Merge git://proxy01.pd.intel.com:9419/git/mesa/mesa into crestline

11 files changed, 324 insertions, 367 deletions

diff --git a/src/mesa/drivers/dri/i965/brw_urb.c b/src/mesa/drivers/dri/i965/brw_urb.c
index 79ff2b2d4db..64f5904ac68 100644
--- a/src/mesa/drivers/dri/i965/brw_urb.c
+++ b/src/mesa/drivers/dri/i965/brw_urb.c
@@ -131,7 +131,7 @@ static void recalculate_urb_fence( struct brw_context *brw )
 
 	 brw->urb.constrained = 1;
 	 
-	 if (check_urb_layout(brw)) {
+	 if (!check_urb_layout(brw)) {
 	    /* This is impossible, given the maximal sizes of urb
 	     * entries and the values for minimum nr of entries
 	     * provided above.
diff --git a/src/mesa/drivers/dri/r300/r300_fragprog.c b/src/mesa/drivers/dri/r300/r300_fragprog.c
index 68a75ec7f00..e01f56d99df 100644
--- a/src/mesa/drivers/dri/r300/r300_fragprog.c
+++ b/src/mesa/drivers/dri/r300/r300_fragprog.c
@@ -934,6 +934,9 @@ static void emit_tex(struct r300_fragment_program *rp,
 	int hwsrc, hwdest;
 	GLuint tempreg = 0;
 
+	uin = cs->used_in_node;
+	din = cs->dest_in_node;
+
 	/* Resolve source/dest to hardware registers */
 	if (opcode != R300_FPITX_OP_KIL) {
 		if (fpi->TexSrcTarget == TEXTURE_RECT_INDEX) {
@@ -958,6 +961,10 @@ static void emit_tex(struct r300_fragment_program *rp,
 			emit_arith(rp, PFS_OP_MAD, tempreg, WRITEMASK_XYZW,
 			           coord, factorreg, pfs_zero, 0);
 
+			/* Ensure correct node indirection */
+			uin = cs->used_in_node;
+			din = cs->dest_in_node;
+
 			hwsrc = t_hw_src(rp, tempreg, GL_TRUE);
 		} else {
 			hwsrc = t_hw_src(rp, coord, GL_TRUE);
@@ -986,8 +993,6 @@ static void emit_tex(struct r300_fragment_program *rp,
 		hwsrc = t_hw_src(rp, coord, GL_TRUE);
 	}
 
-	din = cs->dest_in_node;
-	uin = cs->used_in_node;
 
