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 #include "r300_fs.h"
45
46 #include <limits.h>
47
48 #define IMMD_DWORDS 32
49
r300_translate_primitive(unsigned prim)50 static uint32_t r300_translate_primitive(unsigned prim)
51 {
52 static const int prim_conv[] = {
53 R300_VAP_VF_CNTL__PRIM_POINTS,
54 R300_VAP_VF_CNTL__PRIM_LINES,
55 R300_VAP_VF_CNTL__PRIM_LINE_LOOP,
56 R300_VAP_VF_CNTL__PRIM_LINE_STRIP,
57 R300_VAP_VF_CNTL__PRIM_TRIANGLES,
58 R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP,
59 R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN,
60 R300_VAP_VF_CNTL__PRIM_QUADS,
61 R300_VAP_VF_CNTL__PRIM_QUAD_STRIP,
62 R300_VAP_VF_CNTL__PRIM_POLYGON,
63 -1,
64 -1,
65 -1,
66 -1
67 };
68 unsigned hwprim = prim_conv[prim];
69
70 assert(hwprim != -1);
71 return hwprim;
72 }
73
r300_provoking_vertex_fixes(struct r300_context * r300,unsigned mode)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 MESA_PRIM_TRIANGLE_FAN:
101 color_control |= R300_GA_COLOR_CONTROL_PROVOKING_VERTEX_SECOND;
102 break;
103 case MESA_PRIM_QUADS:
104 case MESA_PRIM_QUAD_STRIP:
105 case MESA_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
r500_emit_index_bias(struct r300_context * r300,int index_bias)119 void r500_emit_index_bias(struct r300_context *r300, int index_bias)
120 {
121 CS_LOCALS(r300);
122
123 BEGIN_CS(2);
124 OUT_CS_REG(R500_VAP_INDEX_OFFSET,
125 (index_bias & 0xFFFFFF) | (index_bias < 0 ? 1<<24 : 0));
126 END_CS;
127 }
128
r300_emit_draw_init(struct r300_context * r300,unsigned mode,unsigned max_index)129 static void r300_emit_draw_init(struct r300_context *r300, unsigned mode,
130 unsigned max_index)
131 {
132 CS_LOCALS(r300);
133
134 assert(max_index < (1 << 24));
135
136 BEGIN_CS(5);
137 OUT_CS_REG(R300_GA_COLOR_CONTROL,
138 r300_provoking_vertex_fixes(r300, mode));
139 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
140 OUT_CS(max_index);
141 OUT_CS(0);
142 END_CS;
143 }
144
145 /* This function splits the index bias value into two parts:
146 * - buffer_offset: the value that can be safely added to buffer offsets
147 * in r300_emit_vertex_arrays (it must yield a positive offset when added to
148 * a vertex buffer offset)
149 * - index_offset: the value that must be manually subtracted from indices
150 * 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)151 static void r300_split_index_bias(struct r300_context *r300, int index_bias,
152 int *buffer_offset, int *index_offset)
153 {
154 struct pipe_vertex_buffer *vb, *vbufs = r300->vertex_buffer;
155 struct pipe_vertex_element *velem = r300->velems->velem;
156 unsigned i, size;
157 int max_neg_bias;
158
159 if (index_bias < 0) {
160 /* See how large index bias we may subtract. We must be careful
161 * here because negative buffer offsets are not allowed
162 * by the DRM API. */
163 max_neg_bias = INT_MAX;
164 for (i = 0; i < r300->velems->count; i++) {
165 vb = &vbufs[velem[i].vertex_buffer_index];
166 size = (vb->buffer_offset + velem[i].src_offset) / velem[i].src_stride;
167 max_neg_bias = MIN2(max_neg_bias, size);
168 }
169
170 /* Now set the minimum allowed value. */
171 *buffer_offset = MAX2(-max_neg_bias, index_bias);
172 } else {
173 /* A positive index bias is OK. */
174 *buffer_offset = index_bias;
175 }
176
177 *index_offset = index_bias - *buffer_offset;
178 }
179
180 enum r300_prepare_flags {
181 PREP_EMIT_STATES = (1 << 0), /* call emit_dirty_state and friends? */
182 PREP_VALIDATE_VBOS = (1 << 1), /* validate VBOs? */
183 PREP_EMIT_VARRAYS = (1 << 2), /* call emit_vertex_arrays? */
184 PREP_EMIT_VARRAYS_SWTCL = (1 << 3), /* call emit_vertex_arrays_swtcl? */
185 PREP_INDEXED = (1 << 4) /* is this draw_elements? */
186 };
187
188 /**
189 * Check if the requested number of dwords is available in the CS and
190 * if not, flush.
191 * \param r300 The context.
192 * \param flags See r300_prepare_flags.
193 * \param cs_dwords The number of dwords to reserve in CS.
