1 /* -*- mode: c; c-basic-offset: 3 -*- */
3 * Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, Inc.,
7 * Permission is hereby granted, free of charge, to any person obtaining a
8 * copy of this software and associated documentation files (the "Software"),
9 * to deal in the Software without restriction, including without limitation
10 * on the rights to use, copy, modify, merge, publish, distribute, sub
11 * license, and/or sell copies of the Software, and to permit persons to whom
12 * the Software is furnished to do so, subject to the following conditions:
14 * The above copyright notice and this permission notice (including the next
15 * paragraph) shall be included in all copies or substantial portions of the
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21 * ATI, PRECISION INSIGHT AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
22 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24 * USE OR OTHER DEALINGS IN THE SOFTWARE.
29 * Gareth Hughes <gareth@valinux.com>
30 * Leif Delgass <ldelgass@retinalburn.net>
31 * Jose Fonseca <j_r_fonseca@yahoo.co.uk>
34 #include "main/context.h"
35 #include "main/macros.h"
36 #include "main/simple_list.h"
37 #include "main/imports.h"
39 #include "mach64_context.h"
40 #include "mach64_ioctl.h"
41 #include "mach64_tex.h"
44 /* Destroy hardware state associated with texture `t'.
46 void mach64DestroyTexObj( mach64ContextPtr mmesa, mach64TexObjPtr t )
50 /* See if it was the driver's current object.
54 for ( i = 0 ; i < mmesa->glCtx->Const.MaxTextureUnits ; i++ )
56 if ( t == mmesa->CurrentTexObj[ i ] ) {
57 assert( t->base.bound & (1 << i) );
58 mmesa->CurrentTexObj[ i ] = NULL;
64 /* Upload the texture image associated with texture `t' at level `level'
65 * at the address relative to `start'.
67 static void mach64UploadAGPSubImage( mach64ContextPtr mmesa,
68 mach64TexObjPtr t, int level,
69 int x, int y, int width, int height )
71 mach64ScreenRec *mach64Screen = mmesa->mach64Screen;
72 struct gl_texture_image *image;
73 int texelsPerDword = 0;
77 /* Ensure we have a valid texture to upload */
78 if ( ( level < 0 ) || ( level > mmesa->glCtx->Const.MaxTextureLevels ) )
81 image = t->base.tObj->Image[0][level];
85 texelBytes = _mesa_get_format_bytes(image->TexFormat);
87 switch ( texelBytes ) {
88 case 1: texelsPerDword = 4; break;
89 case 2: texelsPerDword = 2; break;
90 case 4: texelsPerDword = 1; break;
94 /* FIXME: The subimage index calcs are wrong... */
98 height = image->Height;
101 dwords = width * height / texelsPerDword;
103 #if ENABLE_PERF_BOXES
104 /* Bump the performance counter */
105 mmesa->c_agpTextureBytes += (dwords << 2);
108 if ( MACH64_DEBUG & DEBUG_VERBOSE_API ) {
109 fprintf( stderr, "mach64UploadSubImage: %d,%d of %d,%d at %d,%d\n",
110 width, height, image->Width, image->Height, x, y );
111 fprintf( stderr, " blit ofs: 0x%07x pitch: 0x%x dwords: %d\n",
112 (GLuint)t->bufAddr, (GLint)width, dwords );
118 CARD32 *dst = (CARD32 *)((char *)mach64Screen->agpTextures.map + t->base.memBlock->ofs);
119 const GLubyte *src = (const GLubyte *) image->Data +
120 (y * image->Width + x) * texelBytes;
121 const GLuint bytes = width * height * texelBytes;
122 memcpy(dst, src, bytes);
127 /* Upload the texture image associated with texture `t' at level `level'
128 * at the address relative to `start'.
