1 /*
2 * Copyright 2008 Ben Skeggs
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 */
22
23 #include "nvc0/nvc0_context.h"
24 #include "nvc0/nvc0_resource.h"
25 #include "nvc0/gm107_texture.xml.h"
26 #include "nvc0/nvc0_compute.xml.h"
27 #include "nv50/g80_texture.xml.h"
28 #include "nv50/g80_defs.xml.h"
29
30 #include "util/format/u_format.h"
31
32 #define NVE4_TIC_ENTRY_INVALID 0x000fffff
33 #define NVE4_TSC_ENTRY_INVALID 0xfff00000
34
35 static inline uint32_t
nv50_tic_swizzle(const struct nvc0_format * fmt,unsigned swz,bool tex_int)36 nv50_tic_swizzle(const struct nvc0_format *fmt, unsigned swz, bool tex_int)
37 {
38 switch (swz) {
39 case PIPE_SWIZZLE_X : return fmt->tic.src_x;
40 case PIPE_SWIZZLE_Y: return fmt->tic.src_y;
41 case PIPE_SWIZZLE_Z : return fmt->tic.src_z;
42 case PIPE_SWIZZLE_W: return fmt->tic.src_w;
43 case PIPE_SWIZZLE_1:
44 return tex_int ? G80_TIC_SOURCE_ONE_INT : G80_TIC_SOURCE_ONE_FLOAT;
45 case PIPE_SWIZZLE_0:
46 default:
47 return G80_TIC_SOURCE_ZERO;
48 }
49 }
50
51 struct pipe_sampler_view *
nvc0_create_sampler_view(struct pipe_context * pipe,struct pipe_resource * res,const struct pipe_sampler_view * templ)52 nvc0_create_sampler_view(struct pipe_context *pipe,
53 struct pipe_resource *res,
54 const struct pipe_sampler_view *templ)
55 {
56 uint32_t flags = 0;
57
58 if (templ->target == PIPE_TEXTURE_RECT || templ->target == PIPE_BUFFER)
59 flags |= NV50_TEXVIEW_SCALED_COORDS;
60
61 return nvc0_create_texture_view(pipe, res, templ, flags, templ->target);
62 }
63
64 static struct pipe_sampler_view *
gm107_create_texture_view(struct pipe_context * pipe,struct pipe_resource * texture,const struct pipe_sampler_view * templ,uint32_t flags,enum pipe_texture_target target)65 gm107_create_texture_view(struct pipe_context *pipe,
66 struct pipe_resource *texture,
67 const struct pipe_sampler_view *templ,
68 uint32_t flags,
69 enum pipe_texture_target target)
70 {
71 const struct util_format_description *desc;
72 const struct nvc0_format *fmt;
73 uint64_t address;
74 uint32_t *tic;
75 uint32_t swz[4];
76 uint32_t width, height;
77 uint32_t depth;
78 struct nv50_tic_entry *view;
79 struct nv50_miptree *mt;
80 bool tex_int;
81
82 view = MALLOC_STRUCT(nv50_tic_entry);
83 if (!view)
84 return NULL;
85 mt = nv50_miptree(texture);
86
87 view->pipe = *templ;
88 view->pipe.reference.count = 1;
89 view->pipe.texture = NULL;
90 view->pipe.context = pipe;
91
92 view->id = -1;
93 view->bindless = 0;
94
95 pipe_resource_reference(&view->pipe.texture, texture);
96
97 tic = &view->tic[0];
98
99 desc = util_format_description(view->pipe.format);
100 tex_int = util_format_is_pure_integer(view->pipe.format);
101
102 fmt = &nvc0_format_table[view->pipe.format];
103 swz[0] = nv50_tic_swizzle(fmt, view->pipe.swizzle_r, tex_int);
104 swz[1] = nv50_tic_swizzle(fmt, view->pipe.swizzle_g, tex_int);
105 swz[2] = nv50_tic_swizzle(fmt, view->pipe.swizzle_b, tex_int);
106 swz[3] = nv50_tic_swizzle(fmt, view->pipe.swizzle_a, tex_int);
107
108 tic[0] = fmt->tic.format << GM107_TIC2_0_COMPONENTS_SIZES__SHIFT;
109 tic[0] |= fmt->tic.type_r << GM107_TIC2_0_R_DATA_TYPE__SHIFT;
110 tic[0] |= fmt->tic.type_g << GM107_TIC2_0_G_DATA_TYPE__SHIFT;
111 tic[0] |= fmt->tic.type_b << GM107_TIC2_0_B_DATA_TYPE__SHIFT;
112 tic[0] |= fmt->tic.type_a << GM107_TIC2_0_A_DATA_TYPE__SHIFT;
113 tic[0] |= swz[0] << GM107_TIC2_0_X_SOURCE__SHIFT;
114 tic[0] |= swz[1] << GM107_TIC2_0_Y_SOURCE__SHIFT;
115 tic[0] |= swz[2] << GM107_TIC2_0_Z_SOURCE__SHIFT;
116 tic[0] |= swz[3] << GM107_TIC2_0_W_SOURCE__SHIFT;
117
118 address = mt->base.address;
119
120 tic[3] = GM107_TIC2_3_LOD_ANISO_QUALITY_2;
121 tic[4] = GM107_TIC2_4_SECTOR_PROMOTION_PROMOTE_TO_2_V;
122 tic[4] |= GM107_TIC2_4_BORDER_SIZE_SAMPLER_COLOR;
123
124 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
125 tic[4] |= GM107_TIC2_4_SRGB_CONVERSION;
126
127 if (!(flags & NV50_TEXVIEW_SCALED_COORDS))
128 tic[5] = GM107_TIC2_5_NORMALIZED_COORDS;
129 else
130 tic[5] = 0;
131
132 /* check for linear storage type */
133 if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
134 if (texture->target == PIPE_BUFFER) {
135 assert(!(tic[5] & GM107_TIC2_5_NORMALIZED_COORDS));
136 width = view->pipe.u.buf.size / (desc->block.bits / 8) - 1;
137 address +=
138 view->pipe.u.buf.offset;
139 tic[2] = GM107_TIC2_2_HEADER_VERSION_ONE_D_BUFFER;
140 tic[3] |= width >> 16;
141 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_ONE_D_BUFFER;
142 tic[4] |= width & 0xffff;
143 } else {
144 assert(!(mt->level[0].pitch & 0x1f));
145 /* must be 2D texture without mip maps */
146 tic[2] = GM107_TIC2_2_HEADER_VERSION_PITCH;
147 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_TWO_D_NO_MIPMAP;
148 tic[3] |= mt->level[0].pitch >> 5;
149 tic[4] |= mt->base.base.width0 - 1;
150 tic[5] |= 0 << GM107_TIC2_5_DEPTH_MINUS_ONE__SHIFT;
151 tic[5] |= mt->base.base.height0 - 1;
152 }
153 tic[1] = address;
154 tic[2] |= address >> 32;
155 tic[6] = 0;
156 tic[7] = 0;
157 return &view->pipe;
158 }
159
160 tic[2] = GM107_TIC2_2_HEADER_VERSION_BLOCKLINEAR;
161 tic[3] |=
162 ((mt->level[0].tile_mode & 0x0f0) >> 4 << 3) |
163 ((mt->level[0].tile_mode & 0xf00) >> 8 << 6);
164
165 depth = MAX2(mt->base.base.array_size, mt->base.base.depth0);
166
167 if (mt->base.base.array_size > 1) {
168 /* there doesn't seem to be a base layer field in TIC */
169 address += view->pipe.u.tex.first_layer * mt->layer_stride;
170 depth = view->pipe.u.tex.last_layer - view->pipe.u.tex.first_layer + 1;
171 }
172 tic[1] = address;
173 tic[2] |= address >> 32;
174
175 switch (target) {
176 case PIPE_TEXTURE_1D:
177 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_ONE_D;
178 break;
179 case PIPE_TEXTURE_2D:
180 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_TWO_D;
181 break;
182 case PIPE_TEXTURE_RECT:
183 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_TWO_D;
184 break;
185 case PIPE_TEXTURE_3D:
186 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_THREE_D;
187 break;
188 case PIPE_TEXTURE_CUBE:
189 depth /= 6;
190 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_CUBEMAP;
191 break;
192 case PIPE_TEXTURE_1D_ARRAY:
193 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_ONE_D_ARRAY;
194 break;
195 case PIPE_TEXTURE_2D_ARRAY:
196 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_TWO_D_ARRAY;
197 break;
198 case PIPE_TEXTURE_CUBE_ARRAY:
199 depth /= 6;
200 tic[4] |= GM107_TIC2_4_TEXTURE_TYPE_CUBE_ARRAY;
201 break;
202 default:
203 unreachable("unexpected/invalid texture target");
204 }
205
206 tic[3] |= (flags & NV50_TEXVIEW_FILTER_MSAA8) ?
