05daf1cec4344b33b3210ce3669308b534c2149b
[mesa.git] / src / mesa / drivers / dri / radeon / radeon_mipmap_tree.c
1 /*
2 * Copyright (C) 2009 Maciej Cencora.
3 * Copyright (C) 2008 Nicolai Haehnle.
4 *
5 * All Rights Reserved.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining
8 * a copy of this software and associated documentation files (the
9 * "Software"), to deal in the Software without restriction, including
10 * without limitation the rights to use, copy, modify, merge, publish,
11 * distribute, sublicense, and/or sell copies of the Software, and to
12 * permit persons to whom the Software is furnished to do so, subject to
13 * the following conditions:
14 *
15 * The above copyright notice and this permission notice (including the
16 * next paragraph) shall be included in all copies or substantial
17 * portions of the Software.
18 *
19 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
20 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
21 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
22 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
23 * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
24 * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
25 * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26 *
27 */
28
29 #include "radeon_mipmap_tree.h"
30
31 #include <errno.h>
32 #include <unistd.h>
33
34 #include "main/simple_list.h"
35 #include "main/teximage.h"
36 #include "main/texobj.h"
37 #include "main/enums.h"
38 #include "radeon_texture.h"
39 #include "radeon_tile.h"
40
41 static unsigned get_aligned_compressed_row_stride(
42 gl_format format,
43 unsigned width,
44 unsigned minStride)
45 {
46 const unsigned blockBytes = _mesa_get_format_bytes(format);
47 unsigned blockWidth, blockHeight;
48 unsigned stride;
49
50 _mesa_get_format_block_size(format, &blockWidth, &blockHeight);
51
52 /* Count number of blocks required to store the given width.
53 * And then multiple it with bytes required to store a block.
54 */
55 stride = (width + blockWidth - 1) / blockWidth * blockBytes;
56
57 /* Round the given minimum stride to the next full blocksize.
58 * (minStride + blockBytes - 1) / blockBytes * blockBytes
59 */
60 if ( stride < minStride )
61 stride = (minStride + blockBytes - 1) / blockBytes * blockBytes;
62
63 radeon_print(RADEON_TEXTURE, RADEON_TRACE,
64 "%s width %u, minStride %u, block(bytes %u, width %u):"
65 "stride %u\n",
66 __func__, width, minStride,
67 blockBytes, blockWidth,
68 stride);
69
70 return stride;
71 }
72
73 unsigned get_texture_image_size(
74 gl_format format,
75 unsigned rowStride,
76 unsigned height,
77 unsigned depth,
78 unsigned tiling)
79 {
80 if (_mesa_is_format_compressed(format)) {
81 unsigned blockWidth, blockHeight;
82
83 _mesa_get_format_block_size(format, &blockWidth, &blockHeight);
84
85 return rowStride * ((height + blockHeight - 1) / blockHeight) * depth;
86 } else if (tiling) {
87 /* Need to align height to tile height */
88 unsigned tileWidth, tileHeight;
89
90 get_tile_size(format, &tileWidth, &tileHeight);
91 tileHeight--;
92
93 height = (height + tileHeight) & ~tileHeight;
94 }
95
96 return rowStride * height * depth;
97 }
98
99 unsigned get_texture_image_row_stride(radeonContextPtr rmesa, gl_format format, unsigned width, unsigned tiling, GLuint target)
100 {
101 if (_mesa_is_format_compressed(format)) {
102 return get_aligned_compressed_row_stride(format, width, rmesa->texture_compressed_row_align);
103 } else {
104 unsigned row_align;
105
106 if (!_mesa_is_pow_two(width) || target == GL_TEXTURE_RECTANGLE) {
107 row_align = rmesa->texture_rect_row_align - 1;
108 } else if (tiling) {
109 unsigned tileWidth, tileHeight;
110 get_tile_size(format, &tileWidth, &tileHeight);
111 row_align = tileWidth * _mesa_get_format_bytes(format) - 1;
112 } else {
113 row_align = rmesa->texture_row_align - 1;
114 }
115
116 return (_mesa_format_row_stride(format, width) + row_align) & ~row_align;
117 }
118 }
119
120 /**
121 * Compute sizes and fill in offset and blit information for the given
122 * image (determined by \p face and \p level).
