nouveau: switch to libdrm_nouveau-2.0
[mesa.git] / src / gallium / drivers / nv50 / nv50_vbo.c
1 /*
2 * Copyright 2010 Christoph Bumiller
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
18 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
19 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
20 * SOFTWARE.
21 */
22
23 #include "pipe/p_context.h"
24 #include "pipe/p_state.h"
25 #include "util/u_inlines.h"
26 #include "util/u_format.h"
27 #include "translate/translate.h"
28
29 #include "nv50_context.h"
30 #include "nv50_resource.h"
31
32 #include "nv50_3d.xml.h"
33
34 void
35 nv50_vertex_state_delete(struct pipe_context *pipe,
36 void *hwcso)
37 {
38 struct nv50_vertex_stateobj *so = hwcso;
39
40 if (so->translate)
41 so->translate->release(so->translate);
42 FREE(hwcso);
43 }
44
45 void *
46 nv50_vertex_state_create(struct pipe_context *pipe,
47 unsigned num_elements,
48 const struct pipe_vertex_element *elements)
49 {
50 struct nv50_vertex_stateobj *so;
51 struct translate_key transkey;
52 unsigned i;
53
54 so = MALLOC(sizeof(*so) +
55 num_elements * sizeof(struct nv50_vertex_element));
56 if (!so)
57 return NULL;
58 so->num_elements = num_elements;
59 so->instance_elts = 0;
60 so->instance_bufs = 0;
61 so->need_conversion = FALSE;
62
63 transkey.nr_elements = 0;
64 transkey.output_stride = 0;
65
66 for (i = 0; i < num_elements; ++i) {
67 const struct pipe_vertex_element *ve = &elements[i];
68 const unsigned vbi = ve->vertex_buffer_index;
69 enum pipe_format fmt = ve->src_format;
70
71 so->element[i].pipe = elements[i];
72 so->element[i].state = nv50_format_table[fmt].vtx;
73
74 if (!so->element[i].state) {
75 switch (util_format_get_nr_components(fmt)) {
76 case 1: fmt = PIPE_FORMAT_R32_FLOAT; break;
77 case 2: fmt = PIPE_FORMAT_R32G32_FLOAT; break;
78 case 3: fmt = PIPE_FORMAT_R32G32B32_FLOAT; break;
79 case 4: fmt = PIPE_FORMAT_R32G32B32A32_FLOAT; break;
80 default:
81 assert(0);
82 return NULL;
83 }
84 so->element[i].state = nv50_format_table[fmt].vtx;
85 so->need_conversion = TRUE;
86 }
87 so->element[i].state |= i;
88
89 if (1) {
90 unsigned j = transkey.nr_elements++;
91
92 transkey.element[j].type = TRANSLATE_ELEMENT_NORMAL;
93 transkey.element[j].input_format = ve->src_format;
94 transkey.element[j].input_buffer = vbi;
95 transkey.element[j].input_offset = ve->src_offset;
96 transkey.element[j].instance_divisor = ve->instance_divisor;
97
98 transkey.element[j].output_format = fmt;
99 transkey.element[j].output_offset = transkey.output_stride;
100 transkey.output_stride += (util_format_get_stride(fmt, 1) + 3) & ~3;
101
102 if (unlikely(ve->instance_divisor)) {
103 so->instance_elts |= 1 << i;
104 so->instance_bufs |= 1 << vbi;
105 }
106 }
107 }
108
109 so->translate = translate_create(&transkey);
110 so->vertex_size = transkey.output_stride / 4;
111 so->packet_vertex_limit = NV04_PFIFO_MAX_PACKET_LEN /
112 MAX2(so->vertex_size, 1);
113
114 return so;
115 }
116
117 #define NV50_3D_VERTEX_ATTRIB_INACTIVE \
118 NV50_3D_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT | \
119 NV50_3D_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 | \
120 NV50_3D_VERTEX_ARRAY_ATTRIB_CONST
121
122 static void
123 nv50_emit_vtxattr(struct nv50_context *nv50, struct pipe_vertex_buffer *vb,
124 struct pipe_vertex_element *ve, unsigned attr)
125 {
126 const void *data;
127 struct nouveau_pushbuf *push = nv50->base.pushbuf;
128 struct nv04_resource *res = nv04_resource(vb->buffer);
129 float v[4];
130 const unsigned nc = util_format_get_nr_components(ve->src_format);
131
132 data = nouveau_resource_map_offset(&nv50->base, res, vb->buffer_offset +
133 ve->src_offset, NOUVEAU_BO_RD);
134
