Merge branch 'mesa_7_6_branch' into mesa_7_7_branch
[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/texcompress.h"
36 #include "main/teximage.h"
37 #include "main/texobj.h"
38 #include "radeon_texture.h"
39
40 static unsigned get_aligned_compressed_row_stride(
41 gl_format format,
42 unsigned width,
43 unsigned minStride)
44 {
45 const unsigned blockSize = _mesa_get_format_bytes(format);
46 unsigned blockWidth, blockHeight, numXBlocks;
47
48 _mesa_get_format_block_size(format, &blockWidth, &blockHeight);
49 numXBlocks = (width + blockWidth - 1) / blockWidth;
50
51 while (numXBlocks * blockSize < minStride)
52 {
53 ++numXBlocks;
54 }
55
56 return numXBlocks * blockSize;
57 }
58
59 static unsigned get_compressed_image_size(
60 gl_format format,
61 unsigned rowStride,
62 unsigned height)
63 {
64 unsigned blockWidth, blockHeight;
65
66 _mesa_get_format_block_size(format, &blockWidth, &blockHeight);
67
68 return rowStride * ((height + blockHeight - 1) / blockHeight);
69 }
70
71 /**
72 * Compute sizes and fill in offset and blit information for the given
73 * image (determined by \p face and \p level).
74 *
75 * \param curOffset points to the offset at which the image is to be stored
76 * and is updated by this function according to the size of the image.
77 */
78 static void compute_tex_image_offset(radeonContextPtr rmesa, radeon_mipmap_tree *mt,
79 GLuint face, GLuint level, GLuint* curOffset)
80 {
81 radeon_mipmap_level *lvl = &mt->levels[level];
82 uint32_t row_align;
83
84 /* Find image size in bytes */
85 if (_mesa_is_format_compressed(mt->mesaFormat)) {
86 lvl->rowstride = get_aligned_compressed_row_stride(mt->mesaFormat, lvl->width, rmesa->texture_compressed_row_align);
87 lvl->size = get_compressed_image_size(mt->mesaFormat, lvl->rowstride, lvl->height);
88 } else if (mt->target == GL_TEXTURE_RECTANGLE_NV) {
89 row_align = rmesa->texture_rect_row_align - 1;
90 lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
91 lvl->size = lvl->rowstride * lvl->height;
92 } else if (mt->tilebits & RADEON_TXO_MICRO_TILE) {
93 /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
94 * though the actual offset may be different (if texture is less than
95 * 32 bytes width) to the untiled case */
96 lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) * 2 + 31) & ~31;
97 lvl->size = lvl->rowstride * ((lvl->height + 1) / 2) * lvl->depth;
98 } else {
99 row_align = rmesa->texture_row_align - 1;
100 lvl->rowstride = (_mesa_format_row_stride(mt->mesaFormat, lvl->width) + row_align) & ~row_align;
101 lvl->size = lvl->rowstride * lvl->height * lvl->depth;
102 }
103 assert(lvl->size > 0);
104
105 /* All images are aligned to a 32-byte offset */
106 *curOffset = (*curOffset + 0x1f) & ~0x1f;
107 lvl->faces[face].offset = *curOffset;
108 *curOffset += lvl->size;
109
110 if (RADEON_DEBUG & RADEON_TEXTURE)
111 fprintf(stderr,
112 "level %d, face %d: rs:%d %dx%d at %d\n",
113 level, face, lvl->rowstride, lvl->width, lvl->height, lvl->faces[face].offset);
114 }
115
116 static GLuint minify(GLuint size, GLuint levels)
117 {
118 size = size >> levels;
119 if (size < 1)
120 size = 1;
121 return size;
122 }
123
124
125 static void calculate_miptree_layout_r100(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
126 {
127 GLuint curOffset, i, face, level;
128
129 assert(mt->numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
130
131 curOffset = 0;
132 for(face = 0; face < mt->faces; face++) {
133
134 for(i = 0, level = mt->baseLevel; i < mt->numLevels; i++, level++) {
135 mt->levels[level].valid = 1;
136 mt->levels[level].width = minify(mt->width0, i);
137 mt->levels[level].height = minify(mt->height0, i);
138 mt->levels[level].depth = minify(mt->depth0, i);
139 compute_tex_image_offset(rmesa, mt, face, level, &curOffset);
140 }
141 }
142
143 /* Note the required size in memory */
144 mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
145 }
146
147 static void calculate_miptree_layout_r300(radeonContextPtr rmesa, radeon_mipmap_tree *mt)
148 {
149 GLuint curOffset, i, level;
150
151 assert(mt->numLevels <= rmesa->glCtx->Const.MaxTextureLevels);
152
153 curOffset = 0;
154 for(i = 0, level = mt->baseLevel; i < mt->numLevels; i++, level++) {
155 GLuint face;
156
157 mt->levels[level].valid = 1;
158 mt->levels[level].width = minify(mt->width0, i);
159 mt->levels[level].height = minify(mt->height0, i);
160 mt->levels[level].depth = minify(mt->depth0, i);
161
162 for(face = 0; face < mt->faces; face++)
163 compute_tex_image_offset(rmesa, mt, face, level, &curOffset);
164 }
165
166 /* Note the required size in memory */
167 mt->totalsize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
168 }
169
170 /**
171 * Create a new mipmap tree, calculate its layout and allocate memory.