194 * \return TRUE if the CS was flushed
195 */
r300_reserve_cs_dwords(struct r300_context * r300,enum r300_prepare_flags flags,unsigned cs_dwords)196 static bool r300_reserve_cs_dwords(struct r300_context *r300,
197 enum r300_prepare_flags flags,
198 unsigned cs_dwords)
199 {
200 bool flushed = false;
201 bool emit_states = flags & PREP_EMIT_STATES;
202 bool emit_vertex_arrays = flags & PREP_EMIT_VARRAYS;
203 bool emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
204
205 /* Add dirty state, index offset, and AOS. */
206 if (emit_states)
207 cs_dwords += r300_get_num_dirty_dwords(r300);
208
209 if (r300->screen->caps.is_r500)
210 cs_dwords += 2; /* emit_index_offset */
211
212 if (emit_vertex_arrays)
213 cs_dwords += 55; /* emit_vertex_arrays */
214
215 if (emit_vertex_arrays_swtcl)
216 cs_dwords += 7; /* emit_vertex_arrays_swtcl */
217
218 cs_dwords += r300_get_num_cs_end_dwords(r300);
219
220 /* Reserve requested CS space. */
221 if (!r300->rws->cs_check_space(&r300->cs, cs_dwords)) {
222 r300_flush(&r300->context, PIPE_FLUSH_ASYNC, NULL);
223 flushed = true;
224 }
225
226 return flushed;
227 }
228
229 /**
230 * Validate buffers and emit dirty state.
231 * \param r300 The context.
232 * \param flags See r300_prepare_flags.
233 * \param index_buffer The index buffer to validate. The parameter may be NULL.
234 * \param buffer_offset The offset passed to emit_vertex_arrays.
235 * \param index_bias The index bias to emit.
236 * \param instance_id Index of instance to render
237 * \return TRUE if rendering should be skipped
238 */
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)239 static bool r300_emit_states(struct r300_context *r300,
240 enum r300_prepare_flags flags,
241 struct pipe_resource *index_buffer,
242 int buffer_offset,
243 int index_bias, int instance_id)
244 {
245 bool emit_states = flags & PREP_EMIT_STATES;
246 bool emit_vertex_arrays = flags & PREP_EMIT_VARRAYS;
247 bool emit_vertex_arrays_swtcl = flags & PREP_EMIT_VARRAYS_SWTCL;
248 bool indexed = flags & PREP_INDEXED;
249 bool validate_vbos = flags & PREP_VALIDATE_VBOS;
250
251 /* Validate buffers and emit dirty state if needed. */
252 if (emit_states || (emit_vertex_arrays && validate_vbos)) {
253 if (!r300_emit_buffer_validate(r300, validate_vbos,
254 index_buffer)) {
255 fprintf(stderr, "r300: CS space validation failed. "
256 "(not enough memory?) Skipping rendering.\n");
257 return false;
258 }
259 }
260
261 if (emit_states)
262 r300_emit_dirty_state(r300);
263
264 if (r300->screen->caps.is_r500) {
265 if (r300->screen->caps.has_tcl)
266 r500_emit_index_bias(r300, index_bias);
267 else
268 r500_emit_index_bias(r300, 0);
269 }
270
271 if (emit_vertex_arrays &&
272 (r300->vertex_arrays_dirty ||
273 r300->vertex_arrays_indexed != indexed ||
274 r300->vertex_arrays_offset != buffer_offset ||
275 r300->vertex_arrays_instance_id != instance_id)) {
276 r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id);
277
278 r300->vertex_arrays_dirty = false;
279 r300->vertex_arrays_indexed = indexed;
280 r300->vertex_arrays_offset = buffer_offset;
281 r300->vertex_arrays_instance_id = instance_id;
282 }
283
284 if (emit_vertex_arrays_swtcl)
285 r300_emit_vertex_arrays_swtcl(r300, indexed);
286
287 return true;
288 }
289
290 /**
291 * Check if the requested number of dwords is available in the CS and
292 * if not, flush. Then validate buffers and emit dirty state.
293 * \param r300 The context.
294 * \param flags See r300_prepare_flags.
295 * \param index_buffer The index buffer to validate. The parameter may be NULL.
296 * \param cs_dwords The number of dwords to reserve in CS.
297 * \param buffer_offset The offset passed to emit_vertex_arrays.
298 * \param index_bias The index bias to emit.
299 * \param instance_id The instance to render.
300 * \return TRUE if rendering should be skipped
301 */
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)302 static bool r300_prepare_for_rendering(struct r300_context *r300,
303 enum r300_prepare_flags flags,
304 struct pipe_resource *index_buffer,
305 unsigned cs_dwords,
306 int buffer_offset,
307 int index_bias,
308 int instance_id)
309 {
310 /* Make sure there is enough space in the command stream and emit states. */
311 if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
312 flags |= PREP_EMIT_STATES;
313
314 return r300_emit_states(r300, flags, index_buffer, buffer_offset,
315 index_bias, instance_id);
316 }
317
immd_is_good_idea(struct r300_context * r300,unsigned count)318 static bool immd_is_good_idea(struct r300_context *r300,
319 unsigned count)
320 {
321 if (DBG_ON(r300, DBG_NO_IMMD)) {
322 return false;
323 }
324
325 if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
326 return false;
327 }
328
329 /* Buffers can only be used for read by r300 (except query buffers, but
330 * those can't be bound by an gallium frontend as vertex buffers). */
331 return true;
332 }
333
334 /*****************************************************************************
335 * The HWTCL draw functions. *
336 ****************************************************************************/
337
r300_draw_arrays_immediate(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)338 static void r300_draw_arrays_immediate(struct r300_context *r300,