207 GM107_TIC2_3_USE_HEADER_OPT_CONTROL :
208 GM107_TIC2_3_LOD_ANISO_QUALITY_HIGH |
209 GM107_TIC2_3_LOD_ISO_QUALITY_HIGH;
210
211 if (flags & (NV50_TEXVIEW_ACCESS_RESOLVE | NV50_TEXVIEW_IMAGE_GM107)) {
212 width = mt->base.base.width0 << mt->ms_x;
213 height = mt->base.base.height0 << mt->ms_y;
214 } else {
215 width = mt->base.base.width0;
216 height = mt->base.base.height0;
217 }
218
219 tic[4] |= width - 1;
220
221 tic[5] |= (height - 1) & 0xffff;
222 tic[5] |= (depth - 1) << GM107_TIC2_5_DEPTH_MINUS_ONE__SHIFT;
223 tic[3] |= mt->base.base.last_level << GM107_TIC2_3_MAX_MIP_LEVEL__SHIFT;
224
225 /* sampling points: (?) */
226 if ((flags & NV50_TEXVIEW_ACCESS_RESOLVE) && mt->ms_x > 1) {
227 tic[6] = GM107_TIC2_6_ANISO_FINE_SPREAD_MODIFIER_CONST_TWO;
228 tic[6] |= GM107_TIC2_6_MAX_ANISOTROPY_2_TO_1;
229 } else {
230 tic[6] = GM107_TIC2_6_ANISO_FINE_SPREAD_FUNC_TWO;
231 tic[6] |= GM107_TIC2_6_ANISO_COARSE_SPREAD_FUNC_ONE;
232 }
233
234 tic[7] = (view->pipe.u.tex.last_level << 4) | view->pipe.u.tex.first_level;
235 tic[7] |= mt->ms_mode << GM107_TIC2_7_MULTI_SAMPLE_COUNT__SHIFT;
236
237 return &view->pipe;
238 }
239
240 struct pipe_sampler_view *
gm107_create_texture_view_from_image(struct pipe_context * pipe,const struct pipe_image_view * view)241 gm107_create_texture_view_from_image(struct pipe_context *pipe,
242 const struct pipe_image_view *view)
243 {
244 struct nv04_resource *res = nv04_resource(view->resource);
245 struct pipe_sampler_view templ = {};
246 enum pipe_texture_target target;
247 uint32_t flags = 0;
248
249 if (!res)
250 return NULL;
251 target = res->base.target;
252
253 if (target == PIPE_TEXTURE_CUBE || target == PIPE_TEXTURE_CUBE_ARRAY)
254 target = PIPE_TEXTURE_2D_ARRAY;
255
256 templ.format = view->format;
257 templ.swizzle_r = PIPE_SWIZZLE_X;
258 templ.swizzle_g = PIPE_SWIZZLE_Y;
259 templ.swizzle_b = PIPE_SWIZZLE_Z;
260 templ.swizzle_a = PIPE_SWIZZLE_W;
261
262 if (target == PIPE_BUFFER) {
263 templ.u.buf.offset = view->u.buf.offset;
264 templ.u.buf.size = view->u.buf.size;
265 } else {
266 templ.u.tex.first_layer = view->u.tex.first_layer;
267 templ.u.tex.last_layer = view->u.tex.last_layer;
268 templ.u.tex.first_level = templ.u.tex.last_level = view->u.tex.level;
269 }
270
271 flags = NV50_TEXVIEW_SCALED_COORDS | NV50_TEXVIEW_IMAGE_GM107;
272
273 return nvc0_create_texture_view(pipe, &res->base, &templ, flags, target);
274 }
275
276 static struct pipe_sampler_view *
gf100_create_texture_view(struct pipe_context * pipe,struct pipe_resource * texture,const struct pipe_sampler_view * templ,uint32_t flags,enum pipe_texture_target target)277 gf100_create_texture_view(struct pipe_context *pipe,
278 struct pipe_resource *texture,
279 const struct pipe_sampler_view *templ,
280 uint32_t flags,
281 enum pipe_texture_target target)
282 {
283 const struct util_format_description *desc;
284 const struct nvc0_format *fmt;
285 uint64_t address;
286 uint32_t *tic;
287 uint32_t swz[4];
288 uint32_t width, height;
289 uint32_t depth;
290 uint32_t tex_fmt;
291 struct nv50_tic_entry *view;
292 struct nv50_miptree *mt;
293 bool tex_int;
294
295 view = MALLOC_STRUCT(nv50_tic_entry);
296 if (!view)
297 return NULL;
298 mt = nv50_miptree(texture);
299
300 view->pipe = *templ;
301 view->pipe.reference.count = 1;
302 view->pipe.texture = NULL;
303 view->pipe.context = pipe;
304
305 view->id = -1;
306 view->bindless = 0;
307
308 pipe_resource_reference(&view->pipe.texture, texture);
309
310 tic = &view->tic[0];
311
312 desc = util_format_description(view->pipe.format);
313
314 fmt = &nvc0_format_table[view->pipe.format];
315
316 tex_int = util_format_is_pure_integer(view->pipe.format);
317 tex_fmt = fmt->tic.format & 0x3f;
318
319 swz[0] = nv50_tic_swizzle(fmt, view->pipe.swizzle_r, tex_int);
320 swz[1] = nv50_tic_swizzle(fmt, view->pipe.swizzle_g, tex_int);
321 swz[2] = nv50_tic_swizzle(fmt, view->pipe.swizzle_b, tex_int);
322 swz[3] = nv50_tic_swizzle(fmt, view->pipe.swizzle_a, tex_int);
323 tic[0] = (tex_fmt << G80_TIC_0_COMPONENTS_SIZES__SHIFT) |
324 (fmt->tic.type_r << G80_TIC_0_R_DATA_TYPE__SHIFT) |
325 (fmt->tic.type_g << G80_TIC_0_G_DATA_TYPE__SHIFT) |
326 (fmt->tic.type_b << G80_TIC_0_B_DATA_TYPE__SHIFT) |
327 (fmt->tic.type_a << G80_TIC_0_A_DATA_TYPE__SHIFT) |
328 (swz[0] << G80_TIC_0_X_SOURCE__SHIFT) |
329 (swz[1] << G80_TIC_0_Y_SOURCE__SHIFT) |
330 (swz[2] << G80_TIC_0_Z_SOURCE__SHIFT) |
331 (swz[3] << G80_TIC_0_W_SOURCE__SHIFT) |
332 ((fmt->tic.format & 0x40) << (GK20A_TIC_0_USE_COMPONENT_SIZES_EXTENDED__SHIFT - 6));
333
334 address = mt->base.address;
335
336 tic[2] = 0x10001000 | G80_TIC_2_BORDER_SOURCE_COLOR;
337
338 if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB)
339 tic[2] |= G80_TIC_2_SRGB_CONVERSION;
340
341 if (!(flags & NV50_TEXVIEW_SCALED_COORDS))
342 tic[2] |= G80_TIC_2_NORMALIZED_COORDS;
343
344 /* check for linear storage type */
345 if (unlikely(!nouveau_bo_memtype(nv04_resource(texture)->bo))) {
346 if (texture->target == PIPE_BUFFER) {
347 assert(!(tic[2] & G80_TIC_2_NORMALIZED_COORDS));
348 address +=
349 view->pipe.u.buf.offset;
350 tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_ONE_D_BUFFER;
351 tic[3] = 0;
352 tic[4] = /* width */
353 view->pipe.u.buf.size / (desc->block.bits / 8);
354 tic[5] = 0;
355 } else {
356 /* must be 2D texture without mip maps */
357 tic[2] |= G80_TIC_2_LAYOUT_PITCH | G80_TIC_2_TEXTURE_TYPE_TWO_D_NO_MIPMAP;
358 tic[3] = mt->level[0].pitch;
359 tic[4] = mt->base.base.width0;
360 tic[5] = (1 << 16) | mt->base.base.height0;
361 }
362 tic[6] =
363 tic[7] = 0;
364 tic[1] = address;
365 tic[2] |= address >> 32;
366 return &view->pipe;
367 }
368
369 tic[2] |=
370 ((mt->level[0].tile_mode & 0x0f0) << (22 - 4)) |
371 ((mt->level[0].tile_mode & 0xf00) << (25 - 8));
372
373 depth = MAX2(mt->base.base.array_size, mt->base.base.depth0);
374
375 if (mt->base.base.array_size > 1) {
376 /* there doesn't seem to be a base layer field in TIC */
377 address += view->pipe.u.tex.first_layer * mt->layer_stride;
378 depth = view->pipe.u.tex.last_layer - view->pipe.u.tex.first_layer + 1;
379 }
380 tic[1] = address;
381 tic[2] |= address >> 32;
382
383 switch (target) {
384 case PIPE_TEXTURE_1D:
385 tic[2] |= G80_TIC_2_TEXTURE_TYPE_ONE_D;
386 break;
387 case PIPE_TEXTURE_2D:
388 tic[2] |= G80_TIC_2_TEXTURE_TYPE_TWO_D;
389 break;
390 case PIPE_TEXTURE_RECT:
391 tic[2] |= G80_TIC_2_TEXTURE_TYPE_TWO_D;
392 break;
393 case PIPE_TEXTURE_3D:
394 tic[2] |= G80_TIC_2_TEXTURE_TYPE_THREE_D;
395 break;
396 case PIPE_TEXTURE_CUBE:
397 depth /= 6;
398 tic[2] |= G80_TIC_2_TEXTURE_TYPE_CUBEMAP;
399 break;
400 case PIPE_TEXTURE_1D_ARRAY:
401 tic[2] |= G80_TIC_2_TEXTURE_TYPE_ONE_D_ARRAY;
402 break;
403 case PIPE_TEXTURE_2D_ARRAY:
404 tic[2] |= G80_TIC_2_TEXTURE_TYPE_TWO_D_ARRAY;
405 break;
406 case PIPE_TEXTURE_CUBE_ARRAY:
407 depth /= 6;
408 tic[2] |= G80_TIC_2_TEXTURE_TYPE_CUBE_ARRAY;
409 break;
410 default:
411 unreachable("unexpected/invalid texture target");
412 }
413
414 tic[3] = (flags & NV50_TEXVIEW_FILTER_MSAA8) ? 0x20000000 : 0x00300000;
415
416 if (flags & NV50_TEXVIEW_ACCESS_RESOLVE) {
417 width = mt->base.base.width0 << mt->ms_x;
418 height = mt->base.base.height0 << mt->ms_y;
419 } else {
420 width = mt->base.base.width0;
421 height = mt->base.base.height0;
422 }
423
424 tic[4] = (1 << 31) | width;
425
426 tic[5] = height & 0xffff;
427 tic[5] |= depth << 16;
428 tic[5] |= mt->base.base.last_level << 28;
429
430 /* sampling points: (?) */
431 if (flags & NV50_TEXVIEW_ACCESS_RESOLVE)
432 tic[6] = (mt->ms_x > 1) ? 0x88000000 : 0x03000000;
