Search
lxdream.org :: lxdream/src/pvr2/texcache.c
lxdream 0.9.1
released Jun 29
Download Now
filename src/pvr2/texcache.c
changeset 300:a8ae4aeb0b42
prev291:cb3e8f0ac24a
next311:5be79c6b4367
author nkeynes
date Wed Jan 17 09:21:55 2007 +0000 (13 years ago)
permissions -rw-r--r--
last change Fix RGB565 format
Fix miscalculation of byte size
view annotate diff log raw
     1 /**
     2  * $Id: texcache.c,v 1.15 2007-01-17 09:21:55 nkeynes Exp $
     3  *
     4  * Texture cache. Responsible for maintaining a working set of OpenGL 
     5  * textures. 
     6  *
     7  *
     8  * Copyright (c) 2005 Nathan Keynes.
     9  *
    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.
    14  *
    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.
    19  */
    21 #include <assert.h>
    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.
    27  */
    28 #define MAX_TEXTURES 64
    30 /**
    31  * Data structure:
    32  *
    33  * Main operations:
    34  *    find entry by texture_addr
    35  *    add new entry
    36  *    move entry to tail of lru list
    37  *    remove entry
    38  */
    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;
    49     GLuint texture_id;
    50     texcache_entry_index next;
    51     uint32_t lru_count;
    52 } *texcache_entry_t;
    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];
    58 /**
    59  * Initialize the texture cache.
    60  */
    61 void texcache_init( )
    62 {
    63     int i;
    64     for( i=0; i<PVR2_RAM_PAGES; i++ ) {
    65 	texcache_page_lookup[i] = EMPTY_ENTRY;
    66     }
    67     for( i=0; i<MAX_TEXTURES; i++ ) {
    68 	texcache_free_list[i] = i;
    69     }
    70     texcache_free_ptr = 0;
    71     texcache_ref_counter = 0;
    72 }
    74 /**
    75  * Setup the initial texture ids (must be called after the GL context is
    76  * prepared)
    77  */
    78 void texcache_gl_init( )
    79 {
    80     int i;
    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];
    86     }
    87 }
    89 /**
    90  * Flush all textures from the cache, returning them to the free list.
    91  */
    92 void texcache_flush( )
    93 {
    94     int i;
    95     /* clear structures */
    96     for( i=0; i<PVR2_RAM_PAGES; i++ ) {
    97 	texcache_page_lookup[i] = EMPTY_ENTRY;
    98     }
    99     for( i=0; i<MAX_TEXTURES; i++ ) {
   100 	texcache_free_list[i] = i;
   101     }
   102     texcache_free_ptr = 0;
   103     texcache_ref_counter = 0;
   104 }
   106 /**
   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.
   110  */    
   111 void texcache_shutdown( )
   112 {
   113     GLuint texids[MAX_TEXTURES];
   114     int i;
   115     texcache_flush();
   117     for( i=0; i<MAX_TEXTURES; i++ ) {
   118 	texids[i] = texcache_active_list[i].texture_id;
   119     }
   120     glDeleteTextures( MAX_TEXTURES, texids );
   121 }
   123 /**
   124  * Evict all textures contained in the page identified by a texture address.
   125  */
   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 )
   130 	return;
   131     assert( texcache_free_ptr >= 0 );
   132     do {
   133 	texcache_entry_t entry = &texcache_active_list[idx];	
   134 	/* release entry */
   135 	texcache_free_ptr--;
   136 	texcache_free_list[texcache_free_ptr] = idx;
   137 	idx = entry->next;
   138 	entry->next = EMPTY_ENTRY;
   139     } while( idx != EMPTY_ENTRY );
   140     texcache_page_lookup[texture_page] = EMPTY_ENTRY;
   141 }
   143 /**
   144  * Evict a single texture from the cache.
   145  * @return the slot of the evicted texture.