 	/* Indirection if source has been written in this node, or if the
 	 * dest has been read/written in this node
diff --git a/src/mesa/drivers/dri/r300/r300_texmem.c b/src/mesa/drivers/dri/r300/r300_texmem.c
index f531b54d115..c527677cd0c 100644
--- a/src/mesa/drivers/dri/r300/r300_texmem.c
+++ b/src/mesa/drivers/dri/r300/r300_texmem.c
@@ -303,195 +303,190 @@ static void r300UploadRectSubImage(r300ContextPtr rmesa,
  * Upload the texture image associated with texture \a t at the specified
  * level at the address relative to \a start.
  */
-static void uploadSubImage( r300ContextPtr rmesa, r300TexObjPtr t, 
+static void uploadSubImage( r300ContextPtr rmesa, r300TexObjPtr t,
 			    GLint hwlevel,
 			    GLint x, GLint y, GLint width, GLint height,
 			    GLuint face )
 {
-   struct gl_texture_image *texImage = NULL;
-   GLuint offset;
-   GLint imageWidth, imageHeight;
-   GLint ret;
-   drm_radeon_texture_t tex;
-   drm_radeon_tex_image_t tmp;
-   const int level = hwlevel + t->base.firstLevel;
-
-   if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
-      fprintf( stderr, "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n", 
-	       __FUNCTION__, (void *)t, (void *)t->base.tObj,
-	       level, width, height, face );
-   }
-
-   ASSERT(face < 6);
-
-   /* Ensure we have a valid texture to upload */
-   if ( ( hwlevel < 0 ) || ( hwlevel >= RADEON_MAX_TEXTURE_LEVELS ) ) {
-      _mesa_problem(NULL, "bad texture level in %s", __FUNCTION__);
-      return;
-   }
-
-   texImage = t->base.tObj->Image[face][level];
-
-   if ( !texImage ) {
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: texImage %d is NULL!\n", __FUNCTION__, level );
-      return;
-   }
-   if ( !texImage->Data ) {
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is NULL!\n", __FUNCTION__ );
-      return;
-   }
-
-
-   if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-      assert(level == 0);
-      assert(hwlevel == 0);
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is rectangular\n", __FUNCTION__);
-      r300UploadRectSubImage( rmesa, t, texImage, x, y, width, height );
-      return;
-   }
-   else if (texImage->IsClientData) {
-      if ( RADEON_DEBUG & DEBUG_TEXTURE )
-	 fprintf( stderr, "%s: image data is in GART client storage\n",
-		  __FUNCTION__);
-      r300UploadGARTClientSubImage( rmesa, t, texImage, hwlevel,
-				   x, y, width, height );
-      return;
-   }
-   else if ( RADEON_DEBUG & DEBUG_TEXTURE )
-      fprintf( stderr, "%s: image data is in normal memory\n",
-	       __FUNCTION__);
-      
-
-   imageWidth = texImage->Width;
-   imageHeight = texImage->Height;
-
-   offset = t->bufAddr + t->base.totalSize / 6 * face;
-
-   if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
-      GLint imageX = 0;
-      GLint imageY = 0;
-      GLint blitX = t->image[face][hwlevel].x;
-      GLint blitY = t->image[face][hwlevel].y;
-      GLint blitWidth = t->image[face][hwlevel].width;
-      GLint blitHeight = t->image[face][hwlevel].height;
-      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
-	       imageWidth, imageHeight, imageX, imageY );
-      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
-	       blitWidth, blitHeight, blitX, blitY );
-      fprintf( stderr, "       blit ofs: 0x%07x level: %d/%d\n",
-	       (GLuint)offset, hwlevel, level );
-   }
-
-   t->image[face][hwlevel].data = texImage->Data;
-
-   /* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct.
-    * NOTE: we're always use a 1KB-wide blit and I8 texture format.
-    * We used to use 1, 2 and 4-byte texels and used to use the texture
-    * width to dictate the blit width - but that won't work for compressed
-    * textures. (Brian)
-    * NOTE: can't do that with texture tiling. (sroland)
-    */
-   tex.offset = offset;
-   tex.image = &tmp;
-   /* copy (x,y,width,height,data) */
-   memcpy( &tmp, &t->image[face][hwlevel], sizeof(tmp) );
-   
-   if (texImage->TexFormat->TexelBytes > 4) {
-      const int log2TexelBytes = (3 + (texImage->TexFormat->TexelBytes >> 4));
-      tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */
-      tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1);
-      tex.height = imageHeight;
-      tex.width = imageWidth << log2TexelBytes;
-      tex.offset += (tmp.x << log2TexelBytes) & ~1023;
-      tmp.x = tmp.x % (1024 >> log2TexelBytes);
-      tmp.width = tmp.width << log2TexelBytes;
-   }
-   else if (texImage->TexFormat->TexelBytes) {
-      /* use multi-byte upload scheme */
-      tex.height = imageHeight;
-      tex.width = imageWidth;
-      switch(texImage->TexFormat->TexelBytes) {
-      case 1:
-	tex.format = RADEON_TXFORMAT_I8;
-	break;
-      case 2:
-	tex.format = RADEON_TXFORMAT_AI88;
-	break;
-      case 4:
-	tex.format = RADEON_TXFORMAT_ARGB8888;
-	break;
-      }
-      tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1);
-      tex.offset += tmp.x & ~1023;
-      tmp.x = tmp.x % 1024;
-#if 1
-      if (t->tile_bits & R300_TXO_MICRO_TILE) {
-	 /* need something like "tiled coordinates" ? */
-	 tmp.y = tmp.x / (tex.pitch * 128) * 2;
-	 tmp.x = tmp.x % (tex.pitch * 128) / 2 / texImage->TexFormat->TexelBytes;
-	 tex.pitch |= RADEON_DST_TILE_MICRO >> 22;
-      }
-      else 
-#endif
-      {
-	 tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1);
-      }
+	struct gl_texture_image *texImage = NULL;
+	GLuint offset;
+	GLint imageWidth, imageHeight;
+	GLint ret;
+	drm_radeon_texture_t tex;
+	drm_radeon_tex_image_t tmp;
+	const int level = hwlevel + t->base.firstLevel;
+
+	if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
+		fprintf( stderr, "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n",
+			__FUNCTION__, (void *)t, (void *)t->base.tObj,
+			level, width, height, face );
+	}
+
+	ASSERT(face < 6);
+
+	/* Ensure we have a valid texture to upload */
+	if ( ( hwlevel < 0 ) || ( hwlevel >= RADEON_MAX_TEXTURE_LEVELS ) ) {
+		_mesa_problem(NULL, "bad texture level in %s", __FUNCTION__);
+		return;
+	}
+
+	texImage = t->base.tObj->Image[face][level];
+
+	if ( !texImage ) {
+		if ( RADEON_DEBUG & DEBUG_TEXTURE )
+		fprintf( stderr, "%s: texImage %d is NULL!\n", __FUNCTION__, level );
+		return;
+	}
+	if ( !texImage->Data ) {
+		if ( RADEON_DEBUG & DEBUG_TEXTURE )
+		fprintf( stderr, "%s: image data is NULL!\n", __FUNCTION__ );
+		return;
+	}
+
+
+	if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+		assert(level == 0);
+		assert(hwlevel == 0);
+		if ( RADEON_DEBUG & DEBUG_TEXTURE )
+		fprintf( stderr, "%s: image data is rectangular\n", __FUNCTION__);
+		r300UploadRectSubImage( rmesa, t, texImage, x, y, width, height );
+		return;
+	} else if (texImage->IsClientData) {
+		if ( RADEON_DEBUG & DEBUG_TEXTURE )
+		fprintf( stderr, "%s: image data is in GART client storage\n",
+			__FUNCTION__);
