nv50: reinstate dedicated constbuf push path
[mesa.git] / src / gallium / drivers / nouveau / nv50 / nv50_shader_state.c
1 /*
2 * Copyright 2008 Ben Skeggs
3 * Copyright 2010 Christoph Bumiller
4 *
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:
11 *
12 * The above copyright notice and this permission notice shall be included in
13 * all copies or substantial portions of the Software.
14 *
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.
22 */
23
24 #include "pipe/p_context.h"
25 #include "pipe/p_defines.h"
26 #include "pipe/p_state.h"
27 #include "util/u_inlines.h"
28
29 #include "nv50/nv50_context.h"
30 #include "nv50/nv50_query_hw.h"
31
32 #include "nv50/nv50_compute.xml.h"
33
34 void
35 nv50_constbufs_validate(struct nv50_context *nv50)
36 {
37 struct nouveau_pushbuf *push = nv50->base.pushbuf;
38 unsigned s;
39
40 for (s = 0; s < 3; ++s) {
41 unsigned p;
42
43 if (s == PIPE_SHADER_FRAGMENT)
44 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_FRAGMENT;
45 else
46 if (s == PIPE_SHADER_GEOMETRY)
47 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_GEOMETRY;
48 else
49 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_VERTEX;
50
51 while (nv50->constbuf_dirty[s]) {
52 const unsigned i = (unsigned)ffs(nv50->constbuf_dirty[s]) - 1;
53
54 assert(i < NV50_MAX_PIPE_CONSTBUFS);
55 nv50->constbuf_dirty[s] &= ~(1 << i);
56
57 if (nv50->constbuf[s][i].user) {
58 const unsigned b = NV50_CB_PVP + s;
59 unsigned start = 0;
60 unsigned words = nv50->constbuf[s][0].size / 4;
61 if (i) {
62 NOUVEAU_ERR("user constbufs only supported in slot 0\n");
63 continue;
64 }
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);
69 }
70 while (words) {
71 unsigned nr = MIN2(words, NV04_PFIFO_MAX_PACKET_LEN);
72
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);
78
79 start += nr;
80 words -= nr;
81 }
82 } else {
83 struct nv04_resource *res =
84 nv04_resource(nv50->constbuf[s][i].u.buf);
85 if (res) {
86 /* TODO: allocate persistent bindings */
87 const unsigned b = s * 16 + i;
88
89 assert(nouveau_resource_mapped_by_gpu(&res->base));
90
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);
98
99 BCTX_REFN(nv50->bufctx_3d, 3D_CB(s, i), res, RD);
100
101 nv50->cb_dirty = 1; /* Force cache flush for UBO. */
102 res->cb_bindings[s] |= 1 << i;
103 } else {
104 BEGIN_NV04(push, NV50_3D(SET_PROGRAM_CB), 1);
105 PUSH_DATA (push, (i << 8) | p | 0);
106 }
107 if (i == 0)
108 nv50->state.uniform_buffer_bound[s] = false;
109 }
110 }
111 }
112 }
113
114 static bool
115 nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog)
116 {
117 if (!prog->translated) {
118 prog->translated = nv50_program_translate(
119 prog, nv50->screen->base.device->chipset, &nv50->base.debug);
120 if (!prog->translated)
121 return false;
122 } else
123 if (prog->mem)
124 return true;
125
126 return nv50_program_upload_code(nv50, prog);
127 }
128
129 static inline void
130 nv50_program_update_context_state(struct nv50_context *nv50,
131 struct nv50_program *prog, int stage)
132 {
133 const unsigned flags = NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR;
134
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;
142 } else {
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);
146 }
147 }
148
149 void
150 nv50_vertprog_validate(struct nv50_context *nv50)
151 {
152 struct nouveau_pushbuf *push = nv50->base.pushbuf;
153 struct nv50_program *vp = nv50->vertprog;
154
155 if (!nv50_program_validate(nv50, vp))
156 return;