339 const struct pipe_draw_info *info,
340 const struct pipe_draw_start_count_bias *draw)
341 {
342 struct pipe_vertex_element* velem;
343 struct pipe_vertex_buffer* vbuf;
344 unsigned vertex_element_count = r300->velems->count;
345 unsigned i, v, vbi;
346
347 /* Size of the vertex, in dwords. */
348 unsigned vertex_size = r300->velems->vertex_size_dwords;
349
350 /* The number of dwords for this draw operation. */
351 unsigned dwords = 4 + draw->count * vertex_size;
352
353 /* Size of the vertex element, in dwords. */
354 unsigned size[PIPE_MAX_ATTRIBS];
355
356 /* Stride to the same attrib in the next vertex in the vertex buffer,
357 * in dwords. */
358 unsigned stride[PIPE_MAX_ATTRIBS];
359
360 /* Mapped vertex buffers. */
361 uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
362 uint32_t* mapelem[PIPE_MAX_ATTRIBS];
363
364 CS_LOCALS(r300);
365
366 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
367 return;
368
369 /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
370 for (i = 0; i < vertex_element_count; i++) {
371 velem = &r300->velems->velem[i];
372 size[i] = r300->velems->format_size[i] / 4;
373 vbi = velem->vertex_buffer_index;
374 vbuf = &r300->vertex_buffer[vbi];
375 stride[i] = velem->src_stride / 4;
376
377 /* Map the buffer. */
378 if (!map[vbi]) {
379 map[vbi] = (uint32_t*)r300->rws->buffer_map(r300->rws,
380 r300_resource(vbuf->buffer.resource)->buf,
381 &r300->cs, PIPE_MAP_READ | PIPE_MAP_UNSYNCHRONIZED);
382 map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * draw->start;
383 }
384 mapelem[i] = map[vbi] + (velem->src_offset / 4);
385 }
386
387 r300_emit_draw_init(r300, info->mode, draw->count-1);
388
389 BEGIN_CS(dwords);
390 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
391 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, draw->count * vertex_size);
392 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (draw->count << 16) |
393 r300_translate_primitive(info->mode));
394
395 /* Emit vertices. */
396 for (v = 0; v < draw->count; v++) {
397 for (i = 0; i < vertex_element_count; i++) {
398 OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
399 }
400 }
401 END_CS;
402 }
403
r300_emit_draw_arrays(struct r300_context * r300,unsigned mode,unsigned count)404 static void r300_emit_draw_arrays(struct r300_context *r300,
405 unsigned mode,
406 unsigned count)
407 {
408 bool alt_num_verts = count > 65535;
409 CS_LOCALS(r300);
410
411 if (count >= (1 << 24)) {
412 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
413 "refusing to render.\n", count);
414 return;
415 }
416
417 r300_emit_draw_init(r300, mode, count-1);
418
419 BEGIN_CS(2 + (alt_num_verts ? 2 : 0));
420 if (alt_num_verts) {
421 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
422 }
423 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
424 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
425 r300_translate_primitive(mode) |
426 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
427 END_CS;
428 }
429
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)430 static void r300_emit_draw_elements(struct r300_context *r300,
431 struct pipe_resource* indexBuffer,
432 unsigned indexSize,
433 unsigned max_index,
434 unsigned mode,
435 unsigned start,
436 unsigned count,
437 uint16_t *imm_indices3)
438 {
439 uint32_t count_dwords, offset_dwords;
440 bool alt_num_verts = count > 65535;
441 CS_LOCALS(r300);
442
443 if (count >= (1 << 24)) {
444 fprintf(stderr, "r300: Got a huge number of vertices: %i, "
445 "refusing to render (max_index: %i).\n", count, max_index);
446 return;
447 }
448
449 DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, max %u\n",
450 count, max_index);
451
452 r300_emit_draw_init(r300, mode, max_index);
453
454 /* If start is odd, render the first triangle with indices embedded
455 * in the command stream. This will increase start by 3 and make it
456 * even. We can then proceed without a fallback. */
457 if (indexSize == 2 && (start & 1) &&
458 mode == MESA_PRIM_TRIANGLES) {
459 BEGIN_CS(4);
460 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 2);
461 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (3 << 16) |
462 R300_VAP_VF_CNTL__PRIM_TRIANGLES);
463 OUT_CS(imm_indices3[1] << 16 | imm_indices3[0]);
464 OUT_CS(imm_indices3[2]);
465 END_CS;
466
467 start += 3;
468 count -= 3;
469 if (!count)
470 return;
471 }
472
473 offset_dwords = indexSize * start / sizeof(uint32_t);
474
475 BEGIN_CS(8 + (alt_num_verts ? 2 : 0));
476 if (alt_num_verts) {
477 OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
478 }
479 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
480 if (indexSize == 4) {
481 count_dwords = count;
482 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
483 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
484 r300_translate_primitive(mode) |
485 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
486 } else {
487 count_dwords = (count + 1) / 2;
488 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
489 r300_translate_primitive(mode) |
490 (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
491 }
492
493 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