135 util_format_read_4f(ve->src_format, v, 0, data, 0, 0, 0, 1, 1);
136
137 switch (nc) {
138 case 4:
139 BEGIN_NV04(push, NV50_3D(VTX_ATTR_4F_X(attr)), 4);
140 PUSH_DATAf(push, v[0]);
141 PUSH_DATAf(push, v[1]);
142 PUSH_DATAf(push, v[2]);
143 PUSH_DATAf(push, v[3]);
144 break;
145 case 3:
146 BEGIN_NV04(push, NV50_3D(VTX_ATTR_3F_X(attr)), 3);
147 PUSH_DATAf(push, v[0]);
148 PUSH_DATAf(push, v[1]);
149 PUSH_DATAf(push, v[2]);
150 break;
151 case 2:
152 BEGIN_NV04(push, NV50_3D(VTX_ATTR_2F_X(attr)), 2);
153 PUSH_DATAf(push, v[0]);
154 PUSH_DATAf(push, v[1]);
155 break;
156 case 1:
157 if (attr == nv50->vertprog->vp.edgeflag) {
158 BEGIN_NV04(push, NV50_3D(EDGEFLAG), 1);
159 PUSH_DATA (push, v[0] ? 1 : 0);
160 }
161 BEGIN_NV04(push, NV50_3D(VTX_ATTR_1F(attr)), 1);
162 PUSH_DATAf(push, v[0]);
163 break;
164 default:
165 assert(0);
166 break;
167 }
168 }
169
170 static INLINE void
171 nv50_vbuf_range(struct nv50_context *nv50, int vbi,
172 uint32_t *base, uint32_t *size)
173 {
174 if (unlikely(nv50->vertex->instance_bufs & (1 << vbi))) {
175 /* TODO: use min and max instance divisor to get a proper range */
176 *base = 0;
177 *size = nv50->vtxbuf[vbi].buffer->width0;
178 } else {
179 assert(nv50->vbo_max_index != ~0);
180 *base = nv50->vbo_min_index * nv50->vtxbuf[vbi].stride;
181 *size = (nv50->vbo_max_index -
182 nv50->vbo_min_index + 1) * nv50->vtxbuf[vbi].stride;
183 }
184 }
185
186 static void
187 nv50_prevalidate_vbufs(struct nv50_context *nv50)
188 {
189 struct pipe_vertex_buffer *vb;
190 struct nv04_resource *buf;
191 int i;
192 uint32_t base, size;
193
194 nv50->vbo_fifo = nv50->vbo_user = 0;
195
196 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_VERTEX);
197
198 for (i = 0; i < nv50->num_vtxbufs; ++i) {
199 vb = &nv50->vtxbuf[i];
200 if (!vb->stride)
201 continue;
202 buf = nv04_resource(vb->buffer);
203
204 /* NOTE: user buffers with temporary storage count as mapped by GPU */
205 if (!nouveau_resource_mapped_by_gpu(vb->buffer)) {
206 if (nv50->vbo_push_hint) {
207 nv50->vbo_fifo = ~0;
208 continue;
209 } else {
210 if (buf->status & NOUVEAU_BUFFER_STATUS_USER_MEMORY) {
211 nv50->vbo_user |= 1 << i;
212 assert(vb->stride > vb->buffer_offset);
213 nv50_vbuf_range(nv50, i, &base, &size);
214 nouveau_user_buffer_upload(&nv50->base, buf, base, size);
215 } else {
216 nouveau_buffer_migrate(&nv50->base, buf, NOUVEAU_BO_GART);
217 }
218 nv50->base.vbo_dirty = TRUE;
219 }
220 }
221 BCTX_REFN(nv50->bufctx_3d, VERTEX, buf, RD);
222 }
223 }
224
225 static void
226 nv50_update_user_vbufs(struct nv50_context *nv50)
227 {
228 struct nouveau_pushbuf *push = nv50->base.pushbuf;
229 uint32_t base, offset, size;
230 int i;
231 uint32_t written = 0;
232
233 nouveau_bufctx_reset(nv50->bufctx_3d, NV50_BIND_VERTEX);
234
235 for (i = 0; i < nv50->vertex->num_elements; ++i) {
236 struct pipe_vertex_element *ve = &nv50->vertex->element[i].pipe;
237 const int b = ve->vertex_buffer_index;
238 struct pipe_vertex_buffer *vb = &nv50->vtxbuf[b];
239 struct nv04_resource *buf = nv04_resource(vb->buffer);
240
241 if (!(nv50->vbo_user & (1 << b))) {
242 BCTX_REFN(nv50->bufctx_3d, VERTEX, buf, RD);
243 continue;
244 }
245
246 if (!vb->stride) {
247 nv50_emit_vtxattr(nv50, vb, ve, i);
248 continue;
249 }
250 nv50_vbuf_range(nv50, b, &base, &size);
251
252 if (!(written & (1 << b))) {
253 written |= 1 << b;
254 nouveau_user_buffer_upload(&nv50->base, buf, base, size);
255 }
256 offset = vb->buffer_offset + ve->src_offset;
257
258 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
259 PUSH_DATAh(push, buf->address + base + size - 1);
260 PUSH_DATA (push, buf->address + base + size - 1);
261 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_START_HIGH(i)), 2);
