2 * Copyright 2010 Christoph Bumiller
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.
23 #include "pipe/p_context.h"
24 #include "pipe/p_state.h"
25 #include "util/u_inlines.h"
26 #include "util/format/u_format.h"
27 #include "translate/translate.h"
29 #include "nv50/nv50_context.h"
30 #include "nv50/nv50_query_hw.h"
31 #include "nv50/nv50_resource.h"
33 #include "nv50/nv50_3d.xml.h"
36 nv50_vertex_state_delete(struct pipe_context
*pipe
,
39 struct nv50_vertex_stateobj
*so
= hwcso
;
42 so
->translate
->release(so
->translate
);
47 nv50_vertex_state_create(struct pipe_context
*pipe
,
48 unsigned num_elements
,
49 const struct pipe_vertex_element
*elements
)
51 struct nv50_vertex_stateobj
*so
;
52 struct translate_key transkey
;
55 so
= MALLOC(sizeof(*so
) +
56 num_elements
* sizeof(struct nv50_vertex_element
));
59 so
->num_elements
= num_elements
;
60 so
->instance_elts
= 0;
61 so
->instance_bufs
= 0;
62 so
->need_conversion
= false;
64 memset(so
->vb_access_size
, 0, sizeof(so
->vb_access_size
));
66 for (i
= 0; i
< PIPE_MAX_ATTRIBS
; ++i
)
67 so
->min_instance_div
[i
] = 0xffffffff;
69 transkey
.nr_elements
= 0;
70 transkey
.output_stride
= 0;
72 for (i
= 0; i
< num_elements
; ++i
) {
73 const struct pipe_vertex_element
*ve
= &elements
[i
];
74 const unsigned vbi
= ve
->vertex_buffer_index
;
76 enum pipe_format fmt
= ve
->src_format
;
78 so
->element
[i
].pipe
= elements
[i
];
79 so
->element
[i
].state
= nv50_vertex_format
[fmt
].vtx
;
81 if (!so
->element
[i
].state
) {
82 switch (util_format_get_nr_components(fmt
)) {
83 case 1: fmt
= PIPE_FORMAT_R32_FLOAT
; break;
84 case 2: fmt
= PIPE_FORMAT_R32G32_FLOAT
; break;
85 case 3: fmt
= PIPE_FORMAT_R32G32B32_FLOAT
; break;
86 case 4: fmt
= PIPE_FORMAT_R32G32B32A32_FLOAT
; break;
92 so
->element
[i
].state
= nv50_vertex_format
[fmt
].vtx
;
93 so
->need_conversion
= true;
94 pipe_debug_message(&nouveau_context(pipe
)->debug
, FALLBACK
,
95 "Converting vertex element %d, no hw format %s",
96 i
, util_format_name(ve
->src_format
));
98 so
->element
[i
].state
|= i
;
100 size
= util_format_get_blocksize(fmt
);
101 if (so
->vb_access_size
[vbi
] < (ve
->src_offset
+ size
))
102 so
->vb_access_size
[vbi
] = ve
->src_offset
+ size
;
105 unsigned j
= transkey
.nr_elements
++;
107 transkey
.element
[j
].type
= TRANSLATE_ELEMENT_NORMAL
;
108 transkey
.element
[j
].input_format
= ve
->src_format
;
109 transkey
.element
[j
].input_buffer
= vbi
;
110 transkey
.element
[j
].input_offset
= ve
->src_offset
;
111 transkey
.element
[j
].instance_divisor
= ve
->instance_divisor
;
113 transkey
.element
[j
].output_format
= fmt
;
114 transkey
.element
[j
].output_offset
= transkey
.output_stride
;
115 transkey
.output_stride
+= (util_format_get_stride(fmt
, 1) + 3) & ~3;
117 if (unlikely(ve
->instance_divisor
)) {
118 so
->instance_elts
|= 1 << i
;
119 so
->instance_bufs
|= 1 << vbi
;
120 if (ve
->instance_divisor
< so
->min_instance_div
[vbi
])
121 so
->min_instance_div
[vbi
] = ve
->instance_divisor
;
126 so
->translate
= translate_create(&transkey