494 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
495 (0 << R300_INDX_BUFFER_SKIP_SHIFT));
496 OUT_CS(offset_dwords << 2);
497 OUT_CS(count_dwords);
498 OUT_CS_RELOC(r300_resource(indexBuffer));
499 END_CS;
500 }
501
r300_draw_elements_immediate(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)502 static void r300_draw_elements_immediate(struct r300_context *r300,
503 const struct pipe_draw_info *info,
504 const struct pipe_draw_start_count_bias *draw)
505 {
506 const uint8_t *ptr1;
507 const uint16_t *ptr2;
508 const uint32_t *ptr4;
509 unsigned index_size = info->index_size;
510 unsigned i, count_dwords = index_size == 4 ? draw->count :
511 (draw->count + 1) / 2;
512 CS_LOCALS(r300);
513
514 /* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
515 if (!r300_prepare_for_rendering(r300,
516 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
517 PREP_INDEXED, NULL, 2+count_dwords, 0, draw->index_bias, -1))
518 return;
519
520 r300_emit_draw_init(r300, info->mode, info->max_index);
521
522 BEGIN_CS(2 + count_dwords);
523 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
524
525 switch (index_size) {
526 case 1:
527 ptr1 = (uint8_t*)info->index.user;
528 ptr1 += draw->start;
529
530 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
531 r300_translate_primitive(info->mode));
532
533 if (draw->index_bias && !r300->screen->caps.is_r500) {
534 for (i = 0; i < draw->count-1; i += 2)
535 OUT_CS(((ptr1[i+1] + draw->index_bias) << 16) |
536 (ptr1[i] + draw->index_bias));
537
538 if (draw->count & 1)
539 OUT_CS(ptr1[i] + draw->index_bias);
540 } else {
541 for (i = 0; i < draw->count-1; i += 2)
542 OUT_CS(((ptr1[i+1]) << 16) |
543 (ptr1[i] ));
544
545 if (draw->count & 1)
546 OUT_CS(ptr1[i]);
547 }
548 break;
549
550 case 2:
551 ptr2 = (uint16_t*)info->index.user;
552 ptr2 += draw->start;
553
554 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
555 r300_translate_primitive(info->mode));
556
557 if (draw->index_bias && !r300->screen->caps.is_r500) {
558 for (i = 0; i < draw->count-1; i += 2)
559 OUT_CS(((ptr2[i+1] + draw->index_bias) << 16) |
560 (ptr2[i] + draw->index_bias));
561
562 if (draw->count & 1)
563 OUT_CS(ptr2[i] + draw->index_bias);
564 } else {
565 OUT_CS_TABLE(ptr2, count_dwords);
566 }
567 break;
568
569 case 4:
570 ptr4 = (uint32_t*)info->index.user;
571 ptr4 += draw->start;
572
573 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (draw->count << 16) |
574 R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
575 r300_translate_primitive(info->mode));
576
577 if (draw->index_bias && !r300->screen->caps.is_r500) {
578 for (i = 0; i < draw->count; i++)
579 OUT_CS(ptr4[i] + draw->index_bias);
580 } else {
581 OUT_CS_TABLE(ptr4, count_dwords);
582 }
583 break;
584 }
585 END_CS;
586 }
587
r300_draw_elements(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,int instance_id)588 static void r300_draw_elements(struct r300_context *r300,
589 const struct pipe_draw_info *info,
590 const struct pipe_draw_start_count_bias *draw,
591 int instance_id)
592 {
593 struct pipe_resource *indexBuffer =
594 info->has_user_indices ? NULL : info->index.resource;
595 unsigned indexSize = info->index_size;
596 struct pipe_resource* orgIndexBuffer = indexBuffer;
597 unsigned start = draw->start;
598 unsigned count = draw->count;
599 bool alt_num_verts = r300->screen->caps.is_r500 &&
600 count > 65536;
601 unsigned short_count;
602 int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
603 uint16_t indices3[3];
604
605 if (draw->index_bias && !r300->screen->caps.is_r500) {
606 r300_split_index_bias(r300, draw->index_bias, &buffer_offset,
607 &index_offset);
608 }
609
610 r300_translate_index_buffer(r300, info, &indexBuffer,
611 &indexSize, index_offset, &start, count);
612
613 /* Fallback for misaligned ushort indices. */
614 if (indexSize == 2 && (start & 1) && indexBuffer) {
615 /* If we got here, then orgIndexBuffer == indexBuffer. */
616 uint16_t *ptr = r300->rws->buffer_map(r300->rws, r300_resource(orgIndexBuffer)->buf,
617 &r300->cs,
618 PIPE_MAP_READ |
619 PIPE_MAP_UNSYNCHRONIZED);
620
621 if (info->mode == MESA_PRIM_TRIANGLES) {
622 memcpy(indices3, ptr + start, 6);
623 } else {
624 /* Copy the mapped index buffer directly to the upload buffer.
625 * The start index will be aligned simply from the fact that
626 * every sub-buffer in the upload buffer is aligned. */
627 r300_upload_index_buffer(r300, &indexBuffer, indexSize, &start,
628 count, (uint8_t*)ptr);
629 }
630 } else {
631 if (info->has_user_indices)
632 r300_upload_index_buffer(r300, &indexBuffer, indexSize,
633 &start, count,
634 info->index.user);
635 }
636
637 /* 19 dwords for emit_draw_elements. Give up if the function fails. */
638 if (!r300_prepare_for_rendering(r300,
639 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS |
640 PREP_INDEXED, indexBuffer, 19, buffer_offset, draw->index_bias,
641 instance_id))
642 goto done;
643
644 if (alt_num_verts || count <= 65535) {
645 r300_emit_draw_elements(r300, indexBuffer, indexSize,
646 info->max_index, info->mode, start, count,
647 indices3);
648 } else {
649 do {
650 /* The maximum must be divisible by 4 and 3,
651 * so that quad and triangle lists are split correctly.
