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 #include "r300_vs.h"
44
45 #include <limits.h>
46
47 #define IMMD_DWORDS 32
48
r300_translate_primitive(unsigned prim)49 static uint32_t r300_translate_primitive(unsigned prim)
50 {
51 static const int prim_conv[] = {
52 R300_VAP_VF_CNTL__PRIM_POINTS,
53 R300_VAP_VF_CNTL__PRIM_LINES,
54 R300_VAP_VF_CNTL__PRIM_LINE_LOOP,
55 R300_VAP_VF_CNTL__PRIM_LINE_STRIP,
56 R300_VAP_VF_CNTL__PRIM_TRIANGLES,
57 R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP,
58 R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN,
59 R300_VAP_VF_CNTL__PRIM_QUADS,
60 R300_VAP_VF_CNTL__PRIM_QUAD_STRIP,
61 R300_VAP_VF_CNTL__PRIM_POLYGON,
62 -1,
63 -1,
64 -1,
65 -1
66 };
67 unsigned hwprim = prim_conv[prim];
68
69 assert(hwprim != -1);
70 return hwprim;
71 }
72
r300_provoking_vertex_fixes(struct r300_context * r300,unsigned mode)73 static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
74 unsigned mode)
75 {
76 struct r300_rs_state* rs = (struct r300_rs_state*)r300->rs_state.state;
77 uint32_t color_control = rs->color_control;
78
79 /* By default (see r300_state.c:r300_create_rs_state) color_control is
80 * initialized to provoking the first vertex.
81 *
82 * Triangle fans must be reduced to the second vertex, not the first, in
83 * Gallium flatshade-first mode, as per the GL spec.
84 * (http://www.opengl.org/registry/specs/ARB/provoking_vertex.txt)
85 *
86 * Quads never provoke correctly in flatshade-first mode. The first
87 * vertex is never considered as provoking, so only the second, third,
88 * and fourth vertices can be selected, and both "third" and "last" modes
89 * select the fourth vertex. This is probably due to D3D lacking quads.
90 *
91 * Similarly, polygons reduce to the first, not the last, vertex, when in
92 * "last" mode, and all other modes start from the second vertex.
93 *
94 * ~ C.
95 */
96
97 if (rs->rs.flatshade_first) {
98 switch (mode) {
99 case PIPE_PRIM_TRIANGLE_FAN:
100 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
101 break;
102 case PIPE_PRIM_QUADS:
103 case PIPE_PRIM_QUAD_STRIP:
104 case PIPE_PRIM_POLYGON:
105 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
106 break;
107 default:
108 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_FIRST;
109 break;
110 }
111 } else {
112 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_LAST;
113 }
114
115 return color_control;
116 }
117
r500_emit_index_bias(struct r300_context * r300,int index_bias)118 void r500_emit_index_bias(struct r300_context *r300, int index_bias)
119 {
120 CS_LOCALS(r300);
121
122 BEGIN_CS(2);
123 OUT_CS_REG(R500_VAP_INDEX_OFFSET,
124 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
125 END_CS;
126 }
127
r300_emit_draw_init(struct r300_context * r300,unsigned mode,unsigned max_index)128 static void r300_emit_draw_init(struct r300_context *r300, unsigned mode,
129 unsigned max_index)
130 {
131 CS_LOCALS(r300);
132
133 assert(max_index < (1 << 24));
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(0);
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. */
r300_split_index_bias(struct r300_context * r300,int index_bias,int * buffer_offset,int * index_offset)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->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_EMIT_STATES = (1 << 0), /* call emit_dirty_state and friends? */
181 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */
182 PREP_EMIT_VARRAYS = (1 << 2), /* call emit_vertex_arrays? */
183 PREP_EMIT_VARRAYS_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 */
r300_reserve_cs_dwords(struct r300_context * r300,enum r300_prepare_flags flags,unsigned cs_dwords)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 emit_states = flags & PREP_EMIT_STATES;
201 boolean emit_vertex_arrays = flags & PREP_EMIT_VARRAYS;
202 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
203
204 /* Add dirty state, index offset, and AOS. */
205 if (emit_states)
206 cs_dwords += r300_get_num_dirty_dwords(r300);
207
208 if (r300->screen->caps.is_r500)
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 cs_dwords += r300_get_num_cs_end_dwords(r300);
218
219 /* Reserve requested CS space. */
220 if (!r300->rws->cs_check_space(&r300->cs, cs_dwords)) {
221 r300_flush(&r300->context, PIPE_FLUSH_ASYNC, NULL);
222 flushed = TRUE;
223 }
224
225 return flushed;
226 }
227
228 /**
229 * Validate buffers and emit dirty state.
230 * \param r300 The context.
231 * \param flags See r300_prepare_flags.
232 * \param index_buffer The index buffer to validate. The parameter may be NULL.
233 * \param buffer_offset The offset passed to emit_vertex_arrays.
234 * \param index_bias The index bias to emit.