172 */
173 static radeon_mipmap_tree* radeon_miptree_create(radeonContextPtr rmesa,
174 GLenum target, gl_format mesaFormat, GLuint baseLevel, GLuint numLevels,
175 GLuint width0, GLuint height0, GLuint depth0, GLuint tilebits)
176 {
177 radeon_mipmap_tree *mt = CALLOC_STRUCT(_radeon_mipmap_tree);
178
179 mt->mesaFormat = mesaFormat;
180 mt->refcount = 1;
181 mt->target = target;
182 mt->faces = (target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
183 mt->baseLevel = baseLevel;
184 mt->numLevels = numLevels;
185 mt->width0 = width0;
186 mt->height0 = height0;
187 mt->depth0 = depth0;
188 mt->tilebits = tilebits;
189
190 if (rmesa->radeonScreen->chip_family >= CHIP_FAMILY_R300)
191 calculate_miptree_layout_r300(rmesa, mt);
192 else
193 calculate_miptree_layout_r100(rmesa, mt);
194
195 mt->bo = radeon_bo_open(rmesa->radeonScreen->bom,
196 0, mt->totalsize, 1024,
197 RADEON_GEM_DOMAIN_VRAM,
198 0);
199
200 return mt;
201 }
202
203 void radeon_miptree_reference(radeon_mipmap_tree *mt, radeon_mipmap_tree **ptr)
204 {
205 assert(!*ptr);
206
207 mt->refcount++;
208 assert(mt->refcount > 0);
209
210 *ptr = mt;
211 }
212
213 void radeon_miptree_unreference(radeon_mipmap_tree **ptr)
214 {
215 radeon_mipmap_tree *mt = *ptr;
216 if (!mt)
217 return;
218
219 assert(mt->refcount > 0);
220
221 mt->refcount--;
222 if (!mt->refcount) {
223 radeon_bo_unref(mt->bo);
224 free(mt);
225 }
226
227 *ptr = 0;
228 }
229
230 /**
231 * Calculate min and max LOD for the given texture object.
232 * @param[in] tObj texture object whose LOD values to calculate
233 * @param[out] pminLod minimal LOD
234 * @param[out] pmaxLod maximal LOD
235 */
236 static void calculate_min_max_lod(struct gl_texture_object *tObj,
237 unsigned *pminLod, unsigned *pmaxLod)
238 {
239 int minLod, maxLod;
240 /* Yes, this looks overly complicated, but it's all needed.
241 */
242 switch (tObj->Target) {
243 case GL_TEXTURE_1D:
244 case GL_TEXTURE_2D:
245 case GL_TEXTURE_3D:
246 case GL_TEXTURE_CUBE_MAP:
247 if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) {
248 /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
249 */
250 minLod = maxLod = tObj->BaseLevel;
251 } else {
252 minLod = tObj->BaseLevel + (GLint)(tObj->MinLod);
253 minLod = MAX2(minLod, tObj->BaseLevel);
254 minLod = MIN2(minLod, tObj->MaxLevel);
255 maxLod = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5);
256 maxLod = MIN2(maxLod, tObj->MaxLevel);
257 maxLod = MIN2(maxLod, tObj->Image[0][minLod]->MaxLog2 + minLod);
258 maxLod = MAX2(maxLod, minLod); /* need at least one level */
259 }
260 break;
261 case GL_TEXTURE_RECTANGLE_NV:
262 case GL_TEXTURE_4D_SGIS:
263 minLod = maxLod = 0;
264 break;
265 default:
266 return;
267 }
268
269 /* save these values */
270 *pminLod = minLod;
271 *pmaxLod = maxLod;
272 }
273
274 /**
275 * Checks whether the given miptree can hold the given texture image at the
276 * given face and level.