+		r300UploadGARTClientSubImage( rmesa, t, texImage, hwlevel,
+					x, y, width, height );
+		return;
+	} else if ( RADEON_DEBUG & DEBUG_TEXTURE )
+		fprintf( stderr, "%s: image data is in normal memory\n",
+			__FUNCTION__);
+
+
+	imageWidth = texImage->Width;
+	imageHeight = texImage->Height;
+
+	offset = t->bufAddr + t->base.totalSize / 6 * face;
+
+	if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
+		GLint imageX = 0;
+		GLint imageY = 0;
+		GLint blitX = t->image[face][hwlevel].x;
+		GLint blitY = t->image[face][hwlevel].y;
+		GLint blitWidth = t->image[face][hwlevel].width;
+		GLint blitHeight = t->image[face][hwlevel].height;
+		fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
+			imageWidth, imageHeight, imageX, imageY );
+		fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
+			blitWidth, blitHeight, blitX, blitY );
+		fprintf( stderr, "       blit ofs: 0x%07x level: %d/%d\n",
+			(GLuint)offset, hwlevel, level );
+	}
+
+	t->image[face][hwlevel].data = texImage->Data;
+
+	/* Init the DRM_RADEON_TEXTURE command / drm_radeon_texture_t struct.
+	 * NOTE: we're always use a 1KB-wide blit and I8 texture format.
+	 * We used to use 1, 2 and 4-byte texels and used to use the texture
+	 * width to dictate the blit width - but that won't work for compressed
+	 * textures. (Brian)
+	 * NOTE: can't do that with texture tiling. (sroland)
+	 */
+	tex.offset = offset;
+	tex.image = &tmp;
+	/* copy (x,y,width,height,data) */
+	memcpy( &tmp, &t->image[face][hwlevel], sizeof(tmp) );
+
+	if (texImage->TexFormat->TexelBytes > 4) {
+		const int log2TexelBytes = (3 + (texImage->TexFormat->TexelBytes >> 4));
+		tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */
+		tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1);
+		tex.height = imageHeight;
+		tex.width = imageWidth << log2TexelBytes;
+		tex.offset += (tmp.x << log2TexelBytes) & ~1023;
+		tmp.x = tmp.x % (1024 >> log2TexelBytes);
+		tmp.width = tmp.width << log2TexelBytes;
+	} else if (texImage->TexFormat->TexelBytes) {
+		/* use multi-byte upload scheme */
+		tex.height = imageHeight;
+		tex.width = imageWidth;
+		switch(texImage->TexFormat->TexelBytes) {
+		case 1:
+			tex.format = RADEON_TXFORMAT_I8;
+			break;
+		case 2:
+			tex.format = RADEON_TXFORMAT_AI88;
+			break;
+		case 4:
+			tex.format = RADEON_TXFORMAT_ARGB8888;
+			break;
+		}
+		tex.pitch = MAX2((texImage->Width * texImage->TexFormat->TexelBytes) / 64, 1);
+		tex.offset += tmp.x & ~1023;
+		tmp.x = tmp.x % 1024;
+
+		if (t->tile_bits & R300_TXO_MICRO_TILE) {
+			/* need something like "tiled coordinates" ? */
+			tmp.y = tmp.x / (tex.pitch * 128) * 2;
+			tmp.x = tmp.x % (tex.pitch * 128) / 2 / texImage->TexFormat->TexelBytes;
+			tex.pitch |= RADEON_DST_TILE_MICRO >> 22;
+		} else {
+			tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1);
+		}
 #if 1
-      if ((t->tile_bits & R300_TXO_MACRO_TILE) &&
-	 (texImage->Width * texImage->TexFormat->TexelBytes >= 256) &&
-	 ((!(t->tile_bits & R300_TXO_MICRO_TILE) && (texImage->Height >= 8)) ||
-	    (texImage->Height >= 16))) {
-	 /* weird: R200 disables macro tiling if mip width is smaller than 256 bytes,
-	    OR if height is smaller than 8 automatically, but if micro tiling is active
-	    the limit is height 16 instead ? */
-	 tex.pitch |= RADEON_DST_TILE_MACRO >> 22;
-      }
+		if ((t->tile_bits & R300_TXO_MACRO_TILE) &&
+		    (texImage->Width * texImage->TexFormat->TexelBytes >= 256) &&
+		    ((!(t->tile_bits & R300_TXO_MICRO_TILE) && (texImage->Height >= 8)) ||
+		     (texImage->Height >= 16))) {
+			/* weird: R200 disables macro tiling if mip width is smaller than 256 bytes,
+			   OR if height is smaller than 8 automatically, but if micro tiling is active
+			   the limit is height 16 instead ? */
+			tex.pitch |= RADEON_DST_TILE_MACRO >> 22;
+		}
 #endif
-   }
-   else {
-      /* In case of for instance 8x8 texture (2x2 dxt blocks), padding after the first two blocks is
-         needed (only with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */
-      /* set tex.height to 1/4 since 1 "macropixel" (dxt-block) has 4 real pixels. Needed
-         so the kernel module reads the right amount of data. */
-      tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */
-      tex.pitch = (R300_BLIT_WIDTH_BYTES / 64);
-      tex.height = (imageHeight + 3) / 4;
-      tex.width = (imageWidth + 3) / 4;
-      if ((t->format & R300_TX_FORMAT_DXT1) == R300_TX_FORMAT_DXT1)
-      {
-           tex.width *= 8;
-      } else {
-           tex.width *= 16;
-      }
-   }
-
-   LOCK_HARDWARE( &rmesa->radeon );
-   do {
-      ret = drmCommandWriteRead( rmesa->radeon.dri.fd, DRM_RADEON_TEXTURE,
-                                 &tex, sizeof(drm_radeon_texture_t) );
-      if (ret) {
-	 if (RADEON_DEBUG & DEBUG_IOCTL)
-	    fprintf(stderr, "DRM_RADEON_TEXTURE:  again!\n");
-	 usleep(1);
-      }
-   } while ( ret == -EAGAIN );
-
-   UNLOCK_HARDWARE( &rmesa->radeon );
-
-   if ( ret ) {
-      fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret );
-      fprintf( stderr, "   offset=0x%08x\n",
-	       offset );
-      fprintf( stderr, "   image width=%d height=%d\n",
-	       imageWidth, imageHeight );
-      fprintf( stderr, "    blit width=%d height=%d data=%p\n",
-	       t->image[face][hwlevel].width, t->image[face][hwlevel].height,
-	       t->image[face][hwlevel].data );
-      exit( 1 );
-   }
+	} else {
+		/* In case of for instance 8x8 texture (2x2 dxt blocks),
+		   padding after the first two blocks is needed (only
+		   with dxt1 since 2 dxt3/dxt5 blocks already use 32 Byte). */
+		/* set tex.height to 1/4 since 1 "macropixel" (dxt-block)
+		   has 4 real pixels. Needed so the kernel module reads
+		   the right amount of data. */
+		tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */
+		tex.pitch = (R300_BLIT_WIDTH_BYTES / 64);
+		tex.height = (imageHeight + 3) / 4;
+		tex.width = (imageWidth + 3) / 4;
+		if ((t->format & R300_TX_FORMAT_DXT1) == R300_TX_FORMAT_DXT1)
+		{
+			tex.width *= 8;
+		} else {
+			tex.width *= 16;
+		}
+	}
+
+	LOCK_HARDWARE( &rmesa->radeon );
+	do {
+		ret = drmCommandWriteRead( rmesa->radeon.dri.fd, DRM_RADEON_TEXTURE,
+		                           &tex, sizeof(drm_radeon_texture_t) );
+		if (ret) {
+		if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "DRM_RADEON_TEXTURE:  again!\n");
+		usleep(1);
+		}
+	} while ( ret == -EAGAIN );
+
+	UNLOCK_HARDWARE( &rmesa->radeon );
+
+	if ( ret ) {
+		fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret );
+		fprintf( stderr, "   offset=0x%08x\n",
+				offset );
+		fprintf( stderr, "   image width=%d height=%d\n",
+				imageWidth, imageHeight );
+		fprintf( stderr, "    blit width=%d height=%d data=%p\n",
+				t->image[face][hwlevel].width, t->image[face][hwlevel].height,
+				t->image[face][hwlevel].data );
+		exit( 1 );
+	}
 }
 
