2 * Copyright 2008 Ben Skeggs
3 * Copyright 2010 Christoph Bumiller
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
19 * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF
20 * OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 #include "pipe/p_context.h"
25 #include "pipe/p_defines.h"
26 #include "pipe/p_state.h"
27 #include "util/u_inlines.h"
29 #include "nv50_context.h"
32 nv50_transfer_constbuf(struct nv50_context
*nv50
,
33 struct pipe_resource
*buf
, unsigned size
, unsigned cbi
)
35 struct pipe_context
*pipe
= &nv50
->pipe
;
36 struct pipe_transfer
*transfer
;
37 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
38 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
40 unsigned count
, start
;
42 map
= pipe_buffer_map(pipe
, buf
, PIPE_TRANSFER_READ
, &transfer
);
46 count
= (buf
->width0
+ 3) / 4;
50 unsigned nr
= AVAIL_RING(chan
);
56 nr
= MIN2(count
, nr
- 7);
59 nv50_screen_reloc_constbuf(nv50
->screen
, cbi
);
61 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 1);
62 OUT_RING (chan
, (start
<< 8) | cbi
);
63 BEGIN_RING_NI(chan
, tesla
, NV50TCL_CB_DATA(0), nr
);
64 OUT_RINGp (chan
, map
, nr
);
71 pipe_buffer_unmap(pipe
, buf
, transfer
);
75 nv50_program_validate_data(struct nv50_context
*nv50
, struct nv50_program
*p
)
77 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
78 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
82 uint32_t *data
= p
->immd
;
83 unsigned count
= p
->immd_size
/ 4;
87 unsigned nr
= AVAIL_RING(chan
);
93 nr
= MIN2(count
, nr
- 7);
96 nv50_screen_reloc_constbuf(nv50
->screen
, NV50_CB_PMISC
);
98 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 1);
99 OUT_RING (chan
, (start
<< 8) | NV50_CB_PMISC
);
100 BEGIN_RING_NI(chan
, tesla
, NV50TCL_CB_DATA(0), nr
);
101 OUT_RINGp (chan
, data
, nr
);
109 /* If the state tracker doesn't change the constbuf, and it is first
110 * validated with a program that doesn't use it, this check prevents
111 * it from even being uploaded. */
113 if (p->parm_size == 0)
118 case PIPE_SHADER_VERTEX
:
121 case PIPE_SHADER_FRAGMENT
:
124 case PIPE_SHADER_GEOMETRY
:
132 nv50_transfer_constbuf(nv50
, nv50
->constbuf
[p
->type
], p
->parm_size
, cbi
);
136 nv50_program_validate_code(struct nv50_context
*nv50
, struct nv50_program
*p
)
138 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
139 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
140 struct nouveau_grobj
*eng2d
= nv50
->screen
->eng2d
;
143 unsigned size
= p
->code_size
;
144 uint32_t *data
= p
->code
;
146 assert(p
->translated
);
148 /* TODO: use a single bo (for each type) for shader code */
151 ret
= nouveau_bo_new(chan
->device
, NOUVEAU_BO_VRAM
, 0x100, size
, &p
->bo
);
154 offset
= p
->code_start
= 0;
156 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_FORMAT
, 2);
157 OUT_RING (chan
, NV50_2D_DST_FORMAT_R8_UNORM
);
159 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_PITCH
, 1);
160 OUT_RING (chan
, 0x40000);
161 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_WIDTH
, 2);
162 OUT_RING (chan
, 0x10000);
166 unsigned nr
= size
/ 4;
168 if (AVAIL_RING(chan
) < 32)
171 nr
= MIN2(nr
, AVAIL_RING(chan
) - 18);
177 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_ADDRESS_HIGH
, 2);
178 OUT_RELOCh(chan
, p
->bo
, offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
);
179 OUT_RELOCl(chan
, p
->bo
, offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
);
180 BEGIN_RING(chan
, eng2d
, NV50_2D_SIFC_BITMAP_ENABLE
, 2);
182 OUT_RING (chan
, NV50_2D_SIFC_FORMAT_R8_UNORM
);
183 BEGIN_RING(chan
, eng2d
, NV50_2D_SIFC_WIDTH
, 10);
184 OUT_RING (chan
, nr
* 4);
195 BEGIN_RING_NI(chan
, eng2d
, NV50_2D_SIFC_DATA
, nr
);
196 OUT_RINGp (chan
, data
, nr
);
203 BEGIN_RING(chan
, tesla
, NV50TCL_CODE_CB_FLUSH
, 1);
208 nv50_vp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
210 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
211 struct nouveau_stateobj
*so
= so_new(5, 7, 2);
213 nv50_program_validate_code(nv50
, p
);
215 so_method(so
, tesla
, NV50TCL_VP_ADDRESS_HIGH
, 2);
216 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
217 NOUVEAU_BO_HIGH
, 0, 0);
218 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