262 PUSH_DATAh(push, buf->address + offset);
263 PUSH_DATA (push, buf->address + offset);
264
265 BCTX_REFN(nv50->bufctx_3d, VERTEX, buf, RD);
266 }
267 nv50->base.vbo_dirty = TRUE;
268 }
269
270 static INLINE void
271 nv50_release_user_vbufs(struct nv50_context *nv50)
272 {
273 uint32_t vbo_user = nv50->vbo_user;
274
275 while (vbo_user) {
276 int i = ffs(vbo_user) - 1;
277 vbo_user &= ~(1 << i);
278
279 nouveau_buffer_release_gpu_storage(nv04_resource(nv50->vtxbuf[i].buffer));
280 }
281 }
282
283 void
284 nv50_vertex_arrays_validate(struct nv50_context *nv50)
285 {
286 struct nouveau_pushbuf *push = nv50->base.pushbuf;
287 struct nv50_vertex_stateobj *vertex = nv50->vertex;
288 struct pipe_vertex_buffer *vb;
289 struct nv50_vertex_element *ve;
290 unsigned i;
291
292 if (unlikely(vertex->need_conversion)) {
293 nv50->vbo_fifo = ~0;
294 nv50->vbo_user = 0;
295 } else {
296 nv50_prevalidate_vbufs(nv50);
297 }
298
299 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_ATTRIB(0)), vertex->num_elements);
300 for (i = 0; i < vertex->num_elements; ++i) {
301 ve = &vertex->element[i];
302 vb = &nv50->vtxbuf[ve->pipe.vertex_buffer_index];
303
304 if (likely(vb->stride) || nv50->vbo_fifo) {
305 PUSH_DATA(push, ve->state);
306 } else {
307 PUSH_DATA(push, ve->state | NV50_3D_VERTEX_ARRAY_ATTRIB_CONST);
308 nv50->vbo_fifo &= ~(1 << i);
309 }
310 }
311
312 for (i = 0; i < vertex->num_elements; ++i) {
313 struct nv04_resource *res;
314 unsigned size, offset;
315
316 ve = &vertex->element[i];
317 vb = &nv50->vtxbuf[ve->pipe.vertex_buffer_index];
318
319 if (unlikely(ve->pipe.instance_divisor)) {
320 if (!(nv50->state.instance_elts & (1 << i))) {
321 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
322 PUSH_DATA (push, 1);
323 }
324 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_DIVISOR(i)), 1);
325 PUSH_DATA (push, ve->pipe.instance_divisor);
326 } else
327 if (unlikely(nv50->state.instance_elts & (1 << i))) {
328 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
329 PUSH_DATA (push, 0);
330 }
331
332 res = nv04_resource(vb->buffer);
333
334 if (nv50->vbo_fifo || unlikely(vb->stride == 0)) {
335 if (!nv50->vbo_fifo)
336 nv50_emit_vtxattr(nv50, vb, &ve->pipe, i);
337 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 1);
338 PUSH_DATA (push, 0);
339 continue;
340 }
341
342 size = vb->buffer->width0;
343 offset = ve->pipe.src_offset + vb->buffer_offset;
344
345 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 1);
346 PUSH_DATA (push, NV50_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
347 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
348 PUSH_DATAh(push, res->address + size - 1);
349 PUSH_DATA (push, res->address + size - 1);
350 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_START_HIGH(i)), 2);
351 PUSH_DATAh(push, res->address + offset);
352 PUSH_DATA (push, res->address + offset);
353 }
354 for (; i < nv50->state.num_vtxelts; ++i) {
355 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_ATTRIB(i)), 1);
356 PUSH_DATA (push, NV50_3D_VERTEX_ATTRIB_INACTIVE);
357 if (unlikely(nv50->state.instance_elts & (1 << i))) {
358 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
359 PUSH_DATA (push, 0);
360 }
361 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FETCH(i)), 1);
362 PUSH_DATA (push, 0);
363 }
364
365 nv50->state.num_vtxelts = vertex->num_elements;
366 nv50->state.instance_elts = vertex->instance_elts;
367 }
368
369 #define NV50_PRIM_GL_CASE(n) \
370 case PIPE_PRIM_##n: return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n
371
372 static INLINE unsigned
373 nv50_prim_gl(unsigned prim)
374 {
375 switch (prim) {
376 NV50_PRIM_GL_CASE(POINTS);
377 NV50_PRIM_GL_CASE(LINES);