235 * \param instance_id Index of instance to render
236 * \return TRUE if rendering should be skipped
237 */
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)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, int instance_id)
243 {
244 boolean emit_states = flags & PREP_EMIT_STATES;
245 boolean emit_vertex_arrays = flags & PREP_EMIT_VARRAYS;
246 boolean emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
247 boolean indexed = flags & PREP_INDEXED;
248 boolean validate_vbos = flags & PREP_VALIDATE_VBOS;
249
250 /* Validate buffers and emit dirty state if needed. */
251 if (emit_states || (emit_vertex_arrays && validate_vbos)) {
252 if (!r300_emit_buffer_validate(r300, validate_vbos,
253 index_buffer)) {
254 fprintf(stderr, "r300: CS space validation failed. "
255 "(not enough memory?) Skipping rendering.\n");
256 return FALSE;
257 }
258 }
259
260 if (emit_states)
261 r300_emit_dirty_state(r300);
262
263 if (r300->screen->caps.is_r500) {
264 if (r300->screen->caps.has_tcl)
265 r500_emit_index_bias(r300, index_bias);
266 else
267 r500_emit_index_bias(r300, 0);
268 }
269
270 if (emit_vertex_arrays &&
271 (r300->vertex_arrays_dirty ||
272 r300->vertex_arrays_indexed != indexed ||
273 r300->vertex_arrays_offset != buffer_offset ||
274 r300->vertex_arrays_instance_id != instance_id)) {
275 r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id);
276
277 r300->vertex_arrays_dirty = FALSE;
278 r300->vertex_arrays_indexed = indexed;
279 r300->vertex_arrays_offset = buffer_offset;
280 r300->vertex_arrays_instance_id = instance_id;
281 }
282
283 if (emit_vertex_arrays_swtcl)
284 r300_emit_vertex_arrays_swtcl(r300, indexed);
285
286 return TRUE;
287 }
288
289 /**
290 * Check if the requested number of dwords is available in the CS and
291 * if not, flush. Then validate buffers and emit dirty state.
292 * \param r300 The context.
293 * \param flags See r300_prepare_flags.
294 * \param index_buffer The index buffer to validate. The parameter may be NULL.
295 * \param cs_dwords The number of dwords to reserve in CS.
296 * \param buffer_offset The offset passed to emit_vertex_arrays.
297 * \param index_bias The index bias to emit.
298 * \param instance_id The instance to render.
299 * \return TRUE if rendering should be skipped
300 */
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)301 static boolean r300_prepare_for_rendering(struct r300_context *r300,
302 enum r300_prepare_flags flags,
303 struct pipe_resource *index_buffer,
304 unsigned cs_dwords,
305 int buffer_offset,
306 int index_bias,
307 int instance_id)
308 {
309 /* Make sure there is enough space in the command stream and emit states. */
310 if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
311 flags |= PREP_EMIT_STATES;
312
313 return r300_emit_states(r300, flags, index_buffer, buffer_offset,
314 index_bias, instance_id);
315 }
316
immd_is_good_idea(struct r300_context * r300,unsigned count)317 static boolean immd_is_good_idea(struct r300_context *r300,
318 unsigned count)
319 {
320 if (DBG_ON(r300, DBG_NO_IMMD)) {
321 return FALSE;
322 }
323
324 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
325 return FALSE;
326 }
327
328 /* Buffers can only be used for read by r300 (except query buffers, but
329 * those can't be bound by an gallium frontend as vertex buffers). */
330 return TRUE;
331 }
332
333 /*****************************************************************************
334 * The HWTCL draw functions. *
335 ****************************************************************************/
336
r300_draw_arrays_immediate(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)337 static void r300_draw_arrays_immediate(struct r300_context *r300,
338 const struct pipe_draw_info *info,
339 const struct pipe_draw_start_count_bias *draw)
340 {
341 struct pipe_vertex_element* velem;
342 struct pipe_vertex_buffer* vbuf;
343 unsigned vertex_element_count = r300->velems->count;
344 unsigned i, v, vbi;
345
346 /* Size of the vertex, in dwords. */
347 unsigned vertex_size = r300->velems->vertex_size_dwords;
348
349 /* The number of dwords for this draw operation. */
350 unsigned dwords = 4 + draw->count * vertex_size;
351
352 /* Size of the vertex element, in dwords. */
353 unsigned size[PIPE_MAX_ATTRIBS];
354
355 /* Stride to the same attrib in the next vertex in the vertex buffer,
356 * in dwords. */
357 unsigned stride[PIPE_MAX_ATTRIBS];
358
359 /* Mapped vertex buffers. */
360 uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
361 uint32_t* mapelem[PIPE_MAX_ATTRIBS];
362
363 CS_LOCALS(r300);
364
365 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
366 return;
367
368 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
369 for (i = 0; i < vertex_element_count; i++) {
370 velem = &r300->velems->velem[i];
371 size[i] = r300->velems->format_size[i] / 4;
372 vbi = velem->vertex_buffer_index;
373 vbuf = &r300->vertex_buffer[vbi];
374 stride[i] = vbuf->stride / 4;
375
376 /* Map the buffer. */
377 if (!map[vbi]) {
378 map[vbi] = (uint32_t*)r300->rws->buffer_map(r300->rws,
379 r300_resource(vbuf->buffer.resource)->buf,
380 &r300->cs, PIPE_MAP_READ | PIPE_MAP_UNSYNCHRONIZED);
381 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * draw->start;