123 *
124 * \param curOffset points to the offset at which the image is to be stored
125 * and is updated by this function according to the size of the image.
126 */
127 static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree *mt,
128 GLuint face, GLuint level, GLuint* curOffset)
129 {
130 radeon_mipmap_level *lvl = &mt->levels[level];
131 GLuint height;
132
133 height = _mesa_next_pow_two_32(lvl->height);
134
135 lvl->rowstride = get_texture_image_row_stride(rmesa, mt->mesaFormat, lvl->width, mt->tilebits, mt->target);
136 lvl->size = get_texture_image_size(mt->mesaFormat, lvl->rowstride, height, lvl->depth, mt->tilebits);
137
138 assert(lvl->size > 0);
139
140 lvl->faces[face].offset = *curOffset;
141 *curOffset += lvl->size;
142
143 radeon_print(RADEON_TEXTURE, RADEON_TRACE,
144 "%s(%p) level %d, face %d: rs:%d %dx%d at %d\n",
145 __func__, rmesa,
146 level, face,
147 lvl->rowstride, lvl->width, height, lvl->faces[face].offset);
148 }
149
150 static GLuint minify(GLuint size, GLuint levels)
151 {
152 size = size >> levels;
153 if (size < 1)
154 size = 1;
155 return size;
156 }
157
158
159 static void calculate_miptree_layout(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
160 {
161 GLuint curOffset, i, face, level;
162
163 assert(mt->numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
164
165 curOffset = 0;
166 for(face = 0; face < mt->faces; face++) {
167
168 for(i = 0, level = mt->baseLevel; i < mt->numLevels; i++, level++) {
169 mt->levels[level].valid = 1;
170 mt->levels[level].width = minify(mt->width0, i);
171 mt->levels[level].height = minify(mt->height0, i);
172 mt->levels[level].depth = minify(mt->depth0, i);
173 compute_tex_image_offset(rmesa, mt, face, level, &curOffset);
174 }
175 }
176
177 /* Note the required size in memory */
178 mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
179
180 radeon_print(RADEON_TEXTURE, RADEON_TRACE,
181 "%s(%p, %p) total size %d\n",
182 __func__, rmesa, mt, mt->totalsize);
183 }
184
185 /**
186 * Create a new mipmap tree, calculate its layout and allocate memory.
187 */
188 static radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa,
189 GLenum target, gl_format mesaFormat, GLuint baseLevel, GLuint numLevels,
190 GLuint width0, GLuint height0, GLuint depth0, GLuint tilebits)
191 {
192 radeon_mipmap_tree *mt = CALLOC_STRUCT(_radeon_mipmap_tree);
193
194 radeon_print(RADEON_TEXTURE, RADEON_NORMAL,
195 "%s(%p) new tree is %p.\n",
196 __func__, rmesa, mt);
197
198 mt->mesaFormat = mesaFormat;
199 mt->refcount = 1;
200 mt->target = target;
201 mt->faces = (target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
202 mt->baseLevel = baseLevel;
203 mt->numLevels = numLevels;
204 mt->width0 = width0;
205 mt->height0 = height0;
206 mt->depth0 = depth0;
207 mt->tilebits = tilebits;
208
209 calculate_miptree_layout(rmesa, mt);
210
211 mt->bo = radeon_bo_open(rmesa->radeonScreen->bom,
212 0, mt->totalsize, 1024,
213 RADEON_GEM_DOMAIN_VRAM,
214 0);
215
216 return mt;
217 }
218
219 void radeon_miptree_reference(radeon_mipmap_tree *mt, radeon_mipmap_tree **ptr)
220 {
221 assert(!*ptr);
222
223 mt->refcount++;
224 assert(mt->refcount > 0);
225
226 *ptr = mt;
227 }
228
229 void radeon_miptree_unreference(radeon_mipmap_tree **ptr)
230 {
231 radeon_mipmap_tree *mt = *ptr;
232 if (!mt)
233 return;
234
235 assert(mt->refcount > 0);
236
237 mt->refcount--;
238 if (!mt->refcount) {
239 radeon_bo_unref(mt->bo);
240 free(mt);
241 }
242
243 *ptr = 0;
244 }
245
246 /**
247 * Calculate min and max LOD for the given texture object.
