filename | src/pvr2/texcache.c |
changeset | 647:d6fecb7750b9 |
prev | 645:a7392098299c |
next | 1066:ddffe9d2b332 |
author | nkeynes |
date | Sun Mar 16 04:49:19 2008 +0000 (16 years ago) |
branch | lxdream-render |
permissions | -rw-r--r-- |
last change | Use max-z rather than min-z for tri sort (still wrong for some cases of course, but consistent with prior behaviour) |
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 <string.h>
23 #include "pvr2/pvr2.h"
25 /** Specifies the maximum number of OpenGL
26 * textures we're willing to have open at a time. If more are
27 * needed, textures will be evicted in LRU order.
28 */
29 #define MAX_TEXTURES 256
31 /**
32 * Data structure:
33 *
34 * Main operations:
35 * find entry by texture_addr
36 * add new entry
37 * move entry to tail of lru list
38 * remove entry
39 */
41 typedef signed short texcache_entry_index;
42 #define EMPTY_ENTRY -1
44 static texcache_entry_index texcache_free_ptr = 0;
45 static GLuint texcache_free_list[MAX_TEXTURES];
47 typedef struct texcache_entry {
48 uint32_t texture_addr;
49 int width, height, mode;
50 GLuint texture_id;
51 texcache_entry_index next;
52 uint32_t lru_count;
53 } *texcache_entry_t;
55 static texcache_entry_index texcache_page_lookup[PVR2_RAM_PAGES];
56 static uint32_t texcache_ref_counter;
57 static struct texcache_entry texcache_active_list[MAX_TEXTURES];
59 /**
60 * Initialize the texture cache.
61 */
62 void texcache_init( )
63 {
64 int i;
65 for( i=0; i<PVR2_RAM_PAGES; i++ ) {
66 texcache_page_lookup[i] = EMPTY_ENTRY;
67 }
68 for( i=0; i<MAX_TEXTURES; i++ ) {
69 texcache_free_list[i] = i;
70 texcache_active_list[i].texture_addr = -1;
71 texcache_active_list[i].next = EMPTY_ENTRY;
72 }
73 texcache_free_ptr = 0;
74 texcache_ref_counter = 0;
75 }
77 /**
78 * Setup the initial texture ids (must be called after the GL context is
79 * prepared)
80 */
81 void texcache_gl_init( )
82 {
83 int i;
84 GLuint texids[MAX_TEXTURES];
86 glGenTextures( MAX_TEXTURES, texids );
87 for( i=0; i<MAX_TEXTURES; i++ ) {
88 texcache_active_list[i].texture_id = texids[i];
89 }
90 }
92 /**
93 * Flush all textures from the cache, returning them to the free list.
94 */
95 void texcache_flush( )
96 {
97 int i;
98 /* clear structures */
99 for( i=0; i<PVR2_RAM_PAGES; i++ ) {
100 texcache_page_lookup[i] = EMPTY_ENTRY;
101 }
102 for( i=0; i<MAX_TEXTURES; i++ ) {
103 texcache_free_list[i] = i;
104 texcache_active_list[i].next = EMPTY_ENTRY;
105 }
106 texcache_free_ptr = 0;
107 texcache_ref_counter = 0;
108 }
110 /**
111 * Flush all textures and delete. The cache will be non-functional until
112 * the next call to texcache_init(). This would typically be done if
113 * switching GL targets.