652 *
653 * Strips, loops, and fans won't work. */
654 short_count = MIN2(count, 65532);
655
656 r300_emit_draw_elements(r300, indexBuffer, indexSize,
657 info->max_index,
658 info->mode, start, short_count, indices3);
659
660 start += short_count;
661 count -= short_count;
662
663 /* 15 dwords for emit_draw_elements */
664 if (count) {
665 if (!r300_prepare_for_rendering(r300,
666 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS | PREP_INDEXED,
667 indexBuffer, 19, buffer_offset, draw->index_bias,
668 instance_id))
669 goto done;
670 }
671 } while (count);
672 }
673
674 done:
675 if (indexBuffer != orgIndexBuffer) {
676 pipe_resource_reference( &indexBuffer, NULL );
677 }
678 }
679
r300_draw_arrays(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw,int instance_id)680 static void r300_draw_arrays(struct r300_context *r300,
681 const struct pipe_draw_info *info,
682 const struct pipe_draw_start_count_bias *draw,
683 int instance_id)
684 {
685 bool alt_num_verts = r300->screen->caps.is_r500 &&
686 draw->count > 65536;
687 unsigned start = draw->start;
688 unsigned count = draw->count;
689 unsigned short_count;
690
691 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
692 if (!r300_prepare_for_rendering(r300,
693 PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS,
694 NULL, 9, start, 0, instance_id))
695 return;
696
697 if (alt_num_verts || count <= 65535) {
698 r300_emit_draw_arrays(r300, info->mode, count);
699 } else {
700 do {
701 /* The maximum must be divisible by 4 and 3,
702 * so that quad and triangle lists are split correctly.
703 *
704 * Strips, loops, and fans won't work. */
705 short_count = MIN2(count, 65532);
706 r300_emit_draw_arrays(r300, info->mode, short_count);
707
708 start += short_count;
709 count -= short_count;
710
711 /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
712 if (count) {
713 if (!r300_prepare_for_rendering(r300,
714 PREP_VALIDATE_VBOS | PREP_EMIT_VARRAYS, NULL, 9,
715 start, 0, instance_id))
716 return;
717 }
718 } while (count);
719 }
720 }
721
r300_draw_arrays_instanced(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)722 static void r300_draw_arrays_instanced(struct r300_context *r300,
723 const struct pipe_draw_info *info,
724 const struct pipe_draw_start_count_bias *draw)
725 {
726 int i;
727
728 for (i = 0; i < info->instance_count; i++)
729 r300_draw_arrays(r300, info, draw, i);
730 }
731
r300_draw_elements_instanced(struct r300_context * r300,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)732 static void r300_draw_elements_instanced(struct r300_context *r300,
733 const struct pipe_draw_info *info,
734 const struct pipe_draw_start_count_bias *draw)
735 {
736 int i;
737
738 for (i = 0; i < info->instance_count; i++)
739 r300_draw_elements(r300, info, draw, i);
740 }
741
r300_max_vertex_count(struct r300_context * r300)742 static unsigned r300_max_vertex_count(struct r300_context *r300)
743 {
744 unsigned i, nr = r300->velems->count;
745 struct pipe_vertex_element *velems = r300->velems->velem;
746 unsigned result = ~0;
747
748 for (i = 0; i < nr; i++) {
749 struct pipe_vertex_buffer *vb =
750 &r300->vertex_buffer[velems[i].vertex_buffer_index];
751 unsigned size, max_count, value;
752
753 /* We're not interested in constant and per-instance attribs. */
754 if (!vb->buffer.resource ||
755 !velems[i].src_stride ||
756 velems[i].instance_divisor) {
757 continue;
758 }
759
760 size = vb->buffer.resource->width0;
761
762 /* Subtract buffer_offset. */
763 value = vb->buffer_offset;
764 if (value >= size) {
765 return 0;
766 }
767 size -= value;
768
769 /* Subtract src_offset. */
770 value = velems[i].src_offset;
771 if (value >= size) {
772 return 0;
773 }
774 size -= value;
775
776 /* Compute the max count. */
777 max_count = 1 + size / velems[i].src_stride;
778 result = MIN2(result, max_count);
779 }
780 return result;
781 }
782
783
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)784 static void r300_draw_vbo(struct pipe_context* pipe,
785 const struct pipe_draw_info *dinfo,
786 unsigned drawid_offset,
787 const struct pipe_draw_indirect_info *indirect,
788 const struct pipe_draw_start_count_bias *draws,
789 unsigned num_draws)
790 {
791 if (num_draws > 1) {
792 util_draw_multi(pipe, dinfo, drawid_offset, indirect, draws, num_draws);
793 return;
794 }
795
796 struct r300_context* r300 = r300_context(pipe);
797 struct pipe_draw_info info = *dinfo;
798 struct pipe_draw_start_count_bias draw = draws[0];
799
800 if (r300->skip_rendering ||
801 !u_trim_pipe_prim(info.mode, &draw.count)) {
802 return;
803 }
804
805 if (r300->sprite_coord_enable != 0 ||
806 r300_fs(r300)->shader->inputs.pcoord != ATTR_UNUSED) {
807 if ((info.mode == MESA_PRIM_POINTS) != r300->is_point) {
808 r300->is_point = !r300->is_point;
809 r300_mark_atom_dirty(r300, &r300->rs_block_state);
810 }
811 }
812
813 r300_update_derived_state(r300);
814
815 /* Skip draw if we failed to compile the vertex shader. */
816 if (r300_vs(r300)->shader->dummy)
817 return;
818
819 /* Draw. */
820 if (info.index_size) {
821 unsigned max_count = r300_max_vertex_count(r300);
822
823 if (!max_count) {
824 fprintf(stderr, "r300: Skipping a draw command. There is a buffer "
825 " which is too small to be used for rendering.\n");
826 return;
827 }
828
829 if (max_count == ~0) {
