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 unsigned nr_components
= util_format_get_nr_components(nv50
->vtxelt
[i
].src_format
);
229 util_format_read_4f(nv50
->vtxelt
[i
].src_format
,
230 v
, 0, data
, 0, 0, 0, 1, 1);
232 switch (nr_components
) {
234 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_4F_X(i
), 4);
235 OUT_RINGf (chan
, v
[0]);
236 OUT_RINGf (chan
, v
[1]);
237 OUT_RINGf (chan
, v
[2]);
238 OUT_RINGf (chan
, v
[3]);
241 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_3F_X(i
), 3);
242 OUT_RINGf (chan
, v
[0]);
243 OUT_RINGf (chan
, v
[1]);
244 OUT_RINGf (chan
, v
[2]);
247 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_2F_X(i
), 2);
248 OUT_RINGf (chan
, v
[0]);
249 OUT_RINGf (chan
, v
[1]);
252 BEGIN_RING(chan
, tesla
, NV50TCL_VTX_ATTR_1F(i
), 1);
253 OUT_RINGf (chan
, v
[0]);
262 init_per_instance_arrays_immd(struct nv50_context
*nv50
,
263 unsigned startInstance
,
264 unsigned pos
[16], unsigned step
[16])
266 struct nouveau_bo
*bo
;
267 unsigned i
, b
, count
= 0;
269 for (i
= 0; i
< nv50
->vtxelt_nr
; ++i
) {
270 if (!nv50
->vtxelt
[i
].instance_divisor
)
273 b
= nv50
->vtxelt
[i
].vertex_buffer_index
;
275 pos
[i
] = nv50
->vtxelt
[i
].src_offset
+
276 nv50
->vtxbuf
[b
].buffer_offset
+
277 startInstance
* nv50
->vtxbuf
[b
].stride
;
278 step
[i
] = startInstance
% nv50
->vtxelt
[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;
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(nv50
->vtxelt_nr
, nv50
->vtxelt_nr
* 2, nv50
->vtxelt_nr
* 2);
308 for (i
= 0; i
< nv50
->vtxelt_nr
; ++i
) {
309 if (!nv50
->vtxelt
[i
].instance_divisor
)
312 b
= nv50
->vtxelt
[i
].vertex_buffer_index
;
314 pos
[i
] = nv50
->vtxelt
[i
].src_offset
+
315 nv50
->vtxbuf
[b
].buffer_offset
+
316 startInstance
* nv50
->vtxbuf
[b
].stride
;
318 if (!startInstance
) {
322 step
[i
] = startInstance
% nv50
->vtxelt
[i
].instance_divisor
;
324 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
326 so_method(so
, tesla
, NV50TCL_VERTEX_ARRAY_START_HIGH(i
), 2);
327 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_HIGH
, 0, 0);
328 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_LOW
, 0, 0);
331 if (count
&& startInstance
) {
332 so_ref (so
, &nv50
->state
.instbuf
); /* for flush notify */
333 so_emit(chan
, nv50
->state
.instbuf
);
341 step_per_instance_arrays_immd(struct nv50_context
*nv50
,
342 unsigned pos
[16], unsigned step
[16])
344 struct nouveau_bo
*bo
;
347 for (i
= 0; i
< nv50
->vtxelt_nr
; ++i
) {
348 if (!nv50
->vtxelt
[i
].instance_divisor
)
350 if (++step
[i
] != nv50
->vtxelt
[i
].instance_divisor
)
352 b
= nv50
->vtxelt
[i
].vertex_buffer_index
;
353 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
356 pos
[i
] += nv50
->vtxbuf
[b
].stride
;
358 nv50_set_static_vtxattr(nv50
, i
, (uint8_t *)bo
->map
+ pos
[i
]);
363 step_per_instance_arrays(struct nv50_context
*nv50
,
364 unsigned pos
[16], unsigned step
[16])
366 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
367 struct nouveau_channel
*chan
= tesla
->channel
;
368 struct nouveau_bo
*bo
;
369 struct nouveau_stateobj
*so
;
371 const uint32_t rl
= NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
;
373 if (nv50
->vbo_fifo
) {
374 step_per_instance_arrays_immd(nv50
, pos
, step
);
378 so
= so_new(nv50
->vtxelt_nr