   146  */
   147 static texcache_entry_index texcache_evict( void )
   148 {
   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;
   152     int i;
   153     for( i=1; i<MAX_TEXTURES; i++ ) {
   154 	/* FIXME: account for rollover */
   155 	if( texcache_active_list[i].lru_count < lru_value ) {
   156 	    slot = i;
   157 	    lru_value = texcache_active_list[i].lru_count;
   158 	}
   159     }
   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;
   167     } else {
   168 	texcache_entry_index idx = texcache_page_lookup[evict_page];
   169 	texcache_entry_index next;
   170 	do {
   171 	    next = texcache_active_list[idx].next;
   172 	    if( next == slot ) {
   173 		texcache_active_list[idx].next = replace_next;
   174 		break;
   175 	    }
   176 	    idx = next;
   177 	} while( next != EMPTY_ENTRY );
   178     }
   179     return slot;
   180 }
   182 static void detwiddle_pal8_to_32(int x1, int y1, int size, int totsize,
   183 				 char **in, uint32_t *out, uint32_t *pal) {
   184     if (size == 1) {
   185 	out[y1 * totsize + x1] = pal[**in];
   186 	(*in)++;
   187     } else {
   188 	int ns = size>>1;
   189 	detwiddle_pal8_to_32(x1, y1, ns, totsize, in, out, pal);
   190 	detwiddle_pal8_to_32(x1, y1+ns, ns, totsize, in, out, pal);
   191 	detwiddle_pal8_to_32(x1+ns, y1, ns, totsize, in, out, pal);
   192 	detwiddle_pal8_to_32(x1+ns, y1+ns, ns, totsize, in, out, pal);
   193     }
   194 }
   196 static void detwiddle_pal8_to_16(int x1, int y1, int size, int totsize,
   197 				 char **in, uint16_t *out, uint16_t *pal) {
   198     if (size == 1) {
   199 	out[y1 * totsize + x1] = pal[**in];
   200 	(*in)++;
   201     } else {
   202 	int ns = size>>1;
   203 	detwiddle_pal8_to_16(x1, y1, ns, totsize, in, out, pal);
   204 	detwiddle_pal8_to_16(x1, y1+ns, ns, totsize, in, out, pal);
   205 	detwiddle_pal8_to_16(x1+ns, y1, ns, totsize, in, out, pal);
   206 	detwiddle_pal8_to_16(x1+ns, y1+ns, ns, totsize, in, out, pal);
   207     }
   208 }
   210 static void detwiddle_16_to_16(int x1, int y1, int size, int totsize,
   211 			       uint16_t **in, uint16_t *out ) {
   212     if (size == 1) {
   213 	out[y1 * totsize + x1] = **in;
   214 	(*in)++;
   215     } else {
   216 	int ns = size>>1;
   217 	detwiddle_16_to_16(x1, y1, ns, totsize, in, out);
   218 	detwiddle_16_to_16(x1, y1+ns, ns, totsize, in, out);
   219 	detwiddle_16_to_16(x1+ns, y1, ns, totsize, in, out);
   220 	detwiddle_16_to_16(x1+ns, y1+ns, ns, totsize, in, out);
   221     }
   222 }
   224 #define VQ_CODEBOOK_SIZE 2048 /* 256 entries * 4 pixels per quad * 2 byte pixels */
   226 struct vq_codebook {
   227     uint16_t quad[256][4];
   228 };
   230 static void detwiddle_vq_to_16(int x1, int y1, int size, int totsize,
   231 		   uint8_t **in, uint16_t *out, struct vq_codebook *codebook ) {
   232     if( size == 2 ) {
   233 	uint8_t code = **in;
   234 	(*in)++;
   235 	out[y1 * totsize + x1] = codebook->quad[code][0];
   236 	out[y1 * totsize + x1 + 1] = codebook->quad[code][1];
   237 	out[(y1+1) * totsize + x1] = codebook->quad[code][2];
   238 	out[(y1+1) * totsize + x1 + 1] = codebook->quad[code][3];
   239     } else {
   240 	int ns = size>>1;
   241 	detwiddle_vq_to_16(x1, y1, ns, totsize, in, out, codebook);
   242 	detwiddle_vq_to_16(x1, y1+ns, ns, totsize, in, out, codebook);
   243 	detwiddle_vq_to_16(x1+ns, y1, ns, totsize, in, out, codebook);
   244 	detwiddle_vq_to_16(x1+ns, y1+ns, ns, totsize, in, out, codebook);