114 */
115 void texcache_shutdown( )
116 {
117 GLuint texids[MAX_TEXTURES];
118 int i;
119 texcache_flush();
121 for( i=0; i<MAX_TEXTURES; i++ ) {
122 texids[i] = texcache_active_list[i].texture_id;
123 }
124 glDeleteTextures( MAX_TEXTURES, texids );
125 }
127 static void texcache_evict( int slot )
128 {
129 /* Remove the selected slot from the lookup table */
130 assert( texcache_active_list[slot].texture_addr != -1 );
131 uint32_t evict_page = texcache_active_list[slot].texture_addr >> 12;
132 texcache_entry_index replace_next = texcache_active_list[slot].next;
133 texcache_active_list[slot].texture_addr = -1;
134 texcache_active_list[slot].next = EMPTY_ENTRY; /* Just for safety */
135 if( texcache_page_lookup[evict_page] == slot ) {
136 texcache_page_lookup[evict_page] = replace_next;
137 } else {
138 texcache_entry_index idx = texcache_page_lookup[evict_page];
139 texcache_entry_index next;
140 do {
141 next = texcache_active_list[idx].next;
142 if( next == slot ) {
143 assert( idx != replace_next );
144 texcache_active_list[idx].next = replace_next;
145 break;
146 }
147 idx = next;
148 } while( next != EMPTY_ENTRY );
149 }
150 }
152 /**
153 * Evict a single texture from the cache.
154 * @return the slot of the evicted texture.
155 */
156 static texcache_entry_index texcache_evict_lru( void )
157 {
158 /* Full table scan - take over the entry with the lowest lru value */
159 texcache_entry_index slot = 0;
160 int lru_value = texcache_active_list[0].lru_count;
161 int i;
162 for( i=1; i<MAX_TEXTURES; i++ ) {
163 /* FIXME: account for rollover */
164 if( texcache_active_list[i].lru_count < lru_value ) {
165 slot = i;
166 lru_value = texcache_active_list[i].lru_count;
167 }
168 }
169 texcache_evict(slot);
171 return slot;
172 }
174 /**
175 * Evict all textures contained in the page identified by a texture address.
176 */
177 void texcache_invalidate_page( uint32_t texture_addr ) {
178 uint32_t texture_page = texture_addr >> 12;
179 texcache_entry_index idx = texcache_page_lookup[texture_page];
180 if( idx == EMPTY_ENTRY )
181 return;
182 assert( texcache_free_ptr >= 0 );
183 do {
184 texcache_entry_t entry = &texcache_active_list[idx];
185 entry->texture_addr = -1;
186 /* release entry */
187 texcache_free_ptr--;
188 texcache_free_list[texcache_free_ptr] = idx;
189 idx = entry->next;
190 entry->next = EMPTY_ENTRY;
191 } while( idx != EMPTY_ENTRY );
192 texcache_page_lookup[texture_page] = EMPTY_ENTRY;
193 }
195 /**
196 * Mark all textures that use the palette table as needing a re-read (ie
197 * for when the palette is changed. We could track exactly which ones are
198 * affected, but it's not clear that the extra maintanence overhead is
199 * worthwhile.
200 */
201 void texcache_invalidate_palette( )
202 {
203 int i;
204 for( i=0; i<MAX_TEXTURES; i++ ) {
205 if( texcache_active_list[i].texture_addr != -1 &&
206 PVR2_TEX_IS_PALETTE(texcache_active_list[i].mode) ) {
207 texcache_evict( i );
208 texcache_free_ptr--;
209 texcache_free_list[texcache_free_ptr] = i;
210 }
211 }
212 }
214 static void decode_pal8_to_32( uint32_t *out, uint8_t *in, int inbytes, uint32_t *pal )
215 {
216 int i;
217 for( i=0; i<inbytes; i++ ) {
218 *out++ = pal[*in++];
219 }
220 }
222 static void decode_pal8_to_16( uint16_t *out, uint8_t *in, int inbytes, uint32_t *pal )
223 {
224 int i;
225 for( i=0; i<inbytes; i++ ) {
226 *out++ = (uint16_t)pal[*in++];
227 }
228 }
