2 * $Id: texcache.c,v 1.21 2007-01-25 10:12:54 nkeynes Exp $
4 * Texture cache. Responsible for maintaining a working set of OpenGL
8 * Copyright (c) 2005 Nathan Keynes.
10 * This program is free software; you can redistribute it and/or modify
11 * it under the terms of the GNU General Public License as published by
12 * the Free Software Foundation; either version 2 of the License, or
13 * (at your option) any later version.
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
22 #include "pvr2/pvr2.h"
24 /** Specifies the maximum number of OpenGL
25 * textures we're willing to have open at a time. If more are
26 * needed, textures will be evicted in LRU order.
28 #define MAX_TEXTURES 128
34 * find entry by texture_addr
36 * move entry to tail of lru list
40 typedef signed short texcache_entry_index;
41 #define EMPTY_ENTRY 0xFF
43 static texcache_entry_index texcache_free_ptr = 0;
44 static GLuint texcache_free_list[MAX_TEXTURES];
46 typedef struct texcache_entry {
47 uint32_t texture_addr;
48 int width, height, mode;
50 texcache_entry_index next;
54 static uint8_t texcache_page_lookup[PVR2_RAM_PAGES];
55 static uint32_t texcache_ref_counter;
56 static struct texcache_entry texcache_active_list[MAX_TEXTURES];
59 * Initialize the texture cache.
64 for( i=0; i<PVR2_RAM_PAGES; i++ ) {
65 texcache_page_lookup[i] = EMPTY_ENTRY;
67 for( i=0; i<MAX_TEXTURES; i++ ) {
68 texcache_free_list[i] = i;
70 texcache_free_ptr = 0;
71 texcache_ref_counter = 0;
75 * Setup the initial texture ids (must be called after the GL context is
78 void texcache_gl_init( )
81 GLuint texids[MAX_TEXTURES];
83 glGenTextures( MAX_TEXTURES, texids );
84 for( i=0; i<MAX_TEXTURES; i++ ) {
85 texcache_active_list[i].texture_id = texids[i];
90 * Flush all textures from the cache, returning them to the free list.
92 void texcache_flush( )
95 /* clear structures */
96 for( i=0; i<PVR2_RAM_PAGES; i++ ) {
97 texcache_page_lookup[i] = EMPTY_ENTRY;
99 for( i=0; i<MAX_TEXTURES; i++ ) {
100 texcache_free_list[i] = i;
102 texcache_free_ptr = 0;
103 texcache_ref_counter = 0;
107 * Flush all textures and delete. The cache will be non-functional until
108 * the next call to texcache_init(). This would typically be done if
109 * switching GL targets.
111 void texcache_shutdown( )
113 GLuint texids[MAX_TEXTURES];
117 for( i=0; i<MAX_TEXTURES; i++ ) {
118 texids[i] = texcache_active_list[i].texture_id;
120 glDeleteTextures( MAX_TEXTURES, texids );
124 * Evict all textures contained in the page identified by a texture address.
126 void texcache_invalidate_page( uint32_t texture_addr ) {
127 uint32_t texture_page = texture_addr >> 12;
128 texcache_entry_index idx = texcache_page_lookup[texture_page];
129 if( idx == EMPTY_ENTRY )
131 assert( texcache_free_ptr >= 0 );
133 texcache_entry_t entry = &texcache_active_list[idx];
136 texcache_free_list[texcache_free_ptr] = idx;
138 entry->next = EMPTY_ENTRY;
139 } while( idx != EMPTY_ENTRY );
140 texcache_page_lookup[texture_page] = EMPTY_ENTRY;
144 * Evict a single texture from the cache.
145 * @return the slot of the evicted texture.
