2 * Copyright (C) 2009 Maciej Cencora.
3 * Copyright (C) 2008 Nicolai Haehnle.
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:
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.
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.
29 #include "radeon_mipmap_tree.h"
34 #include "main/simple_list.h"
35 #include "main/teximage.h"
36 #include "main/texobj.h"
37 #include "radeon_texture.h"
39 static unsigned get_aligned_compressed_row_stride(
44 const unsigned blockSize
= _mesa_get_format_bytes(format
);
45 unsigned blockWidth
, blockHeight
, numXBlocks
;
47 _mesa_get_format_block_size(format
, &blockWidth
, &blockHeight
);
48 numXBlocks
= (width
+ blockWidth
- 1) / blockWidth
;
50 while (numXBlocks
* blockSize
< minStride
)
55 return numXBlocks
* blockSize
;
58 static unsigned get_compressed_image_size(
63 unsigned blockWidth
, blockHeight
;
65 _mesa_get_format_block_size(format
, &blockWidth
, &blockHeight
);
67 return rowStride
* ((height
+ blockHeight
- 1) / blockHeight
);
70 static int find_next_power_of_two(GLuint value
)
84 * Compute sizes and fill in offset and blit information for the given
85 * image (determined by \p face and \p level).
87 * \param curOffset points to the offset at which the image is to be stored
88 * and is updated by this function according to the size of the image.
90 static void compute_tex_image_offset(radeonContextPtr rmesa
, radeon_mipmap_tree
*mt
,
91 GLuint face
, GLuint level
, GLuint
* curOffset
)
93 radeon_mipmap_level
*lvl
= &mt
->levels
[level
];
97 height
= find_next_power_of_two(lvl
->height
);
99 /* Find image size in bytes */
100 if (_mesa_is_format_compressed(mt
->mesaFormat
)) {
101 lvl
->rowstride
= get_aligned_compressed_row_stride(mt
->mesaFormat
, lvl
->width
, rmesa
->texture_compressed_row_align
);
102 lvl
->size
= get_compressed_image_size(mt
->mesaFormat
, lvl
->rowstride
, height
);
103 } else if (mt
->target
== GL_TEXTURE_RECTANGLE_NV
) {
104 row_align
= rmesa
->texture_rect_row_align
- 1;
105 lvl
->rowstride
= (_mesa_format_row_stride(mt
->mesaFormat
, lvl
->width
) + row_align
) & ~row_align
;
106 lvl
->size
= lvl
->rowstride
* height
;
107 } else if (mt
->tilebits
& RADEON_TXO_MICRO_TILE
) {
108 /* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
109 * though the actual offset may be different (if texture is less than
110 * 32 bytes width) to the untiled case */
111 lvl
->rowstride
= (_mesa_format_row_stride(mt
->mesaFormat
, lvl
->width
) * 2 + 31) & ~31;
112 lvl
->size
= lvl
->rowstride
* ((height
+ 1) / 2) * lvl
->depth
;
114 row_align
= rmesa
->texture_row_align
- 1;
115 lvl
->rowstride
= (_mesa_format_row_stride(mt
->mesaFormat
, lvl
->width
) + row_align
) & ~row_align
;
116 lvl
->size
= lvl
->rowstride
* height
* lvl
->depth
;
118 assert(lvl
->size
> 0);
120 /* All images are aligned to a 32-byte offset */
121 *curOffset
= (*curOffset
+ 0x1f) & ~0x1f;
122 lvl
->faces
[face
].offset
= *curOffset
;
123 *curOffset
+= lvl
->size
;
125 if (RADEON_DEBUG
& RADEON_TEXTURE
)
127 "level %d, face %d: rs:%d %dx%d at %d\n",
128 level
, face
, lvl
->rowstride
, lvl
->width
, height
, lvl
->faces
[face
].offset
);
131 static GLuint
minify(GLuint size
, GLuint levels
)
133 size
= size
>> levels
;
140 static void calculate_miptree_layout_r100(radeonContextPtr rmesa
