d73f7c7f213f80d8f5e2b99bd0df1ce715d24a3a
[mesa.git] / src / gallium / drivers / 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 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
21 * 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_context.h"
30
31 void
32 nv50_constbufs_validate(struct nv50_context *nv50)
33 {
34 struct nouveau_channel *chan = nv50->screen->base.channel;
35 unsigned s;
36
37 for (s = 0; s < 3; ++s) {
38 struct nv04_resource *res;
39 int i;
40 unsigned p, b;
41
42 if (s == PIPE_SHADER_FRAGMENT)
43 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_FRAGMENT;
44 else
45 if (s == PIPE_SHADER_GEOMETRY)
46 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_GEOMETRY;
47 else
48 p = NV50_3D_SET_PROGRAM_CB_PROGRAM_VERTEX;
49
50 while (nv50->constbuf_dirty[s]) {
51 struct nouveau_bo *bo;
52 unsigned start = 0;
53 unsigned words = 0;
54
55 i = ffs(nv50->constbuf_dirty[s]) - 1;
56 nv50->constbuf_dirty[s] &= ~(1 << i);
57
58 res = nv04_resource(nv50->constbuf[s][i]);
59 if (!res) {
60 if (i != 0) {
61 BEGIN_RING(chan, RING_3D(SET_PROGRAM_CB), 1);
62 OUT_RING (chan, (i << 8) | p | 0);
63 }
64 continue;
65 }
66
67 if (i == 0) {
68 b = NV50_CB_PVP + s;
69
70 /* always upload GL uniforms through CB DATA */
71 bo = nv50->screen->uniforms;
72 words = res->base.width0 / 4;
73 } else {
74 b = s * 16 + i;
75
76 assert(0);
77
78 if (!nouveau_resource_mapped_by_gpu(&res->base)) {
79 nouveau_buffer_migrate(&nv50->base, res, NOUVEAU_BO_VRAM);
80
81 BEGIN_RING(chan, RING_3D(CODE_CB_FLUSH), 1);
82 OUT_RING (chan, 0);
83 }
84 MARK_RING (chan, 6, 2);
85 BEGIN_RING(chan, RING_3D(CB_DEF_ADDRESS_HIGH), 3);
86 OUT_RESRCh(chan, res, 0, NOUVEAU_BO_RD);
87 OUT_RESRCl(chan, res, 0, NOUVEAU_BO_RD);
88 OUT_RING (chan, (b << 16) | (res->base.width0 & 0xffff));
89 BEGIN_RING(chan, RING_3D(SET_PROGRAM_CB), 1);
90 OUT_RING (chan, (b << 12) | (i << 8) | p | 1);
91
92 bo = res->bo;
93
94 nv50_bufctx_add_resident(nv50, NV50_BUFCTX_CONSTANT, res,
95 res->domain | NOUVEAU_BO_RD);
96 }
97
98 if (words) {
99 MARK_RING(chan, 8, 1);
100
101 nouveau_bo_validate(chan, bo, res->domain | NOUVEAU_BO_WR);
102 }
103
104 while (words) {
105 unsigned nr = AVAIL_RING(chan);
106
107 if (nr < 16) {
108 FIRE_RING(chan);
109 nouveau_bo_validate(chan, bo, res->domain | NOUVEAU_BO_WR);
110 continue;
111 }
112 nr = MIN2(MIN2(nr - 3, words), NV04_PFIFO_MAX_PACKET_LEN);
113
114 BEGIN_RING(chan, RING_3D(CB_ADDR), 1);
115 OUT_RING (chan, (start << 8) | b);
116 BEGIN_RING_NI(chan, RING_3D(CB_DATA(0)), nr);
117 OUT_RINGp (chan, &res->data[start * 4], nr);
118
119 start += nr;
120 words -= nr;
121 }
122 }
123 }
124 }
125
126 static boolean
127 nv50_program_validate(struct nv50_context *nv50, struct nv50_program *prog)
128 {