277 */
278 GLboolean radeon_miptree_matches_image(radeon_mipmap_tree *mt,
279 struct gl_texture_image *texImage, GLuint face, GLuint level)
280 {
281 radeon_mipmap_level *lvl;
282
283 if (face >= mt->faces)
284 return GL_FALSE;
285
286 if (texImage->TexFormat != mt->mesaFormat)
287 return GL_FALSE;
288
289 lvl = &mt->levels[level];
290 if (!lvl->valid ||
291 lvl->width != texImage->Width ||
292 lvl->height != texImage->Height ||
293 lvl->depth != texImage->Depth)
294 return GL_FALSE;
295
296 return GL_TRUE;
297 }
298
299 /**
300 * Checks whether the given miptree has the right format to store the given texture object.
301 */
302 static GLboolean radeon_miptree_matches_texture(radeon_mipmap_tree *mt, struct gl_texture_object *texObj)
303 {
304 struct gl_texture_image *firstImage;
305 unsigned numLevels;
306 radeon_mipmap_level *mtBaseLevel;
307
308 if (texObj->BaseLevel < mt->baseLevel)
309 return GL_FALSE;
310
311 mtBaseLevel = &mt->levels[texObj->BaseLevel - mt->baseLevel];
312 firstImage = texObj->Image[0][texObj->BaseLevel];
313 numLevels = MIN2(texObj->MaxLevel - texObj->BaseLevel + 1, firstImage->MaxLog2 + 1);
314
315 if (RADEON_DEBUG & RADEON_TEXTURE) {
316 fprintf(stderr, "Checking if miptree %p matches texObj %p\n", mt, texObj);
317 fprintf(stderr, "target %d vs %d\n", mt->target, texObj->Target);
318 fprintf(stderr, "format %d vs %d\n", mt->mesaFormat, firstImage->TexFormat);
319 fprintf(stderr, "numLevels %d vs %d\n", mt->numLevels, numLevels);
320 fprintf(stderr, "width0 %d vs %d\n", mtBaseLevel->width, firstImage->Width);
321 fprintf(stderr, "height0 %d vs %d\n", mtBaseLevel->height, firstImage->Height);
322 fprintf(stderr, "depth0 %d vs %d\n", mtBaseLevel->depth, firstImage->Depth);
323 if (mt->target == texObj->Target &&
324 mt->mesaFormat == firstImage->TexFormat &&
325 mt->numLevels >= numLevels &&
326 mtBaseLevel->width == firstImage->Width &&
327 mtBaseLevel->height == firstImage->Height &&
328 mtBaseLevel->depth == firstImage->Depth) {
329 fprintf(stderr, "MATCHED\n");
330 } else {
331 fprintf(stderr, "NOT MATCHED\n");
332 }
333 }
334
335 return (mt->target == texObj->Target &&
336 mt->mesaFormat == firstImage->TexFormat &&
337 mt->numLevels >= numLevels &&
338 mtBaseLevel->width == firstImage->Width &&
339 mtBaseLevel->height == firstImage->Height &&
340 mtBaseLevel->depth == firstImage->Depth);
341 }
342
343 /**
344 * Try to allocate a mipmap tree for the given texture object.
345 * @param[in] rmesa radeon context
346 * @param[in] t radeon texture object
347 */
348 void radeon_try_alloc_miptree(radeonContextPtr rmesa, radeonTexObj *t)
349 {
350 struct gl_texture_object *texObj = &t->base;
351 struct gl_texture_image *texImg = texObj->Image[0][texObj->BaseLevel];
352 GLuint numLevels;
353
354 assert(!t->mt);
355
356 if (!texImg)
357 return;
358
359 numLevels = MIN2(texObj->MaxLevel - texObj->BaseLevel + 1, texImg->MaxLog2 + 1);
360
361 t->mt = radeon_miptree_create(rmesa, t->base.Target,
362 texImg->TexFormat, texObj->BaseLevel,
363 numLevels, texImg->Width, texImg->Height,
364 texImg->Depth, t->tile_bits);
365 }
366
367 /* Although we use the image_offset[] array to store relative offsets
368 * to cube faces, Mesa doesn't know anything about this and expects
369 * each cube face to be treated as a separate image.