830 /* There are no per-vertex vertex elements. Use the hardware maximum. */
831 max_count = 0xffffff;
832 }
833
834 info.max_index = max_count - 1;
835
836 if (info.instance_count <= 1) {
837 if (draw.count <= 8 && info.has_user_indices) {
838 r300_draw_elements_immediate(r300, &info, &draw);
839 } else {
840 r300_draw_elements(r300, &info, &draw, -1);
841 }
842 } else {
843 r300_draw_elements_instanced(r300, &info, &draw);
844 }
845 } else {
846 if (info.instance_count <= 1) {
847 if (immd_is_good_idea(r300, draw.count)) {
848 r300_draw_arrays_immediate(r300, &info, &draw);
849 } else {
850 r300_draw_arrays(r300, &info, &draw, -1);
851 }
852 } else {
853 r300_draw_arrays_instanced(r300, &info, &draw);
854 }
855 }
856 }
857
858 /****************************************************************************
859 * The rest of this file is for SW TCL rendering only. Please be polite and *
860 * keep these functions separated so that they are easier to locate. ~C. *
861 ***************************************************************************/
862
863 /* 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)864 static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
865 const struct pipe_draw_info *info,
866 unsigned drawid_offset,
867 const struct pipe_draw_indirect_info *indirect,
868 const struct pipe_draw_start_count_bias *draws,
869 unsigned num_draws)
870 {
871 if (num_draws > 1) {
872 util_draw_multi(pipe, info, drawid_offset, indirect, draws, num_draws);
873 return;
874 }
875
876 struct r300_context* r300 = r300_context(pipe);
877 struct pipe_draw_start_count_bias draw = draws[0];
878
879 if (r300->skip_rendering) {
880 return;
881 }
882
883 if (!u_trim_pipe_prim(info->mode, &draw.count))
884 return;
885
886 if (info->index_size) {
887 draw_set_indexes(r300->draw,
888 info->has_user_indices ?
889 info->index.user :
890 r300_resource(info->index.resource)->malloced_buffer,
891 info->index_size, ~0);
892 }
893
894 if (r300->sprite_coord_enable != 0 ||
895 r300_fs(r300)->shader->inputs.pcoord != ATTR_UNUSED) {
896 if ((info->mode == MESA_PRIM_POINTS) != r300->is_point) {
897 r300->is_point = !r300->is_point;
898 r300_mark_atom_dirty(r300, &r300->rs_block_state);
899 }
900 }
901
902
903 r300_update_derived_state(r300);
904
905 draw_vbo(r300->draw, info, drawid_offset, NULL, &draw, 1, 0);
906 draw_flush(r300->draw);
907 }
908
909 /* Object for rendering using Draw. */
910 struct r300_render {
911 /* Parent class */
912 struct vbuf_render base;
913
914 /* Pipe context */
915 struct r300_context* r300;
916
917 /* Vertex information */
918 size_t vertex_size;
919 unsigned prim;
920 unsigned hwprim;
921
922 /* VBO */
923 size_t vbo_max_used;
924 uint8_t *vbo_ptr;
925 };
926
927 static inline struct r300_render*
r300_render(struct vbuf_render * render)928 r300_render(struct vbuf_render* render)
929 {
930 return (struct r300_render*)render;
931 }
932
933 static const struct vertex_info*
r300_render_get_vertex_info(struct vbuf_render * render)934 r300_render_get_vertex_info(struct vbuf_render* render)
935 {
936 struct r300_render* r300render = r300_render(render);
937 struct r300_context* r300 = r300render->r300;
938
939 return &r300->vertex_info;
940 }
941
r300_render_allocate_vertices(struct vbuf_render * render,uint16_t vertex_size,uint16_t count)942 static bool r300_render_allocate_vertices(struct vbuf_render* render,
943 uint16_t vertex_size,
944 uint16_t count)
945 {
946 struct r300_render* r300render = r300_render(render);
947 struct r300_context* r300 = r300render->r300;
948 struct radeon_winsys *rws = r300->rws;
949 size_t size = (size_t)vertex_size * (size_t)count;
950
951 DBG(r300, DBG_DRAW, "r300: render_allocate_vertices (size: %d)\n", size);
952
953 if (!r300->vbo || size + r300->draw_vbo_offset > r300->vbo->size) {
954 radeon_bo_reference(r300->rws, &r300->vbo, NULL);
955 r300->vbo = NULL;
956 r300render->vbo_ptr = NULL;
957
958 r300->vbo = rws->buffer_create(rws,
959 MAX2(R300_MAX_DRAW_VBO_SIZE, size),
960 R300_BUFFER_ALIGNMENT,
961 RADEON_DOMAIN_GTT,
962 RADEON_FLAG_NO_INTERPROCESS_SHARING);
963 if (!r300->vbo) {
964 return false;
965 }
966 r300->draw_vbo_offset = 0;
967 r300render->vbo_ptr = rws->buffer_map(rws, r300->vbo, &r300->cs,
968 PIPE_MAP_WRITE);
969 }
970
971 r300render->vertex_size = vertex_size;
972 return true;
973 }
974
r300_render_map_vertices(struct vbuf_render * render)975 static void* r300_render_map_vertices(struct vbuf_render* render)
976 {
977 struct r300_render* r300render = r300_render(render);
978 struct r300_context* r300 = r300render->r300;
979
980 DBG(r300, DBG_DRAW, "r300: render_map_vertices\n");
981
982 assert(r300render->vbo_ptr);
983 return r300render->vbo_ptr + r300->draw_vbo_offset;
984 }
985
r300_render_unmap_vertices(struct vbuf_render * render,uint16_t min,uint16_t max)986 static void r300_render_unmap_vertices(struct vbuf_render* render,
987 uint16_t min,
988 uint16_t max)
989 {
990 struct r300_render* r300render = r300_render(render);
991 struct r300_context* r300 = r300render->r300;
992
993 DBG(r300, DBG_DRAW, "r300: render_unmap_vertices\n");
994
995 r300render->vbo_max_used = MAX2(r300render->vbo_max_used,
996 r300render->vertex_size * (max + 1));
997 }
998
r300_render_release_vertices(struct vbuf_render * render)999 static void r300_render_release_vertices(struct vbuf_render* render)