433 else
434 tic[6] = 0x03000000;
435
436 tic[7] = (view->pipe.u.tex.last_level << 4) | view->pipe.u.tex.first_level;
437 tic[7] |= mt->ms_mode << 12;
438
439 return &view->pipe;
440 }
441
442 struct pipe_sampler_view *
nvc0_create_texture_view(struct pipe_context * pipe,struct pipe_resource * texture,const struct pipe_sampler_view * templ,uint32_t flags,enum pipe_texture_target target)443 nvc0_create_texture_view(struct pipe_context *pipe,
444 struct pipe_resource *texture,
445 const struct pipe_sampler_view *templ,
446 uint32_t flags,
447 enum pipe_texture_target target)
448 {
449 if (nvc0_context(pipe)->screen->tic.maxwell)
450 return gm107_create_texture_view(pipe, texture, templ, flags, target);
451 return gf100_create_texture_view(pipe, texture, templ, flags, target);
452 }
453
454 bool
nvc0_update_tic(struct nvc0_context * nvc0,struct nv50_tic_entry * tic,struct nv04_resource * res)455 nvc0_update_tic(struct nvc0_context *nvc0, struct nv50_tic_entry *tic,
456 struct nv04_resource *res)
457 {
458 uint64_t address = res->address;
459 if (res->base.target != PIPE_BUFFER)
460 return false;
461 address += tic->pipe.u.buf.offset;
462 if (tic->tic[1] == (uint32_t)address &&
463 (tic->tic[2] & 0xff) == address >> 32)
464 return false;
465
466 tic->tic[1] = address;
467 tic->tic[2] &= 0xffffff00;
468 tic->tic[2] |= address >> 32;
469
470 if (tic->id >= 0) {
471 nvc0->base.push_data(&nvc0->base, nvc0->screen->txc, tic->id * 32,
472 NV_VRAM_DOMAIN(&nvc0->screen->base), 32,
473 tic->tic);
474 return true;
475 }
476
477 return false;
478 }
479
480 bool
nvc0_validate_tic(struct nvc0_context * nvc0,int s)481 nvc0_validate_tic(struct nvc0_context *nvc0, int s)
482 {
483 uint32_t commands[32];
484 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
485 unsigned i;
486 unsigned n = 0;
487 bool need_flush = false;
488
489 for (i = 0; i < nvc0->num_textures[s]; ++i) {
490 struct nv50_tic_entry *tic = nv50_tic_entry(nvc0->textures[s][i]);
491 struct nv04_resource *res;
492 const bool dirty = !!(nvc0->textures_dirty[s] & (1 << i));
493
494 if (!tic) {
495 if (dirty)
496 commands[n++] = (i << 1) | 0;
497 continue;
498 }
499 res = nv04_resource(tic->pipe.texture);
500 need_flush |= nvc0_update_tic(nvc0, tic, res);
501
502 if (tic->id < 0) {
503 tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);
504
505 nvc0->base.push_data(&nvc0->base, nvc0->screen->txc, tic->id * 32,
506 NV_VRAM_DOMAIN(&nvc0->screen->base), 32,
507 tic->tic);
508 need_flush = true;
509 } else
510 if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
511 if (unlikely(s == 5))
512 BEGIN_NVC0(push, NVC0_CP(TEX_CACHE_CTL), 1);
513 else
514 BEGIN_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 1);
515 PUSH_DATA (push, (tic->id << 4) | 1);
516 NOUVEAU_DRV_STAT(&nvc0->screen->base, tex_cache_flush_count, 1);
517 }
518 nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
519
520 res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
521 res->status |= NOUVEAU_BUFFER_STATUS_GPU_READING;
522
523 if (!dirty)
524 continue;
525 commands[n++] = (tic->id << 9) | (i << 1) | 1;
526
527 if (unlikely(s == 5))
528 BCTX_REFN(nvc0->bufctx_cp, CP_TEX(i), res, RD);
529 else
530 BCTX_REFN(nvc0->bufctx_3d, 3D_TEX(s, i), res, RD);
531 }
532 for (; i < nvc0->state.num_textures[s]; ++i)
533 commands[n++] = (i << 1) | 0;
534
535 nvc0->state.num_textures[s] = nvc0->num_textures[s];
536
537 if (n) {
538 if (unlikely(s == 5))
539 BEGIN_NIC0(push, NVC0_CP(BIND_TIC), n);
540 else
541 BEGIN_NIC0(push, NVC0_3D(BIND_TIC(s)), n);
542 PUSH_DATAp(push, commands, n);
543 }
544 nvc0->textures_dirty[s] = 0;
545
546 return need_flush;
547 }
548
549 static bool
nve4_validate_tic(struct nvc0_context * nvc0,unsigned s)550 nve4_validate_tic(struct nvc0_context *nvc0, unsigned s)
551 {
552 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
553 unsigned i;
554 bool need_flush = false;
555
556 for (i = 0; i < nvc0->num_textures[s]; ++i) {
557 struct nv50_tic_entry *tic = nv50_tic_entry(nvc0->textures[s][i]);
558 struct nv04_resource *res;
559 const bool dirty = !!(nvc0->textures_dirty[s] & (1 << i));
560
561 if (!tic) {
562 nvc0->tex_handles[s][i] |= NVE4_TIC_ENTRY_INVALID;
563 continue;
564 }
565 res = nv04_resource(tic->pipe.texture);
566 need_flush |= nvc0_update_tic(nvc0, tic, res);
567
568 if (tic->id < 0) {
569 tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);
570
571 nvc0->base.push_data(&nvc0->base, nvc0->screen->txc, tic->id * 32,
572 NV_VRAM_DOMAIN(&nvc0->screen->base), 32,
573 tic->tic);
574 need_flush = true;
575 } else
576 if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
577 BEGIN_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 1);
578 PUSH_DATA (push, (tic->id << 4) | 1);
579 }
580 nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
581
582 res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
583 res->status |= NOUVEAU_BUFFER_STATUS_GPU_READING;
584
585 nvc0->tex_handles[s][i] &= ~NVE4_TIC_ENTRY_INVALID;
586 nvc0->tex_handles[s][i] |= tic->id;
587 if (dirty)
588 BCTX_REFN(nvc0->bufctx_3d, 3D_TEX(s, i), res, RD);
589 }
590 for (; i < nvc0->state.num_textures[s]; ++i) {
591 nvc0->tex_handles[s][i] |= NVE4_TIC_ENTRY_INVALID;
592 nvc0->textures_dirty[s] |= 1 << i;
593 }
594
595 nvc0->state.num_textures[s] = nvc0->num_textures[s];
596
597 return need_flush;
598 }
599
nvc0_validate_textures(struct nvc0_context * nvc0)600 void nvc0_validate_textures(struct nvc0_context *nvc0)
601 {
602 bool need_flush = false;
603 int i;
604
605 for (i = 0; i < 5; i++) {
606 if (nvc0->screen->base.class_3d >= NVE4_3D_CLASS)
607 need_flush |= nve4_validate_tic(nvc0, i);
608 else
609 need_flush |= nvc0_validate_tic(nvc0, i);
610 }
611
612 if (need_flush) {
613 BEGIN_NVC0(nvc0->base.pushbuf, NVC0_3D(TIC_FLUSH), 1);
614 PUSH_DATA (nvc0->base.pushbuf, 0);
615 }
616
617 /* Invalidate all CP textures because they are aliased. */
618 for (int i = 0; i < nvc0->num_textures[5]; i++)
619 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_TEX(i));
620 nvc0->textures_dirty[5] = ~0;
621 nvc0->dirty_cp |= NVC0_NEW_CP_TEXTURES;
622 }
623
624 bool
nvc0_validate_tsc(struct nvc0_context * nvc0,int s)625 nvc0_validate_tsc(struct nvc0_context *nvc0, int s)
626 {
627 uint32_t commands[16];
628 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
629 unsigned i;
630 unsigned n = 0;
631 bool need_flush = false;
632
633 for (i = 0; i < nvc0->num_samplers[s]; ++i) {
634 struct nv50_tsc_entry *tsc = nv50_tsc_entry(nvc0->samplers[s][i]);
635
636 if (!(nvc0->samplers_dirty[s] & (1 << i)))
637 continue;
638 if (!tsc) {
639 commands[n++] = (i << 4) | 0;
640 continue;
641 }
642 nvc0->seamless_cube_map = tsc->seamless_cube_map;
643 if (tsc->id < 0) {
644 tsc->id = nvc0_screen_tsc_alloc(nvc0->screen, tsc);
645
646 nvc0_m2mf_push_linear(&nvc0->base, nvc0->screen->txc,
647 65536 + tsc->id * 32, NV_VRAM_DOMAIN(&nvc0->screen->base),
648 32, tsc->tsc);
649 need_flush = true;
650 }
651 nvc0->screen->tsc.lock[tsc->id / 32] |= 1 << (tsc->id % 32);
652
653 commands[n++] = (tsc->id << 12) | (i << 4) | 1;
654 }
655 for (; i < nvc0->state.num_samplers[s]; ++i)
656 commands[n++] = (i << 4) | 0;
657
658 nvc0->state.num_samplers[s] = nvc0->num_samplers[s];
659
660 // TXF, in unlinked tsc mode, will always use sampler 0. So we have to
661 // ensure that it remains bound. Its contents don't matter, all samplers we
662 // ever create have the SRGB_CONVERSION bit set, so as long as the first
663 // entry is initialized, we're good to go. This is the only bit that has
664 // any effect on what TXF does.