382 }
383 mapelem[i] = map[vbi] + (velem->src_offset / 4);
384 }
385
386 r300_emit_draw_init(r300, info->mode, draw->count-1);
387
388 BEGIN_CS(dwords);
389 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
390 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, draw->count * vertex_size);
391 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (draw->count << 16) |
392 r300_translate_primitive(info->mode));
393
394 /* Emit vertices. */
395 for (v = 0; v < draw->count; v++) {
396 for (i = 0; i < vertex_element_count; i++) {
397 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
398 }
399 }
400 END_CS;
401 }
402
r300_emit_draw_arrays(struct r300_context * r300,unsigned mode,unsigned count)403 static void r300_emit_draw_arrays(struct r300_context *r300,
404 unsigned mode,
405 unsigned count)
406 {
407 boolean alt_num_verts = count > 65535;
408 CS_LOCALS(r300);
409
410 if (count >= (1 << 24)) {
411 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
412 "refusing to render.\n", count);
413 return;
414 }
415
416 r300_emit_draw_init(r300, mode, count-1);
417
418 BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
419 if (alt_num_verts) {
420 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
421 }
422 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
423 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
424 r300_translate_primitive(mode) |
425 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
426 END_CS;
427 }
428
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)429 static void r300_emit_draw_elements(struct r300_context *r300,
430 struct pipe_resource* indexBuffer,
431 unsigned indexSize,
432 unsigned max_index,
433 unsigned mode,
434 unsigned start,
435 unsigned count,
436 uint16_t *imm_indices3)
437 {
438 uint32_t count_dwords, offset_dwords;
439 boolean alt_num_verts = count > 65535;
440 CS_LOCALS(r300);
441
442 if (count >= (1 << 24)) {
443 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
444 "refusing to render (max_index: %i).\n", count, max_index);
445 return;
446 }
447
448 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, max %u\n",
449 count, max_index);
450
451 r300_emit_draw_init(r300, mode, max_index);
452
453 /* If start is odd, render the first triangle with indices embedded
454 * in the command stream. This will increase start by 3 and make it
455 * even. We can then proceed without a fallback. */
456 if (indexSize == 2 && (start & 1) &&
457 mode == PIPE_PRIM_TRIANGLES) {
458 BEGIN_CS(4);
459 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
460 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
461 R300_VAP_VF_CNTL__PRIM_TRIANGLES);
462 OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
463 OUT_CS(imm_indices3[2]);
464 END_CS;
465
466 start += 3;
467 count -= 3;
468 if (!count)
469 return;
470 }
471
472 offset_dwords = indexSize * start / sizeof(uint32_t);
473
474 BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
475 if (alt_num_verts) {
476 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
477 }
478 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
479 if (indexSize == 4) {
480 count_dwords = count;
481 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
482 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
483 r300_translate_primitive(mode) |
484 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
485 } else {
486 count_dwords = (count + 1) / 2;
487 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
488 r300_translate_primitive(mode) |
489 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
490 }
491
492 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
493 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
494 (0 << R300_INDX_BUFFER_SKIP_SHIFT));
495 OUT_CS(offset_dwords << 2);
496 OUT_CS(count_dwords);
497 OUT_CS_RELOC(r300_resource(indexBuffer));
498 END_CS;
499 }
500
r300_draw_elements_immediate(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)501 static void r300_draw_elements_immediate(struct r300_context *r300,
502 const struct pipe_draw_info *info,
503 const struct pipe_draw_start_count_bias *draw)
504 {
505 const uint8_t *ptr1;
506 const uint16_t *ptr2;
507 const uint32_t *ptr4;
508 unsigned index_size = info->index_size;
509 unsigned i, count_dwords = index_size == 4 ? draw->count :
510 (draw->count + 1) / 2;
511 CS_LOCALS(r300);
512
513 /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
514 if (!r300_prepare_for_rendering(r300,
515 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
516 PREP_INDEXED, NULL, 2+count_dwords, 0, draw->index_bias, -1))
517 return;
518
519 r300_emit_draw_init(r300, info->mode, info->max_index);
520
521 BEGIN_CS(2 + count_dwords);
522 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
523
524 switch (index_size) {
525 case 1:
526 ptr1 = (uint8_t*)info->index.user;
527 ptr1 += draw->start;
528
529 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
530 r300_translate_primitive(info->mode));
531
532 if (draw->index_bias && !r300->screen->caps.is_r500) {
533 for (i = 0; i < draw->count-1; i += 2)
534 OUT_CS(((ptr1[i+1] + draw->index_bias) << 16) |
535 (ptr1[i] + draw->index_bias));
536
537 if (draw->count & 1)