248 * @param[in] tObj texture object whose LOD values to calculate
249 * @param[out] pminLod minimal LOD
250 * @param[out] pmaxLod maximal LOD
251 */
252 static void calculate_min_max_lod(struct gl_texture_object *tObj,
253 unsigned *pminLod, unsigned *pmaxLod)
254 {
255 int minLod, maxLod;
256 /* Yes, this looks overly complicated, but it's all needed.
257 */
258 switch (tObj->Target) {
259 case GL_TEXTURE_1D:
260 case GL_TEXTURE_2D:
261 case GL_TEXTURE_3D:
262 case GL_TEXTURE_CUBE_MAP:
263 if (tObj->Sampler.MinFilter == GL_NEAREST || tObj->Sampler.MinFilter == GL_LINEAR) {
264 /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
265 */
266 minLod = maxLod = tObj->BaseLevel;
267 } else {
268 minLod = tObj->BaseLevel + (GLint)(tObj->Sampler.MinLod);
269 minLod = MAX2(minLod, tObj->BaseLevel);
270 minLod = MIN2(minLod, tObj->MaxLevel);
271 maxLod = tObj->BaseLevel + (GLint)(tObj->Sampler.MaxLod + 0.5);
272 maxLod = MIN2(maxLod, tObj->MaxLevel);
273 maxLod = MIN2(maxLod, tObj->Image[0][minLod]->MaxLog2 + minLod);
274 maxLod = MAX2(maxLod, minLod); /* need at least one level */
275 }
276 break;
277 case GL_TEXTURE_RECTANGLE_NV:
278 case GL_TEXTURE_4D_SGIS:
279 minLod = maxLod = 0;
280 break;
281 default:
282 return;
283 }
284
285 radeon_print(RADEON_TEXTURE, RADEON_TRACE,
286 "%s(%p) target %s, min %d, max %d.\n",
287 __func__, tObj,
288 _mesa_lookup_enum_by_nr(tObj->Target),
289 minLod, maxLod);
290
291 /* save these values */
292 *pminLod = minLod;
293 *pmaxLod = maxLod;
294 }
295
296 /**
297 * Checks whether the given miptree can hold the given texture image at the
298 * given face and level.
299 */
300 GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
301 struct gl_texture_image *texImage, GLuint face, GLuint level)
302 {
303 radeon_mipmap_level *lvl;
304
305 if (face >= mt->faces)
306 return GL_FALSE;
307
308 if (texImage->TexFormat != mt->mesaFormat)
309 return GL_FALSE;
310
311 lvl = &mt->levels[level];
312 if (!lvl->valid ||
313 lvl->width != texImage->Width ||
314 lvl->height != texImage->Height ||
315 lvl->depth != texImage->Depth)
316 return GL_FALSE;
317
318 return GL_TRUE;
319 }
320
321 /**
322 * Checks whether the given miptree has the right format to store the given texture object.
