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/format/u_format.h"
33 #include "util/u_draw.h"
34 #include "util/u_memory.h"
35 #include "util/u_upload_mgr.h"
36 #include "util/u_prim.h"
37
38 #include "r300_cs.h"
39 #include "r300_context.h"
40 #include "r300_screen_buffer.h"
41 #include "r300_emit.h"
42 #include "r300_reg.h"
43
44 #include <limits.h>
45
46 #define IMMD_DWORDS 32
47
r300_translate_primitive(unsigned prim)48 static uint32_t r300_translate_primitive(unsigned prim)
49 {
50 static const int prim_conv[] = {
51 R300_VAP_VF_CNTL__PRIM_POINTS,
52 R300_VAP_VF_CNTL__PRIM_LINES,
53 R300_VAP_VF_CNTL__PRIM_LINE_LOOP,
54 R300_VAP_VF_CNTL__PRIM_LINE_STRIP,
55 R300_VAP_VF_CNTL__PRIM_TRIANGLES,
56 R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP,
57 R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN,
58 R300_VAP_VF_CNTL__PRIM_QUADS,
59 R300_VAP_VF_CNTL__PRIM_QUAD_STRIP,
60 R300_VAP_VF_CNTL__PRIM_POLYGON,
61 -1,
62 -1,
63 -1,
64 -1
65 };
66 unsigned hwprim = prim_conv[prim];
67
68 assert(hwprim != -1);
69 return hwprim;
70 }
71
r300_provoking_vertex_fixes(struct r300_context * r300,unsigned mode)72 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
73 unsigned mode)
74 {
75 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
76 uint32_t color_control = rs->color_control;
77
78 /* By default (see r300_state.c:r300_create_rs_state) color_control is
79 * initialized to provoking the first vertex.
80 *
81 * Triangle fans must be reduced to the second vertex, not the first, in
82 * Gallium flatshade-first mode, as per the GL spec.
83 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
84 *
85 * Quads never provoke correctly in flatshade-first mode. The first
86 * vertex is never considered as provoking, so only the second, third,
87 * and fourth vertices can be selected, and both "third" and "last" modes
88 * select the fourth vertex. This is probably due to D3D lacking quads.
89 *
90 * Similarly, polygons reduce to the first, not the last, vertex, when in
91 * "last" mode, and all other modes start from the second vertex.
92 *
93 * ~ C.
94 */
95
96 if (rs->rs.flatshade_first) {
97 switch (mode) {
98 case PIPE_PRIM_TRIANGLE_FAN:
99 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
100 break;
101 case PIPE_PRIM_QUADS:
102 case PIPE_PRIM_QUAD_STRIP:
103 case PIPE_PRIM_POLYGON:
104 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
105 break;
106 default:
107 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
108 break;
109 }
110 } else {
111 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
112 }
113
114 return color_control;
115 }
116
r500_emit_index_bias(struct r300_context * r300,int index_bias)117 void r500_emit_index_bias(struct r300_context *r300, int index_bias)
118 {
119 CS_LOCALS(r300);
120
121 BEGIN_CS(2);
122 OUT_CS_REG(R500_VAP_INDEX_OFFSET,
123 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
124 END_CS;
125 }
126
r300_emit_draw_init(struct r300_context * r300,unsigned mode,unsigned max_index)127 static void r300_emit_draw_init(struct r300_context *r300, unsigned mode,
128 unsigned max_index)
129 {
130 CS_LOCALS(r300);
131
132 assert(max_index < (1 << 24));
133
134 BEGIN_CS(5);
135 OUT_CS_REG(R300_GA_COLOR_CONTROL,
136 r300_provoking_vertex_fixes(r300, mode));
137 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
138 OUT_CS(max_index);
139 OUT_CS(0);
140 END_CS;
141 }
142
143 /* This function splits the index bias value into two parts:
144 * - buffer_offset: the value that can be safely added to buffer offsets
145 * in r300_emit_vertex_arrays (it must yield a positive offset when added to
146 * a vertex buffer offset)
147 * - index_offset: the value that must be manually subtracted from indices
148 * in an index buffer to achieve negative offsets. */
r300_split_index_bias(struct r300_context * r300,int index_bias,int * buffer_offset,int * index_offset)149 static void r300_split_index_bias(struct r300_context *r300, int index_bias,
150 int *buffer_offset, int *index_offset)
151 {
152 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer;
153 struct pipe_vertex_element *velem = r300->velems->velem;
154 unsigned i, size;
155 int max_neg_bias;
156
157 if (index_bias < 0) {
158 /* See how large index bias we may subtract. We must be careful
159 * here because negative buffer offsets are not allowed
160 * by the DRM API. */
161 max_neg_bias = INT_MAX;
162 for (i = 0; i < r300->velems->count; i++) {
163 vb = &vbufs[velem[i].vertex_buffer_index];
164 size = (vb->buffer_offset + velem[i].src_offset) / vb->stride;
165 max_neg_bias = MIN2(max_neg_bias, size);
166 }
167
168 /* Now set the minimum allowed value. */
169 *buffer_offset = MAX2(-max_neg_bias, index_bias);
170 } else {
171 /* A positive index bias is OK. */
172 *buffer_offset = index_bias;
173 }
174
175 *index_offset = index_bias - *buffer_offset;
176 }
177
178 enum r300_prepare_flags {
179 PREP_EMIT_STATES = (1 << 0), /* call emit_dirty_state and friends? */
180 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */
181 PREP_EMIT_VARRAYS = (1 << 2), /* call emit_vertex_arrays? */
182 PREP_EMIT_VARRAYS_SWTCL = (1 << 3), /* call emit_vertex_arrays_swtcl? */
183 PREP_INDEXED = (1 << 4) /* is this draw_elements? */
184 };
185
186 /**
187 * Check if the requested number of dwords is available in the CS and
188 * if not, flush.
