Pull the texmem i915 driver onto a new branch closer to the current trunk.

Compiles but otherwise untested.

38 files changed, 3102 insertions, 6341 deletions

diff --git a/src/mesa/drivers/dri/i915/Makefile b/src/mesa/drivers/dri/i915/Makefile
index 805abf75e0..56f1104142 100644
--- a/src/mesa/drivers/dri/i915/Makefile
+++ b/src/mesa/drivers/dri/i915/Makefile
@@ -7,16 +7,6 @@ LIBNAME = i915_dri.so
 MINIGLX_SOURCES = server/intel_dri.c
 
 DRIVER_SOURCES = \
-	i915_context.c \
-	i915_debug.c \
-	i915_fragprog.c \
-	i915_metaops.c \
-	i915_program.c \
-	i915_state.c \
-	i915_tex.c \
-	i915_texprog.c \
-	i915_texstate.c \
-	i915_vtbl.c \
 	i830_context.c \
 	i830_metaops.c \
 	i830_state.c \
@@ -24,18 +14,44 @@ DRIVER_SOURCES = \
 	i830_tex.c \
 	i830_texstate.c \
 	i830_vtbl.c \
+	intel_render.c \
+	intel_regions.c \
+	intel_buffer_objects.c \
 	intel_batchbuffer.c \
+	intel_mipmap_tree.c \
+	i915_tex_layout.c \
+	intel_tex_image.c \
+	intel_tex_subimage.c \
+	intel_tex_copy.c \
+	intel_tex_validate.c \
+	intel_tex_format.c \
+	intel_tex.c \
+	intel_pixel.c \
+	intel_pixel_copy.c \
+	intel_pixel_read.c \
+	intel_pixel_draw.c \
+	intel_buffers.c \
+	intel_blit.c \
+	i915_tex.c \
+	i915_texstate.c \
+	i915_context.c \
+	i915_debug.c \
+	i915_fragprog.c \
+	i915_metaops.c \
+	i915_program.c \
+	i915_state.c \
+	i915_texprog.c \
+	i915_vtbl.c \
 	intel_context.c \
 	intel_ioctl.c \
-	intel_pixel.c \
-	intel_render.c \
 	intel_rotate.c \
 	intel_screen.c \
 	intel_span.c \
 	intel_state.c \
-	intel_tex.c \
-	intel_texmem.c \
-	intel_tris.c 
+	intel_tris.c \
+	intel_fbo.c \
+	intel_depthstencil.c \
+	intel_bufmgr.c
 
 C_SOURCES = \
 	$(COMMON_SOURCES) \
diff --git a/src/mesa/drivers/dri/i915/i830_context.c b/src/mesa/drivers/dri/i915/i830_context.c
index d16b153fac..b66ca31059 100644
--- a/src/mesa/drivers/dri/i915/i830_context.c
+++ b/src/mesa/drivers/dri/i915/i830_context.c
@@ -58,10 +58,9 @@ GLboolean i830CreateContext( const __GLcontextModes *mesaVis,
 			    void *sharedContextPrivate)
 {
    struct dd_function_table functions;
-   i830ContextPtr i830 = (i830ContextPtr) CALLOC_STRUCT(i830_context);
-   intelContextPtr intel = &i830->intel;
+   struct i830_context *i830 = CALLOC_STRUCT(i830_context);
+   struct intel_context *intel = &i830->intel;
    GLcontext *ctx = &intel->ctx;
-   GLuint i;
    if (!i830) return GL_FALSE;
 
    i830InitVtbl( i830 );
@@ -77,34 +76,14 @@ GLboolean i830CreateContext( const __GLcontextModes *mesaVis,
    intel->ctx.Const.MaxTextureImageUnits = I830_TEX_UNITS;
    intel->ctx.Const.MaxTextureCoordUnits = I830_TEX_UNITS;
 
-   intel->nr_heaps = 1;
-   intel->texture_heaps[0] = 
-      driCreateTextureHeap( 0, intel,
-			    intel->intelScreen->tex.size,
-			    12,
-			    I830_NR_TEX_REGIONS,
-			    intel->sarea->texList,
-			    (unsigned *) & intel->sarea->texAge,
-			    & intel->swapped,
-			    sizeof( struct i830_texture_object ),
-			    (destroy_texture_object_t *)intelDestroyTexObj );
-
-   /* FIXME: driCalculateMaxTextureLevels assumes that mipmaps are tightly
-    * FIXME: packed, but they're not in Intel graphics hardware.
+   /* Advertise the full hardware capabilities.  The new memory
+    * manager should cope much better with overload situations:
     */
-   intel->ctx.Const.MaxTextureUnits = I830_TEX_UNITS;
-   i = driQueryOptioni( &intel->intelScreen->optionCache, "allow_large_textures");
-   driCalculateMaxTextureLevels( intel->texture_heaps,
-				 intel->nr_heaps,
-				 &intel->ctx.Const,
-				 4,
-				 11, /* max 2D texture size is 2048x2048 */
-				 8,  /* max 3D texture size is 256^3 */
-				 10, /* max CUBE texture size is 1024x1024 */
-				 11, /* max RECT. supported */
-				 12,
-				 GL_FALSE,
-				 i );
+   ctx->Const.MaxTextureLevels = 12;
+   ctx->Const.Max3DTextureLevels = 9;
+   ctx->Const.MaxCubeTextureLevels = 11;
+   ctx->Const.MaxTextureRectSize = (1<<11);
+   ctx->Const.MaxTextureUnits = I830_TEX_UNITS;
 
    _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 
 		       18 * sizeof(GLfloat) );
@@ -114,7 +93,7 @@ GLboolean i830CreateContext( const __GLcontextModes *mesaVis,
    driInitExtensions( ctx, i830_extensions, GL_FALSE );
 
    i830InitState( i830 );
-
+   i830InitMetaFuncs( i830 );
 
    _tnl_allow_vertex_fog( ctx, 1 ); 
    _tnl_allow_pixel_fog( ctx, 0 ); 
diff --git a/src/mesa/drivers/dri/i915/i830_context.h b/src/mesa/drivers/dri/i915/i830_context.h
index d5811e6c34..7cf9b8af64 100644
--- a/src/mesa/drivers/dri/i915/i830_context.h
+++ b/src/mesa/drivers/dri/i915/i830_context.h
@@ -39,6 +39,7 @@
 #define I830_UPLOAD_CTX              0x1
 #define I830_UPLOAD_BUFFERS          0x2
 #define I830_UPLOAD_STIPPLE          0x4
+#define I830_UPLOAD_INVARIENT        0x8
 #define I830_UPLOAD_TEX(i)           (0x10<<(i))
 #define I830_UPLOAD_TEXBLEND(i)      (0x100<<(i))
 #define I830_UPLOAD_TEX_ALL          (0x0f0)
@@ -48,17 +49,15 @@
  */
 #define I830_DESTREG_CBUFADDR0 0
 #define I830_DESTREG_CBUFADDR1 1
-#define I830_DESTREG_CBUFADDR2 2
-#define I830_DESTREG_DBUFADDR0 3
-#define I830_DESTREG_DBUFADDR1 4
-#define I830_DESTREG_DBUFADDR2 5
-#define I830_DESTREG_DV0 6
-#define I830_DESTREG_DV1 7
-#define I830_DESTREG_SENABLE 8
-#define I830_DESTREG_SR0 9
-#define I830_DESTREG_SR1 10
-#define I830_DESTREG_SR2 11
-#define I830_DEST_SETUP_SIZE 12
+#define I830_DESTREG_DBUFADDR0 2
+#define I830_DESTREG_DBUFADDR1 3
+#define I830_DESTREG_DV0 4
+#define I830_DESTREG_DV1 5
+#define I830_DESTREG_SENABLE 6
+#define I830_DESTREG_SR0 7
+#define I830_DESTREG_SR1 8
+#define I830_DESTREG_SR2 9
+#define I830_DEST_SETUP_SIZE 10
 
 #define I830_CTXREG_STATE1		0
 #define I830_CTXREG_STATE2		1
@@ -84,14 +83,13 @@
 #define I830_STP_SETUP_SIZE    2
 
 #define I830_TEXREG_TM0LI      0 /* load immediate 2 texture map n */
-#define I830_TEXREG_TM0S0      1
-#define I830_TEXREG_TM0S1      2
-#define I830_TEXREG_TM0S2      3
-#define I830_TEXREG_TM0S3      4
-#define I830_TEXREG_TM0S4      5
-#define I830_TEXREG_MCS	       6	/* _3DSTATE_MAP_COORD_SETS */
-#define I830_TEXREG_CUBE       7	/* _3DSTATE_MAP_SUBE */
-#define I830_TEX_SETUP_SIZE    8
+#define I830_TEXREG_TM0S1      1
+#define I830_TEXREG_TM0S2      2
+#define I830_TEXREG_TM0S3      3
+#define I830_TEXREG_TM0S4      4
+#define I830_TEXREG_MCS	       5	/* _3DSTATE_MAP_COORD_SETS */
+#define I830_TEXREG_CUBE       6	/* _3DSTATE_MAP_SUBE */
+#define I830_TEX_SETUP_SIZE    7
 
 #define I830_TEXBLEND_SIZE	12	/* (4 args + op) * 2 + COLOR_FACTOR */
 
@@ -110,6 +108,17 @@ struct i830_hw_state {
    GLuint Tex[I830_TEX_UNITS][I830_TEX_SETUP_SIZE];
    GLuint TexBlend[I830_TEX_UNITS][I830_TEXBLEND_SIZE];
    GLuint TexBlendWordsUsed[I830_TEX_UNITS];
+
+   struct intel_region *draw_region;
+   struct intel_region *depth_region;
+
+   /* Regions aren't actually that appropriate here as the memory may
+    * be from a PBO or FBO.  Just use the buffer id.  Will have to do
+    * this for draw and depth for FBO's...
+    */
+   GLuint tex_buffer[I830_TEX_UNITS];
+   GLuint tex_offset[I830_TEX_UNITS];
+   
    GLuint emitted;		/* I810_UPLOAD_* */
    GLuint active;
 };
@@ -118,15 +127,12 @@ struct i830_context
 {
    struct intel_context intel;
    
+   GLuint lodbias_tm0s3[MAX_TEXTURE_UNITS];
    DECLARE_RENDERINPUTS(last_index_bitset);
 
    struct i830_hw_state meta, initial, state, *current;
 };
 
-typedef struct i830_context *i830ContextPtr;
-typedef struct i830_texture_object *i830TextureObjectPtr;
-
-#define I830_CONTEXT(ctx)	((i830ContextPtr)(ctx))
 
 
 
@@ -148,7 +154,7 @@ do {						\
 /* i830_vtbl.c
  */
 extern void 
-i830InitVtbl( i830ContextPtr i830 );
+i830InitVtbl( struct i830_context *i830 );
 
 /* i830_context.c
  */
@@ -160,22 +166,19 @@ i830CreateContext( const __GLcontextModes *mesaVis,
 /* i830_tex.c, i830_texstate.c
  */
 extern void 
-i830UpdateTextureState( intelContextPtr intel );
+i830UpdateTextureState( struct intel_context *intel );
 
 extern void 
 i830InitTextureFuncs( struct dd_function_table *functions );
 
-extern intelTextureObjectPtr
-i830AllocTexObj( struct gl_texture_object *tObj );
-
 /* i830_texblend.c
  */
-extern GLuint i830SetTexEnvCombine(i830ContextPtr i830,
+extern GLuint i830SetTexEnvCombine(struct i830_context *i830,
     const struct gl_tex_env_combine_state * combine, GLint blendUnit,
      GLuint texel_op, GLuint *state, const GLfloat *factor );
 
 extern void 
-i830EmitTextureBlend( i830ContextPtr i830 );
+i830EmitTextureBlend( struct i830_context *i830 );
 
 
 /* i830_state.c
@@ -184,34 +187,29 @@ extern void
 i830InitStateFuncs( struct dd_function_table *functions );
 
 extern void 
-i830EmitState( i830ContextPtr i830 );
+i830EmitState( struct i830_context *i830 );
 
 extern void 
-i830InitState( i830ContextPtr i830 );
+i830InitState( struct i830_context *i830 );
 
 /* i830_metaops.c
  */
-extern GLboolean
-i830TryTextureReadPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *pack,
-			  GLvoid *pixels );
-
-extern GLboolean
-i830TryTextureDrawPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *unpack,
-			  const GLvoid *pixels );
-
 extern void 
-i830ClearWithTris( intelContextPtr intel, GLbitfield mask,
-		   GLboolean all, GLint cx, GLint cy, GLint cw, GLint ch);
+i830InitMetaFuncs( struct i830_context *i830 );
 
 extern void
-i830RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
+i830RotateWindow(struct intel_context *intel, __DRIdrawablePrivate *dPriv,
                  GLuint srcBuf);
 
