transfer[i]);
794 draw_set_mapped_vertex_buffer(r300->draw, i, buf);
795 }
796
797 draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL);
798
799 draw_arrays(r300->draw, mode, start, count);
800
801 /* XXX Not sure whether this is the best fix.
802 * It prevents CS from being rejected and weird assertion failures. */
803 draw_flush(r300->draw);
804
805 for (i = 0; i < r300->vertex_buffer_count; i++) {
806 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
807 vb_transfer[i]);
808 draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
809 }
810 }
811
812 /* SW TCL elements, using Draw. */
813 static void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
814 struct pipe_resource* indexBuffer,
815 unsigned indexSize,
816 int indexBias,
817 unsigned minIndex,
818 unsigned maxIndex,
819 unsigned mode,
820 unsigned start,
821 unsigned count)
822 {
823 struct r300_context* r300 = r300_context(pipe);
824 struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
825 struct pipe_transfer *ib_transfer;
826 int i;
827 void* indices;
828
829 if (r300->skip_rendering) {
830 return;
831 }
832
833 if (!u_trim_pipe_prim(mode, &count)) {
834 return;
835 }
836
837 r300_update_derived_state(r300);
838
839 for (i = 0; i < r300->vertex_buffer_count; i++) {
840 void* buf = pipe_buffer_map(pipe,
841 r300->vertex_buffer[i].buffer,
842 PIPE_TRANSFER_READ,
843 &vb_transfer[i]);
844 draw_set_mapped_vertex_buffer(r300->draw, i, buf);
845 }
846
847 indices = pipe_buffer_map(pipe, indexBuffer,
848 PIPE_TRANSFER_READ, &ib_transfer);
849 draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias,
850 minIndex, maxIndex, indices);
851
852 draw_arrays(r300->draw, mode, start, count);
853
854 /* XXX Not sure whether this is the best fix.
855 * It prevents CS from being rejected and weird assertion failures. */
856 draw_flush(r300->draw);
857
858 for (i = 0; i < r300->vertex_buffer_count; i++) {
859 pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
860 vb_transfer[i]);
861 draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
862 }
863
864 pipe_buffer_unmap(pipe, indexBuffer,
865 ib_transfer);
866 draw_set_mapped_element_buffer_range(r300->draw, 0, 0,
867 start, start + count - 1,
868 NULL);
869 }
870
871 /* Object for rendering using Draw. */
872 struct r300_render {
873 /* Parent class */
874 struct vbuf_render base;
875
876 /* Pipe context */
877 struct r300_context* r300;
878
879 /* Vertex information */
880 size_t vertex_size;
881 unsigned prim;
882 unsigned hwprim;
883
884 /* VBO */
885 struct pipe_resource* vbo;
886 size_t vbo_size;
887 size_t vbo_offset;
888 size_t vbo_max_used;
889 void * vbo_ptr;
890
891 struct pipe_transfer *vbo_transfer;
892 };
893
894 static INLINE struct r300_render*
895 r300_render(struct vbuf_render* render)
896 {
897 return (struct r300_render*)render;
898 }
899
900 static const struct vertex_info*
901 r300_render_get_vertex_info(struct vbuf_render* render)
902 {
903 struct r300_render* r300render = r300_render(render);
904 struct r300_context* r300 = r300render->r300;
905
906 return &r300->vertex_info;
907 }
908
909 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
910 ushort vertex_size,
911 ushort count)
912 {
913 struct r300_render* r300render = r300_render(render);
914 struct r300_context* r300 = r300render->r300;
915 struct pipe_screen* screen = r300->context.screen;
916 size_t size = (size_t)vertex_size * (size_t)count;
917
918 if (size + r300render->vbo_offset > r300render->vbo_size)
919 {
920 pipe_resource_reference(&r300->vbo, NULL);
921 r300render->vbo = pipe_buffer_create(screen,
922 PIPE_BIND_VERTEX_BUFFER,
923 R300_MAX_DRAW_VBO_SIZE);