130 static void mach64UploadLocalSubImage( mach64ContextPtr mmesa,
131 mach64TexObjPtr t, int level,
132 int x, int y, int width, int height )
134 struct gl_texture_image *image;
135 int texelsPerDword = 0;
136 int imageWidth, imageHeight;
139 const int maxdwords = (MACH64_BUFFER_MAX_DWORDS - (MACH64_HOSTDATA_BLIT_OFFSET / 4));
140 CARD32 pitch, offset;
144 /* Ensure we have a valid texture to upload */
145 if ( ( level < 0 ) || ( level > mmesa->glCtx->Const.MaxTextureLevels ) )
148 image = t->base.tObj->Image[0][level];
152 texelBytes = _mesa_get_format_bytes(image->TexFormat);
154 switch ( texelBytes ) {
155 case 1: texelsPerDword = 4; break;
156 case 2: texelsPerDword = 2; break;
157 case 4: texelsPerDword = 1; break;
161 /* FIXME: The subimage index calcs are wrong... */
164 width = image->Width;
165 height = image->Height;
168 imageWidth = image->Width;
169 imageHeight = image->Height;
171 format = t->textureFormat;
173 /* The texel upload routines have a minimum width, so force the size
176 if ( imageWidth < texelsPerDword ) {
179 factor = texelsPerDword / imageWidth;
180 imageWidth = texelsPerDword;
181 imageHeight /= factor;
182 if ( imageHeight == 0 ) {
183 /* In this case, the texel converter will actually walk a
184 * texel or two off the end of the image, but normal malloc
185 * alignment should prevent it from ever causing a fault.
191 /* We can't upload to a pitch less than 64 texels so we will need to
192 * linearly upload all modified rows for textures smaller than this.
193 * This makes the x/y/width/height different for the blitter and the
196 if ( imageWidth >= 64 ) {
197 /* The texture walker and the blitter look identical */
198 pitch = imageWidth >> 3;
204 start = (y * imageWidth) & ~63;
205 end = (y + height) * imageWidth;
207 if ( end - start < 64 ) {
208 /* Handle the case where the total number of texels
216 /* Upload some number of full 64 texel blit rows */
217 factor = 64 / imageWidth;
228 /* Fixed pitch of 64 */
232 dwords = width * height / texelsPerDword;
235 #if ENABLE_PERF_BOXES
236 /* Bump the performance counter */
237 mmesa->c_textureBytes += (dwords << 2);
240 if ( MACH64_DEBUG & DEBUG_VERBOSE_API ) {
241 fprintf( stderr, "mach64UploadSubImage: %d,%d of %d,%d at %d,%d\n",
242 width, height, image->Width, image->Height, x, y );
243 fprintf( stderr, " blit ofs: 0x%07x pitch: 0x%x dwords: %d\n",
244 (GLuint)offset, (GLint)width, dwords );
247 /* Subdivide the texture if required (account for the registers added by the drm) */
248 if ( dwords <= maxdwords ) {
251 rows = (maxdwords * texelsPerDword) / (2 * width);
254 for ( i = 0, remaining = height ;
256 remaining -= rows, y += rows, i++ )
258 height = MIN2(remaining, rows);
263 const GLubyte *src = (const GLubyte *) image->Data +
264 (y * image->Width + x) * texelBytes;
266 mach64FireBlitLocked( mmesa, (void *)src, offset, pitch, format,
267 x, y, width, height );
272 mmesa->new_state |= MACH64_NEW_CONTEXT;
273 mmesa->dirty |= MACH64_UPLOAD_CONTEXT | MACH64_UPLOAD_MISC;
277 /* Upload the texture images associated with texture `t'. This might
278 * require removing our own and/or other client's texture objects to
279 * make room for these images.