, radeon_mipmap_tree
*mt
)
142 GLuint curOffset
, i
, face
, level
;
144 assert(mt
->numLevels
<= rmesa
->glCtx
->Const
.MaxTextureLevels
);
147 for(face
= 0; face
< mt
->faces
; face
++) {
149 for(i
= 0, level
= mt
->baseLevel
; i
< mt
->numLevels
; i
++, level
++) {
150 mt
->levels
[level
].valid
= 1;
151 mt
->levels
[level
].width
= minify(mt
->width0
, i
);
152 mt
->levels
[level
].height
= minify(mt
->height0
, i
);
153 mt
->levels
[level
].depth
= minify(mt
->depth0
, i
);
154 compute_tex_image_offset(rmesa
, mt
, face
, level
, &curOffset
);
158 /* Note the required size in memory */
159 mt
->totalsize
= (curOffset
+ RADEON_OFFSET_MASK
) & ~RADEON_OFFSET_MASK
;
162 static void calculate_miptree_layout_r300(radeonContextPtr rmesa
, radeon_mipmap_tree
*mt
)
164 GLuint curOffset
, i
, level
;
166 assert(mt
->numLevels
<= rmesa
->glCtx
->Const
.MaxTextureLevels
);
169 for(i
= 0, level
= mt
->baseLevel
; i
< mt
->numLevels
; i
++, level
++) {
172 mt
->levels
[level
].valid
= 1;
173 mt
->levels
[level
].width
= minify(mt
->width0
, i
);
174 mt
->levels
[level
].height
= minify(mt
->height0
, i
);
175 mt
->levels
[level
].depth
= minify(mt
->depth0
, i
);
177 for(face
= 0; face
< mt
->faces
; face
++)
178 compute_tex_image_offset(rmesa
, mt
, face
, level
, &curOffset
);
179 /* r600 cube levels seems to be aligned to 8 faces but
180 * we have separate register for 1'st level offset so add
181 * 2 image alignment after 1'st mip level */
182 if(rmesa
->radeonScreen
->chip_family
>= CHIP_FAMILY_R600
&&
183 mt
->target
== GL_TEXTURE_CUBE_MAP
&& level
>= 1)
184 curOffset
+= 2 * mt
->levels
[level
].size
;
187 /* Note the required size in memory */
188 mt
->totalsize
= (curOffset
+ RADEON_OFFSET_MASK
) & ~RADEON_OFFSET_MASK
;
192 * Create a new mipmap tree, calculate its layout and allocate memory.
194 static radeon_mipmap_tree
* radeon_miptree_create(radeonContextPtr rmesa
,
195 GLenum target
, gl_format mesaFormat
, GLuint baseLevel
, GLuint numLevels
,
196 GLuint width0
, GLuint height0
, GLuint depth0
, GLuint tilebits
)
198 radeon_mipmap_tree
*mt
= CALLOC_STRUCT(_radeon_mipmap_tree
);
200 mt
->mesaFormat
= mesaFormat
;
203 mt
->faces
= (target
== GL_TEXTURE_CUBE_MAP
) ? 6 : 1;
204 mt
->baseLevel
= baseLevel
;
205 mt
->numLevels
= numLevels
;
207 mt
->height0
= height0
;
209 mt
->tilebits
= tilebits
;
211 if (rmesa
->radeonScreen
->chip_family
>= CHIP_FAMILY_R300
)
212 calculate_miptree_layout_r300(rmesa
, mt
);
214 calculate_miptree_layout_r100(rmesa
, mt
);
216 mt
->bo
= radeon_bo_open(rmesa
->radeonScreen
->bom
,
217 0, mt
->totalsize
, 1024,
218 RADEON_GEM_DOMAIN_VRAM
,
224 void radeon_miptree_reference(radeon_mipmap_tree
*mt
, radeon_mipmap_tree
**ptr
)
229 assert(mt
->refcount
> 0);
234 void radeon_miptree_unreference(radeon_mipmap_tree
**ptr
)
236 radeon_mipmap_tree
*mt
= *ptr
;
240 assert(mt
->refcount
> 0);
244 radeon_bo_unref(mt
->bo
);
252 * Calculate min and max LOD for the given texture object.
253 * @param[in] tObj texture object whose LOD values to calculate
254 * @param[out] pminLod minimal LOD
255 * @param[out] pmaxLod maximal LOD
257 static void calculate_min_max_lod(struct gl_texture_object
*tObj
,
258 unsigned *pminLod
, unsigned *pmaxLod
)
261 /* Yes, this looks overly complicated, but it's all needed.
