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"
30 #include "nv50_transfer.h"
33 nv50_transfer_constbuf(struct nv50_context
*nv50
,
34 struct pipe_resource
*buf
, unsigned size
, unsigned cbi
)
36 struct pipe_context
*pipe
= &nv50
->pipe
;
37 struct pipe_transfer
*transfer
;
38 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
39 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
41 unsigned count
, start
;
43 map
= pipe_buffer_map(pipe
, buf
, PIPE_TRANSFER_READ
, &transfer
);
47 count
= MIN2(buf
->width0
, size
);
54 /* FIXME: emit relocs for unsuiTed MM */
55 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 1);
56 OUT_RING (chan
, (start
<< 8) | cbi
);
57 BEGIN_RING_NI(chan
, tesla
, NV50TCL_CB_DATA(0), nr
);
58 OUT_RINGp (chan
, map
, nr
);
65 pipe_buffer_unmap(pipe
, buf
, transfer
);
69 nv50_program_validate_data(struct nv50_context
*nv50
, struct nv50_program
*p
)
71 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
72 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
76 uint32_t *data
= p
->immd
;
77 unsigned count
= p
->immd_size
/ 4;
84 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 1);
85 OUT_RING (chan
, (start
<< 8) | NV50_CB_PMISC
);
86 BEGIN_RING_NI(chan
, tesla
, NV50TCL_CB_DATA(0), nr
);
87 OUT_RINGp (chan
, data
, nr
);
95 if (p
->parm_size
== 0)
99 case PIPE_SHADER_VERTEX
:
102 case PIPE_SHADER_FRAGMENT
:
105 case PIPE_SHADER_GEOMETRY
:
113 nv50_transfer_constbuf(nv50
, nv50
->constbuf
[p
->type
], p
->parm_size
, cbi
);
117 nv50_program_validate_code(struct nv50_context
*nv50
, struct nv50_program
*p
)
119 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
120 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
121 struct nouveau_grobj
*eng2d
= nv50
->screen
->eng2d
;
124 unsigned size
= p
->code_size
;
125 uint32_t *data
= p
->code
;
127 assert(p
->translated
);
129 /* TODO: use a single bo (for each type) for shader code */
132 ret
= nouveau_bo_new(chan
->device
, NOUVEAU_BO_VRAM
, 0x100, size
, &p
->bo
);
135 offset
= p
->code_start
= 0;
137 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_FORMAT
, 2);
138 OUT_RING (chan
, NV50_2D_DST_FORMAT_R8_UNORM
);
140 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_PITCH
, 1);
141 OUT_RING (chan
, 0x40000);
142 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_WIDTH
, 2);
143 OUT_RING (chan
, 0x10000);
147 unsigned nr
= size
/ 4;
149 if (AVAIL_RING(chan
) < 32)
152 nr
= MIN2(nr
, AVAIL_RING(chan
) - 18);
158 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_ADDRESS_HIGH
, 2);
159 OUT_RELOCh(chan
, p
->bo
, offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
);
160 OUT_RELOCl(chan
, p
->bo
, offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
);
161 BEGIN_RING(chan
, eng2d
, NV50_2D_SIFC_BITMAP_ENABLE
, 2);
163 OUT_RING (chan
, NV50_2D_SIFC_FORMAT_R8_UNORM
);
164 BEGIN_RING(chan
, eng2d
, NV50_2D_SIFC_WIDTH
, 10);
165 OUT_RING (chan
, nr
* 4);
176 BEGIN_RING_NI(chan
, eng2d
, NV50_2D_SIFC_DATA
, nr
);
177 OUT_RINGp (chan
, data
, nr
);
184 BEGIN_RING(chan
, tesla
, NV50TCL_CODE_CB_FLUSH
, 1);
189 nv50_vp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
191 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
192 struct nouveau_stateobj
*so
= so_new(5, 7, 2);
194 nv50_program_validate_code(nv50
, p
);
196 so_method(so
, tesla
, NV50TCL_VP_ADDRESS_HIGH
, 2);
197 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
198 NOUVEAU_BO_HIGH
, 0, 0);
199 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
200 NOUVEAU_BO_LOW
, 0, 0);
201 so_method(so
, tesla
, NV50TCL_VP_ATTR_EN_0
, 2);
202 so_data (so
, p
->vp
.attrs
[0]);
203 so_data (so
, p
->vp
.attrs
[1]);
204 so_method(so
, tesla
, NV50TCL_VP_REG_ALLOC_RESULT
, 1);
205 so_data (so