230 static void decode_pal4_to_32( uint32_t *out, uint8_t *in, int inbytes, uint32_t *pal )
231 {
232 int i;
233 for( i=0; i<inbytes; i++ ) {
234 *out++ = pal[*in & 0x0F];
235 *out++ = pal[(*in >> 4)];
236 in++;
237 }
238 }
241 static void decode_pal4_to_16( uint16_t *out, uint8_t *in, int inbytes, uint32_t *pal )
242 {
243 int i;
244 for( i=0; i<inbytes; i++ ) {
245 *out++ = (uint16_t)pal[*in & 0x0F];
246 *out++ = (uint16_t)pal[(*in >> 4)];
247 in++;
248 }
249 }
251 #define VQ_CODEBOOK_SIZE 2048 /* 256 entries * 4 pixels per quad * 2 byte pixels */
253 struct vq_codebook {
254 uint16_t quad[256][4];
255 };
257 static void vq_get_codebook( struct vq_codebook *codebook,
258 uint16_t *input )
259 {
260 /* Detwiddle the codebook, for the sake of my own sanity if nothing else */
261 uint16_t *p = (uint16_t *)input;
262 int i;
263 for( i=0; i<256; i++ ) {
264 codebook->quad[i][0] = *p++;
265 codebook->quad[i][2] = *p++;
266 codebook->quad[i][1] = *p++;
267 codebook->quad[i][3] = *p++;
268 }
269 }
271 static void vq_decode( uint16_t *output, unsigned char *input, int width, int height,
272 struct vq_codebook *codebook ) {
273 int i,j;
275 uint8_t *c = (uint8_t *)input;
276 for( j=0; j<height; j+=2 ) {
277 for( i=0; i<width; i+=2 ) {
278 uint8_t code = *c++;
279 output[i + j*width] = codebook->quad[code][0];
280 output[i + 1 + j*width] = codebook->quad[code][1];
281 output[i + (j+1)*width] = codebook->quad[code][2];
282 output[i + 1 + (j+1)*width] = codebook->quad[code][3];
283 }
284 }
285 }
287 static inline uint32_t yuv_to_rgb32( float y, float u, float v )
288 {
289 u -= 128;
290 v -= 128;
291 int r = (int)(y + v*1.375);
292 int g = (int)(y - u*0.34375 - v*0.6875);
293 int b = (int)(y + u*1.71875);
294 if( r > 255 ) { r = 255; } else if( r < 0 ) { r = 0; }
295 if( g > 255 ) { g = 255; } else if( g < 0 ) { g = 0; }
296 if( b > 255 ) { b = 255; } else if( b < 0 ) { b = 0; }
297 return 0xFF000000 | (r<<16) | (g<<8) | (b);
298 }
301 /**
302 * Convert raster YUV texture data into RGB32 data - most GL implementations don't
303 * directly support this format unfortunately. The input data is formatted as
304 * 32 bits = 2 horizontal pixels, UYVY. This is currently done rather inefficiently
305 * in floating point.
306 */
307 static void yuv_decode( uint32_t *output, uint32_t *input, int width, int height )
308 {
309 int x, y;
310 uint32_t *p = input;
311 for( y=0; y<height; y++ ) {
312 for( x=0; x<width; x+=2 ) {
313 float u = (float)(*p & 0xFF);
314 float y0 = (float)( (*p>>8)&0xFF );
315 float v = (float)( (*p>>16)&0xFF );
316 float y1 = (float)( (*p>>24)&0xFF );
317 *output++ = yuv_to_rgb32( y0, u, v );
318 *output++ = yuv_to_rgb32( y1, u, v );
319 p++;
320 }
321 }
322 }
324 /**
325 * Load texture data from the given address and parameters into the currently
326 * bound OpenGL texture.
327 */
328 static void texcache_load_texture( uint32_t texture_addr, int width, int height,
329 int mode ) {
330 int bpp_shift = 1; /* bytes per (output) pixel as a power of 2 */
331 GLint intFormat = GL_RGBA, format, type;
332 int tex_format = mode & PVR2_TEX_FORMAT_MASK;
333 struct vq_codebook codebook;
334 GLint filter = GL_LINEAR;
336 glPixelStorei( GL_UNPACK_ROW_LENGTH, 0 );
337 /* Decode the format parameters */
338 switch( tex_format ) {
339 case PVR2_TEX_FORMAT_IDX4:
340 case PVR2_TEX_FORMAT_IDX8:
341 /* For indexed-colour modes, we need to lookup the palette control
342 * word to determine the de-indexed texture format.