263 switch (tObj
->Target
) {
267 case GL_TEXTURE_CUBE_MAP
:
268 if (tObj
->MinFilter
== GL_NEAREST
|| tObj
->MinFilter
== GL_LINEAR
) {
269 /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
271 minLod
= maxLod
= tObj
->BaseLevel
;
273 minLod
= tObj
->BaseLevel
+ (GLint
)(tObj
->MinLod
);
274 minLod
= MAX2(minLod
, tObj
->BaseLevel
);
275 minLod
= MIN2(minLod
, tObj
->MaxLevel
);
276 maxLod
= tObj
->BaseLevel
+ (GLint
)(tObj
->MaxLod
+ 0.5);
277 maxLod
= MIN2(maxLod
, tObj
->MaxLevel
);
278 maxLod
= MIN2(maxLod
, tObj
->Image
[0][minLod
]->MaxLog2
+ minLod
);
279 maxLod
= MAX2(maxLod
, minLod
); /* need at least one level */
282 case GL_TEXTURE_RECTANGLE_NV
:
283 case GL_TEXTURE_4D_SGIS
:
290 /* save these values */
296 * Checks whether the given miptree can hold the given texture image at the
297 * given face and level.
299 GLboolean
radeon_miptree_matches_image(radeon_mipmap_tree
*mt
,
300 struct gl_texture_image
*texImage
, GLuint face
, GLuint level
)
302 radeon_mipmap_level
*lvl
;
304 if (face
>= mt
->faces
)
307 if (texImage
->TexFormat
!= mt
->mesaFormat
)
310 lvl
= &mt
->levels
[level
];
312 lvl
->width
!= texImage
->Width
||
313 lvl
->height
!= texImage
->Height
||
314 lvl
->depth
!= texImage
->Depth
)
321 * Checks whether the given miptree has the right format to store the given texture object.
323 static GLboolean
radeon_miptree_matches_texture(radeon_mipmap_tree
*mt
, struct gl_texture_object
*texObj
)
325 struct gl_texture_image
*firstImage
;
327 radeon_mipmap_level
*mtBaseLevel
;
329 if (texObj
->BaseLevel
< mt
->baseLevel
)
332 mtBaseLevel
= &mt
->levels
[texObj
->BaseLevel
- mt
->baseLevel
];
333 firstImage
= texObj
->Image
[0][texObj
->BaseLevel
];
334 numLevels
= MIN2(texObj
->MaxLevel
- texObj
->BaseLevel
+ 1, firstImage
->MaxLog2
+ 1);
336 if (RADEON_DEBUG
& RADEON_TEXTURE
) {
337 fprintf(stderr
, "Checking if miptree %p matches texObj %p\n", mt
, texObj
);
338 fprintf(stderr
, "target %d vs %d\n", mt
->target
, texObj
->Target
);
339 fprintf(stderr
, "format %d vs %d\n", mt
->mesaFormat
, firstImage
->TexFormat
);
340 fprintf(stderr
, "numLevels %d vs %d\n", mt
->numLevels
, numLevels
);
341 fprintf(stderr
, "width0 %d vs %d\n", mtBaseLevel
->width
, firstImage
->Width
);
342 fprintf(stderr
, "height0 %d vs %d\n", mtBaseLevel
->height
, firstImage
->Height
);
343 fprintf(stderr
, "depth0 %d vs %d\n", mtBaseLevel
->depth
, firstImage
->Depth
);
344 if (mt
->target
== texObj
->Target
&&
345 mt
->mesaFormat
== firstImage
->TexFormat
&&
346 mt
->numLevels
>= numLevels
&&
347 mtBaseLevel
->width
== firstImage
->Width
&&
348 mtBaseLevel
->height
== firstImage
->Height
&&
349 mtBaseLevel
->depth
== firstImage
->Depth
) {
350 fprintf(stderr
, "MATCHED\n");
352 fprintf(stderr
, "NOT MATCHED\n");
356 return (mt
->target
== texObj
->Target
&&
357 mt
->mesaFormat
== firstImage
->TexFormat
&&
358 mt
->numLevels
>= numLevels
&&
359 mtBaseLevel
->width
== firstImage
->Width
&&
360 mtBaseLevel
->height
== firstImage
->Height
&&
361 mtBaseLevel
->depth
== firstImage
->Depth
);
365 * Try to allocate a mipmap tree for the given texture object.