157 nv50_program_update_context_state(nv50, vp, 0);
158
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);
168 }
169
170 void
171 nv50_fragprog_validate(struct nv50_context *nv50)
172 {
173 struct nouveau_pushbuf *push = nv50->base.pushbuf;
174 struct nv50_program *fp = nv50->fragprog;
175 struct pipe_rasterizer_state *rast = &nv50->rast->pipe;
176
177 if (fp->fp.force_persample_interp != rast->force_persample_interp) {
178 /* Force the program to be reuploaded, which will trigger interp fixups
179 * to get applied
180 */
181 if (fp->mem)
182 nouveau_heap_free(&fp->mem);
183
184 fp->fp.force_persample_interp = rast->force_persample_interp;
185 }
186
187 if (fp->mem && !(nv50->dirty_3d & (NV50_NEW_3D_FRAGPROG | NV50_NEW_3D_MIN_SAMPLES)))
188 return;
189
190 if (!nv50_program_validate(nv50, fp))
191 return;
192 nv50_program_update_context_state(nv50, fp, 1);
193
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);
204
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)
208 PUSH_DATA(push,
209 NVA3_3D_FP_MULTISAMPLE_FORCE_PER_SAMPLE |
210 (NVA3_3D_FP_MULTISAMPLE_EXPORT_SAMPLE_MASK *
211 fp->fp.has_samplemask));
212 else
213 PUSH_DATA(push, 0);
214 }
215 }
216
217 void
218 nv50_gmtyprog_validate(struct nv50_context *nv50)
219 {
220 struct nouveau_pushbuf *push = nv50->base.pushbuf;
221 struct nv50_program *gp = nv50->gmtyprog;
222
223 if (gp) {
224 if (!nv50_program_validate(nv50, gp))
225 return;
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);
236
237 nv50->state.prim_size = gp->gp.prim_type; /* enum matches vertex count */
238 }
239 nv50_program_update_context_state(nv50, gp, 2);
240
241 /* GP_ENABLE is updated in linkage validation */
242 }
243
244 void
245 nv50_compprog_validate(struct nv50_context *nv50)
246 {
247 struct nouveau_pushbuf *push = nv50->base.pushbuf;
248 struct nv50_program *cp = nv50->compprog;
249
250 if (cp && !nv50_program_validate(nv50, cp))
251 return;
252
253 BEGIN_NV04(push, NV50_CP(CODE_CB_FLUSH), 1);
254 PUSH_DATA (push, 0);
255 }
256
257 static void
258 nv50_sprite_coords_validate(struct nv50_context *nv50)
259 {
260 struct nouveau_pushbuf *push = nv50->base.pushbuf;
261 uint32_t pntc[8], mode;
262 struct nv50_program *fp = nv50->fragprog;
263 unsigned i, c;
264 unsigned m = (nv50->state.interpolant_ctrl >> 8) & 0xff;
265
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)
270 PUSH_DATA(push, 0);
271
272 nv50->state.point_sprite = false;
273 }
274 return;
275 } else {
276 nv50->state.point_sprite = true;
277 }
278
279 memset(pntc, 0, sizeof(pntc));
280
281 for (i = 0; i < fp->in_nr; i++) {
282 unsigned n = util_bitcount(fp->in[i].mask);
283
284 if (fp->in[i].sn != TGSI_SEMANTIC_GENERIC) {
285 m += n;
286 continue;
287 }
288 if (!(nv50->rast->pipe.sprite_coord_enable & (1 << fp->in[i].si))) {
289 m += n;
290 continue;
291 }
292
293 for (c = 0; c < 4; ++c) {
294 if (fp->in[i].mask & (1 << c)) {
295 pntc[m / 8] |= (c + 1) << ((m % 8) * 4);
296 ++m;
297 }
298 }
299 }
300
301 if (nv50->rast->pipe.sprite_coord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
302 mode = 0x00;
303 else
304 mode = 0x10;
305
306 BEGIN_NV04(push, NV50_3D(POINT_SPRITE_CTRL), 1);
307 PUSH_DATA (push, mode);
308
309 BEGIN_NV04(push, NV50_3D(POINT_COORD_REPLACE_MAP(0)), 8);
310 PUSH_DATAp(push, pntc, 8);
311 }
312
313 /* Validate state derived from shaders and the rasterizer cso. */
314 void
315 nv50_validate_derived_rs(struct nv50_context *nv50)
316 {
317 struct nouveau_pushbuf *push = nv50->base.pushbuf;