1000 {
1001 struct r300_render* r300render = r300_render(render);
1002 struct r300_context* r300 = r300render->r300;
1003
1004 DBG(r300, DBG_DRAW, "r300: render_release_vertices\n");
1005
1006 r300->draw_vbo_offset += r300render->vbo_max_used;
1007 r300render->vbo_max_used = 0;
1008 }
1009
r300_render_set_primitive(struct vbuf_render * render,enum mesa_prim prim)1010 static void r300_render_set_primitive(struct vbuf_render* render,
1011 enum mesa_prim prim)
1012 {
1013 struct r300_render* r300render = r300_render(render);
1014
1015 r300render->prim = prim;
1016 r300render->hwprim = r300_translate_primitive(prim);
1017 }
1018
r300_render_draw_arrays(struct vbuf_render * render,unsigned start,unsigned count)1019 static void r300_render_draw_arrays(struct vbuf_render* render,
1020 unsigned start,
1021 unsigned count)
1022 {
1023 struct r300_render* r300render = r300_render(render);
1024 struct r300_context* r300 = r300render->r300;
1025 uint8_t* ptr;
1026 unsigned i;
1027 unsigned dwords = 6;
1028
1029 CS_LOCALS(r300);
1030 (void) i; (void) ptr;
1031
1032 assert(start == 0);
1033 assert(count < (1 << 16));
1034
1035 DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
1036
1037 if (!r300_prepare_for_rendering(r300,
1038 PREP_EMIT_STATES | PREP_EMIT_VARRAYS_SWTCL,
1039 NULL, dwords, 0, 0, -1)) {
1040 return;
1041 }
1042
1043 BEGIN_CS(dwords);
1044 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1045 r300_provoking_vertex_fixes(r300, r300render->prim));
1046 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
1047 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
1048 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
1049 r300render->hwprim);
1050 END_CS;
1051 }
1052
r300_render_draw_elements(struct vbuf_render * render,const uint16_t * indices,uint count)1053 static void r300_render_draw_elements(struct vbuf_render* render,
1054 const uint16_t* indices,
1055 uint count)
1056 {
1057 struct r300_render* r300render = r300_render(render);
1058 struct r300_context* r300 = r300render->r300;
1059 unsigned max_index = (r300->vbo->size - r300->draw_vbo_offset) /
1060 (r300render->r300->vertex_info.size * 4) - 1;
1061 struct pipe_resource *index_buffer = NULL;
1062 unsigned index_buffer_offset;
1063
1064 CS_LOCALS(r300);
1065 DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
1066
1067 u_upload_data(r300->uploader, 0, count * 2, 4, indices,
1068 &index_buffer_offset, &index_buffer);
1069 if (!index_buffer) {
1070 return;
1071 }
1072
1073 if (!r300_prepare_for_rendering(r300,
1074 PREP_EMIT_STATES |
1075 PREP_EMIT_VARRAYS_SWTCL | PREP_INDEXED,
1076 index_buffer, 12, 0, 0, -1)) {
1077 pipe_resource_reference(&index_buffer, NULL);
1078 return;
1079 }
1080
1081 BEGIN_CS(12);
1082 OUT_CS_REG(R300_GA_COLOR_CONTROL,
1083 r300_provoking_vertex_fixes(r300, r300render->prim));
1084 OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, max_index);
1085
1086 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
1087 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
1088 r300render->hwprim);
1089
1090 OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
1091 OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2));
1092 OUT_CS(index_buffer_offset);
1093 OUT_CS((count + 1) / 2);
1094 OUT_CS_RELOC(r300_resource(index_buffer));
1095 END_CS;
1096
1097 pipe_resource_reference(&index_buffer, NULL);
1098 }
1099
r300_render_destroy(struct vbuf_render * render)1100 static void r300_render_destroy(struct vbuf_render* render)
1101 {
1102 FREE(render);
1103 }
1104
r300_render_create(struct r300_context * r300)1105 static struct vbuf_render* r300_render_create(struct r300_context* r300)
1106 {
1107 struct r300_render* r300render = CALLOC_STRUCT(r300_render);
1108
1109 r300render->r300 = r300;
1110
1111 r300render->base.max_vertex_buffer_bytes = R300_MAX_DRAW_VBO_SIZE;
1112 r300render->base.max_indices = 16 * 1024;
1113
1114 r300render->base.get_vertex_info = r300_render_get_vertex_info;
1115 r300render->base.allocate_vertices = r300_render_allocate_vertices;
1116 r300render->base.map_vertices = r300_render_map_vertices;
1117 r300render->base.unmap_vertices = r300_render_unmap_vertices;
1118 r300render->base.set_primitive = r300_render_set_primitive;
1119 r300render->base.draw_elements = r300_render_draw_elements;
1120 r300render->base.draw_arrays = r300_render_draw_arrays;
1121 r300render->base.release_vertices = r300_render_release_vertices;
1122 r300render->base.destroy = r300_render_destroy;
1123
1124 return &r300render->base;
1125 }
1126
r300_draw_stage(struct r300_context * r300)1127 struct draw_stage* r300_draw_stage(struct r300_context* r300)
1128 {
1129 struct vbuf_render* render;
1130 struct draw_stage* stage;
1131
1132 render = r300_render_create(r300);
1133
1134 if (!render) {
1135 return NULL;
1136 }
1137
1138 stage = draw_vbuf_stage(r300->draw, render);
1139
1140 if (!stage) {
1141 render->destroy(render);
1142 return NULL;
1143 }
1144
1145 draw_set_render(r300->draw, render);
1146
1147 return stage;
1148 }
1149
1150 /****************************************************************************
1151 * End of SW TCL functions *
1152 ***************************************************************************/
1153
1154 /* This functions is used to draw a rectangle for the blitter module.