, p
->max_out
);
206 so_method(so
, tesla
, NV50TCL_VP_REG_ALLOC_TEMP
, 1);
207 so_data (so
, p
->max_gpr
);
208 so_method(so
, tesla
, NV50TCL_VP_START_ID
, 1);
209 so_data (so
, p
->code_start
);
216 nv50_fp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
218 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
219 struct nouveau_stateobj
*so
= so_new(6, 7, 2);
221 nv50_program_validate_code(nv50
, p
);
223 so_method(so
, tesla
, NV50TCL_FP_ADDRESS_HIGH
, 2);
224 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
225 NOUVEAU_BO_HIGH
, 0, 0);
226 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
227 NOUVEAU_BO_LOW
, 0, 0);
228 so_method(so
, tesla
, NV50TCL_FP_REG_ALLOC_TEMP
, 1);
229 so_data (so
, p
->max_gpr
);
230 so_method(so
, tesla
, NV50TCL_FP_RESULT_COUNT
, 1);
231 so_data (so
, p
->max_out
);
232 so_method(so
, tesla
, NV50TCL_FP_CONTROL
, 1);
233 so_data (so
, p
->fp
.flags
[0]);
234 so_method(so
, tesla
, NV50TCL_FP_CTRL_UNK196C
, 1);
235 so_data (so
, p
->fp
.flags
[1]);
236 so_method(so
, tesla
, NV50TCL_FP_START_ID
, 1);
237 so_data (so
, p
->code_start
);
244 nv50_gp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
246 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
247 struct nouveau_stateobj
*so
= so_new(6, 7, 2);
249 nv50_program_validate_code(nv50
, p
);
251 so_method(so
, tesla
, NV50TCL_GP_ADDRESS_HIGH
, 2);
252 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
253 NOUVEAU_BO_HIGH
, 0, 0);
254 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
255 NOUVEAU_BO_LOW
, 0, 0);
256 so_method(so
, tesla
, NV50TCL_GP_REG_ALLOC_TEMP
, 1);
257 so_data (so
, p
->max_gpr
);
258 so_method(so
, tesla
, NV50TCL_GP_REG_ALLOC_RESULT
, 1);
259 so_data (so
, p
->max_out
);
260 so_method(so
, tesla
, NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE
, 1);
261 so_data (so
, p
->gp
.prim_type
);
262 so_method(so
, tesla
, NV50TCL_GP_VERTEX_OUTPUT_COUNT
, 1);
263 so_data (so
, p
->gp
.vert_count
);
264 so_method(so
, tesla
, NV50TCL_GP_START_ID
, 1);
265 so_data (so
, p
->code_start
);
272 nv50_program_validate(struct nv50_program
*p
)
274 p
->translated
= nv50_program_tx(p
);
275 assert(p
->translated
);
276 return p
->translated
;
279 struct nouveau_stateobj
*
280 nv50_vertprog_validate(struct nv50_context
*nv50
)
282 struct nv50_program
*p
= nv50
->vertprog
;
283 struct nouveau_stateobj
*so
= NULL
;
285 if (!p
->translated
) {
286 if (nv50_program_validate(p
))
287 nv50_vp_update_stateobj(nv50
, p
);
292 if (nv50
->dirty
& NV50_NEW_VERTPROG_CB
)
293 nv50_program_validate_data(nv50
, p
);
295 if (!(nv50
->dirty
& NV50_NEW_VERTPROG
))
298 nv50_program_validate_code(nv50
, p
);
304 struct nouveau_stateobj
*
305 nv50_fragprog_validate(struct nv50_context
*nv50
)
307 struct nv50_program
*p
= nv50
->fragprog
;
308 struct nouveau_stateobj
*so
= NULL
;
310 if (!p
->translated
) {
311 if (nv50_program_validate(p
))
312 nv50_fp_update_stateobj(nv50
, p
);
317 if (nv50
->dirty
& NV50_NEW_FRAGPROG_CB
)
318 nv50_program_validate_data(nv50
, p
);
320 if (!(nv50
->dirty
& NV50_NEW_FRAGPROG
))
323 nv50_program_validate_code(nv50
, p
);
329 struct nouveau_stateobj
*
330 nv50_geomprog_validate(struct nv50_context
*nv50
)
332 struct nv50_program
*p
= nv50
->geomprog
;
333 struct nouveau_stateobj
*so
= NULL
;
335 if (!p
->translated
) {
336 if (nv50_program_validate(p
))
337 nv50_gp_update_stateobj(nv50
, p
);
342 if (nv50
->dirty
& NV50_NEW_GEOMPROG_CB
)
343 nv50_program_validate_data(nv50
, p
);
345 if (!(nv50
->dirty
& NV50_NEW_GEOMPROG
))
348 nv50_program_validate_code(nv50
, p
);
354 /* XXX: this might not work correctly in all cases yet: we assume that
355 * an FP generic input that is not written in the VP is gl_PointCoord.