665 if ((nvc0->samplers_dirty[s] & 1) && !nvc0->samplers[s][0]) {
666 if (n == 0)
667 n = 1;
668 // We're guaranteed that the first command refers to the first slot, so
669 // we're not overwriting a valid entry.
670 commands[0] = (0 << 12) | (0 << 4) | 1;
671 }
672
673 if (n) {
674 if (unlikely(s == 5))
675 BEGIN_NIC0(push, NVC0_CP(BIND_TSC), n);
676 else
677 BEGIN_NIC0(push, NVC0_3D(BIND_TSC(s)), n);
678 PUSH_DATAp(push, commands, n);
679 }
680 nvc0->samplers_dirty[s] = 0;
681
682 return need_flush;
683 }
684
685 bool
nve4_validate_tsc(struct nvc0_context * nvc0,int s)686 nve4_validate_tsc(struct nvc0_context *nvc0, int s)
687 {
688 unsigned i;
689 bool need_flush = false;
690
691 for (i = 0; i < nvc0->num_samplers[s]; ++i) {
692 struct nv50_tsc_entry *tsc = nv50_tsc_entry(nvc0->samplers[s][i]);
693
694 if (!tsc) {
695 nvc0->tex_handles[s][i] |= NVE4_TSC_ENTRY_INVALID;
696 continue;
697 }
698 if (tsc->id < 0) {
699 tsc->id = nvc0_screen_tsc_alloc(nvc0->screen, tsc);
700
701 nve4_p2mf_push_linear(&nvc0->base, nvc0->screen->txc,
702 65536 + tsc->id * 32,
703 NV_VRAM_DOMAIN(&nvc0->screen->base),
704 32, tsc->tsc);
705 need_flush = true;
706 }
707 nvc0->screen->tsc.lock[tsc->id / 32] |= 1 << (tsc->id % 32);
708
709 nvc0->tex_handles[s][i] &= ~NVE4_TSC_ENTRY_INVALID;
710 nvc0->tex_handles[s][i] |= tsc->id << 20;
711 }
712 for (; i < nvc0->state.num_samplers[s]; ++i) {
713 nvc0->tex_handles[s][i] |= NVE4_TSC_ENTRY_INVALID;
714 nvc0->samplers_dirty[s] |= 1 << i;
715 }
716
717 nvc0->state.num_samplers[s] = nvc0->num_samplers[s];
718
719 return need_flush;
720 }
721
nvc0_validate_samplers(struct nvc0_context * nvc0)722 void nvc0_validate_samplers(struct nvc0_context *nvc0)
723 {
724 bool need_flush = false;
725 int i;
726
727 for (i = 0; i < 5; i++) {
728 if (nvc0->screen->base.class_3d >= NVE4_3D_CLASS)
729 need_flush |= nve4_validate_tsc(nvc0, i);
730 else
731 need_flush |= nvc0_validate_tsc(nvc0, i);
732 }
733
734 if (need_flush) {
735 BEGIN_NVC0(nvc0->base.pushbuf, NVC0_3D(TSC_FLUSH), 1);
736 PUSH_DATA (nvc0->base.pushbuf, 0);
737 }
738
739 /* Invalidate all CP samplers because they are aliased. */
740 nvc0->samplers_dirty[5] = ~0;
741 nvc0->dirty_cp |= NVC0_NEW_CP_SAMPLERS;
742 }
743
744 void
nvc0_upload_tsc0(struct nvc0_context * nvc0)745 nvc0_upload_tsc0(struct nvc0_context *nvc0)
746 {
747 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
748 u32 data[8] = { G80_TSC_0_SRGB_CONVERSION };
749 nvc0->base.push_data(&nvc0->base, nvc0->screen->txc,
750 65536 /*+ tsc->id * 32*/,
751 NV_VRAM_DOMAIN(&nvc0->screen->base), 32, data);
752 BEGIN_NVC0(push, NVC0_3D(TSC_FLUSH), 1);
753 PUSH_DATA (push, 0);
754 }
755
756 /* Upload the "diagonal" entries for the possible texture sources ($t == $s).
757 * At some point we might want to get a list of the combinations used by a
758 * shader and fill in those entries instead of having it extract the handles.
759 */
760 void
nve4_set_tex_handles(struct nvc0_context * nvc0)761 nve4_set_tex_handles(struct nvc0_context *nvc0)
762 {
763 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
764 struct nvc0_screen *screen = nvc0->screen;
765 unsigned s;
766
767 if (nvc0->screen->base.class_3d < NVE4_3D_CLASS)
768 return;
769
770 for (s = 0; s < 5; ++s) {
771 uint32_t dirty = nvc0->textures_dirty[s] | nvc0->samplers_dirty[s];
772 if (!dirty)
773 continue;
774 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
775 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
776 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
777 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
778 do {
779 int i = ffs(dirty) - 1;
780 dirty &= ~(1 << i);
781
782 BEGIN_NVC0(push, NVC0_3D(CB_POS), 2);
783 PUSH_DATA (push, NVC0_CB_AUX_TEX_INFO(i));
784 PUSH_DATA (push, nvc0->tex_handles[s][i]);
785 } while (dirty);
786
787 nvc0->textures_dirty[s] = 0;
788 nvc0->samplers_dirty[s] = 0;
789 }
790 }
791
792 static uint64_t
nve4_create_texture_handle(struct pipe_context * pipe,struct pipe_sampler_view * view,const struct pipe_sampler_state * sampler)793 nve4_create_texture_handle(struct pipe_context *pipe,
794 struct pipe_sampler_view *view,
795 const struct pipe_sampler_state *sampler)
796 {
797 /* We have to create persistent handles that won't change for these objects
798 * That means that we have to upload them into place and lock them so that
799 * they can't be kicked out later.
800 */
801 struct nvc0_context *nvc0 = nvc0_context(pipe);
802 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
803 struct nv50_tic_entry *tic = nv50_tic_entry(view);
804 struct nv50_tsc_entry *tsc = pipe->create_sampler_state(pipe, sampler);
805 struct pipe_sampler_view *v = NULL;
806
807 tsc->id = nvc0_screen_tsc_alloc(nvc0->screen, tsc);
808 if (tsc->id < 0)
809 goto fail;
810
811 if (tic->id < 0) {
812 tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);
813 if (tic->id < 0)
814 goto fail;
815
816 nve4_p2mf_push_linear(&nvc0->base, nvc0->screen->txc, tic->id * 32,
817 NV_VRAM_DOMAIN(&nvc0->screen->base), 32,
818 tic->tic);
819
820 IMMED_NVC0(push, NVC0_3D(TIC_FLUSH), 0);
821 }
822
823 nve4_p2mf_push_linear(&nvc0->base, nvc0->screen->txc,
824 65536 + tsc->id * 32,
825 NV_VRAM_DOMAIN(&nvc0->screen->base),
826 32, tsc->tsc);
827
828 IMMED_NVC0(push, NVC0_3D(TSC_FLUSH), 0);
829
830 // Add an extra reference to this sampler view effectively held by this
831 // texture handle. This is to deal with the sampler view being dereferenced
832 // before the handle is. However we need the view to still be live until the
833 // handle to it is deleted.
