2 * Copyright 2008 Ben Skeggs
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 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
23 #include "pipe/p_context.h"
24 #include "pipe/p_state.h"
25 #include "util/u_inlines.h"
26 #include "util/u_format.h"
28 #include "nv50_context.h"
31 nv50_push_elements_u08(struct nv50_context
*, uint8_t *, unsigned);
34 nv50_push_elements_u16(struct nv50_context
*, uint16_t *, unsigned);
37 nv50_push_elements_u32(struct nv50_context
*, uint32_t *, unsigned);
40 nv50_push_arrays(struct nv50_context
*, unsigned, unsigned);
42 #define NV50_USING_LOATHED_EDGEFLAG(ctx) ((ctx)->vertprog->cfg.edgeflag_in < 16)
44 static INLINE
unsigned
45 nv50_prim(unsigned mode
)
48 case PIPE_PRIM_POINTS
: return NV50TCL_VERTEX_BEGIN_POINTS
;
49 case PIPE_PRIM_LINES
: return NV50TCL_VERTEX_BEGIN_LINES
;
50 case PIPE_PRIM_LINE_LOOP
: return NV50TCL_VERTEX_BEGIN_LINE_LOOP
;
51 case PIPE_PRIM_LINE_STRIP
: return NV50TCL_VERTEX_BEGIN_LINE_STRIP
;
52 case PIPE_PRIM_TRIANGLES
: return NV50TCL_VERTEX_BEGIN_TRIANGLES
;
53 case PIPE_PRIM_TRIANGLE_STRIP
:
54 return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP
;
55 case PIPE_PRIM_TRIANGLE_FAN
: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN
;
56 case PIPE_PRIM_QUADS
: return NV50TCL_VERTEX_BEGIN_QUADS
;
57 case PIPE_PRIM_QUAD_STRIP
: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP
;
58 case PIPE_PRIM_POLYGON
: return NV50TCL_VERTEX_BEGIN_POLYGON
;
59 case PIPE_PRIM_LINES_ADJACENCY
:
60 return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY
;
61 case PIPE_PRIM_LINE_STRIP_ADJACENCY
:
62 return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY
;
63 case PIPE_PRIM_TRIANGLES_ADJACENCY
:
64 return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY
;
65 case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY
:
66 return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY
;
71 NOUVEAU_ERR("invalid primitive type %d\n", mode
);
72 return NV50TCL_VERTEX_BEGIN_POINTS
;
75 static INLINE
uint32_t
76 nv50_vbo_type_to_hw(enum pipe_format format
)
78 const struct util_format_description
*desc
;
80 desc
= util_format_description(format
);
83 switch (desc
->channel
[0].type
) {
84 case UTIL_FORMAT_TYPE_FLOAT
:
85 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_FLOAT
;
86 case UTIL_FORMAT_TYPE_UNSIGNED
:
87 if (desc
->channel
[0].normalized
) {
88 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UNORM
;
90 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_USCALED
;
91 case UTIL_FORMAT_TYPE_SIGNED
:
92 if (desc
->channel
[0].normalized
) {
93 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SNORM
;
95 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SSCALED
;
97 case PIPE_FORMAT_TYPE_UINT:
98 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_UINT;
99 case PIPE_FORMAT_TYPE_SINT:
100 return NV50TCL_VERTEX_ARRAY_ATTRIB_TYPE_SINT; */
106 static INLINE
uint32_t
107 nv50_vbo_size_to_hw(unsigned size
, unsigned nr_c
)
109 static const uint32_t hw_values
[] = {
111 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8
,
112 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8
,
113 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8
,
114 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_8_8_8_8
,
115 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16
,
116 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16
,
117 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16
,
118 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_16_16_16_16
,
120 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32
,
121 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32
,
122 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32
,
123 NV50TCL_VERTEX_ARRAY_ATTRIB_FORMAT_32_32_32_32
};
125 /* we'd also have R11G11B10 and R10G10B10A2 */
127 assert(nr_c
> 0 && nr_c
<= 4);
133 return hw_values
[size
+ (nr_c
- 1)];
136 static INLINE
uint32_t
137 nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element
*ve
)
139 uint32_t hw_type
, hw_size
;
140 enum pipe_format pf
= ve
->src_format
;
141 const struct util_format_description
*desc
;
142 unsigned size
, nr_components
;
144 desc
= util_format_description(pf
);
147 size
= util_format_get_component_bits(pf
, UTIL_FORMAT_COLORSPACE_RGB
, 0);
148 nr_components
= util_format_get_nr_components(pf
);
150 hw_type
= nv50_vbo_type_to_hw(pf
);
151 hw_size
= nv50_vbo_size_to_hw(size
, nr_components
);
153 if (!hw_type
|| !hw_size
) {
154 NOUVEAU_ERR("unsupported vbo format: %s\n", util_format_name(pf
));
159 if (desc
->swizzle
[0] == UTIL_FORMAT_SWIZZLE_Z
) /* BGRA */
160 hw_size
|= (1 << 31); /* no real swizzle bits :-( */
162 return (hw_type
| hw_size
);
165 /* For instanced drawing from user buffers, hitting the FIFO repeatedly
166 * with the same vertex data is probably worse than uploading all data.