   245     }	
   246 }
   248 static void vq_get_codebook( struct vq_codebook *codebook, 
   249 				uint16_t *input )
   250 {
   251     /* Detwiddle the codebook, for the sake of my own sanity if nothing else */
   252     uint16_t *p = (uint16_t *)input;
   253     int i;
   254     for( i=0; i<256; i++ ) {
   255 	codebook->quad[i][0] = *p++;
   256 	codebook->quad[i][2] = *p++;
   257 	codebook->quad[i][1] = *p++;
   258 	codebook->quad[i][3] = *p++;
   259     }
   260 }    
   263 static void vq_decode( int width, int height, char *input, uint16_t *output,
   264 		       struct vq_codebook *codebook, int twiddled ) {
   265     int i,j;
   267     uint8_t *c = (uint8_t *)input;
   268     if( twiddled ) {
   269 	detwiddle_vq_to_16( 0, 0, width, width, &c, output, codebook );
   270     } else {
   271 	for( j=0; j<height; j+=2 ) {
   272 	    for( i=0; i<width; i+=2 ) {
   273 		uint8_t code = *c;
   274 		output[i + j*width] = codebook->quad[code][0];
   275 		output[i + 1 + j*width] = codebook->quad[code][1];
   276 		output[i + (j+1)*width] = codebook->quad[code][2];
   277 		output[i + 1 + (j+1)*width] = codebook->quad[code][3];
   278 	    }
   279 	}
   280     }
   281 }
   283 static inline uint32_t yuv_to_rgb32( float y, float u, float v )
   284 {
   285     u -= 128;
   286     v -= 128;
   287     int r = (int)(y + v*1.375);
   288     int g = (int)(y - u*0.34375 - v*0.6875);
   289     int b = (int)(y + u*1.71875);
   290     if( r > 255 ) { r = 255; } else if( r < 0 ) { r = 0; }
   291     if( g > 255 ) { g = 255; } else if( g < 0 ) { g = 0; }
   292     if( b > 255 ) { b = 255; } else if( b < 0 ) { b = 0; }
   293     return 0xFF000000 | (r<<16) | (g<<8) | (b);
   294 }
   297 /**
   298  * Convert non-twiddled YUV texture data into RGB32 data - most GL implementations don't
   299  * directly support this format unfortunately. The input data is formatted as
   300  * 32 bits = 2 horizontal pixels, UYVY. This is currently done rather inefficiently
   301  * in floating point.
   302  */
   303 static void yuv_decode( int width, int height, uint32_t *input, uint32_t *output )
   304 {
   305     int x, y;
   306     uint32_t *p = input;
   307     for( y=0; y<height; y++ ) {
   308 	for( x=0; x<width; x+=2 ) {
   309 	    float u = (float)(*p & 0xFF);
   310 	    float y0 = (float)( (*p>>8)&0xFF );
   311 	    float v = (float)( (*p>>16)&0xFF );
   312 	    float y1 = (float)( (*p>>24)&0xFF );
   313 	    *output++ = yuv_to_rgb32( y0, u, v ); 
   314 	    *output++ = yuv_to_rgb32( y1, u, v );
   315 	    p++;
   316 	}
   317     }
   318 }
   320 /**
   321  * Load texture data from the given address and parameters into the currently
   322  * bound OpenGL texture.
   323  */
   324 static texcache_load_texture( uint32_t texture_addr, int width, int height,
   325 			      int mode ) {
   326     int bpp_shift = 1; /* bytes per (output) pixel as a power of 2 */
   327     GLint intFormat, format, type;
   328     int tex_format = mode & PVR2_TEX_FORMAT_MASK;
   329     struct vq_codebook codebook;
   330     GLint filter = GL_LINEAR;
   332     /* Decode the format parameters */
   333     switch( tex_format ) {
   334     case PVR2_TEX_FORMAT_IDX4:
   335 	ERROR( "4-bit indexed textures not supported" );
   336     case PVR2_TEX_FORMAT_IDX8:
   337 	/* For indexed-colour modes, we need to lookup the palette control
   338 	 * word to determine the de-indexed texture format.