147 static texcache_entry_index texcache_evict( void )
149 /* Full table scan - take over the entry with the lowest lru value */
150 texcache_entry_index slot = 0;
151 int lru_value = texcache_active_list[0].lru_count;
153 for( i=1; i<MAX_TEXTURES; i++ ) {
154 /* FIXME: account for rollover */
155 if( texcache_active_list[i].lru_count < lru_value ) {
157 lru_value = texcache_active_list[i].lru_count;
161 /* Remove the selected slot from the lookup table */
162 uint32_t evict_page = texcache_active_list[slot].texture_addr >> 12;
163 texcache_entry_index replace_next = texcache_active_list[slot].next;
164 texcache_active_list[slot].next = EMPTY_ENTRY; /* Just for safety */
165 if( texcache_page_lookup[evict_page] == slot ) {
166 texcache_page_lookup[evict_page] = replace_next;
168 texcache_entry_index idx = texcache_page_lookup[evict_page];
169 texcache_entry_index next;
171 next = texcache_active_list[idx].next;
173 texcache_active_list[idx].next = replace_next;
177 } while( next != EMPTY_ENTRY );
182 static void decode_pal8_to_32( uint32_t *out, uint8_t *in, int inbytes, uint32_t *pal )
185 for( i=0; i<inbytes; i++ ) {
190 static void decode_pal8_to_16( uint16_t *out, uint8_t *in, int inbytes, uint32_t *pal )
193 for( i=0; i<inbytes; i++ ) {
194 *out++ = (uint16_t)pal[*in++];
198 static void decode_pal4_to_32( uint32_t *out, uint8_t *in, int inbytes, uint32_t *pal )
201 for( i=0; i<inbytes; i++ ) {
202 *out++ = pal[*in & 0x0F];
203 *out++ = pal[(*in >> 4)];
209 static void decode_pal4_to_16( uint16_t *out, uint8_t *in, int inbytes, uint32_t *pal )
212 for( i=0; i<inbytes; i++ ) {
213 *out++ = (uint16_t)pal[*in & 0x0F];
214 *out++ = (uint16_t)pal[(*in >> 4)];
219 #define VQ_CODEBOOK_SIZE 2048 /* 256 entries * 4 pixels per quad * 2 byte pixels */
222 uint16_t quad[256][4];
225 static void vq_get_codebook( struct vq_codebook *codebook,
228 /* Detwiddle the codebook, for the sake of my own sanity if nothing else */
229 uint16_t *p = (uint16_t *)input;
231 for( i=0; i<256; i++ ) {
232 codebook->quad[i][0] = *p++;
233 codebook->quad[i][2] = *p++;
234 codebook->quad[i][1] = *p++;
235 codebook->quad[i][3] = *p++;
239 static void vq_decode( uint16_t *output, char *input, int width, int height,
240 struct vq_codebook *codebook ) {
243 uint8_t *c = (uint8_t *)input;
244 for( j=0; j<height; j+=2 ) {
245 for( i=0; i<width; i+=2 ) {
247 output[i + j*width] = codebook->quad[code][0];
248 output[i + 1 + j*width] = codebook->quad[code][1];
249 output[i + (j+1)*width] = codebook->quad[code][2];
250 output[i + 1 + (j+1)*width] = codebook->quad[code][3];
255 static inline uint32_t yuv_to_rgb32( float y, float u, float v )
259 int r = (int)(y + v*1.375);
260 int g = (int)(y - u*0.34375 - v*0.6875);
261 int b = (int)(y + u*1.71875);
262 if( r > 255 ) { r = 255; } else if( r < 0 ) { r = 0; }
263 if( g > 255 ) { g = 255; } else if( g < 0 ) { g = 0; }
264 if( b > 255 ) { b = 255; } else if( b < 0 ) { b = 0; }
265 return 0xFF000000 | (r<<16) | (g<<8) | (b);
270 * Convert raster YUV texture data into RGB32 data - most GL implementations don't
271 * directly support this format unfortunately. The input data is formatted as
272 * 32 bits = 2 horizontal pixels, UYVY. This is currently done rather inefficiently
275 static void yuv_decode( uint32_t *output, uint32_t *input, int width, int height )
279 for( y=0; y<height; y++ ) {
280 for( x=0; x<width; x+=2 ) {
281 float u = (float)(*p & 0xFF);
282 float y0 = (float)( (*p>>8)&0xFF );
283 float v = (float)( (*p>>16)&0xFF );
284 float y1 = (float)( (*p>>24)&0xFF );
285 *output++ = yuv_to_rgb32( y0, u, v );
286 *output++ = yuv_to_rgb32( y1, u, v );
293 * Load texture data from the given address and parameters into the currently
294 * bound OpenGL texture.