538 OUT_CS(ptr1[i] + draw->index_bias);
539 } else {
540 for (i = 0; i < draw->count-1; i += 2)
541 OUT_CS(((ptr1[i+1]) << 16) |
542 (ptr1[i] ));
543
544 if (draw->count & 1)
545 OUT_CS(ptr1[i]);
546 }
547 break;
548
549 case 2:
550 ptr2 = (uint16_t*)info->index.user;
551 ptr2 += draw->start;
552
553 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
554 r300_translate_primitive(info->mode));
555
556 if (draw->index_bias && !r300->screen->caps.is_r500) {
557 for (i = 0; i < draw->count-1; i += 2)
558 OUT_CS(((ptr2[i+1] + draw->index_bias) << 16) |
559 (ptr2[i] + draw->index_bias));
560
561 if (draw->count & 1)
562 OUT_CS(ptr2[i] + draw->index_bias);
563 } else {
564 OUT_CS_TABLE(ptr2, count_dwords);
565 }
566 break;
567
568 case 4:
569 ptr4 = (uint32_t*)info->index.user;
570 ptr4 += draw->start;
571
572 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
573 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
574 r300_translate_primitive(info->mode));
575
576 if (draw->index_bias && !r300->screen->caps.is_r500) {
577 for (i = 0; i < draw->count; i++)
578 OUT_CS(ptr4[i] + draw->index_bias);
579 } else {
580 OUT_CS_TABLE(ptr4, count_dwords);
581 }
582 break;
583 }
584 END_CS;
585 }
586
r300_draw_elements(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,int instance_id)587 static void r300_draw_elements(struct r300_context *r300,
588 const struct pipe_draw_info *info,
589 const struct pipe_draw_start_count_bias *draw,
590 int instance_id)
591 {
592 struct pipe_resource *indexBuffer =
593 info->has_user_indices ? NULL : info->index.resource;
594 unsigned indexSize = info->index_size;
595 struct pipe_resource* orgIndexBuffer = indexBuffer;
596 unsigned start = draw->start;
597 unsigned count = draw->count;
598 boolean alt_num_verts = r300->screen->caps.is_r500 &&
599 count > 65536;
600 unsigned short_count;
601 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
602 uint16_t indices3[3];
603
604 if (draw->index_bias && !r300->screen->caps.is_r500) {
605 r300_split_index_bias(r300, draw->index_bias, &buffer_offset,
606 &index_offset);
607 }
608
609 r300_translate_index_buffer(r300, info, &indexBuffer,
610 &indexSize, index_offset, &start, count);
611
612 /* Fallback for misaligned ushort indices. */
613 if (indexSize == 2 && (start & 1) && indexBuffer) {
614 /* If we got here, then orgIndexBuffer == indexBuffer. */
615 uint16_t *ptr = r300->rws->buffer_map(r300->rws, r300_resource(orgIndexBuffer)->buf,
616 &r300->cs,
617 PIPE_MAP_READ |
618 PIPE_MAP_UNSYNCHRONIZED);
619
620 if (info->mode == PIPE_PRIM_TRIANGLES) {
621 memcpy(indices3, ptr + start, 6);
622 } else {
623 /* Copy the mapped index buffer directly to the upload buffer.
624 * The start index will be aligned simply from the fact that
625 * every sub-buffer in the upload buffer is aligned. */
626 r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
627 count, (uint8_t*)ptr);
628 }
629 } else {
630 if (info->has_user_indices)
631 r300_upload_index_buffer(r300, &indexBuffer, indexSize,
632 &start, count,
633 info->index.user);
634 }
635
636 /* 19 dwords for emit_draw_elements. Give up if the function fails. */
637 if (!r300_prepare_for_rendering(r300,
638 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
639 PREP_INDEXED, indexBuffer, 19, buffer_offset, draw->index_bias,
640 instance_id))
641 goto done;
642
643 if (alt_num_verts || count <= 65535) {
644 r300_emit_draw_elements(r300, indexBuffer, indexSize,
645 info->max_index, info->mode, start, count,
646 indices3);
647 } else {
648 do {
649 /* The maximum must be divisible by 4 and 3,
650 * so that quad and triangle lists are split correctly.
651 *
652 * Strips, loops, and fans won't work. */
653 short_count = MIN2(count, 65532);
654
655 r300_emit_draw_elements(r300, indexBuffer, indexSize,
656 info->max_index,
657 info->mode, start, short_count, indices3);
658
659 start += short_count;
660 count -= short_count;
661
662 /* 15 dwords for emit_draw_elements */
663 if (count) {
664 if (!r300_prepare_for_rendering(r300,
665 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
666 indexBuffer, 19, buffer_offset, draw->index_bias,
667 instance_id))
668 goto done;
669 }
670 } while (count);
671 }
672
673 done:
674 if (indexBuffer != orgIndexBuffer) {
675 pipe_resource_reference( &indexBuffer, NULL );
676 }
677 }
678
r300_draw_arrays(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,int instance_id)679 static void r300_draw_arrays(struct r300_context *r300,
680 const struct pipe_draw_info *info,
681 const struct pipe_draw_start_count_bias *draw,
682 int instance_id)
683 {
684 boolean alt_num_verts = r300->screen->caps.is_r500 &&
685 draw->count > 65536;
686 unsigned start = draw->start;
687 unsigned count = draw->count;
688 unsigned short_count;
689
690 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
691 if (!r300_prepare_for_rendering(r300,
692 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
693 NULL, 9, start, 0, instance_id))
694 return;
695
696 if (alt_num_verts || count <= 65535) {
697 r300_emit_draw_arrays(r300, info->mode, count);
698 } else {
699 do {
700 /* The maximum must be divisible by 4 and 3,
701 * so that quad and triangle lists are split correctly.