189 * \param r300 The context.
190 * \param flags See r300_prepare_flags.
191 * \param cs_dwords The number of dwords to reserve in CS.
192 * \return TRUE if the CS was flushed
193 */
r300_reserve_cs_dwords(struct r300_context * r300,enum r300_prepare_flags flags,unsigned cs_dwords)194 static boolean r300_reserve_cs_dwords(struct r300_context *r300,
195 enum r300_prepare_flags flags,
196 unsigned cs_dwords)
197 {
198 boolean flushed = FALSE;
199 boolean emit_states = flags & PREP_EMIT_STATES;
200 boolean emit_vertex_arrays = flags & PREP_EMIT_VARRAYS;
201 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
202
203 /* Add dirty state, index offset, and AOS. */
204 if (emit_states)
205 cs_dwords += r300_get_num_dirty_dwords(r300);
206
207 if (r300->screen->caps.is_r500)
208 cs_dwords += 2; /* emit_index_offset */
209
210 if (emit_vertex_arrays)
211 cs_dwords += 55; /* emit_vertex_arrays */
212
213 if (emit_vertex_arrays_swtcl)
214 cs_dwords += 7; /* emit_vertex_arrays_swtcl */
215
216 cs_dwords += r300_get_num_cs_end_dwords(r300);
217
218 /* Reserve requested CS space. */
219 if (!r300->rws->cs_check_space(&r300->cs, cs_dwords, false)) {
220 r300_flush(&r300->context, PIPE_FLUSH_ASYNC, NULL);
221 flushed = TRUE;
222 }
223
224 return flushed;
225 }
226
227 /**
228 * Validate buffers and emit dirty state.
229 * \param r300 The context.
230 * \param flags See r300_prepare_flags.
231 * \param index_buffer The index buffer to validate. The parameter may be NULL.
232 * \param buffer_offset The offset passed to emit_vertex_arrays.
233 * \param index_bias The index bias to emit.
234 * \param instance_id Index of instance to render
235 * \return TRUE if rendering should be skipped
236 */
r300_emit_states(struct r300_context * r300,enum r300_prepare_flags flags,struct pipe_resource * index_buffer,int buffer_offset,int index_bias,int instance_id)237 static boolean r300_emit_states(struct r300_context *r300,
238 enum r300_prepare_flags flags,
239 struct pipe_resource *index_buffer,
240 int buffer_offset,
241 int index_bias, int instance_id)
242 {
243 boolean emit_states = flags & PREP_EMIT_STATES;
244 boolean emit_vertex_arrays = flags & PREP_EMIT_VARRAYS;
245 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
246 boolean indexed = flags & PREP_INDEXED;
247 boolean validate_vbos = flags & PREP_VALIDATE_VBOS;
248
249 /* Validate buffers and emit dirty state if needed. */
250 if (emit_states || (emit_vertex_arrays && validate_vbos)) {
251 if (!r300_emit_buffer_validate(r300, validate_vbos,
252 index_buffer)) {
253 fprintf(stderr, "r300: CS space validation failed. "
254 "(not enough memory?) Skipping rendering.\n");
255 return FALSE;
256 }
257 }
258
259 if (emit_states)
260 r300_emit_dirty_state(r300);
261
262 if (r300->screen->caps.is_r500) {
263 if (r300->screen->caps.has_tcl)
264 r500_emit_index_bias(r300, index_bias);
265 else
266 r500_emit_index_bias(r300, 0);
267 }
268
269 if (emit_vertex_arrays &&
270 (r300->vertex_arrays_dirty ||
271 r300->vertex_arrays_indexed != indexed ||
272 r300->vertex_arrays_offset != buffer_offset ||
273 r300->vertex_arrays_instance_id != instance_id)) {
274 r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id);
275
276 r300->vertex_arrays_dirty = FALSE;
277 r300->vertex_arrays_indexed = indexed;
278 r300->vertex_arrays_offset = buffer_offset;
279 r300->vertex_arrays_instance_id = instance_id;
280 }
281
282 if (emit_vertex_arrays_swtcl)
283 r300_emit_vertex_arrays_swtcl(r300, indexed);
284
285 return TRUE;
286 }
287
288 /**
289 * Check if the requested number of dwords is available in the CS and
290 * if not, flush. Then validate buffers and emit dirty state.
291 * \param r300 The context.
292 * \param flags See r300_prepare_flags.
293 * \param index_buffer The index buffer to validate. The parameter may be NULL.
294 * \param cs_dwords The number of dwords to reserve in CS.
295 * \param buffer_offset The offset passed to emit_vertex_arrays.
296 * \param index_bias The index bias to emit.
297 * \param instance_id The instance to render.
