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 OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
19 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
20 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
21 * OTHER DEALINGS IN THE SOFTWARE.
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/nv50_context.h"
30 #include "nv50/nv50_query_hw.h"
32 #include "nv50/nv50_compute.xml.h"
35 nv50_constbufs_validate(struct nv50_context
*nv50
)
37 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
40 for (s
= 0; s
< 3; ++s
) {
43 if (s
== PIPE_SHADER_FRAGMENT
)
44 p
= NV50_3D_SET_PROGRAM_CB_PROGRAM_FRAGMENT
;
46 if (s
== PIPE_SHADER_GEOMETRY
)
47 p
= NV50_3D_SET_PROGRAM_CB_PROGRAM_GEOMETRY
;
49 p
= NV50_3D_SET_PROGRAM_CB_PROGRAM_VERTEX
;
51 while (nv50
->constbuf_dirty
[s
]) {
52 const unsigned i
= (unsigned)ffs(nv50
->constbuf_dirty
[s
]) - 1;
54 assert(i
< NV50_MAX_PIPE_CONSTBUFS
);
55 nv50
->constbuf_dirty
[s
] &= ~(1 << i
);
57 if (nv50
->constbuf
[s
][i
].user
) {
58 const unsigned b
= NV50_CB_PVP
+ s
;
60 unsigned words
= nv50
->constbuf
[s
][0].size
/ 4;
62 NOUVEAU_ERR("user constbufs only supported in slot 0\n");
65 if (!nv50
->state
.uniform_buffer_bound
[s
]) {
66 nv50
->state
.uniform_buffer_bound
[s
] = true;
67 BEGIN_NV04(push
, NV50_3D(SET_PROGRAM_CB
), 1);
68 PUSH_DATA (push
, (b
<< 12) | (i
<< 8) | p
| 1);
71 unsigned nr
= MIN2(words
, NV04_PFIFO_MAX_PACKET_LEN
);
73 PUSH_SPACE(push
, nr
+ 3);
74 BEGIN_NV04(push
, NV50_3D(CB_ADDR
), 1);
75 PUSH_DATA (push
, (start
<< 8) | b
);
76 BEGIN_NI04(push
, NV50_3D(CB_DATA(0)), nr
);
77 PUSH_DATAp(push
, &nv50
->constbuf
[s
][0].u
.data
[start
* 4], nr
);
83 struct nv04_resource
*res
=
84 nv04_resource(nv50
->constbuf
[s
][i
].u
.buf
);
86 /* TODO: allocate persistent bindings */
87 const unsigned b
= s
* 16 + i
;
89 assert(nouveau_resource_mapped_by_gpu(&res
->base
));
91 BEGIN_NV04(push
, NV50_3D(CB_DEF_ADDRESS_HIGH
), 3);
92 PUSH_DATAh(push
, res
->address
+ nv50
->constbuf
[s
][i
].offset
);
93 PUSH_DATA (push
, res
->address
+ nv50
->constbuf
[s
][i
].offset
);
94 PUSH_DATA (push
, (b
<< 16) |
95 (nv50
->constbuf
[s
][i
].size
& 0xffff));
96 BEGIN_NV04(push
, NV50_3D(SET_PROGRAM_CB
), 1);
97 PUSH_DATA (push
, (b
<< 12) | (i
<< 8) | p
| 1);
99 BCTX_REFN(nv50
->bufctx_3d
, 3D_CB(s
, i
), res
, RD
);
101 nv50
->cb_dirty
= 1; /* Force cache flush for UBO. */
102 res
->cb_bindings
[s
] |= 1 << i
;
104 BEGIN_NV04(push
, NV50_3D(SET_PROGRAM_CB
), 1);
105 PUSH_DATA (push
, (i
<< 8) | p
| 0);
108 nv50
->state
.uniform_buffer_bound
[s
] = false;
115 nv50_program_validate(struct nv50_context
*nv50
, struct nv50_program
*prog
)
117 if (!prog
->translated
) {
118 prog
->translated
= nv50_program_translate(
119 prog
, nv50
->screen
->base
.device
->chipset
, &nv50
->base
.debug
);
120 if (!prog
->translated
)
126 return nv50_program_upload_code(nv50
, prog
);
130 nv50_program_update_context_state(struct nv50_context
*nv50
,
131 struct nv50_program
*prog
, int stage
)
133 const unsigned flags
= NOUVEAU_BO_VRAM
| NOUVEAU_BO_RDWR
;
135 if (prog
&& prog
