Merge remote branch 'origin/master' into pipe-video
[mesa.git] / src / gallium / drivers / nvc0 / nvc0_state_validate.c
1
2 #include "util/u_math.h"
3
4 #include "nvc0_context.h"
5
6 static void
7 nvc0_validate_zcull(struct nvc0_context *nvc0)
8 {
9 struct nouveau_channel *chan = nvc0->screen->base.channel;
10 struct pipe_framebuffer_state *fb = &nvc0->framebuffer;
11 struct nvc0_surface *sf = nvc0_surface(fb->zsbuf);
12 struct nvc0_miptree *mt = nvc0_miptree(sf->base.texture);
13 struct nouveau_bo *bo = mt->base.bo;
14 uint32_t size;
15 uint32_t offset = align(mt->total_size, 1 << 17);
16 unsigned width, height;
17
18 assert(mt->base.base.depth0 == 1 && mt->base.base.array_size < 2);
19
20 size = mt->total_size * 2;
21
22 height = align(fb->height, 32);
23 width = fb->width % 224;
24 if (width)
25 width = fb->width + (224 - width);
26 else
27 width = fb->width;
28
29 MARK_RING (chan, 23, 4);
30 BEGIN_RING(chan, RING_3D_(0x1590), 1); /* ZCULL_REGION_INDEX (bits 0x3f) */
31 OUT_RING (chan, 0);
32 BEGIN_RING(chan, RING_3D_(0x07e8), 2); /* ZCULL_ADDRESS_A_HIGH */
33 OUT_RELOCh(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
34 OUT_RELOCl(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
35 offset += 1 << 17;
36 BEGIN_RING(chan, RING_3D_(0x07f0), 2); /* ZCULL_ADDRESS_B_HIGH */
37 OUT_RELOCh(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
38 OUT_RELOCl(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
39 BEGIN_RING(chan, RING_3D_(0x07e0), 2);
40 OUT_RING (chan, size);
41 OUT_RING (chan, size >> 16);
42 BEGIN_RING(chan, RING_3D_(0x15c8), 1); /* bits 0x3 */
43 OUT_RING (chan, 2);
44 BEGIN_RING(chan, RING_3D_(0x07c0), 4); /* ZCULL dimensions */
45 OUT_RING (chan, width);
46 OUT_RING (chan, height);
47 OUT_RING (chan, 1);
48 OUT_RING (chan, 0);
49 BEGIN_RING(chan, RING_3D_(0x15fc), 2);
50 OUT_RING (chan, 0); /* bits 0xffff */
51 OUT_RING (chan, 0); /* bits 0xffff */
52 BEGIN_RING(chan, RING_3D_(0x1958), 1);
53 OUT_RING (chan, 0); /* bits ~0 */
54 }
55
56 static void
57 nvc0_validate_fb(struct nvc0_context *nvc0)
58 {
59 struct nouveau_channel *chan = nvc0->screen->base.channel;
60 struct pipe_framebuffer_state *fb = &nvc0->framebuffer;
61 unsigned i;
62 boolean serialize = FALSE;
63
64 nvc0_bufctx_reset(nvc0, NVC0_BUFCTX_FRAME);
65
66 BEGIN_RING(chan, RING_3D(RT_CONTROL), 1);
67 OUT_RING (chan, (076543210 << 4) | fb->nr_cbufs);
68 BEGIN_RING(chan, RING_3D(SCREEN_SCISSOR_HORIZ), 2);
69 OUT_RING (chan, fb->width << 16);
70 OUT_RING (chan, fb->height << 16);
71
72 MARK_RING(chan, 9 * fb->nr_cbufs, 2 * fb->nr_cbufs);
73
74 for (i = 0; i < fb->nr_cbufs; ++i) {
75 struct nvc0_miptree *mt = nvc0_miptree(fb->cbufs[i]->texture);