129 struct nouveau_resource *heap;
130 int ret;
131 unsigned size;
132
133 if (!prog->translated) {
134 prog->translated = nv50_program_translate(prog);
135 if (!prog->translated)
136 return FALSE;
137 } else
138 if (prog->res)
139 return TRUE;
140
141 if (prog->type == PIPE_SHADER_FRAGMENT) heap = nv50->screen->fp_code_heap;
142 else
143 if (prog->type == PIPE_SHADER_GEOMETRY) heap = nv50->screen->gp_code_heap;
144 else
145 heap = nv50->screen->vp_code_heap;
146
147 size = align(prog->code_size, 0x100);
148
149 ret = nouveau_resource_alloc(heap, size, prog, &prog->res);
150 if (ret) {
151 NOUVEAU_ERR("out of code space for shader type %i\n", prog->type);
152 return FALSE;
153 }
154 prog->code_base = prog->res->start;
155
156 nv50_relocate_program(prog, prog->code_base, 0);
157
158 nv50_sifc_linear_u8(&nv50->base, nv50->screen->code,
159 (prog->type << NV50_CODE_BO_SIZE_LOG2) + prog->code_base,
160 NOUVEAU_BO_VRAM, prog->code_size, prog->code);
161
162 BEGIN_RING(nv50->screen->base.channel, RING_3D(CODE_CB_FLUSH), 1);
163 OUT_RING (nv50->screen->base.channel, 0);
164
165 return TRUE;
166 }
167
168 void
169 nv50_vertprog_validate(struct nv50_context *nv50)
170 {
171 struct nouveau_channel *chan = nv50->screen->base.channel;
172 struct nv50_program *vp = nv50->vertprog;
173
174 if (nv50->clip.nr > vp->vp.clpd_nr) {
175 if (vp->translated)
176 nv50_program_destroy(nv50, vp);
177 vp->vp.clpd_nr = nv50->clip.nr;
178 }
179
180 if (!nv50_program_validate(nv50, vp))
181 return;
182
183 BEGIN_RING(chan, RING_3D(VP_ATTR_EN(0)), 2);
184 OUT_RING (chan, vp->vp.attrs[0]);
185 OUT_RING (chan, vp->vp.attrs[1]);
186 BEGIN_RING(chan, RING_3D(VP_REG_ALLOC_RESULT), 1);
187 OUT_RING (chan, vp->max_out);
188 BEGIN_RING(chan, RING_3D(VP_REG_ALLOC_TEMP), 1);
189 OUT_RING (chan, vp->max_gpr);
190 BEGIN_RING(chan, RING_3D(VP_START_ID), 1);
191 OUT_RING (chan, vp->code_base);
192 }
193
194 void
195 nv50_fragprog_validate(struct nv50_context *nv50)
196 {
197 struct nouveau_channel *chan = nv50->screen->base.channel;
198 struct nv50_program *fp = nv50->fragprog;
199
200 if (!nv50_program_validate(nv50, fp))
201 return;
202
203 BEGIN_RING(chan, RING_3D(FP_REG_ALLOC_TEMP), 1);
204 OUT_RING (chan, fp->max_gpr);
205 BEGIN_RING(chan, RING_3D(FP_RESULT_COUNT), 1);
206 OUT_RING (chan, fp->max_out);
207 BEGIN_RING(chan, RING_3D(FP_CONTROL), 1);
208 OUT_RING (chan, fp->fp.flags[0]);
209 BEGIN_RING(chan, RING_3D(FP_CTRL_UNK196C), 1);
210 OUT_RING (chan, fp->fp.flags[1]);
211 BEGIN_RING(chan, RING_3D(FP_START_ID), 1);
212 OUT_RING (chan, fp->code_base);
213 }
214
215 void
216 nv50_gmtyprog_validate(struct nv50_context *nv50)
217 {
218 struct nouveau_channel *chan = nv50->screen->base.channel;
219 struct nv50_program *gp = nv50->gmtyprog;