323 */
324 static GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj)
325 {
326 struct gl_texture_image *firstImage;
327 unsigned numLevels;
328 radeon_mipmap_level *mtBaseLevel;
329
330 if (texObj->BaseLevel < mt->baseLevel)
331 return GL_FALSE;
332
333 mtBaseLevel = &mt->levels[texObj->BaseLevel - mt->baseLevel];
334 firstImage = texObj->Image[0][texObj->BaseLevel];
335 numLevels = MIN2(texObj->MaxLevel - texObj->BaseLevel + 1, firstImage->MaxLog2 + 1);
336
337 if (radeon_is_debug_enabled(RADEON_TEXTURE,RADEON_TRACE)) {
338 fprintf(stderr, "Checking if miptree %p matches texObj %p\n", mt, texObj);
339 fprintf(stderr, "target %d vs %d\n", mt->target, texObj->Target);
340 fprintf(stderr, "format %d vs %d\n", mt->mesaFormat, firstImage->TexFormat);
341 fprintf(stderr, "numLevels %d vs %d\n", mt->numLevels, numLevels);
342 fprintf(stderr, "width0 %d vs %d\n", mtBaseLevel->width, firstImage->Width);
343 fprintf(stderr, "height0 %d vs %d\n", mtBaseLevel->height, firstImage->Height);
344 fprintf(stderr, "depth0 %d vs %d\n", mtBaseLevel->depth, firstImage->Depth);
345 if (mt->target == texObj->Target &&
346 mt->mesaFormat == firstImage->TexFormat &&
347 mt->numLevels >= numLevels &&
348 mtBaseLevel->width == firstImage->Width &&
349 mtBaseLevel->height == firstImage->Height &&
350 mtBaseLevel->depth == firstImage->Depth) {
351 fprintf(stderr, "MATCHED\n");
352 } else {
353 fprintf(stderr, "NOT MATCHED\n");
354 }
355 }
356
357 return (mt->target == texObj->Target &&
358 mt->mesaFormat == firstImage->TexFormat &&
359 mt->numLevels >= numLevels &&
360 mtBaseLevel->width == firstImage->Width &&
361 mtBaseLevel->height == firstImage->Height &&
362 mtBaseLevel->depth == firstImage->Depth);
363 }
364
365 /**
366 * Try to allocate a mipmap tree for the given texture object.
367 * @param[in] rmesa radeon context
368 * @param[in] t radeon texture object
369 */
370 void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t)
371 {
372 struct gl_texture_object *texObj = &t->base;
373 struct gl_texture_image *texImg = texObj->Image[0][texObj->BaseLevel];
374 GLuint numLevels;
375
376 assert(!t->mt);
377
378 if (!texImg) {
379 radeon_warning("%s(%p) No image in given texture object(%p).\n",
380 __func__, rmesa, t);
381 return;
382 }
383
384
385 numLevels = MIN2(texObj->MaxLevel - texObj->BaseLevel + 1, texImg->MaxLog2 + 1);
386
387 t->mt = radeon_miptree_create(rmesa, t->base.Target,
388 texImg->TexFormat, texObj->BaseLevel,
389 numLevels, texImg->Width, texImg->Height,
390 texImg->Depth, t->tile_bits);
391 }
392
393 GLuint
394 radeon_miptree_image_offset(radeon_mipmap_tree *mt,
395 GLuint face, GLuint level)
396 {
397 if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
398 return (mt->levels[level].faces[face].offset);
399 else
400 return mt->levels[level].faces[0].offset;
401 }
402
403 /**
404 * Ensure that the given image is stored in the given miptree from now on.
405 */
406 static void migrate_image_to_miptree(radeon_mipmap_tree *mt,
407 radeon_texture_image *image,
408 int face, int level)
409 {
410 radeon_mipmap_level *dstlvl = &mt->levels[level];
411 unsigned char *dest;
412
413 assert(image->mt != mt);
414 assert(dstlvl->valid);
415 assert(dstlvl->width == image->base.Base.Width);
416 assert(dstlvl->height == image->base.Base.Height);
417 assert(dstlvl->depth == image->base.Base.Depth);
418
419 radeon_print(RADEON_TEXTURE, RADEON_VERBOSE,
420 "%s miptree %p, image %p, face %d, level %d.\n",
421 __func__, mt, image, face, level);
422
423 radeon_bo_map(mt->bo, GL_TRUE);
424 dest = mt->bo->ptr + dstlvl->faces[face].offset;
425
426 if (image->mt) {
427 /* Format etc. should match, so we really just need a memcpy().