76 struct nvc0_surface *sf = nvc0_surface(fb->cbufs[i]);
77 struct nouveau_bo *bo = mt->base.bo;
78 uint32_t offset = sf->offset;
79
80 BEGIN_RING(chan, RING_3D(RT_ADDRESS_HIGH(i)), 9);
81 OUT_RELOCh(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
82 OUT_RELOCl(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
83 OUT_RING (chan, sf->width);
84 OUT_RING (chan, sf->height);
85 OUT_RING (chan, nvc0_format_table[sf->base.format].rt);
86 OUT_RING (chan, (mt->layout_3d << 16) |
87 mt->level[sf->base.u.tex.level].tile_mode);
88 OUT_RING (chan, sf->base.u.tex.first_layer + sf->depth);
89 OUT_RING (chan, mt->layer_stride >> 2);
90 OUT_RING (chan, sf->base.u.tex.first_layer);
91
92 if (mt->base.status & NOUVEAU_BUFFER_STATUS_GPU_READING)
93 serialize = TRUE;
94 mt->base.status |= NOUVEAU_BUFFER_STATUS_GPU_WRITING;
95 mt->base.status &= ~NOUVEAU_BUFFER_STATUS_GPU_READING;
96
97 /* only register for writing, otherwise we'd always serialize here */
98 nvc0_bufctx_add_resident(nvc0, NVC0_BUFCTX_FRAME, &mt->base,
99 NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
100 }
101
102 if (fb->zsbuf) {
103 struct nvc0_miptree *mt = nvc0_miptree(fb->zsbuf->texture);
104 struct nvc0_surface *sf = nvc0_surface(fb->zsbuf);
105 struct nouveau_bo *bo = mt->base.bo;
106 int unk = mt->base.base.target == PIPE_TEXTURE_2D;
107 uint32_t offset = sf->offset;
108
109 MARK_RING (chan, 12, 2);
110 BEGIN_RING(chan, RING_3D(ZETA_ADDRESS_HIGH), 5);
111 OUT_RELOCh(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
112 OUT_RELOCl(chan, bo, offset, NOUVEAU_BO_VRAM | NOUVEAU_BO_RDWR);
113 OUT_RING (chan, nvc0_format_table[fb->zsbuf->format].rt);
114 OUT_RING (chan, mt->level[sf->base.u.tex.level].tile_mode);
115 OUT_RING (chan, mt->layer_stride >> 2);
116 BEGIN_RING(chan, RING_3D(ZETA_ENABLE), 1);
117 OUT_RING (chan, 1);
118 BEGIN_RING(chan, RING_3D(ZETA_HORIZ), 3);
119 OUT_RING (chan, sf->width);
120 OUT_RING (chan, sf->height);
121 OUT_RING (chan, (unk << 16) |
122 (sf->base.u.tex.first_layer + sf->depth));
123 BEGIN_RING(chan, RING_3D(ZETA_BASE_LAYER), 1);
124 OUT_RING (chan, sf->base.u.tex.first_layer);
125
126 if (mt->base.status & NOUVEAU_BUFFER_STATUS_GPU_READING)
127 serialize = TRUE;
128 mt->base.status |= NOUVEAU_BUFFER_STATUS_GPU_WRITING;
129 mt->base.status &= ~NOUVEAU_BUFFER_STATUS_GPU_READING;
130
131 nvc0_bufctx_add_resident(nvc0, NVC0_BUFCTX_FRAME, &mt->base,
132 NOUVEAU_BO_VRAM | NOUVEAU_BO_WR);
133 } else {
134 BEGIN_RING(chan, RING_3D(ZETA_ENABLE), 1);
135 OUT_RING (chan, 0);
136 }
137
138 if (serialize) {
139 BEGIN_RING(chan, RING_3D(SERIALIZE), 1);
140 OUT_RING (chan, 0);
141 }
142 }
143
144 static void