);
127 so
->vertex_size
= transkey
.output_stride
/ 4;
128 so
->packet_vertex_limit
= NV04_PFIFO_MAX_PACKET_LEN
/
129 MAX2(so
->vertex_size
, 1);
134 #define NV50_3D_VERTEX_ATTRIB_INACTIVE \
135 NV50_3D_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT | \
136 NV50_3D_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32 | \
137 NV50_3D_VERTEX_ARRAY_ATTRIB_CONST
140 nv50_emit_vtxattr(struct nv50_context
*nv50
, struct pipe_vertex_buffer
*vb
,
141 struct pipe_vertex_element
*ve
, unsigned attr
)
143 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
144 const void *data
= (const uint8_t *)vb
->buffer
.user
+ ve
->src_offset
;
146 const unsigned nc
= util_format_get_nr_components(ve
->src_format
);
148 assert(vb
->is_user_buffer
);
150 util_format_unpack_rgba(ve
->src_format
, v
, data
, 1);
154 BEGIN_NV04(push
, NV50_3D(VTX_ATTR_4F_X(attr
)), 4);
155 PUSH_DATAf(push
, v
[0]);
156 PUSH_DATAf(push
, v
[1]);
157 PUSH_DATAf(push
, v
[2]);
158 PUSH_DATAf(push
, v
[3]);
161 BEGIN_NV04(push
, NV50_3D(VTX_ATTR_3F_X(attr
)), 3);
162 PUSH_DATAf(push
, v
[0]);
163 PUSH_DATAf(push
, v
[1]);
164 PUSH_DATAf(push
, v
[2]);
167 BEGIN_NV04(push
, NV50_3D(VTX_ATTR_2F_X(attr
)), 2);
168 PUSH_DATAf(push
, v
[0]);
169 PUSH_DATAf(push
, v
[1]);
172 if (attr
== nv50
->vertprog
->vp
.edgeflag
) {
173 BEGIN_NV04(push
, NV50_3D(EDGEFLAG
), 1);
174 PUSH_DATA (push
, v
[0] ? 1 : 0);
176 BEGIN_NV04(push
, NV50_3D(VTX_ATTR_1F(attr
)), 1);
177 PUSH_DATAf(push
, v
[0]);
186 nv50_user_vbuf_range(struct nv50_context
*nv50
, unsigned vbi
,
187 uint32_t *base
, uint32_t *size
)
189 assert(vbi
< PIPE_MAX_ATTRIBS
);
190 if (unlikely(nv50
->vertex
->instance_bufs
& (1 << vbi
))) {
191 /* TODO: use min and max instance divisor to get a proper range */
193 *size
= nv50
->vtxbuf
[vbi
].buffer
.resource
->width0
;
195 /* NOTE: if there are user buffers, we *must* have index bounds */
196 assert(nv50
->vb_elt_limit
!= ~0);
197 *base
= nv50
->vb_elt_first
* nv50
->vtxbuf
[vbi
].stride
;
198 *size
= nv50
->vb_elt_limit
* nv50
->vtxbuf
[vbi
].stride
+
199 nv50
->vertex
->vb_access_size
[vbi
];
204 nv50_upload_user_buffers(struct nv50_context
*nv50
,
205 uint64_t addrs
[], uint32_t limits
[])
209 assert(nv50
->num_vtxbufs
<= PIPE_MAX_ATTRIBS
);
210 for (b
= 0; b
< nv50
->num_vtxbufs
; ++b
) {
211 struct nouveau_bo
*bo
;
212 const struct pipe_vertex_buffer
*vb
= &nv50
->vtxbuf
[b
];
215 if (!(nv50
->vbo_user
& (1 << b
)) || !vb
->stride
)
217 nv50_user_vbuf_range(nv50
, b
, &base
, &size
);
219 limits
[b
] = base
+ size
- 1;
220 addrs
[b
] = nouveau_scratch_data(&nv50
->base
, vb
->buffer
.user
, base
, size
,
223 BCTX_REFN_bo(nv50
->bufctx_3d
, 3D_VERTEX_TMP
, NOUVEAU_BO_GART
|
226 nv50
->base
.vbo_dirty
= true;
230 nv50_update_user_vbufs(struct nv50_context
*nv50
)
232 uint64_t address
[PIPE_MAX_ATTRIBS
];
233 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
235 uint32_t written
= 0;
237 for (i
= 0; i
< nv50
->vertex
->num_elements
; ++i
) {
238 struct pipe_vertex_element
*ve
= &nv50
->vertex
->element
[i
].pipe