 /**
diff --git a/src/mesa/drivers/dri/r300/r300_texstate.c b/src/mesa/drivers/dri/r300/r300_texstate.c
index 4bc0ea14f81..14b0c6063b7 100644
--- a/src/mesa/drivers/dri/r300/r300_texstate.c
+++ b/src/mesa/drivers/dri/r300/r300_texstate.c
@@ -218,7 +218,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 		if (rmesa->texmicrotile  && (tObj->Target != GL_TEXTURE_RECTANGLE_NV) &&
 		   /* texrect might be able to use micro tiling too in theory? */
 		   (baseImage->Height > 1)) {
-			
+
 			/* allow 32 (bytes) x 1 mip (which will use two times the space
 			   the non-tiled version would use) max if base texture is large enough */
 			if ((numLevels == 1) ||
@@ -228,7 +228,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 				t->tile_bits |= R300_TXO_MICRO_TILE;
 			}
 		}
-		
+
 		if (tObj->Target != GL_TEXTURE_RECTANGLE_NV) {
 			/* we can set macro tiling even for small textures, they will be untiled anyway */
 			t->tile_bits |= R300_TXO_MACRO_TILE;
@@ -237,91 +237,85 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 #endif
 
 	for (i = 0; i < numLevels; i++) {
-	  const struct gl_texture_image *texImage;
-	  GLuint size;
-	  
-	  texImage = tObj->Image[0][i + t->base.firstLevel];
-	  if (!texImage)
-	    break;
-	  
-	  /* find image size in bytes */
-	  if (texImage->IsCompressed) {
-	    if ((t->format & R300_TX_FORMAT_DXT1) == R300_TX_FORMAT_DXT1) {
-	      // fprintf(stderr,"DXT 1 %d %08X\n", texImage->Width, t->format);
-	      if ((texImage->Width + 3) < 8) /* width one block */
-		size = texImage->CompressedSize * 4;
-	      else if ((texImage->Width + 3) < 16)
-		size = texImage->CompressedSize * 2;
-	      else size = texImage->CompressedSize;
-	    }
-	    else /* DXT3/5, 16 bytes per block */
-	    {
-	      WARN_ONCE("DXT 3/5 suffers from multitexturing problems!\n");
-	      // fprintf(stderr,"DXT 3/5 %d\n", texImage->Width);
-	      if ((texImage->Width + 3) < 8)
-		size = texImage->CompressedSize * 2;
-	      else size = texImage->CompressedSize;
-	    }
-	    
-	  } else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-	    size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height;
-	    blitWidth = 64 / texelBytes;
-	  } else if (t->tile_bits & R300_TXO_MICRO_TILE) {
-		/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
-		   though the actual offset may be different (if texture is less than
-		   32 bytes width) to the untiled case */
-		int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
-		size = (w * ((texImage->Height + 1) / 2)) * texImage->Depth;
-		blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-	  } else {
-	    int w = (texImage->Width * texelBytes + 31) & ~31;
-	    size = w * texImage->Height * texImage->Depth;
-	    blitWidth = MAX2(texImage->Width, 64 / texelBytes);
-	  }
-	  assert(size > 0);
-	  
-	  if(0)
-	    fprintf(stderr, "w=%d h=%d d=%d tb=%d intFormat=%d\n", texImage->Width, texImage->Height,
-		    texImage->Depth, texImage->TexFormat->TexelBytes,
-		    texImage->InternalFormat);
-	  
-	  /* Align to 32-byte offset.  It is faster to do this unconditionally
-	   * (no branch penalty).
-	   */
-	  
-	  curOffset = (curOffset + 0x1f) & ~0x1f;
-	  
-	  if (texelBytes) {
-	    t->image[0][i].x = curOffset; /* fix x and y coords up later together with offset */
-	    t->image[0][i].y = 0;
-	    t->image[0][i].width = MIN2(size / texelBytes, blitWidth);
-	    t->image[0][i].height = (size / texelBytes) / t->image[0][i].width;
-	  } else {
-	    t->image[0][i].x = curOffset % R300_BLIT_WIDTH_BYTES;
-	    t->image[0][i].y = curOffset / R300_BLIT_WIDTH_BYTES;
-	    t->image[0][i].width = MIN2(size, R300_BLIT_WIDTH_BYTES);
-	    t->image[0][i].height = size / t->image[0][i].width;
-	  }
-#if 0
-	  /* for debugging only and only  applicable to non-rectangle targets */
-	  assert(size % t->image[0][i].width == 0);
-	  assert(t->image[0][i].x == 0
-		 || (size < R300_BLIT_WIDTH_BYTES
-		     && t->image[0][i].height == 1));
-#endif
-	  
-	  if (0)
-	    fprintf(stderr,
-		    "level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
-		    i, texImage->Width, texImage->Height,
-		    t->image[0][i].x, t->image[0][i].y,
-		    t->image[0][i].width, t->image[0][i].height,
-		    size, curOffset);
-	  
-	  curOffset += size;
-	  
+		const struct gl_texture_image *texImage;
+		GLuint size;
+
+		texImage = tObj->Image[0][i + t->base.firstLevel];
+		if (!texImage)
+			break;
+
+		/* find image size in bytes */
+		if (texImage->IsCompressed) {
+			if ((t->format & R300_TX_FORMAT_DXT1) == R300_TX_FORMAT_DXT1) {
+				// fprintf(stderr,"DXT 1 %d %08X\n", texImage->Width, t->format);
+				if ((texImage->Width + 3) < 8) /* width one block */
+					size = texImage->CompressedSize * 4;
+				else if ((texImage->Width + 3) < 16)
+					size = texImage->CompressedSize * 2;
+				else
+					size = texImage->CompressedSize;
+			} else {
+				/* DXT3/5, 16 bytes per block */
+				WARN_ONCE("DXT 3/5 suffers from multitexturing problems!\n");
+				// fprintf(stderr,"DXT 3/5 %d\n", texImage->Width);
+				if ((texImage->Width + 3) < 8)
+					size = texImage->CompressedSize * 2;
+				else
+					size = texImage->CompressedSize;
+			}
+		} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+			size = ((texImage->Width * texelBytes + 63) & ~63) * texImage->Height;
+			blitWidth = 64 / texelBytes;
+		} else if (t->tile_bits & R300_TXO_MICRO_TILE) {
+			/* tile pattern is 16 bytes x2. mipmaps stay 32 byte aligned,
+				though the actual offset may be different (if texture is less than
+				32 bytes width) to the untiled case */
+			int w = (texImage->Width * texelBytes * 2 + 31) & ~31;
+			size = (w * ((texImage->Height + 1) / 2)) * texImage->Depth;
+			blitWidth = MAX2(texImage->Width, 64 / texelBytes);
+		} else {
+			int w = (texImage->Width * texelBytes + 31) & ~31;
+			size = w * texImage->Height * texImage->Depth;
+			blitWidth = MAX2(texImage->Width, 64 / texelBytes);
+		}
+		assert(size > 0);
+
+		if(0)
+			fprintf(stderr, "w=%d h=%d d=%d tb=%d intFormat=%d\n",
+					texImage->Width, texImage->Height,
+					texImage->Depth, texImage->TexFormat->TexelBytes,
+					texImage->InternalFormat);
+
+		/* Align to 32-byte offset.  It is faster to do this unconditionally
+		 * (no branch penalty).
+		 */
+
+		curOffset = (curOffset + 0x1f) & ~0x1f;
+
+		if (texelBytes) {
+			/* fix x and y coords up later together with offset */
+			t->image[0][i].x = curOffset;
+			t->image[0][i].y = 0;
+			t->image[0][i].width = MIN2(size / texelBytes, blitWidth);
+			t->image[0][i].height = (size / texelBytes) / t->image[0][i].width;
+		} else {
+			t->image[0][i].x = curOffset % R300_BLIT_WIDTH_BYTES;
+			t->image[0][i].y = curOffset / R300_BLIT_WIDTH_BYTES;
+			t->image[0][i].width = MIN2(size, R300_BLIT_WIDTH_BYTES);
+			t->image[0][i].height = size / t->image[0][i].width;
+		}
+
+		if (0)
+			fprintf(stderr,
+					"level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
+					i, texImage->Width, texImage->Height,
+					t->image[0][i].x, t->image[0][i].y,
+					t->image[0][i].width, t->image[0][i].height,
+					size, curOffset);
+
+		curOffset += size;
 	}
-	
+
 	/* Align the total size of texture memory block.
 	 */
 	t->base.totalSize =
@@ -361,7 +355,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	} else if (tObj->Target == GL_TEXTURE_CUBE_MAP) {
 		ASSERT(log2Width == log2Height);
 		t->format |= R300_TX_FORMAT_CUBIC_MAP;
-		
+
 		t->format_x |= R200_TEXCOORD_CUBIC_ENV;
 		t->pp_cubic_faces = ((log2Width << R200_FACE_WIDTH_1_SHIFT) |
 				     (log2Height << R200_FACE_HEIGHT_1_SHIFT) |
@@ -377,7 +371,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 		ASSERT(log2Width == log2Height);
 		t->format |= R300_TX_FORMAT_CUBIC_MAP;
 	}
-	
+
 	t->size = (((tObj->Image[0][t->base.firstLevel]->Width - 1) << R300_TX_WIDTHMASK_SHIFT)
 			|((tObj->Image[0][t->base.firstLevel]->Height - 1) << R300_TX_HEIGHTMASK_SHIFT))
 			|((numLevels - 1) << R300_TX_MAX_MIP_LEVEL_SHIFT);
diff --git a/src/mesa/drivers/x11/xm_buffer.c b/src/mesa/drivers/x11/xm_buffer.c
index c1fa23328f0..bb8fe31ce88 100644
--- a/src/mesa/drivers/x11/xm_buffer.c
+++ b/src/mesa/drivers/x11/xm_buffer.c
@@ -422,18 +422,6 @@ xmesa_delete_framebuffer(struct gl_framebuffer *fb)
       XMesaDestroyImage( b->rowimage );
    }
 