, nv50
->vtxelt_nr
* 2, nv50
->vtxelt_nr
* 2);
380 for (i
= 0; i
< nv50
->vtxelt_nr
; ++i
) {
381 if (!nv50
->vtxelt
[i
].instance_divisor
)
383 b
= nv50
->vtxelt
[i
].vertex_buffer_index
;
385 if (++step
[i
] == nv50
->vtxelt
[i
].instance_divisor
) {
387 pos
[i
] += nv50
->vtxbuf
[b
].stride
;
390 bo
= nouveau_bo(nv50
->vtxbuf
[b
].buffer
);
392 so_method(so
, tesla
, NV50TCL_VERTEX_ARRAY_START_HIGH(i
), 2);
393 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_HIGH
, 0, 0);
394 so_reloc (so
, bo
, pos
[i
], rl
| NOUVEAU_BO_LOW
, 0, 0);
397 so_ref (so
, &nv50
->state
.instbuf
); /* for flush notify */
400 so_emit(chan
, nv50
->state
.instbuf
);
404 nv50_unmap_vbufs(struct nv50_context
*nv50
)
408 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
)
409 if (nouveau_bo(nv50
->vtxbuf
[i
].buffer
)->map
)
410 nouveau_bo_unmap(nouveau_bo(nv50
->vtxbuf
[i
].buffer
));
414 nv50_draw_arrays_instanced(struct pipe_context
*pipe
,
415 unsigned mode
, unsigned start
, unsigned count
,
416 unsigned startInstance
, unsigned instanceCount
)
418 struct nv50_context
*nv50
= nv50_context(pipe
);
419 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
420 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
421 unsigned i
, nz_divisors
;
422 unsigned step
[16], pos
[16];
424 if (!NV50_USING_LOATHED_EDGEFLAG(nv50
))
425 nv50_upload_user_vbufs(nv50
);
427 nv50_state_validate(nv50
);
429 nz_divisors
= init_per_instance_arrays(nv50
, startInstance
, pos
, step
);
431 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 2);
432 OUT_RING (chan
, NV50_CB_AUX
| (24 << 8));
433 OUT_RING (chan
, startInstance
);
435 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
436 OUT_RING (chan
, nv50_prim(mode
));
439 nv50_push_arrays(nv50
, start
, count
);
441 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
442 OUT_RING (chan
, start
);
443 OUT_RING (chan
, count
);
445 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
448 for (i
= 1; i
< instanceCount
; i
++) {
449 if (nz_divisors
) /* any non-zero array divisors ? */
450 step_per_instance_arrays(nv50
, pos
, step
);
452 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
453 OUT_RING (chan
, nv50_prim(mode
) | (1 << 28));
456 nv50_push_arrays(nv50
, start
, count
);
458 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
459 OUT_RING (chan
, start
);
460 OUT_RING (chan
, count
);
462 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
465 nv50_unmap_vbufs(nv50
);
467 so_ref(NULL
, &nv50
->state
.instbuf
);
471 nv50_draw_arrays(struct pipe_context
*pipe
, unsigned mode
, unsigned start
,
474 struct nv50_context
*nv50
= nv50_context(pipe
);
475 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
476 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
479 nv50_state_validate(nv50
);
481 BEGIN_RING(chan
, tesla
, 0x142c, 1);
483 BEGIN_RING(chan
, tesla
, 0x142c, 1);
486 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
487 OUT_RING (chan
, nv50_prim(mode
));
490 ret
= nv50_push_arrays(nv50
, start
, count
);
492 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BUFFER_FIRST
, 2);
493 OUT_RING (chan
, start
);
494 OUT_RING (chan
, count
);
497 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
500 nv50_unmap_vbufs(nv50
);
502 /* XXX: not sure what to do if ret != TRUE: flush and retry?