834 pipe_sampler_view_reference(&v, view);
835 p_atomic_inc(&tic->bindless);
836
837 nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
838 nvc0->screen->tsc.lock[tsc->id / 32] |= 1 << (tsc->id % 32);
839
840 return 0x100000000ULL | (tsc->id << 20) | tic->id;
841
842 fail:
843 pipe->delete_sampler_state(pipe, tsc);
844 return 0;
845 }
846
847 static bool
view_bound(struct nvc0_context * nvc0,struct pipe_sampler_view * view)848 view_bound(struct nvc0_context *nvc0, struct pipe_sampler_view *view) {
849 for (int s = 0; s < 6; s++) {
850 for (int i = 0; i < nvc0->num_textures[s]; i++)
851 if (nvc0->textures[s][i] == view)
852 return true;
853 }
854 return false;
855 }
856
857 static void
nve4_delete_texture_handle(struct pipe_context * pipe,uint64_t handle)858 nve4_delete_texture_handle(struct pipe_context *pipe, uint64_t handle)
859 {
860 struct nvc0_context *nvc0 = nvc0_context(pipe);
861 uint32_t tic = handle & NVE4_TIC_ENTRY_INVALID;
862 uint32_t tsc = (handle & NVE4_TSC_ENTRY_INVALID) >> 20;
863 struct nv50_tic_entry *entry = nvc0->screen->tic.entries[tic];
864
865 if (entry) {
866 struct pipe_sampler_view *view = &entry->pipe;
867 assert(entry->bindless);
868 p_atomic_dec(&entry->bindless);
869 if (!view_bound(nvc0, view))
870 nvc0_screen_tic_unlock(nvc0->screen, entry);
871 pipe_sampler_view_reference(&view, NULL);
872 }
873
874 pipe->delete_sampler_state(pipe, nvc0->screen->tsc.entries[tsc]);
875 }
876
877 static void
nve4_make_texture_handle_resident(struct pipe_context * pipe,uint64_t handle,bool resident)878 nve4_make_texture_handle_resident(struct pipe_context *pipe,
879 uint64_t handle, bool resident)
880 {
881 struct nvc0_context *nvc0 = nvc0_context(pipe);
882 if (resident) {
883 struct nvc0_resident *res = calloc(1, sizeof(struct nvc0_resident));
884 struct nv50_tic_entry *tic =
885 nvc0->screen->tic.entries[handle & NVE4_TIC_ENTRY_INVALID];
886 assert(tic);
887 assert(tic->bindless);
888
889 res->handle = handle;
890 res->buf = nv04_resource(tic->pipe.texture);
891 res->flags = NOUVEAU_BO_RD;
892 list_add(&res->list, &nvc0->tex_head);
893 } else {
894 list_for_each_entry_safe(struct nvc0_resident, pos, &nvc0->tex_head, list) {
895 if (pos->handle == handle) {
896 list_del(&pos->list);
897 free(pos);
898 break;
899 }
900 }
901 }
902 }
903
904 static const uint8_t nve4_su_format_map[PIPE_FORMAT_COUNT];
905 static const uint16_t nve4_su_format_aux_map[PIPE_FORMAT_COUNT];
906 static const uint16_t nve4_suldp_lib_offset[PIPE_FORMAT_COUNT];
907
908 static void
nvc0_get_surface_dims(const struct pipe_image_view * view,int * width,int * height,int * depth)909 nvc0_get_surface_dims(const struct pipe_image_view *view,
910 int *width, int *height, int *depth)
911 {
912 struct nv04_resource *res = nv04_resource(view->resource);
913 int level;
914
915 *width = *height = *depth = 1;
916 if (res->base.target == PIPE_BUFFER) {
917 *width = view->u.buf.size / util_format_get_blocksize(view->format);
918 return;
919 }
920
921 level = view->u.tex.level;
922 *width = u_minify(view->resource->width0, level);
923 *height = u_minify(view->resource->height0, level);
924 *depth = u_minify(view->resource->depth0, level);
925
926 switch (res->base.target) {
927 case PIPE_TEXTURE_1D_ARRAY:
928 case PIPE_TEXTURE_2D_ARRAY:
929 case PIPE_TEXTURE_CUBE:
930 case PIPE_TEXTURE_CUBE_ARRAY:
931 *depth = view->u.tex.last_layer - view->u.tex.first_layer + 1;
932 break;
933 case PIPE_TEXTURE_1D:
934 case PIPE_TEXTURE_2D:
935 case PIPE_TEXTURE_RECT:
936 case PIPE_TEXTURE_3D:
937 break;
938 default:
939 assert(!"unexpected texture target");
940 break;
941 }
942 }
943
944 void
nvc0_mark_image_range_valid(const struct pipe_image_view * view)945 nvc0_mark_image_range_valid(const struct pipe_image_view *view)
946 {
947 struct nv04_resource *res = (struct nv04_resource *)view->resource;
948
949 assert(view->resource->target == PIPE_BUFFER);
950
951 util_range_add(&res->base, &res->valid_buffer_range,
952 view->u.buf.offset,
953 view->u.buf.offset + view->u.buf.size);
954 }
955
956 void
nve4_set_surface_info(struct nouveau_pushbuf * push,const struct pipe_image_view * view,struct nvc0_context * nvc0)957 nve4_set_surface_info(struct nouveau_pushbuf *push,
958 const struct pipe_image_view *view,
959 struct nvc0_context *nvc0)
960 {
961 struct nvc0_screen *screen = nvc0->screen;
962 struct nv04_resource *res;
963 uint64_t address;
964 uint32_t *const info = push->cur;
965 int width, height, depth;
966 uint8_t log2cpp;
967
968 if (view && !nve4_su_format_map[view->format])
969 NOUVEAU_ERR("unsupported surface format, try is_format_supported() !\n");
970
971 push->cur += 16;
972
973 if (!view || !nve4_su_format_map[view->format]) {
974 memset(info, 0, 16 * sizeof(*info));
975
976 info[0] = 0xbadf0000;
977 info[1] = 0x80004000;
978 info[12] = nve4_suldp_lib_offset[PIPE_FORMAT_R32G32B32A32_UINT] +
979 screen->lib_code->start;
980 return;
981 }
982 res = nv04_resource(view->resource);
983
984 address = res->address;
985
986 /* get surface dimensions based on the target. */
987 nvc0_get_surface_dims(view, &width, &height, &depth);
988
989 info[8] = width;
990 info[9] = height;
991 info[10] = depth;
992 switch (res->base.target) {
993 case PIPE_TEXTURE_1D_ARRAY:
994 info[11] = 1;
995 break;
996 case PIPE_TEXTURE_2D:
997 case PIPE_TEXTURE_RECT:
998 info[11] = 2;
999 break;
1000 case PIPE_TEXTURE_3D:
1001 info[11] = 3;
1002 break;
1003 case PIPE_TEXTURE_2D_ARRAY:
1004 case PIPE_TEXTURE_CUBE:
1005 case PIPE_TEXTURE_CUBE_ARRAY:
1006 info[11] = 4;
1007 break;
1008 default:
1009 info[11] = 0;
1010 break;
1011 }
1012 log2cpp = (0xf000 & nve4_su_format_aux_map[view->format]) >> 12;
1013
1014 /* Stick the blockwidth (ie. number of bytes per pixel) to check if the
1015 * format doesn't mismatch. */
1016 info[12] = util_format_get_blocksize(view->format);
1017
1018 /* limit in bytes for raw access */
1019 info[13] = (0x06 << 22) | ((width << log2cpp) - 1);
1020
1021 info[1] = nve4_su_format_map[view->format];
1022
1023 #if 0
1024 switch (util_format_get_blocksizebits(view->format)) {
1025 case 16: info[1] |= 1 << 16; break;
1026 case 32: info[1] |= 2 << 16; break;
1027 case 64: info[1] |= 3 << 16; break;
1028 case 128: info[1] |= 4 << 16; break;
1029 default:
1030 break;
1031 }
1032 #else
1033 info[1] |= log2cpp << 16;
1034 info[1] |= 0x4000;
1035 info[1] |= (0x0f00 & nve4_su_format_aux_map[view->format]);
1036 #endif
1037
1038 if (res->base.target == PIPE_BUFFER) {
1039 address += view->u.buf.offset;
1040
1041 info[0] = address >> 8;
1042 info[2] = width - 1;
1043 info[2] |= (0xff & nve4_su_format_aux_map[view->format]) << 22;
1044 info[3] = 0;
1045 info[4] = 0;
1046 info[5] = 0;
1047 info[6] = 0;
1048 info[7] = 0;
1049 info[14] = 0;
1050 info[15] = 0;
1051 } else {
1052 struct nv50_miptree *mt = nv50_miptree(&res->base);
1053 struct nv50_miptree_level *lvl = &mt->level[view->u.tex.level];
1054 unsigned z = view->u.tex.first_layer;
1055
1056 if (!mt->layout_3d) {
1057 address += mt->layer_stride * z;
1058 z = 0;
1059 }
1060
1061 address += lvl->offset;
1062
1063 info[0] = address >> 8;
1064 info[2] = (width << mt->ms_x) - 1;
1065 /* NOTE: this is really important: */
1066 info[2] |= (0xff & nve4_su_format_aux_map[view->format]) << 22;
1067 info[3] = (0x88 << 24) | (lvl->pitch / 64);
1068 info[4] = (height << mt->ms_y) - 1;
1069 info[4] |= (lvl->tile_mode & 0x0f0) << 25;
1070 info[4] |= NVC0_TILE_SHIFT_Y(lvl->tile_mode) << 22;
1071 info[5] = mt->layer_stride >> 8;
1072 info[6] = depth - 1;
1073 info[6] |= (lvl->tile_mode & 0xf00) << 21;
1074 info[6] |= NVC0_TILE_SHIFT_Z(lvl->tile_mode) << 22;
1075 info[7] = mt->layout_3d ? 1 : 0;
1076 info[7] |= z << 16;
1077 info[14] = mt->ms_x;
1078 info[15] = mt->ms_y;
1079 }
1080 }
1081
1082 static inline void
nvc0_set_surface_info(struct nouveau_pushbuf * push,const struct pipe_image_view * view,uint64_t address,int width,int height,int depth)1083 nvc0_set_surface_info(struct nouveau_pushbuf *push,
1084 const struct pipe_image_view *view, uint64_t address,
1085 int width, int height, int depth)
1086 {
1087 struct nv04_resource *res;
1088 uint32_t *const info = push->cur;
1089
1090 push->cur += 16;
1091
1092 /* Make sure to always initialize the surface information area because it's
1093 * used to check if the given image is bound or not. */
1094 memset(info, 0, 16 * sizeof(*info));
1095
1096 if (!view || !view->resource)
1097 return;
1098 res = nv04_resource(view->resource);
1099
1100 /* Stick the image dimensions for the imageSize() builtin. */
1101 info[8] = width;
1102 info[9] = height;
1103 info[10] = depth;
1104
1105 /* Stick the blockwidth (ie. number of bytes per pixel) to calculate pixel
1106 * offset and to check if the format doesn't mismatch. */
1107 info[12] = util_format_get_blocksize(view->format);
1108
1109 if (res->base.target == PIPE_BUFFER) {
1110 info[0] = address >> 8;
1111 info[2] = width;
1112 } else {
1113 struct nv50_miptree *mt = nv50_miptree(&res->base);
1114
1115 info[0] = address >> 8;
1116 info[2] = width;
1117 info[4] = height;
1118 info[5] = mt->layer_stride >> 8;
1119 info[6] = depth;
1120 info[14] = mt->ms_x;
1121 info[15] = mt->ms_y;
1122 }
1123 }
1124
1125 void
nvc0_validate_suf(struct nvc0_context * nvc0,int s)1126 nvc0_validate_suf(struct nvc0_context *nvc0, int s)
1127 {
1128 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
1129 struct nvc0_screen *screen = nvc0->screen;
1130
1131 for (int i = 0; i < NVC0_MAX_IMAGES; ++i) {
1132 struct pipe_image_view *view = &nvc0->images[s][i];
1133 int width, height, depth;
1134 uint64_t address = 0;
1135
1136 if (s == 5)
1137 BEGIN_NVC0(push, NVC0_CP(IMAGE(i)), 6);
1138 else
1139 BEGIN_NVC0(push, NVC0_3D(IMAGE(i)), 6);