298 * \return TRUE if rendering should be skipped
299 */
r300_prepare_for_rendering(struct r300_context * r300,enum r300_prepare_flags flags,struct pipe_resource * index_buffer,unsigned cs_dwords,int buffer_offset,int index_bias,int instance_id)300 static boolean r300_prepare_for_rendering(struct r300_context *r300,
301 enum r300_prepare_flags flags,
302 struct pipe_resource *index_buffer,
303 unsigned cs_dwords,
304 int buffer_offset,
305 int index_bias,
306 int instance_id)
307 {
308 /* Make sure there is enough space in the command stream and emit states. */
309 if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
310 flags |= PREP_EMIT_STATES;
311
312 return r300_emit_states(r300, flags, index_buffer, buffer_offset,
313 index_bias, instance_id);
314 }
315
immd_is_good_idea(struct r300_context * r300,unsigned count)316 static boolean immd_is_good_idea(struct r300_context *r300,
317 unsigned count)
318 {
319 if (DBG_ON(r300, DBG_NO_IMMD)) {
320 return FALSE;
321 }
322
323 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
324 return FALSE;
325 }
326
327 /* Buffers can only be used for read by r300 (except query buffers, but
328 * those can't be bound by an gallium frontend as vertex buffers). */
329 return TRUE;
330 }
331
332 /*****************************************************************************
333 * The HWTCL draw functions. *
334 ****************************************************************************/
335
r300_draw_arrays_immediate(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)336 static void r300_draw_arrays_immediate(struct r300_context *r300,
337 const struct pipe_draw_info *info,
338 const struct pipe_draw_start_count_bias *draw)
339 {
340 struct pipe_vertex_element* velem;
341 struct pipe_vertex_buffer* vbuf;
342 unsigned vertex_element_count = r300->velems->count;
343 unsigned i, v, vbi;
344
345 /* Size of the vertex, in dwords. */
346 unsigned vertex_size = r300->velems->vertex_size_dwords;
347
348 /* The number of dwords for this draw operation. */
349 unsigned dwords = 4 + draw->count * vertex_size;
350
351 /* Size of the vertex element, in dwords. */
352 unsigned size[PIPE_MAX_ATTRIBS];
353
354 /* Stride to the same attrib in the next vertex in the vertex buffer,
355 * in dwords. */
356 unsigned stride[PIPE_MAX_ATTRIBS];
357
358 /* Mapped vertex buffers. */
359 uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
360 uint32_t* mapelem[PIPE_MAX_ATTRIBS];
361
362 CS_LOCALS(r300);
363
364 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
365 return;
366
367 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
368 for (i = 0; i < vertex_element_count; i++) {
369 velem = &r300->velems->velem[i];
370 size[i] = r300->velems->format_size[i] / 4;
371 vbi = velem->vertex_buffer_index;
372 vbuf = &r300->vertex_buffer[vbi];
373 stride[i] = vbuf->stride / 4;
374
375 /* Map the buffer. */
376 if (!map[vbi]) {
377 map[vbi] = (uint32_t*)r300->rws->buffer_map(r300->rws,
378 r300_resource(vbuf->buffer.resource)->buf,
379 &r300->cs, PIPE_MAP_READ | PIPE_MAP_UNSYNCHRONIZED);
380 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * draw->start;
381 }
382 mapelem[i] = map[vbi] + (velem->src_offset / 4);
383 }
384
385 r300_emit_draw_init(r300, info->mode, draw->count-1);
386
387 BEGIN_CS(dwords);
388 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
389 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, draw->count * vertex_size);
390 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (draw->count << 16) |
391 r300_translate_primitive(info->mode));
392
393 /* Emit vertices. */
394 for (v = 0; v < draw->count; v++) {
395 for (i = 0; i < vertex_element_count; i++) {
396 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
397 }
398 }
399 END_CS;
400 }
401
r300_emit_draw_arrays(struct r300_context * r300,unsigned mode,unsigned count)402 static void r300_emit_draw_arrays(struct r300_context *r300,
403 unsigned mode,
404 unsigned count)
405 {
406 boolean alt_num_verts = count > 65535;
407 CS_LOCALS(r300);
408
409 if (count >= (1 << 24)) {
410 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
411 "refusing to render.\n", count);
412 return;
413 }
414
415 r300_emit_draw_init(r300, mode, count-1);
416
417 BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
418 if (alt_num_verts) {
419 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
420 }
421 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
422 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
423 r300_translate_primitive(mode) |
424 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
425 END_CS;
426 }
427
r300_emit_draw_elements(struct r300_context * r300,struct pipe_resource * indexBuffer,unsigned indexSize,unsigned max_index,unsigned mode,unsigned start,unsigned count,uint16_t * imm_indices3)428 static void r300_emit_draw_elements(struct r300_context *r300,
429 struct pipe_resource* indexBuffer,
430 unsigned indexSize,
431 unsigned max_index,
432 unsigned mode,
433 unsigned start,
434 unsigned count,
435 uint16_t *imm_indices3)
436 {
437 uint32_t count_dwords, offset_dwords;
438 boolean alt_num_verts = count > 65535;
439 CS_LOCALS(r300);
440
441 if (count >= (1 << 24)) {
442 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
443 "refusing to render (max_index: %i).\n", count, max_index);
444 return;
445 }
446
447 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, max %u\n",
448 count, max_index);
449
450 r300_emit_draw_init(r300, mode, max_index);
451
452 /* If start is odd, render the first triangle with indices embedded