;
239 const unsigned b
= ve
->vertex_buffer_index
;
240 struct pipe_vertex_buffer
*vb
;
243 assert(b
< PIPE_MAX_ATTRIBS
);
244 vb
= &nv50
->vtxbuf
[b
];
246 if (!(nv50
->vbo_user
& (1 << b
)))
250 nv50_emit_vtxattr(nv50
, vb
, ve
, i
);
253 nv50_user_vbuf_range(nv50
, b
, &base
, &size
);
255 if (!(written
& (1 << b
))) {
256 struct nouveau_bo
*bo
;
257 const uint32_t bo_flags
= NOUVEAU_BO_GART
| NOUVEAU_BO_RD
;
259 address
[b
] = nouveau_scratch_data(&nv50
->base
, vb
->buffer
.user
,
262 BCTX_REFN_bo(nv50
->bufctx_3d
, 3D_VERTEX_TMP
, bo_flags
, bo
);
265 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_LIMIT_HIGH(i
)), 2);
266 PUSH_DATAh(push
, address
[b
] + base
+ size
- 1);
267 PUSH_DATA (push
, address
[b
] + base
+ size
- 1);
268 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_START_HIGH(i
)), 2);
269 PUSH_DATAh(push
, address
[b
] + ve
->src_offset
);
270 PUSH_DATA (push
, address
[b
] + ve
->src_offset
);
272 nv50
->base
.vbo_dirty
= true;
276 nv50_release_user_vbufs(struct nv50_context
*nv50
)
278 if (nv50
->vbo_user
) {
279 nouveau_bufctx_reset(nv50
->bufctx_3d
, NV50_BIND_3D_VERTEX_TMP
);
280 nouveau_scratch_done(&nv50
->base
);
285 nv50_vertex_arrays_validate(struct nv50_context
*nv50
)
287 uint64_t addrs
[PIPE_MAX_ATTRIBS
];
288 uint32_t limits
[PIPE_MAX_ATTRIBS
];
289 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
290 struct nv50_vertex_stateobj
*vertex
= nv50
->vertex
;
291 struct pipe_vertex_buffer
*vb
;
292 struct nv50_vertex_element
*ve
;
296 const unsigned n
= MAX2(vertex
->num_elements
, nv50
->state
.num_vtxelts
);
298 if (unlikely(vertex
->need_conversion
))
301 if (nv50
->vbo_user
& ~nv50
->vbo_constant
)
302 nv50
->vbo_fifo
= nv50
->vbo_push_hint
? ~0 : 0;
306 if (!nv50
->vbo_fifo
) {
307 /* if vertex buffer was written by GPU - flush VBO cache */
308 assert(nv50
->num_vtxbufs
<= PIPE_MAX_ATTRIBS
);
309 for (i
= 0; i
< nv50
->num_vtxbufs
; ++i
) {
310 struct nv04_resource
*buf
= nv04_resource(nv50
->vtxbuf
[i
].buffer
.resource
);
311 if (!nv50
->vtxbuf
[i
].is_user_buffer
&&
312 buf
&& buf
->status
& NOUVEAU_BUFFER_STATUS_GPU_WRITING
) {
313 buf
->status
&= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING
;
314 nv50
->base
.vbo_dirty
= true;
319 /* update vertex format state */
320 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_ATTRIB(0)), n
);
321 if (nv50
->vbo_fifo
) {
322 nv50
->state
.num_vtxelts
= vertex
->num_elements
;
323 for (i
= 0; i
< vertex
->num_elements
; ++i
)
324 PUSH_DATA (push
, vertex
->element
[i
].state
);
326 PUSH_DATA (push
, NV50_3D_VERTEX_ATTRIB_INACTIVE
);
327 for (i
= 0; i
< n
; ++i
) {
328 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FETCH(i
)), 1);
333 for (i
= 0; i
< vertex
->num_elements
; ++i
) {
334 const unsigned b
= vertex
->element
[i
].pipe
.vertex_buffer_index
;
336 assert(b
< PIPE_MAX_ATTRIBS
);
337 ve
= &vertex
->element
[i
];
338 vb
= &nv50
->vtxbuf
[b
];
340 if (likely(vb
->stride
) || !(nv50
->vbo_user
& (1 << b
)))
341 PUSH_DATA(push
, ve
->state
);
343 PUSH_DATA(push