145 nvc0_validate_blend_colour(struct nvc0_context *nvc0)
146 {
147 struct nouveau_channel *chan = nvc0->screen->base.channel;
148
149 BEGIN_RING(chan, RING_3D(BLEND_COLOR(0)), 4);
150 OUT_RINGf (chan, nvc0->blend_colour.color[0]);
151 OUT_RINGf (chan, nvc0->blend_colour.color[1]);
152 OUT_RINGf (chan, nvc0->blend_colour.color[2]);
153 OUT_RINGf (chan, nvc0->blend_colour.color[3]);
154 }
155
156 static void
157 nvc0_validate_stencil_ref(struct nvc0_context *nvc0)
158 {
159 struct nouveau_channel *chan = nvc0->screen->base.channel;
160 const ubyte *ref = &nvc0->stencil_ref.ref_value[0];
161
162 IMMED_RING(chan, RING_3D(STENCIL_FRONT_FUNC_REF), ref[0]);
163 IMMED_RING(chan, RING_3D(STENCIL_BACK_FUNC_REF), ref[1]);
164 }
165
166 static void
167 nvc0_validate_stipple(struct nvc0_context *nvc0)
168 {
169 struct nouveau_channel *chan = nvc0->screen->base.channel;
170 unsigned i;
171
172 BEGIN_RING(chan, RING_3D(POLYGON_STIPPLE_PATTERN(0)), 32);
173 for (i = 0; i < 32; ++i)
174 OUT_RING(chan, util_bswap32(nvc0->stipple.stipple[i]));
175 }
176
177 static void
178 nvc0_validate_scissor(struct nvc0_context *nvc0)
179 {
180 struct nouveau_channel *chan = nvc0->screen->base.channel;
181 struct pipe_scissor_state *s = &nvc0->scissor;
182
183 if (!(nvc0->dirty & NVC0_NEW_SCISSOR) &&
184 nvc0->rast->pipe.scissor == nvc0->state.scissor)
185 return;
186 nvc0->state.scissor = nvc0->rast->pipe.scissor;
187
188 BEGIN_RING(chan, RING_3D(SCISSOR_HORIZ(0)), 2);
189 if (nvc0->rast->pipe.scissor) {
190 OUT_RING(chan, (s->maxx << 16) | s->minx);
191 OUT_RING(chan, (s->maxy << 16) | s->miny);
192 } else {
193 OUT_RING(chan, (0xffff << 16) | 0);
194 OUT_RING(chan, (0xffff << 16) | 0);
195 }
196 }
197
198 static void
199 nvc0_validate_viewport(struct nvc0_context *nvc0)
200 {
201 struct nouveau_channel *chan = nvc0->screen->base.channel;
202 struct pipe_viewport_state *vp = &nvc0->viewport;
203 int x, y, w, h;
204 float zmin, zmax;
205
206 BEGIN_RING(chan, RING_3D(VIEWPORT_TRANSLATE_X(0)), 3);
207 OUT_RINGf (chan, vp->translate[0]);
208 OUT_RINGf (chan, vp->translate[1]);
209 OUT_RINGf (chan, vp->translate[2]);
210 BEGIN_RING(chan, RING_3D(VIEWPORT_SCALE_X(0)), 3);
211 OUT_RINGf (chan, vp->scale[0]);
212 OUT_RINGf (chan, vp->scale[1]);
213 OUT_RINGf (chan, vp->scale[2]);
214
215 /* now set the viewport rectangle to viewport dimensions for clipping */
216
217 x = util_iround(MAX2(0.0f, vp->translate[0] - fabsf(vp->scale[0])));
218 y = util_iround(MAX2(0.0f, vp->translate[1] - fabsf(vp->scale[1])));
219 w = util_iround(vp->translate[0] + fabsf(vp->scale[0])) - x;
220 h = util_iround(vp->translate[1] + fabsf(vp->scale[1])) - y;
221