453 * in the command stream. This will increase start by 3 and make it
454 * even. We can then proceed without a fallback. */
455 if (indexSize == 2 && (start & 1) &&
456 mode == PIPE_PRIM_TRIANGLES) {
457 BEGIN_CS(4);
458 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
459 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
460 R300_VAP_VF_CNTL__PRIM_TRIANGLES);
461 OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
462 OUT_CS(imm_indices3[2]);
463 END_CS;
464
465 start += 3;
466 count -= 3;
467 if (!count)
468 return;
469 }
470
471 offset_dwords = indexSize * start / sizeof(uint32_t);
472
473 BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
474 if (alt_num_verts) {
475 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
476 }
477 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
478 if (indexSize == 4) {
479 count_dwords = count;
480 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
481 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
482 r300_translate_primitive(mode) |
483 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
484 } else {
485 count_dwords = (count + 1) / 2;
486 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
487 r300_translate_primitive(mode) |
488 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
489 }
490
491 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
492 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
493 (0 << R300_INDX_BUFFER_SKIP_SHIFT));
494 OUT_CS(offset_dwords << 2);
495 OUT_CS(count_dwords);
496 OUT_CS_RELOC(r300_resource(indexBuffer));
497 END_CS;
498 }
499
r300_draw_elements_immediate(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)500 static void r300_draw_elements_immediate(struct r300_context *r300,
501 const struct pipe_draw_info *info,
502 const struct pipe_draw_start_count_bias *draw)
503 {
504 const uint8_t *ptr1;
505 const uint16_t *ptr2;
506 const uint32_t *ptr4;
507 unsigned index_size = info->index_size;
508 unsigned i, count_dwords = index_size == 4 ? draw->count :
509 (draw->count + 1) / 2;
510 CS_LOCALS(r300);
511
512 /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
513 if (!r300_prepare_for_rendering(r300,
514 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
515 PREP_INDEXED, NULL, 2+count_dwords, 0, draw->index_bias, -1))
516 return;
517
518 r300_emit_draw_init(r300, info->mode, info->max_index);
519
520 BEGIN_CS(2 + count_dwords);
521 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
522
523 switch (index_size) {
524 case 1:
525 ptr1 = (uint8_t*)info->index.user;
526 ptr1 += draw->start;
527
528 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
529 r300_translate_primitive(info->mode));
530
531 if (draw->index_bias && !r300->screen->caps.is_r500) {
532 for (i = 0; i < draw->count-1; i += 2)
533 OUT_CS(((ptr1[i+1] + draw->index_bias) << 16) |
534 (ptr1[i] + draw->index_bias));
535
536 if (draw->count & 1)
537 OUT_CS(ptr1[i] + draw->index_bias);
538 } else {
539 for (i = 0; i < draw->count-1; i += 2)
540 OUT_CS(((ptr1[i+1]) << 16) |
541 (ptr1[i] ));
542
543 if (draw->count & 1)
544 OUT_CS(ptr1[i]);
545 }
546 break;
547
548 case 2:
549 ptr2 = (uint16_t*)info->index.user;
550 ptr2 += draw->start;
551
552 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
553 r300_translate_primitive(info->mode));
554
555 if (draw->index_bias && !r300->screen->caps.is_r500) {
556 for (i = 0; i < draw->count-1; i += 2)
557 OUT_CS(((ptr2[i+1] + draw->index_bias) << 16) |
558 (ptr2[i] + draw->index_bias));
559
560 if (draw->count & 1)
561 OUT_CS(ptr2[i] + draw->index_bias);
562 } else {
563 OUT_CS_TABLE(ptr2, count_dwords);
564 }
565 break;
566
567 case 4:
568 ptr4 = (uint32_t*)info->index.user;
569 ptr4 += draw->start;
570
571 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
572 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
573 r300_translate_primitive(info->mode));
574
575 if (draw->index_bias && !r300->screen->caps.is_r500) {
576 for (i = 0; i < draw->count; i++)
577 OUT_CS(ptr4[i] + draw->index_bias);
578 } else {
579 OUT_CS_TABLE(ptr4, count_dwords);
580 }
581 break;
582 }
583 END_CS;
584 }
585
r300_draw_elements(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,int instance_id)586 static void r300_draw_elements(struct r300_context *r300,
587 const struct pipe_draw_info *info,
588 const struct pipe_draw_start_count_bias *draw,
589 int instance_id)
590 {
591 struct pipe_resource *indexBuffer =
592 info->has_user_indices ? NULL : info->index.resource;
593 unsigned indexSize = info->index_size;
594 struct pipe_resource* orgIndexBuffer = indexBuffer;
595 unsigned start = draw->start;
596 unsigned count = draw->count;
597 boolean alt_num_verts = r300->screen->caps.is_r500 &&
598 count > 65536;
599 unsigned short_count;
600 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
601 uint16_t indices3[3];
602
603 if (draw->index_bias && !r300->screen->caps.is_r500) {
604 r300_split_index_bias(r300, draw->index_bias, &buffer_offset,
605 &index_offset);
606 }
607
608 r300_translate_index_buffer(r300, info, &indexBuffer,
609 &indexSize, index_offset, &start, count);
610
611 /* Fallback for misaligned ushort indices. */
612 if (indexSize == 2 && (start & 1) && indexBuffer) {
613 /* If we got here, then orgIndexBuffer == indexBuffer. */
614 uint16_t *ptr = r300->rws->buffer_map(r300->rws, r300_resource(orgIndexBuffer)->buf,
615 &r300->cs,
616 PIPE_MAP_READ |
617 PIPE_MAP_UNSYNCHRONIZED);
618
619 if (info->mode == PIPE_PRIM_TRIANGLES) {
620 memcpy(indices3, ptr + start, 6);
621 } else {
622 /* Copy the mapped index buffer directly to the upload buffer.