507 static INLINE boolean
508 nv50_draw_elements_inline_u08(struct nv50_context
*nv50
, uint8_t *map
,
509 unsigned start
, unsigned count
)
511 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
512 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
517 return nv50_push_elements_u08(nv50
, map
, count
);
520 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, 1);
521 OUT_RING (chan
, map
[0]);
527 unsigned nr
= count
> 2046 ? 2046 : count
;
530 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
, nr
>> 1);
531 for (i
= 0; i
< nr
; i
+= 2)
532 OUT_RING (chan
, (map
[i
+ 1] << 16) | map
[i
]);
540 static INLINE boolean
541 nv50_draw_elements_inline_u16(struct nv50_context
*nv50
, uint16_t *map
,
542 unsigned start
, unsigned count
)
544 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
545 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
550 return nv50_push_elements_u16(nv50
, map
, count
);
553 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, 1);
554 OUT_RING (chan
, map
[0]);
560 unsigned nr
= count
> 2046 ? 2046 : count
;
563 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
, nr
>> 1);
564 for (i
= 0; i
< nr
; i
+= 2)
565 OUT_RING (chan
, (map
[i
+ 1] << 16) | map
[i
]);
573 static INLINE boolean
574 nv50_draw_elements_inline_u32(struct nv50_context
*nv50
, uint32_t *map
,
575 unsigned start
, unsigned count
)
577 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
578 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
583 return nv50_push_elements_u32(nv50
, map
, count
);
586 unsigned nr
= count
> 2047 ? 2047 : count
;
588 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
, nr
);
589 OUT_RINGp (chan
, map
, nr
);
598 nv50_draw_elements_inline(struct nv50_context
*nv50
,
599 void *map
, unsigned indexSize
,
600 unsigned start
, unsigned count
)
604 nv50_draw_elements_inline_u08(nv50
, map
, start
, count
);
607 nv50_draw_elements_inline_u16(nv50
, map
, start
, count
);
610 nv50_draw_elements_inline_u32(nv50
, map
, start
, count
);
616 nv50_draw_elements_instanced(struct pipe_context
*pipe
,
617 struct pipe_buffer
*indexBuffer
,
619 unsigned mode
, unsigned start
, unsigned count
,
620 unsigned startInstance
, unsigned instanceCount
)
622 struct nv50_context
*nv50
= nv50_context(pipe
);
623 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
624 struct nouveau_channel
*chan
= tesla
->channel
;
625 struct pipe_screen
*pscreen
= pipe
->screen
;
627 unsigned i
, nz_divisors
;
628 unsigned step
[16], pos
[16];
630 map
= pipe_buffer_map(pscreen
, indexBuffer
, PIPE_BUFFER_USAGE_CPU_READ
);
632 if (!NV50_USING_LOATHED_EDGEFLAG(nv50
))
633 nv50_upload_user_vbufs(nv50
);
635 nv50_state_validate(nv50
);
637 nz_divisors
= init_per_instance_arrays(nv50
, startInstance
, pos
, step
);
639 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 2);
640 OUT_RING (chan
, NV50_CB_AUX
| (24 << 8));
641 OUT_RING (chan
, startInstance
);
643 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
644 OUT_RING (chan
, nv50_prim(mode
));
646 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
648 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
651 for (i
= 1; i
< instanceCount
; ++i
) {
652 if (nz_divisors