+/*======================================================================
+ * Inline conversion functions.  These are better-typed than the
+ * macros used previously:
+ */
+static INLINE struct i830_context *
+i830_context( GLcontext *ctx )
+{
+   return (struct i830_context *)ctx;
+}
+
 #endif
 
diff --git a/src/mesa/drivers/dri/i915/i830_metaops.c b/src/mesa/drivers/dri/i915/i830_metaops.c
index e7215dfaf0..c8100658e3 100644
--- a/src/mesa/drivers/dri/i915/i830_metaops.c
+++ b/src/mesa/drivers/dri/i915/i830_metaops.c
@@ -34,41 +34,33 @@
 #include "intel_screen.h"
 #include "intel_batchbuffer.h"
 #include "intel_ioctl.h"
+#include "intel_regions.h"
 
 #include "i830_context.h"
 #include "i830_reg.h"
 
 /* A large amount of state doesn't need to be uploaded.
  */
-#define ACTIVE (I830_UPLOAD_TEXBLEND(0) |	\
-		I830_UPLOAD_STIPPLE |		\
+#define ACTIVE (I830_UPLOAD_INVARIENT |         \
 		I830_UPLOAD_CTX |		\
 		I830_UPLOAD_BUFFERS |		\
+		I830_UPLOAD_STIPPLE |		\
+		I830_UPLOAD_TEXBLEND(0) |	\
 		I830_UPLOAD_TEX(0))		
 
 
 #define SET_STATE( i830, STATE )		\
 do {						\
-   assert(!i830->intel.prim.flush); \
-   i830->current->emitted = 0;			\
+   i830->current->emitted &= ~ACTIVE;			\
    i830->current = &i830->STATE;		\
-   i830->current->emitted = 0;			\
+   i830->current->emitted &= ~ACTIVE;			\
 } while (0)
 
-/* Operations where the 3D engine is decoupled temporarily from the
- * current GL state and used for other purposes than simply rendering
- * incoming triangles.
- */
-static void set_initial_state( i830ContextPtr i830 )
-{
-   memcpy(&i830->meta, &i830->initial, sizeof(i830->meta) );
-   i830->meta.active = ACTIVE;
-   i830->meta.emitted = 0;
-}
-
 
-static void set_no_depth_stencil_write( i830ContextPtr i830 )
+static void set_no_stencil_write( struct intel_context *intel )
 {
+   struct i830_context *i830 = i830_context(&intel->ctx);
+
    /* ctx->Driver.Enable( ctx, GL_STENCIL_TEST, GL_FALSE )
     */
    i830->meta.Ctx[I830_CTXREG_ENABLES_1] &= ~ENABLE_STENCIL_TEST;
@@ -76,6 +68,12 @@ static void set_no_depth_stencil_write( i830ContextPtr i830 )
    i830->meta.Ctx[I830_CTXREG_ENABLES_1] |= DISABLE_STENCIL_TEST;
    i830->meta.Ctx[I830_CTXREG_ENABLES_2] |= DISABLE_STENCIL_WRITE;
 
+   i830->meta.emitted &= ~I830_UPLOAD_CTX;
+}
+
+static void set_no_depth_write( struct intel_context *intel )
+{
+   struct i830_context *i830 = i830_context(&intel->ctx);
 
    /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
     */
@@ -87,25 +85,43 @@ static void set_no_depth_stencil_write( i830ContextPtr i830 )
    i830->meta.emitted &= ~I830_UPLOAD_CTX;
 }
 
+/* Set depth unit to replace.
+ */
+static void set_depth_replace( struct intel_context *intel )
+{
+   struct i830_context *i830 = i830_context(&intel->ctx);
+
+   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
+    * ctx->Driver.DepthMask( ctx, GL_TRUE )
+    */
+   i830->meta.Ctx[I830_CTXREG_ENABLES_1] &= ~ENABLE_DIS_DEPTH_TEST_MASK;
+   i830->meta.Ctx[I830_CTXREG_ENABLES_2] &= ~ENABLE_DIS_DEPTH_WRITE_MASK;
+   i830->meta.Ctx[I830_CTXREG_ENABLES_1] |= ENABLE_DEPTH_TEST;
+   i830->meta.Ctx[I830_CTXREG_ENABLES_2] |= ENABLE_DEPTH_WRITE;
+
+   /* ctx->Driver.DepthFunc( ctx, GL_ALWAYS )
+    */
+   i830->meta.Ctx[I830_CTXREG_STATE3] &= ~DEPTH_TEST_FUNC_MASK;
+   i830->meta.Ctx[I830_CTXREG_STATE3] |= (ENABLE_DEPTH_TEST_FUNC |
+					  DEPTH_TEST_FUNC(COMPAREFUNC_ALWAYS));
+
+   i830->meta.emitted &= ~I830_UPLOAD_CTX;
+}
+
+
 /* Set stencil unit to replace always with the reference value.
  */
-static void set_stencil_replace( i830ContextPtr i830,
+static void set_stencil_replace( struct intel_context *intel,
 				 GLuint s_mask,
 				 GLuint s_clear)
 {
+   struct i830_context *i830 = i830_context(&intel->ctx);
+
    /* ctx->Driver.Enable( ctx, GL_STENCIL_TEST, GL_TRUE )
     */
    i830->meta.Ctx[I830_CTXREG_ENABLES_1] |= ENABLE_STENCIL_TEST;
    i830->meta.Ctx[I830_CTXREG_ENABLES_2] |= ENABLE_STENCIL_WRITE;
 
-
-   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
-    */
-   i830->meta.Ctx[I830_CTXREG_ENABLES_1] &= ~ENABLE_DIS_DEPTH_TEST_MASK;
-   i830->meta.Ctx[I830_CTXREG_ENABLES_2] &= ~ENABLE_DIS_DEPTH_WRITE_MASK;
-   i830->meta.Ctx[I830_CTXREG_ENABLES_1] |= DISABLE_DEPTH_TEST;
-   i830->meta.Ctx[I830_CTXREG_ENABLES_2] |= DISABLE_DEPTH_WRITE;
-
    /* ctx->Driver.StencilMask( ctx, s_mask )
     */
    i830->meta.Ctx[I830_CTXREG_STATE4] &= ~MODE4_ENABLE_STENCIL_WRITE_MASK;
@@ -141,8 +157,10 @@ static void set_stencil_replace( i830ContextPtr i830,
 }
 
 
-static void set_color_mask( i830ContextPtr i830, GLboolean state )
+static void set_color_mask( struct intel_context *intel, GLboolean state )
 {
+   struct i830_context *i830 = i830_context(&intel->ctx);
+
    const GLuint mask = ((1 << WRITEMASK_RED_SHIFT) |
 			(1 << WRITEMASK_GREEN_SHIFT) |
 			(1 << WRITEMASK_BLUE_SHIFT) |
@@ -161,8 +179,9 @@ static void set_color_mask( i830ContextPtr i830, GLboolean state )
 /* Installs a one-stage passthrough texture blend pipeline.  Is there
  * more that can be done to turn off texturing?
  */
-static void set_no_texture( i830ContextPtr i830 )
+static void set_no_texture( struct intel_context *intel )
 {
+   struct i830_context *i830 = i830_context(&intel->ctx);
    static const struct gl_tex_env_combine_state comb = {
       GL_NONE, GL_NONE,
       { GL_TEXTURE, 0, 0, }, { GL_TEXTURE, 0, 0, },
@@ -181,11 +200,12 @@ static void set_no_texture( i830ContextPtr i830 )
 /* Set up a single element blend stage for 'replace' texturing with no
  * funny ops.
  */
-static void enable_texture_blend_replace( i830ContextPtr i830 )
+static void set_texture_blend_replace( struct intel_context *intel )
 {
+   struct i830_context *i830 = i830_context(&intel->ctx);
    static const struct gl_tex_env_combine_state comb = {
       GL_REPLACE, GL_REPLACE,
-      { GL_TEXTURE, GL_TEXTURE, GL_TEXTURE }, { GL_TEXTURE, GL_TEXTURE, GL_TEXTURE, },
+      { GL_TEXTURE, GL_TEXTURE, GL_TEXTURE, }, { GL_TEXTURE, GL_TEXTURE, GL_TEXTURE, },
       { GL_SRC_COLOR, GL_SRC_COLOR, GL_SRC_COLOR }, { GL_SRC_ALPHA, GL_SRC_ALPHA, GL_SRC_ALPHA },
       0, 0, 1, 1
    };
@@ -206,29 +226,90 @@ static void enable_texture_blend_replace( i830ContextPtr i830 )
 /* Set up an arbitary piece of memory as a rectangular texture
  * (including the front or back buffer).
  */
-static void set_tex_rect_source( i830ContextPtr i830,
-				 GLuint offset,
-				 GLuint width, 
-				 GLuint height,
-				 GLuint pitch, /* in bytes */
-				 GLuint textureFormat )
+static GLboolean set_tex_rect_source( struct intel_context *intel,
+				      GLuint buffer,
+				      GLuint offset,
+				      GLuint pitch,
+				      GLuint height,
+				      GLenum format,
+				      GLenum type)
 {
-   GLint numLevels = 1;
+   struct i830_context *i830 = i830_context(&intel->ctx);
    GLuint *setup = i830->meta.Tex[0];
+   GLint numLevels = 1;
+   GLuint textureFormat;
+   GLuint cpp;
+
+   /* A full implementation of this would do the upload through
+    * glTexImage2d, and get all the conversion operations at that
+    * point.  We are restricted, but still at least have access to the
+    * fragment program swizzle.
+    */
+   switch (format) {
+   case GL_BGRA:
+      switch (type) {
+      case GL_UNSIGNED_INT_8_8_8_8_REV:
+      case GL_UNSIGNED_BYTE:
+	 textureFormat = (MAPSURF_32BIT | MT_32BIT_ARGB8888);
+	 cpp = 4;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+   case GL_RGBA:
+      switch (type) {
+      case GL_UNSIGNED_INT_8_8_8_8_REV:
+      case GL_UNSIGNED_BYTE:
+	 textureFormat = (MAPSURF_32BIT | MT_32BIT_ABGR8888);
+	 cpp = 4;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+   case GL_BGR:
+      switch (type) {
+      case GL_UNSIGNED_SHORT_5_6_5_REV:
+	 textureFormat = (MAPSURF_16BIT | MT_16BIT_RGB565);
+	 cpp = 2;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+   case GL_RGB:
+      switch (type) {
+      case GL_UNSIGNED_SHORT_5_6_5:
+	 textureFormat = (MAPSURF_16BIT | MT_16BIT_RGB565);
+	 cpp = 2;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   default:
+      return GL_FALSE;
+   }
 
-/*    fprintf(stderr, "%s: offset: %x w: %d h: %d pitch %d format %x\n", */
-/* 	   __FUNCTION__, offset, width, height, pitch, textureFormat ); */
+   i830->meta.tex_buffer[0] = buffer;
+   i830->meta.tex_offset[0] = offset;
 
    setup[I830_TEXREG_TM0LI] = (_3DSTATE_LOAD_STATE_IMMEDIATE_2 | 
 			       (LOAD_TEXTURE_MAP0 << 0) | 4);
-   setup[I830_TEXREG_TM0S0] = (TM0S0_USE_FENCE | offset);
    setup[I830_TEXREG_TM0S1] = (((height - 1) << TM0S1_HEIGHT_SHIFT) |
-			       ((width - 1) << TM0S1_WIDTH_SHIFT) |
+			       ((pitch - 1) << TM0S1_WIDTH_SHIFT) |
 			       textureFormat);
-   setup[I830_TEXREG_TM0S2] = ((((pitch / 4) - 1) << TM0S2_PITCH_SHIFT));   
-   setup[I830_TEXREG_TM0S3] &= ~TM0S3_MAX_MIP_MASK;
-   setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIN_MIP_MASK;
-   setup[I830_TEXREG_TM0S3] |= ((numLevels - 1)*4) << TM0S3_MIN_MIP_SHIFT;
+   setup[I830_TEXREG_TM0S2] = (((((pitch * cpp) / 4) - 1) << TM0S2_PITCH_SHIFT)  |
+			       TM0S2_CUBE_FACE_ENA_MASK);   
+
+   setup[I830_TEXREG_TM0S3] = ( (((numLevels - 1)*4) << TM0S3_MIN_MIP_SHIFT) |
+				(FILTER_NEAREST << TM0S3_MIN_FILTER_SHIFT) |
+				(MIPFILTER_NONE << TM0S3_MIP_FILTER_SHIFT) |
+				(FILTER_NEAREST << TM0S3_MAG_FILTER_SHIFT));
+
+   setup[I830_TEXREG_CUBE] = (_3DSTATE_MAP_CUBE | MAP_UNIT(0));
 
    setup[I830_TEXREG_MCS] = (_3DSTATE_MAP_COORD_SET_CMD |
 			     MAP_UNIT(0) |
@@ -241,44 +322,17 @@ static void set_tex_rect_source( i830ContextPtr i830,
 			     TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_WRAP));
 
    i830->meta.emitted &= ~I830_UPLOAD_TEX(0);
+   return GL_TRUE;
 }
 
 
-/* Select between front and back draw buffers.
- */
-static void set_draw_region( i830ContextPtr i830,
-			      const intelRegion *region )
-{
-   i830->meta.Buffer[I830_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i830->meta.Buffer[I830_DESTREG_CBUFADDR2] = region->offset;
-   i830->meta.emitted &= ~I830_UPLOAD_BUFFERS;
-}
-
-/* Setup an arbitary draw format, useful for targeting
- * texture or agp memory.
- */
-#if 0
-static void set_draw_format( i830ContextPtr i830,
-			     GLuint format,
-			     GLuint depth_format)
-{
-   i830->meta.Buffer[I830_DESTREG_DV1] = (DSTORG_HORT_BIAS(0x8) | /* .5 */
-					  DSTORG_VERT_BIAS(0x8) | /* .5 */
-					  format |
-					  DEPTH_IS_Z |
-					  depth_format);
-}
-#endif
-
-
-static void set_vertex_format( i830ContextPtr i830 )
+static void set_vertex_format( struct intel_context *intel )
 {
+   struct i830_context *i830 = i830_context(&intel->ctx);
    i830->meta.Ctx[I830_CTXREG_VF] =  (_3DSTATE_VFT0_CMD |
 				      VFT0_TEX_COUNT(1) |
 				      VFT0_DIFFUSE |
-				      VFT0_SPEC |
-				      VFT0_XYZW);
+				      VFT0_XYZ);
    i830->meta.Ctx[I830_CTXREG_VF2] = (_3DSTATE_VFT1_CMD |
 				      VFT1_TEX0_FMT(TEXCOORDFMT_2D) |
 				      VFT1_TEX1_FMT(TEXCOORDFMT_2D) | 
@@ -288,626 +342,127 @@ static void set_vertex_format( i830ContextPtr i830 )
 }
 