222 zmin = vp->translate[2] - fabsf(vp->scale[2]);
223 zmax = vp->translate[2] + fabsf(vp->scale[2]);
224
225 BEGIN_RING(chan, RING_3D(VIEWPORT_HORIZ(0)), 2);
226 OUT_RING (chan, (w << 16) | x);
227 OUT_RING (chan, (h << 16) | y);
228 BEGIN_RING(chan, RING_3D(DEPTH_RANGE_NEAR(0)), 2);
229 OUT_RINGf (chan, zmin);
230 OUT_RINGf (chan, zmax);
231 }
232
233 static void
234 nvc0_validate_clip(struct nvc0_context *nvc0)
235 {
236 struct nouveau_channel *chan = nvc0->screen->base.channel;
237 uint32_t clip;
238
239 if (nvc0->clip.depth_clamp) {
240 clip =
241 NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1 |
242 NVC0_3D_VIEW_VOLUME_CLIP_CTRL_DEPTH_CLAMP_NEAR |
243 NVC0_3D_VIEW_VOLUME_CLIP_CTRL_DEPTH_CLAMP_FAR |
244 NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK12_UNK2;
245 } else {
246 clip = NVC0_3D_VIEW_VOLUME_CLIP_CTRL_UNK1_UNK1;
247 }
248
249 BEGIN_RING(chan, RING_3D(VIEW_VOLUME_CLIP_CTRL), 1);
250 OUT_RING (chan, clip);
251
252 if (nvc0->clip.nr) {
253 struct nouveau_bo *bo = nvc0->screen->uniforms;
254
255 MARK_RING (chan, 6 + nvc0->clip.nr * 4, 2);
256 BEGIN_RING(chan, RING_3D(CB_SIZE), 3);
257 OUT_RING (chan, 256);
258 OUT_RELOCh(chan, bo, 5 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
259 OUT_RELOCl(chan, bo, 5 << 16, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
260 BEGIN_RING_1I(chan, RING_3D(CB_POS), nvc0->clip.nr * 4 + 1);
261 OUT_RING (chan, 0);
262 OUT_RINGp (chan, &nvc0->clip.ucp[0][0], nvc0->clip.nr * 4);
263
264 BEGIN_RING(chan, RING_3D(VP_CLIP_DISTANCE_ENABLE), 1);
265 OUT_RING (chan, (1 << nvc0->clip.nr) - 1);
266 } else {
267 IMMED_RING(chan, RING_3D(VP_CLIP_DISTANCE_ENABLE), 0);
268 }
269 }
270
271 static void
272 nvc0_validate_blend(struct nvc0_context *nvc0)
273 {
274 struct nouveau_channel *chan = nvc0->screen->base.channel;
275
276 WAIT_RING(chan, nvc0->blend->size);
277 OUT_RINGp(chan, nvc0->blend->state, nvc0->blend->size);
278 }
279
280 static void
281 nvc0_validate_zsa(struct nvc0_context *nvc0)
282 {
283 struct nouveau_channel *chan = nvc0->screen->base.channel;
284
285 WAIT_RING(chan, nvc0->zsa->size);
286 OUT_RINGp(chan, nvc0->zsa->state, nvc0->zsa->size);
287 }
288
289 static void
290 nvc0_validate_rasterizer(struct nvc0_context *nvc0)
291 {
292 struct nouveau_channel *chan = nvc0->screen->base.channel;
293
294 WAIT_RING(chan, nvc0->rast->size);
295 OUT_RINGp(chan, nvc0->rast->state, nvc0->rast->size);
296 }
297
298 static void
299 nvc0_constbufs_validate(struct nvc0_context *nvc0)
300 {
301 struct nouveau_channel *chan = nvc0->screen->base.channel;
302 struct nouveau_bo *bo;
303 unsigned s;
304
305 for (s = 0; s < 5; ++s) {
306 struct nv04_resource *res;
307 int i;
