2 * Copyright 2012 Red Hat Inc.
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:
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
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 OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
26 #include "util/u_format.h"
27 #include "util/u_inlines.h"
28 #include "translate/translate.h"
30 #include "nouveau_fence.h"
31 #include "nv_object.xml.h"
32 #include "nv30/nv30-40_3d.xml.h"
33 #include "nv30/nv30_context.h"
34 #include "nv30/nv30_format.h"
37 nv30_emit_vtxattr(struct nv30_context
*nv30
, struct pipe_vertex_buffer
*vb
,
38 struct pipe_vertex_element
*ve
, unsigned attr
)
40 const unsigned nc
= util_format_get_nr_components(ve
->src_format
);
41 struct nouveau_pushbuf
*push
= nv30
->base
.pushbuf
;
42 struct nv04_resource
*res
= nv04_resource(vb
->buffer
);
43 const struct util_format_description
*desc
=
44 util_format_description(ve
->src_format
);
48 data
= nouveau_resource_map_offset(&nv30
->base
, res
, vb
->buffer_offset
+
49 ve
->src_offset
, NOUVEAU_BO_RD
);
51 desc
->unpack_rgba_float(v
, 0, data
, 0, 1, 1);
55 BEGIN_NV04(push
, NV30_3D(VTX_ATTR_4F(attr
)), 4);
56 PUSH_DATAf(push
, v
[0]);
57 PUSH_DATAf(push
, v
[1]);
58 PUSH_DATAf(push
, v
[2]);
59 PUSH_DATAf(push
, v
[3]);
62 BEGIN_NV04(push
, NV30_3D(VTX_ATTR_3F(attr
)), 3);
63 PUSH_DATAf(push
, v
[0]);
64 PUSH_DATAf(push
, v
[1]);
65 PUSH_DATAf(push
, v
[2]);
68 BEGIN_NV04(push
, NV30_3D(VTX_ATTR_2F(attr
)), 2);
69 PUSH_DATAf(push
, v
[0]);
70 PUSH_DATAf(push
, v
[1]);
73 BEGIN_NV04(push
, NV30_3D(VTX_ATTR_1F(attr
)), 1);
74 PUSH_DATAf(push
, v
[0]);
83 nv30_vbuf_range(struct nv30_context
*nv30
, int vbi
,
84 uint32_t *base
, uint32_t *size
)
86 assert(nv30
->vbo_max_index
!= ~0);
87 *base
= nv30
->vbo_min_index
* nv30
->vtxbuf
[vbi
].stride
;
88 *size
= (nv30
->vbo_max_index
-
89 nv30
->vbo_min_index
+ 1) * nv30
->vtxbuf
[vbi
].stride
;
93 nv30_prevalidate_vbufs(struct nv30_context
*nv30
)
95 struct pipe_vertex_buffer
*vb
;
96 struct nv04_resource
*buf
;
100 nv30
->vbo_fifo
= nv30
->vbo_user
= 0;
102 for (i
= 0; i
< nv30
->num_vtxbufs
; i
++) {
103 vb
= &nv30
->vtxbuf
[i
];
104 if (!vb
->stride
|| !vb
->buffer
) /* NOTE: user_buffer not implemented */
106 buf
= nv04_resource(vb
->buffer
);
108 /* NOTE: user buffers with temporary storage count as mapped by GPU */
109 if (!nouveau_resource_mapped_by_gpu(vb
->buffer
)) {
110 if (nv30
->vbo_push_hint
) {
114 if (buf
->status
& NOUVEAU_BUFFER_STATUS_USER_MEMORY
) {
115 nv30
->vbo_user
|= 1 << i
;
116 assert(vb
->stride
> vb
->buffer_offset
);
117 nv30_vbuf_range(nv30
, i
, &base
, &size
);
118 nouveau_user_buffer_upload(&nv30
->base
, buf
, base
, size
);
120 nouveau_buffer_migrate(&nv30
->base
, buf
, NOUVEAU_BO_GART
);
122 nv30
->base
.vbo_dirty
= true;
129 nv30_update_user_vbufs(struct nv30_context
*nv30
)
131 struct nouveau_pushbuf
*push
= nv30
->base
.pushbuf
;
132 uint32_t base
, offset
, size
;
134 uint32_t written
= 0;
136 for (i
= 0; i
< nv30
->vertex
->num_elements
; i
++) {
137 struct pipe_vertex_element
*ve
= &nv30
->vertex
->pipe
[i
];
138 const int b
= ve
->vertex_buffer_index
;
139 struct pipe_vertex_buffer
*vb
= &nv30
->vtxbuf
[b
];
140 struct nv04_resource
*buf
= nv04_resource(vb
->buffer
);
142 if (!(nv30
->vbo_user
& (1 << b
)))