220
221 if (!gp) /* GP_ENABLE is updated in linkage validation */
222 return;
223 if (!nv50_program_validate(nv50, gp))
224 return;
225
226 BEGIN_RING(chan, RING_3D(GP_REG_ALLOC_TEMP), 1);
227 OUT_RING (chan, gp->max_gpr);
228 BEGIN_RING(chan, RING_3D(GP_REG_ALLOC_RESULT), 1);
229 OUT_RING (chan, gp->max_out);
230 BEGIN_RING(chan, RING_3D(GP_OUTPUT_PRIMITIVE_TYPE), 1);
231 OUT_RING (chan, gp->gp.prim_type);
232 BEGIN_RING(chan, RING_3D(GP_VERTEX_OUTPUT_COUNT), 1);
233 OUT_RING (chan, gp->gp.vert_count);
234 BEGIN_RING(chan, RING_3D(GP_START_ID), 1);
235 OUT_RING (chan, gp->code_base);
236 }
237
238 static void
239 nv50_sprite_coords_validate(struct nv50_context *nv50)
240 {
241 struct nouveau_channel *chan = nv50->screen->base.channel;
242 uint32_t pntc[8], mode;
243 struct nv50_program *fp = nv50->fragprog;
244 unsigned i, c;
245 unsigned m = (nv50->state.interpolant_ctrl >> 8) & 0xff;
246
247 if (!nv50->rast->pipe.point_quad_rasterization) {
248 if (nv50->state.point_sprite) {
249 BEGIN_RING(chan, RING_3D(POINT_COORD_REPLACE_MAP(0)), 8);
250 for (i = 0; i < 8; ++i)
251 OUT_RING(chan, 0);
252
253 nv50->state.point_sprite = FALSE;
254 }
255 return;
256 } else {
257 nv50->state.point_sprite = TRUE;
258 }
259
260 memset(pntc, 0, sizeof(pntc));
261
262 for (i = 0; i < fp->in_nr; i++) {
263 unsigned n = util_bitcount(fp->in[i].mask);
264
265 if (fp->in[i].sn != TGSI_SEMANTIC_GENERIC) {
266 m += n;
267 continue;
268 }
269 if (!(nv50->rast->pipe.sprite_coord_enable & (1 << fp->in[i].si))) {
270 m += n;
271 continue;
272 }
273
274 for (c = 0; c < 4; ++c) {
275 if (fp->in[i].mask & (1 << c)) {
276 pntc[m / 8] |= (c + 1) << ((m % 8) * 4);
277 ++m;
278 }
279 }
280 }
281
282 if (nv50->rast->pipe.sprite_coord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
283 mode = 0x00;
284 else
285 mode = 0x10;
286
287 BEGIN_RING(chan, RING_3D(POINT_SPRITE_CTRL), 1);
288 OUT_RING (chan, mode);
289
290 BEGIN_RING(chan, RING_3D(POINT_COORD_REPLACE_MAP(0)), 8);
291 OUT_RINGp (chan, pntc, 8);
292 }
293
294 /* Validate state derived from shaders and the rasterizer cso. */
295 void
296 nv50_validate_derived_rs(struct nv50_context *nv50)
297 {
298 struct nouveau_channel *chan = nv50->screen->base.channel;
299 uint32_t color, psize;
300
301 nv50_sprite_coords_validate(nv50);
302
303 if (nv50->dirty & NV50_NEW_FRAGPROG)
304 return;
305 psize = nv50->state.semantic_psize & ~NV50_3D_MAP_SEMANTIC_3_PTSZ_EN__MASK;
306 color = nv50->state.semantic_color & ~NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
307
308 if (nv50->rast->pipe.clamp_vertex_color)
309 color |= NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
310
311 if (color != nv50->state.semantic_color) {
312 nv50->state.semantic_color = color;