1140
1141 if (view->resource) {
1142 struct nv04_resource *res = nv04_resource(view->resource);
1143 unsigned rt = nvc0_format_table[view->format].rt;
1144
1145 if (util_format_is_depth_or_stencil(view->format))
1146 rt = rt << 12;
1147 else
1148 rt = (rt << 4) | (0x14 << 12);
1149
1150 /* get surface dimensions based on the target. */
1151 nvc0_get_surface_dims(view, &width, &height, &depth);
1152
1153 address = res->address;
1154 if (res->base.target == PIPE_BUFFER) {
1155 unsigned blocksize = util_format_get_blocksize(view->format);
1156
1157 address += view->u.buf.offset;
1158 assert(!(address & 0xff));
1159
1160 if (view->access & PIPE_IMAGE_ACCESS_WRITE)
1161 nvc0_mark_image_range_valid(view);
1162
1163 PUSH_DATAh(push, address);
1164 PUSH_DATA (push, address);
1165 PUSH_DATA (push, align(width * blocksize, 0x100));
1166 PUSH_DATA (push, NVC0_3D_IMAGE_HEIGHT_LINEAR | 1);
1167 PUSH_DATA (push, rt);
1168 PUSH_DATA (push, 0);
1169 } else {
1170 struct nv50_miptree *mt = nv50_miptree(view->resource);
1171 struct nv50_miptree_level *lvl = &mt->level[view->u.tex.level];
1172 const unsigned z = view->u.tex.first_layer;
1173
1174 if (mt->layout_3d) {
1175 address += nvc0_mt_zslice_offset(mt, view->u.tex.level, z);
1176 if (depth >= 1) {
1177 pipe_debug_message(&nvc0->base.debug, CONFORMANCE,
1178 "3D images are not supported!");
1179 debug_printf("3D images are not supported!\n");
1180 }
1181 } else {
1182 address += mt->layer_stride * z;
1183 }
1184 address += lvl->offset;
1185
1186 PUSH_DATAh(push, address);
1187 PUSH_DATA (push, address);
1188 PUSH_DATA (push, width << mt->ms_x);
1189 PUSH_DATA (push, height << mt->ms_y);
1190 PUSH_DATA (push, rt);
1191 PUSH_DATA (push, lvl->tile_mode & 0xff); /* mask out z-tiling */
1192 }
1193
1194 if (s == 5)
1195 BCTX_REFN(nvc0->bufctx_cp, CP_SUF, res, RDWR);
1196 else
1197 BCTX_REFN(nvc0->bufctx_3d, 3D_SUF, res, RDWR);
1198 } else {
1199 PUSH_DATA(push, 0);
1200 PUSH_DATA(push, 0);
1201 PUSH_DATA(push, 0);
1202 PUSH_DATA(push, 0);
1203 PUSH_DATA(push, 0x14000);
1204 PUSH_DATA(push, 0);
1205 }
1206
1207 /* stick surface information into the driver constant buffer */
1208 if (s == 5)
1209 BEGIN_NVC0(push, NVC0_CP(CB_SIZE), 3);
1210 else
1211 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1212 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1213 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
1214 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
1215 if (s == 5)
1216 BEGIN_1IC0(push, NVC0_CP(CB_POS), 1 + 16);
1217 else
1218 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 16);
1219 PUSH_DATA (push, NVC0_CB_AUX_SU_INFO(i));
1220
1221 nvc0_set_surface_info(push, view, address, width, height, depth);
1222 }
1223 }
1224
1225 static inline void
nvc0_update_surface_bindings(struct nvc0_context * nvc0)1226 nvc0_update_surface_bindings(struct nvc0_context *nvc0)
1227 {
1228 nvc0_validate_suf(nvc0, 4);
1229
1230 /* Invalidate all COMPUTE images because they are aliased with FRAGMENT. */
1231 nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_SUF);
1232 nvc0->dirty_cp |= NVC0_NEW_CP_SURFACES;
1233 nvc0->images_dirty[5] |= nvc0->images_valid[5];
1234 }
1235
1236 static void
gm107_validate_surfaces(struct nvc0_context * nvc0,struct pipe_image_view * view,int stage,int slot)1237 gm107_validate_surfaces(struct nvc0_context *nvc0,
1238 struct pipe_image_view *view, int stage, int slot)
1239 {
1240 struct nv04_resource *res = nv04_resource(view->resource);
1241 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
1242 struct nvc0_screen *screen = nvc0->screen;
1243 struct nv50_tic_entry *tic;
1244
1245 tic = nv50_tic_entry(nvc0->images_tic[stage][slot]);
1246
1247 res = nv04_resource(tic->pipe.texture);
1248 nvc0_update_tic(nvc0, tic, res);
1249
1250 if (tic->id < 0) {
1251 tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);
1252
1253 /* upload the texture view */
1254 nve4_p2mf_push_linear(&nvc0->base, nvc0->screen->txc, tic->id * 32,
1255 NV_VRAM_DOMAIN(&nvc0->screen->base), 32, tic->tic);
1256
1257 BEGIN_NVC0(push, NVC0_3D(TIC_FLUSH), 1);
1258 PUSH_DATA (push, 0);
1259 } else
1260 if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
1261 BEGIN_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 1);
1262 PUSH_DATA (push, (tic->id << 4) | 1);
1263 }
1264 nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
1265
1266 res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
1267 res->status |= NOUVEAU_BUFFER_STATUS_GPU_READING;
1268
1269 BCTX_REFN(nvc0->bufctx_3d, 3D_SUF, res, RD);
1270
1271 /* upload the texture handle */
1272 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1273 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1274 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(stage));
1275 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(stage));
1276 BEGIN_NVC0(push, NVC0_3D(CB_POS), 2);
1277 PUSH_DATA (push, NVC0_CB_AUX_TEX_INFO(slot + 32));
1278 PUSH_DATA (push, tic->id);
1279 }
1280
1281 static inline void
nve4_update_surface_bindings(struct nvc0_context * nvc0)1282 nve4_update_surface_bindings(struct nvc0_context *nvc0)
1283 {
1284 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
1285 struct nvc0_screen *screen = nvc0->screen;
1286 int i, j, s;
1287
1288 for (s = 0; s < 5; s++) {
1289 if (!nvc0->images_dirty[s])
1290 continue;
1291
1292 for (i = 0; i < NVC0_MAX_IMAGES; ++i) {
1293 struct pipe_image_view *view = &nvc0->images[s][i];
1294
1295 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1296 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1297 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
1298 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
1299 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 16);
1300 PUSH_DATA (push, NVC0_CB_AUX_SU_INFO(i));
1301
1302 if (view->resource) {
1303 struct nv04_resource *res = nv04_resource(view->resource);
1304
1305 if (res->base.target == PIPE_BUFFER) {
1306 if (view->access & PIPE_IMAGE_ACCESS_WRITE)
1307 nvc0_mark_image_range_valid(view);
1308 }
1309
1310 nve4_set_surface_info(push, view, nvc0);
1311 BCTX_REFN(nvc0->bufctx_3d, 3D_SUF, res, RDWR);
1312
1313 if (nvc0->screen->base.class_3d >= GM107_3D_CLASS)
1314 gm107_validate_surfaces(nvc0, view, s, i);
1315 } else {
1316 for (j = 0; j < 16; j++)
1317 PUSH_DATA(push, 0);
1318 }
1319 }
1320 }
1321 }
1322
1323 void
nvc0_validate_surfaces(struct nvc0_context * nvc0)1324 nvc0_validate_surfaces(struct nvc0_context *nvc0)
1325 {
1326 if (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) {
1327 nve4_update_surface_bindings(nvc0);
1328 } else {
1329 nvc0_update_surface_bindings(nvc0);
1330 }
1331 }
1332
1333 static uint64_t
nve4_create_image_handle(struct pipe_context * pipe,const struct pipe_image_view * view)1334 nve4_create_image_handle(struct pipe_context *pipe,
1335 const struct pipe_image_view *view)
1336 {
1337 struct nvc0_context *nvc0 = nvc0_context(pipe);
1338 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
1339 struct nvc0_screen *screen = nvc0->screen;
1340 int i = screen->img.next, s;
1341
1342 while (screen->img.entries[i]) {
1343 i = (i + 1) & (NVE4_IMG_MAX_HANDLES - 1);
1344 if (i == screen->img.next)
1345 return 0;
1346 }
1347
1348 screen->img.next = (i + 1) & (NVE4_IMG_MAX_HANDLES - 1);
1349 screen->img.entries[i] = calloc(1, sizeof(struct pipe_image_view));
1350 *screen->img.entries[i] = *view;
1351
1352 for (s = 0; s < 6; s++) {
1353 BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
1354 PUSH_DATA (push, NVC0_CB_AUX_SIZE);
1355 PUSH_DATAh(push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
1356 PUSH_DATA (push, screen->uniform_bo->offset + NVC0_CB_AUX_INFO(s));
1357 BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 16);
1358 PUSH_DATA (push, NVC0_CB_AUX_BINDLESS_INFO(i));
1359 nve4_set_surface_info(push, view, nvc0);
1360 }
1361
1362 return 0x100000000ULL | i;
1363 }
1364
1365 static void
nve4_delete_image_handle(struct pipe_context * pipe,uint64_t handle)1366 nve4_delete_image_handle(struct pipe_context *pipe, uint64_t handle)
1367 {
1368 struct nvc0_context *nvc0 = nvc0_context(pipe);
1369 struct nvc0_screen *screen = nvc0->screen;
1370 int i = handle & (NVE4_IMG_MAX_HANDLES - 1);
1371
1372 free(screen->img.entries[i]);
1373 screen->img.entries[i] = NULL;
1374 }
1375
1376 static void
nve4_make_image_handle_resident(struct pipe_context * pipe,uint64_t handle,unsigned access,bool resident)1377 nve4_make_image_handle_resident(struct pipe_context *pipe, uint64_t handle,
1378 unsigned access, bool resident)
1379 {
1380 struct nvc0_context *nvc0 = nvc0_context(pipe);
1381 struct nvc0_screen *screen = nvc0->screen;
1382
1383 if (resident) {
1384 struct nvc0_resident *res = calloc(1, sizeof(struct nvc0_resident));
1385 struct pipe_image_view *view =
1386 screen->img.entries[handle & (NVE4_IMG_MAX_HANDLES - 1)];
1387 assert(view);
1388
1389 if (view->resource->target == PIPE_BUFFER &&
1390 access & PIPE_IMAGE_ACCESS_WRITE)
1391 nvc0_mark_image_range_valid(view);
1392 res->handle = handle;
1393 res->buf = nv04_resource(view->resource);
1394 res->flags = (access & 3) << 8;
1395 list_add(&res->list, &nvc0->img_head);
1396 } else {
1397 list_for_each_entry_safe(struct nvc0_resident, pos, &nvc0->img_head, list) {
1398 if (pos->handle == handle) {
1399 list_del(&pos->list);
1400 free(pos);
1401 break;
1402 }
1403 }
1404 }
1405 }
1406
1407 static uint64_t
gm107_create_image_handle(struct pipe_context * pipe,const struct pipe_image_view * view)1408 gm107_create_image_handle(struct pipe_context *pipe,
1409 const struct pipe_image_view *view)
1410 {
1411 /* GM107+ use TIC handles to reference images. As such, image handles are
1412 * just the TIC id.