   339 	 */
   340 	switch( MMIO_READ( PVR2, RENDER_PALETTE ) & 0x03 ) {
   341 	case 0: /* ARGB1555 */
   342 	    intFormat = GL_RGB5_A1;
   343 	    format = GL_RGBA;
   344 	    type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
   345 	    break;
   346 	case 1:  /* RGB565 */
   347 	    intFormat = GL_RGB5;
   348 	    format = GL_RGB;
   349 	    type = GL_UNSIGNED_SHORT_5_6_5;
   350 	    break;
   351 	case 2: /* ARGB4444 */
   352 	    intFormat = GL_RGBA4;
   353 	    format = GL_BGRA;
   354 	    type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
   355 	    break;
   356 	case 3: /* ARGB8888 */
   357 	    intFormat = GL_RGBA8;
   358 	    format = GL_BGRA;
   359 	    type = GL_UNSIGNED_INT_8_8_8_8_REV;
   360 	    bpp_shift = 2;
   361 	    break;
   362 	}
   363 	break;
   365     case PVR2_TEX_FORMAT_ARGB1555:
   366 	intFormat = GL_RGB5_A1;
   367 	format = GL_RGBA;
   368 	type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
   369 	break;
   370     case PVR2_TEX_FORMAT_RGB565:
   371 	intFormat = GL_RGB5;
   372 	format = GL_RGB;
   373 	type = GL_UNSIGNED_SHORT_5_6_5;
   374 	break;
   375     case PVR2_TEX_FORMAT_ARGB4444:
   376 	intFormat = GL_RGBA4;
   377 	format = GL_BGRA;
   378 	type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
   379 	break;
   380     case PVR2_TEX_FORMAT_YUV422:
   381 	/* YUV422 isn't directly supported by most implementations, so decode
   382 	 * it to a (reasonably) standard ARGB32.
   383 	 */
   384 	bpp_shift = 2;
   385 	intFormat = GL_RGBA8;
   386 	format = GL_BGRA;
   387 	type = GL_UNSIGNED_INT_8_8_8_8_REV;
   388 	break;
   389     case PVR2_TEX_FORMAT_BUMPMAP:
   390 	ERROR( "Bumpmap not supported" );
   391 	break;
   392     }
   394     if( PVR2_TEX_IS_STRIDE(mode) ) {
   395 	/* Stride textures cannot be mip-mapped, compressed, indexed or twiddled */
   396 	uint32_t stride = (MMIO_READ( PVR2, RENDER_TEXSIZE ) & 0x003F) << 5;
   397 	char data[(width*height) << bpp_shift];
   398 	if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
   399 	    char tmp[(width*height)<<1];
   400 	    pvr2_vram64_read_stride( tmp, width<<1, texture_addr, stride<<1, height );
   401 	    yuv_decode(width, height, (uint32_t *)tmp, (uint32_t *)data );
   402 	} else {
   403 	    pvr2_vram64_read_stride( data, width<<bpp_shift, texture_addr, stride<<bpp_shift, height );
   404 	}
   405 	glTexImage2D( GL_TEXTURE_2D, 0, intFormat, width, height, 0, format, type, data );
   406 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
   407 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
   408 	return;
   409     } 
   411     int level=0, last_level = 0, mip_width = width, mip_height = height, mip_bytes;
   412     if( PVR2_TEX_IS_MIPMAPPED(mode) ) {
   413 	int i;
   414 	for( i=0; 1<<(i+1) < width; i++ );
   415 	last_level = i;
   416 	mip_width = width >> i;
   417 	mip_height= height >> i;
   418 	filter = GL_LINEAR_MIPMAP_LINEAR;
   419     }
   420     mip_bytes = (mip_width * mip_height) << bpp_shift;
   422     if( PVR2_TEX_IS_COMPRESSED(mode) ) {
   423 	uint16_t tmp[VQ_CODEBOOK_SIZE];
   424 	pvr2_vram64_read( (char *)tmp, texture_addr, VQ_CODEBOOK_SIZE );
   425 	texture_addr += VQ_CODEBOOK_SIZE;
   426 	vq_get_codebook( &codebook, tmp );
   427     }
   429     for( level=last_level; level>= 0; level-- ) {
   430 	char data[mip_bytes];
   431 	/* load data from image, detwiddling/uncompressing as required */
   432 	if( tex_format == PVR2_TEX_FORMAT_IDX8 ) {
   433 	    int inputlength = mip_bytes >> bpp_shift;
   434 	    int bank = (mode >> 25) &0x03;
   435 	    char *palette = mmio_region_PVR2PAL.mem + (bank * (256 << bpp_shift));
   436 	    char tmp[inputlength];
   437 	    char *p = tmp;
   438 	    pvr2_vram64_read( tmp, texture_addr, inputlength );
   439 	    if( bpp_shift == 2 ) {
   440 		detwiddle_pal8_to_32( 0, 0, mip_width, mip_width, &p, 
   441 				      (uint32_t *)data, (uint32_t *)palette );
   442 	    } else {
   443 		detwiddle_pal8_to_16( 0, 0, mip_width, mip_width, &p,
   444 				      (uint16_t *)data, (uint16_t *)palette );
   445 	    }