 
-static void draw_quad(i830ContextPtr i830, 
-		      GLfloat x0, GLfloat x1,
-		      GLfloat y0, GLfloat y1, 
-		      GLubyte red, GLubyte green,
-		      GLubyte blue, GLubyte alpha,
-		      GLfloat s0, GLfloat s1,
-		      GLfloat t0, GLfloat t1 )
+static void meta_import_pixel_state( struct intel_context *intel )
 {
-   GLuint vertex_size = 8;
-   GLuint *vb = intelEmitInlinePrimitiveLocked( &i830->intel, 
-						PRIM3D_TRIFAN, 
-						4*vertex_size,
-						vertex_size );
-   intelVertex tmp;
-   int i;
-
+   struct i830_context *i830 = i830_context(&intel->ctx);
+
+   i830->meta.Ctx[I830_CTXREG_STATE1] = i830->state.Ctx[I830_CTXREG_STATE1];
+   i830->meta.Ctx[I830_CTXREG_STATE2] = i830->state.Ctx[I830_CTXREG_STATE2];
+   i830->meta.Ctx[I830_CTXREG_STATE3] = i830->state.Ctx[I830_CTXREG_STATE3];
+   i830->meta.Ctx[I830_CTXREG_STATE4] = i830->state.Ctx[I830_CTXREG_STATE4];
+   i830->meta.Ctx[I830_CTXREG_STATE5] = i830->state.Ctx[I830_CTXREG_STATE5];
+   i830->meta.Ctx[I830_CTXREG_IALPHAB] = i830->state.Ctx[I830_CTXREG_IALPHAB];
+   i830->meta.Ctx[I830_CTXREG_STENCILTST] = i830->state.Ctx[I830_CTXREG_STENCILTST];
+   i830->meta.Ctx[I830_CTXREG_ENABLES_1] = i830->state.Ctx[I830_CTXREG_ENABLES_1];
+   i830->meta.Ctx[I830_CTXREG_ENABLES_2] = i830->state.Ctx[I830_CTXREG_ENABLES_2];
+   i830->meta.Ctx[I830_CTXREG_AA] = i830->state.Ctx[I830_CTXREG_AA];
+   i830->meta.Ctx[I830_CTXREG_FOGCOLOR] = i830->state.Ctx[I830_CTXREG_FOGCOLOR];
+   i830->meta.Ctx[I830_CTXREG_BLENDCOLOR0] = i830->state.Ctx[I830_CTXREG_BLENDCOLOR0];
+   i830->meta.Ctx[I830_CTXREG_BLENDCOLOR1] = i830->state.Ctx[I830_CTXREG_BLENDCOLOR1];
+   i830->meta.Ctx[I830_CTXREG_MCSB0] = i830->state.Ctx[I830_CTXREG_MCSB0];
+   i830->meta.Ctx[I830_CTXREG_MCSB1] = i830->state.Ctx[I830_CTXREG_MCSB1];
    
-/*    fprintf(stderr, "%s: %f,%f-%f,%f 0x%x%x%x%x %f,%f-%f,%f\n", */
-/* 	   __FUNCTION__, */
-/* 	   x0,y0,x1,y1,red,green,blue,alpha,s0,t0,s1,t1); */
-
-
-   /* initial vertex, left bottom */
-   tmp.v.x = x0;
-   tmp.v.y = y0;
-   tmp.v.z = 1.0;
-   tmp.v.w = 1.0;
-   tmp.v.color.red = red;
-   tmp.v.color.green = green;
-   tmp.v.color.blue = blue;
-   tmp.v.color.alpha = alpha;
-   tmp.v.specular.red = 0;
-   tmp.v.specular.green = 0;
-   tmp.v.specular.blue = 0;
-   tmp.v.specular.alpha = 0;
-   tmp.v.u0 = s0;
-   tmp.v.v0 = t0;
-   for (i = 0 ; i < 8 ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* right bottom */
-   vb += 8;
-   tmp.v.x = x1;
-   tmp.v.u0 = s1;
-   for (i = 0 ; i < 8 ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* right top */
-   vb += 8;
-   tmp.v.y = y1;
-   tmp.v.v0 = t1;
-   for (i = 0 ; i < 8 ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* left top */
-   vb += 8;
-   tmp.v.x = x0;
-   tmp.v.u0 = s0;
-   for (i = 0 ; i < 8 ; i++)
-      vb[i] = tmp.ui[i];
-
-/*    fprintf(stderr, "%s: DV1: %x\n",  */
-/* 	   __FUNCTION__, i830->meta.Buffer[I830_DESTREG_DV1]); */
-}
-
-static void draw_poly(i830ContextPtr i830, 
-		      GLubyte red, GLubyte green, GLubyte blue, GLubyte alpha,
-                      GLuint numVerts,
-                      GLfloat verts[][2],
-                      GLfloat texcoords[][2])
-{
-   GLuint vertex_size = 8;
-   GLuint *vb = intelEmitInlinePrimitiveLocked( &i830->intel, 
-						PRIM3D_TRIFAN, 
-						numVerts * vertex_size,
-						vertex_size );
-   intelVertex tmp;
-   int i, k;
-
-   /* initial constant vertex fields */
-   tmp.v.z = 1.0;
-   tmp.v.w = 1.0; 
-   tmp.v.color.red = red;
-   tmp.v.color.green = green;
-   tmp.v.color.blue = blue;
-   tmp.v.color.alpha = alpha;
-   tmp.v.specular.red = 0;
-   tmp.v.specular.green = 0;
-   tmp.v.specular.blue = 0;
-   tmp.v.specular.alpha = 0;
-
-   for (k = 0; k < numVerts; k++) {
-      tmp.v.x = verts[k][0];
-      tmp.v.y = verts[k][1];
-      tmp.v.u0 = texcoords[k][0];
-      tmp.v.v0 = texcoords[k][1];
-
-      for (i = 0 ; i < vertex_size ; i++)
-         vb[i] = tmp.ui[i];
-
-      vb += vertex_size;
-   }
-}
-
-void 
-i830ClearWithTris(intelContextPtr intel, GLbitfield mask,
-		  GLboolean all,
-		  GLint cx, GLint cy, GLint cw, GLint ch)
-{
-   i830ContextPtr i830 = I830_CONTEXT( intel );
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   intelScreenPrivate *screen = intel->intelScreen;
-   int x0, y0, x1, y1;
-
-   INTEL_FIREVERTICES(intel);
-   SET_STATE( i830, meta );
-   set_initial_state( i830 );
-/*    set_no_texture( i830 ); */
-   set_vertex_format( i830 ); 
-
-   LOCK_HARDWARE(intel);
-
-   if(!all) {
-      x0 = cx;
-      y0 = cy;
-      x1 = x0 + cw;
-      y1 = y0 + ch;
-   } else {
-      x0 = 0;
-      y0 = 0;
-      x1 = x0 + dPriv->w;
-      y1 = y0 + dPriv->h;
-   }
-
-   /* Don't do any clipping to screen - these are window coordinates.
-    * The active cliprects will be applied as for any other geometry.
-    */
 
-   if(mask & BUFFER_BIT_FRONT_LEFT) {
-      set_no_depth_stencil_write( i830 );
-      set_color_mask( i830, GL_TRUE );
-      set_draw_region( i830, &screen->front );
-      draw_quad(i830, x0, x1, y0, y1,
-		intel->clear_red, intel->clear_green,
-		intel->clear_blue, intel->clear_alpha,
-		0, 0, 0, 0);
-   }
-
-   if(mask & BUFFER_BIT_BACK_LEFT) {
-      set_no_depth_stencil_write( i830 );
-      set_color_mask( i830, GL_TRUE );
-      set_draw_region( i830, &screen->back );
-
-      draw_quad(i830, x0, x1, y0, y1,
-		intel->clear_red, intel->clear_green,
-		intel->clear_blue, intel->clear_alpha,
-		0, 0, 0, 0);
-   }
-
-   if(mask & BUFFER_BIT_STENCIL) {
-      set_stencil_replace( i830, 
-			   intel->ctx.Stencil.WriteMask[0], 
-			   intel->ctx.Stencil.Clear);
-
-      set_color_mask( i830, GL_FALSE );
-      set_draw_region( i830, &screen->front );
-      draw_quad( i830, x0, x1, y0, y1, 0, 0, 0, 0, 0, 0, 0, 0 );
-   }
+   i830->meta.Ctx[I830_CTXREG_STATE3] &= ~CULLMODE_MASK;
+   i830->meta.Stipple[I830_STPREG_ST1] &= ~ST1_ENABLE;
+   i830->meta.emitted &= ~I830_UPLOAD_CTX;
 
-   UNLOCK_HARDWARE(intel);
 
-   INTEL_FIREVERTICES(intel);
-   SET_STATE( i830, state );
+   i830->meta.Buffer[I830_DESTREG_SENABLE] = i830->state.Buffer[I830_DESTREG_SENABLE];
+   i830->meta.Buffer[I830_DESTREG_SR1] = i830->state.Buffer[I830_DESTREG_SR1];
+   i830->meta.Buffer[I830_DESTREG_SR2] = i830->state.Buffer[I830_DESTREG_SR2];
+   i830->meta.emitted &= ~I830_UPLOAD_BUFFERS;
 }
 
 
-#if 0
 
-GLboolean
-i830TryTextureReadPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *pack,
-			  GLvoid *pixels )
+/* Select between front and back draw buffers.
+ */
+static void meta_draw_region( struct intel_context *intel,
+			      struct intel_region *draw_region,
+			      struct intel_region *depth_region )
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   intelScreenPrivate *screen = i830->intel.intelScreen;
-   GLint pitch = pack->RowLength ? pack->RowLength : width;
-   __DRIdrawablePrivate *dPriv = i830->intel.driDrawable;
-   int textureFormat;
-   GLenum glTextureFormat;
-   int src_offset = i830->meta.Buffer[I830_DESTREG_CBUFADDR2];
-   int destOffset = intelAgpOffsetFromVirtual( &i830->intel, pixels);
-   int destFormat, depthFormat, destPitch;
-   drm_clip_rect_t tmp;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-
-   if (	ctx->_ImageTransferState ||
-	pack->SwapBytes ||
-	pack->LsbFirst ||
-	!pack->Invert) {
-      fprintf(stderr, "%s: check_color failed\n", __FUNCTION__);
-      return GL_FALSE;
-   }
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   GLuint format;
+   GLuint depth_format = DEPTH_FRMT_16_FIXED;
 