343 */
344 switch( MMIO_READ( PVR2, RENDER_PALETTE ) & 0x03 ) {
345 case 0: /* ARGB1555 */
346 format = GL_BGRA;
347 type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
348 break;
349 case 1: /* RGB565 */
350 intFormat = GL_RGB;
351 format = GL_RGB;
352 type = GL_UNSIGNED_SHORT_5_6_5;
353 break;
354 case 2: /* ARGB4444 */
355 format = GL_BGRA;
356 type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
357 break;
358 case 3: /* ARGB8888 */
359 format = GL_BGRA;
360 type = GL_UNSIGNED_BYTE;
361 bpp_shift = 2;
362 break;
363 default:
364 return; /* Can't happen, but it makes gcc stop complaining */
365 }
366 break;
368 case PVR2_TEX_FORMAT_ARGB1555:
369 format = GL_BGRA;
370 type = GL_UNSIGNED_SHORT_1_5_5_5_REV;
371 break;
372 case PVR2_TEX_FORMAT_RGB565:
373 intFormat = GL_RGB;
374 format = GL_RGB;
375 type = GL_UNSIGNED_SHORT_5_6_5;
376 break;
377 case PVR2_TEX_FORMAT_ARGB4444:
378 format = GL_BGRA;
379 type = GL_UNSIGNED_SHORT_4_4_4_4_REV;
380 break;
381 case PVR2_TEX_FORMAT_YUV422:
382 /* YUV422 isn't directly supported by most implementations, so decode
383 * it to a (reasonably) standard ARGB32.
384 */
385 bpp_shift = 2;
386 format = GL_BGRA;
387 type = GL_UNSIGNED_BYTE;
388 break;
389 case PVR2_TEX_FORMAT_BUMPMAP:
390 ERROR( "Bumpmap not supported" );
391 return;
392 default:
393 ERROR( "Undefined texture format" );
394 return;
395 }
397 if( PVR2_TEX_IS_STRIDE(mode) && tex_format != PVR2_TEX_FORMAT_IDX4 &&
398 tex_format != PVR2_TEX_FORMAT_IDX8 ) {
399 /* Stride textures cannot be mip-mapped, compressed, indexed or twiddled */
400 uint32_t stride = (MMIO_READ( PVR2, RENDER_TEXSIZE ) & 0x003F) << 5;
401 unsigned char data[(width*height) << bpp_shift];
402 if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
403 unsigned char tmp[(width*height)<<1];
404 pvr2_vram64_read_stride( tmp, width<<1, texture_addr, stride<<1, height );
405 yuv_decode( (uint32_t *)data, (uint32_t *)tmp, width, height );
406 } else {
407 pvr2_vram64_read_stride( data, width<<bpp_shift, texture_addr, stride<<bpp_shift, height );
408 }
409 glTexImage2D( GL_TEXTURE_2D, 0, intFormat, width, height, 0, format, type, data );
410 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
411 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
412 return;
413 }
415 int level=0, last_level = 0, mip_width = width, mip_height = height, src_bytes, dest_bytes;
416 if( PVR2_TEX_IS_MIPMAPPED(mode) ) {
417 int i;
418 for( i=0; 1<<i < width; i++ );
419 last_level = i;
420 mip_width = 2;
421 mip_height= 2;
422 filter = GL_LINEAR_MIPMAP_LINEAR;
423 }
424 dest_bytes = (mip_width * mip_height) << bpp_shift;
425 src_bytes = dest_bytes; // Modes will change this (below)
427 if( PVR2_TEX_IS_COMPRESSED(mode) ) {
428 uint16_t tmp[VQ_CODEBOOK_SIZE];
429 pvr2_vram64_read( (unsigned char *)tmp, texture_addr, VQ_CODEBOOK_SIZE );
430 texture_addr += VQ_CODEBOOK_SIZE;
431 vq_get_codebook( &codebook, tmp );
432 }
434 for( level=last_level; level>= 0; level-- ) {
435 unsigned char data[dest_bytes];