169 nv50_upload_vtxbuf(struct nv50_context
*nv50
, unsigned i
)
171 struct nv50_screen
*nscreen
= nv50
->screen
;
172 struct pipe_screen
*pscreen
= &nscreen
->base
.base
;
173 struct pipe_buffer
*buf
= nscreen
->strm_vbuf
[i
];
174 struct pipe_vertex_buffer
*vb
= &nv50
->vtxbuf
[i
];
176 unsigned size
= align(vb
->buffer
->size
, 4096);
178 if (buf
&& buf
->size
< size
)
179 pipe_buffer_reference(&nscreen
->strm_vbuf
[i
], NULL
);
181 if (!nscreen
->strm_vbuf
[i
]) {
182 nscreen
->strm_vbuf
[i
] = pipe_buffer_create(
183 pscreen
, 0, PIPE_BUFFER_USAGE_VERTEX
, size
);
184 buf
= nscreen
->strm_vbuf
[i
];
187 src
= pipe_buffer_map(pscreen
, vb
->buffer
, PIPE_BUFFER_USAGE_CPU_READ
);
190 src
+= vb
->buffer_offset
;
192 size
= (vb
->max_index
+ 1) * vb
->stride
+ 16; /* + 16 is for stride 0 */
193 if (vb
->buffer_offset
+ size
> vb
->buffer
->size
)
194 size
= vb
->buffer
->size
- vb
->buffer_offset
;
196 pipe_buffer_write(pscreen
, buf
, vb
->buffer_offset
, size
, src
);
197 pipe_buffer_unmap(pscreen
, vb
->buffer
);
199 vb
->buffer
= buf
; /* don't pipe_reference, this is a private copy */
204 nv50_upload_user_vbufs(struct nv50_context
*nv50
)
209 nv50
->dirty
|= NV50_NEW_ARRAYS
;
210 if (!(nv50
->dirty
& NV50_NEW_ARRAYS
))
213 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
) {
214 if (nv50
->vtxbuf
[i
].buffer
->usage
& PIPE_BUFFER_USAGE_VERTEX
)
216 nv50_upload_vtxbuf(nv50
, i
);
221 nv50_set_static_vtxattr(struct nv50_context
*nv50
, unsigned i
, void *data
)
223 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
224 struct nouveau_channel
*chan
= tesla
->channel
;
226 enum pipe_format pf
= nv50
->vtxelt
->pipe
[i
].src_format
;
227 unsigned nr_components
= util_format_get_nr_components(pf
);
229 util_format_read_4f(pf
, v
, 0, data
, 0, 0, 0, 1, 1);
231 switch (nr_components
) {
233 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_4F_X(i
), 4);
234 OUT_RINGf (chan
, v
[0]);
235 OUT_RINGf (chan
, v
[1]);
236 OUT_RINGf (chan
, v
[2]);
237 OUT_RINGf (chan
, v
[3]);
240 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_3F_X(i
), 3);
241 OUT_RINGf (chan
, v
[0]);
242 OUT_RINGf (chan
, v
[1]);
243 OUT_RINGf (chan
, v
[2]);
246 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_2F_X(i
), 2);
247 OUT_RINGf (chan
, v
[0]);
248 OUT_RINGf (chan
, v
[1]);
251 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_1F(i
), 1);
252 OUT_RINGf (chan
, v
[0]);
261 init_per_instance_arrays_immd(struct nv50_context
*nv50
,
262 unsigned startInstance
,
263 unsigned pos
[16], unsigned step
[16])
265 struct nouveau_bo
*bo
;
266 unsigned i
, b
, count
= 0;
268 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; ++i
) {
269 if (!nv50
->vtxelt
->pipe
[i
].instance_divisor
)
272 b
= nv50
->vtxelt
->pipe
[i
].vertex_buffer_index
;
274 pos
[i
] = nv50
->vtxelt
->pipe
[i
].src_offset
+
275 nv50
->vtxbuf
[b
].buffer_offset
+
276 startInstance
* nv50
->vtxbuf
[b
].stride
;
277 step
[i
] = startInstance
%
278 nv50
->vtxelt
->pipe
[i
].instance_divisor
;
280 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
282 nouveau_bo_map(bo
, NOUVEAU_BO_RD
);
284 nv50_set_static_vtxattr(nv50
, i
, (uint8_t *)bo
->map
+ pos
[i
]);
291 init_per_instance_arrays(struct nv50_context
*nv50
,
292 unsigned startInstance
,
293 unsigned pos
[16], unsigned step
[16])
295 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
296 struct nouveau_channel
*chan
= tesla
->channel
;
297 struct nouveau_bo
*bo
;
298 struct nouveau_stateobj
*so
;
299 unsigned i
, b
, count
= 0, num_elements
= nv50
->vtxelt
->num_elements
;
300 const uint32_t rl
= NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
;
303 return init_per_instance_arrays_immd(nv50
, startInstance
,
306 so
