Add support for texture compression to R300 driver

This isn't perfect, texcmp still has some issues with the small textures..
but its a good start

6 files changed, 409 insertions, 214 deletions

diff --git a/src/mesa/drivers/dri/r300/r300_cmdbuf.c b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
index 9ef74dd9ab..f24ebe1a91 100644
--- a/src/mesa/drivers/dri/r300/r300_cmdbuf.c
+++ b/src/mesa/drivers/dri/r300/r300_cmdbuf.c
@@ -587,6 +587,7 @@ void r300EmitBlit(r300ContextPtr rmesa,
 							__FUNCTION__);
 
 	cmd[0].header.cmd_type = R300_CMD_PACKET3;
+	cmd[0].header.pad0 = R300_CMD_PACKET3_RAW;
 	cmd[1].i = R200_CP_CMD_BITBLT_MULTI | (5 << 16);
 	cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
 		    RADEON_GMC_DST_PITCH_OFFSET_CNTL |
diff --git a/src/mesa/drivers/dri/r300/r300_context.c b/src/mesa/drivers/dri/r300/r300_context.c
index 927cc88fda..db8b0a6e11 100644
--- a/src/mesa/drivers/dri/r300/r300_context.c
+++ b/src/mesa/drivers/dri/r300/r300_context.c
@@ -380,6 +380,14 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 
 	driInitExtensions(ctx, card_extensions, GL_TRUE);
 	
+	if (r300->radeon.glCtx->Mesa_DXTn) {
+	  _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
+	  _mesa_enable_extension( ctx, "GL_S3_s3tc" );
+	}
+	else if (driQueryOptionb (&r300->radeon.optionCache, "force_s3tc_enable")) {
+	  _mesa_enable_extension( ctx, "GL_EXT_texture_compression_s3tc" );
+	}
+
 	radeonInitSpanFuncs(ctx);
 	r300InitCmdBuf(r300);
 	r300InitState(r300);
diff --git a/src/mesa/drivers/dri/r300/r300_tex.c b/src/mesa/drivers/dri/r300/r300_tex.c
index e6c8c18011..44368c2819 100644
--- a/src/mesa/drivers/dri/r300/r300_tex.c
+++ b/src/mesa/drivers/dri/r300/r300_tex.c
@@ -433,6 +433,22 @@ static const struct gl_texture_format *r300ChooseTextureFormat(GLcontext * ctx,
 		else
 			return &_mesa_texformat_ycbcr_rev;
 
+	case GL_RGB_S3TC:
+	case GL_RGB4_S3TC:
+	case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
+	  return &_mesa_texformat_rgb_dxt1;
+	  
+	case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
+	  return &_mesa_texformat_rgba_dxt1;
+	  
+	case GL_RGBA_S3TC:
+	case GL_RGBA4_S3TC:
+	case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
+	  return &_mesa_texformat_rgba_dxt3;
+	  
+	case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
+	  return &_mesa_texformat_rgba_dxt5;
+	  
 	default:
 		_mesa_problem(ctx,
 			      "unexpected internalFormat 0x%x in r300ChooseTextureFormat",
@@ -706,6 +722,115 @@ static void r300TexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
 	t->dirty_images[face] |= (1 << level);
 }
 