, ve
->state
| NV50_3D_VERTEX_ARRAY_ATTRIB_CONST
);
346 PUSH_DATA(push
, NV50_3D_VERTEX_ATTRIB_INACTIVE
);
348 /* update per-instance enables */
349 mask
= vertex
->instance_elts
^ nv50
->state
.instance_elts
;
351 const int i
= ffs(mask
) - 1;
353 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_PER_INSTANCE(i
)), 1);
354 PUSH_DATA (push
, (vertex
->instance_elts
>> i
) & 1);
356 nv50
->state
.instance_elts
= vertex
->instance_elts
;
358 if (nv50
->vbo_user
& ~nv50
->vbo_constant
)
359 nv50_upload_user_buffers(nv50
, addrs
, limits
);
361 /* update buffers and set constant attributes */
362 for (i
= 0; i
< vertex
->num_elements
; ++i
) {
363 uint64_t address
, limit
;
364 const unsigned b
= vertex
->element
[i
].pipe
.vertex_buffer_index
;
366 assert(b
< PIPE_MAX_ATTRIBS
);
367 ve
= &vertex
->element
[i
];
368 vb
= &nv50
->vtxbuf
[b
];
370 if (unlikely(nv50
->vbo_constant
& (1 << b
))) {
371 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FETCH(i
)), 1);
373 nv50_emit_vtxattr(nv50
, vb
, &ve
->pipe
, i
);
376 if (nv50
->vbo_user
& (1 << b
)) {
377 address
= addrs
[b
] + ve
->pipe
.src_offset
;
378 limit
= addrs
[b
] + limits
[b
];
380 if (!vb
->buffer
.resource
) {
381 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FETCH(i
)), 1);
385 struct nv04_resource
*buf
= nv04_resource(vb
->buffer
.resource
);
386 if (!(refd
& (1 << b
))) {
388 BCTX_REFN(nv50
->bufctx_3d
, 3D_VERTEX
, buf
, RD
);
390 address
= buf
->address
+ vb
->buffer_offset
+ ve
->pipe
.src_offset
;
391 limit
= buf
->address
+ buf
->base
.width0
- 1;
394 if (unlikely(ve
->pipe
.instance_divisor
)) {
395 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FETCH(i
)), 4);
396 PUSH_DATA (push
, NV50_3D_VERTEX_ARRAY_FETCH_ENABLE
| vb
->stride
);
397 PUSH_DATAh(push
, address
);
398 PUSH_DATA (push
, address
);
399 PUSH_DATA (push
, ve
->pipe
.instance_divisor
);
401 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FETCH(i
)), 3);
402 PUSH_DATA (push
, NV50_3D_VERTEX_ARRAY_FETCH_ENABLE
| vb
->stride
);
403 PUSH_DATAh(push
, address
);
404 PUSH_DATA (push
, address
);
406 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_LIMIT_HIGH(i
)), 2);
407 PUSH_DATAh(push
, limit
);
408 PUSH_DATA (push
, limit
);
410 for (; i
< nv50
->state
.num_vtxelts
; ++i
) {
411 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FETCH(i
)), 1);
414 nv50
->state
.num_vtxelts
= vertex
->num_elements
;
417 #define NV50_PRIM_GL_CASE(n) \
418 case PIPE_PRIM_##n: return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_##n
420 static inline unsigned
421 nv50_prim_gl(unsigned prim
)
424 NV50_PRIM_GL_CASE(POINTS
);
425 NV50_PRIM_GL_CASE(LINES
);
426 NV50_PRIM_GL_CASE(LINE_LOOP
);
427 NV50_PRIM_GL_CASE(LINE_STRIP
);
428 NV50_PRIM_GL_CASE(TRIANGLES
);
429 NV50_PRIM_GL_CASE(TRIANGLE_STRIP
);
430 NV50_PRIM_GL_CASE(TRIANGLE_FAN
);
431 NV50_PRIM_GL_CASE(QUADS
);
432 NV50_PRIM_GL_CASE(QUAD_STRIP
);
433 NV50_PRIM_GL_CASE(POLYGON
);
434 NV50_PRIM_GL_CASE(LINES_ADJACENCY
);
435 NV50_PRIM_GL_CASE(LINE_STRIP_ADJACENCY
);