378 NV50_PRIM_GL_CASE(LINE_LOOP);
379 NV50_PRIM_GL_CASE(LINE_STRIP);
380 NV50_PRIM_GL_CASE(TRIANGLES);
381 NV50_PRIM_GL_CASE(TRIANGLE_STRIP);
382 NV50_PRIM_GL_CASE(TRIANGLE_FAN);
383 NV50_PRIM_GL_CASE(QUADS);
384 NV50_PRIM_GL_CASE(QUAD_STRIP);
385 NV50_PRIM_GL_CASE(POLYGON);
386 NV50_PRIM_GL_CASE(LINES_ADJACENCY);
387 NV50_PRIM_GL_CASE(LINE_STRIP_ADJACENCY);
388 NV50_PRIM_GL_CASE(TRIANGLES_ADJACENCY);
389 NV50_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY);
390 default:
391 return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS;
392 break;
393 }
394 }
395
396 static void
397 nv50_draw_arrays(struct nv50_context *nv50,
398 unsigned mode, unsigned start, unsigned count,
399 unsigned instance_count)
400 {
401 struct nouveau_pushbuf *push = nv50->base.pushbuf;
402 unsigned prim;
403
404 if (nv50->state.index_bias) {
405 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
406 PUSH_DATA (push, 0);
407 nv50->state.index_bias = 0;
408 }
409
410 prim = nv50_prim_gl(mode);
411
412 while (instance_count--) {
413 BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
414 PUSH_DATA (push, prim);
415 BEGIN_NV04(push, NV50_3D(VERTEX_BUFFER_FIRST), 2);
416 PUSH_DATA (push, start);
417 PUSH_DATA (push, count);
418 BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
419 PUSH_DATA (push, 0);
420
421 prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
422 }
423 }
424
425 static void
426 nv50_draw_elements_inline_u08(struct nouveau_pushbuf *push, uint8_t *map,
427 unsigned start, unsigned count)
428 {
429 map += start;
430
431 if (count & 3) {
432 unsigned i;
433 BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U32), count & 3);
434 for (i = 0; i < (count & 3); ++i)
435 PUSH_DATA(push, *map++);
436 count &= ~3;
437 }
438 while (count) {
439 unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 4) / 4;
440
441 BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U8), nr);
442 for (i = 0; i < nr; ++i) {
443 PUSH_DATA(push,
444 (map[3] << 24) | (map[2] << 16) | (map[1] << 8) | map[0]);
445 map += 4;
446 }
447 count -= nr * 4;
448 }
449 }
450
451 static void
452 nv50_draw_elements_inline_u16(struct nouveau_pushbuf *push, uint16_t *map,
453 unsigned start, unsigned count)
454 {
455 map += start;
456
457 if (count & 1) {
458 count &= ~1;
459 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U32), 1);
460 PUSH_DATA (push, *map++);
461 }
462 while (count) {
463 unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
464
465 BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U16), nr);
466 for (i = 0; i < nr; ++i) {
467 PUSH_DATA(push, (map[1] << 16) | map[0]);
468 map += 2;
469 }
470 count -= nr * 2;
471 }
472 }
473
474 static void
475 nv50_draw_elements_inline_u32(struct nouveau_pushbuf *push, uint32_t *map,
476 unsigned start, unsigned count)
477 {
478 map += start;
479
480 while (count) {
481 const unsigned nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN);
482
483 BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U32), nr);
484 PUSH_DATAp(push, map, nr);
485
486 map += nr;
487 count -= nr;
488 }
489 }
490
491 static void
492 nv50_draw_elements_inline_u32_short(struct nouveau_pushbuf *push, uint32_t *map,
493 unsigned start, unsigned count)
494 {
495 map += start;
496
497 if (count & 1) {
498 count--;
499 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U32), 1);
500 PUSH_DATA (push, *map++);
501 }
502 while (count) {
503 unsigned i, nr = MIN2(count, NV04_PFIFO_MAX_PACKET_LEN * 2) / 2;
504
505 BEGIN_NI04(push, NV50_3D(VB_ELEMENT_U16), nr);
506 for (i = 0; i < nr; ++i) {
507 PUSH_DATA(push, (map[1] << 16) | map[0]);
508 map += 2;
509 }
510 count -= nr * 2;
511 }
512 }
513
514 static void