-   switch (screen->fbFormat) {
-   case DV_PF_565:
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-      glTextureFormat = GL_RGB;
-      break;
-   case DV_PF_555:
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB1555;
-      glTextureFormat = GL_RGBA;
-      break;
-   case DV_PF_8888:
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-      glTextureFormat = GL_RGBA;
-      break;
-   default:
-      fprintf(stderr, "%s: textureFormat failed %x\n", __FUNCTION__,
-	      screen->fbFormat);
-      return GL_FALSE;
-   }
-
-
-   switch (type) {
-   case GL_UNSIGNED_SHORT_5_6_5: 
-      if (format != GL_RGB) return GL_FALSE;
-      destFormat = COLR_BUF_RGB565; 
-      depthFormat = DEPTH_FRMT_16_FIXED;
-      destPitch = pitch * 2;
-      break;
-   case GL_UNSIGNED_INT_8_8_8_8_REV: 
-      if (format != GL_BGRA) return GL_FALSE;
-      destFormat = COLR_BUF_ARGB8888; 
-      depthFormat = DEPTH_FRMT_24_FIXED_8_OTHER;
-      destPitch = pitch * 4;
-      break;
-   default:
-      fprintf(stderr, "%s: destFormat failed %s\n", __FUNCTION__,
-	      _mesa_lookup_enum_by_nr(type));
-      return GL_FALSE;
-   }
-
-   destFormat |= (0x02<<24);
-
-/*    fprintf(stderr, "type: %s destFormat: %x\n", */
-/* 	   _mesa_lookup_enum_by_nr(type), */
-/* 	   destFormat); */
-
-   intelFlush( ctx );
-
-   SET_STATE( i830, meta );
-   set_initial_state( i830 );
-   set_no_depth_stencil_write( i830 );
-
-   LOCK_HARDWARE( intel );
-   {
-      intelWaitForIdle( intel ); /* required by GL */
-
-      if (!driClipRectToFramebuffer(ctx->ReadBuffer, &x, &y, &width, &height)) {
-	 UNLOCK_HARDWARE( intel );
-	 SET_STATE(i830, state);
-	 fprintf(stderr, "%s: cliprect failed\n", __FUNCTION__);
-	 return GL_TRUE;
-      }
-
-#if 0
-      /* FIXME -- Just emit the correct state
-       */
-      if (i830SetParam(i830->driFd, I830_SETPARAM_CBUFFER_PITCH, 
-		      destPitch) != 0) {
-	 UNLOCK_HARDWARE( intel );
-	 SET_STATE(i830, state);
-	 fprintf(stderr, "%s: setparam failed\n", __FUNCTION__);
-	 return GL_FALSE;
-      }
-#endif
-
-
-      y = dPriv->h - y - height;
-      x += dPriv->x;
-      y += dPriv->y;
-
-
-      /* Set the frontbuffer up as a large rectangular texture.
-       */
-      set_tex_rect_source( i830, 
-			   src_offset, 
-			   screen->width, 
-			   screen->height, 
-			   screen->front.pitch, 
-			   textureFormat ); 
-   
+   intel_region_release(intel, &i830->meta.draw_region);
+   intel_region_reference(&i830->meta.draw_region, draw_region);
    
-      enable_texture_blend_replace( i830 ); 
-
-
-      /* Set the 3d engine to draw into the agp memory
-       */
-
-      set_draw_region( i830, destOffset ); 
-      set_draw_format( i830, destFormat, depthFormat );  
+   intel_region_release(intel, &i830->meta.depth_region);   
+   intel_region_reference(&i830->meta.depth_region, depth_region);
 
+   /* XXX FBO: grab code from i915 meta_draw_region */
 
-      /* Draw a single quad, no cliprects:
-       */
-      i830->intel.numClipRects = 1;
-      i830->intel.pClipRects = &tmp;
-      i830->intel.pClipRects[0].x1 = 0;
-      i830->intel.pClipRects[0].y1 = 0;
-      i830->intel.pClipRects[0].x2 = width;
-      i830->intel.pClipRects[0].y2 = height;
-
-      draw_quad( i830, 
-		 0, width, 0, height, 
-		 0, 255, 0, 0, 
-		 x, x+width, y, y+height );
-
-      intelWindowMoved( intel );
-   }
-   UNLOCK_HARDWARE( intel );
-   intelFinish( ctx ); /* required by GL */
-
-   SET_STATE( i830, state );
-   return GL_TRUE;
-}
-
-
-GLboolean
-i830TryTextureDrawPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *unpack,
-			  const GLvoid *pixels )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   GLint pitch = unpack->RowLength ? unpack->RowLength : width;
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   int textureFormat;
-   GLenum glTextureFormat;
-   int dst_offset = i830->meta.Buffer[I830_DESTREG_CBUFADDR2];
-   int src_offset = intelAgpOffsetFromVirtual( intel, pixels );
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Todo -- upload images that aren't in agp space, then texture
-    * from them.  
+   /* XXX: 555 support?
     */
+   if (draw_region->cpp == 2)
+      format = DV_PF_565;
+   else
+      format = DV_PF_8888;
 
-   if ( !intelIsAgpMemory( intel, pixels, pitch*height ) ) {
-      fprintf(stderr, "%s: intelIsAgpMemory failed\n", __FUNCTION__);
-      return GL_FALSE;
+   if (depth_region) {
+      if (depth_region->cpp == 2)
+	 depth_format = DEPTH_FRMT_16_FIXED;
+      else
+	 depth_format = DEPTH_FRMT_24_FIXED_8_OTHER;
    }
 
-   /* Todo -- don't want to clobber all the drawing state like we do
-    * for readpixels -- most of this state can be handled just fine.
-    */
-   if (	ctx->_ImageTransferState ||
-	unpack->SwapBytes ||
-	unpack->LsbFirst ||
-	ctx->Color.AlphaEnabled || 
-	ctx->Depth.Test ||
-	ctx->Fog.Enabled ||
-	ctx->Scissor.Enabled ||
-	ctx->Stencil.Enabled ||
-	!ctx->Color.ColorMask[0] ||
-	!ctx->Color.ColorMask[1] ||
-	!ctx->Color.ColorMask[2] ||
-	!ctx->Color.ColorMask[3] ||
-	ctx->Color.ColorLogicOpEnabled ||
-	ctx->Texture._EnabledUnits ||
-	ctx->Depth.OcclusionTest) {
-      fprintf(stderr, "%s: other tests failed\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   /* Todo -- remove these restrictions:
-    */
-   if (ctx->Pixel.ZoomX != 1.0F ||
-       ctx->Pixel.ZoomY != -1.0F)
-      return GL_FALSE;
-
-
-
-   switch (type) {
-   case GL_UNSIGNED_SHORT_1_5_5_5_REV:
-      if (format != GL_BGRA) return GL_FALSE;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB1555;
-      glTextureFormat = GL_RGBA;
-      break;
-   case GL_UNSIGNED_SHORT_5_6_5: 
-      if (format != GL_RGB) return GL_FALSE;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-      glTextureFormat = GL_RGB;
-      break;
-   case GL_UNSIGNED_SHORT_8_8_MESA: 
-      if (format != GL_YCBCR_MESA) return GL_FALSE;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_SWAPY 
-/*  		       | TM0S1_COLORSPACE_CONVERSION */
-	 );
-      glTextureFormat = GL_YCBCR_MESA;
-      break;
-   case GL_UNSIGNED_SHORT_8_8_REV_MESA: 
-      if (format != GL_YCBCR_MESA) return GL_FALSE;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_NORMAL 
-/* 		       | TM0S1_COLORSPACE_CONVERSION */
-	 );
-      glTextureFormat = GL_YCBCR_MESA;
-      break;
-   case GL_UNSIGNED_INT_8_8_8_8_REV: 
-      if (format != GL_BGRA) return GL_FALSE;
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-      glTextureFormat = GL_RGBA;
-      break;
-   default:
-      fprintf(stderr, "%s: destFormat failed\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   intelFlush( ctx );
-
-   SET_STATE( i830, meta );
-
-   LOCK_HARDWARE( intel );
-   {
-      intelWaitForIdle( intel ); /* required by GL */
-
-      y -= height;			/* cope with pixel zoom */
-
-      if (!driClipRectToFramebuffer(ctx->ReadBuffer, &x, &y, &width, &height)) {
-	 UNLOCK_HARDWARE( intel );
-	 SET_STATE(i830, state);
-	 fprintf(stderr, "%s: cliprect failed\n", __FUNCTION__);
-	 return GL_TRUE;
-      }
-
-
-      y = dPriv->h - y - height;
-
-      set_initial_state( i830 );
-
-      /* Set the pixel image up as a rectangular texture.
-       */
-      set_tex_rect_source( i830, 
-			   src_offset, 
-			   width, 
-			   height, 
-			   pitch, /* XXXX!!!! -- /2 sometimes */
-			   textureFormat ); 
-   
-   
-      enable_texture_blend_replace( i830 ); 
-
-   
-      /* Draw to the current draw buffer:
-       */
-      set_draw_offset( i830, dst_offset );
-
-      /* Draw a quad, use regular cliprects
-       */
-/*       fprintf(stderr, "x: %d y: %d width %d height %d\n", x, y, width, height); */
-
-      draw_quad( i830, 
-		 x, x+width, y, y+height,
-		 0, 255, 0, 0, 
-		 0, width, 0, height );
-
-      intelWindowMoved( intel );
-   }
-   UNLOCK_HARDWARE( intel );
-   intelFinish( ctx ); /* required by GL */
-   
-   SET_STATE(i830, state);
+   i830->meta.Buffer[I830_DESTREG_DV1] = (DSTORG_HORT_BIAS(0x8) | /* .5 */
+					  DSTORG_VERT_BIAS(0x8) | /* .5 */
+					  format |
+					  DEPTH_IS_Z |
+					  depth_format);
 
-   return GL_TRUE;
+   i830->meta.emitted &= ~I830_UPLOAD_BUFFERS;
 }
 
-#endif
 
-/**
- * Copy the window contents named by dPriv to the rotated (or reflected)
- * color buffer.
- * srcBuf is BUFFER_BIT_FRONT_LEFT or BUFFER_BIT_BACK_LEFT to indicate the source.
+/* Operations where the 3D engine is decoupled temporarily from the
+ * current GL state and used for other purposes than simply rendering
+ * incoming triangles.
  */
-void
-i830RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                 GLuint srcBuf)
+static void install_meta_state( struct intel_context *intel )
 {
-   i830ContextPtr i830 = I830_CONTEXT( intel );
-   intelScreenPrivate *screen = intel->intelScreen;
-   const GLuint cpp = screen->cpp;
-   drm_clip_rect_t fullRect;
-   GLuint textureFormat, srcOffset, srcPitch;
-   const drm_clip_rect_t *clipRects;
-   int numClipRects;
-   int i;
-
-   int xOrig, yOrig;
-   int origNumClipRects;
-   drm_clip_rect_t *origRects;
-
-   /*
-    * set up hardware state
-    */
-   intelFlush( &intel->ctx );
-
-   SET_STATE( i830, meta ); 
-   set_initial_state( i830 ); 
-   set_no_texture( i830 ); 
-   set_vertex_format( i830 ); 
-   set_no_depth_stencil_write( i830 );
-   set_color_mask( i830, GL_FALSE );
-
-   LOCK_HARDWARE(intel);
-
-   /* save current drawing origin and cliprects (restored at end) */
-   xOrig = intel->drawX;
-   yOrig = intel->drawY;
-   origNumClipRects = intel->numClipRects;
-   origRects = intel->pClipRects;
-
-   if (!intel->numClipRects)
-      goto done;
-
-   /*
-    * set drawing origin, cliprects for full-screen access to rotated screen
-    */
-   fullRect.x1 = 0;
-   fullRect.y1 = 0;
-   fullRect.x2 = screen->rotatedWidth;
-   fullRect.y2 = screen->rotatedHeight;
-   intel->drawX = 0;
-   intel->drawY = 0;
-   intel->numClipRects = 1;
-   intel->pClipRects = &fullRect;
-
-   set_draw_region( i830, &screen->rotated );
-
-   if (cpp == 4)
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-   else
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   memcpy(&i830->meta, &i830->initial, sizeof(i830->meta) );
 
-   if (srcBuf == BUFFER_BIT_FRONT_LEFT) {
-      srcPitch = screen->front.pitch;   /* in bytes */
-      srcOffset = screen->front.offset; /* bytes */
-      clipRects = dPriv->pClipRects;
-      numClipRects = dPriv->numClipRects;
-   }
-   else {
-      srcPitch = screen->back.pitch;   /* in bytes */
-      srcOffset = screen->back.offset; /* bytes */
-      clipRects = dPriv->pBackClipRects;
-      numClipRects = dPriv->numBackClipRects;
-   }
+   i830->meta.active = ACTIVE;
+   i830->meta.emitted = 0;
 
-   /* set the whole screen up as a texture to avoid alignment issues */
-   set_tex_rect_source(i830,
-                       srcOffset,
-                       screen->width,
-                       screen->height,
-                       srcPitch,
-                       textureFormat);
+   SET_STATE(i830, meta);
+   set_vertex_format(intel);
+   set_no_texture(intel);
+}
 
-   enable_texture_blend_replace(i830);
+static void leave_meta_state( struct intel_context *intel )
+{
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   intel_region_release(intel, &i830->meta.draw_region);
+   intel_region_release(intel, &i830->meta.depth_region);
+/*    intel_region_release(intel, &i830->meta.tex_region[0]); */
+   SET_STATE(i830, state);
+}
 
-   /*
-    * loop over the source window's cliprects
-    */
-   for (i = 0; i < numClipRects; i++) {
-      int srcX0 = clipRects[i].x1;
-      int srcY0 = clipRects[i].y1;
-      int srcX1 = clipRects[i].x2;
-      int srcY1 = clipRects[i].y2;
-      GLfloat verts[4][2], tex[4][2];
-      int j;
-
-      /* build vertices for four corners of clip rect */
-      verts[0][0] = srcX0;  verts[0][1] = srcY0;
-      verts[1][0] = srcX1;  verts[1][1] = srcY0;
-      verts[2][0] = srcX1;  verts[2][1] = srcY1;
-      verts[3][0] = srcX0;  verts[3][1] = srcY1;
-
-      /* .. and texcoords */
-      tex[0][0] = srcX0;  tex[0][1] = srcY0;
-      tex[1][0] = srcX1;  tex[1][1] = srcY0;
-      tex[2][0] = srcX1;  tex[2][1] = srcY1;
-      tex[3][0] = srcX0;  tex[3][1] = srcY1;
-
-      /* transform coords to rotated screen coords */
-
-      for (j = 0; j < 4; j++) {
-         matrix23TransformCoordf(&screen->rotMatrix,
-                                 &verts[j][0], &verts[j][1]);
-      }
 