308
309 while (nvc0->constbuf_dirty[s]) {
310 unsigned base = 0;
311 unsigned offset = 0, words = 0;
312 boolean rebind = TRUE;
313
314 i = ffs(nvc0->constbuf_dirty[s]) - 1;
315 nvc0->constbuf_dirty[s] &= ~(1 << i);
316
317 res = nv04_resource(nvc0->constbuf[s][i]);
318 if (!res) {
319 BEGIN_RING(chan, RING_3D(CB_BIND(s)), 1);
320 OUT_RING (chan, (i << 4) | 0);
321 if (i == 0)
322 nvc0->state.uniform_buffer_bound[s] = 0;
323 continue;
324 }
325
326 if (!nouveau_resource_mapped_by_gpu(&res->base)) {
327 if (i == 0) {
328 base = s << 16;
329 bo = nvc0->screen->uniforms;
330
331 if (nvc0->state.uniform_buffer_bound[s] >= res->base.width0)
332 rebind = FALSE;
333 else
334 nvc0->state.uniform_buffer_bound[s] =
335 align(res->base.width0, 0x100);
336 } else {
337 bo = res->bo;
338 }
339 #if 0
340 nvc0_m2mf_push_linear(nvc0, bo, NOUVEAU_BO_VRAM,
341 base, res->base.width0, res->data);
342 BEGIN_RING(chan, RING_3D_(0x021c), 1);
343 OUT_RING (chan, 0x1111);
344 #else
345 words = res->base.width0 / 4;
346 #endif
347 } else {
348 bo = res->bo;
349 if (i == 0)
350 nvc0->state.uniform_buffer_bound[s] = 0;
351 }
352
353 if (bo != nvc0->screen->uniforms)
354 nvc0_bufctx_add_resident(nvc0, NVC0_BUFCTX_CONSTANT, res,
355 NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
356
357 if (rebind) {
358 MARK_RING (chan, 4, 2);
359 BEGIN_RING(chan, RING_3D(CB_SIZE), 3);
360 OUT_RING (chan, align(res->base.width0, 0x100));
361 OUT_RELOCh(chan, bo, base, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
362 OUT_RELOCl(chan, bo, base, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
363 BEGIN_RING(chan, RING_3D(CB_BIND(s)), 1);
364 OUT_RING (chan, (i << 4) | 1);
365 }
366
367 while (words) {
368 unsigned nr = AVAIL_RING(chan);
369
370 if (nr < 16) {
371 FIRE_RING(chan);
372 continue;
373 }
374 nr = MIN2(MIN2(nr - 6, words), NV04_PFIFO_MAX_PACKET_LEN - 1);
375
376 MARK_RING (chan, nr + 5, 2);
377 BEGIN_RING(chan, RING_3D(CB_SIZE), 3);
378 OUT_RING (chan, align(res->base.width0, 0x100));
379 OUT_RELOCh(chan, bo, base, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
380 OUT_RELOCl(chan, bo, base, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD);
381 BEGIN_RING_1I(chan, RING_3D(CB_POS), nr + 1);
382 OUT_RING (chan, offset);
383 OUT_RINGp (chan, &res->data[offset], nr);
384
385 offset += nr * 4;
386 words -= nr;
387 }
388 }
389 }
390 }
391
392 static void
393 nvc0_validate_derived_1(struct nvc0_context *nvc0)
394 {
395 struct nouveau_channel *chan = nvc0->screen->base.channel;
396 boolean early_z;
397
398 early_z = nvc0->fragprog->fp.early_z && !nvc0->zsa->pipe.alpha.enabled;
399
400 if (early_z != nvc0->state.early_z) {
