Search
lxdream.org :: lxdream/src/pvr2/texcache.c
lxdream 0.9.1
released Jun 29
Download Now
filename src/pvr2/texcache.c
changeset 561:533f6b478071
prev477:9a373f2ff009
next635:76c63aac3590
next653:3202ff01d48e
author nkeynes
date Tue Jan 01 05:08:38 2008 +0000 (12 years ago)
branchlxdream-mmu
permissions -rw-r--r--
last change Enable Id keyword on all source files
view annotate diff log raw
     1 /**
     2  * $Id$
     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 256
    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 -1
    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 texcache_entry_index 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 	texcache_active_list[i].texture_addr = -1;
    70     }
    71     texcache_free_ptr = 0;
    72     texcache_ref_counter = 0;
    73 }
    75 /**
    76  * Setup the initial texture ids (must be called after the GL context is
    77  * prepared)
    78  */
    79 void texcache_gl_init( )
    80 {
    81     int i;
    82     GLuint texids[MAX_TEXTURES];
    84     glGenTextures( MAX_TEXTURES, texids );
    85     for( i=0; i<MAX_TEXTURES; i++ ) {
    86 	texcache_active_list[i].texture_id = texids[i];
    87     }
    88 }
    90 /**
    91  * Flush all textures from the cache, returning them to the free list.
    92  */
    93 void texcache_flush( )
    94 {
    95     int i;
    96     /* clear structures */
    97     for( i=0; i<PVR2_RAM_PAGES; i++ ) {
    98 	texcache_page_lookup[i] = EMPTY_ENTRY;
    99     }
   100     for( i=0; i<MAX_TEXTURES; i++ ) {
   101 	texcache_free_list[i] = i;
   102     }
   103     texcache_free_ptr = 0;
   104     texcache_ref_counter = 0;
   105 }
   107 /**
   108  * Flush all textures and delete. The cache will be non-functional until
   109  * the next call to texcache_init(). This would typically be done if
   110  * switching GL targets.
   111  */    
   112 void texcache_shutdown( )
   113 {
   114     GLuint texids[MAX_TEXTURES];
   115     int i;
   116     texcache_flush();
   118     for( i=0; i<MAX_TEXTURES; i++ ) {
   119 	texids[i] = texcache_active_list[i].texture_id;
   120     }
   121     glDeleteTextures( MAX_TEXTURES, texids );
   122 }
   124 static void texcache_evict( int slot )
   125 {
   126     /* Remove the selected slot from the lookup table */
   127     assert( texcache_active_list[slot].texture_addr != -1 );
   128     uint32_t evict_page = texcache_active_list[slot].texture_addr >> 12;
   129     texcache_entry_index replace_next = texcache_active_list[slot].next;
   130     texcache_active_list[slot].texture_addr = -1;
   131     texcache_active_list[slot].next = EMPTY_ENTRY; /* Just for safety */
   132     if( texcache_page_lookup[evict_page] == slot ) {
   133 	texcache_page_lookup[evict_page] = replace_next;
   134     } else {
   135 	texcache_entry_index idx = texcache_page_lookup[evict_page];
   136 	texcache_entry_index next;
   137 	do {
   138 	    next = texcache_active_list[idx].next;
   139 	    if( next == slot ) {
   140 		assert( idx != replace_next );
   141 		texcache_active_list[idx].next = replace_next;
   142 		break;
   143 	    }
   144 	    idx = next;
   145 	} while( next != EMPTY_ENTRY );
   146     }
   147 }
   149 /**
   150  * Evict a single texture from the cache.
   151  * @return the slot of the evicted texture.
   152  */
   153 static texcache_entry_index texcache_evict_lru( void )
   154 {
   155     /* Full table scan - take over the entry with the lowest lru value */
   156     texcache_entry_index slot = 0;
   157     int lru_value = texcache_active_list[0].lru_count;
   158     int i;
   159     for( i=1; i<MAX_TEXTURES; i++ ) {
   160 	/* FIXME: account for rollover */
   161 	if( texcache_active_list[i].lru_count < lru_value ) {
   162 	    slot = i;
   163 	    lru_value = texcache_active_list[i].lru_count;
   164 	}
   165     }
   166     texcache_evict(slot);
   168     return slot;
   169 }
   171 /**
   172  * Evict all textures contained in the page identified by a texture address.