-      /* draw polygon to map source image to dest region */
-      draw_poly(i830, 255, 255, 255, 255, 4, verts, tex);
 
-   } /* cliprect loop */
+void i830InitMetaFuncs( struct i830_context *i830 )
+{
+   i830->intel.vtbl.install_meta_state = install_meta_state;
+   i830->intel.vtbl.leave_meta_state = leave_meta_state;
+   i830->intel.vtbl.meta_no_depth_write = set_no_depth_write;
+   i830->intel.vtbl.meta_no_stencil_write = set_no_stencil_write;
+   i830->intel.vtbl.meta_stencil_replace = set_stencil_replace;
+   i830->intel.vtbl.meta_depth_replace = set_depth_replace;
+   i830->intel.vtbl.meta_color_mask = set_color_mask;
+   i830->intel.vtbl.meta_no_texture = set_no_texture;
+   i830->intel.vtbl.meta_texture_blend_replace = set_texture_blend_replace;
+   i830->intel.vtbl.meta_tex_rect_source = set_tex_rect_source;
+   i830->intel.vtbl.meta_draw_region = meta_draw_region;
+   i830->intel.vtbl.meta_import_pixel_state = meta_import_pixel_state;
+}
 
-   assert(!intel->prim.flush); 
-   intelFlushBatchLocked( intel, GL_FALSE, GL_FALSE, GL_FALSE );
 
- done:
-   /* restore original drawing origin and cliprects */
-   intel->drawX = xOrig;
-   intel->drawY = yOrig;
-   intel->numClipRects = origNumClipRects;
-   intel->pClipRects = origRects;
 
-   UNLOCK_HARDWARE(intel);
 
-   SET_STATE( i830, state );
-}
 
diff --git a/src/mesa/drivers/dri/i915/i830_reg.h b/src/mesa/drivers/dri/i915/i830_reg.h
index 98cee2f214..ff08683d7e 100644
--- a/src/mesa/drivers/dri/i915/i830_reg.h
+++ b/src/mesa/drivers/dri/i915/i830_reg.h
@@ -407,7 +407,7 @@
 #define LOGICOP_SET			0xf
 #define MODE4_ENABLE_STENCIL_TEST_MASK	((1<<17)|(0xff00))
 #define ENABLE_STENCIL_TEST_MASK	(1<<17)
-#define STENCIL_TEST_MASK(x)		((x)<<8)
+#define STENCIL_TEST_MASK(x)		(((x)&0xff)<<8)
 #define MODE4_ENABLE_STENCIL_WRITE_MASK	((1<<16)|(0x00ff))
 #define ENABLE_STENCIL_WRITE_MASK	(1<<16)
 #define STENCIL_WRITE_MASK(x)		((x)&0xff)
diff --git a/src/mesa/drivers/dri/i915/i830_state.c b/src/mesa/drivers/dri/i915/i830_state.c
index ec9dca6231..5c0a3bffc2 100644
--- a/src/mesa/drivers/dri/i915/i830_state.c
+++ b/src/mesa/drivers/dri/i915/i830_state.c
@@ -36,6 +36,7 @@
 
 #include "intel_screen.h"
 #include "intel_batchbuffer.h"
+#include "intel_fbo.h"
 
 #include "i830_context.h"
 #include "i830_reg.h"
@@ -44,7 +45,7 @@ static void
 i830StencilFuncSeparate(GLcontext *ctx, GLenum face, GLenum func, GLint ref,
                         GLuint mask)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    int test = intel_translate_compare_func(func);
 
    mask = mask & 0xff;
@@ -69,7 +70,7 @@ i830StencilFuncSeparate(GLcontext *ctx, GLenum face, GLenum func, GLint ref,
 static void
 i830StencilMaskSeparate(GLcontext *ctx, GLenum face, GLuint mask)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
 
    if (INTEL_DEBUG&DEBUG_DRI)
       fprintf(stderr, "%s : mask 0x%x\n", __FUNCTION__, mask);
@@ -86,7 +87,7 @@ static void
 i830StencilOpSeparate(GLcontext *ctx, GLenum face, GLenum fail, GLenum zfail,
                       GLenum zpass)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    int fop, dfop, dpop;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -193,7 +194,7 @@ i830StencilOpSeparate(GLcontext *ctx, GLenum face, GLenum fail, GLenum zfail,
 
 static void i830AlphaFunc(GLcontext *ctx, GLenum func, GLfloat ref)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    int test = intel_translate_compare_func(func);
    GLubyte refByte;
    GLuint refInt;
@@ -221,7 +222,7 @@ static void i830AlphaFunc(GLcontext *ctx, GLenum func, GLfloat ref)
  */
 static void i830EvalLogicOpBlendState(GLcontext *ctx)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
 
    I830_STATECHANGE(i830, I830_UPLOAD_CTX);
 
@@ -245,7 +246,7 @@ static void i830EvalLogicOpBlendState(GLcontext *ctx)
 
 static void i830BlendColor(GLcontext *ctx, const GLfloat color[4])
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    GLubyte r, g, b, a;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -268,7 +269,7 @@ static void i830BlendColor(GLcontext *ctx, const GLfloat color[4])
  */
 static void i830_set_blend_state( GLcontext * ctx )
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    int funcA;
    int funcRGB;
    int eqnA;
@@ -406,7 +407,7 @@ static void i830BlendFuncSeparate(GLcontext *ctx, GLenum sfactorRGB,
 
 static void i830DepthFunc(GLcontext *ctx, GLenum func)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    int test = intel_translate_compare_func(func);
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -420,7 +421,7 @@ static void i830DepthFunc(GLcontext *ctx, GLenum func)
 
 static void i830DepthMask(GLcontext *ctx, GLboolean flag)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
 
    if (INTEL_DEBUG&DEBUG_DRI)
       fprintf(stderr, "%s flag (%d)\n", __FUNCTION__, flag);
@@ -443,7 +444,7 @@ static void i830DepthMask(GLcontext *ctx, GLboolean flag)
  */
 static void i830PolygonStipple( GLcontext *ctx, const GLubyte *mask )
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    const GLubyte *m = mask;
    GLubyte p[4];
    int i,j,k;
@@ -496,15 +497,14 @@ static void i830PolygonStipple( GLcontext *ctx, const GLubyte *mask )
 static void i830Scissor(GLcontext *ctx, GLint x, GLint y, 
 			GLsizei w, GLsizei h)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   intelScreenPrivate *screen = i830->intel.intelScreen;
+   struct i830_context *i830 = i830_context(ctx);
    int x1, y1, x2, y2;
 
-   if (!i830->intel.driDrawable)
+   if (!ctx->DrawBuffer)
       return;
 
    x1 = x;
-   y1 = i830->intel.driDrawable->h - (y + h);
+   y1 = ctx->DrawBuffer->Height - (y + h);
    x2 = x + w - 1;
    y2 = y1 + h - 1;
 
@@ -512,16 +512,10 @@ static void i830Scissor(GLcontext *ctx, GLint x, GLint y,
       fprintf(stderr, "[%s] x(%d) y(%d) w(%d) h(%d)\n", __FUNCTION__,
 	      x, y, w, h);
 
-   if (x1 < 0) x1 = 0;
-   if (y1 < 0) y1 = 0;
-   if (x2 < 0) x2 = 0;
-   if (y2 < 0) y2 = 0;
-
-   if (x2 >= screen->width) x2 = screen->width-1;
-   if (y2 >= screen->height) y2 = screen->height-1;
-   if (x1 >= screen->width) x1 = screen->width-1;
-   if (y1 >= screen->height) y1 = screen->height-1;
-
+   x1 = CLAMP(x1, 0, ctx->DrawBuffer->Width - 1);
+   y1 = CLAMP(y1, 0, ctx->DrawBuffer->Height - 1);
+   x2 = CLAMP(x2, 0, ctx->DrawBuffer->Width - 1);
+   y2 = CLAMP(y2, 0, ctx->DrawBuffer->Height - 1);
 
    I830_STATECHANGE(i830, I830_UPLOAD_BUFFERS);
    i830->state.Buffer[I830_DESTREG_SR1] = (y1 << 16) | (x1 & 0xffff);
@@ -530,7 +524,7 @@ static void i830Scissor(GLcontext *ctx, GLint x, GLint y,
 
 static void i830LogicOp(GLcontext *ctx, GLenum opcode)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    int tmp = intel_translate_logic_op( opcode );
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -545,7 +539,7 @@ static void i830LogicOp(GLcontext *ctx, GLenum opcode)
 
 static void i830CullFaceFrontFace(GLcontext *ctx, GLenum unused)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    GLuint mode;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -573,7 +567,7 @@ static void i830CullFaceFrontFace(GLcontext *ctx, GLenum unused)
 
 static void i830LineWidth( GLcontext *ctx, GLfloat widthf )
 {
-   i830ContextPtr i830 = I830_CONTEXT( ctx );
+   struct i830_context *i830 = i830_context( ctx );
    int width;
    int state5;
 
@@ -594,7 +588,7 @@ static void i830LineWidth( GLcontext *ctx, GLfloat widthf )
 
 static void i830PointSize(GLcontext *ctx, GLfloat size)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    GLint point_size = (int)size;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -616,7 +610,7 @@ static void i830ColorMask(GLcontext *ctx,
 			  GLboolean r, GLboolean g,
 			  GLboolean b, GLboolean a)
 {
-   i830ContextPtr i830 = I830_CONTEXT( ctx );
+   struct i830_context *i830 = i830_context( ctx );
    GLuint tmp = 0;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -638,7 +632,7 @@ static void i830ColorMask(GLcontext *ctx,
 
 static void update_specular( GLcontext *ctx )
 {
-   i830ContextPtr i830 = I830_CONTEXT( ctx );
+   struct i830_context *i830 = i830_context( ctx );
 
    I830_STATECHANGE(i830, I830_UPLOAD_CTX);
    i830->state.Ctx[I830_CTXREG_ENABLES_1] &= ~ENABLE_SPEC_ADD_MASK;
@@ -664,7 +658,7 @@ static void i830LightModelfv(GLcontext *ctx, GLenum pname,
  */
 static void i830ShadeModel(GLcontext *ctx, GLenum mode)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
    I830_STATECHANGE(i830, I830_UPLOAD_CTX);
 
 
@@ -690,7 +684,7 @@ static void i830ShadeModel(GLcontext *ctx, GLenum mode)
  */
 static void i830Fogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
 
    if (INTEL_DEBUG&DEBUG_DRI)
       fprintf(stderr, "%s\n", __FUNCTION__);
@@ -710,7 +704,7 @@ static void i830Fogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 
 static void i830Enable(GLcontext *ctx, GLenum cap, GLboolean state)
 {
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
+   struct i830_context *i830 = i830_context(ctx);
 
    switch(cap) {
    case GL_LIGHTING:
@@ -806,20 +800,28 @@ static void i830Enable(GLcontext *ctx, GLenum cap, GLboolean state)
       break;
 
    case GL_STENCIL_TEST:
-      if (i830->intel.hw_stencil) {
-	 I830_STATECHANGE(i830, I830_UPLOAD_CTX);
-
-	 if (state) {
-	    i830->state.Ctx[I830_CTXREG_ENABLES_1] |= ENABLE_STENCIL_TEST;
-	    i830->state.Ctx[I830_CTXREG_ENABLES_2] |= ENABLE_STENCIL_WRITE;
-	 } else {
-	    i830->state.Ctx[I830_CTXREG_ENABLES_1] &= ~ENABLE_STENCIL_TEST;
-	    i830->state.Ctx[I830_CTXREG_ENABLES_2] &= ~ENABLE_STENCIL_WRITE;
-	    i830->state.Ctx[I830_CTXREG_ENABLES_1] |= DISABLE_STENCIL_TEST;
-	    i830->state.Ctx[I830_CTXREG_ENABLES_2] |= DISABLE_STENCIL_WRITE;
-	 }
-      } else {
-	 FALLBACK( &i830->intel, I830_FALLBACK_STENCIL, state );
+      {
+         GLboolean hw_stencil = GL_FALSE;
+         if (ctx->DrawBuffer) {
+            struct intel_renderbuffer *irbStencil
+               = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL);
+            hw_stencil = (irbStencil && irbStencil->region);
+         }
+         if (hw_stencil) {
+            I830_STATECHANGE(i830, I830_UPLOAD_CTX);
+
+            if (state) {
+               i830->state.Ctx[I830_CTXREG_ENABLES_1] |= ENABLE_STENCIL_TEST;
+               i830->state.Ctx[I830_CTXREG_ENABLES_2] |= ENABLE_STENCIL_WRITE;
+            } else {
+               i830->state.Ctx[I830_CTXREG_ENABLES_1] &= ~ENABLE_STENCIL_TEST;
+               i830->state.Ctx[I830_CTXREG_ENABLES_2] &= ~ENABLE_STENCIL_WRITE;
+               i830->state.Ctx[I830_CTXREG_ENABLES_1] |= DISABLE_STENCIL_TEST;
+               i830->state.Ctx[I830_CTXREG_ENABLES_2] |= DISABLE_STENCIL_WRITE;
+            }
+         } else {
+            FALLBACK( &i830->intel, I830_FALLBACK_STENCIL, state );
+         }
       }
       break;
 