1155 *
1156 * If we rendered a quad, the pixels on the main diagonal
1157 * would be computed and stored twice, which makes the clear/copy codepaths
1158 * 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)1159 void r300_blitter_draw_rectangle(struct blitter_context *blitter,
1160 void *vertex_elements_cso,
1161 blitter_get_vs_func get_vs,
1162 int x1, int y1, int x2, int y2,
1163 float depth, unsigned num_instances,
1164 enum blitter_attrib_type type,
1165 const union blitter_attrib *attrib)
1166 {
1167 struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
1168 unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
1169 unsigned last_is_point = r300->is_point;
1170 unsigned width = x2 - x1;
1171 unsigned height = y2 - y1;
1172 unsigned vertex_size =
1173 type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
1174 unsigned dwords = 13 + vertex_size +
1175 (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY ? 7 : 0);
1176 static const union blitter_attrib zeros;
1177 CS_LOCALS(r300);
1178
1179 /* XXX workaround for a lockup in MSAA resolve on SWTCL chipsets, this
1180 * function most probably doesn't handle type=NONE correctly */
1181 if ((!r300->screen->caps.has_tcl && type == UTIL_BLITTER_ATTRIB_NONE) ||
1182 type == UTIL_BLITTER_ATTRIB_TEXCOORD_XYZW ||
1183 num_instances > 1) {
1184 util_blitter_draw_rectangle(blitter, vertex_elements_cso, get_vs,
1185 x1, y1, x2, y2,
1186 depth, num_instances, type, attrib);
1187 return;
1188 }
1189
1190 if (r300->skip_rendering)
1191 return;
1192
1193 r300->context.bind_vertex_elements_state(&r300->context, vertex_elements_cso);
1194 r300->context.bind_vs_state(&r300->context, get_vs(blitter));
1195
1196 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) {
1197 r300->sprite_coord_enable = 1;
1198 r300->is_point = true;
1199 }
1200
1201 r300_update_derived_state(r300);
1202
1203 /* Mark some states we don't care about as non-dirty. */
1204 r300->viewport_state.dirty = false;
1205
1206 if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
1207 goto done;
1208
1209 DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
1210
1211 BEGIN_CS(dwords);
1212 /* Set up GA. */
1213 OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
1214
1215 if (type == UTIL_BLITTER_ATTRIB_TEXCOORD_XY) {
1216 /* Set up the GA to generate texcoords. */
1217 OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
1218 (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
1219 OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
1220 OUT_CS_32F(attrib->texcoord.x1);
1221 OUT_CS_32F(attrib->texcoord.y2);
1222 OUT_CS_32F(attrib->texcoord.x2);
1223 OUT_CS_32F(attrib->texcoord.y1);
1224 }
1225
1226 /* Set up VAP controls. */
1227 OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
1228 OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
1229 OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
1230 OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
1231 OUT_CS(1);
1232 OUT_CS(0);
1233
1234 /* Draw. */
1235 OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
1236 OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
1237 R300_VAP_VF_CNTL__PRIM_POINTS);
1238
1239 OUT_CS_32F(x1 + width * 0.5f);
1240 OUT_CS_32F(y1 + height * 0.5f);
1241 OUT_CS_32F(depth);
1242 OUT_CS_32F(1);
1243
1244 if (vertex_size == 8) {
1245 if (!attrib)
1246 attrib = &zeros;
1247 OUT_CS_TABLE(attrib->color, 4);
1248 }
1249 END_CS;
1250
1251 done:
1252 /* Restore the state. */
1253 r300_mark_atom_dirty(r300, &r300->rs_state);
1254 r300_mark_atom_dirty(r300, &r300->viewport_state);
1255
1256 r300->sprite_coord_enable = last_sprite_coord_enable;
1257 r300->is_point = last_is_point;
1258 }
1259
r300_init_render_functions(struct r300_context * r300)1260 void r300_init_render_functions(struct r300_context *r300)
1261 {
1262 /* Set draw functions based on presence of HW TCL. */
1263 if (r300->screen->caps.has_tcl) {
1264 r300->context.draw_vbo = r300_draw_vbo;
1265 } else {
1266 r300->context.draw_vbo = r300_swtcl_draw_vbo;
1267 }
1268
1269 /* Plug in the two-sided stencil reference value fallback if needed. */
1270 if (!r300->screen->caps.is_r500)
1271 r300_plug_in_stencil_ref_fallback(r300);
1272 }
1273