   173  */
   174 void texcache_invalidate_page( uint32_t texture_addr ) {
   175     uint32_t texture_page = texture_addr >> 12;
   176     texcache_entry_index idx = texcache_page_lookup[texture_page];
   177     if( idx == EMPTY_ENTRY )
   178 	return;
   179     assert( texcache_free_ptr >= 0 );
   180     do {
   181 	texcache_entry_t entry = &texcache_active_list[idx];
   182 	entry->texture_addr = -1;
   183 	/* release entry */
   184 	texcache_free_ptr--;
   185 	texcache_free_list[texcache_free_ptr] = idx;
   186 	idx = entry->next;
   187 	entry->next = EMPTY_ENTRY;
   188     } while( idx != EMPTY_ENTRY );
   189     texcache_page_lookup[texture_page] = EMPTY_ENTRY;
   190 }
   192 /**
   193  * Mark all textures that use the palette table as needing a re-read (ie 
   194  * for when the palette is changed. We could track exactly which ones are 
   195  * affected, but it's not clear that the extra maintanence overhead is 
   196  * worthwhile.
   197  */
   198 void texcache_invalidate_palette( )
   199 {
   200     int i;
   201     for( i=0; i<MAX_TEXTURES; i++ ) {
   202 	if( texcache_active_list[i].texture_addr != -1 &&
   203 	    PVR2_TEX_IS_PALETTE(texcache_active_list[i].mode) ) {
   204 	    texcache_evict( i );
   205 	    texcache_free_ptr--;
   206 	    texcache_free_list[texcache_free_ptr] = i;
   207 	}
   208     }
   209 }
   211 static void decode_pal8_to_32( uint32_t *out, uint8_t *in, int inbytes, uint32_t *pal )
   212 {
   213     int i;
   214     for( i=0; i<inbytes; i++ ) {
   215 	*out++ = pal[*in++];
   216     }
   217 }
   219 static void decode_pal8_to_16( uint16_t *out, uint8_t *in, int inbytes, uint32_t *pal )
   220 {
   221     int i;
   222     for( i=0; i<inbytes; i++ ) {
   223 	*out++ = (uint16_t)pal[*in++];
   224     }
   225 }
   227 static void decode_pal4_to_32( uint32_t *out, uint8_t *in, int inbytes, uint32_t *pal )
   228 {
   229     int i;
   230     for( i=0; i<inbytes; i++ ) {
   231 	*out++ = pal[*in & 0x0F];
   232 	*out++ = pal[(*in >> 4)];
   233 	in++;
   234     }
   235 }
   238 static void decode_pal4_to_16( uint16_t *out, uint8_t *in, int inbytes, uint32_t *pal )
   239 {
   240     int i;
   241     for( i=0; i<inbytes; i++ ) {
   242 	*out++ = (uint16_t)pal[*in & 0x0F];
   243 	*out++ = (uint16_t)pal[(*in >> 4)];
   244 	in++;
   245     }
   246 }
   248 #define VQ_CODEBOOK_SIZE 2048 /* 256 entries * 4 pixels per quad * 2 byte pixels */
   250 struct vq_codebook {
   251     uint16_t quad[256][4];
   252 };
   254 static void vq_get_codebook( struct vq_codebook *codebook, 
   255 				uint16_t *input )
   256 {
   257     /* Detwiddle the codebook, for the sake of my own sanity if nothing else */
   258     uint16_t *p = (uint16_t *)input;
   259     int i;
   260     for( i=0; i<256; i++ ) {
   261 	codebook->quad[i][0] = *p++;
   262 	codebook->quad[i][2] = *p++;
   263 	codebook->quad[i][1] = *p++;
   264 	codebook->quad[i][3] = *p++;
   265     }
   266 }    
   268 static void vq_decode( uint16_t *output, unsigned char *input, int width, int height, 
   269 		       struct vq_codebook *codebook ) {
   270     int i,j;
   272     uint8_t *c = (uint8_t *)input;
   273     for( j=0; j<height; j+=2 ) {
   274 	for( i=0; i<width; i+=2 ) {
   275 	    uint8_t code = *c++;
   276 	    output[i + j*width] = codebook->quad[code][0];
   277 	    output[i + 1 + j*width] = codebook->quad[code][1];
   278 	    output[i + (j+1)*width] = codebook->quad[code][2];
   279 	    output[i + 1 + (j+1)*width] = codebook->quad[code][3];
   280 	}
   281     }
   282 }
   284 static inline uint32_t yuv_to_rgb32( float y, float u, float v )
   285 {
   286     u -= 128;
   287     v -= 128;
   288     int r = (int)(y + v*1.375);
   289     int g = (int)(y - u*0.34375 - v*0.6875);
   290     int b = (int)(y + u*1.71875);
   291     if( r > 255 ) { r = 255; } else if( r < 0 ) { r = 0; }
   292     if( g > 255 ) { g = 255; } else if( g < 0 ) { g = 0; }
   293     if( b > 255 ) { b = 255; } else if( b < 0 ) { b = 0; }
   294     return 0xFF000000 | (r<<16) | (g<<8) | (b);
   295 }
   298 /**
   299  * Convert raster YUV texture data into RGB32 data - most GL implementations don't
   300  * directly support this format unfortunately. The input data is formatted as
   301  * 32 bits = 2 horizontal pixels, UYVY. This is currently done rather inefficiently
   302  * in floating point.