-   /* Note that XMesaBuffer renderbuffers normally have a refcount of 2
-    * (creation + binding) so we need to explicitly delete/unbind them here.
-    */
-   if (b->frontxrb) {
-      _mesa_unreference_renderbuffer((struct gl_renderbuffer **) &b->frontxrb);
-      ASSERT(b->frontxrb == NULL);
-   }
-   if (b->backxrb) {
-      _mesa_unreference_renderbuffer((struct gl_renderbuffer **) &b->backxrb);
-      ASSERT(b->backxrb == NULL);
-   }
-
    _mesa_free_framebuffer_data(fb);
    _mesa_free(fb);
 }
diff --git a/src/mesa/main/api_loopback.c b/src/mesa/main/api_loopback.c
index 717ef1fc8f0..efe5a77d581 100644
--- a/src/mesa/main/api_loopback.c
+++ b/src/mesa/main/api_loopback.c
@@ -146,7 +146,7 @@ static void GLAPIENTRY
 loopback_Color3iv_f( const GLint *v )
 {
    COLORF( INT_TO_FLOAT(v[0]), INT_TO_FLOAT(v[1]),
-	   INT_TO_FLOAT(v[2]), INT_TO_FLOAT(v[3]) );
+	   INT_TO_FLOAT(v[2]), 1.0 );
 }
 