366 * @param[in] rmesa radeon context
367 * @param[in] t radeon texture object
369 void radeon_try_alloc_miptree(radeonContextPtr rmesa
, radeonTexObj
*t
)
371 struct gl_texture_object
*texObj
= &t
->base
;
372 struct gl_texture_image
*texImg
= texObj
->Image
[0][texObj
->BaseLevel
];
380 numLevels
= MIN2(texObj
->MaxLevel
- texObj
->BaseLevel
+ 1, texImg
->MaxLog2
+ 1);
382 t
->mt
= radeon_miptree_create(rmesa
, t
->base
.Target
,
383 texImg
->TexFormat
, texObj
->BaseLevel
,
384 numLevels
, texImg
->Width
, texImg
->Height
,
385 texImg
->Depth
, t
->tile_bits
);
389 radeon_miptree_image_offset(radeon_mipmap_tree
*mt
,
390 GLuint face
, GLuint level
)
392 if (mt
->target
== GL_TEXTURE_CUBE_MAP_ARB
)
393 return (mt
->levels
[level
].faces
[face
].offset
);
395 return mt
->levels
[level
].faces
[0].offset
;
399 * Ensure that the given image is stored in the given miptree from now on.
401 static void migrate_image_to_miptree(radeon_mipmap_tree
*mt
,
402 radeon_texture_image
*image
,
405 radeon_mipmap_level
*dstlvl
= &mt
->levels
[level
];
408 assert(image
->mt
!= mt
);
409 assert(dstlvl
->valid
);
410 assert(dstlvl
->width
== image
->base
.Width
);
411 assert(dstlvl
->height
== image
->base
.Height
);
412 assert(dstlvl
->depth
== image
->base
.Depth
);
414 radeon_bo_map(mt
->bo
, GL_TRUE
);
415 dest
= mt
->bo
->ptr
+ dstlvl
->faces
[face
].offset
;
418 /* Format etc. should match, so we really just need a memcpy().
419 * In fact, that memcpy() could be done by the hardware in many
420 * cases, provided that we have a proper memory manager.
422 assert(mt
->mesaFormat
== image
->base
.TexFormat
);
424 radeon_mipmap_level
*srclvl
= &image
->mt
->levels
[image
->mtlevel
];
426 /* TODO: bring back these assertions once the FBOs are fixed */
428 assert(image
->mtlevel
== level
);
429 assert(srclvl
->size
== dstlvl
->size
);
430 assert(srclvl
->rowstride
== dstlvl
->rowstride
);
433 radeon_bo_map(image
->mt
->bo
, GL_FALSE
);
436 image
->mt
->bo
->ptr
+ srclvl
->faces
[face
].offset
,
438 radeon_bo_unmap(image
->mt
->bo
);
440 radeon_miptree_unreference(&image
->mt
);
441 } else if (image
->base
.Data
) {
442 /* This condition should be removed, it's here to workaround
443 * a segfault when mapping textures during software fallbacks.
445 const uint32_t srcrowstride
= _mesa_format_row_stride(image
->base
.TexFormat
, image
->base
.Width
);
446 uint32_t rows
= image
->base
.Height
* image
->base
.Depth
;
448 if (_mesa_is_format_compressed(image
->base
.TexFormat
)) {
449 uint32_t blockWidth
, blockHeight
;
450 _mesa_get_format_block_size(image
->base
.TexFormat
, &blockWidth
, &blockHeight
);
451 rows
= (rows
+ blockHeight
- 1) / blockHeight
;
454 copy_rows(dest
, dstlvl
->rowstride
, image
->base
.Data
, srcrowstride
,
457 _mesa_free_texmemory(image
->base
.Data
);
458 image
->base
.Data
= 0;
461 radeon_bo_unmap(mt
->bo
);
463 radeon_miptree_reference(mt
, &image
->mt
);
464 image
->mtface
= face
;
465 image
->mtlevel
= level
;
469 * Filter matching miptrees, and select one with the most of data.
470 * @param[in] texObj radeon texture object
471 * @param[in] firstLevel first texture level to check
472 * @param[in] lastLevel last texture level to check
474 static radeon_mipmap_tree
* get_biggest_matching_miptree(radeonTexObj
*texObj
,
478 const unsigned numLevels
= lastLevel
- firstLevel
+ 1;
479 unsigned *mtSizes
= calloc(numLevels
, sizeof(unsigned));
480 radeon_mipmap_tree
**mts
= calloc(numLevels
, sizeof(radeon_mipmap_tree
*));
481 unsigned mtCount
= 0;
482 unsigned maxMtIndex
= 0;
483 radeon_mipmap_tree
*tmp
;
485 for (unsigned level
= firstLevel
; level
<= lastLevel
; ++level
) {
486 radeon_texture_image
*img
= get_radeon_texture_image(texObj
->base
.Image
[0][level
]);
488 // TODO: why this hack??