@@ -844,7 +846,7 @@ static void i830Enable(GLcontext *ctx, GLenum cap, GLboolean state)
 }
 
 
-static void i830_init_packets( i830ContextPtr i830 )
+static void i830_init_packets( struct i830_context *i830 )
 {
    intelScreenPrivate *screen = i830->intel.intelScreen;
 
@@ -902,6 +904,7 @@ static void i830_init_packets( i830ContextPtr i830 )
 					     DISABLE_COLOR_BLEND |
 					     DISABLE_DEPTH_TEST);
 
+#if 000 /* XXX all the stencil enable state is set in i830Enable(), right? */
    if (i830->intel.hw_stencil) {
       i830->state.Ctx[I830_CTXREG_ENABLES_2] = (_3DSTATE_ENABLES_2_CMD |
 						ENABLE_STENCIL_WRITE |
@@ -911,7 +914,9 @@ static void i830_init_packets( i830ContextPtr i830 )
 						/* set no color comps disabled */
 						ENABLE_COLOR_WRITE |
 						ENABLE_DEPTH_WRITE);
-   } else {
+   } else
+#endif
+   {
       i830->state.Ctx[I830_CTXREG_ENABLES_2] = (_3DSTATE_ENABLES_2_CMD |
 						DISABLE_STENCIL_WRITE |
 						ENABLE_TEX_CACHE |
@@ -1012,13 +1017,10 @@ static void i830_init_packets( i830ContextPtr i830 )
       (BUF_3D_ID_DEPTH |
        BUF_3D_PITCH(screen->depth.pitch) |  /* pitch in bytes */
        BUF_3D_USE_FENCE);
-   i830->state.Buffer[I830_DESTREG_DBUFADDR2] = screen->depth.offset;
-
 
    i830->state.Buffer[I830_DESTREG_DV0] = _3DSTATE_DST_BUF_VARS_CMD;
 
    switch (screen->fbFormat) {
-   case DV_PF_555:
    case DV_PF_565:
       i830->state.Buffer[I830_DESTREG_DV1] = (DSTORG_HORT_BIAS(0x8) | /* .5 */
 					      DSTORG_VERT_BIAS(0x8) | /* .5 */
@@ -1068,7 +1070,7 @@ void i830InitStateFuncs( struct dd_function_table *functions )
    functions->StencilOpSeparate = i830StencilOpSeparate;
 }
 
-void i830InitState( i830ContextPtr i830 )
+void i830InitState( struct i830_context *i830 )
 {
    GLcontext *ctx = &i830->intel.ctx;
 
@@ -1080,7 +1082,8 @@ void i830InitState( i830ContextPtr i830 )
 
    i830->current = &i830->state;
    i830->state.emitted = 0;
-   i830->state.active = (I830_UPLOAD_TEXBLEND(0) |
+   i830->state.active = (I830_UPLOAD_INVARIENT |
+			 I830_UPLOAD_TEXBLEND(0) |
 			 I830_UPLOAD_STIPPLE |
 			 I830_UPLOAD_CTX |
 			 I830_UPLOAD_BUFFERS);
diff --git a/src/mesa/drivers/dri/i915/i830_tex.c b/src/mesa/drivers/dri/i915/i830_tex.c
index 3c4aedb35c..a230eb2e46 100644
--- a/src/mesa/drivers/dri/i915/i830_tex.c
+++ b/src/mesa/drivers/dri/i915/i830_tex.c
@@ -45,261 +45,12 @@
 
 
 
-
-/**
- * Set the texture wrap modes.
- * 
- * The i830M (and related graphics cores) do not support GL_CLAMP.  The Intel
- * drivers for "other operating systems" implement GL_CLAMP as
- * GL_CLAMP_TO_EDGE, so the same is done here.
- * 
- * \param t Texture object whose wrap modes are to be set
- * \param swrap Wrap mode for the \a s texture coordinate
- * \param twrap Wrap mode for the \a t texture coordinate
- */
-static void i830SetTexWrapping(i830TextureObjectPtr tex,
-			       GLenum swrap, 
-			       GLenum twrap)
-{
-   tex->Setup[I830_TEXREG_MCS] &= ~(TEXCOORD_ADDR_U_MASK|TEXCOORD_ADDR_V_MASK);
-
-   switch( swrap ) {
-   case GL_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_WRAP);
-      break;
-   case GL_CLAMP:
-   case GL_CLAMP_TO_EDGE:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_CLAMP);
-      break;
-   case GL_CLAMP_TO_BORDER:
-      tex->Setup[I830_TEXREG_MCS] |= 
-			TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_CLAMP_BORDER);
-      break;
-   case GL_MIRRORED_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |= 
-			TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_MIRROR);
-      break;
-   default:
-      break;
-   }
-
-   switch( twrap ) {
-   case GL_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_WRAP);
-      break;
-   case GL_CLAMP:
-   case GL_CLAMP_TO_EDGE:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_CLAMP);
-      break;
-   case GL_CLAMP_TO_BORDER:
-      tex->Setup[I830_TEXREG_MCS] |= 
-			TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_CLAMP_BORDER);
-      break;
-   case GL_MIRRORED_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |=
-			TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_MIRROR);
-      break;
-   default:
-      break;
-   }
-}
-
-
-/**
- * Set the texture magnification and minification modes.
- * 
- * \param t Texture whose filter modes are to be set
- * \param minf Texture minification mode
- * \param magf Texture magnification mode
- * \param bias LOD bias for this texture unit.
- */
-
-static void i830SetTexFilter( i830TextureObjectPtr t, GLenum minf, GLenum magf,
-			      GLfloat maxanisotropy )
-{
-   int minFilt = 0, mipFilt = 0, magFilt = 0;
-
-   if(INTEL_DEBUG&DEBUG_DRI)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if ( maxanisotropy > 1.0 ) {
-      minFilt = FILTER_ANISOTROPIC;
-      magFilt = FILTER_ANISOTROPIC;
-   }
-   else {
-      switch (minf) {
-      case GL_NEAREST:
-	 minFilt = FILTER_NEAREST;
-	 mipFilt = MIPFILTER_NONE;
-	 break;
-      case GL_LINEAR:
-	 minFilt = FILTER_LINEAR;
-	 mipFilt = MIPFILTER_NONE;
-	 break;
-      case GL_NEAREST_MIPMAP_NEAREST:
-	 minFilt = FILTER_NEAREST;
-	 mipFilt = MIPFILTER_NEAREST;
-	 break;
-      case GL_LINEAR_MIPMAP_NEAREST:
-	 minFilt = FILTER_LINEAR;
-	 mipFilt = MIPFILTER_NEAREST;
-	 break;
-      case GL_NEAREST_MIPMAP_LINEAR:
-	 minFilt = FILTER_NEAREST;
-	 mipFilt = MIPFILTER_LINEAR;
-	 break;
-      case GL_LINEAR_MIPMAP_LINEAR:
-	 minFilt = FILTER_LINEAR;
-	 mipFilt = MIPFILTER_LINEAR;
-	 break;
-      default:
-	 break;
-      }
-
-      switch (magf) {
-      case GL_NEAREST:
-	 magFilt = FILTER_NEAREST;
-	 break;
-      case GL_LINEAR:
-	 magFilt = FILTER_LINEAR;
-	 break;
-      default:
-	 break;
-      }  
-   }
-
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIN_FILTER_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIP_FILTER_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MAG_FILTER_MASK;
-   t->Setup[I830_TEXREG_TM0S3] |= ((minFilt << TM0S3_MIN_FILTER_SHIFT) |
-				   (mipFilt << TM0S3_MIP_FILTER_SHIFT) |
-				   (magFilt << TM0S3_MAG_FILTER_SHIFT));
-}
-
-static void i830SetTexBorderColor(i830TextureObjectPtr t, GLubyte color[4])
-{
-   if(INTEL_DEBUG&DEBUG_DRI)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-    t->Setup[I830_TEXREG_TM0S4] = 
-        INTEL_PACKCOLOR8888(color[0],color[1],color[2],color[3]);
-}
-
-
-/**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-intelTextureObjectPtr i830AllocTexObj( struct gl_texture_object *texObj )
-{
-   i830TextureObjectPtr t = CALLOC_STRUCT( i830_texture_object );
-   if ( !t ) 
-      return NULL;
-
-   texObj->DriverData = t;
-   t->intel.base.tObj = texObj;
-   t->intel.dirty = I830_UPLOAD_TEX_ALL;
-   make_empty_list( &t->intel.base );
-
-   t->Setup[I830_TEXREG_TM0LI] = 0; /* not used */
-   t->Setup[I830_TEXREG_TM0S0] = 0;
-   t->Setup[I830_TEXREG_TM0S1] = 0;
-   t->Setup[I830_TEXREG_TM0S2] = 0;
-   t->Setup[I830_TEXREG_TM0S3] = 0;
-   t->Setup[I830_TEXREG_MCS] = (_3DSTATE_MAP_COORD_SET_CMD |
-				MAP_UNIT(0) |
-				ENABLE_TEXCOORD_PARAMS |
-				TEXCOORDS_ARE_NORMAL |
-				TEXCOORDTYPE_CARTESIAN |
-				ENABLE_ADDR_V_CNTL |
-				TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_WRAP) |
-				ENABLE_ADDR_U_CNTL |
-				TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_WRAP));
-
-   
-   i830SetTexWrapping( t, texObj->WrapS, texObj->WrapT );
-   i830SetTexFilter( t, texObj->MinFilter, texObj->MagFilter, 
-		     texObj->MaxAnisotropy );
-   i830SetTexBorderColor( t, texObj->_BorderChan );
-
-   return &t->intel;
-}
-
-
-static void i830TexParameter( GLcontext *ctx, GLenum target,
-			      struct gl_texture_object *tObj,
-			      GLenum pname, const GLfloat *params )
-{
-   i830TextureObjectPtr t = (i830TextureObjectPtr) tObj->DriverData;
-   if (!t)
-      return;
-
-   switch (pname) {
-   case GL_TEXTURE_MIN_FILTER:
-   case GL_TEXTURE_MAG_FILTER:
-   case GL_TEXTURE_MAX_ANISOTROPY_EXT:
-      i830SetTexFilter( t, tObj->MinFilter, tObj->MagFilter,
-			tObj->MaxAnisotropy);
-      break;
-
-   case GL_TEXTURE_WRAP_S:
-   case GL_TEXTURE_WRAP_T:
-      i830SetTexWrapping( t, tObj->WrapS, tObj->WrapT );
-      break;
-  
-   case GL_TEXTURE_BORDER_COLOR:
-      i830SetTexBorderColor( t, tObj->_BorderChan );
-      break;
-
-   case GL_TEXTURE_BASE_LEVEL:
-   case GL_TEXTURE_MAX_LEVEL:
-   case GL_TEXTURE_MIN_LOD:
-   case GL_TEXTURE_MAX_LOD:
-      /* The i830 and its successors can do a lot of this without
-       * reloading the textures.  A project for someone?
-       */
-      intelFlush( ctx );
-      driSwapOutTextureObject( (driTextureObject *) t );
-      break;
-
-   default:
-      return;
-   }
-
-   t->intel.dirty = I830_UPLOAD_TEX_ALL;
-}
-
-
 static void i830TexEnv( GLcontext *ctx, GLenum target, 
 			GLenum pname, const GLfloat *param )
 {
-   i830ContextPtr i830 = I830_CONTEXT( ctx );
-   GLuint unit = ctx->Texture.CurrentUnit;
 
    switch (pname) {
    case GL_TEXTURE_ENV_COLOR: 
-#if 0
-   {
-      GLubyte r, g, b, a;
-      GLuint col;
-      
-      UNCLAMPED_FLOAT_TO_UBYTE(r, param[RCOMP]);
-      UNCLAMPED_FLOAT_TO_UBYTE(g, param[GCOMP]);
-      UNCLAMPED_FLOAT_TO_UBYTE(b, param[BCOMP]);
-      UNCLAMPED_FLOAT_TO_UBYTE(a, param[ACOMP]);
-
-      col = ((a << 24) | (r << 16) | (g << 8) | b);
-
-      if (col != i830->state.TexEnv[unit][I830_TEXENVREG_COL1]) {
-	 I830_STATECHANGE(i830, I830_UPLOAD_TEXENV);
-	 i830->state.TexEnv[unit][I830_TEXENVREG_COL1] = col;
-      }
-
-      break;
-   }
-#endif
    case GL_TEXTURE_ENV_MODE:
    case GL_COMBINE_RGB:
    case GL_COMBINE_ALPHA:
@@ -320,13 +71,13 @@ static void i830TexEnv( GLcontext *ctx, GLenum target,
       break;
 
    case GL_TEXTURE_LOD_BIAS: {
+      struct i830_context *i830 = i830_context( ctx );
+      GLuint unit = ctx->Texture.CurrentUnit;
       int b = (int) ((*param) * 16.0);
       if (b > 63) b = 63;
       if (b < -64) b = -64;
       I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_TM0S3] &= ~TM0S3_LOD_BIAS_MASK;
-      i830->state.Tex[unit][I830_TEXREG_TM0S3] |= 
-	 ((b << TM0S3_LOD_BIAS_SHIFT) & TM0S3_LOD_BIAS_MASK);
+      i830->lodbias_tm0s3[unit] = ((b << TM0S3_LOD_BIAS_SHIFT) & TM0S3_LOD_BIAS_MASK);
       break;
    }
 