924 r300render->vbo_offset = 0;
925 r300render->vbo_size = R300_MAX_DRAW_VBO_SIZE;
926 }
927
928 r300render->vertex_size = vertex_size;
929 r300->vbo = r300render->vbo;
930 r300->vbo_offset = r300render->vbo_offset;
931
932 return (r300render->vbo) ? TRUE : FALSE;
933 }
934
935 static void* r300_render_map_vertices(struct vbuf_render* render)
936 {
937 struct r300_render* r300render = r300_render(render);
938
939 assert(!r300render->vbo_transfer);
940
941 r300render->vbo_ptr = pipe_buffer_map(&r300render->r300->context,
942 r300render->vbo,
943 PIPE_TRANSFER_WRITE,
944 &r300render->vbo_transfer);
945
946 return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
947 }
948
949 static void r300_render_unmap_vertices(struct vbuf_render* render,
950 ushort min,
951 ushort max)
952 {
953 struct r300_render* r300render = r300_render(render);
954 struct pipe_context* context = &r300render->r300->context;
955
956 assert(r300render->vbo_transfer);
957
958 r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
959 r300render->vertex_size * (max + 1));
960 pipe_buffer_unmap(context, r300render->vbo, r300render->vbo_transfer);
961
962 r300render->vbo_transfer = NULL;
963 }
964
965 static void r300_render_release_vertices(struct vbuf_render* render)
966 {
967 struct r300_render* r300render = r300_render(render);
968
969 r300render->vbo_offset += r300render->vbo_max_used;
970 r300render->vbo_max_used = 0;
971 }
972
973 static boolean r300_render_set_primitive(struct vbuf_render* render,
974 unsigned prim)
975 {
976 struct r300_render* r300render = r300_render(render);
977
978 r300render->prim = prim;
979 r300render->hwprim = r300_translate_primitive(prim);
980
981 return TRUE;
982 }
983
984 static void r300_render_draw_arrays(struct vbuf_render* render,
985 unsigned start,
986 unsigned count)
987 {
988 struct r300_render* r300render = r300_render(render);
989 struct r300_context* r300 = r300render->r300;
990 uint8_t* ptr;
991 unsigned i;
992 unsigned dwords = 6;
993
994 CS_LOCALS(r300);
995
996 (void) i; (void) ptr;
997
998 r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
999 NULL, dwords, 0, 0, NULL);
1000
1001 DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
1002
1003 /* Uncomment to dump all VBOs rendered through this interface.
1004 * Slow and noisy!
1005 ptr = pipe_buffer_map(&r300render->r300->context,
1006 r300render->vbo, PIPE_TRANSFER_READ,
1007 &r300render->vbo_transfer);
1008
1009 for (i = 0; i < count; i++) {
1010 printf("r300: Vertex %d\n", i);
1011 draw_dump_emitted_vertex(&r300->vertex_info, ptr);
1012 ptr += r300->vertex_info.size * 4;
1013 printf("\n");
1014 }
1015
1016 pipe_buffer_unmap(&r300render->r300->context, r300render->vbo,
1017 r300render->vbo_transfer);
1018 */
1019
1020 BEGIN_CS(dwords);
1021 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1022 r300_provoking_vertex_fixes(r300, r300render->prim));
1023 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1024 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1025 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1026 r300render->hwprim);
1027 END_CS;
1028 }
1029
1030 static void r300_render_draw_elements(struct vbuf_render* render,
1031 const ushort* indices,
1032 uint count)
1033 {
1034 struct r300_render* r300render = r300_render(render);
1035 struct r300_context* r300 = r300render->r300;
1036 int i;
1037 unsigned end_cs_dwords;
1038 unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) /
1039 (r300render->r300->vertex_info.size * 4) - 1;
1040 unsigned short_count;
1041 struct r300_cs_info cs_info;
1042
1043 CS_LOCALS(r300);
1044
1045 /* Reserve at least 256 dwords.