296 static texcache_load_texture( uint32_t texture_addr, int width, int height,
298 int bpp_shift = 1; /* bytes per (output) pixel as a power of 2 */
299 GLint intFormat, format, type;
300 int tex_format = mode & PVR2_TEX_FORMAT_MASK;
301 struct vq_codebook codebook;
302 GLint filter = GL_LINEAR;
304 /* Decode the format parameters */
305 switch( tex_format ) {
306 case PVR2_TEX_FORMAT_IDX4:
307 case PVR2_TEX_FORMAT_IDX8:
308 /* For indexed-colour modes, we need to lookup the palette control
309 * word to determine the de-indexed texture format.
311 switch( MMIO_READ( PVR2, RENDER_PALETTE ) & 0x03 ) {
312 case 0: /* ARGB1555 */
313 intFormat = GL_RGB5_A1;
315 type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
320 type = GL_UNSIGNED_SHORT_5_6_5;
322 case 2: /* ARGB4444 */
323 intFormat = GL_RGBA4;
325 type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
327 case 3: /* ARGB8888 */
328 intFormat = GL_RGBA8;
330 type = GL_UNSIGNED_INT_8_8_8_8_REV;
336 case PVR2_TEX_FORMAT_ARGB1555:
337 intFormat = GL_RGB5_A1;
339 type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
341 case PVR2_TEX_FORMAT_RGB565:
344 type = GL_UNSIGNED_SHORT_5_6_5;
346 case PVR2_TEX_FORMAT_ARGB4444:
347 intFormat = GL_RGBA4;
349 type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
351 case PVR2_TEX_FORMAT_YUV422:
352 /* YUV422 isn't directly supported by most implementations, so decode
353 * it to a (reasonably) standard ARGB32.
356 intFormat = GL_RGBA8;
358 type = GL_UNSIGNED_INT_8_8_8_8_REV;
360 case PVR2_TEX_FORMAT_BUMPMAP:
361 ERROR( "Bumpmap not supported" );
365 if( PVR2_TEX_IS_STRIDE(mode) && tex_format != PVR2_TEX_FORMAT_IDX4 &&
366 tex_format != PVR2_TEX_FORMAT_IDX8 ) {
367 /* Stride textures cannot be mip-mapped, compressed, indexed or twiddled */
368 uint32_t stride = (MMIO_READ( PVR2, RENDER_TEXSIZE ) & 0x003F) << 5;
369 char data[(width*height) << bpp_shift];
370 if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
371 char tmp[(width*height)<<1];
372 pvr2_vram64_read_stride( tmp, width<<1, texture_addr, stride<<1, height );
373 yuv_decode( (uint32_t *)data, (uint32_t *)tmp, width, height );
375 pvr2_vram64_read_stride( data, width<<bpp_shift, texture_addr, stride<<bpp_shift, height );
377 glTexImage2D( GL_TEXTURE_2D, 0, intFormat, width, height, 0, format, type, data );
378 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
379 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
383 int level=0, last_level = 0, mip_width = width, mip_height = height, mip_bytes;
384 if( PVR2_TEX_IS_MIPMAPPED(mode) ) {
386 for( i=0; 1<<i < width; i++ );
390 filter = GL_LINEAR_MIPMAP_LINEAR;
392 mip_bytes = (mip_width * mip_height) << bpp_shift;
394 if( PVR2_TEX_IS_COMPRESSED(mode) ) {
395 uint16_t tmp[VQ_CODEBOOK_SIZE];
396 pvr2_vram64_read( (char *)tmp, texture_addr, VQ_CODEBOOK_SIZE );
397 texture_addr += VQ_CODEBOOK_SIZE;
398 vq_get_codebook( &codebook, tmp );
401 for( level=last_level; level>= 0; level-- ) {
402 char data[mip_bytes];
403 /* load data from image, detwiddling/uncompressing as required */
404 if( tex_format == PVR2_TEX_FORMAT_IDX8 ) {
405 int inputlength = mip_bytes >> bpp_shift;
406 int bank = (mode >> 25) &0x03;
407 uint32_t *palette = ((uint32_t *)mmio_region_PVR2PAL.mem) + (bank<<8);
408 char tmp[inputlength];
409 pvr2_vram64_read_twiddled_8( tmp, texture_addr, mip_width, mip_height );
410 if( bpp_shift == 2 ) {
411 decode_pal8_to_32( (uint32_t *)data, tmp, inputlength, palette );
413 decode_pal8_to_16( (uint16_t *)data, tmp, inputlength, palette );
415 } else if( tex_format == PVR2_TEX_FORMAT_IDX4 ) {
416 int inputlength = (mip_width * mip_height) >> 1;
417 int bank = (mode >>21 ) & 0x3F;