) /* any non-zero array divisors ? */
653 step_per_instance_arrays(nv50
, pos
, step
);
655 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
656 OUT_RING (chan
, nv50_prim(mode
) | (1 << 28));
658 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
660 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
663 nv50_unmap_vbufs(nv50
);
665 so_ref(NULL
, &nv50
->state
.instbuf
);
669 nv50_draw_elements(struct pipe_context
*pipe
,
670 struct pipe_buffer
*indexBuffer
, unsigned indexSize
,
671 unsigned mode
, unsigned start
, unsigned count
)
673 struct nv50_context
*nv50
= nv50_context(pipe
);
674 struct nouveau_channel
*chan
= nv50
->screen
->tesla
->channel
;
675 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
676 struct pipe_screen
*pscreen
= pipe
->screen
;
679 nv50_state_validate(nv50
);
681 BEGIN_RING(chan
, tesla
, 0x142c, 1);
683 BEGIN_RING(chan
, tesla
, 0x142c, 1);
686 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_BEGIN
, 1);
687 OUT_RING (chan
, nv50_prim(mode
));
689 if (!nv50
->vbo_fifo
&& indexSize
== 4) {
690 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U32
| 0x30000, 0);
691 OUT_RING (chan
, count
);
692 nouveau_pushbuf_submit(chan
, nouveau_bo(indexBuffer
),
693 start
<< 2, count
<< 2);
695 if (!nv50
->vbo_fifo
&& indexSize
== 2) {
696 unsigned vb_start
= (start
& ~1);
697 unsigned vb_end
= (start
+ count
+ 1) & ~1;
698 unsigned dwords
= (vb_end
- vb_start
) >> 1;
700 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16_SETUP
, 1);
701 OUT_RING (chan
, ((start
& 1) << 31) | count
);
702 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16
| 0x30000, 0);
703 OUT_RING (chan
, dwords
);
704 nouveau_pushbuf_submit(chan
, nouveau_bo(indexBuffer
),
705 vb_start
<< 1, dwords
<< 2);
706 BEGIN_RING(chan
, tesla
, NV50TCL_VB_ELEMENT_U16_SETUP
, 1);
709 map
= pipe_buffer_map(pscreen
, indexBuffer
,
710 PIPE_BUFFER_USAGE_CPU_READ
);
711 nv50_draw_elements_inline(nv50
, map
, indexSize
, start
, count
);
712 nv50_unmap_vbufs(nv50
);
713 pipe_buffer_unmap(pscreen
, indexBuffer
);
716 BEGIN_RING(chan
, tesla
, NV50TCL_VERTEX_END
, 1);
720 static INLINE boolean
721 nv50_vbo_static_attrib(struct nv50_context
*nv50
, unsigned attrib
,
722 struct nouveau_stateobj
**pso
,
723 struct pipe_vertex_element
*ve
,
724 struct pipe_vertex_buffer
*vb
)
727 struct nouveau_stateobj
*so
;
728 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
729 struct nouveau_bo
*bo
= nouveau_bo(vb
->buffer
);
732 unsigned nr_components
= util_format_get_nr_components(ve
->src_format
);
734 ret
= nouveau_bo_map(bo
, NOUVEAU_BO_RD
);
738 util_format_read_4f(ve
->src_format
, v
, 0, (uint8_t *)bo
->map
+
739 (vb
->buffer_offset
+ ve
->src_offset
), 0,
743 *pso
= so
= so_new(nv50
->vtxelt_nr
, nv50
->vtxelt_nr
* 4, 0);
745 switch (nr_components
) {
747 so_method(so
, tesla
, NV50TCL_VTX_ATTR_4F_X(attrib
), 4);
748 so_data (so
, fui(v
[0]));
749 so_data (so
, fui(v
[1]));
750 so_data (so
, fui(v
[2]));
751 so_data (so
, fui(v
[3]));
754 so_method(so
, tesla
, NV50TCL_VTX_ATTR_3F_X(attrib
), 3);
755 so_data (so
, fui(v
[0]));
756 so_data (so
, fui(v
[1]));
757 so_data (so
, fui(v
[2]));
760 so_method(so
, tesla
, NV50TCL_VTX_ATTR_2F_X(attrib
