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;
53 /* FIXME: emit relocs for unsuiTed MM */
54 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 1);
55 OUT_RING (chan
, (start
<< 8) | cbi
);
56 BEGIN_RING_NI(chan
, tesla
, NV50TCL_CB_DATA(0), nr
);
57 OUT_RINGp (chan
, map
, nr
);
64 pipe_buffer_unmap(pipe
, buf
, transfer
);
68 nv50_program_validate_data(struct nv50_context
*nv50
, struct nv50_program
*p
)
70 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
71 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
75 uint32_t *data
= p
->immd
;
76 unsigned count
= p
->immd_size
/ 4;
83 BEGIN_RING(chan
, tesla
, NV50TCL_CB_ADDR
, 1);
84 OUT_RING (chan
, (start
<< 8) | NV50_CB_PMISC
);
85 BEGIN_RING_NI(chan
, tesla
, NV50TCL_CB_DATA(0), nr
);
86 OUT_RINGp (chan
, data
, nr
);
94 /* If the state tracker doesn't change the constbuf, and it is first
95 * validated with a program that doesn't use it, this check prevents
96 * it from even being uploaded. */
98 if (p->parm_size == 0)
103 case PIPE_SHADER_VERTEX
:
106 case PIPE_SHADER_FRAGMENT
:
109 case PIPE_SHADER_GEOMETRY
:
117 nv50_transfer_constbuf(nv50
, nv50
->constbuf
[p
->type
], p
->parm_size
, cbi
);
121 nv50_program_validate_code(struct nv50_context
*nv50
, struct nv50_program
*p
)
123 struct nouveau_channel
*chan
= nv50
->screen
->base
.channel
;
124 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
125 struct nouveau_grobj
*eng2d
= nv50
->screen
->eng2d
;
128 unsigned size
= p
->code_size
;
129 uint32_t *data
= p
->code
;
131 assert(p
->translated
);
133 /* TODO: use a single bo (for each type) for shader code */
136 ret
= nouveau_bo_new(chan
->device
, NOUVEAU_BO_VRAM
, 0x100, size
, &p
->bo
);
139 offset
= p
->code_start
= 0;
141 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_FORMAT
, 2);
142 OUT_RING (chan
, NV50_2D_DST_FORMAT_R8_UNORM
);
144 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_PITCH
, 1);
145 OUT_RING (chan
, 0x40000);
146 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_WIDTH
, 2);
147 OUT_RING (chan
, 0x10000);
151 unsigned nr
= size
/ 4;
153 if (AVAIL_RING(chan
) < 32)
156 nr
= MIN2(nr
, AVAIL_RING(chan
) - 18);
162 BEGIN_RING(chan
, eng2d
, NV50_2D_DST_ADDRESS_HIGH
, 2);
163 OUT_RELOCh(chan
, p
->bo
, offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
);
164 OUT_RELOCl(chan
, p
->bo
, offset
, NOUVEAU_BO_VRAM
| NOUVEAU_BO_WR
);
165 BEGIN_RING(chan
, eng2d
, NV50_2D_SIFC_BITMAP_ENABLE
, 2);
167 OUT_RING (chan
, NV50_2D_SIFC_FORMAT_R8_UNORM
);
168 BEGIN_RING(chan
, eng2d
, NV50_2D_SIFC_WIDTH
, 10);
169 OUT_RING (chan
, nr
* 4);
180 BEGIN_RING_NI(chan
, eng2d
, NV50_2D_SIFC_DATA
, nr
);
181 OUT_RINGp (chan
, data
, nr
);
188 BEGIN_RING(chan
, tesla
, NV50TCL_CODE_CB_FLUSH
, 1);
193 nv50_vp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
195 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
196 struct nouveau_stateobj
*so
= so_new(5, 7, 2);
198 nv50_program_validate_code(nv50
, p
);
200 so_method(so
, tesla
, NV50TCL_VP_ADDRESS_HIGH
, 2);
201 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
202 NOUVEAU_BO_HIGH
, 0, 0);
203 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
204 NOUVEAU_BO_LOW
, 0, 0);
205 so_method(so
, tesla
, NV50TCL_VP_ATTR_EN_0
, 2);
206 so_data (so
, p
->vp
.attrs
[0]);
207 so_data (so
, p
->vp
.attrs
[1]);
208 so_method(so
, tesla
, NV50TCL_VP_REG_ALLOC_RESULT
, 1);
209 so_data (so
, p
->max_out
);
210 so_method(so
, tesla
, NV50TCL_VP_REG_ALLOC_TEMP
, 1);
211 so_data (so
, p
->max_gpr
);
212 so_method(so
, tesla
, NV50TCL_VP_START_ID
, 1);