219 NOUVEAU_BO_LOW
, 0, 0);
220 so_method(so
, tesla
, NV50TCL_VP_ATTR_EN_0
, 2);
221 so_data (so
, p
->vp
.attrs
[0]);
222 so_data (so
, p
->vp
.attrs
[1]);
223 so_method(so
, tesla
, NV50TCL_VP_REG_ALLOC_RESULT
, 1);
224 so_data (so
, p
->max_out
);
225 so_method(so
, tesla
, NV50TCL_VP_REG_ALLOC_TEMP
, 1);
226 so_data (so
, p
->max_gpr
);
227 so_method(so
, tesla
, NV50TCL_VP_START_ID
, 1);
228 so_data (so
, p
->code_start
);
235 nv50_fp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
237 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
238 struct nouveau_stateobj
*so
= so_new(6, 7, 2);
240 nv50_program_validate_code(nv50
, p
);
242 so_method(so
, tesla
, NV50TCL_FP_ADDRESS_HIGH
, 2);
243 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
244 NOUVEAU_BO_HIGH
, 0, 0);
245 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
246 NOUVEAU_BO_LOW
, 0, 0);
247 so_method(so
, tesla
, NV50TCL_FP_REG_ALLOC_TEMP
, 1);
248 so_data (so
, p
->max_gpr
);
249 so_method(so
, tesla
, NV50TCL_FP_RESULT_COUNT
, 1);
250 so_data (so
, p
->max_out
);
251 so_method(so
, tesla
, NV50TCL_FP_CONTROL
, 1);
252 so_data (so
, p
->fp
.flags
[0]);
253 so_method(so
, tesla
, NV50TCL_FP_CTRL_UNK196C
, 1);
254 so_data (so
, p
->fp
.flags
[1]);
255 so_method(so
, tesla
, NV50TCL_FP_START_ID
, 1);
256 so_data (so
, p
->code_start
);
263 nv50_gp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
265 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
266 struct nouveau_stateobj
*so
= so_new(6, 7, 2);
268 nv50_program_validate_code(nv50
, p
);
270 so_method(so
, tesla
, NV50TCL_GP_ADDRESS_HIGH
, 2);
271 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
272 NOUVEAU_BO_HIGH
, 0, 0);
273 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
274 NOUVEAU_BO_LOW
, 0, 0);
275 so_method(so
, tesla
, NV50TCL_GP_REG_ALLOC_TEMP
, 1);
276 so_data (so
, p
->max_gpr
);
277 so_method(so
, tesla
, NV50TCL_GP_REG_ALLOC_RESULT
, 1);
278 so_data (so
, p
->max_out
);
279 so_method(so
, tesla
, NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE
, 1);
280 so_data (so
, p
->gp
.prim_type
);
281 so_method(so
, tesla
, NV50TCL_GP_VERTEX_OUTPUT_COUNT
, 1);
282 so_data (so
, p
->gp
.vert_count
);
283 so_method(so
, tesla
, NV50TCL_GP_START_ID
, 1);
284 so_data (so
, p
->code_start
);
291 nv50_program_validate(struct nv50_program
*p
)
293 p
->translated
= nv50_program_tx(p
);
294 assert(p
->translated
);
295 return p
->translated
;
299 nv50_program_validate_common(struct nv50_context
*nv50
, struct nv50_program
*p
)
301 nv50_program_validate_code(nv50
, p
);
304 nv50
->req_lmem
|= 1 << p
->type
;
306 nv50
->req_lmem
&= ~(1 << p
->type
);
309 struct nouveau_stateobj
*
310 nv50_vertprog_validate(struct nv50_context
*nv50
)
312 struct nv50_program
*p
= nv50
->vertprog
;
313 struct nouveau_stateobj
*so
= NULL
;
315 if (!p
->translated
) {
316 if (nv50_program_validate(p
))
317 nv50_vp_update_stateobj(nv50
, p
);
322 if (nv50
->dirty
& NV50_NEW_VERTPROG_CB
)
323 nv50_program_validate_data(nv50
, p
);
325 if (!(nv50
->dirty
& NV50_NEW_VERTPROG
))
328 nv50_program_validate_common(nv50
, p
);
334 struct nouveau_stateobj
*
335 nv50_fragprog_validate(struct nv50_context
*nv50
)
337 struct nv50_program
*p
= nv50
->fragprog
;
338 struct nouveau_stateobj
*so
= NULL
;
340 if (!p
->translated
) {
341 if (nv50_program_validate(p
))
342 nv50_fp_update_stateobj(nv50
, p
);
347 if (nv50
->dirty
& NV50_NEW_FRAGPROG_CB
)
348 nv50_program_validate_data(nv50
, p
);
350 if (!(nv50
->dirty
& NV50_NEW_FRAGPROG
))
353 nv50_program_validate_common(nv50
, p
);
359 struct nouveau_stateobj
*
360 nv50_geomprog_validate(struct nv50_context
*nv50
)
362 struct nv50_program
*p
= nv50
->geomprog
;
363 struct nouveau_stateobj
*so
= NULL
;
365 if (!p
->translated
) {
366 if (nv50_program_validate(p
))
367 nv50_gp_update_stateobj(nv50
, p
);
372 if (nv50
->dirty
& NV50_NEW_GEOMPROG_CB
)
373 nv50_program_validate_data(nv50
, p
);
375 if (!(nv50
->dirty
& NV50_NEW_GEOMPROG
))
378 nv50_program_validate_common(nv50
, p
);
384 /* XXX: this might not work correctly in all cases yet: we assume that
385 * an FP generic input that is not written in the VP is gl_PointCoord.