313 BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_0), 1);
314 OUT_RING (chan, color);
315 }
316
317 if (nv50->rast->pipe.point_size_per_vertex)
318 psize |= NV50_3D_MAP_SEMANTIC_3_PTSZ_EN__MASK;
319
320 if (psize != nv50->state.semantic_psize) {
321 nv50->state.semantic_psize = psize;
322 BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_3), 1);
323 OUT_RING (chan, psize);
324 }
325 }
326
327 static int
328 nv50_vec4_map(uint8_t *map, int mid, uint32_t lin[4],
329 struct nv50_varying *in, struct nv50_varying *out)
330 {
331 int c;
332 uint8_t mv = out->mask, mf = in->mask, oid = out->hw;
333
334 for (c = 0; c < 4; ++c) {
335 if (mf & 1) {
336 if (in->linear)
337 lin[mid / 32] |= 1 << (mid % 32);
338 if (mv & 1)
339 map[mid] = oid;
340 else
341 if (c == 3)
342 map[mid] |= 1;
343 ++mid;
344 }
345
346 oid += mv & 1;
347 mf >>= 1;
348 mv >>= 1;
349 }
350
351 return mid;
352 }
353
354 void
355 nv50_fp_linkage_validate(struct nv50_context *nv50)
356 {
357 struct nouveau_channel *chan = nv50->screen->base.channel;
358 struct nv50_program *vp = nv50->gmtyprog ? nv50->gmtyprog : nv50->vertprog;
359 struct nv50_program *fp = nv50->fragprog;
360 struct nv50_varying dummy;
361 int i, n, c, m;
362 uint32_t primid = 0;
363 uint32_t psiz = 0x000;
364 uint32_t interp = fp->fp.interp;
365 uint32_t colors = fp->fp.colors;
366 uint32_t lin[4];
367 uint8_t map[64];
368
369 memset(lin, 0x00, sizeof(lin));
370
371 /* XXX: in buggy-endian mode, is the first element of map (u32)0x000000xx
372 * or is it the first byte ?
373 */
374 memset(map, nv50->gmtyprog ? 0x80 : 0x40, sizeof(map));
375
376 dummy.mask = 0xf; /* map all components of HPOS */
377 dummy.linear = 0;
378 m = nv50_vec4_map(map, 0, lin, &dummy, &vp->out[0]);
379
380 for (c = 0; c < vp->vp.clpd_nr; ++c)
381 map[m++] = vp->vp.clpd + c;
382
383 colors |= m << 8; /* adjust BFC0 id */
384
385 /* if light_twoside is active, FFC0_ID == BFC0_ID is invalid */
386 if (nv50->rast->pipe.light_twoside) {
387 for (i = 0; i < 2; ++i)
388 m = nv50_vec4_map(map, m, lin,
389 &fp->in[fp->vp.bfc[i]], &vp->out[vp->vp.bfc[i]]);
390 }
391 colors += m - 4; /* adjust FFC0 id */
392 interp |= m << 8; /* set map id where 'normal' FP inputs start */
393
394 dummy.mask = 0x0;
395 for (i = 0; i < fp->in_nr; ++i) {
396 for (n = 0; n < vp->out_nr; ++n)
397 if (vp->out[n].sn == fp->in[i].sn &&
398 vp->out[n].si == fp->in[i].si)
399 break;
400 m = nv50_vec4_map(map, m, lin,
401 &fp->in[i], (n < vp->out_nr) ? &vp->out[n] : &dummy);
402 }
403
404 /* PrimitiveID either is replaced by the system value, or
405 * written by the geometry shader into an output register
406 */
407 if (fp->gp.primid < 0x40) {
408 primid = m;
409 map[m++] = vp->gp.primid;
410 }
411