   303  */
   304 static void yuv_decode( uint32_t *output, uint32_t *input, int width, int height )
   305 {
   306     int x, y;
   307     uint32_t *p = input;
   308     for( y=0; y<height; y++ ) {
   309 	for( x=0; x<width; x+=2 ) {
   310 	    float u = (float)(*p & 0xFF);
   311 	    float y0 = (float)( (*p>>8)&0xFF );
   312 	    float v = (float)( (*p>>16)&0xFF );
   313 	    float y1 = (float)( (*p>>24)&0xFF );
   314 	    *output++ = yuv_to_rgb32( y0, u, v ); 
   315 	    *output++ = yuv_to_rgb32( y1, u, v );
   316 	    p++;
   317 	}
   318     }
   319 }
   321 /**
   322  * Load texture data from the given address and parameters into the currently
   323  * bound OpenGL texture.
   324  */
   325 static void texcache_load_texture( uint32_t texture_addr, int width, int height,
   326 			      int mode ) {
   327     int bpp_shift = 1; /* bytes per (output) pixel as a power of 2 */
   328     GLint intFormat = GL_RGBA, format, type;
   329     int tex_format = mode & PVR2_TEX_FORMAT_MASK;
   330     struct vq_codebook codebook;
   331     GLint filter = GL_LINEAR;
   333     glPixelStorei( GL_UNPACK_ROW_LENGTH, 0 );
   334     /* Decode the format parameters */
   335     switch( tex_format ) {
   336     case PVR2_TEX_FORMAT_IDX4:
   337     case PVR2_TEX_FORMAT_IDX8:
   338 	/* For indexed-colour modes, we need to lookup the palette control
   339 	 * word to determine the de-indexed texture format.
   340 	 */
   341 	switch( MMIO_READ( PVR2, RENDER_PALETTE ) & 0x03 ) {
   342 	case 0: /* ARGB1555 */
   343 	    format = GL_BGRA;
   344 	    type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
   345 	    break;
   346 	case 1:  /* RGB565 */
   347 	    intFormat = GL_RGB;
   348 	    format = GL_RGB;
   349 	    type = GL_UNSIGNED_SHORT_5_6_5;
   350 	    break;
   351 	case 2: /* ARGB4444 */
   352 	    format = GL_BGRA;
   353 	    type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
   354 	    break;
   355 	case 3: /* ARGB8888 */
   356 	    format = GL_BGRA;
   357 	    type = GL_UNSIGNED_BYTE;
   358 	    bpp_shift = 2;
   359 	    break;
   360 	default:
   361 	    return; /* Can't happen, but it makes gcc stop complaining */
   362 	}
   363 	break;
   365     case PVR2_TEX_FORMAT_ARGB1555:
   366 	format = GL_BGRA;
   367 	type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
   368 	break;
   369     case PVR2_TEX_FORMAT_RGB565:
   370 	intFormat = GL_RGB;
   371 	format = GL_RGB;
   372 	type = GL_UNSIGNED_SHORT_5_6_5;
   373 	break;
   374     case PVR2_TEX_FORMAT_ARGB4444:
   375 	format = GL_BGRA;
   376 	type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
   377 	break;
   378     case PVR2_TEX_FORMAT_YUV422:
   379 	/* YUV422 isn't directly supported by most implementations, so decode
   380 	 * it to a (reasonably) standard ARGB32.