318 uint32_t color, psize;
319
320 nv50_sprite_coords_validate(nv50);
321
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);
326 }
327
328 if (nv50->dirty_3d & NV50_NEW_3D_FRAGPROG)
329 return;
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;
332
333 if (nv50->rast->pipe.clamp_vertex_color)
334 color |= NV50_3D_SEMANTIC_COLOR_CLMP_EN;
335
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);
340 }
341
342 if (nv50->rast->pipe.point_size_per_vertex)
343 psize |= NV50_3D_SEMANTIC_PTSZ_PTSZ_EN__MASK;
344
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);
349 }
350 }
351
352 static int
353 nv50_vec4_map(uint8_t *map, int mid, uint32_t lin[4],
354 struct nv50_varying *in, struct nv50_varying *out)
355 {
356 int c;
357 uint8_t mv = out->mask, mf = in->mask, oid = out->hw;
358
359 for (c = 0; c < 4; ++c) {
360 if (mf & 1) {
361 if (in->linear)
362 lin[mid / 32] |= 1 << (mid % 32);
363 if (mv & 1)
364 map[mid] = oid;
365 else
366 if (c == 3)
367 map[mid] |= 1;
368 ++mid;
369 }
370
371 oid += mv & 1;
372 mf >>= 1;
373 mv >>= 1;
374 }
375
376 return mid;
377 }
378
379 void
380 nv50_fp_linkage_validate(struct nv50_context *nv50)
381 {
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;
386 int i, n, c, m;
387 uint32_t primid = 0;
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);
394 uint32_t lin[4];
395 uint8_t map[64];
396 uint8_t so_map[64];
397
398 if (!(nv50->dirty_3d & (NV50_NEW_3D_VERTPROG |
399 NV50_NEW_3D_FRAGPROG |
400 NV50_NEW_3D_GMTYPROG))) {
401 uint8_t bfc, ffc;
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)
404 >> 8;
405 if (nv50->rast->pipe.light_twoside == ((ffc == bfc) ? 0 : 1))
406 return;
407 }
408
409 memset(lin, 0x00, sizeof(lin));
410
411 /* XXX: in buggy-endian mode, is the first element of map (u32)0x000000xx
412 * or is it the first byte ?
413 */
414 memset(map, nv50->gmtyprog ? 0x80 : 0x40, sizeof(map));
415
416 dummy.mask = 0xf; /* map all components of HPOS */
417 dummy.linear = 0;
418 m = nv50_vec4_map(map, 0, lin, &dummy, &vp->out[0]);
419
420 for (c = 0; c < clpd_nr; ++c)
421 map[m++] = vp->vp.clpd[c / 4] + (c % 4);
422
423 colors |= m << 8; /* adjust BFC0 id */
424
425 dummy.mask = 0x0;
426
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) {
430 n = vp->vp.bfc[i];
431 if (fp->vp.bfc[i] >= fp->in_nr)
432 continue;
433 m = nv50_vec4_map(map, m, lin, &fp->in[fp->vp.bfc[i]],
434 (n < vp->out_nr) ? &vp->out[n] : &dummy);
435 }
436 }
437 colors += m - 4; /* adjust FFC0 id */
438 interp |= m << 8; /* set map id where 'normal' FP inputs start */
439
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)
444 break;
445 switch (fp->in[i].sn) {
446 case TGSI_SEMANTIC_PRIMID:
447 primid = m;
448 break;
449 case TGSI_SEMANTIC_LAYER:
450 layerid = m;
451 break;
452 case TGSI_SEMANTIC_VIEWPORT_INDEX:
453 viewportid = m;
454 break;
455 }
456 m = nv50_vec4_map(map, m, lin,
457 &fp->in[i], (n < vp->out_nr) ? &vp->out[n] : &dummy);
458 }
459
460 if (vp->gp.has_layer && !layerid) {
461 layerid = m;
462 map[m++] = vp->gp.layerid;
463 }
464
465 if (vp->gp.has_viewport && !viewportid) {
466 viewportid = m;
467 map[m++] = vp->gp.viewportid;
468 }
469
470 if (nv50->rast->pipe.point_size_per_vertex) {
471 psiz = (m << 4) | 1;
472 map[m++] = vp->vp.psiz;
473 }
474
475 if (nv50->rast->pipe.clamp_vertex_color)
476 colors |= NV50_3D_SEMANTIC_COLOR_CLMP_EN;
477
478 if (unlikely(vp->so)) {
479 /* Slot i in STRMOUT_MAP specifies the offset where slot i in RESULT_MAP
480 * gets written.