1046 *
1047 * Below we manage the CS space manually because there may be more
1048 * indices than it can fit in CS. */
1049 r300_prepare_for_rendering(r300,
1050 PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
1051 NULL, 256, 0, 0, &end_cs_dwords);
1052
1053 while (count) {
1054 r300->rws->get_cs_info(r300->rws, &cs_info);
1055
1056 short_count = MIN2(count, (cs_info.free - end_cs_dwords - 6) * 2);
1057
1058 BEGIN_CS(6 + (short_count+1)/2);
1059 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1060 r300_provoking_vertex_fixes(r300, r300render->prim));
1061 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1062 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (short_count+1)/2);
1063 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (short_count << 16) |
1064 r300render->hwprim);
1065 for (i = 0; i < short_count-1; i += 2) {
1066 OUT_CS(indices[i+1] << 16 | indices[i]);
1067 }
1068 if (short_count % 2) {
1069 OUT_CS(indices[short_count-1]);
1070 }
1071 END_CS;
1072
1073 /* OK now subtract the emitted indices and see if we need to emit
1074 * another draw packet. */
1075 indices += short_count;
1076 count -= short_count;
1077
1078 if (count) {
1079 r300_prepare_for_rendering(r300,
1080 PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
1081 NULL, 256, 0, 0, &end_cs_dwords);
1082 }
1083 }
1084 }
1085
1086 static void r300_render_destroy(struct vbuf_render* render)
1087 {
1088 FREE(render);
1089 }
1090
1091 static struct vbuf_render* r300_render_create(struct r300_context* r300)
1092 {
1093 struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1094
1095 r300render->r300 = r300;
1096
1097 /* XXX find real numbers plz */
1098 r300render->base.max_vertex_buffer_bytes = 128 * 1024;
1099 r300render->base.max_indices = 16 * 1024;
1100
1101 r300render->base.get_vertex_info = r300_render_get_vertex_info;
1102 r300render->base.allocate_vertices = r300_render_allocate_vertices;
1103 r300render->base.map_vertices = r300_render_map_vertices;
1104 r300render->base.unmap_vertices = r300_render_unmap_vertices;
1105 r300render->base.set_primitive = r300_render_set_primitive;
1106 r300render->base.draw_elements = r300_render_draw_elements;
1107 r300render->base.draw_arrays = r300_render_draw_arrays;
1108 r300render->base.release_vertices = r300_render_release_vertices;
1109 r300render->base.destroy = r300_render_destroy;
1110
1111 r300render->vbo = NULL;
1112 r300render->vbo_size = 0;
1113 r300render->vbo_offset = 0;
1114
1115 return &r300render->base;
1116 }
1117
1118 struct draw_stage* r300_draw_stage(struct r300_context* r300)
1119 {
1120 struct vbuf_render* render;
1121 struct draw_stage* stage;
1122
1123 render = r300_render_create(r300);
1124
1125 if (!render) {
1126 return NULL;
1127 }
1128
1129 stage = draw_vbuf_stage(r300->draw, render);
1130
1131 if (!stage) {
1132 render->destroy(render);
1133 return NULL;
1134 }
1135
1136 draw_set_render(r300->draw, render);
1137
1138 return stage;
1139 }
1140
1141 /****************************************************************************
1142 * Two-sided stencil reference value fallback. It's designed to be as much
1143 * separate from rest of the driver as possible.
1144 ***************************************************************************/
1145
1146 struct r300_stencilref_context {
1147 void (*draw_arrays)(struct pipe_context *pipe,
1148 unsigned mode, unsigned start, unsigned count);
1149
1150 void (*draw_range_elements)(
1151 struct pipe_context *pipe, struct pipe_resource *indexBuffer,
1152 unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
1153 unsigned mode, unsigned start, unsigned count);
1154
1155 uint32_t rs_cull_mode;
1156 uint32_t zb_stencilrefmask;
1157 ubyte ref_value_front;
1158 };
1159
1160 static boolean r300_stencilref_needed(struct r300_context *r300)
1161 {
1162 struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
1163
1164 return dsa->two_sided_stencil_ref ||
1165 (dsa->two_sided &&
1166 r300->stencil_ref.ref_value[0] != r300->stencil_ref.ref_value[1]);
1167 }
1168
1169 /* Set drawing for front faces. */
1170 static void r300_stencilref_begin(struct r300_context *r300)
1171 {
1172 struct r300_stencilref_context *sr = r300->stencilref_fallback;
1173 struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state;