146 nv30_emit_vtxattr(nv30
, vb
, ve
, i
);
149 nv30_vbuf_range(nv30
, b
, &base
, &size
);
151 if (!(written
& (1 << b
))) {
153 nouveau_user_buffer_upload(&nv30
->base
, buf
, base
, size
);
156 offset
= vb
->buffer_offset
+ ve
->src_offset
;
158 BEGIN_NV04(push
, NV30_3D(VTXBUF(i
)), 1);
159 PUSH_RESRC(push
, NV30_3D(VTXBUF(i
)), BUFCTX_VTXTMP
, buf
, offset
,
160 NOUVEAU_BO_LOW
| NOUVEAU_BO_RD
,
161 0, NV30_3D_VTXBUF_DMA1
);
163 nv30
->base
.vbo_dirty
= true;
167 nv30_release_user_vbufs(struct nv30_context
*nv30
)
169 uint32_t vbo_user
= nv30
->vbo_user
;
172 int i
= ffs(vbo_user
) - 1;
173 vbo_user
&= ~(1 << i
);
175 nouveau_buffer_release_gpu_storage(nv04_resource(nv30
->vtxbuf
[i
].buffer
));
178 nouveau_bufctx_reset(nv30
->bufctx
, BUFCTX_VTXTMP
);
182 nv30_vbo_validate(struct nv30_context
*nv30
)
184 struct nouveau_pushbuf
*push
= nv30
->base
.pushbuf
;
185 struct nv30_vertex_stateobj
*vertex
= nv30
->vertex
;
186 struct pipe_vertex_element
*ve
;
187 struct pipe_vertex_buffer
*vb
;
188 unsigned i
, redefine
;
190 nouveau_bufctx_reset(nv30
->bufctx
, BUFCTX_VTXBUF
);
191 if (!nv30
->vertex
|| nv30
->draw_flags
)
194 #ifdef PIPE_ARCH_BIG_ENDIAN
195 if (1) { /* Figure out where the buffers are getting messed up */
197 if (unlikely(vertex
->need_conversion
)) {
202 nv30_prevalidate_vbufs(nv30
);
205 if (!PUSH_SPACE(push
, 128))
208 redefine
= MAX2(vertex
->num_elements
, nv30
->state
.num_vtxelts
);
212 BEGIN_NV04(push
, NV30_3D(VTXFMT(0)), redefine
);
214 for (i
= 0; i
< vertex
->num_elements
; i
++) {
215 ve
= &vertex
->pipe
[i
];
216 vb
= &nv30
->vtxbuf
[ve
->vertex_buffer_index
];
218 if (likely(vb
->stride
) || nv30
->vbo_fifo
)
219 PUSH_DATA (push
, (vb
->stride
<< 8) | vertex
->element
[i
].state
);
221 PUSH_DATA (push
, NV30_3D_VTXFMT_TYPE_V32_FLOAT
);
224 for (; i
< nv30
->state
.num_vtxelts
; i
++) {
225 PUSH_DATA (push
, NV30_3D_VTXFMT_TYPE_V32_FLOAT
);
228 for (i
= 0; i
< vertex
->num_elements
; i
++) {
229 struct nv04_resource
*res
;
233 ve
= &vertex
->pipe
[i
];
234 vb
= &nv30
->vtxbuf
[ve
->vertex_buffer_index
];
235 user
= (nv30
->vbo_user
& (1 << ve
->vertex_buffer_index
));
237 res
= nv04_resource(vb
->buffer
);
239 if (nv30
->vbo_fifo
|| unlikely(vb
->stride
== 0)) {
241 nv30_emit_vtxattr(nv30
, vb
, ve
, i
);
245 offset
= ve
->src_offset
+ vb
->buffer_offset
;
247 BEGIN_NV04(push
, NV30_3D(VTXBUF(i
)), 1);
248 PUSH_RESRC(push
, NV30_3D(VTXBUF(i
)), user
? BUFCTX_VTXTMP
: BUFCTX_VTXBUF
,
249 res
, offset
, NOUVEAU_BO_LOW
| NOUVEAU_BO_RD
,
250 0, NV30_3D_VTXBUF_DMA1
);
253 nv30
->state
.num_vtxelts
= vertex
->num_elements
;
257 nv30_vertex_state_create(struct pipe_context
*pipe
, unsigned num_elements
,
258 const struct pipe_vertex_element
*elements
)
260 struct nv30_vertex_stateobj
*so
;
261 struct translate_key transkey
;
264 so
= MALLOC(sizeof(*so
) + sizeof(*so
->element
) * num_elements
);
267 memcpy(so
->pipe
, elements
, sizeof(*elements
) * num_elements
);
268 so
->num_elements
= num_elements
;
269 so
->need_conversion
= false;
271 transkey
.nr_elements
= 0;
272 transkey
.output_stride
= 0;
274 for (i
= 0; i
< num_elements
; i
++) {
275 const struct pipe_vertex_element
*ve
= &elements
[i
];
276 const unsigned vbi
= ve
->vertex_buffer_index
;
277 enum pipe_format fmt
= ve
->src_format
;
279 so
->element
[i