481 *
482 * TODO:
483 * Inverting vp->so->map (output -> offset) would probably speed this up.
484 */
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)
488 continue;
489 for (c = 0; c < m; ++c)
490 if (map[c] == vp->so->map[i] && !so_map[c])
491 break;
492 if (c == m) {
493 c = m;
494 map[m++] = vp->so->map[i];
495 }
496 so_map[c] = 0x80 | i;
497 }
498 for (c = m; c & 3; ++c)
499 so_map[c] = 0;
500 }
501
502 n = (m + 3) / 4;
503 assert(m <= 64);
504
505 if (unlikely(nv50->gmtyprog)) {
506 BEGIN_NV04(push, NV50_3D(GP_RESULT_MAP_SIZE), 1);
507 PUSH_DATA (push, m);
508 BEGIN_NV04(push, NV50_3D(GP_RESULT_MAP(0)), n);
509 PUSH_DATAp(push, map, n);
510 } else {
511 BEGIN_NV04(push, NV50_3D(VP_GP_BUILTIN_ATTR_EN), 1);
512 PUSH_DATA (push, vp->vp.attrs[2] | fp->vp.attrs[2]);
513
514 BEGIN_NV04(push, NV50_3D(SEMANTIC_PRIM_ID), 1);
515 PUSH_DATA (push, primid);
516
517 assert(m > 0);
518 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP_SIZE), 1);
519 PUSH_DATA (push, m);
520 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP(0)), n);
521 PUSH_DATAp(push, map, n);
522 }
523
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);
530
531 BEGIN_NV04(push, NV50_3D(SEMANTIC_VIEWPORT), 1);
532 PUSH_DATA (push, viewportid);
533
534 BEGIN_NV04(push, NV50_3D(LAYER), 1);
535 PUSH_DATA (push, vp->gp.has_layer << 16);
536
537 BEGIN_NV04(push, NV50_3D(FP_INTERPOLANT_CTRL), 1);
538 PUSH_DATA (push, interp);
539
540 nv50->state.interpolant_ctrl = interp;
541
542 nv50->state.semantic_color = colors;
543 nv50->state.semantic_psize = psiz;
544
545 BEGIN_NV04(push, NV50_3D(NOPERSPECTIVE_BITMAP(0)), 4);
546 PUSH_DATAp(push, lin, 4);
547
548 BEGIN_NV04(push, NV50_3D(GP_ENABLE), 1);
549 PUSH_DATA (push, nv50->gmtyprog ? 1 : 0);
550
551 if (vp->so) {
552 BEGIN_NV04(push, NV50_3D(STRMOUT_MAP(0)), n);
553 PUSH_DATAp(push, so_map, n);
554 }
555 }
556
557 static int
558 nv50_vp_gp_mapping(uint8_t *map, int m,
559 struct nv50_program *vp, struct nv50_program *gp)
560 {
561 int i, j, c;
562
563 for (i = 0; i < gp->in_nr; ++i) {
564 uint8_t oid = 0, mv = 0, mg = gp->in[i].mask;
565
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;
570 oid = vp->out[j].hw;
571 break;
572 }
573 }
574
575 for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) {
576 if (mg & mv & 1)
577 map[m++] = oid;
578 else
579 if (mg & 1)
580 map[m++] = (c == 3) ? 0x41 : 0x40;
581 oid += mv & 1;
582 }
583 }
584 if (!m)
585 map[m++] = 0;
586 return m;
587 }
588
589 void
590 nv50_gp_linkage_validate(struct nv50_context *nv50)
591 {
592 struct nouveau_pushbuf *push = nv50->base.pushbuf;