213 so_data (so
, p
->code_start
);
220 nv50_fp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
222 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
223 struct nouveau_stateobj
*so
= so_new(6, 7, 2);
225 nv50_program_validate_code(nv50
, p
);
227 so_method(so
, tesla
, NV50TCL_FP_ADDRESS_HIGH
, 2);
228 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
229 NOUVEAU_BO_HIGH
, 0, 0);
230 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
231 NOUVEAU_BO_LOW
, 0, 0);
232 so_method(so
, tesla
, NV50TCL_FP_REG_ALLOC_TEMP
, 1);
233 so_data (so
, p
->max_gpr
);
234 so_method(so
, tesla
, NV50TCL_FP_RESULT_COUNT
, 1);
235 so_data (so
, p
->max_out
);
236 so_method(so
, tesla
, NV50TCL_FP_CONTROL
, 1);
237 so_data (so
, p
->fp
.flags
[0]);
238 so_method(so
, tesla
, NV50TCL_FP_CTRL_UNK196C
, 1);
239 so_data (so
, p
->fp
.flags
[1]);
240 so_method(so
, tesla
, NV50TCL_FP_START_ID
, 1);
241 so_data (so
, p
->code_start
);
248 nv50_gp_update_stateobj(struct nv50_context
*nv50
, struct nv50_program
*p
)
250 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
251 struct nouveau_stateobj
*so
= so_new(6, 7, 2);
253 nv50_program_validate_code(nv50
, p
);
255 so_method(so
, tesla
, NV50TCL_GP_ADDRESS_HIGH
, 2);
256 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
257 NOUVEAU_BO_HIGH
, 0, 0);
258 so_reloc (so
, p
->bo
, 0, NOUVEAU_BO_VRAM
| NOUVEAU_BO_RD
|
259 NOUVEAU_BO_LOW
, 0, 0);
260 so_method(so
, tesla
, NV50TCL_GP_REG_ALLOC_TEMP
, 1);
261 so_data (so
, p
->max_gpr
);
262 so_method(so
, tesla
, NV50TCL_GP_REG_ALLOC_RESULT
, 1);
263 so_data (so
, p
->max_out
);
264 so_method(so
, tesla
, NV50TCL_GP_OUTPUT_PRIMITIVE_TYPE
, 1);
265 so_data (so
, p
->gp
.prim_type
);
266 so_method(so
, tesla
, NV50TCL_GP_VERTEX_OUTPUT_COUNT
, 1);
267 so_data (so
, p
->gp
.vert_count
);
268 so_method(so
, tesla
, NV50TCL_GP_START_ID
, 1);
269 so_data (so
, p
->code_start
);
276 nv50_program_validate(struct nv50_program
*p
)
278 p
->translated
= nv50_program_tx(p
);
279 assert(p
->translated
);
280 return p
->translated
;
283 struct nouveau_stateobj
*
284 nv50_vertprog_validate(struct nv50_context
*nv50
)
286 struct nv50_program
*p
= nv50
->vertprog
;
287 struct nouveau_stateobj
*so
= NULL
;
289 if (!p
->translated
) {
290 if (nv50_program_validate(p
))
291 nv50_vp_update_stateobj(nv50
, p
);
296 if (nv50
->dirty
& NV50_NEW_VERTPROG_CB
)
297 nv50_program_validate_data(nv50
, p
);
299 if (!(nv50
->dirty
& NV50_NEW_VERTPROG
))
302 nv50_program_validate_code(nv50
, p
);
308 struct nouveau_stateobj
*
309 nv50_fragprog_validate(struct nv50_context
*nv50
)
311 struct nv50_program
*p
= nv50
->fragprog
;
312 struct nouveau_stateobj
*so
= NULL
;
314 if (!p
->translated
) {
315 if (nv50_program_validate(p
))
316 nv50_fp_update_stateobj(nv50
, p
);
321 if (nv50
->dirty
& NV50_NEW_FRAGPROG_CB
)
322 nv50_program_validate_data(nv50
, p
);
324 if (!(nv50
->dirty
& NV50_NEW_FRAGPROG
))
327 nv50_program_validate_code(nv50
, p
);
333 struct nouveau_stateobj
*
334 nv50_geomprog_validate(struct nv50_context
*nv50
)
336 struct nv50_program
*p
= nv50
->geomprog
;
337 struct nouveau_stateobj
*so
= NULL
;
339 if (!p
->translated
) {
340 if (nv50_program_validate(p
))
341 nv50_gp_update_stateobj(nv50
, p
);
346 if (nv50
->dirty
& NV50_NEW_GEOMPROG_CB
)
347 nv50_program_validate_data(nv50
, p
);
349 if (!(nv50
->dirty
& NV50_NEW_GEOMPROG
))
352 nv50_program_validate_code(nv50
, p
);
358 /* XXX: this might not work correctly in all cases yet: we assume that
359 * an FP generic input that is not written in the VP is gl_PointCoord.