428 * In fact, that memcpy() could be done by the hardware in many
429 * cases, provided that we have a proper memory manager.
430 */
431 assert(mt->mesaFormat == image->base.Base.TexFormat);
432
433 radeon_mipmap_level *srclvl = &image->mt->levels[image->base.Base.Level];
434
435 assert(image->base.Base.Level == level);
436 assert(srclvl->size == dstlvl->size);
437 assert(srclvl->rowstride == dstlvl->rowstride);
438
439 radeon_bo_map(image->mt->bo, GL_FALSE);
440
441 memcpy(dest,
442 image->mt->bo->ptr + srclvl->faces[face].offset,
443 dstlvl->size);
444 radeon_bo_unmap(image->mt->bo);
445
446 radeon_miptree_unreference(&image->mt);
447 } else if (image->base.Data) {
448 /* This condition should be removed, it's here to workaround
449 * a segfault when mapping textures during software fallbacks.
450 */
451 radeon_print(RADEON_FALLBACKS, RADEON_IMPORTANT,
452 "%s Trying to map texture in sowftware fallback.\n",
453 __func__);
454 const uint32_t srcrowstride = _mesa_format_row_stride(image->base.Base.TexFormat, image->base.Base.Width);
455 uint32_t rows = image->base.Base.Height * image->base.Base.Depth;
456
457 if (_mesa_is_format_compressed(image->base.Base.TexFormat)) {
458 uint32_t blockWidth, blockHeight;
459 _mesa_get_format_block_size(image->base.Base.TexFormat, &blockWidth, &blockHeight);
460 rows = (rows + blockHeight - 1) / blockHeight;
461 }
462
463 copy_rows(dest, dstlvl->rowstride, image->base.Data, srcrowstride,
464 rows, srcrowstride);
465
466 _mesa_align_free(image->base.Data);
467 image->base.Data = 0;
468 }
469
470 radeon_bo_unmap(mt->bo);
471
472 radeon_miptree_reference(mt, &image->mt);
473 }
474
475 /**
476 * Filter matching miptrees, and select one with the most of data.
477 * @param[in] texObj radeon texture object
478 * @param[in] firstLevel first texture level to check
479 * @param[in] lastLevel last texture level to check
480 */
481 static radeon_mipmap_tree * get_biggest_matching_miptree(radeonTexObj *texObj,
482 unsigned firstLevel,
483 unsigned lastLevel)
484 {
485 const unsigned numLevels = lastLevel - firstLevel + 1;
486 unsigned *mtSizes = calloc(numLevels, sizeof(unsigned));
487 radeon_mipmap_tree **mts = calloc(numLevels, sizeof(radeon_mipmap_tree *));
488 unsigned mtCount = 0;
489 unsigned maxMtIndex = 0;
490 radeon_mipmap_tree *tmp;
491 unsigned int level;
492 int i;
493
494 for (level = firstLevel; level <= lastLevel; ++level) {
495 radeon_texture_image *img = get_radeon_texture_image(texObj->base.Image[0][level]);
496 unsigned found = 0;
497 // TODO: why this hack??
498 if (!img)
499 break;
500
501 if (!img->mt)
502 continue;
503
504 for (i = 0; i < mtCount; ++i) {
505 if (mts[i] == img->mt) {
506 found = 1;
507 mtSizes[i] += img->mt->levels[img->base.Base.Level].size;
508 break;
509 }
510 }
511
512 if (!found && radeon_miptree_matches_texture(img->mt, &texObj->base)) {
513 mtSizes[mtCount] = img->mt->levels[img->base.Base.Level].size;
514 mts[mtCount] = img->mt;
515 mtCount++;
516 }
517 }
518
519 if (mtCount == 0) {
520 free(mtSizes);
521 free(mts);
522 return NULL;
523 }
524
525 for (i = 1; i < mtCount; ++i) {
526 if (mtSizes[i] > mtSizes[maxMtIndex]) {
527 maxMtIndex = i;
528 }
529 }
530
531 tmp = mts[maxMtIndex];
532 free(mtSizes);
533 free(mts);
534
535 return tmp;
536 }
537
538 /**
539 * Validate texture mipmap tree.