->tls_space
) {
136 if (nv50
->state
.new_tls_space
)
137 nouveau_bufctx_reset(nv50
->bufctx_3d
, NV50_BIND_3D_TLS
);
138 if (!nv50
->state
.tls_required
|| nv50
->state
.new_tls_space
)
139 BCTX_REFN_bo(nv50
->bufctx_3d
, 3D_TLS
, flags
, nv50
->screen
->tls_bo
);
140 nv50
->state
.new_tls_space
= false;
141 nv50
->state
.tls_required
|= 1 << stage
;
143 if (nv50
->state
.tls_required
== (1 << stage
))
144 nouveau_bufctx_reset(nv50
->bufctx_3d
, NV50_BIND_3D_TLS
);
145 nv50
->state
.tls_required
&= ~(1 << stage
);
150 nv50_vertprog_validate(struct nv50_context
*nv50
)
152 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
153 struct nv50_program
*vp
= nv50
->vertprog
;
155 if (!nv50_program_validate(nv50
, vp
))
157 nv50_program_update_context_state(nv50
, vp
, 0);
159 BEGIN_NV04(push
, NV50_3D(VP_ATTR_EN(0)), 2);
160 PUSH_DATA (push
, vp
->vp
.attrs
[0]);
161 PUSH_DATA (push
, vp
->vp
.attrs
[1]);
162 BEGIN_NV04(push
, NV50_3D(VP_REG_ALLOC_RESULT
), 1);
163 PUSH_DATA (push
, vp
->max_out
);
164 BEGIN_NV04(push
, NV50_3D(VP_REG_ALLOC_TEMP
), 1);
165 PUSH_DATA (push
, vp
->max_gpr
);
166 BEGIN_NV04(push
, NV50_3D(VP_START_ID
), 1);
167 PUSH_DATA (push
, vp
->code_base
);
171 nv50_fragprog_validate(struct nv50_context
*nv50
)
173 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
174 struct nv50_program
*fp
= nv50
->fragprog
;
175 struct pipe_rasterizer_state
*rast
= &nv50
->rast
->pipe
;
177 if (fp
->fp
.force_persample_interp
!= rast
->force_persample_interp
) {
178 /* Force the program to be reuploaded, which will trigger interp fixups
182 nouveau_heap_free(&fp
->mem
);
184 fp
->fp
.force_persample_interp
= rast
->force_persample_interp
;
187 if (fp
->mem
&& !(nv50
->dirty_3d
& (NV50_NEW_3D_FRAGPROG
| NV50_NEW_3D_MIN_SAMPLES
)))
190 if (!nv50_program_validate(nv50
, fp
))
192 nv50_program_update_context_state(nv50
, fp
, 1);
194 BEGIN_NV04(push
, NV50_3D(FP_REG_ALLOC_TEMP
), 1);
195 PUSH_DATA (push
, fp
->max_gpr
);
196 BEGIN_NV04(push
, NV50_3D(FP_RESULT_COUNT
), 1);
197 PUSH_DATA (push
, fp
->max_out
);
198 BEGIN_NV04(push
, NV50_3D(FP_CONTROL
), 1);
199 PUSH_DATA (push
, fp
->fp
.flags
[0]);
200 BEGIN_NV04(push
, NV50_3D(FP_CTRL_UNK196C
), 1);
201 PUSH_DATA (push
, fp
->fp
.flags
[1]);
202 BEGIN_NV04(push
, NV50_3D(FP_START_ID
), 1);
203 PUSH_DATA (push
, fp
->code_base
);
205 if (nv50
->screen
->tesla
->oclass
>= NVA3_3D_CLASS
) {
206 BEGIN_NV04(push
, SUBC_3D(NVA3_3D_FP_MULTISAMPLE
), 1);
207 if (nv50
->min_samples
> 1 || fp
->fp
.has_samplemask
)
209 NVA3_3D_FP_MULTISAMPLE_FORCE_PER_SAMPLE
|
210 (NVA3_3D_FP_MULTISAMPLE_EXPORT_SAMPLE_MASK
*
211 fp
->fp
.has_samplemask
));
218 nv50_gmtyprog_validate(struct nv50_context
*nv50
)
220 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
221 struct nv50_program
*gp
= nv50
->gmtyprog
;
224 if (!nv50_program_validate(nv50
, gp
))
226 BEGIN_NV04(push
, NV50_3D(GP_REG_ALLOC_TEMP
), 1);
227 PUSH_DATA (push
, gp
->max_gpr
);
228 BEGIN_NV04(push
, NV50_3D(GP_REG_ALLOC_RESULT
), 1);
229 PUSH_DATA (push
, gp
->max_out
);
230 BEGIN_NV04(push