].state
= nv30_vtxfmt(pipe
->screen
, fmt
)->hw
;
280 if (!so
->element
[i
].state
) {
281 switch (util_format_get_nr_components(fmt
)) {
282 case 1: fmt
= PIPE_FORMAT_R32_FLOAT
; break;
283 case 2: fmt
= PIPE_FORMAT_R32G32_FLOAT
; break;
284 case 3: fmt
= PIPE_FORMAT_R32G32B32_FLOAT
; break;
285 case 4: fmt
= PIPE_FORMAT_R32G32B32A32_FLOAT
; break;
291 so
->element
[i
].state
= nv30_vtxfmt(pipe
->screen
, fmt
)->hw
;
292 so
->need_conversion
= true;
296 unsigned j
= transkey
.nr_elements
++;
298 transkey
.element
[j
].type
= TRANSLATE_ELEMENT_NORMAL
;
299 transkey
.element
[j
].input_format
= ve
->src_format
;
300 transkey
.element
[j
].input_buffer
= vbi
;
301 transkey
.element
[j
].input_offset
= ve
->src_offset
;
302 transkey
.element
[j
].instance_divisor
= ve
->instance_divisor
;
304 transkey
.element
[j
].output_format
= fmt
;
305 transkey
.element
[j
].output_offset
= transkey
.output_stride
;
306 transkey
.output_stride
+= (util_format_get_stride(fmt
, 1) + 3) & ~3;
310 so
->translate
= translate_create(&transkey
);
311 so
->vtx_size
= transkey
.output_stride
/ 4;
312 so
->vtx_per_packet_max
= NV04_PFIFO_MAX_PACKET_LEN
/ MAX2(so
->vtx_size
, 1);
317 nv30_vertex_state_delete(struct pipe_context
*pipe
, void *hwcso
)
319 struct nv30_vertex_stateobj
*so
= hwcso
;
322 so
->translate
->release(so
->translate
);
327 nv30_vertex_state_bind(struct pipe_context
*pipe
, void *hwcso
)
329 struct nv30_context
*nv30
= nv30_context(pipe
);
331 nv30
->vertex
= hwcso
;
332 nv30
->dirty
|= NV30_NEW_VERTEX
;
336 nv30_draw_arrays(struct nv30_context
*nv30
,
337 unsigned mode
, unsigned start
, unsigned count
,
338 unsigned instance_count
)
340 struct nouveau_pushbuf
*push
= nv30
->base
.pushbuf
;
343 prim
= nv30_prim_gl(mode
);
345 BEGIN_NV04(push
, NV30_3D(VERTEX_BEGIN_END
), 1);
346 PUSH_DATA (push
, prim
);
348 const unsigned mpush
= 2047 * 256;
349 unsigned npush
= (count
> mpush
) ? mpush
: count
;
350 unsigned wpush
= ((npush
+ 255) & ~255) >> 8;
354 BEGIN_NI04(push
, NV30_3D(VB_VERTEX_BATCH
), wpush
);
355 while (npush
>= 256) {
356 PUSH_DATA (push
, 0xff000000 | start
);
362 PUSH_DATA (push
, ((npush
- 1) << 24) | start
);
364 BEGIN_NV04(push
, NV30_3D(VERTEX_BEGIN_END
), 1);
365 PUSH_DATA (push
, NV30_3D_VERTEX_BEGIN_END_STOP
);
369 nv30_draw_elements_inline_u08(struct nouveau_pushbuf
*push
, const uint8_t *map
,
370 unsigned start
, unsigned count
)
375 BEGIN_NV04(push
, NV30_3D(VB_ELEMENT_U32
), 1);
376 PUSH_DATA (push
, *map
++);
381 unsigned npush
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
);
384 BEGIN_NI04(push
, NV30_3D(VB_ELEMENT_U16
), npush
);
386 PUSH_DATA (push
, (map
[1] << 16) | map
[0]);
394 nv30_draw_elements_inline_u16(struct nouveau_pushbuf
*push
, const uint16_t *map
,
395 unsigned start
, unsigned count
)
400 BEGIN_NV04(push
, NV30_3D(VB_ELEMENT_U32
), 1);
401 PUSH_DATA (push
, *map
++);
406 unsigned npush
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
);
409 BEGIN_NI04(push
, NV30_3D(VB_ELEMENT_U16
), npush
);
411 PUSH_DATA (push
, (map
[1] << 16) | map
[0]);
418 nv30_draw_elements_inline_u32(struct nouveau_pushbuf
*push
, const uint32_t *map
,
419 unsigned start
, unsigned count
)
424 const unsigned nr
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
);
426 BEGIN_NI04(push