593 struct nv50_program *vp = nv50->vertprog;
594 struct nv50_program *gp = nv50->gmtyprog;
595 int m = 0;
596 int n;
597 uint8_t map[64];
598
599 if (!gp)
600 return;
601 memset(map, 0, sizeof(map));
602
603 m = nv50_vp_gp_mapping(map, m, vp, gp);
604
605 n = (m + 3) / 4;
606
607 BEGIN_NV04(push, NV50_3D(VP_GP_BUILTIN_ATTR_EN), 1);
608 PUSH_DATA (push, vp->vp.attrs[2] | gp->vp.attrs[2]);
609
610 assert(m > 0);
611 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP_SIZE), 1);
612 PUSH_DATA (push, m);
613 BEGIN_NV04(push, NV50_3D(VP_RESULT_MAP(0)), n);
614 PUSH_DATAp(push, map, n);
615 }
616
617 void
618 nv50_stream_output_validate(struct nv50_context *nv50)
619 {
620 struct nouveau_pushbuf *push = nv50->base.pushbuf;
621 struct nv50_stream_output_state *so;
622 uint32_t ctrl;
623 unsigned i;
624 unsigned prims = ~0;
625
626 so = nv50->gmtyprog ? nv50->gmtyprog->so : nv50->vertprog->so;
627
628 BEGIN_NV04(push, NV50_3D(STRMOUT_ENABLE), 1);
629 PUSH_DATA (push, 0);
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);
633 PUSH_DATA (push, 0);
634 }
635 BEGIN_NV04(push, NV50_3D(STRMOUT_PARAMS_LATCH), 1);
636 PUSH_DATA (push, 1);
637 return;
638 }
639
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);
643 PUSH_DATA (push, 0);
644 }
645
646 ctrl = so->ctrl;
647 if (nv50->screen->base.class_3d >= NVA0_3D_CLASS)
648 ctrl |= NVA0_3D_STRMOUT_BUFFERS_CTRL_LIMIT_MODE_OFFSET;
649
650 BEGIN_NV04(push, NV50_3D(STRMOUT_BUFFERS_CTRL), 1);
651 PUSH_DATA (push, ctrl);
652
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);
656
657 const unsigned n = nv50->screen->base.class_3d >= NVA0_3D_CLASS ? 4 : 3;
658
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]);
665 if (n == 4) {
666 PUSH_DATA(push, targ->pipe.buffer_size);
667 if (!targ->clean) {
668 assert(targ->pq);
669 nv50_hw_query_pushbuf_submit(push, NVA0_3D_STRMOUT_OFFSET(i),
670 nv50_query(targ->pq), 0x4);
671 } else {
672 BEGIN_NV04(push, NVA0_3D(STRMOUT_OFFSET(i)), 1);
673 PUSH_DATA(push, 0);
674 targ->clean = false;
675 }
676 } else {
677 const unsigned limit = targ->pipe.buffer_size /
678 (so->stride[i] * nv50->state.prim_size);
679 prims = MIN2(prims, limit);
680 }
681 targ->stride = so->stride[i];
682 BCTX_REFN(nv50->bufctx_3d, 3D_SO, buf, WR);
683 }
684 if (prims != ~0) {
685 BEGIN_NV04(push, NV50_3D(STRMOUT_PRIMITIVE_LIMIT), 1);
686 PUSH_DATA (push, prims);
687 }
688 BEGIN_NV04(push, NV50_3D(STRMOUT_PARAMS_LATCH), 1);
689 PUSH_DATA (push, 1);
690 BEGIN_NV04(push, NV50_3D(STRMOUT_ENABLE), 1);
691 PUSH_DATA (push, 1);
692 }