495 for (int i
= 0; i
< mtCount
; ++i
) {
496 if (mts
[i
] == img
->mt
) {
498 mtSizes
[i
] += img
->mt
->levels
[img
->mtlevel
].size
;
503 if (!found
&& radeon_miptree_matches_texture(img
->mt
, &texObj
->base
)) {
504 mtSizes
[mtCount
] = img
->mt
->levels
[img
->mtlevel
].size
;
505 mts
[mtCount
] = img
->mt
;
514 for (int i
= 1; i
< mtCount
; ++i
) {
515 if (mtSizes
[i
] > mtSizes
[maxMtIndex
]) {
520 tmp
= mts
[maxMtIndex
];
528 * Validate texture mipmap tree.
529 * If individual images are stored in different mipmap trees
530 * use the mipmap tree that has the most of the correct data.
532 int radeon_validate_texture_miptree(GLcontext
* ctx
, struct gl_texture_object
*texObj
)
534 radeonContextPtr rmesa
= RADEON_CONTEXT(ctx
);
535 radeonTexObj
*t
= radeon_tex_obj(texObj
);
537 if (t
->validated
|| t
->image_override
) {
541 if (texObj
->Image
[0][texObj
->BaseLevel
]->Border
> 0)
544 _mesa_test_texobj_completeness(rmesa
->glCtx
, texObj
);
545 if (!texObj
->_Complete
) {
549 calculate_min_max_lod(&t
->base
, &t
->minLod
, &t
->maxLod
);
551 if (RADEON_DEBUG
& RADEON_TEXTURE
)
552 fprintf(stderr
, "%s: Validating texture %p now, minLod = %d, maxLod = %d\n",
553 __FUNCTION__
, texObj
,t
->minLod
, t
->maxLod
);
555 radeon_mipmap_tree
*dst_miptree
;
556 dst_miptree
= get_biggest_matching_miptree(t
, t
->minLod
, t
->maxLod
);
559 radeon_miptree_unreference(&t
->mt
);
560 radeon_try_alloc_miptree(rmesa
, t
);
562 if (RADEON_DEBUG
& RADEON_TEXTURE
) {
563 fprintf(stderr
, "%s: No matching miptree found, allocated new one %p\n", __FUNCTION__
, t
->mt
);
565 } else if (RADEON_DEBUG
& RADEON_TEXTURE
) {
566 fprintf(stderr
, "%s: Using miptree %p\n", __FUNCTION__
, t
->mt
);
569 const unsigned faces
= texObj
->Target
== GL_TEXTURE_CUBE_MAP
? 6 : 1;
570 unsigned face
, level
;
571 radeon_texture_image
*img
;
572 /* Validate only the levels that will actually be used during rendering */
573 for (face
= 0; face
< faces
; ++face
) {
574 for (level
= t
->minLod
; level
<= t
->maxLod
; ++level
) {
575 img
= get_radeon_texture_image(texObj
->Image
[face
][level
]);
577 if (RADEON_DEBUG
& RADEON_TEXTURE
) {
578 fprintf(stderr
, "Checking image level %d, face %d, mt %p ... ", level
, face
, img
->mt
);
581 if (img
->mt
!= dst_miptree
) {
582 if (RADEON_DEBUG
& RADEON_TEXTURE
) {
583 fprintf(stderr
, "MIGRATING\n");
585 struct radeon_bo
*src_bo
= (img
->mt
) ? img
->mt
->bo
: img
->bo
;
586 if (src_bo
&& radeon_bo_is_referenced_by_cs(src_bo
, rmesa
->cmdbuf
.cs
)) {
587 radeon_firevertices(rmesa
);
589 migrate_image_to_miptree(dst_miptree
, img
, face
, level
);
590 } else if (RADEON_DEBUG
& RADEON_TEXTURE
) {
591 fprintf(stderr
, "OK\n");
596 t
->validated
= GL_TRUE
;
601 uint32_t get_base_teximage_offset(radeonTexObj
*texObj
)
606 return radeon_miptree_image_offset(texObj
->mt
, 0, texObj
->minLod
);