, NV50_3D(GP_OUTPUT_PRIMITIVE_TYPE
), 1);
231 PUSH_DATA (push
, gp
->gp
.prim_type
);
232 BEGIN_NV04(push
, NV50_3D(GP_VERTEX_OUTPUT_COUNT
), 1);
233 PUSH_DATA (push
, gp
->gp
.vert_count
);
234 BEGIN_NV04(push
, NV50_3D(GP_START_ID
), 1);
235 PUSH_DATA (push
, gp
->code_base
);
237 nv50
->state
.prim_size
= gp
->gp
.prim_type
; /* enum matches vertex count */
239 nv50_program_update_context_state(nv50
, gp
, 2);
241 /* GP_ENABLE is updated in linkage validation */
245 nv50_compprog_validate(struct nv50_context
*nv50
)
247 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
248 struct nv50_program
*cp
= nv50
->compprog
;
250 if (cp
&& !nv50_program_validate(nv50
, cp
))
253 BEGIN_NV04(push
, NV50_CP(CODE_CB_FLUSH
), 1);
258 nv50_sprite_coords_validate(struct nv50_context
*nv50
)
260 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
261 uint32_t pntc
[8], mode
;
262 struct nv50_program
*fp
= nv50
->fragprog
;
264 unsigned m
= (nv50
->state
.interpolant_ctrl
>> 8) & 0xff;
266 if (!nv50
->rast
->pipe
.point_quad_rasterization
) {
267 if (nv50
->state
.point_sprite
) {
268 BEGIN_NV04(push
, NV50_3D(POINT_COORD_REPLACE_MAP(0)), 8);
269 for (i
= 0; i
< 8; ++i
)
272 nv50
->state
.point_sprite
= false;
276 nv50
->state
.point_sprite
= true;
279 memset(pntc
, 0, sizeof(pntc
));
281 for (i
= 0; i
< fp
->in_nr
; i
++) {
282 unsigned n
= util_bitcount(fp
->in
[i
].mask
);
284 if (fp
->in
[i
].sn
!= TGSI_SEMANTIC_GENERIC
) {
288 if (!(nv50
->rast
->pipe
.sprite_coord_enable
& (1 << fp
->in
[i
].si
))) {
293 for (c
= 0; c
< 4; ++c
) {
294 if (fp
->in
[i
].mask
& (1 << c
)) {
295 pntc
[m
/ 8] |= (c
+ 1) << ((m
% 8) * 4);
301 if (nv50
->rast
->pipe
.sprite_coord_mode
== PIPE_SPRITE_COORD_LOWER_LEFT
)
306 BEGIN_NV04(push
, NV50_3D(POINT_SPRITE_CTRL
), 1);
307 PUSH_DATA (push
, mode
);
309 BEGIN_NV04(push
, NV50_3D(POINT_COORD_REPLACE_MAP(0)), 8);
310 PUSH_DATAp(push
, pntc
, 8);
313 /* Validate state derived from shaders and the rasterizer cso. */
315 nv50_validate_derived_rs(struct nv50_context
*nv50
)
317 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
318 uint32_t color
, psize
;
320 nv50_sprite_coords_validate(nv50
);
322 if (nv50
->state
.rasterizer_discard
!= nv50
->rast
->pipe
.rasterizer_discard
) {
323 nv50
->state
.rasterizer_discard
= nv50
->rast
->pipe
.rasterizer_discard
;
324 BEGIN_NV04(push
, NV50_3D(RASTERIZE_ENABLE
), 1);
325 PUSH_DATA (push
, !nv50
->rast
->pipe
.rasterizer_discard
);
328 if (nv50
->dirty_3d
& NV50_NEW_3D_FRAGPROG
)
330 psize
= nv50
->state
.semantic_psize
& ~NV50_3D_SEMANTIC_PTSZ_PTSZ_EN__MASK
;
331 color
= nv50
->state
.semantic_color
& ~NV50_3D_SEMANTIC_COLOR_CLMP_EN
;
333 if (nv50
->rast
->pipe
.clamp_vertex_color
)
334 color
|= NV50_3D_SEMANTIC_COLOR_CLMP_EN
;
336 if (color
!= nv50
->state
.semantic_color
) {
337 nv50
->state
.semantic_color
= color
;
338 BEGIN_NV04(push
, NV50_3D(SEMANTIC_COLOR
), 1);
339 PUSH_DATA (push
, color
);
342 if (nv50
->rast
->pipe
.point_size_per_vertex
)
343 psize
|= NV50_3D_SEMANTIC_PTSZ_PTSZ_EN__MASK
;
345 if (psize
!= nv50
->state
.semantic_psize
) {
