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 nv50_state_validate(nv50
);
430 nz_divisors
= init_per_instance_arrays(nv50
, startInstance
, pos
, step
);
432 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 2);
433 OUT_RING (chan
, NV50_CB_AUX
| (24 << 8));
434 OUT_RING (chan
, startInstance
);
436 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
437 OUT_RING (chan
, nv50_prim(mode
));
440 nv50_push_arrays(nv50
, start
, count
);
442 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
443 OUT_RING (chan
, start
);
444 OUT_RING (chan
, count
);
446 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
449 for (i
= 1; i
< instanceCount
; i
++) {
450 if (nz_divisors
) /* any non-zero array divisors ? */
451 step_per_instance_arrays(nv50
, pos
, step
);
453 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
454 OUT_RING (chan
, nv50_prim(mode
) | (1 << 28));
457 nv50_push_arrays(nv50
, start
, count
);
459 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
460 OUT_RING (chan
, start
);
461 OUT_RING (chan
, count
);
463 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
466 nv50_unmap_vbufs(nv50
);
468 so_ref(NULL
, &nv50
->state
.instbuf
);
472 nv50_draw_arrays(struct pipe_context
*pipe
, unsigned mode
, unsigned start
,
475 struct nv50_context
*nv50
= nv50_context(pipe
);
476 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
477 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
480 nv50_state_validate(nv50
);
482 BEGIN_RING(chan
, tesla
, 0x142c, 1);
484 BEGIN_RING(chan
, tesla
, 0x142c, 1);
487 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
488 OUT_RING (chan
, nv50_prim(mode
));
491 ret
= nv50_push_arrays(nv50
, start
, count
);
493 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
494 OUT_RING (chan
, start
);
495 OUT_RING (chan
, count
);
498 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
501 nv50_unmap_vbufs(nv50
);
503 /* XXX: not sure what to do if ret != TRUE: flush and retry?
508 static INLINE boolean
509 nv50_draw_elements_inline_u08(struct nv50_context
*nv50
, uint8_t *map
,
510 unsigned start
, unsigned count
)
512 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
513 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
518 return nv50_push_elements_u08(nv50
, map
, count
);
521 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, 1);
522 OUT_RING (chan
, map
[0]);
528 unsigned nr
= count
> 2046 ? 2046 : count
;
531 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
, nr
>> 1);
532 for (i
= 0; i
< nr
; i
+= 2)
533 OUT_RING (chan
, (map
[i
+ 1] << 16) | map
[i
]);
541 static INLINE boolean
542 nv50_draw_elements_inline_u16(struct nv50_context
*nv50
, uint16_t *map
,
543 unsigned start
, unsigned count
)
545 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
546 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
551 return nv50_push_elements_u16(nv50
, map
, count
);
554 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, 1);
555 OUT_RING (chan
, map
[0]);
561 unsigned nr
= count
> 2046 ? 2046 : count
;
564 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
, nr
>> 1);
565 for (i
= 0; i
< nr
; i
+= 2)
566 OUT_RING (chan
, (map
[i
+ 1] << 16) | map
[i
]);
574 static INLINE boolean
575 nv50_draw_elements_inline_u32(struct nv50_context
*nv50
, uint32_t *map
,
576 unsigned start
, unsigned count
)
578 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
579 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
584 return nv50_push_elements_u32(nv50
, map
, count
);
587 unsigned nr
= count
> 2047 ? 2047 : count
;
589 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, nr
);
590 OUT_RINGp (chan
, map
, nr
);
599 nv50_draw_elements_inline(struct nv50_context
*nv50
,
600 void *map
, unsigned indexSize
,