418 uint32_t *palette = ((uint32_t *)mmio_region_PVR2PAL.mem) + (bank<<4);
419 char tmp[inputlength];
420 pvr2_vram64_read_twiddled_4( tmp, texture_addr, mip_width, mip_height );
421 if( bpp_shift == 2 ) {
422 decode_pal4_to_32( (uint32_t *)data, tmp, inputlength, palette );
424 decode_pal4_to_16( (uint16_t *)data, tmp, inputlength, palette );
426 } else if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
427 int inputlength = ((mip_width*mip_height)<<1);
428 char tmp[inputlength];
429 if( PVR2_TEX_IS_TWIDDLED(mode) ) {
430 pvr2_vram64_read_twiddled_16( tmp, texture_addr, mip_width, mip_height );
432 pvr2_vram64_read( tmp, texture_addr, inputlength );
434 yuv_decode( (uint32_t *)data, (uint32_t *)tmp, mip_width, mip_height );
435 } else if( PVR2_TEX_IS_COMPRESSED(mode) ) {
436 int inputlength = ((mip_width*mip_height) >> 2);
437 char tmp[inputlength];
438 if( PVR2_TEX_IS_TWIDDLED(mode) ) {
439 pvr2_vram64_read_twiddled_8( tmp, texture_addr, mip_width>>1, mip_height>>1 );
441 pvr2_vram64_read( tmp, texture_addr, inputlength );
443 vq_decode( (uint16_t *)data, tmp, mip_width, mip_height, &codebook );
444 } else if( PVR2_TEX_IS_TWIDDLED(mode) ) {
445 pvr2_vram64_read_twiddled_16( data, texture_addr, mip_width, mip_height );
447 pvr2_vram64_read( data, texture_addr, mip_bytes );
451 if( level == last_level && level != 0 ) { /* 1x1 stored within a 2x2 */
452 glTexImage2D( GL_TEXTURE_2D, level, intFormat, 1, 1, 0, format, type,
453 data + (3 << bpp_shift) );
455 glTexImage2D( GL_TEXTURE_2D, level, intFormat, mip_width, mip_height, 0, format, type,
457 texture_addr += mip_bytes;
464 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
465 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
469 * Return a texture ID for the texture specified at the supplied address
470 * and given parameters (the same sequence of bytes could in theory have
471 * multiple interpretations). We use the texture address as the primary
472 * index, but allow for multiple instances at each address. The texture
473 * will be bound to the GL_TEXTURE_2D target before being returned.
475 * If the texture has already been bound, return the ID to which it was
476 * bound. Otherwise obtain an unused texture ID and set it up appropriately.
478 GLuint texcache_get_texture( uint32_t texture_addr, int width, int height,
481 uint32_t texture_page = texture_addr >> 12;
482 texcache_entry_index idx = texcache_page_lookup[texture_page];
483 while( idx != EMPTY_ENTRY ) {
484 texcache_entry_t entry = &texcache_active_list[idx];
485 if( entry->texture_addr == texture_addr &&
486 entry->mode == mode &&
487 entry->width == width &&
488 entry->height == height ) {
489 entry->lru_count = texcache_ref_counter++;
490 glBindTexture( GL_TEXTURE_2D, entry->texture_id );
491 return entry->texture_id;
496 /* Not found - check the free list */
499 if( texcache_free_ptr < MAX_TEXTURES ) {
500 slot = texcache_free_list[texcache_free_ptr++];
502 slot = texcache_evict();
505 /* Construct new entry */
506 texcache_active_list[slot].texture_addr = texture_addr;
507 texcache_active_list[slot].width = width;
508 texcache_active_list[slot].height = height;
509 texcache_active_list[slot].mode = mode;
510 texcache_active_list[slot].lru_count = texcache_ref_counter++;
512 /* Add entry to the lookup table */
513 texcache_active_list[slot].next = texcache_page_lookup[texture_page];
514 texcache_page_lookup[texture_page] = slot;
516 /* Construct the GL texture */
517 glBindTexture( GL_TEXTURE_2D, texcache_active_list[slot].texture_id );
518 texcache_load_texture( texture_addr, width, height, mode );
520 return texcache_active_list[slot].texture_id;
.