436 /* load data from image, detwiddling/uncompressing as required */
437 if( tex_format == PVR2_TEX_FORMAT_IDX8 ) {
438 src_bytes = (mip_width * mip_height);
439 int bank = (mode >> 25) &0x03;
440 uint32_t *palette = ((uint32_t *)mmio_region_PVR2PAL.mem) + (bank<<8);
441 unsigned char tmp[src_bytes];
442 pvr2_vram64_read_twiddled_8( tmp, texture_addr, mip_width, mip_height );
443 if( bpp_shift == 2 ) {
444 decode_pal8_to_32( (uint32_t *)data, tmp, src_bytes, palette );
445 } else {
446 decode_pal8_to_16( (uint16_t *)data, tmp, src_bytes, palette );
447 }
448 } else if( tex_format == PVR2_TEX_FORMAT_IDX4 ) {
449 src_bytes = (mip_width * mip_height) >> 1;
450 int bank = (mode >>21 ) & 0x3F;
451 uint32_t *palette = ((uint32_t *)mmio_region_PVR2PAL.mem) + (bank<<4);
452 unsigned char tmp[src_bytes];
453 pvr2_vram64_read_twiddled_4( tmp, texture_addr, mip_width, mip_height );
454 if( bpp_shift == 2 ) {
455 decode_pal4_to_32( (uint32_t *)data, tmp, src_bytes, palette );
456 } else {
457 decode_pal4_to_16( (uint16_t *)data, tmp, src_bytes, palette );
458 }
459 } else if( tex_format == PVR2_TEX_FORMAT_YUV422 ) {
460 src_bytes = ((mip_width*mip_height)<<1);
461 unsigned char tmp[src_bytes];
462 if( PVR2_TEX_IS_TWIDDLED(mode) ) {
463 pvr2_vram64_read_twiddled_16( tmp, texture_addr, mip_width, mip_height );
464 } else {
465 pvr2_vram64_read( tmp, texture_addr, src_bytes );
466 }
467 yuv_decode( (uint32_t *)data, (uint32_t *)tmp, mip_width, mip_height );
468 } else if( PVR2_TEX_IS_COMPRESSED(mode) ) {
469 src_bytes = ((mip_width*mip_height) >> 2);
470 unsigned char tmp[src_bytes];
471 if( PVR2_TEX_IS_TWIDDLED(mode) ) {
472 pvr2_vram64_read_twiddled_8( tmp, texture_addr, mip_width>>1, mip_height>>1 );
473 } else {
474 pvr2_vram64_read( tmp, texture_addr, src_bytes );
475 }
476 vq_decode( (uint16_t *)data, tmp, mip_width, mip_height, &codebook );
477 } else if( PVR2_TEX_IS_TWIDDLED(mode) ) {
478 pvr2_vram64_read_twiddled_16( data, texture_addr, mip_width, mip_height );
479 } else {
480 pvr2_vram64_read( data, texture_addr, src_bytes );
481 }
483 /* Pass to GL */
484 if( level == last_level && level != 0 ) { /* 1x1 stored within a 2x2 */
485 glTexImage2D( GL_TEXTURE_2D, level, intFormat, 1, 1, 0, format, type,
486 data + (3 << bpp_shift) );
487 texture_addr += src_bytes;
488 } else {
489 glTexImage2D( GL_TEXTURE_2D, level, intFormat, mip_width, mip_height, 0, format, type,
490 data );
491 texture_addr += src_bytes;
492 mip_width <<= 1;
493 mip_height <<= 1;
494 dest_bytes <<= 2;
495 src_bytes <<= 2;
496 }
497 }
499 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, filter);
500 glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_LINEAR);
501 }
503 /**
504 * Return a texture ID for the texture specified at the supplied address
505 * and given parameters (the same sequence of bytes could in theory have
506 * multiple interpretations). We use the texture address as the primary
507 * index, but allow for multiple instances at each address. The texture
508 * will be bound to the GL_TEXTURE_2D target before being returned.