358 nv50_pntc_replace(struct nv50_context
*nv50
, uint32_t pntc
[8], unsigned m
)
360 struct nv50_program
*vp
= nv50
->vertprog
;
361 struct nv50_program
*fp
= nv50
->fragprog
;
364 memset(pntc
, 0, 8 * sizeof(uint32_t));
369 for (i
= 0; i
< fp
->in_nr
; i
++) {
370 unsigned j
, n
= util_bitcount(fp
->in
[i
].mask
);
372 if (fp
->in
[i
].sn
!= TGSI_SEMANTIC_GENERIC
) {
377 for (j
= 0; j
< vp
->out_nr
; ++j
)
378 if (vp
->out
[j
].sn
== fp
->in
[i
].sn
&& vp
->out
[j
].si
== fp
->in
[i
].si
)
381 if (j
< vp
->out_nr
) {
382 ubyte en
= nv50
->rasterizer
->pipe
.sprite_coord_enable
;
384 if (!(en
& (1 << vp
->out
[j
].si
))) {
390 /* this is either PointCoord or replaced by sprite coords */
391 for (c
= 0; c
< 4; c
++) {
392 if (!(fp
->in
[i
].mask
& (1 << c
)))
394 pntc
[m
/ 8] |= (c
+ 1) << ((m
% 8) * 4);
398 if (nv50
->rasterizer
->pipe
.sprite_coord_mode
== PIPE_SPRITE_COORD_LOWER_LEFT
)
404 nv50_vec4_map(uint32_t *map32
, int mid
, uint32_t lin
[4],
405 struct nv50_varying
*in
, struct nv50_varying
*out
)
408 uint8_t mv
= out
->mask
, mf
= in
->mask
, oid
= out
->hw
;
409 uint8_t *map
= (uint8_t *)map32
;
411 for (c
= 0; c
< 4; ++c
) {
414 lin
[mid
/ 32] |= 1 << (mid
% 32);
431 struct nouveau_stateobj
*
432 nv50_fp_linkage_validate(struct nv50_context
*nv50
)
434 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
435 struct nv50_program
*vp
;
436 struct nv50_program
*fp
= nv50
->fragprog
;
437 struct nouveau_stateobj
*so
;
438 struct nv50_varying dummy
;
441 uint32_t map
[16], lin
[4], pntc
[8];
443 uint32_t interp
= fp
->fp
.interp
;
444 uint32_t colors
= fp
->fp
.colors
;
445 uint32_t clip
= 0x04;
446 uint32_t psiz
= 0x000;
450 if (nv50
->geomprog
) {
452 memset(map
, 0x80, sizeof(map
));
455 memset(map
, 0x40, sizeof(map
));
457 memset(lin
, 0, sizeof(lin
));
460 dummy
.mask
= 0xf; /* map all components of HPOS */
461 m
= nv50_vec4_map(map
, 0, lin
, &dummy
, &vp
->out
[0]);
463 if (vp
->vp
.clpd
< 0x40) {
464 for (c
= 0; c
< vp
->vp
.clpd_nr
; ++c
) {
465 map
[m
/ 4] |= (vp
->vp
.clpd
+ c
) << ((m
% 4) * 8);
468 clip
|= vp
->vp
.clpd_nr
<< 8;
471 colors
|= m
<< 8; /* adjust BFC0 id */
473 /* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */
474 if (nv50
->rasterizer
->pipe
.light_twoside
) {
475 for (i
= 0; i
< 2; ++i
)
476 m
= nv50_vec4_map(map
, m
, lin
,
477 &fp
->in
[fp
->vp
.bfc
[i
]],
478 &vp
->out
[vp
->vp
.bfc
[i
]]);
481 colors
+= m
- 4; /* adjust FFC0 id */
482 interp
|= m
<< 8; /* set mid where 'normal' FP inputs start */
485 for (i
= 0; i
< fp
->in_nr
; i
++) {
486 for (n
= 0; n
< vp
->out_nr
; ++n
)
487 if (vp
->out
[n
].sn
== fp
->in
[i
].sn
&&
488 vp
->out
[n
].si
== fp
->in
[i
].si
)
491 m
= nv50_vec4_map(map
, m
, lin
,
492 &fp
->in
[i
], (n
< vp
->out_nr
) ? &vp
->out
[n
] : &dummy
);
494 /* PrimitiveID either is replaced by the system value, or