370 *
371 * These functions present that view to mesa:
372 */
373 void
374 radeon_miptree_depth_offsets(radeon_mipmap_tree *mt, GLuint level, GLuint *offsets)
375 {
376 if (mt->target != GL_TEXTURE_3D || mt->faces == 1) {
377 offsets[0] = 0;
378 } else {
379 int i;
380 for (i = 0; i < 6; i++) {
381 offsets[i] = mt->levels[level].faces[i].offset;
382 }
383 }
384 }
385
386 GLuint
387 radeon_miptree_image_offset(radeon_mipmap_tree *mt,
388 GLuint face, GLuint level)
389 {
390 if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
391 return (mt->levels[level].faces[face].offset);
392 else
393 return mt->levels[level].faces[0].offset;
394 }
395
396 /**
397 * Ensure that the given image is stored in the given miptree from now on.
398 */
399 static void migrate_image_to_miptree(radeon_mipmap_tree *mt,
400 radeon_texture_image *image,
401 int face, int level)
402 {
403 radeon_mipmap_level *dstlvl = &mt->levels[level];
404 unsigned char *dest;
405
406 assert(image->mt != mt);
407 assert(dstlvl->valid);
408 assert(dstlvl->width == image->base.Width);
409 assert(dstlvl->height == image->base.Height);
410 assert(dstlvl->depth == image->base.Depth);
411
412 radeon_bo_map(mt->bo, GL_TRUE);
413 dest = mt->bo->ptr + dstlvl->faces[face].offset;
414
415 if (image->mt) {
416 /* Format etc. should match, so we really just need a memcpy().
417 * In fact, that memcpy() could be done by the hardware in many
418 * cases, provided that we have a proper memory manager.
419 */
420 assert(mt->mesaFormat == image->base.TexFormat);
421
422 radeon_mipmap_level *srclvl = &image->mt->levels[image->mtlevel];
423
424 /* TODO: bring back these assertions once the FBOs are fixed */
425 #if 0
426 assert(image->mtlevel == level);
427 assert(srclvl->size == dstlvl->size);
428 assert(srclvl->rowstride == dstlvl->rowstride);
429 #endif
430
431 radeon_bo_map(image->mt->bo, GL_FALSE);
432
433 memcpy(dest,
434 image->mt->bo->ptr + srclvl->faces[face].offset,
435 dstlvl->size);
436 radeon_bo_unmap(image->mt->bo);
437
438 radeon_miptree_unreference(&image->mt);
439 } else if (image->base.Data) {
440 /* This condition should be removed, it's here to workaround
441 * a segfault when mapping textures during software fallbacks.
442 */
443 const uint32_t srcrowstride = _mesa_format_row_stride(image->base.TexFormat, image->base.Width);
444 uint32_t rows = image->base.Height * image->base.Depth;
445
446 if (_mesa_is_format_compressed(image->base.TexFormat)) {
447 uint32_t blockWidth, blockHeight;
448 _mesa_get_format_block_size(image->base.TexFormat, &blockWidth, &blockHeight);
449 rows = (rows + blockHeight - 1) / blockHeight;
450 }
451
452 copy_rows(dest, dstlvl->rowstride, image->base.Data, srcrowstride,
453 rows, srcrowstride);
454
455 _mesa_free_texmemory(image->base.Data);
456 image->base.Data = 0;
457 }
458
459 radeon_bo_unmap(mt->bo);
460
461 radeon_miptree_reference(mt, &image->mt);
462 image->mtface = face;
463 image->mtlevel = level;
464 }
465
466 /**
467 * Filter matching miptrees, and select one with the most of data.
468 * @param[in] texObj radeon texture object
469 * @param[in] firstLevel first texture level to check
470 * @param[in] lastLevel last texture level to check
471 */
472 static radeon_mipmap_tree * get_biggest_matching_miptree(radeonTexObj *texObj,
473 unsigned firstLevel,
474 unsigned lastLevel)
475 {
476 const unsigned numLevels = lastLevel - firstLevel + 1;
477 unsigned *mtSizes = calloc(numLevels, sizeof(unsigned));
478 radeon_mipmap_tree **mts = calloc(numLevels, sizeof(radeon_mipmap_tree *));
479 unsigned mtCount = 0;
480 unsigned maxMtIndex = 0;
481 radeon_mipmap_tree *tmp;
482
483 for (unsigned level = firstLevel; level <= lastLevel; ++level) {
484 radeon_texture_image *img = get_radeon_texture_image(texObj->base.Image[0][level]);
485 unsigned found = 0;
486 // TODO: why this hack??