346 nv50
->state
.semantic_psize
= psize
;
347 BEGIN_NV04(push
, NV50_3D(SEMANTIC_PTSZ
), 1);
348 PUSH_DATA (push
, psize
);
353 nv50_vec4_map(uint8_t *map
, int mid
, uint32_t lin
[4],
354 struct nv50_varying
*in
, struct nv50_varying
*out
)
357 uint8_t mv
= out
->mask
, mf
= in
->mask
, oid
= out
->hw
;
359 for (c
= 0; c
< 4; ++c
) {
362 lin
[mid
/ 32] |= 1 << (mid
% 32);
380 nv50_fp_linkage_validate(struct nv50_context
*nv50
)
382 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
383 struct nv50_program
*vp
= nv50
->gmtyprog
? nv50
->gmtyprog
: nv50
->vertprog
;
384 struct nv50_program
*fp
= nv50
->fragprog
;
385 struct nv50_varying dummy
;
388 uint32_t layerid
= 0;
389 uint32_t viewportid
= 0;
390 uint32_t psiz
= 0x000;
391 uint32_t interp
= fp
->fp
.interp
;
392 uint32_t colors
= fp
->fp
.colors
;
393 uint32_t clpd_nr
= util_last_bit(vp
->vp
.clip_enable
| vp
->vp
.cull_enable
);
398 if (!(nv50
->dirty_3d
& (NV50_NEW_3D_VERTPROG
|
399 NV50_NEW_3D_FRAGPROG
|
400 NV50_NEW_3D_GMTYPROG
))) {
402 ffc
= (nv50
->state
.semantic_color
& NV50_3D_SEMANTIC_COLOR_FFC0_ID__MASK
);
403 bfc
= (nv50
->state
.semantic_color
& NV50_3D_SEMANTIC_COLOR_BFC0_ID__MASK
)
405 if (nv50
->rast
->pipe
.light_twoside
== ((ffc
== bfc
) ? 0 : 1))
409 memset(lin
, 0x00, sizeof(lin
));
411 /* XXX: in buggy-endian mode, is the first element of map (u32)0x000000xx
412 * or is it the first byte ?
414 memset(map
, nv50
->gmtyprog
? 0x80 : 0x40, sizeof(map
));
416 dummy
.mask
= 0xf; /* map all components of HPOS */
418 m
= nv50_vec4_map(map
, 0, lin
, &dummy
, &vp
->out
[0]);
420 for (c
= 0; c
< clpd_nr
; ++c
)
421 map
[m
++] = vp
->vp
.clpd
[c
/ 4] + (c
% 4);
423 colors
|= m
<< 8; /* adjust BFC0 id */
427 /* if light_twoside is active, FFC0_ID == BFC0_ID is invalid */
428 if (nv50
->rast
->pipe
.light_twoside
) {
429 for (i
= 0; i
< 2; ++i
) {
431 if (fp
->vp
.bfc
[i
] >= fp
->in_nr
)
433 m
= nv50_vec4_map(map
, m
, lin
, &fp
->in
[fp
->vp
.bfc
[i
]],
434 (n
< vp
->out_nr
) ? &vp
->out
[n
] : &dummy
);
437 colors
+= m
- 4; /* adjust FFC0 id */
438 interp
|= m
<< 8; /* set map id where 'normal' FP inputs start */
440 for (i
= 0; i
< fp
->in_nr
; ++i
) {
441 for (n
= 0; n
< vp
->out_nr
; ++n
)
442 if (vp
->out
[n
].sn
== fp
->in
[i
].sn
&&
443 vp
->out
[n
].si
== fp
->in
[i
].si
)
445 switch (fp
->in
[i
].sn
) {
446 case TGSI_SEMANTIC_PRIMID
:
449 case TGSI_SEMANTIC_LAYER
:
452 case TGSI_SEMANTIC_VIEWPORT_INDEX
:
456 m
= nv50_vec4_map(map
, m
, lin
,
457 &fp
->in
[i
], (n
< vp
->out_nr
) ? &vp
->out
[n
] : &dummy
);
460 if (vp
->gp
.has_layer
&& !layerid
) {
462 map
[m
++] = vp
->gp
.layerid
;
465 if (vp
->gp
.has_viewport
&& !viewportid
) {
467 map
[m
++] = vp
->gp
.viewportid
;
470 if (nv50
->rast
->pipe
.point_size_per_vertex
) {
472 map
[m
++] = vp
->vp
.psiz
;
475 if (nv50
->rast
->pipe
.clamp_vertex_color
)
476 colors
|= NV50_3D_SEMANTIC_COLOR_CLMP_EN
;
478 if (unlikely(vp
->so
)) {
479 /* Slot i in STRMOUT_MAP specifies the offset where slot i in RESULT_MAP
483 * Inverting vp->so->map (output -> offset) would probably speed this up.