436 NV50_PRIM_GL_CASE(TRIANGLES_ADJACENCY
);
437 NV50_PRIM_GL_CASE(TRIANGLE_STRIP_ADJACENCY
);
439 return NV50_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS
;
444 /* For pre-nva0 transform feedback. */
445 static const uint8_t nv50_pipe_prim_to_prim_size
[PIPE_PRIM_MAX
+ 1] =
447 [PIPE_PRIM_POINTS
] = 1,
448 [PIPE_PRIM_LINES
] = 2,
449 [PIPE_PRIM_LINE_LOOP
] = 2,
450 [PIPE_PRIM_LINE_STRIP
] = 2,
451 [PIPE_PRIM_TRIANGLES
] = 3,
452 [PIPE_PRIM_TRIANGLE_STRIP
] = 3,
453 [PIPE_PRIM_TRIANGLE_FAN
] = 3,
454 [PIPE_PRIM_QUADS
] = 3,
455 [PIPE_PRIM_QUAD_STRIP
] = 3,
456 [PIPE_PRIM_POLYGON
] = 3,
457 [PIPE_PRIM_LINES_ADJACENCY
] = 2,
458 [PIPE_PRIM_LINE_STRIP_ADJACENCY
] = 2,
459 [PIPE_PRIM_TRIANGLES_ADJACENCY
] = 3,
460 [PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
] = 3
464 nv50_draw_arrays(struct nv50_context
*nv50
,
465 unsigned mode
, unsigned start
, unsigned count
,
466 unsigned instance_count
)
468 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
471 if (nv50
->state
.index_bias
) {
472 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_BASE
), 1);
474 if (nv50
->screen
->base
.class_3d
>= NV84_3D_CLASS
) {
475 BEGIN_NV04(push
, NV84_3D(VERTEX_ID_BASE
), 1);
478 nv50
->state
.index_bias
= 0;
481 prim
= nv50_prim_gl(mode
);
483 while (instance_count
--) {
484 BEGIN_NV04(push
, NV50_3D(VERTEX_BEGIN_GL
), 1);
485 PUSH_DATA (push
, prim
);
486 BEGIN_NV04(push
, NV50_3D(VERTEX_BUFFER_FIRST
), 2);
487 PUSH_DATA (push
, start
);
488 PUSH_DATA (push
, count
);
489 BEGIN_NV04(push
, NV50_3D(VERTEX_END_GL
), 1);
492 prim
|= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT
;
497 nv50_draw_elements_inline_u08(struct nouveau_pushbuf
*push
, const uint8_t *map
,
498 unsigned start
, unsigned count
)
504 BEGIN_NI04(push
, NV50_3D(VB_ELEMENT_U32
), count
& 3);
505 for (i
= 0; i
< (count
& 3); ++i
)
506 PUSH_DATA(push
, *map
++);
510 unsigned i
, nr
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
* 4) / 4;
512 BEGIN_NI04(push
, NV50_3D(VB_ELEMENT_U8
), nr
);
513 for (i
= 0; i
< nr
; ++i
) {
515 (map
[3] << 24) | (map
[2] << 16) | (map
[1] << 8) | map
[0]);
523 nv50_draw_elements_inline_u16(struct nouveau_pushbuf
*push
, const uint16_t *map
,
524 unsigned start
, unsigned count
)
530 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_U32
), 1);
531 PUSH_DATA (push
, *map
++);
534 unsigned i
, nr
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
* 2) / 2;
536 BEGIN_NI04(push
, NV50_3D(VB_ELEMENT_U16
), nr
);
537 for (i
= 0; i
< nr
; ++i
) {
538 PUSH_DATA(push
, (map
[1] << 16) | map
[0]);
546 nv50_draw_elements_inline_u32(struct nouveau_pushbuf
*push
, const uint32_t *map
,
547 unsigned start
, unsigned count
)
552 const unsigned nr
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
);
554 BEGIN_NI04(push
, NV50_3D(VB_ELEMENT_U32
), nr
);
555 PUSH_DATAp(push
, map
, nr
);
563 nv50_draw_elements_inline_u32_short(struct nouveau_pushbuf
*push
,
565 unsigned start
, unsigned count
)
571 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_U32
), 1);
572 PUSH_DATA (push
, *map