= so_new(num_elements
, num_elements
* 2, num_elements
* 2);
308 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; ++i
) {
309 if (!nv50
->vtxelt
->pipe
[i
].instance_divisor
)
312 b
= nv50
->vtxelt
->pipe
[i
].vertex_buffer_index
;
314 pos
[i
] = nv50
->vtxelt
->pipe
[i
].src_offset
+
315 nv50
->vtxbuf
[b
].buffer_offset
+
316 startInstance
* nv50
->vtxbuf
[b
].stride
;
318 if (!startInstance
) {
322 step
[i
] = startInstance
%
323 nv50
->vtxelt
->pipe
[i
].instance_divisor
;
325 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
327 so_method(so
, tesla
, NV50TCL_VERTEX_ARRAY_START_HIGH(i
), 2);
328 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_HIGH
, 0, 0);
329 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_LOW
, 0, 0);
332 if (count
&& startInstance
) {
333 so_ref (so
, &nv50
->state
.instbuf
); /* for flush notify */
334 so_emit(chan
, nv50
->state
.instbuf
);
342 step_per_instance_arrays_immd(struct nv50_context
*nv50
,
343 unsigned pos
[16], unsigned step
[16])
345 struct nouveau_bo
*bo
;
348 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; ++i
) {
349 if (!nv50
->vtxelt
->pipe
[i
].instance_divisor
)
351 if (++step
[i
] != nv50
->vtxelt
->pipe
[i
].instance_divisor
)
353 b
= nv50
->vtxelt
->pipe
[i
].vertex_buffer_index
;
354 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
357 pos
[i
] += nv50
->vtxbuf
[b
].stride
;
359 nv50_set_static_vtxattr(nv50
, i
, (uint8_t *)bo
->map
+ pos
[i
]);
364 step_per_instance_arrays(struct nv50_context
*nv50
,
365 unsigned pos
[16], unsigned step
[16])
367 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
368 struct nouveau_channel
*chan
= tesla
->channel
;
369 struct nouveau_bo
*bo
;
370 struct nouveau_stateobj
*so
;
371 unsigned i
, b
, num_elements
= nv50
->vtxelt
->num_elements
;
372 const uint32_t rl
= NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
;
374 if (nv50
->vbo_fifo
) {
375 step_per_instance_arrays_immd(nv50
, pos
, step
);
379 so
= so_new(num_elements
, num_elements
* 2, num_elements
* 2);
381 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; ++i
) {
382 if (!nv50
->vtxelt
->pipe
[i
].instance_divisor
)
384 b
= nv50
->vtxelt
->pipe
[i
].vertex_buffer_index
;
386 if (++step
[i
] == nv50
->vtxelt
->pipe
[i
].instance_divisor
) {
388 pos
[i
] += nv50
->vtxbuf
[b
].stride
;
391 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
393 so_method(so
, tesla
, NV50TCL_VERTEX_ARRAY_START_HIGH(i
), 2);
394 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_HIGH
, 0, 0);
395 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_LOW
, 0, 0);
398 so_ref (so
, &nv50
->state
.instbuf
); /* for flush notify */
401 so_emit(chan
, nv50
->state
.instbuf
);
405 nv50_unmap_vbufs(struct nv50_context
*nv50
)
409 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
)
410 if (nouveau_bo(nv50
->vtxbuf
[i
].buffer
)->map
)
411 nouveau_bo_unmap(nouveau_bo(nv50
->vtxbuf
[i
].buffer
));
415 nv50_draw_arrays_instanced(struct pipe_context
*pipe
,
416 unsigned mode
, unsigned start
, unsigned count
,
417 unsigned startInstance
, unsigned instanceCount
)
419 struct nv50_context
*nv50
= nv50_context(pipe
);
420 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
421 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
422 unsigned i
, nz_divisors
;
423 unsigned step
[16], pos
[16];
425 if (!NV50_USING_LOATHED_EDGEFLAG(nv50
))
426 nv50_upload_user_vbufs(nv50
);
428 if (!nv50_state_validate(nv50
, 0))
430 chan
->flush_notify
= nv50_state_flush_notify
;
432 nz_divisors
= init_per_instance_arrays(nv50
, startInstance
, pos
, step
);
434 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 2);
435 OUT_RING (chan
, NV50_CB_AUX
| (24 << 8));
436 OUT_RING (chan
, startInstance