601 unsigned start
, unsigned count
)
605 nv50_draw_elements_inline_u08(nv50
, map
, start
, count
);
608 nv50_draw_elements_inline_u16(nv50
, map
, start
, count
);
611 nv50_draw_elements_inline_u32(nv50
, map
, start
, count
);
617 nv50_draw_elements_instanced(struct pipe_context
*pipe
,
618 struct pipe_buffer
*indexBuffer
,
620 unsigned mode
, unsigned start
, unsigned count
,
621 unsigned startInstance
, unsigned instanceCount
)
623 struct nv50_context
*nv50
= nv50_context(pipe
);
624 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
625 struct nouveau_channel
*chan
= tesla
->channel
;
626 struct pipe_screen
*pscreen
= pipe
->screen
;
628 unsigned i
, nz_divisors
;
629 unsigned step
[16], pos
[16];
631 map
= pipe_buffer_map(pscreen
, indexBuffer
, PIPE_BUFFER_USAGE_CPU_READ
);
633 if (!NV50_USING_LOATHED_EDGEFLAG(nv50
))
634 nv50_upload_user_vbufs(nv50
);
636 nv50_state_validate(nv50
);
638 nz_divisors
= init_per_instance_arrays(nv50
, startInstance
, pos
, step
);
640 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 2);
641 OUT_RING (chan
, NV50_CB_AUX
| (24 << 8));
642 OUT_RING (chan
, startInstance
);
644 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
645 OUT_RING (chan
, nv50_prim(mode
));
647 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
649 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
652 for (i
= 1; i
< instanceCount
; ++i
) {
653 if (nz_divisors
) /* any non-zero array divisors ? */
654 step_per_instance_arrays(nv50
, pos
, step
);
656 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
657 OUT_RING (chan
, nv50_prim(mode
) | (1 << 28));
659 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
661 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
664 nv50_unmap_vbufs(nv50
);
666 so_ref(NULL
, &nv50
->state
.instbuf
);
670 nv50_draw_elements(struct pipe_context
*pipe
,
671 struct pipe_buffer
*indexBuffer
, unsigned indexSize
,
672 unsigned mode
, unsigned start
, unsigned count
)
674 struct nv50_context
*nv50
= nv50_context(pipe
);
675 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
676 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
677 struct pipe_screen
*pscreen
= pipe
->screen
;
680 nv50_state_validate(nv50
);
682 BEGIN_RING(chan
, tesla
, 0x142c, 1);
684 BEGIN_RING(chan
, tesla
, 0x142c, 1);
687 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
688 OUT_RING (chan
, nv50_prim(mode
));
690 if (!nv50
->vbo_fifo
&& indexSize
== 4) {
691 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
| 0x30000, 0);
692 OUT_RING (chan
, count
);
693 nouveau_pushbuf_submit(chan
, nouveau_bo(indexBuffer
),
694 start
<< 2, count
<< 2);
696 if (!nv50
->vbo_fifo
&& indexSize
== 2) {
697 unsigned vb_start
= (start
& ~1);
698 unsigned vb_end
= (start
+ count
+ 1) & ~1;
699 unsigned dwords
= (vb_end
- vb_start
) >> 1;
701 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16_SETUP
, 1);
702 OUT_RING (chan
, ((start
& 1) << 31) | count
);
703 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
| 0x30000, 0);
704 OUT_RING (chan
, dwords
);
705 nouveau_pushbuf_submit(chan
, nouveau_bo(indexBuffer
),
706 vb_start
<< 1, dwords
<< 2);
707 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16_SETUP
, 1);
710 map
= pipe_buffer_map(pscreen
, indexBuffer
,
711 PIPE_BUFFER_USAGE_CPU_READ
);
712 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
713 nv50_unmap_vbufs(nv50
);
714 pipe_buffer_unmap(pscreen
, indexBuffer
);
717 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
721 static INLINE boolean
722 nv50_vbo_static_attrib(struct nv50_context
*nv50
, unsigned attrib
,
723 struct nouveau_stateobj
**pso
,
724 struct pipe_vertex_element
*ve
,
725 struct pipe_vertex_buffer
*vb
)
728 struct nouveau_stateobj
*so
;
729 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