509 *
510 * If the texture has already been bound, return the ID to which it was
511 * bound. Otherwise obtain an unused texture ID and set it up appropriately.
512 */
513 GLuint texcache_get_texture( uint32_t texture_word, int width, int height )
514 {
515 uint32_t texture_addr = (texture_word & 0x000FFFFF)<<3;
516 uint32_t texture_page = texture_addr >> 12;
517 texcache_entry_index next;
518 texcache_entry_index idx = texcache_page_lookup[texture_page];
519 while( idx != EMPTY_ENTRY ) {
520 texcache_entry_t entry = &texcache_active_list[idx];
521 if( entry->texture_addr == texture_addr &&
522 entry->mode == texture_word &&
523 entry->width == width &&
524 entry->height == height ) {
525 entry->lru_count = texcache_ref_counter++;
526 return entry->texture_id;
527 }
528 idx = entry->next;
529 }
532 /* Not found - check the free list */
533 texcache_entry_index slot = 0;
535 if( texcache_free_ptr < MAX_TEXTURES ) {
536 slot = texcache_free_list[texcache_free_ptr++];
537 } else {
538 slot = texcache_evict_lru();
539 }
541 /* Construct new entry */
542 texcache_active_list[slot].texture_addr = texture_addr;
543 texcache_active_list[slot].width = width;
544 texcache_active_list[slot].height = height;
545 texcache_active_list[slot].mode = texture_word;
546 texcache_active_list[slot].lru_count = texcache_ref_counter++;
548 /* Add entry to the lookup table */
549 next = texcache_page_lookup[texture_page];
550 if( next == slot ) {
551 int i;
552 fprintf( stderr, "Active list: " );
553 for( i=0; i<MAX_TEXTURES; i++ ) {
554 fprintf( stderr, "%d, ", texcache_active_list[i].next );
555 }
556 fprintf( stderr, "\n" );
557 assert( next != slot );
559 }
560 assert( next != slot );
561 texcache_active_list[slot].next = next;
562 texcache_page_lookup[texture_page] = slot;
564 /* Construct the GL texture */
565 glBindTexture( GL_TEXTURE_2D, texcache_active_list[slot].texture_id );
566 texcache_load_texture( texture_addr, width, height, texture_word );
568 return texcache_active_list[slot].texture_id;
569 }
571 /**
572 * Check the integrity of the texcache. Verifies that every cache slot
573 * appears exactly once on either the free list or one page list. For
574 * active slots, the texture address must also match the page it appears on.
575 *
576 */
577 void texcache_integrity_check()
578 {
579 int i;
580 int slot_found[MAX_TEXTURES];
582 memset( slot_found, 0, sizeof(slot_found) );
584 /* Check entries on the free list */
585 for( i= texcache_free_ptr; i< MAX_TEXTURES; i++ ) {
586 int slot = texcache_free_list[i];
587 assert( slot_found[slot] == 0 );
588 assert( texcache_active_list[slot].next == EMPTY_ENTRY );
589 slot_found[slot] = 1;
590 }
592 /* Check entries on the active lists */
593 for( i=0; i< PVR2_RAM_PAGES; i++ ) {
594 int slot = texcache_page_lookup[i];
595 while( slot != EMPTY_ENTRY ) {
596 assert( slot_found[slot] == 0 );
597 assert( (texcache_active_list[slot].texture_addr >> 12) == i );
598 slot_found[slot] = 2;
599 slot = texcache_active_list[slot].next;
600 }
601 }
603 /* Make sure we didn't miss any entries */
604 for( i=0; i<MAX_TEXTURES; i++ ) {
605 assert( slot_found[i] != 0 );
606 }
607 }
.