+static void r300CompressedTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+                              GLint internalFormat,
+                              GLint width, GLint height, GLint border,
+                              GLsizei imageSize, const GLvoid *data,
+                              struct gl_texture_object *texObj,
+                              struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+   GLuint face;
+
+   /* which cube face or ordinary 2D image */
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
+      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
+      ASSERT(face < 6);
+      break;
+   default:
+      face = 0;
+   }
+
+   if ( t != NULL ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) r300AllocTexObj( texObj );
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexImage2D");
+         return;
+      }
+   }
+
+   texImage->IsClientData = GL_FALSE;
+/* can't call this, different parameters. Would never evaluate to true anyway currently
+   if (r300ValidateClientStorage( ctx, target, 
+				  internalFormat,
+				  width, height,
+				  format, type, pixels,
+				  packing, texObj, texImage)) {
+      if (RADEON_DEBUG & DEBUG_TEXTURE)
+	 fprintf(stderr, "%s: Using client storage\n", __FUNCTION__);
+   }
+   else */{
+      if (RADEON_DEBUG & DEBUG_TEXTURE)
+	 fprintf(stderr, "%s: Using normal storage\n", __FUNCTION__);
+
+      /* Normal path: copy (to cached memory) and eventually upload
+       * via another copy to GART memory and then a blit...  Could
+       * eliminate one copy by going straight to (permanent) GART.
+       *
+       * Note, this will call r300ChooseTextureFormat.
+       */
+      _mesa_store_compressed_teximage2d(ctx, target, level, internalFormat, width,
+                                 height, border, imageSize, data, texObj, texImage);
+
+      t->dirty_images[face] |= (1 << level);
+   }
+}
+
+
+static void r300CompressedTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+                                 GLint xoffset, GLint yoffset,
+                                 GLsizei width, GLsizei height,
+                                 GLenum format,
+                                 GLsizei imageSize, const GLvoid *data,
+                                 struct gl_texture_object *texObj,
+                                 struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+   GLuint face;
+
+
+   /* which cube face or ordinary 2D image */
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
+      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
+      ASSERT(face < 6);
+      break;
+   default:
+      face = 0;
+   }
+
+   assert( t ); /* this _should_ be true */
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) r300AllocTexObj( texObj );
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glCompressedTexSubImage3D");
+         return;
+      }
+   }
+
+   _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
+                            height, format, imageSize, data, texObj, texImage);
+
+   t->dirty_images[face] |= (1 << level);
+}
+
 #if ENABLE_HW_3D_TEXTURE
 static void r300TexImage3D(GLcontext * ctx, GLenum target, GLint level,
 			   GLint internalFormat,
@@ -1032,6 +1157,9 @@ void r300InitTextureFuncs(struct dd_function_table *functions)
 	functions->TexEnv = r300TexEnv;
 	functions->TexParameter = r300TexParameter;
 	functions->TexGen = r300TexGen;
+	
+	functions->CompressedTexImage2D	= r300CompressedTexImage2D;
+	functions->CompressedTexSubImage2D	= r300CompressedTexSubImage2D;
 
 	driInitTextureFormats();
 
diff --git a/src/mesa/drivers/dri/r300/r300_texmem.c b/src/mesa/drivers/dri/r300/r300_texmem.c
index ca392762ad..64fed681f1 100644
--- a/src/mesa/drivers/dri/r300/r300_texmem.c
+++ b/src/mesa/drivers/dri/r300/r300_texmem.c
@@ -276,162 +276,185 @@ 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,
-			   GLint hwlevel,
-			   GLint x, GLint y, GLint width, GLint height,
-			   GLuint face)
+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;
-
-	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)
-	 */
-
-	tex.offset = offset;
-	tex.pitch = BLIT_WIDTH_BYTES / 64;
-	tex.format = R200_TXFORMAT_I8;	/* any 1-byte texel format */
-#if 0 /* I am not sure HOSTDATA_BLT actually works.. Experiment here  - V.D */
-	tex.format = R200_TXFORMAT_RGBA8888;	/* any 4-byte texel format */
-#endif
-	if (texImage->TexFormat->TexelBytes) {
-		tex.width = imageWidth * texImage->TexFormat->TexelBytes;	/* in bytes */
-		tex.height = imageHeight;
-	} else {
-		tex.width = imageWidth;	/* compressed */
-		tex.height = imageHeight;
-		if (tex.height < 4)
-			tex.height = 4;
-	}
-	tex.image = &tmp;
+   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) {
+      /* 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 0
-	tex.width /= 4;
+      if (t->tile_bits & R200_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
-
-	/* copy (x,y,width,height,data) */
-	memcpy(&tmp, &t->image[face][hwlevel], sizeof(tmp));
+      {
+	 tmp.x = tmp.x >> (texImage->TexFormat->TexelBytes >> 1);
+      }
 #if 0
-	tex.image->width /=4;
+      if ((t->tile_bits & R200_TXO_MACRO_TILE) &&
+	 (texImage->Width * texImage->TexFormat->TexelBytes >= 256) &&
+	 ((!(t->tile_bits & R200_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
-
-#if 0
-	sleep(1);
-
-	fprintf(stderr, "*** Uploading texture\n");
-		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);
-#endif
-
-	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 = R200_TXFORMAT_I8; /* any 1-byte texel format */
+      tex.pitch = (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 22fbe2e18a..5c51e30637 100644
--- a/src/mesa/drivers/dri/r300/r300_texstate.c
+++ b/src/mesa/drivers/dri/r300/r300_texstate.c
@@ -54,6 +54,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R200_TXFORMAT_AL88      R200_TXFORMAT_AI88
 #define R200_TXFORMAT_YCBCR     R200_TXFORMAT_YVYU422
 #define R200_TXFORMAT_YCBCR_REV R200_TXFORMAT_VYUY422
+#define R200_TXFORMAT_RGB_DXT1  R200_TXFORMAT_DXT1
+#define R200_TXFORMAT_RGBA_DXT1 R200_TXFORMAT_DXT1
+#define R200_TXFORMAT_RGBA_DXT3 R200_TXFORMAT_DXT23
+#define R200_TXFORMAT_RGBA_DXT5 R200_TXFORMAT_DXT45
 