 static void GLAPIENTRY
diff --git a/src/mesa/main/fbobject.c b/src/mesa/main/fbobject.c
index f7e870b49cd..fefa14e5036 100644
--- a/src/mesa/main/fbobject.c
+++ b/src/mesa/main/fbobject.c
@@ -559,7 +559,7 @@ _mesa_IsRenderbufferEXT(GLuint renderbuffer)
 void GLAPIENTRY
 _mesa_BindRenderbufferEXT(GLenum target, GLuint renderbuffer)
 {
-   struct gl_renderbuffer *newRb, *oldRb;
+   struct gl_renderbuffer *newRb;
    GET_CURRENT_CONTEXT(ctx);
 
    ASSERT_OUTSIDE_BEGIN_END(ctx);
@@ -593,21 +593,16 @@ _mesa_BindRenderbufferEXT(GLenum target, GLuint renderbuffer)
 	 }
          ASSERT(newRb->AllocStorage);
          _mesa_HashInsert(ctx->Shared->RenderBuffers, renderbuffer, newRb);
+         newRb->RefCount = 1; /* referenced by hash table */
       }
-      newRb->RefCount++;
    }
    else {
       newRb = NULL;
    }
 
-   oldRb = ctx->CurrentRenderbuffer;
-   if (oldRb) {
-      _mesa_unreference_renderbuffer(&oldRb);
-   }
-
    ASSERT(newRb != &DummyRenderbuffer);
 
-   ctx->CurrentRenderbuffer = newRb;
+   _mesa_reference_renderbuffer(&ctx->CurrentRenderbuffer, newRb);
 }
 