281 void mach64UploadTexImages( mach64ContextPtr mmesa, mach64TexObjPtr t )
283 if ( MACH64_DEBUG & DEBUG_VERBOSE_API ) {
284 fprintf( stderr, "%s( %p, %p )\n",
285 __FUNCTION__, mmesa->glCtx, t );
289 assert(t->base.tObj);
291 if ( !t->base.memBlock ) {
294 /* NULL heaps are skipped */
295 heap = driAllocateTexture( mmesa->texture_heaps, MACH64_NR_TEX_HEAPS,
296 (driTextureObject *) t );
299 fprintf( stderr, "%s: upload texture failure, sz=%d\n", __FUNCTION__,
307 /* Set the base offset of the texture image */
308 assert(t->base.memBlock);
309 t->bufAddr = mmesa->mach64Screen->texOffset[heap] + t->base.memBlock->ofs;
311 /* Force loading the new state into the hardware */
312 mmesa->dirty |= (MACH64_UPLOAD_SCALE_3D_CNTL |
313 MACH64_UPLOAD_TEXTURE);
316 /* Let the world know we've used this memory recently */
317 driUpdateTextureLRU( (driTextureObject *) t );
319 /* Upload any images that are new */
320 if ( t->base.dirty_images[0] ) {
321 const GLint j = t->base.tObj->BaseLevel;
322 if (t->heap == MACH64_AGP_HEAP) {
323 /* Need to make sure any vertex buffers in the queue complete */
324 mach64WaitForIdleLocked( mmesa );
325 mach64UploadAGPSubImage( mmesa, t, j, 0, 0,
326 t->base.tObj->Image[0][j]->Width,
327 t->base.tObj->Image[0][j]->Height );
329 mach64UploadLocalSubImage( mmesa, t, j, 0, 0,
330 t->base.tObj->Image[0][j]->Width,
331 t->base.tObj->Image[0][j]->Height );
334 mmesa->setup.tex_cntl |= MACH64_TEX_CACHE_FLUSH;
335 t->base.dirty_images[0] = 0;
338 mmesa->dirty |= MACH64_UPLOAD_TEXTURE;
342 /* Allocate memory from the same texture heap `heap' for both textures
345 static int mach64AllocateMultiTex( mach64ContextPtr mmesa,
348 int heap, GLboolean alloc_u0 )
350 /* Both objects should be bound */
351 assert( u0->base.bound && u1->base.bound );
354 /* Evict u0 from its current heap */
355 if ( u0->base.memBlock ) {
356 assert( u0->heap != heap );
357 driSwapOutTextureObject( (driTextureObject *) u0 );
360 /* Try to allocate u0 in the chosen heap */
361 u0->heap = driAllocateTexture( &mmesa->texture_heaps[heap], 1,
362 (driTextureObject *) u0 );
364 if ( u0->heap == -1 ) {
369 /* Evict u1 from its current heap */
370 if ( u1->base.memBlock ) {
371 assert( u1->heap != heap );
372 driSwapOutTextureObject( (driTextureObject *) u1 );
375 /* Try to allocate u1 in the same heap as u0 */
376 u1->heap = driAllocateTexture( &mmesa->texture_heaps[heap], 1,
377 (driTextureObject *) u1 );
379 if ( u1->heap == -1 ) {
383 /* Bound objects are not evicted */
384 assert( u0->base.memBlock && u1->base.memBlock );
385 assert( u0->heap == u1->heap );
390 /* The mach64 needs to have both primary and secondary textures in either
391 * local or AGP memory, so we need a "buddy system" to make sure that allocation
392 * succeeds or fails for both textures.
394 void mach64UploadMultiTexImages( mach64ContextPtr mmesa,
398 if ( MACH64_DEBUG & DEBUG_VERBOSE_API ) {
399 fprintf( stderr, "%s( %p, %p %p )\n",
400 __FUNCTION__, mmesa->glCtx, t0, t1 );
404 assert(t0->base.tObj && t1->base.tObj);
406 if ( !t0->base.memBlock || !t1->base.memBlock || t0->heap != t1->heap ) {
407 mach64TexObjPtr u0 = NULL;
408 mach64TexObjPtr u1 = NULL;
409 unsigned totalSize = t0->base.totalSize + t1->base.totalSize;
413 /* Check if one of the textures is already swapped in a heap and the
414 * other texture fits in that heap.