++);
575 unsigned i
, nr
= MIN2(count
, NV04_PFIFO_MAX_PACKET_LEN
* 2) / 2;
577 BEGIN_NI04(push
, NV50_3D(VB_ELEMENT_U16
), nr
);
578 for (i
= 0; i
< nr
; ++i
) {
579 PUSH_DATA(push
, (map
[1] << 16) | map
[0]);
587 nv50_draw_elements(struct nv50_context
*nv50
, bool shorten
,
588 const struct pipe_draw_info
*info
,
589 unsigned mode
, unsigned start
, unsigned count
,
590 unsigned instance_count
, int32_t index_bias
,
593 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
596 prim
= nv50_prim_gl(mode
);
598 if (index_bias
!= nv50
->state
.index_bias
) {
599 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_BASE
), 1);
600 PUSH_DATA (push
, index_bias
);
601 if (nv50
->screen
->base
.class_3d
>= NV84_3D_CLASS
) {
602 BEGIN_NV04(push
, NV84_3D(VERTEX_ID_BASE
), 1);
603 PUSH_DATA (push
, index_bias
);
605 nv50
->state
.index_bias
= index_bias
;
608 if (!info
->has_user_indices
) {
609 struct nv04_resource
*buf
= nv04_resource(info
->index
.resource
);
612 const unsigned base
= buf
->offset
& ~3;
614 start
+= (buf
->offset
& 3) >> (index_size
>> 1);
616 assert(nouveau_resource_mapped_by_gpu(info
->index
.resource
));
618 /* This shouldn't have to be here. The going theory is that the buffer
619 * is being filled in by PGRAPH, and it's not done yet by the time it
620 * gets submitted to PFIFO, which in turn starts immediately prefetching
621 * the not-yet-written data. Ideally this wait would only happen on
622 * pushbuf submit, but it's probably not a big performance difference.
624 if (buf
->fence_wr
&& !nouveau_fence_signalled(buf
->fence_wr
))
625 nouveau_fence_wait(buf
->fence_wr
, &nv50
->base
.debug
);
627 while (instance_count
--) {
628 BEGIN_NV04(push
, NV50_3D(VERTEX_BEGIN_GL
), 1);
629 PUSH_DATA (push
, prim
);
631 nouveau_pushbuf_space(push
, 16, 0, 1);
632 PUSH_REFN(push
, buf
->bo
, NOUVEAU_BO_RD
| buf
->domain
);
634 switch (index_size
) {
636 BEGIN_NL50(push
, NV50_3D(VB_ELEMENT_U32
), count
);
637 nouveau_pushbuf_data(push
, buf
->bo
, base
+ start
* 4, count
* 4);
640 pb_start
= (start
& ~1) * 2;
641 pb_bytes
= ((start
+ count
+ 1) & ~1) * 2 - pb_start
;
643 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_U16_SETUP
), 1);
644 PUSH_DATA (push
, (start
<< 31) | count
);
645 BEGIN_NL50(push
, NV50_3D(VB_ELEMENT_U16
), pb_bytes
/ 4);
646 nouveau_pushbuf_data(push
, buf
->bo
, base
+ pb_start
, pb_bytes
);
647 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_U16_SETUP
), 1);
651 assert(index_size
== 1);
652 pb_start
= start
& ~3;
653 pb_bytes
= ((start
+ count
+ 3) & ~3) - pb_start
;
655 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_U8_SETUP
), 1);
656 PUSH_DATA (push
, (start
<< 30) | count
);
657 BEGIN_NL50(push
, NV50_3D(VB_ELEMENT_U8
), pb_bytes
/ 4);
658 nouveau_pushbuf_data(push
, buf
->bo
, base
+ pb_start
, pb_bytes
);
659 BEGIN_NV04(push
, NV50_3D(VB_ELEMENT_U8_SETUP
), 1);
663 BEGIN_NV04(push
, NV50_3D(VERTEX_END_GL
), 1);
666 prim
|= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT
;
669 const void *data
= info
->index
.user
;