412 if (nv50->rast->pipe.point_size_per_vertex) {
413 psiz = (m << 4) | 1;
414 map[m++] = vp->vp.psiz;
415 }
416
417 if (nv50->rast->pipe.clamp_vertex_color)
418 colors |= NV50_3D_MAP_SEMANTIC_0_CLMP_EN;
419
420 n = (m + 3) / 4;
421 assert(m <= 64);
422
423 if (unlikely(nv50->gmtyprog)) {
424 BEGIN_RING(chan, RING_3D(GP_RESULT_MAP_SIZE), 1);
425 OUT_RING (chan, m);
426 BEGIN_RING(chan, RING_3D(GP_RESULT_MAP(0)), n);
427 OUT_RINGp (chan, map, n);
428 } else {
429 BEGIN_RING(chan, RING_3D(VP_GP_BUILTIN_ATTR_EN), 1);
430 OUT_RING (chan, vp->vp.attrs[2]);
431
432 BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_4), 1);
433 OUT_RING (chan, primid);
434
435 BEGIN_RING(chan, RING_3D(VP_RESULT_MAP_SIZE), 1);
436 OUT_RING (chan, m);
437 BEGIN_RING(chan, RING_3D(VP_RESULT_MAP(0)), n);
438 OUT_RINGp (chan, map, n);
439 }
440
441 BEGIN_RING(chan, RING_3D(MAP_SEMANTIC_0), 4);
442 OUT_RING (chan, colors);
443 OUT_RING (chan, (vp->vp.clpd_nr << 8) | 4);
444 OUT_RING (chan, 0);
445 OUT_RING (chan, psiz);
446
447 BEGIN_RING(chan, RING_3D(FP_INTERPOLANT_CTRL), 1);
448 OUT_RING (chan, interp);
449
450 nv50->state.interpolant_ctrl = interp;
451
452 nv50->state.semantic_color = colors;
453 nv50->state.semantic_psize = psiz;
454
455 BEGIN_RING(chan, RING_3D(NOPERSPECTIVE_BITMAP(0)), 4);
456 OUT_RINGp (chan, lin, 4);
457
458 BEGIN_RING(chan, RING_3D(GP_ENABLE), 1);
459 OUT_RING (chan, nv50->gmtyprog ? 1 : 0);
460 }
461
462 static int
463 nv50_vp_gp_mapping(uint8_t *map, int m,
464 struct nv50_program *vp, struct nv50_program *gp)
465 {
466 int i, j, c;
467
468 for (i = 0; i < gp->in_nr; ++i) {
469 uint8_t oid = 0, mv = 0, mg = gp->in[i].mask;
470
471 for (j = 0; j < vp->out_nr; ++j) {
472 if (vp->out[j].sn == gp->in[i].sn &&
473 vp->out[j].si == gp->in[i].si) {
474 mv = vp->out[j].mask;
475 oid = vp->out[j].hw;
476 break;
477 }
478 }
479
480 for (c = 0; c < 4; ++c, mv >>= 1, mg >>= 1) {
481 if (mg & mv & 1)
482 map[m++] = oid;
483 else
484 if (mg & 1)
485 map[m++] = (c == 3) ? 0x41 : 0x40;
486 oid += mv & 1;
487 }
488 }
489 return m;
490 }
491
492 void
493 nv50_gp_linkage_validate(struct nv50_context *nv50)
494 {
495 struct nouveau_channel *chan = nv50->screen->base.channel;
496 struct nv50_program *vp = nv50->vertprog;
497 struct nv50_program *gp = nv50->gmtyprog;
498 int m = 0;
499 int n;
500 uint8_t map[64];
501
502 if (!gp)
503 return;
504 memset(map, 0, sizeof(map));
505
506 m = nv50_vp_gp_mapping(map, m, vp, gp);
507
508 n = (m + 3) / 4;
509
510 BEGIN_RING(chan, RING_3D(VP_GP_BUILTIN_ATTR_EN), 1);
511 OUT_RING (chan, vp->vp.attrs[2] | gp->vp.attrs[2]);
512
513 BEGIN_RING(chan, RING_3D(VP_RESULT_MAP_SIZE), 1);
514 OUT_RING (chan, m);
515 BEGIN_RING(chan, RING_3D(VP_RESULT_MAP(0)), n);
516 OUT_RINGp (chan, map, n);
517 }