   381 	 */
   382 	bpp_shift = 2;
   383 	format = GL_BGRA;
   384 	type = GL_UNSIGNED_BYTE;
   385 	break;
   386     case PVR2_TEX_FORMAT_BUMPMAP:
   387 	ERROR( "Bumpmap not supported" );
   388 	return;
   389     default:
   390 	ERROR( "Undefined texture format" );
   391 	return;
   392     }
   394     if( PVR2_TEX_IS_STRIDE(mode) && tex_format != PVR2_TEX_FORMAT_IDX4 &&
   395 	tex_format != PVR2_TEX_FORMAT_IDX8 ) {
   396 	/* Stride textures cannot be mip-mapped, compressed, indexed or twiddled */
   397 	uint32_t stride = (MMIO_READ( PVR2, RENDER_TEXSIZE ) & 0x003F) << 5;
   398 	unsigned char data[(width*height) << bpp_shift];
   399 	if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
   400 	    unsigned char tmp[(width*height)<<1];
   401 	    pvr2_vram64_read_stride( tmp, width<<1, texture_addr, stride<<1, height );
   402 	    yuv_decode( (uint32_t *)data, (uint32_t *)tmp, width, height );
   403 	} else {
   404 	    pvr2_vram64_read_stride( data, width<<bpp_shift, texture_addr, stride<<bpp_shift, height );
   405 	}
   406 	glTexImage2D( GL_TEXTURE_2D, 0, intFormat, width, height, 0, format, type, data );
   407 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
   408 	glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
   409 	return;
   410     } 
   412     int level=0, last_level = 0, mip_width = width, mip_height = height, src_bytes, dest_bytes;
   413     if( PVR2_TEX_IS_MIPMAPPED(mode) ) {
   414 	int i;
   415 	for( i=0; 1<<i < width; i++ );
   416 	last_level = i;
   417 	mip_width = 2;
   418 	mip_height= 2;
   419 	filter = GL_LINEAR_MIPMAP_LINEAR;
   420     }
   421     dest_bytes = (mip_width * mip_height) << bpp_shift;
   422     src_bytes = dest_bytes; // Modes will change this (below)
   424     if( PVR2_TEX_IS_COMPRESSED(mode) ) {
   425 	uint16_t tmp[VQ_CODEBOOK_SIZE];
   426 	pvr2_vram64_read( (unsigned char *)tmp, texture_addr, VQ_CODEBOOK_SIZE );
   427 	texture_addr += VQ_CODEBOOK_SIZE;
   428 	vq_get_codebook( &codebook, tmp );
   429     }
   431     for( level=last_level; level>= 0; level-- ) {
   432 	unsigned char data[dest_bytes];
   433 	/* load data from image, detwiddling/uncompressing as required */
   434 	if( tex_format == PVR2_TEX_FORMAT_IDX8 ) {
   435 	    src_bytes = (mip_width * mip_height);
   436 	    int bank = (mode >> 25) &0x03;
   437 	    uint32_t *palette = ((uint32_t *)mmio_region_PVR2PAL.mem) + (bank<<8);
   438 	    unsigned char tmp[src_bytes];
   439 	    pvr2_vram64_read_twiddled_8( tmp, texture_addr, mip_width, mip_height );
   440 	    if( bpp_shift == 2 ) {
   441 		decode_pal8_to_32( (uint32_t *)data, tmp, src_bytes, palette );
   442 	    } else {
   443 		decode_pal8_to_16( (uint16_t *)data, tmp, src_bytes, palette );
   444 	    }
   445 	} else if( tex_format == PVR2_TEX_FORMAT_IDX4 ) {
   446 	    src_bytes = (mip_width * mip_height) >> 1;
   447 	    int bank = (mode >>21 ) & 0x3F;
   448 	    uint32_t *palette = ((uint32_t *)mmio_region_PVR2PAL.mem) + (bank<<4);
   449 	    unsigned char tmp[src_bytes];
   450 	    pvr2_vram64_read_twiddled_4( tmp, texture_addr, mip_width, mip_height );
   451 	    if( bpp_shift == 2 ) {
   452 		decode_pal4_to_32( (uint32_t *)data, tmp, src_bytes, palette );
   453 	    } else {
   454 		decode_pal4_to_16( (uint16_t *)data, tmp, src_bytes, palette );
   455 	    }
   456 	} else if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
   457 	    src_bytes = ((mip_width*mip_height)<<1);
   458 	    unsigned char tmp[src_bytes];
   459 	    if( PVR2_TEX_IS_TWIDDLED(mode) ) {
   460 		pvr2_vram64_read_twiddled_16( tmp, texture_addr, mip_width, mip_height );
   461 	    } else {
   462 		pvr2_vram64_read( tmp, texture_addr, src_bytes );
   463 	    }
   464 	    yuv_decode( (uint32_t *)data, (uint32_t *)tmp, mip_width, mip_height );
   465 	} else if( PVR2_TEX_IS_COMPRESSED(mode) ) {
   466 	    src_bytes = ((mip_width*mip_height) >> 2);
   467 	    unsigned char tmp[src_bytes];