671 while (instance_count
--) {
672 BEGIN_NV04(push
, NV50_3D(VERTEX_BEGIN_GL
), 1);
673 PUSH_DATA (push
, prim
);
674 switch (index_size
) {
676 nv50_draw_elements_inline_u08(push
, data
, start
, count
);
679 nv50_draw_elements_inline_u16(push
, data
, start
, count
);
683 nv50_draw_elements_inline_u32_short(push
, data
, start
, count
);
685 nv50_draw_elements_inline_u32(push
, data
, start
, count
);
691 BEGIN_NV04(push
, NV50_3D(VERTEX_END_GL
), 1);
694 prim
|= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT
;
697 NOUVEAU_DRV_STAT(&nv50
->screen
->base
, draw_calls_indexed
, 1);
701 nva0_draw_stream_output(struct nv50_context
*nv50
,
702 const struct pipe_draw_info
*info
)
704 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
705 struct nv50_so_target
*so
= nv50_so_target(info
->count_from_stream_output
);
706 struct nv04_resource
*res
= nv04_resource(so
->pipe
.buffer
);
707 unsigned num_instances
= info
->instance_count
;
708 unsigned mode
= nv50_prim_gl(info
->mode
);
710 if (unlikely(nv50
->screen
->base
.class_3d
< NVA0_3D_CLASS
)) {
711 /* A proper implementation without waiting doesn't seem possible,
714 NOUVEAU_ERR("draw_stream_output not supported on pre-NVA0 cards\n");
718 if (res
->status
& NOUVEAU_BUFFER_STATUS_GPU_WRITING
) {
719 res
->status
&= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING
;
721 BEGIN_NV04(push
, SUBC_3D(NV50_GRAPH_SERIALIZE
), 1);
723 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FLUSH
), 1);
727 assert(num_instances
);
730 BEGIN_NV04(push
, NV50_3D(VERTEX_BEGIN_GL
), 1);
731 PUSH_DATA (push
, mode
);
732 BEGIN_NV04(push
, NVA0_3D(DRAW_TFB_BASE
), 1);
734 BEGIN_NV04(push
, NVA0_3D(DRAW_TFB_STRIDE
), 1);
735 PUSH_DATA (push
, so
->stride
);
736 nv50_hw_query_pushbuf_submit(push
, NVA0_3D_DRAW_TFB_BYTES
,
737 nv50_query(so
->pq
), 0x4);
738 BEGIN_NV04(push
, NV50_3D(VERTEX_END_GL
), 1);
741 mode
|= NV50_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT
;
742 } while (--num_instances
);
746 nv50_draw_vbo_kick_notify(struct nouveau_pushbuf
*chan
)
748 struct nv50_screen
*screen
= chan
->user_priv
;
750 nouveau_fence_update(&screen
->base
, true);
752 nv50_bufctx_fence(screen
->cur_ctx
->bufctx_3d
, true);
756 nv50_draw_vbo(struct pipe_context
*pipe
, const struct pipe_draw_info
*info
)
758 struct nv50_context
*nv50
= nv50_context(pipe
);
759 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
760 bool tex_dirty
= false;
763 if (info
->index_size
&& !info
->has_user_indices
)
764 BCTX_REFN(nv50
->bufctx_3d
, 3D_INDEX
, nv04_resource(info
->index
.resource
), RD
);
766 /* NOTE: caller must ensure that (min_index + index_bias) is >= 0 */
767 nv50
->vb_elt_first
= info
->min_index
+ info
->index_bias
;
768 nv50
->vb_elt_limit
= info
->max_index
- info
->min_index
;
769 nv50
->instance_off
= info
->start_instance
;
770 nv50
->instance_max
= info
->instance_count
- 1;
772 /* For picking only a few vertices from a large user buffer, push is better,
773 * if index count is larger and we expect repeated vertices, suggest upload.