);
438 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
439 OUT_RING (chan
, nv50_prim(mode
));
442 nv50_push_arrays(nv50
, start
, count
);
444 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
445 OUT_RING (chan
, start
);
446 OUT_RING (chan
, count
);
448 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
451 for (i
= 1; i
< instanceCount
; i
++) {
452 if (nz_divisors
) /* any non-zero array divisors ? */
453 step_per_instance_arrays(nv50
, pos
, step
);
455 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
456 OUT_RING (chan
, nv50_prim(mode
) | (1 << 28));
459 nv50_push_arrays(nv50
, start
, count
);
461 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
462 OUT_RING (chan
, start
);
463 OUT_RING (chan
, count
);
465 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
469 chan
->flush_notify
= NULL
;
470 nv50_unmap_vbufs(nv50
);
472 so_ref(NULL
, &nv50
->state
.instbuf
);
476 nv50_draw_arrays(struct pipe_context
*pipe
, unsigned mode
, unsigned start
,
479 struct nv50_context
*nv50
= nv50_context(pipe
);
480 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
481 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
484 if (!nv50_state_validate(nv50
, 11))
486 chan
->flush_notify
= nv50_state_flush_notify
;
488 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
489 OUT_RING (chan
, nv50_prim(mode
));
492 ret
= nv50_push_arrays(nv50
, start
, count
);
494 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
495 OUT_RING (chan
, start
);
496 OUT_RING (chan
, count
);
499 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
502 chan
->flush_notify
= NULL
;
503 nv50_unmap_vbufs(nv50
);
505 /* XXX: not sure what to do if ret != TRUE: flush and retry?
510 static INLINE boolean
511 nv50_draw_elements_inline_u08(struct nv50_context
*nv50
, uint8_t *map
,
512 unsigned start
, unsigned count
)
514 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
515 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
520 return nv50_push_elements_u08(nv50
, map
, count
);
523 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, 1);
524 OUT_RING (chan
, map
[0]);
530 unsigned nr
= count
> 2046 ? 2046 : count
;
533 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
, nr
>> 1);
534 for (i
= 0; i
< nr
; i
+= 2)
535 OUT_RING (chan
, (map
[i
+ 1] << 16) | map
[i
]);
543 static INLINE boolean
544 nv50_draw_elements_inline_u16(struct nv50_context
*nv50
, uint16_t *map
,
545 unsigned start
, unsigned count
)
547 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
548 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
553 return nv50_push_elements_u16(nv50
, map
, count
);
556 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, 1);
557 OUT_RING (chan
, map
[0]);
563 unsigned nr
= count
> 2046 ? 2046 : count
;
566 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
, nr
>> 1);
567 for (i
= 0; i
< nr
; i
+= 2)
568 OUT_RING (chan
, (map
[i
+ 1] << 16) | map
[i
]);
576 static INLINE boolean
577 nv50_draw_elements_inline_u32(struct nv50_context
*nv50
, uint32_t *map
,
578 unsigned start
, unsigned count
)
580 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
581 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
586 return nv50_push_elements_u32(nv50
, map
, count
);
589 unsigned nr
= count
> 2047 ? 2047 : count
;
591 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, nr
);
592 OUT_RINGp (chan
, map
, nr
);
601 nv50_draw_elements_inline(struct nv50_context
*nv50
,
602 void *map
, unsigned indexSize
,
603 unsigned start
, unsigned count
)
607 nv50_draw_elements_inline_u08(nv50
, map
, start
, count
);
610 nv50_draw_elements_inline_u16(nv50
, map
, start
, count
);
613 nv50_draw_elements_inline_u32(nv50
, map
, start