485 memset(so_map
, 0, sizeof(so_map
));
486 for (i
= 0; i
< vp
->so
->map_size
; ++i
) {
487 if (vp
->so
->map
[i
] == 0xff)
489 for (c
= 0; c
< m
; ++c
)
490 if (map
[c
] == vp
->so
->map
[i
] && !so_map
[c
])
494 map
[m
++] = vp
->so
->map
[i
];
496 so_map
[c
] = 0x80 | i
;
498 for (c
= m
; c
& 3; ++c
)
505 if (unlikely(nv50
->gmtyprog
)) {
506 BEGIN_NV04(push
, NV50_3D(GP_RESULT_MAP_SIZE
), 1);
508 BEGIN_NV04(push
, NV50_3D(GP_RESULT_MAP(0)), n
);
509 PUSH_DATAp(push
, map
, n
);
511 BEGIN_NV04(push
, NV50_3D(VP_GP_BUILTIN_ATTR_EN
), 1);
512 PUSH_DATA (push
, vp
->vp
.attrs
[2] | fp
->vp
.attrs
[2]);
514 BEGIN_NV04(push
, NV50_3D(SEMANTIC_PRIM_ID
), 1);
515 PUSH_DATA (push
, primid
);
518 BEGIN_NV04(push
, NV50_3D(VP_RESULT_MAP_SIZE
), 1);
520 BEGIN_NV04(push
, NV50_3D(VP_RESULT_MAP(0)), n
);
521 PUSH_DATAp(push
, map
, n
);
524 BEGIN_NV04(push
, NV50_3D(GP_VIEWPORT_ID_ENABLE
), 5);
525 PUSH_DATA (push
, vp
->gp
.has_viewport
);
526 PUSH_DATA (push
, colors
);
527 PUSH_DATA (push
, (clpd_nr
<< 8) | 4);
528 PUSH_DATA (push
, layerid
);
529 PUSH_DATA (push
, psiz
);
531 BEGIN_NV04(push
, NV50_3D(SEMANTIC_VIEWPORT
), 1);
532 PUSH_DATA (push
, viewportid
);
534 BEGIN_NV04(push
, NV50_3D(LAYER
), 1);
535 PUSH_DATA (push
, vp
->gp
.has_layer
<< 16);
537 BEGIN_NV04(push
, NV50_3D(FP_INTERPOLANT_CTRL
), 1);
538 PUSH_DATA (push
, interp
);
540 nv50
->state
.interpolant_ctrl
= interp
;
542 nv50
->state
.semantic_color
= colors
;
543 nv50
->state
.semantic_psize
= psiz
;
545 BEGIN_NV04(push
, NV50_3D(NOPERSPECTIVE_BITMAP(0)), 4);
546 PUSH_DATAp(push
, lin
, 4);
548 BEGIN_NV04(push
, NV50_3D(GP_ENABLE
), 1);
549 PUSH_DATA (push
, nv50
->gmtyprog
? 1 : 0);
552 BEGIN_NV04(push
, NV50_3D(STRMOUT_MAP(0)), n
);
553 PUSH_DATAp(push
, so_map
, n
);
558 nv50_vp_gp_mapping(uint8_t *map
, int m
,
559 struct nv50_program
*vp
, struct nv50_program
*gp
)
563 for (i
= 0; i
< gp
->in_nr
; ++i
) {
564 uint8_t oid
= 0, mv
= 0, mg
= gp
->in
[i
].mask
;
566 for (j
= 0; j
< vp
->out_nr
; ++j
) {
567 if (vp
->out
[j
].sn
== gp
->in
[i
].sn
&&
568 vp
->out
[j
].si
== gp
->in
[i
].si
) {
569 mv
= vp
->out
[j
].mask
;
575 for (c
= 0; c
< 4; ++c
, mv
>>= 1, mg
>>= 1) {
580 map
[m
++] = (c
== 3) ? 0x41 : 0x40;
590 nv50_gp_linkage_validate(struct nv50_context
*nv50
)
592 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
593 struct nv50_program
*vp
= nv50
->vertprog
;
594 struct nv50_program
*gp
= nv50
->gmtyprog
;
601 memset(map