 
@@ -632,14 +627,15 @@ _mesa_DeleteRenderbuffersEXT(GLsizei n, const GLuint *renderbuffers)
                _mesa_BindRenderbufferEXT(GL_RENDERBUFFER_EXT, 0);
             }
 
-	    /* remove from hash table immediately, to free the ID */
+	    /* Remove from hash table immediately, to free the ID.
+             * But the object will not be freed until it's no longer
+             * referenced anywhere else.
+             */
 	    _mesa_HashRemove(ctx->Shared->RenderBuffers, renderbuffers[i]);
 
             if (rb != &DummyRenderbuffer) {
-               /* But the object will not be freed until it's no longer
-                * bound in any context.
-                */
-               _mesa_unreference_renderbuffer(&rb);
+               /* no longer referenced by hash table */
+               _mesa_reference_renderbuffer(&rb, NULL);
 	    }
 	 }
       }
diff --git a/src/mesa/main/framebuffer.c b/src/mesa/main/framebuffer.c
index cd4f594aa29..1fd31a53211 100644
--- a/src/mesa/main/framebuffer.c
+++ b/src/mesa/main/framebuffer.c
@@ -70,42 +70,6 @@ compute_depth_max(struct gl_framebuffer *fb)
 
 
 /**
- * Set the framebuffer's _DepthBuffer field, taking care of
- * reference counts, etc.
- */
-static void
-set_depth_renderbuffer(struct gl_framebuffer *fb,
-                       struct gl_renderbuffer *rb)
-{
-   if (fb->_DepthBuffer) {
-      _mesa_unreference_renderbuffer(&fb->_DepthBuffer);
-   }
-   fb->_DepthBuffer = rb;
-   if (rb) {
-      rb->RefCount++;
-   }
-}
-
-
-/**
- * Set the framebuffer's _StencilBuffer field, taking care of
- * reference counts, etc.
- */
-static void
-set_stencil_renderbuffer(struct gl_framebuffer *fb,
-                         struct gl_renderbuffer *rb)
-{
-   if (fb->_StencilBuffer) {
-      _mesa_unreference_renderbuffer(&fb->_StencilBuffer);
-   }
-   fb->_StencilBuffer = rb;
-   if (rb) {
-      rb->RefCount++;
-   }
-}
-
-
-/**
  * Create and initialize a gl_framebuffer object.
  * This is intended for creating _window_system_ framebuffers, not generic
  * framebuffer objects ala GL_EXT_framebuffer_object.
@@ -223,7 +187,7 @@ _mesa_free_framebuffer_data(struct gl_framebuffer *fb)
    for (i = 0; i < BUFFER_COUNT; i++) {
       struct gl_renderbuffer_attachment *att = &fb->Attachment[i];
       if (att->Renderbuffer) {
-         _mesa_unreference_renderbuffer(&att->Renderbuffer);
+         _mesa_reference_renderbuffer(&att->Renderbuffer, NULL);
       }
       if (att->Texture) {
          /* render to texture */
@@ -239,9 +203,9 @@ _mesa_free_framebuffer_data(struct gl_framebuffer *fb)
       att->Texture = NULL;
    }
 
-   /* unbind depth/stencil to decr ref counts */
-   set_depth_renderbuffer(fb, NULL);
-   set_stencil_renderbuffer(fb, NULL);
+   /* unbind _Depth/_StencilBuffer to decr ref counts */
+   _mesa_reference_renderbuffer(&fb->_DepthBuffer, NULL);
+   _mesa_reference_renderbuffer(&fb->_StencilBuffer, NULL);
 }
 
 
@@ -569,13 +533,13 @@ _mesa_update_depth_buffer(GLcontext *ctx,
          /* need to update wrapper */
          struct gl_renderbuffer *wrapper
             = _mesa_new_z24_renderbuffer_wrapper(ctx, depthRb);
-         set_depth_renderbuffer(fb, wrapper);
+         _mesa_reference_renderbuffer(&fb->_DepthBuffer, wrapper);
          ASSERT(fb->_DepthBuffer->Wrapped == depthRb);
       }
    }
    else {
       /* depthRb may be null */
-      set_depth_renderbuffer(fb, depthRb);
+      _mesa_reference_renderbuffer(&fb->_DepthBuffer, depthRb);
    }
 }
 
@@ -610,13 +574,13 @@ _mesa_update_stencil_buffer(GLcontext *ctx,
          /* need to update wrapper */
          struct gl_renderbuffer *wrapper
             = _mesa_new_s8_renderbuffer_wrapper(ctx, stencilRb);
-         set_stencil_renderbuffer(fb, wrapper);
+         _mesa_reference_renderbuffer(&fb->_StencilBuffer, wrapper);
          ASSERT(fb->_StencilBuffer->Wrapped == stencilRb);
       }
    }
    else {
       /* stencilRb may be null */
-      set_stencil_renderbuffer(fb, stencilRb);
+      _mesa_reference_renderbuffer(&fb->_StencilBuffer, stencilRb);
    }
 }
 
diff --git a/src/mesa/main/rbadaptors.c b/src/mesa/main/rbadaptors.c
index 60f4948becf..c1ac0606c86 100644
--- a/src/mesa/main/rbadaptors.c
+++ b/src/mesa/main/rbadaptors.c
@@ -45,7 +45,7 @@ Delete_wrapper(struct gl_renderbuffer *rb)
    /* Decrement reference count on the buffer we're wrapping and delete
     * it if refcount hits zero.
     */
-   _mesa_unreference_renderbuffer(&rb->Wrapped);
+   _mesa_reference_renderbuffer(&rb->Wrapped, NULL);
 