, count
);
619 nv50_draw_elements_instanced(struct pipe_context
*pipe
,
620 struct pipe_buffer
*indexBuffer
,
622 unsigned mode
, unsigned start
, unsigned count
,
623 unsigned startInstance
, unsigned instanceCount
)
625 struct nv50_context
*nv50
= nv50_context(pipe
);
626 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
627 struct nouveau_channel
*chan
= tesla
->channel
;
628 struct pipe_screen
*pscreen
= pipe
->screen
;
630 unsigned i
, nz_divisors
;
631 unsigned step
[16], pos
[16];
633 map
= pipe_buffer_map(pscreen
, indexBuffer
, PIPE_BUFFER_USAGE_CPU_READ
);
635 if (!NV50_USING_LOATHED_EDGEFLAG(nv50
))
636 nv50_upload_user_vbufs(nv50
);
638 if (!nv50_state_validate(nv50
, 0))
640 chan
->flush_notify
= nv50_state_flush_notify
;
642 nz_divisors
= init_per_instance_arrays(nv50
, startInstance
, pos
, step
);
644 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 2);
645 OUT_RING (chan
, NV50_CB_AUX
| (24 << 8));
646 OUT_RING (chan
, startInstance
);
648 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
649 OUT_RING (chan
, nv50_prim(mode
));
651 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
653 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
656 for (i
= 1; i
< instanceCount
; ++i
) {
657 if (nz_divisors
) /* any non-zero array divisors ? */
658 step_per_instance_arrays(nv50
, pos
, step
);
660 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
661 OUT_RING (chan
, nv50_prim(mode
) | (1 << 28));
663 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
665 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
669 chan
->flush_notify
= NULL
;
670 nv50_unmap_vbufs(nv50
);
672 so_ref(NULL
, &nv50
->state
.instbuf
);
676 nv50_draw_elements(struct pipe_context
*pipe
,
677 struct pipe_buffer
*indexBuffer
, unsigned indexSize
,
678 unsigned mode
, unsigned start
, unsigned count
)
680 struct nv50_context
*nv50
= nv50_context(pipe
);
681 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
682 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
683 struct pipe_screen
*pscreen
= pipe
->screen
;
686 if (!nv50_state_validate(nv50
, 14))
688 chan
->flush_notify
= nv50_state_flush_notify
;
690 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
691 OUT_RING (chan
, nv50_prim(mode
));
693 if (!nv50
->vbo_fifo
&& indexSize
== 4) {
694 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
| 0x30000, 0);
695 OUT_RING (chan
, count
);
696 nouveau_pushbuf_submit(chan
, nouveau_bo(indexBuffer
),
697 start
<< 2, count
<< 2);
699 if (!nv50
->vbo_fifo
&& indexSize
== 2) {
700 unsigned vb_start
= (start
& ~1);
701 unsigned vb_end
= (start
+ count
+ 1) & ~1;
702 unsigned dwords
= (vb_end
- vb_start
) >> 1;
704 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16_SETUP
, 1);
705 OUT_RING (chan
, ((start
& 1) << 31) | count
);
706 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
| 0x30000, 0);
707 OUT_RING (chan
, dwords
);
708 nouveau_pushbuf_submit(chan
, nouveau_bo(indexBuffer
),
709 vb_start
<< 1, dwords
<< 2);
710 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16_SETUP
, 1);
713 map
= pipe_buffer_map(pscreen
, indexBuffer
,
714 PIPE_BUFFER_USAGE_CPU_READ
);
715 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
716 nv50_unmap_vbufs(nv50
);
717 pipe_buffer_unmap(pscreen
, indexBuffer
);
720 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
722 chan
->flush_notify
= NULL
;
725 static INLINE boolean
726 nv50_vbo_static_attrib(struct nv50_context
*nv50
, unsigned attrib
,
727 struct nouveau_stateobj
**pso
,
728 struct pipe_vertex_element
*ve
,
729 struct pipe_vertex_buffer
*vb
)
732 struct nouveau_stateobj
*so
;
733 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
734 struct nouveau_bo
*bo
= nouveau_bo(vb