), 2);
761 so_data (so
, fui(v
[0]));
762 so_data (so
, fui(v
[1]));
765 if (attrib
== nv50
->vertprog
->cfg
.edgeflag_in
) {
766 so_method(so
, tesla
, NV50TCL_EDGEFLAG_ENABLE
, 1);
767 so_data (so
, v
[0] ? 1 : 0);
769 so_method(so
, tesla
, NV50TCL_VTX_ATTR_1F(attrib
), 1);
770 so_data (so
, fui(v
[0]));
773 nouveau_bo_unmap(bo
);
777 nouveau_bo_unmap(bo
);
782 nv50_vbo_validate(struct nv50_context
*nv50
)
784 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
785 struct nouveau_stateobj
*vtxbuf
, *vtxfmt
, *vtxattr
;
788 /* don't validate if Gallium took away our buffers */
789 if (nv50
->vtxbuf_nr
== 0)
793 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
)
794 if (nv50
->vtxbuf
[i
].stride
&&
795 !(nv50
->vtxbuf
[i
].buffer
->usage
& PIPE_BUFFER_USAGE_VERTEX
))
796 nv50
->vbo_fifo
= 0xffff;
798 if (NV50_USING_LOATHED_EDGEFLAG(nv50
))
799 nv50
->vbo_fifo
= 0xffff; /* vertprog can't set edgeflag */
801 n_ve
= MAX2(nv50
->vtxelt_nr
, nv50
->state
.vtxelt_nr
);
804 vtxbuf
= so_new(n_ve
* 2, n_ve
* 5, nv50
->vtxelt_nr
* 4);
805 vtxfmt
= so_new(1, n_ve
, 0);
806 so_method(vtxfmt
, tesla
, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve
);
808 for (i
= 0; i
< nv50
->vtxelt_nr
; i
++) {
809 struct pipe_vertex_element
*ve
= &nv50
->vtxelt
[i
];
810 struct pipe_vertex_buffer
*vb
=
811 &nv50
->vtxbuf
[ve
->vertex_buffer_index
];
812 struct nouveau_bo
*bo
= nouveau_bo(vb
->buffer
);
813 uint32_t hw
= nv50_vbo_vtxelt_to_hw(ve
);
816 nv50_vbo_static_attrib(nv50
, i
, &vtxattr
, ve
, vb
)) {
817 so_data(vtxfmt
, hw
| (1 << 4));
819 so_method(vtxbuf
, tesla
,
820 NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
823 nv50
->vbo_fifo
&= ~(1 << i
);
827 if (nv50
->vbo_fifo
) {
828 so_data (vtxfmt
, hw
|
829 (ve
->instance_divisor
? (1 << 4) : i
));
830 so_method(vtxbuf
, tesla
,
831 NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
835 so_data(vtxfmt
, hw
| i
);
837 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_FORMAT(i
), 3);
838 so_data (vtxbuf
, 0x20000000 |
839 (ve
->instance_divisor
? 0 : vb
->stride
));
840 so_reloc (vtxbuf
, bo
, vb
->buffer_offset
+
841 ve
->src_offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
|
842 NOUVEAU_BO_RD
| NOUVEAU_BO_HIGH
, 0, 0);
843 so_reloc (vtxbuf
, bo
, vb
->buffer_offset
+
844 ve
->src_offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
|
845 NOUVEAU_BO_RD
| NOUVEAU_BO_LOW
, 0, 0);
847 /* vertex array limits */
848 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_LIMIT_HIGH(i
), 2);
849 so_reloc (vtxbuf
, bo
, vb
->buffer
->size
- 1,
850 NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
|
851 NOUVEAU_BO_HIGH
, 0, 0);
852 so_reloc (vtxbuf
, bo
, vb
->buffer
->size
- 1,
853 NOUVEAU_BO_VRAM
| NOUVEAU_BO_GART
| NOUVEAU_BO_RD
|
854 NOUVEAU_BO_LOW
, 0, 0);
856 for (; i
< n_ve
; ++i
) {
857 so_data (vtxfmt
, 0x7e080010);
859 so_method(vtxbuf
, tesla
, NV50TCL_VERTEX_ARRAY_FORMAT(i
), 1);
862 nv50
->state
.vtxelt_nr
= nv50
->vtxelt_nr
;
864 so_ref (vtxfmt
, &nv50
->state
.vtxfmt
);
865 so_ref (vtxbuf
, &nv50
->state
.vtxbuf
);
866 so_ref (vtxattr
, &nv50
->state
.vtxattr
);
867 so_ref (NULL
, &vtxbuf
);
868 so_ref (NULL
, &vtxfmt
);
869 so_ref (NULL
, &vtxattr