730 struct nouveau_bo
*bo
= nouveau_bo(vb
->buffer
);
733 unsigned nr_components
= util_format_get_nr_components(ve
->src_format
);
735 ret
= nouveau_bo_map(bo
, NOUVEAU_BO_RD
);
739 util_format_read_4f(ve
->src_format
, v
, 0, (uint8_t *)bo
->map
+
740 (vb
->buffer_offset
+ ve
->src_offset
), 0,
744 *pso
= so
= so_new(nv50
->vtxelt
->num_elements
,
745 nv50
->vtxelt
->num_elements
* 4, 0);
747 switch (nr_components
) {
749 so_method(so
, tesla
, NV50TCL_VTX_ATTR_4F_X(attrib
), 4);
750 so_data (so
, fui(v
[0]));
751 so_data (so
, fui(v
[1]));
752 so_data (so
, fui(v
[2]));
753 so_data (so
, fui(v
[3]));
756 so_method(so
, tesla
, NV50TCL_VTX_ATTR_3F_X(attrib
), 3);
757 so_data (so
, fui(v
[0]));
758 so_data (so
, fui(v
[1]));
759 so_data (so
, fui(v
[2]));
762 so_method(so
, tesla
, NV50TCL_VTX_ATTR_2F_X(attrib
), 2);
763 so_data (so
, fui(v
[0]));
764 so_data (so
, fui(v
[1]));
767 if (attrib
== nv50
->vertprog
->cfg
.edgeflag_in
) {
768 so_method(so
, tesla
, NV50TCL_EDGEFLAG_ENABLE
, 1);
769 so_data (so
, v
[0] ? 1 : 0);
771 so_method(so
, tesla
, NV50TCL_VTX_ATTR_1F(attrib
), 1);
772 so_data (so
, fui(v
[0]));
775 nouveau_bo_unmap(bo
);
779 nouveau_bo_unmap(bo
);
784 nv50_vtxelt_construct(struct nv50_vtxelt_stateobj
*cso
)
788 for (i
= 0; i
< cso
->num_elements
; ++i
) {
789 struct pipe_vertex_element
*ve
= &cso
->pipe
[i
];
791 cso
->hw
[i
] = nv50_vbo_vtxelt_to_hw(ve
);
796 nv50_vbo_validate(struct nv50_context
*nv50
)
798 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
799 struct nouveau_stateobj
*vtxbuf
, *vtxfmt
, *vtxattr
;
802 /* don't validate if Gallium took away our buffers */
803 if (nv50
->vtxbuf_nr
== 0)
807 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
)
808 if (nv50
->vtxbuf
[i
].stride
&&
809 !(nv50
->vtxbuf
[i
].buffer
->usage
& PIPE_BUFFER_USAGE_VERTEX
))
810 nv50
->vbo_fifo
= 0xffff;
812 if (NV50_USING_LOATHED_EDGEFLAG(nv50
))
813 nv50
->vbo_fifo
= 0xffff; /* vertprog can't set edgeflag */
815 n_ve
= MAX2(nv50
->vtxelt
->num_elements
, nv50
->state
.vtxelt_nr
);
818 vtxbuf
= so_new(n_ve
* 2, n_ve
* 5, nv50
->vtxelt
->num_elements
* 4);
819 vtxfmt
= so_new(1, n_ve
, 0);
820 so_method(vtxfmt
, tesla
, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve
);
822 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; i
++) {
823 struct pipe_vertex_element
*ve
= &nv50
->vtxelt
->pipe
[i
];
824 struct pipe_vertex_buffer
*vb
=
825 &nv50
->vtxbuf
[ve
->vertex_buffer_index
];
826 struct nouveau_bo
*bo
= nouveau_bo(vb
->buffer
);
827 uint32_t hw
= nv50
->vtxelt
->hw
[i
];
830 nv50_vbo_static_attrib(nv50
, i
, &vtxattr
, ve
, vb
)) {
831 so_data(vtxfmt
, hw
| (1 << 4));
833 so_method(vtxbuf
, tesla
,
834 NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
837 nv50
->vbo_fifo
&= ~(1 << i
);
841 if (nv50
->vbo_fifo
) {
842 so_data (vtxfmt
, hw
|
843 (ve
->instance_divisor
? (1 << 4) : i
));
844 so_method(vtxbuf
, tesla
,
845 NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
849 so_data(vtxfmt
, hw
| i
);
851 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_FORMAT(i
), 3);
852 so_data (vtxbuf
, 0x20000000 |
853 (ve
->instance_divisor
? 0 : vb
->stride
));
854 so_reloc (vtxbuf
, bo
, vb
->buffer_offset
+
855 ve
->src_offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
|
856 NOUVEAU_BO_RD
| NOUVEAU_BO_HIGH
, 0, 0);
857 so_reloc (vtxbuf
, bo
, vb
->buffer_offset
+
858 ve
->src_offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
|
859 NOUVEAU_BO_RD
| NOUVEAU_BO_LOW
, 0, 0);
861 /* vertex array limits */
862 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i
), 2);
863 so_reloc (vtxbuf
, bo
, vb
->buffer
->size
- 1,
864 NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
|
865 NOUVEAU_BO_HIGH
, 0, 0);
866 so_reloc (vtxbuf
, bo