702 *
703 * Strips, loops, and fans won't work. */
704 short_count = MIN2(count, 65532);
705 r300_emit_draw_arrays(r300, info->mode, short_count);
706
707 start += short_count;
708 count -= short_count;
709
710 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
711 if (count) {
712 if (!r300_prepare_for_rendering(r300,
713 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
714 start, 0, instance_id))
715 return;
716 }
717 } while (count);
718 }
719 }
720
r300_draw_arrays_instanced(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)721 static void r300_draw_arrays_instanced(struct r300_context *r300,
722 const struct pipe_draw_info *info,
723 const struct pipe_draw_start_count_bias *draw)
724 {
725 int i;
726
727 for (i = 0; i < info->instance_count; i++)
728 r300_draw_arrays(r300, info, draw, i);
729 }
730
r300_draw_elements_instanced(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)731 static void r300_draw_elements_instanced(struct r300_context *r300,
732 const struct pipe_draw_info *info,
733 const struct pipe_draw_start_count_bias *draw)
734 {
735 int i;
736
737 for (i = 0; i < info->instance_count; i++)
738 r300_draw_elements(r300, info, draw, i);
739 }
740
r300_max_vertex_count(struct r300_context * r300)741 static unsigned r300_max_vertex_count(struct r300_context *r300)
742 {
743 unsigned i, nr = r300->velems->count;
744 struct pipe_vertex_element *velems = r300->velems->velem;
745 unsigned result = ~0;
746
747 for (i = 0; i < nr; i++) {
748 struct pipe_vertex_buffer *vb =
749 &r300->vertex_buffer[velems[i].vertex_buffer_index];
750 unsigned size, max_count, value;
751
752 /* We're not interested in constant and per-instance attribs. */
753 if (!vb->buffer.resource ||
754 !vb->stride ||
755 velems[i].instance_divisor) {
756 continue;
757 }
758
759 size = vb->buffer.resource->width0;
760
761 /* Subtract buffer_offset. */
762 value = vb->buffer_offset;
763 if (value >= size) {
764 return 0;
765 }
766 size -= value;
767
768 /* Subtract src_offset. */
769 value = velems[i].src_offset;
770 if (value >= size) {
771 return 0;
772 }
773 size -= value;
774
775 /* Subtract format_size. */
776 value = r300->velems->format_size[i];
777 if (value >= size) {
778 return 0;
779 }
780 size -= value;
781
782 /* Compute the max count. */
783 max_count = 1 + size / vb->stride;
784 result = MIN2(result, max_count);
785 }
786 return result;
787 }
788
789
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)790 static void r300_draw_vbo(struct pipe_context* pipe,
791 const struct pipe_draw_info *dinfo,
792 unsigned drawid_offset,
793 const struct pipe_draw_indirect_info *indirect,
794 const struct pipe_draw_start_count_bias *draws,
795 unsigned num_draws)
796 {
797 if (num_draws > 1) {
798 util_draw_multi(pipe, dinfo, drawid_offset, indirect, draws, num_draws);
799 return;
800 }
801
802 struct r300_context* r300 = r300_context(pipe);
803 struct pipe_draw_info info = *dinfo;
804 struct pipe_draw_start_count_bias draw = draws[0];
805
806 if (r300->skip_rendering ||
807 !u_trim_pipe_prim(info.mode, &draw.count)) {
808 return;
809 }
810
811 if (r300->sprite_coord_enable != 0)
812 if ((info.mode == PIPE_PRIM_POINTS) != r300->is_point) {
813 r300->is_point = !r300->is_point;
814 r300_mark_atom_dirty(r300, &r300->rs_block_state);
815 }
816
817 r300_update_derived_state(r300);
818
819 /* Skip draw if we failed to compile the vertex shader. */
820 if (r300_vs(r300)->shader->dummy)
821 return;
822
823 /* Draw. */
824 if (info.index_size) {
825 unsigned max_count = r300_max_vertex_count(r300);
826
827 if (!max_count) {
828 fprintf(stderr, "r300: Skipping a draw command. There is a buffer "
829 " which is too small to be used for rendering.\n");
830 return;
831 }
832
833 if (max_count == ~0) {
834 /* There are no per-vertex vertex elements. Use the hardware maximum. */
835 max_count = 0xffffff;
836 }
837
838 info.max_index = max_count - 1;
839
840 if (info.instance_count <= 1) {
841 if (draw.count <= 8 && info.has_user_indices) {
842 r300_draw_elements_immediate(r300, &info, &draw);
843 } else {
844 r300_draw_elements(r300, &info, &draw, -1);
845 }
846 } else {
847 r300_draw_elements_instanced(r300, &info, &draw);
848 }
849 } else {
850 if (info.instance_count <= 1) {
851 if (immd_is_good_idea(r300, draw.count)) {
852 r300_draw_arrays_immediate(r300, &info, &draw);
853 } else {
854 r300_draw_arrays(r300, &info, &draw, -1);
855 }
856 } else {
857 r300_draw_arrays_instanced(r300, &info, &draw);
858 }
859 }
860 }
861
862 /****************************************************************************
863 * The rest of this file is for SW TCL rendering only. Please be polite and *
864 * keep these functions separated so that they are easier to locate. ~C. *