);
872 typedef void (*pfn_push
)(struct nouveau_channel
*, void *);
874 struct nv50_vbo_emitctx
884 unsigned ve_edgeflag
;
888 emit_vtx_next(struct nouveau_channel
*chan
, struct nv50_vbo_emitctx
*emit
)
892 for (i
= 0; i
< emit
->nr_ve
; ++i
) {
893 emit
->push
[i
](chan
, emit
->map
[i
]);
894 emit
->map
[i
] += emit
->stride
[i
];
899 emit_vtx(struct nouveau_channel
*chan
, struct nv50_vbo_emitctx
*emit
,
904 for (i
= 0; i
< emit
->nr_ve
; ++i
)
905 emit
->push
[i
](chan
, emit
->map
[i
] + emit
->stride
[i
] * vi
);
908 static INLINE boolean
909 nv50_map_vbufs(struct nv50_context
*nv50
)
913 for (i
= 0; i
< nv50
->vtxbuf_nr
; ++i
) {
914 struct pipe_vertex_buffer
*vb
= &nv50
->vtxbuf
[i
];
915 unsigned size
= vb
->stride
* (vb
->max_index
+ 1) + 16;
917 if (nouveau_bo(vb
->buffer
)->map
)
920 size
= vb
->stride
* (vb
->max_index
+ 1) + 16;
921 size
= MIN2(size
, vb
->buffer
->size
);
923 size
= vb
->buffer
->size
;
925 if (nouveau_bo_map_range(nouveau_bo(vb
->buffer
),
926 0, size
, NOUVEAU_BO_RD
))
930 if (i
== nv50
->vtxbuf_nr
)
933 nouveau_bo_unmap(nouveau_bo(nv50
->vtxbuf
[i
].buffer
));
938 emit_b32_1(struct nouveau_channel
*chan
, void *data
)
942 OUT_RING(chan
, v
[0]);
946 emit_b32_2(struct nouveau_channel
*chan
, void *data
)
950 OUT_RING(chan
, v
[0]);
951 OUT_RING(chan
, v
[1]);
955 emit_b32_3(struct nouveau_channel
*chan
, void *data
)
959 OUT_RING(chan
, v
[0]);
960 OUT_RING(chan
, v
[1]);
961 OUT_RING(chan
, v
[2]);
965 emit_b32_4(struct nouveau_channel
*chan
, void *data
)
969 OUT_RING(chan
, v
[0]);
970 OUT_RING(chan
, v
[1]);
971 OUT_RING(chan
, v
[2]);
972 OUT_RING(chan
, v
[3]);
976 emit_b16_1(struct nouveau_channel
*chan
, void *data
)
980 OUT_RING(chan
, v
[0]);
984 emit_b16_3(struct nouveau_channel
*chan
, void *data
)
988 OUT_RING(chan
, (v
[1] << 16) | v
[0]);
989 OUT_RING(chan
, v
[2]);
993 emit_b08_1(struct nouveau_channel
*chan
, void *data
)
997 OUT_RING(chan
, v
[0]);
1001 emit_b08_3(struct nouveau_channel
*chan
, void *data
)
1005 OUT_RING(chan
, (v
[2] << 16) | (v
[1] << 8) | v
[0]);
1009 emit_prepare(struct nv50_context
*nv50
, struct nv50_vbo_emitctx
*emit
,
1014 if (nv50_map_vbufs(nv50
) == FALSE
)
1017 emit
->ve_edgeflag
= nv50
->vertprog
->cfg
.edgeflag_in
;
1019 emit
->edgeflag
= 0.5f
;
1021 emit
->vtx_dwords
= 0;
1023 for (i
= 0; i
< nv50
->vtxelt_nr
; ++i
) {
1024 struct pipe_vertex_element
*ve
;
1025 struct pipe_vertex_buffer
*vb
;
1026 unsigned n
, size
, nr_components
;
1027 const struct util_format_description
*desc
;
1029 ve
= &nv50
->vtxelt
[i
];
1030 vb
= &nv50
->vtxbuf
[ve
->vertex_buffer_index
];
1031 if (!(nv50
->vbo_fifo
& (1 << i
)) || ve
->instance_divisor
)
1035 emit
->stride
[n
] = vb
->stride
;
1036 emit
->map
[n
] = (uint8_t *)nouveau_bo(vb
->buffer
)->map
+
1038 (start
* vb
->stride
+ ve
->src_offset
);
1040 desc
= util_format_description(ve
->src_format
);
1043 size
= util_format_get_component_bits(
1044 ve
->src_format
, UTIL_FORMAT_COLORSPACE_RGB
, 0);
1045 nr_components
= util_format_get_nr_components(ve
->src_format
);
1047 assert(nr_components
> 0 && nr_components
<= 4);
1049 /* It shouldn't be necessary to push the implicit 1s
1050 * for case 3 and size 8 cases 1, 2, 3.