487 if (!img)
488 break;
489
490 if (!img->mt)
491 continue;
492
493 for (int i = 0; i < mtCount; ++i) {
494 if (mts[i] == img->mt) {
495 found = 1;
496 mtSizes[i] += img->mt->levels[img->mtlevel].size;
497 break;
498 }
499 }
500
501 if (!found && radeon_miptree_matches_texture(img->mt, &texObj->base)) {
502 mtSizes[mtCount] = img->mt->levels[img->mtlevel].size;
503 mts[mtCount] = img->mt;
504 mtCount++;
505 }
506 }
507
508 if (mtCount == 0) {
509 return NULL;
510 }
511
512 for (int i = 1; i < mtCount; ++i) {
513 if (mtSizes[i] > mtSizes[maxMtIndex]) {
514 maxMtIndex = i;
515 }
516 }
517
518 tmp = mts[maxMtIndex];
519 free(mtSizes);
520 free(mts);
521
522 return tmp;
523 }
524
525 /**
526 * Validate texture mipmap tree.
527 * If individual images are stored in different mipmap trees
528 * use the mipmap tree that has the most of the correct data.
529 */
530 int radeon_validate_texture_miptree(GLcontext * ctx, struct gl_texture_object *texObj)
531 {
532 radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
533 radeonTexObj *t = radeon_tex_obj(texObj);
534
535 if (t->validated || t->image_override) {
536 return GL_TRUE;
537 }
538
539 if (texObj->Image[0][texObj->BaseLevel]->Border > 0)
540 return GL_FALSE;
541
542 _mesa_test_texobj_completeness(rmesa->glCtx, texObj);
543 if (!texObj->_Complete) {
544 return GL_FALSE;
545 }
546
547 calculate_min_max_lod(&t->base, &t->minLod, &t->maxLod);
548
549 if (RADEON_DEBUG & RADEON_TEXTURE)
550 fprintf(stderr, "%s: Validating texture %p now, minLod = %d, maxLod = %d\n",
551 __FUNCTION__, texObj ,t->minLod, t->maxLod);
552
553 radeon_mipmap_tree *dst_miptree;
554 dst_miptree = get_biggest_matching_miptree(t, t->minLod, t->maxLod);
555
556 if (!dst_miptree) {
557 radeon_miptree_unreference(&t->mt);
558 radeon_try_alloc_miptree(rmesa, t);
559 dst_miptree = t->mt;
560 if (RADEON_DEBUG & RADEON_TEXTURE) {
561 fprintf(stderr, "%s: No matching miptree found, allocated new one %p\n", __FUNCTION__, t->mt);
562 }
563 } else if (RADEON_DEBUG & RADEON_TEXTURE) {
564 fprintf(stderr, "%s: Using miptree %p\n", __FUNCTION__, t->mt);
565 }
566
567 const unsigned faces = texObj->Target == GL_TEXTURE_CUBE_MAP ? 6 : 1;
568 unsigned face, level;
569 radeon_texture_image *img;
570 /* Validate only the levels that will actually be used during rendering */
571 for (face = 0; face < faces; ++face) {
572 for (level = t->minLod; level <= t->maxLod; ++level) {
573 img = get_radeon_texture_image(texObj->Image[face][level]);
574
575 if (RADEON_DEBUG & RADEON_TEXTURE) {
576 fprintf(stderr, "Checking image level %d, face %d, mt %p ... ", level, face, img->mt);
577 }
578
579 if (img->mt != dst_miptree) {
580 if (RADEON_DEBUG & RADEON_TEXTURE) {
581 fprintf(stderr, "MIGRATING\n");
582 }
583 struct radeon_bo *src_bo = (img->mt) ? img->mt->bo : img->bo;
584 if (src_bo && radeon_bo_is_referenced_by_cs(src_bo, rmesa->cmdbuf.cs)) {
585 radeon_firevertices(rmesa);
586 }
587 migrate_image_to_miptree(dst_miptree, img, face, level);
588 } else if (RADEON_DEBUG & RADEON_TEXTURE) {
589 fprintf(stderr, "OK\n");
590 }
591 }
592 }
593
594 t->validated = GL_TRUE;
595
596 return GL_TRUE;
597 }
598
599 uint32_t get_base_teximage_offset(radeonTexObj *texObj)
600 {
601 if (!texObj->mt) {
602 return 0;
603 } else {
604 return radeon_miptree_image_offset(texObj->mt, 0, texObj->minLod);
605 }
606 }