@@ -335,22 +86,10 @@ static void i830TexEnv( GLcontext *ctx, GLenum target,
    }
 }
 
-static void i830BindTexture( GLcontext *ctx, GLenum target,
-			    struct gl_texture_object *texObj )
-{
-   i830TextureObjectPtr tex;
-   
-   if (!texObj->DriverData)
-      i830AllocTexObj( texObj );
-   
-   tex = (i830TextureObjectPtr)texObj->DriverData;
-}
 
 
 
 void i830InitTextureFuncs( struct dd_function_table *functions )
 {
-   functions->BindTexture 		= i830BindTexture;
    functions->TexEnv                    = i830TexEnv;
-   functions->TexParameter              = i830TexParameter;
 }
diff --git a/src/mesa/drivers/dri/i915/i830_texblend.c b/src/mesa/drivers/dri/i915/i830_texblend.c
index 49e0347643..4c1bee58d8 100644
--- a/src/mesa/drivers/dri/i915/i830_texblend.c
+++ b/src/mesa/drivers/dri/i915/i830_texblend.c
@@ -101,7 +101,7 @@ static GLuint emit_factor( GLuint blendUnit, GLuint *state, GLuint count,
 }
 
 
-static __inline__ GLuint GetTexelOp(GLint unit)
+static INLINE GLuint GetTexelOp(GLint unit)
 {
    switch(unit) {
    case 0: return TEXBLENDARG_TEXEL0;
@@ -132,7 +132,7 @@ static __inline__ GLuint GetTexelOp(GLint unit)
  * partial support for the extension?
  */
 GLuint
-i830SetTexEnvCombine(i830ContextPtr i830,
+i830SetTexEnvCombine(struct i830_context *i830,
 		     const struct gl_tex_env_combine_state * combine,
 		     GLint blendUnit,
 		     GLuint texel_op,
@@ -394,7 +394,7 @@ i830SetTexEnvCombine(i830ContextPtr i830,
 }
 
 
-static void emit_texblend( i830ContextPtr i830, GLuint unit, GLuint blendUnit,
+static void emit_texblend( struct i830_context *i830, GLuint unit, GLuint blendUnit,
 			   GLboolean last_stage )
 {
    struct gl_texture_unit *texUnit = &i830->intel.ctx.Texture.Unit[unit];
@@ -423,7 +423,7 @@ static void emit_texblend( i830ContextPtr i830, GLuint unit, GLuint blendUnit,
    I830_ACTIVESTATE(i830, I830_UPLOAD_TEXBLEND(blendUnit), GL_TRUE);
 }
 
-static void emit_passthrough( i830ContextPtr i830 )
+static void emit_passthrough( struct i830_context *i830 )
 {
    GLuint tmp[I830_TEXBLEND_SIZE], tmp_sz;
    GLuint unit = 0;
@@ -442,7 +442,7 @@ static void emit_passthrough( i830ContextPtr i830 )
    I830_ACTIVESTATE(i830, I830_UPLOAD_TEXBLEND(unit), GL_TRUE);
 }
 
-void i830EmitTextureBlend( i830ContextPtr i830 )
+void i830EmitTextureBlend( struct i830_context *i830 )
 {
    GLcontext *ctx = &i830->intel.ctx;
    GLuint unit, last_stage = 0, blendunit = 0;
diff --git a/src/mesa/drivers/dri/i915/i830_texstate.c b/src/mesa/drivers/dri/i915/i830_texstate.c
index ba972dac8f..dc9aca80fa 100644
--- a/src/mesa/drivers/dri/i915/i830_texstate.c
+++ b/src/mesa/drivers/dri/i915/i830_texstate.c
@@ -38,446 +38,284 @@
 #include "intel_screen.h"
 #include "intel_ioctl.h"
 #include "intel_tex.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
 
 #include "i830_context.h"
 #include "i830_reg.h"
 
-static const GLint initial_offsets[6][2] = { {0,0},
-				       {0,2},
-				       {1,0},
-				       {1,2},
-				       {1,1},
-				       {1,3} };
 
-static const GLint step_offsets[6][2] = { {0,2},
-				    {0,2},
-				    {-1,2},
-				    {-1,2},
-				    {-1,1},
-				    {-1,1} };
 
-#define I830_TEX_UNIT_ENABLED(unit)		(1<<unit)
-
-static GLboolean i830SetTexImages( i830ContextPtr i830, 
-				  struct gl_texture_object *tObj )
+static GLuint translate_texture_format( GLuint mesa_format )
 {
-   GLuint total_height, pitch, i, textureFormat;
-   i830TextureObjectPtr t = (i830TextureObjectPtr) tObj->DriverData;
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   GLint firstLevel, lastLevel, numLevels;
-
-   switch( baseImage->TexFormat->MesaFormat ) {
+   switch (mesa_format) {
    case MESA_FORMAT_L8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_L8;
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_L8;
    case MESA_FORMAT_I8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_I8;
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_I8;
    case MESA_FORMAT_A8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_I8; /* Kludge -- check with conform, glean */
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_I8; /* Kludge! */
    case MESA_FORMAT_AL88:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_AY88;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_AY88;
    case MESA_FORMAT_RGB565:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_RGB565;
    case MESA_FORMAT_ARGB1555:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB1555;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_ARGB1555;
    case MESA_FORMAT_ARGB4444:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB4444;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_ARGB4444;
    case MESA_FORMAT_ARGB8888:
-      t->intel.texelBytes = 4;
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-      break;
-
+      return MAPSURF_32BIT | MT_32BIT_ARGB8888;
    case MESA_FORMAT_YCBCR_REV:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_NORMAL | 
-		       TM0S1_COLORSPACE_CONVERSION);
-      break;
-
+      return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
    case MESA_FORMAT_YCBCR:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_SWAPY | /* ??? */
-		       TM0S1_COLORSPACE_CONVERSION);
-      break;
-
+      return (MAPSURF_422 | MT_422_YCRCB_SWAPY);
    case MESA_FORMAT_RGB_FXT1:
    case MESA_FORMAT_RGBA_FXT1:
-     t->intel.texelBytes = 2;
-     textureFormat = MAPSURF_COMPRESSED | MT_COMPRESS_FXT1;
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_FXT1);
    case MESA_FORMAT_RGBA_DXT1:
    case MESA_FORMAT_RGB_DXT1:
-     /* 
-      * DXTn pitches are Width/4 * blocksize in bytes 
-      * for DXT1: blocksize=8 so Width/4*8 = Width * 2 
-      * for DXT3/5: blocksize=16 so Width/4*16 = Width * 4
-      */
-     t->intel.texelBytes = 2;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
-     break;
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
    case MESA_FORMAT_RGBA_DXT3:
-     t->intel.texelBytes = 4;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
-     break;
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
    case MESA_FORMAT_RGBA_DXT5:
-     t->intel.texelBytes = 4;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
    default:
-      fprintf(stderr, "%s: bad image format\n", __FUNCTION__);
+      fprintf(stderr, "%s: bad image format %x\n", __FUNCTION__,
+	      mesa_format);
       abort();
+      return 0;
    }
+}
 
-   /* Compute which mipmap levels we really want to send to the hardware.
-    * This depends on the base image size, GL_TEXTURE_MIN_LOD,
-    * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
-    * Yes, this looks overly complicated, but it's all needed.
-    */
-   driCalculateTextureFirstLastLevel( (driTextureObject *) t );
-
-
-   /* Figure out the amount of memory required to hold all the mipmap
-    * levels.  Choose the smallest pitch to accomodate the largest
-    * mipmap:
-    */
-   firstLevel = t->intel.base.firstLevel;
-   lastLevel = t->intel.base.lastLevel;
-   numLevels = lastLevel - firstLevel + 1;
 
 
-   /* All images must be loaded at this pitch.  Count the number of
-    * lines required:
-    */
-   switch (tObj->Target) {
-   case GL_TEXTURE_CUBE_MAP: {
-      const GLuint dim = tObj->Image[0][firstLevel]->Width;
-      GLuint face;
-
-      pitch = dim * t->intel.texelBytes;
-      pitch *= 2;		/* double pitch for cube layouts */
-      pitch = (pitch + 3) & ~3;
-      
-      total_height = dim * 4;
-
-      for ( face = 0 ; face < 6 ; face++) {
-	 GLuint x = initial_offsets[face][0] * dim;
-	 GLuint y = initial_offsets[face][1] * dim;
-	 GLuint d = dim;
-	 
-	 t->intel.base.dirty_images[face] = ~0;
-
-	 assert(tObj->Image[face][firstLevel]->Width == dim);
-	 assert(tObj->Image[face][firstLevel]->Height == dim);
-
-	 for (i = 0; i < numLevels; i++) {
-	    t->intel.image[face][i].image = tObj->Image[face][firstLevel + i];
-	    if (!t->intel.image[face][i].image) {
-	       fprintf(stderr, "no image %d %d\n", face, i);
-	       break;		/* can't happen */
-	    }
-	 
-	    t->intel.image[face][i].offset = 
-	       y * pitch + x * t->intel.texelBytes;
-	    t->intel.image[face][i].internalFormat = baseImage->_BaseFormat;
-
-	    d >>= 1;
-	    x += step_offsets[face][0] * d;
-	    y += step_offsets[face][1] * d;
-	 }
-      }
-      break;
-   }
-   default:
-      pitch = tObj->Image[0][firstLevel]->Width * t->intel.texelBytes;
-      pitch = (pitch + 3) & ~3;
-      t->intel.base.dirty_images[0] = ~0;
 
-      for ( total_height = i = 0 ; i < numLevels ; i++ ) {
-	 t->intel.image[0][i].image = tObj->Image[0][firstLevel + i];
-	 if (!t->intel.image[0][i].image) 
-	    break;
-	 
-	 t->intel.image[0][i].offset = total_height * pitch;
-	 t->intel.image[0][i].internalFormat = baseImage->_BaseFormat;
-	 if (t->intel.image[0][i].image->IsCompressed)
-	 {
-	   if (t->intel.image[0][i].image->Height > 4)
-	     total_height += t->intel.image[0][i].image->Height/4;
-	   else
-	     total_height += 1;
-	 }
-	 else
-	   total_height += MAX2(2, t->intel.image[0][i].image->Height);
-      }
-      break;
+/* The i915 (and related graphics cores) do not support GL_CLAMP.  The
+ * Intel drivers for "other operating systems" implement GL_CLAMP as
+ * GL_CLAMP_TO_EDGE, so the same is done here.
+ */
+static GLuint translate_wrap_mode( GLenum wrap )
+{
+   switch( wrap ) {
+   case GL_REPEAT: 
+      return TEXCOORDMODE_WRAP;
+   case GL_CLAMP:  
+   case GL_CLAMP_TO_EDGE: 
+      return TEXCOORDMODE_CLAMP; /* not really correct */
+   case GL_CLAMP_TO_BORDER: 
+      return TEXCOORDMODE_CLAMP_BORDER;
+   case GL_MIRRORED_REPEAT: 
+      return TEXCOORDMODE_MIRROR;
+   default: 
+      return TEXCOORDMODE_WRAP;
    }
-
-   t->intel.Pitch = pitch;
-   t->intel.base.totalSize = total_height*pitch;
-   t->intel.max_level = i-1;
-   t->Setup[I830_TEXREG_TM0S1] = 
-      (((tObj->Image[0][firstLevel]->Height - 1) << TM0S1_HEIGHT_SHIFT) |
-       ((tObj->Image[0][firstLevel]->Width - 1) << TM0S1_WIDTH_SHIFT) |
-       textureFormat);
-   t->Setup[I830_TEXREG_TM0S2] = 
-      (((pitch / 4) - 1) << TM0S2_PITCH_SHIFT) |
-      TM0S2_CUBE_FACE_ENA_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MAX_MIP_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIN_MIP_MASK;
-   t->Setup[I830_TEXREG_TM0S3] |= ((numLevels - 1)*4) << TM0S3_MIN_MIP_SHIFT;
-   t->intel.dirty = I830_UPLOAD_TEX_ALL;
-
-   return intelUploadTexImages( &i830->intel, &t->intel, 0 );
 }
 