388 nv50_pntc_replace(struct nv50_context
*nv50
, uint32_t pntc
[8], unsigned m
)
390 struct nv50_program
*vp
= nv50
->vertprog
;
391 struct nv50_program
*fp
= nv50
->fragprog
;
394 memset(pntc
, 0, 8 * sizeof(uint32_t));
399 for (i
= 0; i
< fp
->in_nr
; i
++) {
400 unsigned j
, n
= util_bitcount(fp
->in
[i
].mask
);
402 if (fp
->in
[i
].sn
!= TGSI_SEMANTIC_GENERIC
) {
407 for (j
= 0; j
< vp
->out_nr
; ++j
)
408 if (vp
->out
[j
].sn
== fp
->in
[i
].sn
&& vp
->out
[j
].si
== fp
->in
[i
].si
)
411 if (j
< vp
->out_nr
) {
412 uint32_t en
= nv50
->rasterizer
->pipe
.sprite_coord_enable
;
414 if (!(en
& (1 << vp
->out
[j
].si
))) {
420 /* this is either PointCoord or replaced by sprite coords */
421 for (c
= 0; c
< 4; c
++) {
422 if (!(fp
->in
[i
].mask
& (1 << c
)))
424 pntc
[m
/ 8] |= (c
+ 1) << ((m
% 8) * 4);
428 if (nv50
->rasterizer
->pipe
.sprite_coord_mode
== PIPE_SPRITE_COORD_LOWER_LEFT
)
434 nv50_vec4_map(uint32_t *map32
, int mid
, uint32_t lin
[4],
435 struct nv50_varying
*in
, struct nv50_varying
*out
)
438 uint8_t mv
= out
->mask
, mf
= in
->mask
, oid
= out
->hw
;
439 uint8_t *map
= (uint8_t *)map32
;
441 for (c
= 0; c
< 4; ++c
) {
444 lin
[mid
/ 32] |= 1 << (mid
% 32);
461 struct nouveau_stateobj
*
462 nv50_fp_linkage_validate(struct nv50_context
*nv50
)
464 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
465 struct nv50_program
*vp
;
466 struct nv50_program
*fp
= nv50
->fragprog
;
467 struct nouveau_stateobj
*so
;
468 struct nv50_varying dummy
;
471 uint32_t map
[16], lin
[4], pntc
[8];
473 uint32_t interp
= fp
->fp
.interp
;
474 uint32_t colors
= fp
->fp
.colors
;
475 uint32_t clip
= 0x04;
476 uint32_t psiz
= 0x000;
480 if (nv50
->geomprog
) {
482 memset(map
, 0x80, sizeof(map
));
485 memset(map
, 0x40, sizeof(map
));
487 memset(lin
, 0, sizeof(lin
));
490 dummy
.mask
= 0xf; /* map all components of HPOS */
491 m
= nv50_vec4_map(map
, 0, lin
, &dummy
, &vp
->out
[0]);
493 if (vp
->vp
.clpd
< 0x40) {
494 for (c
= 0; c
< vp
->vp
.clpd_nr
; ++c
) {
495 map
[m
/ 4] |= (vp
->vp
.clpd
+ c
) << ((m
% 4) * 8);
498 clip
|= vp
->vp
.clpd_nr
<< 8;
501 colors
|= m
<< 8; /* adjust BFC0 id */
503 /* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */
504 if (nv50
->rasterizer
->pipe
.light_twoside
) {
505 for (i
= 0; i
< 2; ++i
)
506 m
= nv50_vec4_map(map
, m
, lin
,
507 &fp
->in
[fp
->vp
.bfc
[i
]],
508 &vp
->out
[vp
->vp
.bfc
[i
]]);
511 colors
+= m
- 4; /* adjust FFC0 id */
512 interp
|= m
<< 8; /* set mid where 'normal' FP inputs start */
515 for (i
= 0; i
< fp
->in_nr
; i
++) {
516 for (n
= 0; n
< vp
->out_nr
; ++n
)
517 if (vp
->out
[n
].sn
== fp
->in
[i
].sn
&&
518 vp
->out
[n
].si
== fp
->in
[i
].si
)
521 m
= nv50_vec4_map(map
, m
, lin
,
522 &fp
->in
[i
], (n
< vp
->out_nr
) ? &vp
->out
[n
] : &dummy
);
525 /* PrimitiveID either is replaced by the system value, or