865 ***************************************************************************/
866
867 /* 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)868 static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
869 const struct pipe_draw_info *info,
870 unsigned drawid_offset,
871 const struct pipe_draw_indirect_info *indirect,
872 const struct pipe_draw_start_count_bias *draws,
873 unsigned num_draws)
874 {
875 if (num_draws > 1) {
876 util_draw_multi(pipe, info, drawid_offset, indirect, draws, num_draws);
877 return;
878 }
879
880 struct r300_context* r300 = r300_context(pipe);
881 struct pipe_draw_start_count_bias draw = draws[0];
882
883 if (r300->skip_rendering) {
884 return;
885 }
886
887 if (!u_trim_pipe_prim(info->mode, &draw.count))
888 return;
889
890 if (info->index_size) {
891 draw_set_indexes(r300->draw,
892 info->has_user_indices ?
893 info->index.user :
894 r300_resource(info->index.resource)->malloced_buffer,
895 info->index_size, ~0);
896 }
897
898 if (r300->sprite_coord_enable != 0)
899 if ((info->mode == PIPE_PRIM_POINTS) != r300->is_point) {
900 r300->is_point = !r300->is_point;
901 r300_mark_atom_dirty(r300, &r300->rs_block_state);
902 }
903
904 r300_update_derived_state(r300);
905
906 draw_vbo(r300->draw, info, drawid_offset, NULL, &draw, 1, 0);
907 draw_flush(r300->draw);
908 }
909
910 /* Object for rendering using Draw. */
911 struct r300_render {
912 /* Parent class */
913 struct vbuf_render base;
914
915 /* Pipe context */
916 struct r300_context* r300;
917
918 /* Vertex information */
919 size_t vertex_size;
920 unsigned prim;
921 unsigned hwprim;
922
923 /* VBO */
924 size_t vbo_max_used;
925 uint8_t *vbo_ptr;
926 };
927
928 static inline struct r300_render*
r300_render(struct vbuf_render * render)929 r300_render(struct vbuf_render* render)
930 {
931 return (struct r300_render*)render;
932 }
933
934 static const struct vertex_info*
r300_render_get_vertex_info(struct vbuf_render * render)935 r300_render_get_vertex_info(struct vbuf_render* render)
936 {
937 struct r300_render* r300render = r300_render(render);
938 struct r300_context* r300 = r300render->r300;
939
940 return &r300->vertex_info;
941 }
942
r300_render_allocate_vertices(struct vbuf_render * render,ushort vertex_size,ushort count)943 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
944 ushort vertex_size,
945 ushort count)
946 {
947 struct r300_render* r300render = r300_render(render);
948 struct r300_context* r300 = r300render->r300;
949 struct radeon_winsys *rws = r300->rws;
950 size_t size = (size_t)vertex_size * (size_t)count;
951
952 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
953
954 if (!r300->vbo || size + r300->draw_vbo_offset > r300->vbo->size) {
955 pb_reference(&r300->vbo, NULL);
956 r300->vbo = NULL;
957 r300render->vbo_ptr = NULL;
958
959 r300->vbo = rws->buffer_create(rws,
960 MAX2(R300_MAX_DRAW_VBO_SIZE, size),
961 R300_BUFFER_ALIGNMENT,
962 RADEON_DOMAIN_GTT,
963 RADEON_FLAG_NO_INTERPROCESS_SHARING);
964 if (!r300->vbo) {
965 return FALSE;
966 }
967 r300->draw_vbo_offset = 0;
968 r300render->vbo_ptr = rws->buffer_map(rws, r300->vbo, &r300->cs,
969 PIPE_MAP_WRITE);
970 }
971
972 r300render->vertex_size = vertex_size;
973 return TRUE;
974 }
975
r300_render_map_vertices(struct vbuf_render * render)976 static void* r300_render_map_vertices(struct vbuf_render* render)
977 {
978 struct r300_render* r300render = r300_render(render);
979 struct r300_context* r300 = r300render->r300;
980
981 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
982
983 assert(r300render->vbo_ptr);
984 return r300render->vbo_ptr + r300->draw_vbo_offset;
985 }
986
r300_render_unmap_vertices(struct vbuf_render * render,ushort min,ushort max)987 static void r300_render_unmap_vertices(struct vbuf_render* render,
988 ushort min,
989 ushort max)
990 {
991 struct r300_render* r300render = r300_render(render);
992 struct r300_context* r300 = r300render->r300;
993
994 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
995
996 r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
997 r300render->vertex_size * (max + 1));
998 }
999
r300_render_release_vertices(struct vbuf_render * render)1000 static void r300_render_release_vertices(struct vbuf_render* render)
1001 {
1002 struct r300_render* r300render = r300_render(render);
1003 struct r300_context* r300 = r300render->r300;
1004
1005 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
1006
1007 r300->draw_vbo_offset += r300render->vbo_max_used;
1008 r300render->vbo_max_used = 0;
1009 }
1010
r300_render_set_primitive(struct vbuf_render * render,enum pipe_prim_type prim)1011 static void r300_render_set_primitive(struct vbuf_render* render,
1012 enum pipe_prim_type prim)
1013 {
1014 struct r300_render* r300render = r300_render(render);
1015