1174 struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
1175
1176 /* Save state. */
1177 sr->rs_cull_mode = rs->cull_mode;
1178 sr->zb_stencilrefmask = dsa->stencil_ref_mask;
1179 sr->ref_value_front = r300->stencil_ref.ref_value[0];
1180
1181 /* We *cull* pixels, therefore no need to mask out the bits. */
1182 rs->cull_mode |= R300_CULL_BACK;
1183
1184 r300->rs_state.dirty = TRUE;
1185 }
1186
1187 /* Set drawing for back faces. */
1188 static void r300_stencilref_switch_side(struct r300_context *r300)
1189 {
1190 struct r300_stencilref_context *sr = r300->stencilref_fallback;
1191 struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state;
1192 struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
1193
1194 rs->cull_mode = sr->rs_cull_mode | R300_CULL_FRONT;
1195 dsa->stencil_ref_mask = dsa->stencil_ref_bf;
1196 r300->stencil_ref.ref_value[0] = r300->stencil_ref.ref_value[1];
1197
1198 r300->rs_state.dirty = TRUE;
1199 r300->dsa_state.dirty = TRUE;
1200 }
1201
1202 /* Restore the original state. */
1203 static void r300_stencilref_end(struct r300_context *r300)
1204 {
1205 struct r300_stencilref_context *sr = r300->stencilref_fallback;
1206 struct r300_rs_state *rs = (struct r300_rs_state*)r300->rs_state.state;
1207 struct r300_dsa_state *dsa = (struct r300_dsa_state*)r300->dsa_state.state;
1208
1209 /* Restore state. */
1210 rs->cull_mode = sr->rs_cull_mode;
1211 dsa->stencil_ref_mask = sr->zb_stencilrefmask;
1212 r300->stencil_ref.ref_value[0] = sr->ref_value_front;
1213
1214 r300->rs_state.dirty = TRUE;
1215 r300->dsa_state.dirty = TRUE;
1216 }
1217
1218 static void r300_stencilref_draw_arrays(struct pipe_context *pipe, unsigned mode,
1219 unsigned start, unsigned count)
1220 {
1221 struct r300_context *r300 = r300_context(pipe);
1222 struct r300_stencilref_context *sr = r300->stencilref_fallback;
1223
1224 if (!r300_stencilref_needed(r300)) {
1225 sr->draw_arrays(pipe, mode, start, count);
1226 } else {
1227 r300_stencilref_begin(r300);
1228 sr->draw_arrays(pipe, mode, start, count);
1229 r300_stencilref_switch_side(r300);
1230 sr->draw_arrays(pipe, mode, start, count);
1231 r300_stencilref_end(r300);
1232 }
1233 }
1234
1235 static void r300_stencilref_draw_range_elements(
1236 struct pipe_context *pipe, struct pipe_resource *indexBuffer,
1237 unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
1238 unsigned mode, unsigned start, unsigned count)
1239 {
1240 struct r300_context *r300 = r300_context(pipe);
1241 struct r300_stencilref_context *sr = r300->stencilref_fallback;
1242
1243 if (!r300_stencilref_needed(r300)) {
1244 sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
1245 minIndex, maxIndex, mode, start, count);
1246 } else {
1247 r300_stencilref_begin(r300);
1248 sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
1249 minIndex, maxIndex, mode, start, count);
1250 r300_stencilref_switch_side(r300);
1251 sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
1252 minIndex, maxIndex, mode, start, count);
1253 r300_stencilref_end(r300);
1254 }
1255 }
1256
1257 static void r300_plug_in_stencil_ref_fallback(struct r300_context *r300)
1258 {
1259 r300->stencilref_fallback = CALLOC_STRUCT(r300_stencilref_context);
1260
1261 /* Save original draw functions. */
1262 r300->stencilref_fallback->draw_arrays = r300->context.draw_arrays;
1263 r300->stencilref_fallback->draw_range_elements = r300->context.draw_range_elements;
1264
1265 /* Override the draw functions. */
1266 r300->context.draw_arrays = r300_stencilref_draw_arrays;
1267 r300->context.draw_range_elements = r300_stencilref_draw_range_elements;
1268 }
1269
1270 void r300_init_render_functions(struct r300_context *r300)
1271 {
1272 /* Set generic functions. */
1273 r300->context.draw_elements = r300_draw_elements;
1274
1275 /* Set draw functions based on presence of HW TCL. */
1276 if (r300->screen->caps.has_tcl) {
1277 r300->context.draw_arrays = r300_draw_arrays;
1278 r300->context.draw_range_elements = r300_draw_range_elements;
1279 } else {
1280 r300->context.draw_arrays = r300_swtcl_draw_arrays;
1281 r300->context.draw_range_elements = r300_swtcl_draw_range_elements;
1282 }
1283
1284 /* Plug in two-sided stencil reference value fallback if needed. */
1285 if (!r300->screen->caps.is_r500)
1286 r300_plug_in_stencil_ref_fallback(r300);
1287 }