623 * The start index will be aligned simply from the fact that
624 * every sub-buffer in the upload buffer is aligned. */
625 r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
626 count, (uint8_t*)ptr);
627 }
628 } else {
629 if (info->has_user_indices)
630 r300_upload_index_buffer(r300, &indexBuffer, indexSize,
631 &start, count,
632 info->index.user);
633 }
634
635 /* 19 dwords for emit_draw_elements. Give up if the function fails. */
636 if (!r300_prepare_for_rendering(r300,
637 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
638 PREP_INDEXED, indexBuffer, 19, buffer_offset, draw->index_bias,
639 instance_id))
640 goto done;
641
642 if (alt_num_verts || count <= 65535) {
643 r300_emit_draw_elements(r300, indexBuffer, indexSize,
644 info->max_index, info->mode, start, count,
645 indices3);
646 } else {
647 do {
648 /* The maximum must be divisible by 4 and 3,
649 * so that quad and triangle lists are split correctly.
650 *
651 * Strips, loops, and fans won't work. */
652 short_count = MIN2(count, 65532);
653
654 r300_emit_draw_elements(r300, indexBuffer, indexSize,
655 info->max_index,
656 info->mode, start, short_count, indices3);
657
658 start += short_count;
659 count -= short_count;
660
661 /* 15 dwords for emit_draw_elements */
662 if (count) {
663 if (!r300_prepare_for_rendering(r300,
664 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
665 indexBuffer, 19, buffer_offset, draw->index_bias,
666 instance_id))
667 goto done;
668 }
669 } while (count);
670 }
671
672 done:
673 if (indexBuffer != orgIndexBuffer) {
674 pipe_resource_reference( &indexBuffer, NULL );
675 }
676 }
677
r300_draw_arrays(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,int instance_id)678 static void r300_draw_arrays(struct r300_context *r300,
679 const struct pipe_draw_info *info,
680 const struct pipe_draw_start_count_bias *draw,
681 int instance_id)
682 {
683 boolean alt_num_verts = r300->screen->caps.is_r500 &&
684 draw->count > 65536;
685 unsigned start = draw->start;
686 unsigned count = draw->count;
687 unsigned short_count;
688
689 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
690 if (!r300_prepare_for_rendering(r300,
691 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
692 NULL, 9, start, 0, instance_id))
693 return;
694
695 if (alt_num_verts || count <= 65535) {
696 r300_emit_draw_arrays(r300, info->mode, count);
697 } else {
698 do {
699 /* The maximum must be divisible by 4 and 3,
700 * so that quad and triangle lists are split correctly.
701 *
702 * Strips, loops, and fans won't work. */
703 short_count = MIN2(count, 65532);
704 r300_emit_draw_arrays(r300, info->mode, short_count);
705
706 start += short_count;
707 count -= short_count;
708
709 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
710 if (count) {
711 if (!r300_prepare_for_rendering(r300,
712 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
713 start, 0, instance_id))
714 return;
715 }
716 } while (count);
717 }
718 }
719
r300_draw_arrays_instanced(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)720 static void r300_draw_arrays_instanced(struct r300_context *r300,
721 const struct pipe_draw_info *info,
722 const struct pipe_draw_start_count_bias *draw)
723 {
724 int i;
725
726 for (i = 0; i < info->instance_count; i++)
727 r300_draw_arrays(r300, info, draw, i);
728 }
729
r300_draw_elements_instanced(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)730 static void r300_draw_elements_instanced(struct r300_context *r300,
731 const struct pipe_draw_info *info,
732 const struct pipe_draw_start_count_bias *draw)
733 {
734 int i;
735
736 for (i = 0; i < info->instance_count; i++)
737 r300_draw_elements(r300, info, draw, i);
738 }
739
r300_max_vertex_count(struct r300_context * r300)740 static unsigned r300_max_vertex_count(struct r300_context *r300)
741 {
742 unsigned i, nr = r300->velems->count;
743 struct pipe_vertex_element *velems = r300->velems->velem;
744 unsigned result = ~0;
745
746 for (i = 0; i < nr; i++) {
747 struct pipe_vertex_buffer *vb =
748 &r300->vertex_buffer[velems[i].vertex_buffer_index];
749 unsigned size, max_count, value;
750
751 /* We're not interested in constant and per-instance attribs. */
752 if (!vb->buffer.resource ||
753 !vb->stride ||
754 velems[i].instance_divisor) {
755 continue;
756 }
757
758 size = vb->buffer.resource->width0;
759
760 /* Subtract buffer_offset. */
761 value = vb->buffer_offset;
762 if (value >= size) {
763 return 0;
764 }
765 size -= value;
766
767 /* Subtract src_offset. */
768 value = velems[i].src_offset;
769 if (value >= size) {
770 return 0;
771 }
772 size -= value;
773
774 /* Subtract format_size. */
775 value = r300->velems->format_size[i];
776 if (value >= size) {
777 return 0;
778 }
779 size -= value;
780
781 /* Compute the max count. */
782 max_count = 1 + size / vb->stride;
783 result = MIN2(result, max_count);
784 }
785 return result;
786 }
787
788
r300_draw_vbo(struct pipe_context * pipe,const struct pipe_draw_info * dinfo,unsigned drawid_offset,const struct pipe_draw_indirect_info * indirect,const struct pipe_draw_start_count_bias * draws,unsigned num_draws)789 static void r300_draw_vbo(struct pipe_context* pipe,
790 const struct pipe_draw_info *dinfo,
791 unsigned drawid_offset,
792 const struct pipe_draw_indirect_info *indirect,
793 const struct pipe_draw_start_count_bias *draws,
794 unsigned num_draws)
795 {
796 if (num_draws > 1) {
797 util_draw_multi(pipe, dinfo, drawid_offset, indirect, draws, num_draws);
798 return;
799 }
800