515 nv50_draw_elements(struct nv50_context *nv50, boolean shorten,
516 unsigned mode, unsigned start, unsigned count,
517 unsigned instance_count, int32_t index_bias)
518 {
519 struct nouveau_pushbuf *push = nv50->base.pushbuf;
520 void *data;
521 struct nv04_resource *buf = nv04_resource(nv50->idxbuf.buffer);
522 unsigned prim;
523 const unsigned index_size = nv50->idxbuf.index_size;
524
525 prim = nv50_prim_gl(mode);
526
527 if (index_bias != nv50->state.index_bias) {
528 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_BASE), 1);
529 PUSH_DATA (push, index_bias);
530 nv50->state.index_bias = index_bias;
531 }
532
533 if (nouveau_resource_mapped_by_gpu(nv50->idxbuf.buffer)) {
534 unsigned pb_start;
535 unsigned pb_bytes;
536 const unsigned base = buf->offset;
537
538 start += nv50->idxbuf.offset >> (index_size >> 1);
539
540 while (instance_count--) {
541 BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
542 PUSH_DATA (push, prim);
543
544 nouveau_pushbuf_space(push, 8, 0, 1);
545
546 switch (index_size) {
547 case 4:
548 BEGIN_NL50(push, NV50_3D(VB_ELEMENT_U32), count);
549 nouveau_pushbuf_data(push, buf->bo, base + start * 4, count * 4);
550 break;
551 case 2:
552 pb_start = (start & ~1) * 2;
553 pb_bytes = ((start + count + 1) & ~1) * 2 - pb_start;
554
555 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U16_SETUP), 1);
556 PUSH_DATA (push, (start << 31) | count);
557 BEGIN_NL50(push, NV50_3D(VB_ELEMENT_U16), pb_bytes / 4);
558 nouveau_pushbuf_data(push, buf->bo, base + pb_start, pb_bytes);
559 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U16_SETUP), 1);
560 PUSH_DATA (push, 0);
561 break;
562 default:
563 assert(index_size == 1);
564 pb_start = start & ~3;
565 pb_bytes = ((start + count + 3) & ~3) - pb_start;
566
567 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U8_SETUP), 1);
568 PUSH_DATA (push, (start << 30) | count);
569 BEGIN_NL50(push, NV50_3D(VB_ELEMENT_U8), pb_bytes / 4);
570 nouveau_pushbuf_data(push, buf->bo, base + pb_start, pb_bytes);
571 BEGIN_NV04(push, NV50_3D(VB_ELEMENT_U8_SETUP), 1);
572 PUSH_DATA (push, 0);
573 break;
574 }
575 BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
576 PUSH_DATA (push, 0);
577
578 prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
579 }
580 } else {
581 data = nouveau_resource_map_offset(&nv50->base, buf,
582 nv50->idxbuf.offset, NOUVEAU_BO_RD);
583 if (!data)
584 return;
585
586 while (instance_count--) {
587 BEGIN_NV04(push, NV50_3D(VERTEX_BEGIN_GL), 1);
588 PUSH_DATA (push, prim);
589 switch (index_size) {
590 case 1:
591 nv50_draw_elements_inline_u08(push, data, start, count);
592 break;
593 case 2:
594 nv50_draw_elements_inline_u16(push, data, start, count);
595 break;
596 case 4:
597 if (shorten)
598 nv50_draw_elements_inline_u32_short(push, data, start, count);
599 else
600 nv50_draw_elements_inline_u32(push, data, start, count);
601 break;
602 default:
603 assert(0);
604 return;
605 }
606 BEGIN_NV04(push, NV50_3D(VERTEX_END_GL), 1);
607 PUSH_DATA (push, 0);
608
609 prim |= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
610 }
611 }
612 }
613
614 static void
615 nv50_draw_vbo_kick_notify(struct nouveau_pushbuf *chan)
616 {
617 struct nv50_screen *screen = chan->user_priv;
618
619 nouveau_fence_update(&screen->base, TRUE);
620
621 nv50_bufctx_fence(screen->cur_ctx->bufctx_3d, TRUE);
622 }
623
624 void
625 nv50_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
626 {
627 struct nv50_context *nv50 = nv50_context(pipe);
628 struct nouveau_pushbuf *push = nv50->base.pushbuf;
629
630 /* For picking only a few vertices from a large user buffer, push is better,
631 * if index count is larger and we expect repeated vertices, suggest upload.