1413 */
1414 struct nvc0_context *nvc0 = nvc0_context(pipe);
1415 struct nouveau_pushbuf *push = nvc0->base.pushbuf;
1416 struct pipe_sampler_view *sview =
1417 gm107_create_texture_view_from_image(pipe, view);
1418 struct nv50_tic_entry *tic = nv50_tic_entry(sview);
1419
1420 if (tic == NULL)
1421 goto fail;
1422
1423 tic->bindless = 1;
1424 tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);
1425 if (tic->id < 0)
1426 goto fail;
1427
1428 nve4_p2mf_push_linear(&nvc0->base, nvc0->screen->txc, tic->id * 32,
1429 NV_VRAM_DOMAIN(&nvc0->screen->base), 32,
1430 tic->tic);
1431
1432 IMMED_NVC0(push, NVC0_3D(TIC_FLUSH), 0);
1433
1434 nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
1435
1436 // Compute handle. This will include the TIC as well as some additional
1437 // info regarding the bound 3d surface layer, if applicable.
1438 uint64_t handle = 0x100000000ULL | tic->id;
1439 struct nv04_resource *res = nv04_resource(view->resource);
1440 if (res->base.target == PIPE_TEXTURE_3D) {
1441 handle |= 1 << 11;
1442 handle |= view->u.tex.first_layer << (11 + 16);
1443 }
1444 return handle;
1445
1446 fail:
1447 FREE(tic);
1448 return 0;
1449 }
1450
1451 static void
gm107_delete_image_handle(struct pipe_context * pipe,uint64_t handle)1452 gm107_delete_image_handle(struct pipe_context *pipe, uint64_t handle)
1453 {
1454 struct nvc0_context *nvc0 = nvc0_context(pipe);
1455 int tic = handle & NVE4_TIC_ENTRY_INVALID;
1456 struct nv50_tic_entry *entry = nvc0->screen->tic.entries[tic];
1457 struct pipe_sampler_view *view = &entry->pipe;
1458 assert(entry->bindless == 1);
1459 assert(!view_bound(nvc0, view));
1460 entry->bindless = 0;
1461 nvc0_screen_tic_unlock(nvc0->screen, entry);
1462 pipe_sampler_view_reference(&view, NULL);
1463 }
1464
1465 static void
gm107_make_image_handle_resident(struct pipe_context * pipe,uint64_t handle,unsigned access,bool resident)1466 gm107_make_image_handle_resident(struct pipe_context *pipe, uint64_t handle,
1467 unsigned access, bool resident)
1468 {
1469 struct nvc0_context *nvc0 = nvc0_context(pipe);
1470
1471 if (resident) {
1472 struct nvc0_resident *res = calloc(1, sizeof(struct nvc0_resident));
1473 struct nv50_tic_entry *tic =
1474 nvc0->screen->tic.entries[handle & NVE4_TIC_ENTRY_INVALID];
1475 assert(tic);
1476 assert(tic->bindless);
1477
1478 res->handle = handle;
1479 res->buf = nv04_resource(tic->pipe.texture);
1480 res->flags = (access & 3) << 8;
1481 if (res->buf->base.target == PIPE_BUFFER &&
1482 access & PIPE_IMAGE_ACCESS_WRITE)
1483 util_range_add(&res->buf->base, &res->buf->valid_buffer_range,
1484 tic->pipe.u.buf.offset,
1485 tic->pipe.u.buf.offset + tic->pipe.u.buf.size);
1486 list_add(&res->list, &nvc0->img_head);
1487 } else {
1488 list_for_each_entry_safe(struct nvc0_resident, pos, &nvc0->img_head, list) {
1489 if (pos->handle == handle) {
1490 list_del(&pos->list);
1491 free(pos);
1492 break;
1493 }
1494 }
1495 }
1496 }
1497
1498 void
nvc0_init_bindless_functions(struct pipe_context * pipe)1499 nvc0_init_bindless_functions(struct pipe_context *pipe) {
1500 pipe->create_texture_handle = nve4_create_texture_handle;
1501 pipe->delete_texture_handle = nve4_delete_texture_handle;
1502 pipe->make_texture_handle_resident = nve4_make_texture_handle_resident;
1503
1504 if (nvc0_context(pipe)->screen->base.class_3d < GM107_3D_CLASS) {
1505 pipe->create_image_handle = nve4_create_image_handle;
1506 pipe->delete_image_handle = nve4_delete_image_handle;
1507 pipe->make_image_handle_resident = nve4_make_image_handle_resident;
1508 } else {
1509 pipe->create_image_handle = gm107_create_image_handle;
1510 pipe->delete_image_handle = gm107_delete_image_handle;
1511 pipe->make_image_handle_resident = gm107_make_image_handle_resident;
1512 }
1513 }
1514
1515
1516 static const uint8_t nve4_su_format_map[PIPE_FORMAT_COUNT] =
1517 {
1518 [PIPE_FORMAT_R32G32B32A32_FLOAT] = GK104_IMAGE_FORMAT_RGBA32_FLOAT,
1519 [PIPE_FORMAT_R32G32B32A32_SINT] = GK104_IMAGE_FORMAT_RGBA32_SINT,
1520 [PIPE_FORMAT_R32G32B32A32_UINT] = GK104_IMAGE_FORMAT_RGBA32_UINT,
1521 [PIPE_FORMAT_R16G16B16A16_FLOAT] = GK104_IMAGE_FORMAT_RGBA16_FLOAT,
1522 [PIPE_FORMAT_R16G16B16A16_UNORM] = GK104_IMAGE_FORMAT_RGBA16_UNORM,
1523 [PIPE_FORMAT_R16G16B16A16_SNORM] = GK104_IMAGE_FORMAT_RGBA16_SNORM,
1524 [PIPE_FORMAT_R16G16B16A16_SINT] = GK104_IMAGE_FORMAT_RGBA16_SINT,
1525 [PIPE_FORMAT_R16G16B16A16_UINT] = GK104_IMAGE_FORMAT_RGBA16_UINT,
1526 [PIPE_FORMAT_B8G8R8A8_UNORM] = GK104_IMAGE_FORMAT_BGRA8_UNORM,
1527 [PIPE_FORMAT_R8G8B8A8_UNORM] = GK104_IMAGE_FORMAT_RGBA8_UNORM,
1528 [PIPE_FORMAT_R8G8B8A8_SNORM] = GK104_IMAGE_FORMAT_RGBA8_SNORM,
1529 [PIPE_FORMAT_R8G8B8A8_SINT] = GK104_IMAGE_FORMAT_RGBA8_SINT,
1530 [PIPE_FORMAT_R8G8B8A8_UINT] = GK104_IMAGE_FORMAT_RGBA8_UINT,
1531 [PIPE_FORMAT_R11G11B10_FLOAT] = GK104_IMAGE_FORMAT_R11G11B10_FLOAT,
1532 [PIPE_FORMAT_R10G10B10A2_UNORM] = GK104_IMAGE_FORMAT_RGB10_A2_UNORM,
1533 [PIPE_FORMAT_R10G10B10A2_UINT] = GK104_IMAGE_FORMAT_RGB10_A2_UINT,
1534 [PIPE_FORMAT_R32G32_FLOAT] = GK104_IMAGE_FORMAT_RG32_FLOAT,
1535 [PIPE_FORMAT_R32G32_SINT] = GK104_IMAGE_FORMAT_RG32_SINT,
1536 [PIPE_FORMAT_R32G32_UINT] = GK104_IMAGE_FORMAT_RG32_UINT,
1537 [PIPE_FORMAT_R16G16_FLOAT] = GK104_IMAGE_FORMAT_RG16_FLOAT,
1538 [PIPE_FORMAT_R16G16_UNORM] = GK104_IMAGE_FORMAT_RG16_UNORM,
1539 [PIPE_FORMAT_R16G16_SNORM] = GK104_IMAGE_FORMAT_RG16_SNORM,
1540 [PIPE_FORMAT_R16G16_SINT] = GK104_IMAGE_FORMAT_RG16_SINT,
1541 [PIPE_FORMAT_R16G16_UINT] = GK104_IMAGE_FORMAT_RG16_UINT,
1542 [PIPE_FORMAT_R8G8_UNORM] = GK104_IMAGE_FORMAT_RG8_UNORM,
1543 [PIPE_FORMAT_R8G8_SNORM] = GK104_IMAGE_FORMAT_RG8_SNORM,
1544 [PIPE_FORMAT_R8G8_SINT] = GK104_IMAGE_FORMAT_RG8_SINT,
1545 [PIPE_FORMAT_R8G8_UINT] = GK104_IMAGE_FORMAT_RG8_UINT,
1546 [PIPE_FORMAT_R32_FLOAT] = GK104_IMAGE_FORMAT_R32_FLOAT,
1547 [PIPE_FORMAT_R32_SINT] = GK104_IMAGE_FORMAT_R32_SINT,
1548 [PIPE_FORMAT_R32_UINT] = GK104_IMAGE_FORMAT_R32_UINT,
1549 [PIPE_FORMAT_R16_FLOAT] = GK104_IMAGE_FORMAT_R16_FLOAT,
1550 [PIPE_FORMAT_R16_UNORM] = GK104_IMAGE_FORMAT_R16_UNORM,
1551 [PIPE_FORMAT_R16_SNORM] = GK104_IMAGE_FORMAT_R16_SNORM,
1552 [PIPE_FORMAT_R16_SINT] = GK104_IMAGE_FORMAT_R16_SINT,
1553 [PIPE_FORMAT_R16_UINT] = GK104_IMAGE_FORMAT_R16_UINT,
1554 [PIPE_FORMAT_R8_UNORM] = GK104_IMAGE_FORMAT_R8_UNORM,
1555 [PIPE_FORMAT_R8_SNORM] = GK104_IMAGE_FORMAT_R8_SNORM,
1556 [PIPE_FORMAT_R8_SINT] = GK104_IMAGE_FORMAT_R8_SINT,
1557 [PIPE_FORMAT_R8_UINT] = GK104_IMAGE_FORMAT_R8_UINT,
1558 };
1559
1560 /* Auxiliary format description values for surface instructions.