801 struct r300_context* r300 = r300_context(pipe);
802 struct pipe_draw_info info = *dinfo;
803 struct pipe_draw_start_count_bias draw = draws[0];
804
805 if (r300->skip_rendering ||
806 !u_trim_pipe_prim(info.mode, &draw.count)) {
807 return;
808 }
809
810 r300_update_derived_state(r300);
811
812 /* Draw. */
813 if (info.index_size) {
814 unsigned max_count = r300_max_vertex_count(r300);
815
816 if (!max_count) {
817 fprintf(stderr, "r300: Skipping a draw command. There is a buffer "
818 " which is too small to be used for rendering.\n");
819 return;
820 }
821
822 if (max_count == ~0) {
823 /* There are no per-vertex vertex elements. Use the hardware maximum. */
824 max_count = 0xffffff;
825 }
826
827 info.max_index = max_count - 1;
828
829 if (info.instance_count <= 1) {
830 if (draw.count <= 8 && info.has_user_indices) {
831 r300_draw_elements_immediate(r300, &info, &draw);
832 } else {
833 r300_draw_elements(r300, &info, &draw, -1);
834 }
835 } else {
836 r300_draw_elements_instanced(r300, &info, &draw);
837 }
838 } else {
839 if (info.instance_count <= 1) {
840 if (immd_is_good_idea(r300, draw.count)) {
841 r300_draw_arrays_immediate(r300, &info, &draw);
842 } else {
843 r300_draw_arrays(r300, &info, &draw, -1);
844 }
845 } else {
846 r300_draw_arrays_instanced(r300, &info, &draw);
847 }
848 }
849 }
850
851 /****************************************************************************
852 * The rest of this file is for SW TCL rendering only. Please be polite and *
853 * keep these functions separated so that they are easier to locate. ~C. *
854 ***************************************************************************/
855
856 /* SW TCL elements, using Draw. */
r300_swtcl_draw_vbo(struct pipe_context * pipe,const struct pipe_draw_info * info,unsigned drawid_offset,const struct pipe_draw_indirect_info * indirect,const struct pipe_draw_start_count_bias * draws,unsigned num_draws)857 static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
858 const struct pipe_draw_info *info,
859 unsigned drawid_offset,
860 const struct pipe_draw_indirect_info *indirect,
861 const struct pipe_draw_start_count_bias *draws,
862 unsigned num_draws)
863 {
864 if (num_draws > 1) {
865 util_draw_multi(pipe, info, drawid_offset, indirect, draws, num_draws);
866 return;
867 }
868
869 struct r300_context* r300 = r300_context(pipe);
870 struct pipe_draw_start_count_bias draw = draws[0];
871
872 if (r300->skip_rendering) {
873 return;
874 }
875
876 if (!u_trim_pipe_prim(info->mode, &draw.count))
877 return;
878
879 if (info->index_size) {
880 draw_set_indexes(r300->draw,
881 info->has_user_indices ?
882 info->index.user :
883 r300_resource(info->index.resource)->malloced_buffer,
884 info->index_size, ~0);
885 }
886
887 r300_update_derived_state(r300);
888
889 draw_vbo(r300->draw, info, drawid_offset, NULL, &draw, 1, 0);
890 draw_flush(r300->draw);
891 }
892
893 /* Object for rendering using Draw. */
894 struct r300_render {
895 /* Parent class */
896 struct vbuf_render base;
897
898 /* Pipe context */
899 struct r300_context* r300;
900
901 /* Vertex information */
902 size_t vertex_size;
903 unsigned prim;
904 unsigned hwprim;
905
906 /* VBO */
907 size_t vbo_max_used;
908 uint8_t *vbo_ptr;
909 };
910
911 static inline struct r300_render*
r300_render(struct vbuf_render * render)912 r300_render(struct vbuf_render* render)
913 {
914 return (struct r300_render*)render;
915 }
916
917 static const struct vertex_info*
r300_render_get_vertex_info(struct vbuf_render * render)918 r300_render_get_vertex_info(struct vbuf_render* render)
919 {
920 struct r300_render* r300render = r300_render(render);
921 struct r300_context* r300 = r300render->r300;
922
923 return &r300->vertex_info;
924 }
925
r300_render_allocate_vertices(struct vbuf_render * render,ushort vertex_size,ushort count)926 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
927 ushort vertex_size,
928 ushort count)
929 {
930 struct r300_render* r300render = r300_render(render);
931 struct r300_context* r300 = r300render->r300;
932 struct radeon_winsys *rws = r300->rws;
933 size_t size = (size_t)vertex_size * (size_t)count;
934
935 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
936
937 if (!r300->vbo || size + r300->draw_vbo_offset > r300->vbo->size) {
938 pb_reference(&r300->vbo, NULL);
939 r300->vbo = NULL;
940 r300render->vbo_ptr = NULL;
941
942 r300->vbo = rws->buffer_create(rws,
943 MAX2(R300_MAX_DRAW_VBO_SIZE, size),
944 R300_BUFFER_ALIGNMENT,
945 RADEON_DOMAIN_GTT,
946 RADEON_FLAG_NO_INTERPROCESS_SHARING);
947 if (!r300->vbo) {
948 return FALSE;
949 }
950 r300->draw_vbo_offset = 0;
951 r300render->vbo_ptr = rws->buffer_map(rws, r300->vbo, &r300->cs,
952 PIPE_MAP_WRITE);
953 }
954
955 r300render->vertex_size = vertex_size;
956 return TRUE;
957 }
958
r300_render_map_vertices(struct vbuf_render * render)959 static void* r300_render_map_vertices(struct vbuf_render* render)
960 {
961 struct r300_render* r300render = r300_render(render);
962 struct r300_context* r300 = r300render->r300;
963
964 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
965
966 assert(r300render->vbo_ptr);
967 return r300render->vbo_ptr + r300->draw_vbo_offset;
968 }
969
r300_render_unmap_vertices(struct vbuf_render * render,ushort min,ushort max)970 static void r300_render_unmap_vertices(struct vbuf_render* render,
971 ushort min,
972 ushort max)
973 {
974 struct r300_render* r300render = r300_render(render);
975 struct r300_context* r300 = r300render->r300;
976
977 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
978