362 nv50_pntc_replace(struct nv50_context
*nv50
, uint32_t pntc
[8], unsigned m
)
364 struct nv50_program
*vp
= nv50
->vertprog
;
365 struct nv50_program
*fp
= nv50
->fragprog
;
368 memset(pntc
, 0, 8 * sizeof(uint32_t));
373 for (i
= 0; i
< fp
->in_nr
; i
++) {
374 unsigned j
, n
= util_bitcount(fp
->in
[i
].mask
);
376 if (fp
->in
[i
].sn
!= TGSI_SEMANTIC_GENERIC
) {
381 for (j
= 0; j
< vp
->out_nr
; ++j
)
382 if (vp
->out
[j
].sn
== fp
->in
[i
].sn
&& vp
->out
[j
].si
== fp
->in
[i
].si
)
385 if (j
< vp
->out_nr
) {
386 uint32_t en
= nv50
->rasterizer
->pipe
.sprite_coord_enable
;
388 if (!(en
& (1 << vp
->out
[j
].si
))) {
394 /* this is either PointCoord or replaced by sprite coords */
395 for (c
= 0; c
< 4; c
++) {
396 if (!(fp
->in
[i
].mask
& (1 << c
)))
398 pntc
[m
/ 8] |= (c
+ 1) << ((m
% 8) * 4);
402 if (nv50
->rasterizer
->pipe
.sprite_coord_mode
== PIPE_SPRITE_COORD_LOWER_LEFT
)
408 nv50_vec4_map(uint32_t *map32
, int mid
, uint32_t lin
[4],
409 struct nv50_varying
*in
, struct nv50_varying
*out
)
412 uint8_t mv
= out
->mask
, mf
= in
->mask
, oid
= out
->hw
;
413 uint8_t *map
= (uint8_t *)map32
;
415 for (c
= 0; c
< 4; ++c
) {
418 lin
[mid
/ 32] |= 1 << (mid
% 32);
435 struct nouveau_stateobj
*
436 nv50_fp_linkage_validate(struct nv50_context
*nv50
)
438 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
439 struct nv50_program
*vp
;
440 struct nv50_program
*fp
= nv50
->fragprog
;
441 struct nouveau_stateobj
*so
;
442 struct nv50_varying dummy
;
445 uint32_t map
[16], lin
[4], pntc
[8];
447 uint32_t interp
= fp
->fp
.interp
;
448 uint32_t colors
= fp
->fp
.colors
;
449 uint32_t clip
= 0x04;
450 uint32_t psiz
= 0x000;
454 if (nv50
->geomprog
) {
456 memset(map
, 0x80, sizeof(map
));
459 memset(map
, 0x40, sizeof(map
));
461 memset(lin
, 0, sizeof(lin
));
464 dummy
.mask
= 0xf; /* map all components of HPOS */
465 m
= nv50_vec4_map(map
, 0, lin
, &dummy
, &vp
->out
[0]);
467 if (vp
->vp
.clpd
< 0x40) {
468 for (c
= 0; c
< vp
->vp
.clpd_nr
; ++c
) {
469 map
[m
/ 4] |= (vp
->vp
.clpd
+ c
) << ((m
% 4) * 8);
472 clip
|= vp
->vp
.clpd_nr
<< 8;
475 colors
|= m
<< 8; /* adjust BFC0 id */
477 /* if light_twoside is active, it seems FFC0_ID == BFC0_ID is bad */
478 if (nv50
->rasterizer
->pipe
.light_twoside
) {
479 for (i
= 0; i
< 2; ++i
)
480 m
= nv50_vec4_map(map
, m
, lin
,
481 &fp
->in
[fp
->vp
.bfc
[i
]],
482 &vp
->out
[vp
->vp
.bfc
[i
]]);
485 colors
+= m
- 4; /* adjust FFC0 id */
486 interp
|= m
<< 8; /* set mid where 'normal' FP inputs start */
489 for (i
= 0; i
< fp
->in_nr
; i
++) {
490 for (n
= 0; n
< vp
->out_nr
; ++n
)
491 if (vp
->out
[n
].sn
== fp
->in
[i
].sn
&&
492 vp
->out
[n
].si
== fp
->in
[i
].si
)
495 m
= nv50_vec4_map(map
, m
, lin
,
496 &fp
->in
[i
], (n
< vp
->out_nr
) ? &vp
->out
[n
] : &dummy
);
499 /* PrimitiveID either is replaced by the system value, or