1561 * (log2(bytes per pixel) << 12) | (unk8 << 8) | unk22
1562 */
1563 static const uint16_t nve4_su_format_aux_map[PIPE_FORMAT_COUNT] =
1564 {
1565 [PIPE_FORMAT_R32G32B32A32_FLOAT] = 0x4842,
1566 [PIPE_FORMAT_R32G32B32A32_SINT] = 0x4842,
1567 [PIPE_FORMAT_R32G32B32A32_UINT] = 0x4842,
1568
1569 [PIPE_FORMAT_R16G16B16A16_UNORM] = 0x3933,
1570 [PIPE_FORMAT_R16G16B16A16_SNORM] = 0x3933,
1571 [PIPE_FORMAT_R16G16B16A16_SINT] = 0x3933,
1572 [PIPE_FORMAT_R16G16B16A16_UINT] = 0x3933,
1573 [PIPE_FORMAT_R16G16B16A16_FLOAT] = 0x3933,
1574
1575 [PIPE_FORMAT_R32G32_FLOAT] = 0x3433,
1576 [PIPE_FORMAT_R32G32_SINT] = 0x3433,
1577 [PIPE_FORMAT_R32G32_UINT] = 0x3433,
1578
1579 [PIPE_FORMAT_R10G10B10A2_UNORM] = 0x2a24,
1580 [PIPE_FORMAT_R10G10B10A2_UINT] = 0x2a24,
1581 [PIPE_FORMAT_B8G8R8A8_UNORM] = 0x2a24,
1582 [PIPE_FORMAT_R8G8B8A8_UNORM] = 0x2a24,
1583 [PIPE_FORMAT_R8G8B8A8_SNORM] = 0x2a24,
1584 [PIPE_FORMAT_R8G8B8A8_SINT] = 0x2a24,
1585 [PIPE_FORMAT_R8G8B8A8_UINT] = 0x2a24,
1586 [PIPE_FORMAT_R11G11B10_FLOAT] = 0x2a24,
1587
1588 [PIPE_FORMAT_R16G16_UNORM] = 0x2524,
1589 [PIPE_FORMAT_R16G16_SNORM] = 0x2524,
1590 [PIPE_FORMAT_R16G16_SINT] = 0x2524,
1591 [PIPE_FORMAT_R16G16_UINT] = 0x2524,
1592 [PIPE_FORMAT_R16G16_FLOAT] = 0x2524,
1593
1594 [PIPE_FORMAT_R32_SINT] = 0x2024,
1595 [PIPE_FORMAT_R32_UINT] = 0x2024,
1596 [PIPE_FORMAT_R32_FLOAT] = 0x2024,
1597
1598 [PIPE_FORMAT_R8G8_UNORM] = 0x1615,
1599 [PIPE_FORMAT_R8G8_SNORM] = 0x1615,
1600 [PIPE_FORMAT_R8G8_SINT] = 0x1615,
1601 [PIPE_FORMAT_R8G8_UINT] = 0x1615,
1602
1603 [PIPE_FORMAT_R16_UNORM] = 0x1115,
1604 [PIPE_FORMAT_R16_SNORM] = 0x1115,
1605 [PIPE_FORMAT_R16_SINT] = 0x1115,
1606 [PIPE_FORMAT_R16_UINT] = 0x1115,
1607 [PIPE_FORMAT_R16_FLOAT] = 0x1115,
1608
1609 [PIPE_FORMAT_R8_UNORM] = 0x0206,
1610 [PIPE_FORMAT_R8_SNORM] = 0x0206,
1611 [PIPE_FORMAT_R8_SINT] = 0x0206,
1612 [PIPE_FORMAT_R8_UINT] = 0x0206
1613 };
1614
1615 /* NOTE: These are hardcoded offsets for the shader library.
1616 * TODO: Automate them.
1617 */
1618 static const uint16_t nve4_suldp_lib_offset[PIPE_FORMAT_COUNT] =
1619 {
1620 [PIPE_FORMAT_R32G32B32A32_FLOAT] = 0x218,
1621 [PIPE_FORMAT_R32G32B32A32_SINT] = 0x218,
1622 [PIPE_FORMAT_R32G32B32A32_UINT] = 0x218,
1623 [PIPE_FORMAT_R16G16B16A16_UNORM] = 0x248,
1624 [PIPE_FORMAT_R16G16B16A16_SNORM] = 0x2b8,
1625 [PIPE_FORMAT_R16G16B16A16_SINT] = 0x330,
1626 [PIPE_FORMAT_R16G16B16A16_UINT] = 0x388,
1627 [PIPE_FORMAT_R16G16B16A16_FLOAT] = 0x3d8,
1628 [PIPE_FORMAT_R32G32_FLOAT] = 0x428,
1629 [PIPE_FORMAT_R32G32_SINT] = 0x468,
1630 [PIPE_FORMAT_R32G32_UINT] = 0x468,
1631 [PIPE_FORMAT_R10G10B10A2_UNORM] = 0x4a8,
1632 [PIPE_FORMAT_R10G10B10A2_UINT] = 0x530,
1633 [PIPE_FORMAT_R8G8B8A8_UNORM] = 0x588,
1634 [PIPE_FORMAT_R8G8B8A8_SNORM] = 0x5f8,
1635 [PIPE_FORMAT_R8G8B8A8_SINT] = 0x670,
1636 [PIPE_FORMAT_R8G8B8A8_UINT] = 0x6c8,
1637 [PIPE_FORMAT_B5G6R5_UNORM] = 0x718,
1638 [PIPE_FORMAT_B5G5R5X1_UNORM] = 0x7a0,
1639 [PIPE_FORMAT_R16G16_UNORM] = 0x828,
1640 [PIPE_FORMAT_R16G16_SNORM] = 0x890,
1641 [PIPE_FORMAT_R16G16_SINT] = 0x8f0,
1642 [PIPE_FORMAT_R16G16_UINT] = 0x948,
1643 [PIPE_FORMAT_R16G16_FLOAT] = 0x998,
1644 [PIPE_FORMAT_R32_FLOAT] = 0x9e8,
1645 [PIPE_FORMAT_R32_SINT] = 0xa30,
1646 [PIPE_FORMAT_R32_UINT] = 0xa30,
1647 [PIPE_FORMAT_R8G8_UNORM] = 0xa78,
1648 [PIPE_FORMAT_R8G8_SNORM] = 0xae0,
1649 [PIPE_FORMAT_R8G8_UINT] = 0xb48,
1650 [PIPE_FORMAT_R8G8_SINT] = 0xb98,
1651 [PIPE_FORMAT_R16_UNORM] = 0xbe8,
1652 [PIPE_FORMAT_R16_SNORM] = 0xc48,
1653 [PIPE_FORMAT_R16_SINT] = 0xca0,
1654 [PIPE_FORMAT_R16_UINT] = 0xce8,
1655 [PIPE_FORMAT_R16_FLOAT] = 0xd30,
1656 [PIPE_FORMAT_R8_UNORM] = 0xd88,
1657 [PIPE_FORMAT_R8_SNORM] = 0xde0,
1658 [PIPE_FORMAT_R8_SINT] = 0xe38,
1659 [PIPE_FORMAT_R8_UINT] = 0xe88,
1660 [PIPE_FORMAT_R11G11B10_FLOAT] = 0xed0
1661 };
1662