540 * If individual images are stored in different mipmap trees
541 * use the mipmap tree that has the most of the correct data.
542 */
543 int radeon_validate_texture_miptree(struct gl_context * ctx, struct gl_texture_object *texObj)
544 {
545 radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
546 radeonTexObj *t = radeon_tex_obj(texObj);
547
548 if (t->validated || t->image_override) {
549 return GL_TRUE;
550 }
551
552 if (texObj->Image[0][texObj->BaseLevel]->Border > 0)
553 return GL_FALSE;
554
555 _mesa_test_texobj_completeness(rmesa->glCtx, texObj);
556 if (!texObj->_Complete) {
557 return GL_FALSE;
558 }
559
560 calculate_min_max_lod(&t->base, &t->minLod, &t->maxLod);
561
562 radeon_print(RADEON_TEXTURE, RADEON_NORMAL,
563 "%s: Validating texture %p now, minLod = %d, maxLod = %d\n",
564 __FUNCTION__, texObj ,t->minLod, t->maxLod);
565
566 radeon_mipmap_tree *dst_miptree;
567 dst_miptree = get_biggest_matching_miptree(t, t->base.BaseLevel, t->base.MaxLevel);
568
569 radeon_miptree_unreference(&t->mt);
570 if (!dst_miptree) {
571 radeon_try_alloc_miptree(rmesa, t);
572 radeon_print(RADEON_TEXTURE, RADEON_NORMAL,
573 "%s: No matching miptree found, allocated new one %p\n",
574 __FUNCTION__, t->mt);
575
576 } else {
577 radeon_miptree_reference(dst_miptree, &t->mt);
578 radeon_print(RADEON_TEXTURE, RADEON_NORMAL,
579 "%s: Using miptree %p\n", __FUNCTION__, t->mt);
580 }
581
582 const unsigned faces = texObj->Target == GL_TEXTURE_CUBE_MAP ? 6 : 1;
583 unsigned face, level;
584 radeon_texture_image *img;
585 /* Validate only the levels that will actually be used during rendering */
586 for (face = 0; face < faces; ++face) {
587 for (level = t->minLod; level <= t->maxLod; ++level) {
588 img = get_radeon_texture_image(texObj->Image[face][level]);
589
590 radeon_print(RADEON_TEXTURE, RADEON_TRACE,
591 "Checking image level %d, face %d, mt %p ... ",
592 level, face, img->mt);
593
594 if (img->mt != t->mt) {
595 radeon_print(RADEON_TEXTURE, RADEON_TRACE,
596 "MIGRATING\n");
597
598 struct radeon_bo *src_bo = (img->mt) ? img->mt->bo : img->bo;
599 if (src_bo && radeon_bo_is_referenced_by_cs(src_bo, rmesa->cmdbuf.cs)) {
600 radeon_firevertices(rmesa);
601 }
602 migrate_image_to_miptree(t->mt, img, face, level);
603 } else
604 radeon_print(RADEON_TEXTURE, RADEON_TRACE, "OK\n");
605 }
606 }
607
608 t->validated = GL_TRUE;
609
610 return GL_TRUE;
611 }
612
613 uint32_t get_base_teximage_offset(radeonTexObj *texObj)
614 {
615 if (!texObj->mt) {
616 return 0;
617 } else {
618 return radeon_miptree_image_offset(texObj->mt, 0, texObj->minLod);
619 }
620 }