 #define _COLOR(f) \
     [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f, 0 }
@@ -67,7 +71,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
     [ MESA_FORMAT_ ## f ] = { R200_TXFORMAT_ ## f, R200_YUV_TO_RGB }
 #define _INVALID(f) \
     [ MESA_FORMAT_ ## f ] = { 0xffffffff, 0 }
-#define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_YCBCR_REV) \
+#define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_RGBA_DXT5) \
 			     && tx_table[f].flag )
 
 #define _ASSIGN(entry, format)	\
@@ -95,6 +99,12 @@ static const struct {
 	    _INVALID(CI8),
 	    _YUV(YCBCR),
 	    _YUV(YCBCR_REV),
+	    _INVALID(RGB_FXT1),
+	    _INVALID(RGBA_FXT1),
+	    _COLOR(RGB_DXT1),
+	    _ALPHA(RGBA_DXT1),
+	    _ALPHA(RGBA_DXT3),
+	    _ALPHA(RGBA_DXT5),
 	    };
 
 static const struct {
@@ -129,6 +139,10 @@ static const struct {
 	    _ASSIGN(CI8, R300_EASY_TX_FORMAT(X, X, X, X, X8)),
 	    _ASSIGN(YCBCR, R300_EASY_TX_FORMAT(X, Y, Z, ONE, G8R8_G8B8)|R300_TX_FORMAT_YUV_MODE ),
 	    _ASSIGN(YCBCR_REV, R300_EASY_TX_FORMAT(X, Y, Z, ONE, G8R8_G8B8)|R300_TX_FORMAT_YUV_MODE),
+	    _ASSIGN(RGB_DXT1, R300_EASY_TX_FORMAT(X, Y, Z, ONE, DXT1)),
+	    _ASSIGN(RGBA_DXT1, R300_EASY_TX_FORMAT(X, Y, Z, W, DXT1)),
+	    _ASSIGN(RGBA_DXT3, R300_EASY_TX_FORMAT(X, Y, Z, W, DXT3)),
+	    _ASSIGN(RGBA_DXT5, R300_EASY_TX_FORMAT(Y, Z, W, X, DXT5)),
 	    };
 
 #undef _COLOR
@@ -154,8 +168,8 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	r300TexObjPtr t = (r300TexObjPtr) tObj->DriverData;
 	const struct gl_texture_image *baseImage =
 	    tObj->Image[0][tObj->BaseLevel];
-	GLint curOffset;
-	GLint i;
+	GLint curOffset, blitWidth;
+	GLint i, texelBytes;
 	GLint numLevels;
 	GLint log2Width, log2Height, log2Depth;
 
@@ -180,6 +194,8 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 		return;
 	}
 
+	texelBytes = baseImage->TexFormat->TexelBytes;
+
 	/* Compute which mipmap levels we really want to send to the hardware.
 	 */
 
@@ -197,68 +213,85 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	 * memory organized as a rectangle of width BLIT_WIDTH_BYTES.
 	 */
 	curOffset = 0;
+	blitWidth = BLIT_WIDTH_BYTES;
 