, vb
->buffer
->size
- 1,
867 NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
|
868 NOUVEAU_BO_LOW
, 0, 0);
870 for (; i
< n_ve
; ++i
) {
871 so_data (vtxfmt
, 0x7e080010);
873 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
876 nv50
->state
.vtxelt_nr
= nv50
->vtxelt
->num_elements
;
878 so_ref (vtxfmt
, &nv50
->state
.vtxfmt
);
879 so_ref (vtxbuf
, &nv50
->state
.vtxbuf
);
880 so_ref (vtxattr
, &nv50
->state
.vtxattr
);
881 so_ref (NULL
, &vtxbuf
);
882 so_ref (NULL
, &vtxfmt
);
883 so_ref (NULL
, &vtxattr
);
886 typedef void (*pfn_push
)(struct nouveau_channel
*, void *);
888 struct nv50_vbo_emitctx
898 unsigned ve_edgeflag
;
902 emit_vtx_next(struct nouveau_channel
*chan
, struct nv50_vbo_emitctx
*emit
)
906 for (i
= 0; i
< emit
->nr_ve
; ++i
) {
907 emit
->push
[i
](chan
, emit
->map
[i
]);
908 emit
->map
[i
] += emit
->stride
[i
];
913 emit_vtx(struct nouveau_channel
*chan
, struct nv50_vbo_emitctx
*emit
,
918 for (i
= 0; i
< emit
->nr_ve
; ++i
)
919 emit
->push
[i
](chan
, emit
->map
[i
] + emit
->stride
[i
] * vi
);
922 static INLINE boolean
923 nv50_map_vbufs(struct nv50_context
*nv50
)
927 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
) {
928 struct pipe_vertex_buffer
*vb
= &nv50
->vtxbuf
[i
];
929 unsigned size
= vb
->stride
* (vb
->max_index
+ 1) + 16;
931 if (nouveau_bo(vb
->buffer
)->map
)
934 size
= vb
->stride
* (vb
->max_index
+ 1) + 16;
935 size
= MIN2(size
, vb
->buffer
->size
);
937 size
= vb
->buffer
->size
;
939 if (nouveau_bo_map_range(nouveau_bo(vb
->buffer
),
940 0, size
, NOUVEAU_BO_RD
))
944 if (i
== nv50
->vtxbuf_nr
)
947 nouveau_bo_unmap(nouveau_bo(nv50
->vtxbuf
[i
].buffer
));
952 emit_b32_1(struct nouveau_channel
*chan
, void *data
)
956 OUT_RING(chan
, v
[0]);
960 emit_b32_2(struct nouveau_channel
*chan
, void *data
)
964 OUT_RING(chan
, v
[0]);
965 OUT_RING(chan
, v
[1]);
969 emit_b32_3(struct nouveau_channel
*chan
, void *data
)
973 OUT_RING(chan
, v
[0]);
974 OUT_RING(chan
, v
[1]);
975 OUT_RING(chan
, v
[2]);
979 emit_b32_4(struct nouveau_channel
*chan
, void *data
)
983 OUT_RING(chan
, v
[0]);
984 OUT_RING(chan
, v
[1]);
985 OUT_RING(chan
, v
[2]);
986 OUT_RING(chan
, v
[3]);
990 emit_b16_1(struct nouveau_channel
*chan
, void *data
)
994 OUT_RING(chan
, v
[0]);
998 emit_b16_3(struct nouveau_channel
*chan
, void *data
)
1002 OUT_RING(chan
, (v
[1] << 16) | v
[0]);
1003 OUT_RING(chan
, v
[2]);
1007 emit_b08_1(struct nouveau_channel
*chan
, void *data
)
1011 OUT_RING(chan
, v
[0]);
1015 emit_b08_3(struct nouveau_channel
*chan
, void *data
)
1019 OUT_RING(chan
, (v
[2] << 16) | (v
[1] << 8) | v
[0]);
1023 emit_prepare(struct nv50_context
*nv50
, struct nv50_vbo_emitctx
*emit
,
1028 if (nv50_map_vbufs(nv50
) == FALSE
)
1031 emit
->ve_edgeflag
= nv50
->vertprog
->cfg
.edgeflag_in
;
1033 emit
->edgeflag
= 0.5f
;
1035 emit
->vtx_dwords
= 0;
1037 for (i
= 0; i
< nv50
->vtxelt
->num_elements
; ++i
) {
1038 struct pipe_vertex_element
*ve
;
1039 struct pipe_vertex_buffer
*vb
;
1040 unsigned n
, size
, nr_components
;
1041 const struct util_format_description
*desc
;
1043 ve
= &nv50
->vtxelt
->pipe
[i
];
1044 vb
= &nv50
->vtxbuf
[ve
->vertex_buffer_index
];
1045 if (!(nv50
->vbo_fifo
& (1 << i
)) || ve
->instance_divisor
)
1049 emit
->stride
[n
] = vb
->stride
;
1050 emit
->map
[n
] = (uint8_t *)nouveau_bo(vb
->buffer
)->map
+
1052 (start
* vb
->stride
+ ve
->src_offset
);
1054 desc
= util_format_description(ve
->src_format
);
1057 size
= util_format_get_component_bits(
1058 ve
->src_format
, UTIL_FORMAT_COLORSPACE_RGB
, 0);
1059 nr_components
= util_format_get_nr_components(ve
->src_format
);
1061 assert(nr_components
> 0 && nr_components
<= 4);
1063 /* It shouldn't be necessary to push the implicit 1s
1064 * for case 3 and size 8 cases 1, 2, 3.