    /* delete myself */
    _mesa_delete_renderbuffer(rb);
diff --git a/src/mesa/main/renderbuffer.c b/src/mesa/main/renderbuffer.c
index ded0063c739..49706b52516 100644
--- a/src/mesa/main/renderbuffer.c
+++ b/src/mesa/main/renderbuffer.c
@@ -1473,7 +1473,7 @@ _mesa_init_renderbuffer(struct gl_renderbuffer *rb, GLuint name)
 
    rb->ClassID = 0;
    rb->Name = name;
-   rb->RefCount = 1;
+   rb->RefCount = 0;
    rb->Delete = _mesa_delete_renderbuffer;
 
    /* The rest of these should be set later by the caller of this function or
@@ -2105,9 +2105,7 @@ _mesa_add_renderbuffer(struct gl_framebuffer *fb,
 
    fb->Attachment[bufferName].Type = GL_RENDERBUFFER_EXT;
    fb->Attachment[bufferName].Complete = GL_TRUE;
-   fb->Attachment[bufferName].Renderbuffer = rb;
-
-   rb->RefCount++;
+   _mesa_reference_renderbuffer(&fb->Attachment[bufferName].Renderbuffer, rb);
 }
 
 
@@ -2125,38 +2123,55 @@ _mesa_remove_renderbuffer(struct gl_framebuffer *fb, GLuint bufferName)
    if (!rb)
       return;
 
-   _mesa_unreference_renderbuffer(&rb);
+   _mesa_reference_renderbuffer(&rb, NULL);
 
    fb->Attachment[bufferName].Renderbuffer = NULL;
 }
 
 
 /**
- * Decrement a renderbuffer object's reference count and delete it when
- * the refcount hits zero.
- * Note: we pass the address of a pointer.
+ * Set *ptr to point to rb.  If *ptr points to another renderbuffer,
+ * dereference that buffer first.  The new renderbuffer's refcount will
+ * be incremented.  The old renderbuffer's refcount will be decremented.
  */
 void
-_mesa_unreference_renderbuffer(struct gl_renderbuffer **rb)
+_mesa_reference_renderbuffer(struct gl_renderbuffer **ptr,
+                             struct gl_renderbuffer *rb)
 {
-   assert(rb);
-   if (*rb) {
+   assert(ptr);
+   if (*ptr == rb) {
+      /* no change */
+      return;
+   }
+
+   if (*ptr) {
+      /* Unreference the old renderbuffer */
       GLboolean deleteFlag = GL_FALSE;
+      struct gl_renderbuffer *oldRb = *ptr;
 
-      _glthread_LOCK_MUTEX((*rb)->Mutex);
-      ASSERT((*rb)->RefCount > 0);
-      (*rb)->RefCount--;
-      deleteFlag = ((*rb)->RefCount == 0);
-      _glthread_UNLOCK_MUTEX((*rb)->Mutex);
+      _glthread_LOCK_MUTEX(oldRb->Mutex);
+      ASSERT(oldRb->RefCount > 0);
+      oldRb->RefCount--;
+      /*printf("RB DECR %p to %d\n", (void*) oldRb, oldRb->RefCount);*/
+      deleteFlag = (oldRb->RefCount == 0);
+      _glthread_UNLOCK_MUTEX(oldRb->Mutex);
 
       if (deleteFlag)
-         (*rb)->Delete(*rb);
+         oldRb->Delete(oldRb);
 
-      *rb = NULL;
+      *ptr = NULL;
    }
-}
-
+   assert(!*ptr);
 
+   if (rb) {
+      /* reference new renderbuffer */
+      _glthread_LOCK_MUTEX(rb->Mutex);
+      rb->RefCount++;
+      /*printf("RB REF  %p to %d\n", (void*)rb, rb->RefCount);*/
+      _glthread_UNLOCK_MUTEX(rb->Mutex);
+      *ptr = rb;
+   }
+}
 
 
 /**
@@ -2180,4 +2195,3 @@ _mesa_new_depthstencil_renderbuffer(GLcontext *ctx, GLuint name)
 
    return dsrb;
 }
-
diff --git a/src/mesa/main/renderbuffer.h b/src/mesa/main/renderbuffer.h
index e5f1147b4a6..c9bf8885487 100644
--- a/src/mesa/main/renderbuffer.h
+++ b/src/mesa/main/renderbuffer.h
@@ -102,7 +102,8 @@ extern void
 _mesa_remove_renderbuffer(struct gl_framebuffer *fb, GLuint bufferName);
 
 extern void
-_mesa_unreference_renderbuffer(struct gl_renderbuffer **rb);
+_mesa_reference_renderbuffer(struct gl_renderbuffer **ptr,
+                             struct gl_renderbuffer *rb);
 
 extern struct gl_renderbuffer *
 _mesa_new_depthstencil_renderbuffer(GLcontext *ctx, GLuint name);