->buffer
);
737 unsigned nr_components
= util_format_get_nr_components(ve
->src_format
);
739 ret
= nouveau_bo_map(bo
, NOUVEAU_BO_RD
);
743 util_format_read_4f(ve
->src_format
, v
, 0, (uint8_t *)bo
->map
+
744 (vb
->buffer_offset
+ ve
->src_offset
), 0,
748 *pso
= so
= so_new(nv50
->vtxelt
->num_elements
,
749 nv50
->vtxelt
->num_elements
* 4, 0);
751 switch (nr_components
) {
753 so_method(so
, tesla
, NV50TCL_VTX_ATTR_4F_X(attrib
), 4);
754 so_data (so
, fui(v
[0]));
755 so_data (so
, fui(v
[1]));
756 so_data (so
, fui(v
[2]));
757 so_data (so
, fui(v
[3]));
760 so_method(so
, tesla
, NV50TCL_VTX_ATTR_3F_X(attrib
), 3);
761 so_data (so
, fui(v
[0]));
762 so_data (so
, fui(v
[1]));
763 so_data (so
, fui(v
[2]));
766 so_method(so
, tesla
, NV50TCL_VTX_ATTR_2F_X(attrib
), 2);
767 so_data (so
, fui(v
[0]));
768 so_data (so
, fui(v
[1]));
771 if (attrib
== nv50
->vertprog
->cfg
.edgeflag_in
) {
772 so_method(so
, tesla
, NV50TCL_EDGEFLAG_ENABLE
, 1);
773 so_data (so
, v
[0] ? 1 : 0);
775 so_method(so
, tesla
, NV50TCL_VTX_ATTR_1F(attrib
), 1);
776 so_data (so
, fui(v
[0]));
779 nouveau_bo_unmap(bo
);
783 nouveau_bo_unmap(bo
);
788 nv50_vtxelt_construct(struct nv50_vtxelt_stateobj
*cso
)
792 for (i
= 0; i
< cso
->num_elements
; ++i
) {
793 struct pipe_vertex_element
*ve
= &cso
->pipe
[i
];
795 cso
->hw
[i
] = nv50_vbo_vtxelt_to_hw(ve
);
799 struct nouveau_stateobj
*
800 nv50_vbo_validate(struct nv50_context
*nv50
)
802 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
803 struct nouveau_stateobj
*vtxbuf
, *vtxfmt
, *vtxattr
;
806 /* don't validate if Gallium took away our buffers */
807 if (nv50
->vtxbuf_nr
== 0)
811 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
)
812 if (nv50
->vtxbuf
[i
].stride
&&
813 !(nv50
->vtxbuf
[i
].buffer
->usage
& PIPE_BUFFER_USAGE_VERTEX
))
814 nv50
->vbo_fifo
= 0xffff;
816 if (NV50_USING_LOATHED_EDGEFLAG(nv50
))
817 nv50
->vbo_fifo
= 0xffff; /* vertprog can't set edgeflag */
819 n_ve
= MAX2(nv50
->vtxelt
->num_elements
, nv50
->state
.vtxelt_nr
);
822 vtxbuf
= so_new(n_ve
* 2, n_ve
* 5, nv50
->vtxelt
->num_elements
* 4);
823 vtxfmt
= so_new(1, n_ve
, 0);
824 so_method(vtxfmt
, tesla
, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve
);
826 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; i
++) {
827 struct pipe_vertex_element
*ve
= &nv50
->vtxelt
->pipe
[i
];
828 struct pipe_vertex_buffer
*vb
=
829 &nv50
->vtxbuf
[ve
->vertex_buffer_index
];
830 struct nouveau_bo
*bo
= nouveau_bo(vb
->buffer
);
831 uint32_t hw
= nv50
->vtxelt
->hw
[i
];
834 nv50_vbo_static_attrib(nv50
, i
, &vtxattr
, ve
, vb
)) {
835 so_data(vtxfmt
, hw
| (1 << 4));
837 so_method(vtxbuf
, tesla
,
838 NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
841 nv50
->vbo_fifo
&= ~(1 << i
);
845 if (nv50
->vbo_fifo
) {
846 so_data (vtxfmt
, hw
|
847 (ve
->instance_divisor
? (1 << 4) : i
));
848 so_method(vtxbuf
, tesla
,
849 NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
853 so_data(vtxfmt
, hw
| i
);
855 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_FORMAT(i
), 3);
856 so_data (vtxbuf
, 0x20000000 |
857 (ve
->instance_divisor
? 0 : vb
->stride
));
858 so_reloc (vtxbuf
, bo
, vb
->buffer_offset
+
859 ve
->src_offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
|
860 NOUVEAU_BO_RD
| NOUVEAU_BO_HIGH
, 0, 0);
861 so_reloc (vtxbuf
, bo
, vb
->buffer_offset
+
862 ve
->src_offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
|
863 NOUVEAU_BO_RD
| NOUVEAU_BO_LOW
, 0, 0);
865 /* vertex array limits */
866 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i
), 2);
867 so_reloc (vtxbuf
, bo
, vb
->buffer
->size
- 1,
868 NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
|
869 NOUVEAU_BO_HIGH
, 0, 0);