632 */
633 nv50->vbo_push_hint = /* the 64 is heuristic */
634 !(info->indexed &&
635 ((info->max_index - info->min_index + 64) < info->count));
636
637 nv50->vbo_min_index = info->min_index;
638 nv50->vbo_max_index = info->max_index;
639
640 if (nv50->vbo_push_hint != !!nv50->vbo_fifo)
641 nv50->dirty |= NV50_NEW_ARRAYS;
642
643 if (nv50->vbo_user && !(nv50->dirty & (NV50_NEW_VERTEX | NV50_NEW_ARRAYS)))
644 nv50_update_user_vbufs(nv50);
645
646 nv50_state_validate(nv50, ~0, 8); /* 8 as minimum, we use flush_notify */
647
648 push->kick_notify = nv50_draw_vbo_kick_notify;
649
650 if (nv50->vbo_fifo) {
651 nv50_push_vbo(nv50, info);
652 push->kick_notify = nv50_default_kick_notify;
653 nouveau_pushbuf_bufctx(push, NULL);
654 return;
655 }
656
657 if (nv50->state.instance_base != info->start_instance) {
658 nv50->state.instance_base = info->start_instance;
659 /* NOTE: this does not affect the shader input, should it ? */
660 BEGIN_NV04(push, NV50_3D(VB_INSTANCE_BASE), 1);
661 PUSH_DATA (push, info->start_instance);
662 }
663
664 if (nv50->base.vbo_dirty) {
665 BEGIN_NV04(push, NV50_3D(VERTEX_ARRAY_FLUSH), 1);
666 PUSH_DATA (push, 0);
667 nv50->base.vbo_dirty = FALSE;
668 }
669
670 if (!info->indexed) {
671 nv50_draw_arrays(nv50,
672 info->mode, info->start, info->count,
673 info->instance_count);
674 } else {
675 boolean shorten = info->max_index <= 65535;
676
677 assert(nv50->idxbuf.buffer);
678
679 if (info->primitive_restart != nv50->state.prim_restart) {
680 if (info->primitive_restart) {
681 BEGIN_NV04(push, NV50_3D(PRIM_RESTART_ENABLE), 2);
682 PUSH_DATA (push, 1);
683 PUSH_DATA (push, info->restart_index);
684
685 if (info->restart_index > 65535)
686 shorten = FALSE;
687 } else {
688 BEGIN_NV04(push, NV50_3D(PRIM_RESTART_ENABLE), 1);
689 PUSH_DATA (push, 0);
690 }
691 nv50->state.prim_restart = info->primitive_restart;
692 } else
693 if (info->primitive_restart) {
694 BEGIN_NV04(push, NV50_3D(PRIM_RESTART_INDEX), 1);
695 PUSH_DATA (push, info->restart_index);
696
697 if (info->restart_index > 65535)
698 shorten = FALSE;
699 }
700
701 nv50_draw_elements(nv50, shorten,
702 info->mode, info->start, info->count,
703 info->instance_count, info->index_bias);
704 }
705 push->kick_notify = nv50_default_kick_notify;
706
707 nv50_release_user_vbufs(nv50);
708
709 nouveau_pushbuf_bufctx(push, NULL);
710 }