 	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) {
-			size = texImage->CompressedSize;
-		} else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
-			size =
-			    ((texImage->Width *
-			      texImage->TexFormat->TexelBytes + 63)
-			     & ~63) * texImage->Height;
-		} else {
-			int w =
-			    texImage->Width * texImage->TexFormat->TexelBytes;
-			if (w < 32)
-				w = 32;
-			size = w * texImage->Height * texImage->Depth;
-		}
-		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->IntFormat);
-
-		/* Align to 32-byte offset.  It is faster to do this unconditionally
-		 * (no branch penalty).
-		 */
-
-		curOffset = (curOffset + 0x1f) & ~0x1f;
-
-		t->image[0][i].x = curOffset % BLIT_WIDTH_BYTES;
-		t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES;
-		t->image[0][i].width = MIN2(size, BLIT_WIDTH_BYTES);
-		t->image[0][i].height = size / t->image[0][i].width;
-
+	  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 */
+	    {
+	      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;
+	  } 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->IntFormat);
+	  
+	  /* 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 % BLIT_WIDTH_BYTES;
+	    t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES;
+	    t->image[0][i].width = MIN2(size, 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 < BLIT_WIDTH_BYTES
-			   && t->image[0][i].height == 1));
+	  /* 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 < 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;
-
+	  
+	  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 =
@@ -336,7 +369,7 @@ static void r300SetTexImages(r300ContextPtr rmesa,
 	else
 		t->pitch =
 		    ((tObj->Image[0][t->base.firstLevel]->Width *
-		      baseImage->TexFormat->TexelBytes) + 63) & ~(63);
+		      texelBytes) + 63) & ~(63);
 	t->pitch -= 32;
 
 	t->dirty_state = TEX_ALL;
diff --git a/src/mesa/drivers/dri/r300/radeon_screen.c b/src/mesa/drivers/dri/r300/radeon_screen.c
index 7b66ce78ef..b169b7e26d 100644
--- a/src/mesa/drivers/dri/r300/radeon_screen.c
+++ b/src/mesa/drivers/dri/r300/radeon_screen.c
@@ -74,6 +74,7 @@ DRI_CONF_BEGIN
 		DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB)
 		DRI_CONF_DEF_MAX_ANISOTROPY(1.0, "1.0,2.0,4.0,8.0,16.0")
 		DRI_CONF_NO_NEG_LOD_BIAS(false)
+		DRI_CONF_FORCE_S3TC_ENABLE(false)
 		DRI_CONF_COLOR_REDUCTION(DRI_CONF_COLOR_REDUCTION_DITHER)
 		DRI_CONF_ROUND_MODE(DRI_CONF_ROUND_TRUNC)
 		DRI_CONF_DITHER_MODE(DRI_CONF_DITHER_XERRORDIFF)
@@ -86,7 +87,7 @@ DRI_CONF_BEGIN
 		DRI_CONF_NV_VERTEX_PROGRAM(false)
 	DRI_CONF_SECTION_END
 DRI_CONF_END;
-static const GLuint __driR200NConfigOptions = 13;
+static const GLuint __driR200NConfigOptions = 14;
 
 /* TODO: integrate these into xmlpool.h! */
 #define DRI_CONF_MAX_TEXTURE_IMAGE_UNITS(def,min,max) \
@@ -122,6 +123,7 @@ DRI_CONF_BEGIN
 		DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB)
 		DRI_CONF_DEF_MAX_ANISOTROPY(1.0, "1.0,2.0,4.0,8.0,16.0")
 		DRI_CONF_NO_NEG_LOD_BIAS(false)
+                DRI_CONF_FORCE_S3TC_ENABLE(false)
 		DRI_CONF_COLOR_REDUCTION(DRI_CONF_COLOR_REDUCTION_DITHER)
 		DRI_CONF_ROUND_MODE(DRI_CONF_ROUND_TRUNC)
 		DRI_CONF_DITHER_MODE(DRI_CONF_DITHER_XERRORDIFF)
@@ -130,7 +132,7 @@ DRI_CONF_BEGIN
 		DRI_CONF_NO_RAST(false)
 	DRI_CONF_SECTION_END
 DRI_CONF_END;
-static const GLuint __driR300NConfigOptions = 13;
+static const GLuint __driR300NConfigOptions = 14;
 
 extern const struct dri_extension card_extensions[];