401 nvc0->state.early_z = early_z;
402 IMMED_RING(chan, RING_3D(EARLY_FRAGMENT_TESTS), early_z);
403 }
404 }
405
406 static void
407 nvc0_switch_pipe_context(struct nvc0_context *ctx_to)
408 {
409 struct nvc0_context *ctx_from = ctx_to->screen->cur_ctx;
410
411 if (ctx_from)
412 ctx_to->state = ctx_from->state;
413
414 ctx_to->dirty = ~0;
415
416 if (!ctx_to->vertex)
417 ctx_to->dirty &= ~(NVC0_NEW_VERTEX | NVC0_NEW_ARRAYS);
418
419 if (!ctx_to->vertprog)
420 ctx_to->dirty &= ~NVC0_NEW_VERTPROG;
421 if (!ctx_to->fragprog)
422 ctx_to->dirty &= ~NVC0_NEW_FRAGPROG;
423
424 if (!ctx_to->blend)
425 ctx_to->dirty &= ~NVC0_NEW_BLEND;
426 if (!ctx_to->rast)
427 ctx_to->dirty &= ~NVC0_NEW_RASTERIZER;
428 if (!ctx_to->zsa)
429 ctx_to->dirty &= ~NVC0_NEW_ZSA;
430
431 ctx_to->screen->base.channel->user_private = ctx_to->screen->cur_ctx =
432 ctx_to;
433 }
434
435 static struct state_validate {
436 void (*func)(struct nvc0_context *);
437 uint32_t states;
438 } validate_list[] = {
439 { nvc0_validate_fb, NVC0_NEW_FRAMEBUFFER },
440 { nvc0_validate_blend, NVC0_NEW_BLEND },
441 { nvc0_validate_zsa, NVC0_NEW_ZSA },
442 { nvc0_validate_rasterizer, NVC0_NEW_RASTERIZER },
443 { nvc0_validate_blend_colour, NVC0_NEW_BLEND_COLOUR },
444 { nvc0_validate_stencil_ref, NVC0_NEW_STENCIL_REF },
445 { nvc0_validate_stipple, NVC0_NEW_STIPPLE },
446 { nvc0_validate_scissor, NVC0_NEW_SCISSOR | NVC0_NEW_RASTERIZER },
447 { nvc0_validate_viewport, NVC0_NEW_VIEWPORT },
448 { nvc0_validate_clip, NVC0_NEW_CLIP },
449 { nvc0_vertprog_validate, NVC0_NEW_VERTPROG },
450 { nvc0_tctlprog_validate, NVC0_NEW_TCTLPROG },
451 { nvc0_tevlprog_validate, NVC0_NEW_TEVLPROG },
452 { nvc0_gmtyprog_validate, NVC0_NEW_GMTYPROG },
453 { nvc0_fragprog_validate, NVC0_NEW_FRAGPROG },
454 { nvc0_validate_derived_1, NVC0_NEW_FRAGPROG | NVC0_NEW_ZSA },
455 { nvc0_constbufs_validate, NVC0_NEW_CONSTBUF },
456 { nvc0_validate_textures, NVC0_NEW_TEXTURES },
457 { nvc0_validate_samplers, NVC0_NEW_SAMPLERS },
458 { nvc0_vertex_arrays_validate, NVC0_NEW_VERTEX | NVC0_NEW_ARRAYS },
459 { nvc0_tfb_validate, NVC0_NEW_TFB | NVC0_NEW_TFB_BUFFERS }
460 };
461 #define validate_list_len (sizeof(validate_list) / sizeof(validate_list[0]))
462
463 boolean
464 nvc0_state_validate(struct nvc0_context *nvc0)
465 {
466 unsigned i;
467
468 if (nvc0->screen->cur_ctx != nvc0)
469 nvc0_switch_pipe_context(nvc0);
470
471 if (nvc0->dirty) {
472 for (i = 0; i < validate_list_len; ++i) {
473 struct state_validate *validate = &validate_list[i];
474
475 if (nvc0->dirty & validate->states)
476 validate->func(nvc0);
477 }
478 nvc0->dirty = 0;
479 }
480
481 nvc0_bufctx_emit_relocs(nvc0);
482
483 return TRUE;
484 }