416 if ( t0->base.memBlock && totalSize <= t0->base.heap->size ) {
419 } else if ( t1->base.memBlock && totalSize <= t1->base.heap->size ) {
427 ret = mach64AllocateMultiTex( mmesa, u0, u1, heap, GL_FALSE );
429 /* Both textures are swapped out or collocation is impossible */
433 /* Choose the heap appropriately */
434 heap = MACH64_CARD_HEAP;
436 if ( totalSize > mmesa->texture_heaps[heap]->size ) {
437 heap = MACH64_AGP_HEAP;
440 ret = mach64AllocateMultiTex( mmesa, u0, u1, heap, GL_TRUE );
443 if ( ret == -1 && heap == MACH64_CARD_HEAP ) {
444 /* Try AGP if local memory failed */
445 heap = MACH64_AGP_HEAP;
447 ret = mach64AllocateMultiTex( mmesa, u0, u1, heap, GL_TRUE );
452 * Swap out all textures from the AGP heap and re-run allocation, this
453 * should succeed in all cases.
455 fprintf( stderr, "%s: upload multi-texture failure, sz0=%d sz1=%d\n",
456 __FUNCTION__, t0->base.totalSize, t1->base.totalSize );
460 /* Set the base offset of the texture image */
461 assert(t0->base.memBlock);
462 t0->bufAddr = mmesa->mach64Screen->texOffset[heap] + t0->base.memBlock->ofs;
463 assert(t1->base.memBlock);
464 t1->bufAddr = mmesa->mach64Screen->texOffset[heap] + t1->base.memBlock->ofs;
466 /* Force loading the new state into the hardware */
467 mmesa->dirty |= (MACH64_UPLOAD_SCALE_3D_CNTL |
468 MACH64_UPLOAD_TEXTURE);
471 /* Let the world know we've used this memory recently */
472 driUpdateTextureLRU( (driTextureObject *) t0 );
473 driUpdateTextureLRU( (driTextureObject *) t1 );
475 /* Upload any images that are new */
476 if ( t0->base.dirty_images[0] ) {
477 const GLint j0 = t0->base.tObj->BaseLevel;
478 if (t0->heap == MACH64_AGP_HEAP) {
479 /* Need to make sure any vertex buffers in the queue complete */
480 mach64WaitForIdleLocked( mmesa );
481 mach64UploadAGPSubImage( mmesa, t0, j0, 0, 0,
482 t0->base.tObj->Image[0][j0]->Width,
483 t0->base.tObj->Image[0][j0]->Height );
485 mach64UploadLocalSubImage( mmesa, t0, j0, 0, 0,
486 t0->base.tObj->Image[0][j0]->Width,
487 t0->base.tObj->Image[0][j0]->Height );
489 mmesa->setup.tex_cntl |= MACH64_TEX_CACHE_FLUSH;
490 t0->base.dirty_images[0] = 0;
492 if ( t1->base.dirty_images[0] ) {
493 const GLint j1 = t1->base.tObj->BaseLevel;
494 if (t1->heap == MACH64_AGP_HEAP) {
495 /* Need to make sure any vertex buffers in the queue complete */
496 mach64WaitForIdleLocked( mmesa );
497 mach64UploadAGPSubImage( mmesa, t1, j1, 0, 0,
498 t1->base.tObj->Image[0][j1]->Width,
499 t1->base.tObj->Image[0][j1]->Height );
501 mach64UploadLocalSubImage( mmesa, t1, j1, 0, 0,
502 t1->base.tObj->Image[0][j1]->Width,
503 t1->base.tObj->Image[0][j1]->Height );
506 mmesa->setup.tex_cntl |= MACH64_TEX_CACHE_FLUSH;
507 t1->base.dirty_images[0] = 0;
510 mmesa->dirty |= MACH64_UPLOAD_TEXTURE;