1054 NOUVEAU_ERR("unsupported vtxelt size: %u\n", size
);
1057 switch (nr_components
) {
1058 case 1: emit
->push
[n
] = emit_b32_1
; break;
1059 case 2: emit
->push
[n
] = emit_b32_2
; break;
1060 case 3: emit
->push
[n
] = emit_b32_3
; break;
1061 case 4: emit
->push
[n
] = emit_b32_4
; break;
1063 emit
->vtx_dwords
+= nr_components
;
1066 switch (nr_components
) {
1067 case 1: emit
->push
[n
] = emit_b16_1
; break;
1068 case 2: emit
->push
[n
] = emit_b32_1
; break;
1069 case 3: emit
->push
[n
] = emit_b16_3
; break;
1070 case 4: emit
->push
[n
] = emit_b32_2
; break;
1072 emit
->vtx_dwords
+= (nr_components
+ 1) >> 1;
1075 switch (nr_components
) {
1076 case 1: emit
->push
[n
] = emit_b08_1
; break;
1077 case 2: emit
->push
[n
] = emit_b16_1
; break;
1078 case 3: emit
->push
[n
] = emit_b08_3
; break;
1079 case 4: emit
->push
[n
] = emit_b32_1
; break;
1081 emit
->vtx_dwords
+= 1;
1086 emit
->vtx_max
= 512 / emit
->vtx_dwords
;
1087 if (emit
->ve_edgeflag
< 16)
1094 set_edgeflag(struct nouveau_channel
*chan
,
1095 struct nouveau_grobj
*tesla
,
1096 struct nv50_vbo_emitctx
*emit
, uint32_t index
)
1098 unsigned i
= emit
->ve_edgeflag
;
1101 float f
= *((float *)(emit
->map
[i
] + index
* emit
->stride
[i
]));
1103 if (emit
->edgeflag
!= f
) {
1106 BEGIN_RING(chan
, tesla
, 0x15e4, 1);
1107 OUT_RING (chan
, f
? 1 : 0);
1113 nv50_push_arrays(struct nv50_context
*nv50
, unsigned start
, unsigned count
)
1115 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1116 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1117 struct nv50_vbo_emitctx emit
;
1119 if (emit_prepare(nv50
, &emit
, start
) == FALSE
)
1123 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1124 dw
= nr
* emit
.vtx_dwords
;
1126 set_edgeflag(chan
, tesla
, &emit
, 0); /* nr will be 1 */
1128 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1129 for (i
= 0; i
< nr
; ++i
)
1130 emit_vtx_next(chan
, &emit
);
1139 nv50_push_elements_u32(struct nv50_context
*nv50
, uint32_t *map
, unsigned count
)
1141 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1142 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1143 struct nv50_vbo_emitctx emit
;
1145 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1149 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1150 dw
= nr
* emit
.vtx_dwords
;
1152 set_edgeflag(chan
, tesla
, &emit
, *map
);
1154 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1155 for (i
= 0; i
< nr
; ++i
)
1156 emit_vtx(chan
, &emit
, *map
++);
1165 nv50_push_elements_u16(struct nv50_context
*nv50
, uint16_t *map
, unsigned count
)
1167 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1168 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1169 struct nv50_vbo_emitctx emit
;
1171 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1175 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1176 dw
= nr
* emit
.vtx_dwords
;
1178 set_edgeflag(chan
, tesla
, &emit
, *map
);
1180 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1181 for (i
= 0; i
< nr
; ++i
)
1182 emit_vtx(chan
, &emit
, *map
++);
1191 nv50_push_elements_u08(struct nv50_context
*nv50
, uint8_t *map
, unsigned count
)
1193 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
1194 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
1195 struct nv50_vbo_emitctx emit
;
1197 if (emit_prepare(nv50
, &emit
, 0) == FALSE
)
1201 unsigned i
, dw
, nr
= MIN2(count
, emit
.vtx_max
);
1202 dw
= nr
* emit
.vtx_dwords
;
1204 set_edgeflag(chan
, tesla
, &emit
, *map
);
1206 BEGIN_RING_NI(chan
, tesla
, NV50TCL_VERTEX_DATA
, dw
);
1207 for (i
= 0; i
< nr
; ++i
)
1208 emit_vtx(chan
, &emit
, *map
++);