, 0, sizeof(map
));
603 m
= nv50_vp_gp_mapping(map
, m
, vp
, gp
);
607 BEGIN_NV04(push
, NV50_3D(VP_GP_BUILTIN_ATTR_EN
), 1);
608 PUSH_DATA (push
, vp
->vp
.attrs
[2] | gp
->vp
.attrs
[2]);
611 BEGIN_NV04(push
, NV50_3D(VP_RESULT_MAP_SIZE
), 1);
613 BEGIN_NV04(push
, NV50_3D(VP_RESULT_MAP(0)), n
);
614 PUSH_DATAp(push
, map
, n
);
618 nv50_stream_output_validate(struct nv50_context
*nv50
)
620 struct nouveau_pushbuf
*push
= nv50
->base
.pushbuf
;
621 struct nv50_stream_output_state
*so
;
626 so
= nv50
->gmtyprog
? nv50
->gmtyprog
->so
: nv50
->vertprog
->so
;
628 BEGIN_NV04(push
, NV50_3D(STRMOUT_ENABLE
), 1);
630 if (!so
|| !nv50
->num_so_targets
) {
631 if (nv50
->screen
->base
.class_3d
< NVA0_3D_CLASS
) {
632 BEGIN_NV04(push
, NV50_3D(STRMOUT_PRIMITIVE_LIMIT
), 1);
635 BEGIN_NV04(push
, NV50_3D(STRMOUT_PARAMS_LATCH
), 1);
640 /* previous TFB needs to complete */
641 if (nv50
->screen
->base
.class_3d
< NVA0_3D_CLASS
) {
642 BEGIN_NV04(push
, SUBC_3D(NV50_GRAPH_SERIALIZE
), 1);
647 if (nv50
->screen
->base
.class_3d
>= NVA0_3D_CLASS
)
648 ctrl
|= NVA0_3D_STRMOUT_BUFFERS_CTRL_LIMIT_MODE_OFFSET
;
650 BEGIN_NV04(push
, NV50_3D(STRMOUT_BUFFERS_CTRL
), 1);
651 PUSH_DATA (push
, ctrl
);
653 for (i
= 0; i
< nv50
->num_so_targets
; ++i
) {
654 struct nv50_so_target
*targ
= nv50_so_target(nv50
->so_target
[i
]);
655 struct nv04_resource
*buf
= nv04_resource(targ
->pipe
.buffer
);
657 const unsigned n
= nv50
->screen
->base
.class_3d
>= NVA0_3D_CLASS
? 4 : 3;
659 if (n
== 4 && !targ
->clean
)
660 nv84_hw_query_fifo_wait(push
, nv50_query(targ
->pq
));
661 BEGIN_NV04(push
, NV50_3D(STRMOUT_ADDRESS_HIGH(i
)), n
);
662 PUSH_DATAh(push
, buf
->address
+ targ
->pipe
.buffer_offset
);
663 PUSH_DATA (push
, buf
->address
+ targ
->pipe
.buffer_offset
);
664 PUSH_DATA (push
, so
->num_attribs
[i
]);
666 PUSH_DATA(push
, targ
->pipe
.buffer_size
);
669 nv50_hw_query_pushbuf_submit(push
, NVA0_3D_STRMOUT_OFFSET(i
),
670 nv50_query(targ
->pq
), 0x4);
672 BEGIN_NV04(push
, NVA0_3D(STRMOUT_OFFSET(i
)), 1);
677 const unsigned limit
= targ
->pipe
.buffer_size
/
678 (so
->stride
[i
] * nv50
->state
.prim_size
);
679 prims
= MIN2(prims
, limit
);
681 targ
->stride
= so
->stride
[i
];
682 BCTX_REFN(nv50
->bufctx_3d
, 3D_SO
, buf
, WR
);
685 BEGIN_NV04(push
, NV50_3D(STRMOUT_PRIMITIVE_LIMIT
), 1);
686 PUSH_DATA (push
, prims
);
688 BEGIN_NV04(push
, NV50_3D(STRMOUT_PARAMS_LATCH
), 1);
690 BEGIN_NV04(push
, NV50_3D(STRMOUT_ENABLE
), 1);