   468 	    if( PVR2_TEX_IS_TWIDDLED(mode) ) {
   469 		pvr2_vram64_read_twiddled_8( tmp, texture_addr, mip_width>>1, mip_height>>1 );
   470 	    } else {
   471 		pvr2_vram64_read( tmp, texture_addr, src_bytes );
   472 	    }
   473 	    vq_decode( (uint16_t *)data, tmp, mip_width, mip_height, &codebook );
   474 	} else if( PVR2_TEX_IS_TWIDDLED(mode) ) {
   475 	    pvr2_vram64_read_twiddled_16( data, texture_addr, mip_width, mip_height );
   476 	} else {
   477 	    pvr2_vram64_read( data, texture_addr, src_bytes );
   478 	}
   480 	/* Pass to GL */
   481 	if( level == last_level && level != 0 ) { /* 1x1 stored within a 2x2 */
   482 	    glTexImage2D( GL_TEXTURE_2D, level, intFormat, 1, 1, 0, format, type,
   483 			  data + (3 << bpp_shift) );
   484 	    texture_addr += src_bytes;
   485 	} else {
   486 	    glTexImage2D( GL_TEXTURE_2D, level, intFormat, mip_width, mip_height, 0, format, type,
   487 			  data );
   488 	    texture_addr += src_bytes;
   489 	    mip_width <<= 1;
   490 	    mip_height <<= 1;
   491 	    dest_bytes <<= 2;
   492 	    src_bytes <<= 2;
   493 	}
   494     }
   496     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
   497     glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
   498 }
   500 /**
   501  * Return a texture ID for the texture specified at the supplied address
   502  * and given parameters (the same sequence of bytes could in theory have
   503  * multiple interpretations). We use the texture address as the primary
   504  * index, but allow for multiple instances at each address. The texture
   505  * will be bound to the GL_TEXTURE_2D target before being returned.
   506  * 
   507  * If the texture has already been bound, return the ID to which it was
   508  * bound. Otherwise obtain an unused texture ID and set it up appropriately.
   509  */
   510 GLuint texcache_get_texture( uint32_t texture_addr, int width, int height,
   511 			     int mode )
   512 {
   513     uint32_t texture_page = texture_addr >> 12;
   514     texcache_entry_index next;
   515     texcache_entry_index idx = texcache_page_lookup[texture_page];
   516     while( idx != EMPTY_ENTRY ) {
   517 	texcache_entry_t entry = &texcache_active_list[idx];
   518 	if( entry->texture_addr == texture_addr &&
   519 	    entry->mode == mode &&
   520 	    entry->width == width &&
   521 	    entry->height == height ) {
   522 	    entry->lru_count = texcache_ref_counter++;
   523 	    glBindTexture( GL_TEXTURE_2D, entry->texture_id );
   524 	    return entry->texture_id;
   525 	}
   526         idx = entry->next;
   527     }
   529     /* Not found - check the free list */
   530     texcache_entry_index slot = 0;
   532     if( texcache_free_ptr < MAX_TEXTURES ) {
   533 	slot = texcache_free_list[texcache_free_ptr++];
   534     } else {
   535 	slot = texcache_evict_lru();
   536     }
   538     /* Construct new entry */
   539     texcache_active_list[slot].texture_addr = texture_addr;
   540     texcache_active_list[slot].width = width;
   541     texcache_active_list[slot].height = height;
   542     texcache_active_list[slot].mode = mode;
   543     texcache_active_list[slot].lru_count = texcache_ref_counter++;
   545     /* Add entry to the lookup table */
   546     next = texcache_page_lookup[texture_page];
   547     if( next == slot ) {
   548 	int i;
   549 	fprintf( stderr, "Active list: " );
   550 	for( i=0; i<MAX_TEXTURES; i++ ) {
   551 	    fprintf( stderr, "%d, ", texcache_active_list[i].next );
   552 	}
   553 	fprintf( stderr, "\n" );
   554 	assert( next != slot );
   556     }
   557     assert( next != slot );
   558     texcache_active_list[slot].next = next;
   559     texcache_page_lookup[texture_page] = slot;
   561     /* Construct the GL texture */
   562     glBindTexture( GL_TEXTURE_2D, texcache_active_list[slot].texture_id );
   563     texcache_load_texture( texture_addr, width, height, mode );
   565     return texcache_active_list[slot].texture_id;
   566 }
.