495 * written by the geometry shader into an output register
497 if (fp
->gp
.primid
< 0x40) {
498 map
[m
/ 4] |= vp
->gp
.primid
<< ((m
% 4) * 8);
502 if (nv50
->rasterizer
->pipe
.point_size_per_vertex
) {
503 map
[m
/ 4] |= vp
->vp
.psiz
<< ((m
% 4) * 8);
504 psiz
= (m
++ << 4) | 1;
507 /* now fill the stateobj (at most 28 so_data) */
508 so
= so_new(10, 54, 0);
512 if (vp
->type
== PIPE_SHADER_GEOMETRY
) {
513 so_method(so
, tesla
, NV50TCL_GP_RESULT_MAP_SIZE
, 1);
515 so_method(so
, tesla
, NV50TCL_GP_RESULT_MAP(0), n
);
516 so_datap (so
, map
, n
);
518 so_method(so
, tesla
, NV50TCL_VP_GP_BUILTIN_ATTR_EN
, 1);
519 so_data (so
, vp
->vp
.attrs
[2]);
521 so_method(so
, tesla
, NV50TCL_MAP_SEMANTIC_4
, 1);
522 so_data (so
, primid
);
524 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP_SIZE
, 1);
526 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP(0), n
);
527 so_datap (so
, map
, n
);
530 //colors = 0x01000404;
531 so_method(so
, tesla
, NV50TCL_MAP_SEMANTIC_0
, 4);
532 so_data (so
, colors
);
534 so_data (so
, sysval
);
537 so_method(so
, tesla
, NV50TCL_FP_INTERPOLANT_CTRL
, 1);
538 so_data (so
, interp
);
540 so_method(so
, tesla
, NV50TCL_NOPERSPECTIVE_BITMAP(0), 4);
541 so_datap (so
, lin
, 4);
543 if (nv50
->rasterizer
->pipe
.sprite_coord_enable
) {
544 so_method(so
, tesla
, NV50TCL_POINT_SPRITE_CTRL
, 1);
546 nv50_pntc_replace(nv50
, pntc
, (interp
>> 8) & 0xff));
548 so_method(so
, tesla
, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8);
549 so_datap (so
, pntc
, 8);
552 so_method(so
, tesla
, NV50TCL_GP_ENABLE
, 1);
553 so_data (so
, (vp
->type
== PIPE_SHADER_GEOMETRY
) ? 1 : 0);
559 nv50_vp_gp_mapping(uint32_t *map32
, int m
,
560 struct nv50_program
*vp
, struct nv50_program
*gp
)
562 uint8_t *map
= (uint8_t *)map32
;
565 for (i
= 0; i
< gp
->in_nr
; ++i
) {
566 uint8_t oid
= 0, mv
= 0, mg
= gp
->in
[i
].mask
;
568 for (j
= 0; j
< vp
->out_nr
; ++j
) {
569 if (vp
->out
[j
].sn
== gp
->in
[i
].sn
&&
570 vp
->out
[j
].si
== gp
->in
[i
].si
) {
571 mv
= vp
->out
[j
].mask
;
577 for (c
= 0; c
< 4; ++c
, mv
>>= 1, mg
>>= 1) {
582 map
[m
++] = (c
== 3) ? 0x41 : 0x40;
589 struct nouveau_stateobj
*
590 nv50_gp_linkage_validate(struct nv50_context
*nv50
)
592 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
593 struct nouveau_stateobj
*so
;
594 struct nv50_program
*vp
= nv50
->vertprog
;
595 struct nv50_program
*gp
= nv50
->geomprog
;
601 memset(map
, 0, sizeof(map
));
603 m
= nv50_vp_gp_mapping(map
, m
, vp
, gp
);
605 so
= so_new(3, 24 - 3, 0);
607 so_method(so
, tesla
, NV50TCL_VP_GP_BUILTIN_ATTR_EN
, 1);
608 so_data (so
, vp
->vp
.attrs
[2] | gp
->vp
.attrs
[2]);
611 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP_SIZE
, 1);
615 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP(0), m
);
616 so_datap (so
, map
, m
);