1016 r300render->prim = prim;
1017 r300render->hwprim = r300_translate_primitive(prim);
1018 }
1019
r300_render_draw_arrays(struct vbuf_render * render,unsigned start,unsigned count)1020 static void r300_render_draw_arrays(struct vbuf_render* render,
1021 unsigned start,
1022 unsigned count)
1023 {
1024 struct r300_render* r300render = r300_render(render);
1025 struct r300_context* r300 = r300render->r300;
1026 uint8_t* ptr;
1027 unsigned i;
1028 unsigned dwords = 6;
1029
1030 CS_LOCALS(r300);
1031 (void) i; (void) ptr;
1032
1033 assert(start == 0);
1034 assert(count < (1 << 16));
1035
1036 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1037
1038 if (!r300_prepare_for_rendering(r300,
1039 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1040 NULL, dwords, 0, 0, -1)) {
1041 return;
1042 }
1043
1044 BEGIN_CS(dwords);
1045 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1046 r300_provoking_vertex_fixes(r300, r300render->prim));
1047 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1048 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1049 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1050 r300render->hwprim);
1051 END_CS;
1052 }
1053
r300_render_draw_elements(struct vbuf_render * render,const ushort * indices,uint count)1054 static void r300_render_draw_elements(struct vbuf_render* render,
1055 const ushort* indices,
1056 uint count)
1057 {
1058 struct r300_render* r300render = r300_render(render);
1059 struct r300_context* r300 = r300render->r300;
1060 unsigned max_index = (r300->vbo->size - r300->draw_vbo_offset) /
1061 (r300render->r300->vertex_info.size * 4) - 1;
1062 struct pipe_resource *index_buffer = NULL;
1063 unsigned index_buffer_offset;
1064
1065 CS_LOCALS(r300);
1066 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1067
1068 u_upload_data(r300->uploader, 0, count * 2, 4, indices,
1069 &index_buffer_offset, &index_buffer);
1070 if (!index_buffer) {
1071 return;
1072 }
1073
1074 if (!r300_prepare_for_rendering(r300,
1075 PREP_EMIT_STATES |
1076 PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1077 index_buffer, 12, 0, 0, -1)) {
1078 pipe_resource_reference(&index_buffer, NULL);
1079 return;
1080 }
1081
1082 BEGIN_CS(12);
1083 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1084 r300_provoking_vertex_fixes(r300, r300render->prim));
1085 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1086
1087 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
1088 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
1089 r300render->hwprim);
1090
1091 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
1092 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2));
1093 OUT_CS(index_buffer_offset);
1094 OUT_CS((count + 1) / 2);
1095 OUT_CS_RELOC(r300_resource(index_buffer));
1096 END_CS;
1097
1098 pipe_resource_reference(&index_buffer, NULL);
1099 }
1100
r300_render_destroy(struct vbuf_render * render)1101 static void r300_render_destroy(struct vbuf_render* render)
1102 {
1103 FREE(render);
1104 }
1105
r300_render_create(struct r300_context * r300)1106 static struct vbuf_render* r300_render_create(struct r300_context* r300)
1107 {
1108 struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1109
1110 r300render->r300 = r300;
1111
1112 r300render->base.max_vertex_buffer_bytes = R300_MAX_DRAW_VBO_SIZE;
1113 r300render->base.max_indices = 16 * 1024;
1114
1115 r300render->base.get_vertex_info = r300_render_get_vertex_info;
1116 r300render->base.allocate_vertices = r300_render_allocate_vertices;
1117 r300render->base.map_vertices = r300_render_map_vertices;
1118 r300render->base.unmap_vertices = r300_render_unmap_vertices;
1119 r300render->base.set_primitive = r300_render_set_primitive;
1120 r300render->base.draw_elements = r300_render_draw_elements;
1121 r300render->base.draw_arrays = r300_render_draw_arrays;
1122 r300render->base.release_vertices = r300_render_release_vertices;
1123 r300render->base.destroy = r300_render_destroy;
1124
1125 return &r300render->base;
1126 }
1127
r300_draw_stage(struct r300_context * r300)1128 struct draw_stage* r300_draw_stage(struct r300_context* r300)
1129 {
1130 struct vbuf_render* render;
1131 struct draw_stage* stage;
1132
1133 render = r300_render_create(r300);
1134
1135 if (!render) {
1136 return NULL;
1137 }
1138
1139 stage = draw_vbuf_stage(r300->draw, render);
1140
1141 if (!stage) {
1142 render->destroy(render);
1143 return NULL;
1144 }
1145
1146 draw_set_render(r300->draw, render);
1147
1148 return stage;
1149 }
1150
1151 /****************************************************************************
1152 * End of SW TCL functions *
1153 ***************************************************************************/
1154
1155 /* This functions is used to draw a rectangle for the blitter module.