   446 	} else if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
   447 	    int inputlength = ((mip_width*mip_height)<<1);
   448 	    char tmp[inputlength];
   449 	    pvr2_vram64_read( tmp, texture_addr, inputlength );
   450 	    yuv_decode( mip_width, mip_height, (uint32_t *)tmp, (uint32_t *)data );
   451 	} else if( PVR2_TEX_IS_COMPRESSED(mode) ) {
   452 	    int inputlength = ((mip_width*mip_height) >> 2);
   453 	    char tmp[inputlength];
   454 	    pvr2_vram64_read( tmp, texture_addr, inputlength );
   455 	    vq_decode( mip_width, mip_height, tmp, (uint16_t *)data, &codebook, 
   456 		       PVR2_TEX_IS_TWIDDLED(mode) );
   457 	} else if( PVR2_TEX_IS_TWIDDLED(mode) ) {
   458 	    char tmp[mip_bytes];
   459 	    uint16_t *p = (uint16_t *)tmp;
   460 	    pvr2_vram64_read( tmp, texture_addr, mip_bytes );
   461 	    /* Untwiddle */
   462 	    detwiddle_16_to_16( 0, 0, mip_width, mip_width, &p, (uint16_t *)data );
   463 	} else {
   464 	    pvr2_vram64_read( data, texture_addr, mip_bytes );
   465 	}
   467 	if( PVR2_TEX_IS_MIPMAPPED(mode) && mip_width == 2 ) {
   468 	    /* Opengl requires a 1x1 texture, but the PVR2 doesn't. This should
   469 	     * strictly speaking be the average of the 2x2 texture, but we're
   470 	     * lazy at the moment */
   471 	    glTexImage2D( GL_TEXTURE_2D, level+1, intFormat, 1, 1, 0, format, type, data );
   472 	}
   474 	/* Pass to GL */
   475 	glTexImage2D( GL_TEXTURE_2D, level, intFormat, mip_width, mip_height, 0, format, type,
   476 		      data );
   477 	texture_addr += mip_bytes;
   478 	mip_width <<= 1;
   479 	mip_height <<= 1;
   480 	mip_bytes <<= 2;
   481     }
   483     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
   484     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
   485 }
   487 /**
   488  * Return a texture ID for the texture specified at the supplied address
   489  * and given parameters (the same sequence of bytes could in theory have
   490  * multiple interpretations). We use the texture address as the primary
   491  * index, but allow for multiple instances at each address. The texture
   492  * will be bound to the GL_TEXTURE_2D target before being returned.
   493  * 
   494  * If the texture has already been bound, return the ID to which it was
   495  * bound. Otherwise obtain an unused texture ID and set it up appropriately.
   496  */
   497 GLuint texcache_get_texture( uint32_t texture_addr, int width, int height,
   498 			     int mode )
   499 {
   500     uint32_t texture_page = texture_addr >> 12;
   501     texcache_entry_index idx = texcache_page_lookup[texture_page];
   502     while( idx != EMPTY_ENTRY ) {
   503 	texcache_entry_t entry = &texcache_active_list[idx];
   504 	if( entry->texture_addr == texture_addr &&
   505 	    entry->mode == mode &&
   506 	    entry->width == width &&
   507 	    entry->height == height ) {
   508 	    entry->lru_count = texcache_ref_counter++;
   509 	    glBindTexture( GL_TEXTURE_2D, entry->texture_id );
   510 	    return entry->texture_id;
   511 	}
   512         idx = entry->next;
   513     }
   515     /* Not found - check the free list */
   516     int slot = 0;
   518     if( texcache_free_ptr < MAX_TEXTURES ) {
   519 	slot = texcache_free_list[texcache_free_ptr++];
   520     } else {
   521 	slot = texcache_evict();
   522     }
   524     /* Construct new entry */
   525     texcache_active_list[slot].texture_addr = texture_addr;
   526     texcache_active_list[slot].width = width;
   527     texcache_active_list[slot].height = height;
   528     texcache_active_list[slot].mode = mode;
   529     texcache_active_list[slot].lru_count = texcache_ref_counter++;
   531     /* Add entry to the lookup table */
   532     texcache_active_list[slot].next = texcache_page_lookup[texture_page];
   533     texcache_page_lookup[texture_page] = slot;
   535     /* Construct the GL texture */
   536     glBindTexture( GL_TEXTURE_2D, texcache_active_list[slot].texture_id );
   537     texcache_load_texture( texture_addr, width, height, mode );
   539     return texcache_active_list[slot].texture_id;
   540 }
.