870 so_reloc (vtxbuf
, bo
, vb
->buffer
->size
- 1,
871 NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
|
872 NOUVEAU_BO_LOW
, 0, 0);
874 for (; i
< n_ve
; ++i
) {
875 so_data (vtxfmt
, 0x7e080010);
877 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
880 nv50
->state
.vtxelt_nr
= nv50
->vtxelt
->num_elements
;
882 so_ref (vtxbuf
, &nv50
->state
.vtxbuf
);
883 so_ref (vtxattr
, &nv50
->state
.vtxattr
);
884 so_ref (NULL
, &vtxbuf
);
885 so_ref (NULL
, &vtxattr
);
889 typedef void (*pfn_push
)(struct nouveau_channel
*, void *);
891 struct nv50_vbo_emitctx
901 unsigned ve_edgeflag
;
905 emit_vtx_next(struct nouveau_channel
*chan
, struct nv50_vbo_emitctx
*emit
)
909 for (i
= 0; i
< emit
->nr_ve
; ++i
) {
910 emit
->push
[i
](chan
, emit
->map
[i
]);
911 emit
->map
[i
] += emit
->stride
[i
];
916 emit_vtx(struct nouveau_channel
*chan
, struct nv50_vbo_emitctx
*emit
,
921 for (i
= 0; i
< emit
->nr_ve
; ++i
)
922 emit
->push
[i
](chan
, emit
->map
[i
] + emit
->stride
[i
] * vi
);
925 static INLINE boolean
926 nv50_map_vbufs(struct nv50_context
*nv50
)
930 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
) {
931 struct pipe_vertex_buffer
*vb
= &nv50
->vtxbuf
[i
];
932 unsigned size
= vb
->stride
* (vb
->max_index
+ 1) + 16;
934 if (nouveau_bo(vb
->buffer
)->map
)
937 size
= vb
->stride
* (vb
->max_index
+ 1) + 16;
938 size
= MIN2(size
, vb
->buffer
->size
);
940 size
= vb
->buffer
->size
;
942 if (nouveau_bo_map_range(nouveau_bo(vb
->buffer
),
943 0, size
, NOUVEAU_BO_RD
))
947 if (i
== nv50
->vtxbuf_nr
)
950 nouveau_bo_unmap(nouveau_bo(nv50
->vtxbuf
[i
].buffer
));
955 emit_b32_1(struct nouveau_channel
*chan
, void *data
)
959 OUT_RING(chan
, v
[0]);
963 emit_b32_2(struct nouveau_channel
*chan
, void *data
)
967 OUT_RING(chan
, v
[0]);
968 OUT_RING(chan
, v
[1]);
972 emit_b32_3(struct nouveau_channel
*chan
, void *data
)
976 OUT_RING(chan
, v
[0]);
977 OUT_RING(chan
, v
[1]);
978 OUT_RING(chan
, v
[2]);
982 emit_b32_4(struct nouveau_channel
*chan
, void *data
)
986 OUT_RING(chan
, v
[0]);
987 OUT_RING(chan
, v
[1]);
988 OUT_RING(chan
, v
[2]);
989 OUT_RING(chan
, v
[3]);
993 emit_b16_1(struct nouveau_channel
*chan
, void *data
)
997 OUT_RING(chan
, v
[0]);
1001 emit_b16_3(struct nouveau_channel
*chan
, void *data
)
1005 OUT_RING(chan
, (v
[1] << 16) | v
[0]);
1006 OUT_RING(chan
, v
[2]);
1010 emit_b08_1(struct nouveau_channel
*chan
, void *data
)
1014 OUT_RING(chan
, v
[0]);
1018 emit_b08_3(struct nouveau_channel
*chan
, void *data
)
1022 OUT_RING(chan
, (v
[2] << 16) | (v
[1] << 8) | v
[0]);
1026 emit_prepare(struct nv50_context
*nv50
, struct nv50_vbo_emitctx
*emit
,
1031 if (nv50_map_vbufs(nv50
) == FALSE
)
1034 emit
->ve_edgeflag
= nv50
->vertprog
->cfg
.edgeflag_in
;
1036 emit
->edgeflag
= 0.5f
;
1038 emit
->vtx_dwords
= 0;
1040 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; ++i
) {
1041 struct pipe_vertex_element
*ve
;
1042 struct pipe_vertex_buffer
*vb
;
1043 unsigned n
, size
, nr_components
;
1044 const struct util_format_description
*desc
;
1046 ve
= &nv50
->vtxelt
->pipe
[i
];
1047 vb
= &nv50
->vtxbuf
[ve
->vertex_buffer_index
];
1048 if (!(nv50
->vbo_fifo
& (1 << i
)) || ve
->instance_divisor
)
1052 emit
->stride
[n
] = vb
->stride
;
1053 emit
->map
[n
] = (uint8_t *)nouveau_bo(vb
->buffer
)->map
+
1055 (start
* vb
->stride
+ ve
->src_offset
);
1057 desc
= util_format_description(ve
->src_format
);
1060 size
= util_format_get_component_bits(
1061 ve
->src_format
, UTIL_FORMAT_COLORSPACE_RGB
, 0);
1062 nr_components
= util_format_get_nr_components(ve
->src_format
);
1064 assert(nr_components
> 0 && nr_components
<= 4);
1066 /* It shouldn't be necessary to push the implicit 1s
1067 * for case 3 and size 8 cases 1, 2, 3.