, NV30_3D(VB_ELEMENT_U32
), nr
);
427 PUSH_DATAp(push
, map
, nr
);
435 nv30_draw_elements_inline_u32_short(struct nouveau_pushbuf
*push
,
437 unsigned start
, unsigned count
)
442 BEGIN_NV04(push
, NV30_3D(VB_ELEMENT_U32
), 1);
443 PUSH_DATA (push
, *map
++);
448 unsigned npush
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
);;
451 BEGIN_NI04(push
, NV30_3D(VB_ELEMENT_U16
), npush
);
453 PUSH_DATA (push
, (map
[1] << 16) | map
[0]);
460 nv30_draw_elements(struct nv30_context
*nv30
, bool shorten
,
461 unsigned mode
, unsigned start
, unsigned count
,
462 unsigned instance_count
, int32_t index_bias
)
464 const unsigned index_size
= nv30
->idxbuf
.index_size
;
465 struct nouveau_pushbuf
*push
= nv30
->base
.pushbuf
;
466 struct nouveau_object
*eng3d
= nv30
->screen
->eng3d
;
467 unsigned prim
= nv30_prim_gl(mode
);
469 if (eng3d
->oclass
>= NV40_3D_CLASS
&& index_bias
!= nv30
->state
.index_bias
) {
470 BEGIN_NV04(push
, NV40_3D(VB_ELEMENT_BASE
), 1);
471 PUSH_DATA (push
, index_bias
);
472 nv30
->state
.index_bias
= index_bias
;
475 if (eng3d
->oclass
== NV40_3D_CLASS
&& index_size
> 1 &&
476 nv30
->idxbuf
.buffer
) {
477 struct nv04_resource
*res
= nv04_resource(nv30
->idxbuf
.buffer
);
478 unsigned offset
= nv30
->idxbuf
.offset
;
480 assert(nouveau_resource_mapped_by_gpu(&res
->base
));
482 BEGIN_NV04(push
, NV30_3D(IDXBUF_OFFSET
), 2);
483 PUSH_RESRC(push
, NV30_3D(IDXBUF_OFFSET
), BUFCTX_IDXBUF
, res
, offset
,
484 NOUVEAU_BO_LOW
| NOUVEAU_BO_RD
, 0, 0);
485 PUSH_MTHD (push
, NV30_3D(IDXBUF_FORMAT
), BUFCTX_IDXBUF
, res
->bo
,
486 (index_size
== 2) ? 0x00000010 : 0x00000000,
487 res
->domain
| NOUVEAU_BO_RD
,
488 0, NV30_3D_IDXBUF_FORMAT_DMA1
);
489 BEGIN_NV04(push
, NV30_3D(VERTEX_BEGIN_END
), 1);
490 PUSH_DATA (push
, prim
);
492 const unsigned mpush
= 2047 * 256;
493 unsigned npush
= (count
> mpush
) ? mpush
: count
;
494 unsigned wpush
= ((npush
+ 255) & ~255) >> 8;
498 BEGIN_NI04(push
, NV30_3D(VB_INDEX_BATCH
), wpush
);
499 while (npush
>= 256) {
500 PUSH_DATA (push
, 0xff000000 | start
);
506 PUSH_DATA (push
, ((npush
- 1) << 24) | start
);
508 BEGIN_NV04(push
, NV30_3D(VERTEX_BEGIN_END
), 1);
509 PUSH_DATA (push
, NV30_3D_VERTEX_BEGIN_END_STOP
);
510 PUSH_RESET(push
, BUFCTX_IDXBUF
);
513 if (nv30
->idxbuf
.buffer
)
514 data
= nouveau_resource_map_offset(&nv30
->base
,
515 nv04_resource(nv30
->idxbuf
.buffer
),
516 nv30
->idxbuf
.offset
, NOUVEAU_BO_RD
);
518 data
= nv30
->idxbuf
.user_buffer
;
522 BEGIN_NV04(push
, NV30_3D(VERTEX_BEGIN_END
), 1);
523 PUSH_DATA (push
, prim
);
524 switch (index_size
) {
526 nv30_draw_elements_inline_u08(push
, data
, start
, count
);
529 nv30_draw_elements_inline_u16(push
, data
, start
, count
);
533 nv30_draw_elements_inline_u32_short(push
, data
, start
, count
);
535 nv30_draw_elements_inline_u32(push
, data
, start
, count
);
541 BEGIN_NV04(push
, NV30_3D(VERTEX_BEGIN_END
), 1);
542 PUSH_DATA (push
, NV30_3D_VERTEX_BEGIN_END_STOP
);
547 nv30_draw_vbo(struct pipe_context
*pipe
, const struct pipe_draw_info
*info
)
549 struct nv30_context
*nv30
= nv30_context(pipe
);
550 struct nouveau_pushbuf
*push
= nv30
->base
.pushbuf
;
553 /* For picking only a few vertices from a large user buffer, push is better,
554 * if index count is larger and we expect repeated vertices, suggest upload.