979 r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
980 r300render->vertex_size * (max + 1));
981 }
982
r300_render_release_vertices(struct vbuf_render * render)983 static void r300_render_release_vertices(struct vbuf_render* render)
984 {
985 struct r300_render* r300render = r300_render(render);
986 struct r300_context* r300 = r300render->r300;
987
988 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
989
990 r300->draw_vbo_offset += r300render->vbo_max_used;
991 r300render->vbo_max_used = 0;
992 }
993
r300_render_set_primitive(struct vbuf_render * render,enum pipe_prim_type prim)994 static void r300_render_set_primitive(struct vbuf_render* render,
995 enum pipe_prim_type prim)
996 {
997 struct r300_render* r300render = r300_render(render);
998
999 r300render->prim = prim;
1000 r300render->hwprim = r300_translate_primitive(prim);
1001 }
1002
r300_render_draw_arrays(struct vbuf_render * render,unsigned start,unsigned count)1003 static void r300_render_draw_arrays(struct vbuf_render* render,
1004 unsigned start,
1005 unsigned count)
1006 {
1007 struct r300_render* r300render = r300_render(render);
1008 struct r300_context* r300 = r300render->r300;
1009 uint8_t* ptr;
1010 unsigned i;
1011 unsigned dwords = 6;
1012
1013 CS_LOCALS(r300);
1014 (void) i; (void) ptr;
1015
1016 assert(start == 0);
1017 assert(count < (1 << 16));
1018
1019 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1020
1021 if (!r300_prepare_for_rendering(r300,
1022 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1023 NULL, dwords, 0, 0, -1)) {
1024 return;
1025 }
1026
1027 BEGIN_CS(dwords);
1028 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1029 r300_provoking_vertex_fixes(r300, r300render->prim));
1030 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1031 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1032 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1033 r300render->hwprim);
1034 END_CS;
1035 }
1036
r300_render_draw_elements(struct vbuf_render * render,const ushort * indices,uint count)1037 static void r300_render_draw_elements(struct vbuf_render* render,
1038 const ushort* indices,
1039 uint count)
1040 {
1041 struct r300_render* r300render = r300_render(render);
1042 struct r300_context* r300 = r300render->r300;
1043 unsigned max_index = (r300->vbo->size - r300->draw_vbo_offset) /
1044 (r300render->r300->vertex_info.size * 4) - 1;
1045 struct pipe_resource *index_buffer = NULL;
1046 unsigned index_buffer_offset;
1047
1048 CS_LOCALS(r300);
1049 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1050
1051 u_upload_data(r300->uploader, 0, count * 2, 4, indices,
1052 &index_buffer_offset, &index_buffer);
1053 if (!index_buffer) {
1054 return;
1055 }
1056
1057 if (!r300_prepare_for_rendering(r300,
1058 PREP_EMIT_STATES |
1059 PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1060 index_buffer, 12, 0, 0, -1)) {
1061 pipe_resource_reference(&index_buffer, NULL);
1062 return;
1063 }
1064
1065 BEGIN_CS(12);
1066 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1067 r300_provoking_vertex_fixes(r300, r300render->prim));
1068 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1069
1070 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
1071 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
1072 r300render->hwprim);
1073
1074 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
1075 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2));
1076 OUT_CS(index_buffer_offset);
1077 OUT_CS((count + 1) / 2);
1078 OUT_CS_RELOC(r300_resource(index_buffer));
1079 END_CS;
1080
1081 pipe_resource_reference(&index_buffer, NULL);
1082 }
1083
r300_render_destroy(struct vbuf_render * render)1084 static void r300_render_destroy(struct vbuf_render* render)
1085 {
1086 FREE(render);
1087 }
1088
r300_render_create(struct r300_context * r300)1089 static struct vbuf_render* r300_render_create(struct r300_context* r300)
1090 {
1091 struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1092
1093 r300render->r300 = r300;
1094
1095 r300render->base.max_vertex_buffer_bytes = R300_MAX_DRAW_VBO_SIZE;
1096 r300render->base.max_indices = 16 * 1024;
1097
1098 r300render->base.get_vertex_info = r300_render_get_vertex_info;
1099 r300render->base.allocate_vertices = r300_render_allocate_vertices;
1100 r300render->base.map_vertices = r300_render_map_vertices;
1101 r300render->base.unmap_vertices = r300_render_unmap_vertices;
1102 r300render->base.set_primitive = r300_render_set_primitive;
1103 r300render->base.draw_elements = r300_render_draw_elements;
1104 r300render->base.draw_arrays = r300_render_draw_arrays;
1105 r300render->base.release_vertices = r300_render_release_vertices;
1106 r300render->base.destroy = r300_render_destroy;
1107
1108 return &r300render->base;
1109 }
1110
r300_draw_stage(struct r300_context * r300)1111 struct draw_stage* r300_draw_stage(struct r300_context* r300)
1112 {
1113 struct vbuf_render* render;
1114 struct draw_stage* stage;
1115
1116 render = r300_render_create(r300);
1117
1118 if (!render) {
1119 return NULL;
1120 }
1121
1122 stage = draw_vbuf_stage(r300->draw, render);
1123
1124 if (!stage) {
1125 render->destroy(render);
1126 return NULL;
1127 }
1128
1129 draw_set_render(r300->draw, render);
1130
1131 return stage;
1132 }
1133
1134 /****************************************************************************
1135 * End of SW TCL functions *
1136 ***************************************************************************/
1137
1138 /* This functions is used to draw a rectangle for the blitter module.