526 * written by the geometry shader into an output register
528 if (fp
->gp
.primid
< 0x40) {
530 map
[m
/ 4] = (map
[m
/ 4] & ~(0xff << i
)) | (vp
->gp
.primid
<< i
);
534 if (nv50
->rasterizer
->pipe
.point_size_per_vertex
) {
536 map
[m
/ 4] = (map
[m
/ 4] & ~(0xff << i
)) | (vp
->vp
.psiz
<< i
);
537 psiz
= (m
++ << 4) | 1;
540 /* now fill the stateobj (at most 28 so_data) */
541 so
= so_new(10, 54, 0);
545 if (vp
->type
== PIPE_SHADER_GEOMETRY
) {
546 so_method(so
, tesla
, NV50TCL_GP_RESULT_MAP_SIZE
, 1);
548 so_method(so
, tesla
, NV50TCL_GP_RESULT_MAP(0), n
);
549 so_datap (so
, map
, n
);
551 so_method(so
, tesla
, NV50TCL_VP_GP_BUILTIN_ATTR_EN
, 1);
552 so_data (so
, vp
->vp
.attrs
[2]);
554 so_method(so
, tesla
, NV50TCL_MAP_SEMANTIC_4
, 1);
555 so_data (so
, primid
);
557 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP_SIZE
, 1);
559 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP(0), n
);
560 so_datap (so
, map
, n
);
563 so_method(so
, tesla
, NV50TCL_MAP_SEMANTIC_0
, 4);
564 so_data (so
, colors
);
566 so_data (so
, sysval
);
569 so_method(so
, tesla
, NV50TCL_FP_INTERPOLANT_CTRL
, 1);
570 so_data (so
, interp
);
572 so_method(so
, tesla
, NV50TCL_NOPERSPECTIVE_BITMAP(0), 4);
573 so_datap (so
, lin
, 4);
575 if (nv50
->rasterizer
->pipe
.point_quad_rasterization
) {
576 so_method(so
, tesla
, NV50TCL_POINT_SPRITE_CTRL
, 1);
578 nv50_pntc_replace(nv50
, pntc
, (interp
>> 8) & 0xff));
580 so_method(so
, tesla
, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8);
581 so_datap (so
, pntc
, 8);
584 so_method(so
, tesla
, NV50TCL_GP_ENABLE
, 1);
585 so_data (so
, (vp
->type
== PIPE_SHADER_GEOMETRY
) ? 1 : 0);
591 nv50_vp_gp_mapping(uint32_t *map32
, int m
,
592 struct nv50_program
*vp
, struct nv50_program
*gp
)
594 uint8_t *map
= (uint8_t *)map32
;
597 for (i
= 0; i
< gp
->in_nr
; ++i
) {
598 uint8_t oid
= 0, mv
= 0, mg
= gp
->in
[i
].mask
;
600 for (j
= 0; j
< vp
->out_nr
; ++j
) {
601 if (vp
->out
[j
].sn
== gp
->in
[i
].sn
&&
602 vp
->out
[j
].si
== gp
->in
[i
].si
) {
603 mv
= vp
->out
[j
].mask
;
609 for (c
= 0; c
< 4; ++c
, mv
>>= 1, mg
>>= 1) {
614 map
[m
++] = (c
== 3) ? 0x41 : 0x40;
621 struct nouveau_stateobj
*
622 nv50_gp_linkage_validate(struct nv50_context
*nv50
)
624 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
625 struct nouveau_stateobj
*so
;
626 struct nv50_program
*vp
= nv50
->vertprog
;
627 struct nv50_program
*gp
= nv50
->geomprog
;
633 memset(map
, 0, sizeof(map
));
635 m
= nv50_vp_gp_mapping(map
, m
, vp
, gp
);
637 so
= so_new(3, 24 - 3, 0);
639 so_method(so
, tesla
, NV50TCL_VP_GP_BUILTIN_ATTR_EN
, 1);
640 so_data (so
, vp
->vp
.attrs
[2] | gp
->vp
.attrs
[2]);
643 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP_SIZE
, 1);
647 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP(0), m
);
648 so_datap (so
, map
, m
);