775 nv50
->vbo_push_hint
= /* the 64 is heuristic */
776 !(info
->index_size
&& ((nv50
->vb_elt_limit
+ 64) < info
->count
));
778 if (nv50
->vbo_user
&& !(nv50
->dirty_3d
& (NV50_NEW_3D_ARRAYS
| NV50_NEW_3D_VERTEX
))) {
779 if (!!nv50
->vbo_fifo
!= nv50
->vbo_push_hint
)
780 nv50
->dirty_3d
|= NV50_NEW_3D_ARRAYS
;
783 nv50_update_user_vbufs(nv50
);
786 if (unlikely(nv50
->num_so_targets
&& !nv50
->gmtyprog
))
787 nv50
->state
.prim_size
= nv50_pipe_prim_to_prim_size
[info
->mode
];
789 nv50_state_validate_3d(nv50
, ~0);
791 push
->kick_notify
= nv50_draw_vbo_kick_notify
;
793 for (s
= 0; s
< 3 && !nv50
->cb_dirty
; ++s
) {
794 if (nv50
->constbuf_coherent
[s
])
795 nv50
->cb_dirty
= true;
798 /* If there are any coherent constbufs, flush the cache */
799 if (nv50
->cb_dirty
) {
800 BEGIN_NV04(push
, NV50_3D(CODE_CB_FLUSH
), 1);
802 nv50
->cb_dirty
= false;
805 for (s
= 0; s
< 3 && !tex_dirty
; ++s
) {
806 if (nv50
->textures_coherent
[s
])
811 BEGIN_NV04(push
, NV50_3D(TEX_CACHE_CTL
), 1);
812 PUSH_DATA (push
, 0x20);
815 if (nv50
->screen
->base
.class_3d
>= NVA0_3D_CLASS
&&
816 nv50
->seamless_cube_map
!= nv50
->state
.seamless_cube_map
) {
817 nv50
->state
.seamless_cube_map
= nv50
->seamless_cube_map
;
818 BEGIN_NV04(push
, SUBC_3D(NVA0_3D_TEX_MISC
), 1);
819 PUSH_DATA (push
, nv50
->seamless_cube_map
? NVA0_3D_TEX_MISC_SEAMLESS_CUBE_MAP
: 0);
822 if (nv50
->vertprog
->mul_zero_wins
!= nv50
->state
.mul_zero_wins
) {
823 nv50
->state
.mul_zero_wins
= nv50
->vertprog
->mul_zero_wins
;
824 BEGIN_NV04(push
, NV50_3D(UNK1690
), 1);
825 PUSH_DATA (push
, 0x00010000 * !!nv50
->state
.mul_zero_wins
);
828 if (nv50
->vbo_fifo
) {
829 nv50_push_vbo(nv50
, info
);
833 if (nv50
->state
.instance_base
!= info
->start_instance
) {
834 nv50
->state
.instance_base
= info
->start_instance
;
835 /* NOTE: this does not affect the shader input, should it ? */
836 BEGIN_NV04(push
, NV50_3D(VB_INSTANCE_BASE
), 1);
837 PUSH_DATA (push
, info
->start_instance
);
840 nv50
->base
.vbo_dirty
|= !!nv50
->vtxbufs_coherent
;
842 if (nv50
->base
.vbo_dirty
) {
843 BEGIN_NV04(push
, NV50_3D(VERTEX_ARRAY_FLUSH
), 1);
845 nv50
->base
.vbo_dirty
= false;
848 if (info
->index_size
) {
849 bool shorten
= info
->max_index
<= 65535;
851 if (info
->primitive_restart
!= nv50
->state
.prim_restart
) {
852 if (info
->primitive_restart
) {
853 BEGIN_NV04(push
, NV50_3D(PRIM_RESTART_ENABLE
), 2);
855 PUSH_DATA (push
, info
->restart_index
);
857 if (info
->restart_index
> 65535)
860 BEGIN_NV04(push
, NV50_3D(PRIM_RESTART_ENABLE
), 1);
863 nv50
->state
.prim_restart
= info
->primitive_restart
;
865 if (info
->primitive_restart
) {
866 BEGIN_NV04(push
, NV50_3D(PRIM_RESTART_INDEX
), 1);
867 PUSH_DATA (push
, info
->restart_index
);
869 if (info
->restart_index
> 65535)
873 nv50_draw_elements(nv50
, shorten
, info
,
874 info
->mode
, info
->start
, info
->count
,
875 info
->instance_count
, info
->index_bias
, info
->index_size
);
877 if (unlikely(info
->count_from_stream_output
)) {
878 nva0_draw_stream_output(nv50
, info
);
880 nv50_draw_arrays(nv50
,
881 info
->mode
, info
->start
, info
->count
,
882 info
->instance_count
);
886 push
->kick_notify
= nv50_default_kick_notify
;
888 nv50_release_user_vbufs(nv50
);
890 nouveau_pushbuf_bufctx(push
, NULL
);
892 nouveau_bufctx_reset(nv50
->bufctx_3d
, NV50_BIND_3D_INDEX
);