1139 *
1140 * If we rendered a quad, the pixels on the main diagonal
1141 * would be computed and stored twice, which makes the clear/copy codepaths
1142 * somewhat inefficient. Instead we use a rectangular point sprite. */
r300_blitter_draw_rectangle(struct blitter_context * blitter,void * vertex_elements_cso,blitter_get_vs_func get_vs,int x1,int y1,int x2,int y2,float depth,unsigned num_instances,enum blitter_attrib_type type,const union blitter_attrib * attrib)1143 void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1144 void *vertex_elements_cso,
1145 blitter_get_vs_func get_vs,
1146 int x1, int y1, int x2, int y2,
1147 float depth, unsigned num_instances,
1148 enum blitter_attrib_type type,
1149 const union blitter_attrib *attrib)
1150 {
1151 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1152 unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1153 unsigned width = x2 - x1;
1154 unsigned height = y2 - y1;
1155 unsigned vertex_size =
1156 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1157 unsigned dwords = 13 + vertex_size +
1158 (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY ? 7 : 0);
1159 static const union blitter_attrib zeros;
1160 CS_LOCALS(r300);
1161
1162 /* XXX workaround for a lockup in MSAA resolve on SWTCL chipsets, this
1163 * function most probably doesn't handle type=NONE correctly */
1164 if ((!r300->screen->caps.has_tcl && type == UTIL_BLITTER_ATTRIB_NONE) ||
1165 type == UTIL_BLITTER_ATTRIB_TEXCOORD_XYZW ||
1166 num_instances > 1) {
1167 util_blitter_draw_rectangle(blitter, vertex_elements_cso, get_vs,
1168 x1, y1, x2, y2,
1169 depth, num_instances, type, attrib);
1170 return;
1171 }
1172
1173 if (r300->skip_rendering)
1174 return;
1175
1176 r300->context.bind_vertex_elements_state(&r300->context, vertex_elements_cso);
1177 r300->context.bind_vs_state(&r300->context, get_vs(blitter));
1178
1179 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY)
1180 r300->sprite_coord_enable = 1;
1181
1182 r300_update_derived_state(r300);
1183
1184 /* Mark some states we don't care about as non-dirty. */
1185 r300->viewport_state.dirty = FALSE;
1186
1187 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1188 goto done;
1189
1190 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1191
1192 BEGIN_CS(dwords);
1193 /* Set up GA. */
1194 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1195
1196 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) {
1197 /* Set up the GA to generate texcoords. */
1198 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1199 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1200 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1201 OUT_CS_32F(attrib->texcoord.x1);
1202 OUT_CS_32F(attrib->texcoord.y2);
1203 OUT_CS_32F(attrib->texcoord.x2);
1204 OUT_CS_32F(attrib->texcoord.y1);
1205 }
1206
1207 /* Set up VAP controls. */
1208 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1209 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1210 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1211 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1212 OUT_CS(1);
1213 OUT_CS(0);
1214
1215 /* Draw. */
1216 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1217 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1218 R300_VAP_VF_CNTL__PRIM_POINTS);
1219
1220 OUT_CS_32F(x1 + width * 0.5f);
1221 OUT_CS_32F(y1 + height * 0.5f);
1222 OUT_CS_32F(depth);
1223 OUT_CS_32F(1);
1224
1225 if (vertex_size == 8) {
1226 if (!attrib)
1227 attrib = &zeros;
1228 OUT_CS_TABLE(attrib->color, 4);
1229 }
1230 END_CS;
1231
1232 done:
1233 /* Restore the state. */
1234 r300_mark_atom_dirty(r300, &r300->rs_state);
1235 r300_mark_atom_dirty(r300, &r300->viewport_state);
1236
1237 r300->sprite_coord_enable = last_sprite_coord_enable;
1238 }
1239
r300_init_render_functions(struct r300_context * r300)1240 void r300_init_render_functions(struct r300_context *r300)
1241 {
1242 /* Set draw functions based on presence of HW TCL. */
1243 if (r300->screen->caps.has_tcl) {
1244 r300->context.draw_vbo = r300_draw_vbo;
1245 } else {
1246 r300->context.draw_vbo = r300_swtcl_draw_vbo;
1247 }
1248
1249 /* Plug in the two-sided stencil reference value fallback if needed. */
1250 if (!r300->screen->caps.is_r500)
1251 r300_plug_in_stencil_ref_fallback(r300);
1252 }
1253