1068 NOUVEAU_ERR("unsupported vtxelt size: %u\n", size
);
1071 switch (nr_components
) {
1072 case 1: emit
->push
[n
] = emit_b32_1
; break;
1073 case 2: emit
->push
[n
] = emit_b32_2
; break;
1074 case 3: emit
->push
[n
] = emit_b32_3
; break;
1075 case 4: emit
->push
[n
] = emit_b32_4
; break;
1077 emit
->vtx_dwords
+= nr_components
;
1080 switch (nr_components
) {
1081 case 1: emit
->push
[n
] = emit_b16_1
; break;
1082 case 2: emit
->push
[n
] = emit_b32_1
; break;
1083 case 3: emit
->push
[n
] = emit_b16_3
; break;
1084 case 4: emit
->push
[n
] = emit_b32_2
; break;
1086 emit
->vtx_dwords
+= (nr_components
+ 1) >> 1;
1089 switch (nr_components
) {
1090 case 1: emit
->push
[n
] = emit_b08_1
; break;
1091 case 2: emit
->push
[n
] = emit_b16_1
; break;
1092 case 3: emit
->push
[n
] = emit_b08_3
; break;
1093 case 4: emit
->push
[n
] = emit_b32_1
; break;
1095 emit
->vtx_dwords
+= 1;
1100 emit
->vtx_max
= 512 / emit
->vtx_dwords
;
1101 if (emit
->ve_edgeflag
< 16)
1108 set_edgeflag(struct nouveau_channel
*chan
,
1109 struct nouveau_grobj
*tesla
,
1110 struct nv50_vbo_emitctx
*emit
, uint32_t index
)
1112 unsigned i
= emit
->ve_edgeflag
;
1115 float f
= *((float *)(emit
->map
[i
] + index
* emit
->stride
[i
]));
1117 if (emit
->edgeflag
!= f
) {
1120 BEGIN_RING(chan
, tesla
, 0x15e4, 1);
1121 OUT_RING (chan
, f
? 1 : 0);
1127 nv50_push_arrays(struct nv50_context
*nv50
, unsigned start
, unsigned count
)
1129 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1130 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1131 struct nv50_vbo_emitctx emit
;
1133 if (emit_prepare(nv50
, &emit
, start
) == FALSE
)
1137 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1138 dw
= nr
* emit
.vtx_dwords
;
1140 set_edgeflag(chan
, tesla
, &emit
, 0); /* nr will be 1 */
1142 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1143 for (i
= 0; i
< nr
; ++i
)
1144 emit_vtx_next(chan
, &emit
);
1153 nv50_push_elements_u32(struct nv50_context
*nv50
, uint32_t *map
, unsigned count
)
1155 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1156 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1157 struct nv50_vbo_emitctx emit
;
1159 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1163 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1164 dw
= nr
* emit
.vtx_dwords
;
1166 set_edgeflag(chan
, tesla
, &emit
, *map
);
1168 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1169 for (i
= 0; i
< nr
; ++i
)
1170 emit_vtx(chan
, &emit
, *map
++);
1179 nv50_push_elements_u16(struct nv50_context
*nv50
, uint16_t *map
, unsigned count
)
1181 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1182 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1183 struct nv50_vbo_emitctx emit
;
1185 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1189 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1190 dw
= nr
* emit
.vtx_dwords
;
1192 set_edgeflag(chan
, tesla
, &emit
, *map
);
1194 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1195 for (i
= 0; i
< nr
; ++i
)
1196 emit_vtx(chan
, &emit
, *map
++);
1205 nv50_push_elements_u08(struct nv50_context
*nv50
, uint8_t *map
, unsigned count
)
1207 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1208 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1209 struct nv50_vbo_emitctx emit
;
1211 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1215 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1216 dw
= nr
* emit
.vtx_dwords
;
1218 set_edgeflag(chan
, tesla
, &emit
, *map
);
1220 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1221 for (i
= 0; i
< nr
; ++i
)
1222 emit_vtx(chan
, &emit
, *map
++);