556 nv30
->vbo_push_hint
= /* the 64 is heuristic */
558 ((info
->max_index
- info
->min_index
+ 64) < info
->count
));
560 nv30
->vbo_min_index
= info
->min_index
;
561 nv30
->vbo_max_index
= info
->max_index
;
563 if (nv30
->vbo_push_hint
!= !!nv30
->vbo_fifo
)
564 nv30
->dirty
|= NV30_NEW_ARRAYS
;
566 push
->user_priv
= &nv30
->bufctx
;
567 if (nv30
->vbo_user
&& !(nv30
->dirty
& (NV30_NEW_VERTEX
| NV30_NEW_ARRAYS
)))
568 nv30_update_user_vbufs(nv30
);
570 nv30_state_validate(nv30
, ~0, true);
571 if (nv30
->draw_flags
) {
572 nv30_render_vbo(pipe
, info
);
575 if (nv30
->vbo_fifo
) {
576 nv30_push_vbo(nv30
, info
);
580 for (i
= 0; i
< nv30
->num_vtxbufs
&& !nv30
->base
.vbo_dirty
; ++i
) {
581 if (!nv30
->vtxbuf
[i
].buffer
)
583 if (nv30
->vtxbuf
[i
].buffer
->flags
& PIPE_RESOURCE_FLAG_MAP_COHERENT
)
584 nv30
->base
.vbo_dirty
= true;
587 if (!nv30
->base
.vbo_dirty
&& nv30
->idxbuf
.buffer
&&
588 nv30
->idxbuf
.buffer
->flags
& PIPE_RESOURCE_FLAG_MAP_COHERENT
)
589 nv30
->base
.vbo_dirty
= true;
591 if (nv30
->base
.vbo_dirty
) {
592 BEGIN_NV04(push
, NV30_3D(VTX_CACHE_INVALIDATE_1710
), 1);
594 nv30
->base
.vbo_dirty
= false;
597 if (!info
->indexed
) {
598 nv30_draw_arrays(nv30
,
599 info
->mode
, info
->start
, info
->count
,
600 info
->instance_count
);
602 bool shorten
= info
->max_index
<= 65535;
604 if (info
->primitive_restart
!= nv30
->state
.prim_restart
) {
605 if (info
->primitive_restart
) {
606 BEGIN_NV04(push
, NV40_3D(PRIM_RESTART_ENABLE
), 2);
608 PUSH_DATA (push
, info
->restart_index
);
610 if (info
->restart_index
> 65535)
613 BEGIN_NV04(push
, NV40_3D(PRIM_RESTART_ENABLE
), 1);
616 nv30
->state
.prim_restart
= info
->primitive_restart
;
618 if (info
->primitive_restart
) {
619 BEGIN_NV04(push
, NV40_3D(PRIM_RESTART_INDEX
), 1);
620 PUSH_DATA (push
, info
->restart_index
);
622 if (info
->restart_index
> 65535)
626 nv30_draw_elements(nv30
, shorten
,
627 info
->mode
, info
->start
, info
->count
,
628 info
->instance_count
, info
->index_bias
);
631 nv30_state_release(nv30
);
632 nv30_release_user_vbufs(nv30
);
636 nv30_vbo_init(struct pipe_context
*pipe
)
638 pipe
->create_vertex_elements_state
= nv30_vertex_state_create
;
639 pipe
->delete_vertex_elements_state
= nv30_vertex_state_delete
;
640 pipe
->bind_vertex_elements_state
= nv30_vertex_state_bind
;
641 pipe
->draw_vbo
= nv30_draw_vbo
;