1156 *
1157 * If we rendered a quad, the pixels on the main diagonal
1158 * would be computed and stored twice, which makes the clear/copy codepaths
1159 * 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)1160 void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1161 void *vertex_elements_cso,
1162 blitter_get_vs_func get_vs,
1163 int x1, int y1, int x2, int y2,
1164 float depth, unsigned num_instances,
1165 enum blitter_attrib_type type,
1166 const union blitter_attrib *attrib)
1167 {
1168 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1169 unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1170 unsigned last_is_point = r300->is_point;
1171 unsigned width = x2 - x1;
1172 unsigned height = y2 - y1;
1173 unsigned vertex_size =
1174 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1175 unsigned dwords = 13 + vertex_size +
1176 (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY ? 7 : 0);
1177 static const union blitter_attrib zeros;
1178 CS_LOCALS(r300);
1179
1180 /* XXX workaround for a lockup in MSAA resolve on SWTCL chipsets, this
1181 * function most probably doesn't handle type=NONE correctly */
1182 if ((!r300->screen->caps.has_tcl && type == UTIL_BLITTER_ATTRIB_NONE) ||
1183 type == UTIL_BLITTER_ATTRIB_TEXCOORD_XYZW ||
1184 num_instances > 1) {
1185 util_blitter_draw_rectangle(blitter, vertex_elements_cso, get_vs,
1186 x1, y1, x2, y2,
1187 depth, num_instances, type, attrib);
1188 return;
1189 }
1190
1191 if (r300->skip_rendering)
1192 return;
1193
1194 r300->context.bind_vertex_elements_state(&r300->context, vertex_elements_cso);
1195 r300->context.bind_vs_state(&r300->context, get_vs(blitter));
1196
1197 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) {
1198 r300->sprite_coord_enable = 1;
1199 r300->is_point = true;
1200 }
1201
1202 r300_update_derived_state(r300);
1203
1204 /* Mark some states we don't care about as non-dirty. */
1205 r300->viewport_state.dirty = FALSE;
1206
1207 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1208 goto done;
1209
1210 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1211
1212 BEGIN_CS(dwords);
1213 /* Set up GA. */
1214 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1215
1216 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) {
1217 /* Set up the GA to generate texcoords. */
1218 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1219 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1220 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1221 OUT_CS_32F(attrib->texcoord.x1);
1222 OUT_CS_32F(attrib->texcoord.y2);
1223 OUT_CS_32F(attrib->texcoord.x2);
1224 OUT_CS_32F(attrib->texcoord.y1);
1225 }
1226
1227 /* Set up VAP controls. */
1228 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1229 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1230 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1231 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1232 OUT_CS(1);
1233 OUT_CS(0);
1234
1235 /* Draw. */
1236 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1237 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1238 R300_VAP_VF_CNTL__PRIM_POINTS);
1239
1240 OUT_CS_32F(x1 + width * 0.5f);
1241 OUT_CS_32F(y1 + height * 0.5f);
1242 OUT_CS_32F(depth);
1243 OUT_CS_32F(1);
1244
1245 if (vertex_size == 8) {
1246 if (!attrib)
1247 attrib = &zeros;
1248 OUT_CS_TABLE(attrib->color, 4);
1249 }
1250 END_CS;
1251
1252 done:
1253 /* Restore the state. */
1254 r300_mark_atom_dirty(r300, &r300->rs_state);
1255 r300_mark_atom_dirty(r300, &r300->viewport_state);
1256
1257 r300->sprite_coord_enable = last_sprite_coord_enable;
1258 r300->is_point = last_is_point;
1259 }
1260
r300_init_render_functions(struct r300_context * r300)1261 void r300_init_render_functions(struct r300_context *r300)
1262 {
1263 /* Set draw functions based on presence of HW TCL. */
1264 if (r300->screen->caps.has_tcl) {
1265 r300->context.draw_vbo = r300_draw_vbo;
1266 } else {
1267 r300->context.draw_vbo = r300_swtcl_draw_vbo;
1268 }
1269
1270 /* Plug in the two-sided stencil reference value fallback if needed. */
1271 if (!r300->screen->caps.is_r500)
1272 r300_plug_in_stencil_ref_fallback(r300);
1273 }
1274