1071 NOUVEAU_ERR("unsupported vtxelt size: %u\n", size
);
1074 switch (nr_components
) {
1075 case 1: emit
->push
[n
] = emit_b32_1
; break;
1076 case 2: emit
->push
[n
] = emit_b32_2
; break;
1077 case 3: emit
->push
[n
] = emit_b32_3
; break;
1078 case 4: emit
->push
[n
] = emit_b32_4
; break;
1080 emit
->vtx_dwords
+= nr_components
;
1083 switch (nr_components
) {
1084 case 1: emit
->push
[n
] = emit_b16_1
; break;
1085 case 2: emit
->push
[n
] = emit_b32_1
; break;
1086 case 3: emit
->push
[n
] = emit_b16_3
; break;
1087 case 4: emit
->push
[n
] = emit_b32_2
; break;
1089 emit
->vtx_dwords
+= (nr_components
+ 1) >> 1;
1092 switch (nr_components
) {
1093 case 1: emit
->push
[n
] = emit_b08_1
; break;
1094 case 2: emit
->push
[n
] = emit_b16_1
; break;
1095 case 3: emit
->push
[n
] = emit_b08_3
; break;
1096 case 4: emit
->push
[n
] = emit_b32_1
; break;
1098 emit
->vtx_dwords
+= 1;
1103 emit
->vtx_max
= 512 / emit
->vtx_dwords
;
1104 if (emit
->ve_edgeflag
< 16)
1111 set_edgeflag(struct nouveau_channel
*chan
,
1112 struct nouveau_grobj
*tesla
,
1113 struct nv50_vbo_emitctx
*emit
, uint32_t index
)
1115 unsigned i
= emit
->ve_edgeflag
;
1118 float f
= *((float *)(emit
->map
[i
] + index
* emit
->stride
[i
]));
1120 if (emit
->edgeflag
!= f
) {
1123 BEGIN_RING(chan
, tesla
, 0x15e4, 1);
1124 OUT_RING (chan
, f
? 1 : 0);
1130 nv50_push_arrays(struct nv50_context
*nv50
, unsigned start
, unsigned count
)
1132 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1133 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1134 struct nv50_vbo_emitctx emit
;
1136 if (emit_prepare(nv50
, &emit
, start
) == FALSE
)
1140 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1141 dw
= nr
* emit
.vtx_dwords
;
1143 set_edgeflag(chan
, tesla
, &emit
, 0); /* nr will be 1 */
1145 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1146 for (i
= 0; i
< nr
; ++i
)
1147 emit_vtx_next(chan
, &emit
);
1156 nv50_push_elements_u32(struct nv50_context
*nv50
, uint32_t *map
, unsigned count
)
1158 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1159 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1160 struct nv50_vbo_emitctx emit
;
1162 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1166 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1167 dw
= nr
* emit
.vtx_dwords
;
1169 set_edgeflag(chan
, tesla
, &emit
, *map
);
1171 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1172 for (i
= 0; i
< nr
; ++i
)
1173 emit_vtx(chan
, &emit
, *map
++);
1182 nv50_push_elements_u16(struct nv50_context
*nv50
, uint16_t *map
, unsigned count
)
1184 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1185 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1186 struct nv50_vbo_emitctx emit
;
1188 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1192 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1193 dw
= nr
* emit
.vtx_dwords
;
1195 set_edgeflag(chan
, tesla
, &emit
, *map
);
1197 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1198 for (i
= 0; i
< nr
; ++i
)
1199 emit_vtx(chan
, &emit
, *map
++);
1208 nv50_push_elements_u08(struct nv50_context
*nv50
, uint8_t *map
, unsigned count
)
1210 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1211 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1212 struct nv50_vbo_emitctx emit
;
1214 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1218 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1219 dw
= nr
* emit
.vtx_dwords
;
1221 set_edgeflag(chan
, tesla
, &emit
, *map
);
1223 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1224 for (i
= 0; i
< nr
; ++i
)
1225 emit_vtx(chan
, &emit
, *map
++);