500 * written by the geometry shader into an output register
502 if (fp
->gp
.primid
< 0x40) {
504 map
[m
/ 4] = (map
[m
/ 4] & ~(0xff << i
)) | (vp
->gp
.primid
<< i
);
508 if (nv50
->rasterizer
->pipe
.point_size_per_vertex
) {
510 map
[m
/ 4] = (map
[m
/ 4] & ~(0xff << i
)) | (vp
->vp
.psiz
<< i
);
511 psiz
= (m
++ << 4) | 1;
514 /* now fill the stateobj (at most 28 so_data) */
515 so
= so_new(10, 54, 0);
519 if (vp
->type
== PIPE_SHADER_GEOMETRY
) {
520 so_method(so
, tesla
, NV50TCL_GP_RESULT_MAP_SIZE
, 1);
522 so_method(so
, tesla
, NV50TCL_GP_RESULT_MAP(0), n
);
523 so_datap (so
, map
, n
);
525 so_method(so
, tesla
, NV50TCL_VP_GP_BUILTIN_ATTR_EN
, 1);
526 so_data (so
, vp
->vp
.attrs
[2]);
528 so_method(so
, tesla
, NV50TCL_MAP_SEMANTIC_4
, 1);
529 so_data (so
, primid
);
531 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP_SIZE
, 1);
533 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP(0), n
);
534 so_datap (so
, map
, n
);
537 so_method(so
, tesla
, NV50TCL_MAP_SEMANTIC_0
, 4);
538 so_data (so
, colors
);
540 so_data (so
, sysval
);
543 so_method(so
, tesla
, NV50TCL_FP_INTERPOLANT_CTRL
, 1);
544 so_data (so
, interp
);
546 so_method(so
, tesla
, NV50TCL_NOPERSPECTIVE_BITMAP(0), 4);
547 so_datap (so
, lin
, 4);
549 if (nv50
->rasterizer
->pipe
.point_quad_rasterization
) {
550 so_method(so
, tesla
, NV50TCL_POINT_SPRITE_CTRL
, 1);
552 nv50_pntc_replace(nv50
, pntc
, (interp
>> 8) & 0xff));
554 so_method(so
, tesla
, NV50TCL_POINT_COORD_REPLACE_MAP(0), 8);
555 so_datap (so
, pntc
, 8);
558 so_method(so
, tesla
, NV50TCL_GP_ENABLE
, 1);
559 so_data (so
, (vp
->type
== PIPE_SHADER_GEOMETRY
) ? 1 : 0);
565 nv50_vp_gp_mapping(uint32_t *map32
, int m
,
566 struct nv50_program
*vp
, struct nv50_program
*gp
)
568 uint8_t *map
= (uint8_t *)map32
;
571 for (i
= 0; i
< gp
->in_nr
; ++i
) {
572 uint8_t oid
= 0, mv
= 0, mg
= gp
->in
[i
].mask
;
574 for (j
= 0; j
< vp
->out_nr
; ++j
) {
575 if (vp
->out
[j
].sn
== gp
->in
[i
].sn
&&
576 vp
->out
[j
].si
== gp
->in
[i
].si
) {
577 mv
= vp
->out
[j
].mask
;
583 for (c
= 0; c
< 4; ++c
, mv
>>= 1, mg
>>= 1) {
588 map
[m
++] = (c
== 3) ? 0x41 : 0x40;
595 struct nouveau_stateobj
*
596 nv50_gp_linkage_validate(struct nv50_context
*nv50
)
598 struct nouveau_grobj
*tesla
= nv50
->screen
->tesla
;
599 struct nouveau_stateobj
*so
;
600 struct nv50_program
*vp
= nv50
->vertprog
;
601 struct nv50_program
*gp
= nv50
->geomprog
;
607 memset(map
, 0, sizeof(map
));
609 m
= nv50_vp_gp_mapping(map
, m
, vp
, gp
);
611 so
= so_new(3, 24 - 3, 0);
613 so_method(so
, tesla
, NV50TCL_VP_GP_BUILTIN_ATTR_EN
, 1);
614 so_data (so
, vp
->vp
.attrs
[2] | gp
->vp
.attrs
[2]);
617 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP_SIZE
, 1);
621 so_method(so
, tesla
, NV50TCL_VP_RESULT_MAP(0), m
);
622 so_datap (so
, map
, m
);