 
-static void i830_import_tex_unit( i830ContextPtr i830, 
-			   i830TextureObjectPtr t,
-			   GLuint unit )
+/* Recalculate all state from scratch.  Perhaps not the most
+ * efficient, but this has gotten complex enough that we need
+ * something which is understandable and reliable.
+ */
+static GLboolean i830_update_tex_unit( struct intel_context *intel,
+				       GLuint unit, 
+				       GLuint ss3 )
 {
-   if(INTEL_DEBUG&DEBUG_TEXTURE)
-      fprintf(stderr, "%s unit(%d)\n", __FUNCTION__, unit);
-   
-   if (i830->intel.CurrentTexObj[unit]) 
-      i830->intel.CurrentTexObj[unit]->base.bound &= ~(1U << unit);
-
-   i830->intel.CurrentTexObj[unit] = (intelTextureObjectPtr)t;
-   t->intel.base.bound |= (1 << unit);
-
-   I830_STATECHANGE( i830, I830_UPLOAD_TEX(unit) );
-
-   i830->state.Tex[unit][I830_TEXREG_TM0LI] = (_3DSTATE_LOAD_STATE_IMMEDIATE_2 | 
-					       (LOAD_TEXTURE_MAP0 << unit) | 4);
-   i830->state.Tex[unit][I830_TEXREG_TM0S0] = (TM0S0_USE_FENCE |
-					       t->intel.TextureOffset);
+   GLcontext *ctx = &intel->ctx;
+   struct i830_context *i830 = i830_context(ctx);
+   struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
+   struct intel_texture_object *intelObj = intel_texture_object(tObj);
+   struct gl_texture_image *firstImage;
+   GLuint *state = i830->state.Tex[unit];
 
-   i830->state.Tex[unit][I830_TEXREG_TM0S1] = t->Setup[I830_TEXREG_TM0S1];
-   i830->state.Tex[unit][I830_TEXREG_TM0S2] = t->Setup[I830_TEXREG_TM0S2];
+   memset(state, 0, sizeof(state));
 
-   i830->state.Tex[unit][I830_TEXREG_TM0S3] &= TM0S3_LOD_BIAS_MASK;
-   i830->state.Tex[unit][I830_TEXREG_TM0S3] |= (t->Setup[I830_TEXREG_TM0S3] &
-						~TM0S3_LOD_BIAS_MASK);
+   if (!intel_finalize_mipmap_tree(intel, unit))
+      return GL_FALSE;   
 
-   i830->state.Tex[unit][I830_TEXREG_TM0S4] = t->Setup[I830_TEXREG_TM0S4];
-   i830->state.Tex[unit][I830_TEXREG_MCS] = (t->Setup[I830_TEXREG_MCS] & 
-					     ~MAP_UNIT_MASK);   
-   i830->state.Tex[unit][I830_TEXREG_CUBE] = t->Setup[I830_TEXREG_CUBE];
-   i830->state.Tex[unit][I830_TEXREG_MCS] |= MAP_UNIT(unit);
+   /* Get first image here, since intelObj->firstLevel will get set in
+    * the intel_finalize_mipmap_tree() call above.
+    */
+   firstImage = tObj->Image[0][intelObj->firstLevel];
 
-   t->intel.dirty &= ~I830_UPLOAD_TEX(unit);
-}
+   i830->state.tex_buffer[unit] = intelObj->mt->region->buffer;
+   i830->state.tex_offset[unit] = intel_miptree_image_offset(intelObj->mt, 0,
+							     intelObj->firstLevel); 
 
 
+   state[I830_TEXREG_TM0LI] = (_3DSTATE_LOAD_STATE_IMMEDIATE_2 | 
+			       (LOAD_TEXTURE_MAP0 << unit) | 4);
 
-static GLboolean enable_tex_common( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i830TextureObjectPtr t = (i830TextureObjectPtr)tObj->DriverData;
+/*    state[I830_TEXREG_TM0S0] = (TM0S0_USE_FENCE | */
+/* 			       t->intel.TextureOffset); */
 
-   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
 
-   /* Fallback if there's a texture border */
-   if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 ) {
-      fprintf(stderr, "Texture border\n");
-      return GL_FALSE;
-   }
+   state[I830_TEXREG_TM0S1] =  
+      (((firstImage->Height - 1) << TM0S1_HEIGHT_SHIFT) |
+       ((firstImage->Width - 1) << TM0S1_WIDTH_SHIFT) |
+       translate_texture_format( firstImage->TexFormat->MesaFormat));
 
-   /* Upload teximages (not pipelined)
-    */
-   if (t->intel.base.dirty_images[0]) {
-      if (!i830SetTexImages( i830, tObj )) {
-	 return GL_FALSE;
-      }
-   }
+   state[I830_TEXREG_TM0S2] = 
+      (((((intelObj->mt->pitch * intelObj->mt->cpp) / 4) - 1) << TM0S2_PITCH_SHIFT) |
+       TM0S2_CUBE_FACE_ENA_MASK);
 
-   /* Update state if this is a different texture object to last
-    * time.
-    */
-   if (i830->intel.CurrentTexObj[unit] != &t->intel || 
-       (t->intel.dirty & I830_UPLOAD_TEX(unit))) {
-      i830_import_tex_unit( i830, t, unit);
+   {
+      if (tObj->Target == GL_TEXTURE_CUBE_MAP)
+	 state[I830_TEXREG_CUBE] = (_3DSTATE_MAP_CUBE | MAP_UNIT(unit) |
+				    CUBE_NEGX_ENABLE | 
+				    CUBE_POSX_ENABLE | 
+				    CUBE_NEGY_ENABLE | 
+				    CUBE_POSY_ENABLE | 
+				    CUBE_NEGZ_ENABLE | 
+				    CUBE_POSZ_ENABLE);
+      else
+	 state[I830_TEXREG_CUBE] = (_3DSTATE_MAP_CUBE | MAP_UNIT(unit));
    }
 
-   I830_ACTIVESTATE(i830, I830_UPLOAD_TEX(unit), GL_TRUE);
-
-   return GL_TRUE;
-}
 
-static GLboolean enable_tex_rect( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   GLuint mcs = i830->state.Tex[unit][I830_TEXREG_MCS];
 
-   mcs &= ~TEXCOORDS_ARE_NORMAL;
-   mcs |= TEXCOORDS_ARE_IN_TEXELUNITS;
 
-   if ((mcs != i830->state.Tex[unit][I830_TEXREG_MCS])
-       || (0 != i830->state.Tex[unit][I830_TEXREG_CUBE])) {
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_MCS] = mcs;
-      i830->state.Tex[unit][I830_TEXREG_CUBE] = 0;
-   }
+   {
+      GLuint minFilt, mipFilt, magFilt;
+
+      switch (tObj->MinFilter) {
+      case GL_NEAREST:
+	 minFilt = FILTER_NEAREST;
+	 mipFilt = MIPFILTER_NONE;
+	 break;
+      case GL_LINEAR:
+	 minFilt = FILTER_LINEAR;
+	 mipFilt = MIPFILTER_NONE;
+	 break;
+      case GL_NEAREST_MIPMAP_NEAREST:
+	 minFilt = FILTER_NEAREST;
+	 mipFilt = MIPFILTER_NEAREST;
+	 break;
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 minFilt = FILTER_LINEAR;
+	 mipFilt = MIPFILTER_NEAREST;
+	 break;
+      case GL_NEAREST_MIPMAP_LINEAR:
+	 minFilt = FILTER_NEAREST;
+	 mipFilt = MIPFILTER_LINEAR;
+	 break;
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 minFilt = FILTER_LINEAR;
+	 mipFilt = MIPFILTER_LINEAR;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
 
-   return GL_TRUE;
-}
+      if ( tObj->MaxAnisotropy > 1.0 ) {
+	 minFilt = FILTER_ANISOTROPIC; 
+	 magFilt = FILTER_ANISOTROPIC;
+      }
+      else {
+	 switch (tObj->MagFilter) {
+	 case GL_NEAREST:
+	    magFilt = FILTER_NEAREST;
+	    break;
+	 case GL_LINEAR:
+	    magFilt = FILTER_LINEAR;
+	    break;
+	 default:
+	    return GL_FALSE;
+	 }  
+      }
 
+      state[I830_TEXREG_TM0S3] = i830->lodbias_tm0s3[unit];
 
-static GLboolean enable_tex_2d( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   GLuint mcs = i830->state.Tex[unit][I830_TEXREG_MCS];
+#if 0 
+      /* YUV conversion:
+       */
+      if (firstImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR ||
+	  firstImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR_REV)
+	 state[I830_TEXREG_TM0S3] |= SS2_COLORSPACE_CONVERSION;
+#endif
 
-   mcs &= ~TEXCOORDS_ARE_IN_TEXELUNITS;
-   mcs |= TEXCOORDS_ARE_NORMAL;
+      state[I830_TEXREG_TM0S3] |= ((intelObj->lastLevel - 
+				    intelObj->firstLevel)*4) << TM0S3_MIN_MIP_SHIFT;
 
-   if ((mcs != i830->state.Tex[unit][I830_TEXREG_MCS])
-       || (0 != i830->state.Tex[unit][I830_TEXREG_CUBE])) {
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_MCS] = mcs;
-      i830->state.Tex[unit][I830_TEXREG_CUBE] = 0;
+      state[I830_TEXREG_TM0S3] |= ((minFilt << TM0S3_MIN_FILTER_SHIFT) |
+				   (mipFilt << TM0S3_MIP_FILTER_SHIFT) |
+				   (magFilt << TM0S3_MAG_FILTER_SHIFT));
    }
 
-   return GL_TRUE;
-}
-
- 
-static GLboolean enable_tex_cube( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i830TextureObjectPtr t = (i830TextureObjectPtr)tObj->DriverData;
-   GLuint mcs = i830->state.Tex[unit][I830_TEXREG_MCS];
-   const GLuint cube = CUBE_NEGX_ENABLE | CUBE_POSX_ENABLE
-     | CUBE_NEGY_ENABLE | CUBE_POSY_ENABLE
-     | CUBE_NEGZ_ENABLE | CUBE_POSZ_ENABLE;
-   GLuint face;
-
-   mcs &= ~TEXCOORDS_ARE_IN_TEXELUNITS;
-   mcs |= TEXCOORDS_ARE_NORMAL;
-
-   if ((mcs != i830->state.Tex[unit][I830_TEXREG_MCS])
-       || (cube != i830->state.Tex[unit][I830_TEXREG_CUBE])) {
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_MCS] = mcs;
-      i830->state.Tex[unit][I830_TEXREG_CUBE] = cube;
-   }
+   {
+      GLenum ws = tObj->WrapS;
+      GLenum wt = tObj->WrapT;
+      
 
-   /* Upload teximages (not pipelined)
-    */
-   if ( t->intel.base.dirty_images[0] || t->intel.base.dirty_images[1] ||
-        t->intel.base.dirty_images[2] || t->intel.base.dirty_images[3] ||
-        t->intel.base.dirty_images[4] || t->intel.base.dirty_images[5] ) {
-      i830SetTexImages( i830, tObj );
+      /* 3D textures not available on i830
+       */
+      if (tObj->Target == GL_TEXTURE_3D)
+	 return GL_FALSE;
+	     
+      state[I830_TEXREG_MCS] = (_3DSTATE_MAP_COORD_SET_CMD |
+				MAP_UNIT(unit) |
+				ENABLE_TEXCOORD_PARAMS |
+				ss3 |
+				ENABLE_ADDR_V_CNTL |
+				TEXCOORD_ADDR_V_MODE(translate_wrap_mode(wt)) |
+				ENABLE_ADDR_U_CNTL |
+				TEXCOORD_ADDR_U_MODE(translate_wrap_mode(ws)));
    }
 
-   /* upload (per face) */
-   for (face = 0; face < 6; face++) {
-      if (t->intel.base.dirty_images[face]) {
-	 if (!intelUploadTexImages( &i830->intel, &t->intel, face )) {
-	    return GL_FALSE;
-	 }
-      }
-   }
 
+   state[I830_TEXREG_TM0S4] = INTEL_PACKCOLOR8888(tObj->_BorderChan[0],
+						  tObj->_BorderChan[1],
+						  tObj->_BorderChan[2],
+						  tObj->_BorderChan[3]);
+   
 
+   I830_ACTIVESTATE(i830, I830_UPLOAD_TEX(unit), GL_TRUE);
+   /* memcmp was already disabled, but definitely won't work as the
+    * region might now change and that wouldn't be detected:
+    */
+   I830_STATECHANGE( i830, I830_UPLOAD_TEX(unit) );
    return GL_TRUE;
 }
 
 
-static GLboolean disable_tex( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-
-   /* This is happening too often.  I need to conditionally send diffuse
-    * state to the card.  Perhaps a diffuse dirty flag of some kind.
-    * Will need to change this logic if more than 2 texture units are
-    * used.  We need to only do this up to the last unit enabled, or unit
-    * one if nothing is enabled.
-    */
-
-   if ( i830->intel.CurrentTexObj[unit] != NULL ) {
-      /* The old texture is no longer bound to this texture unit.
-       * Mark it as such.
-       */
-
-      i830->intel.CurrentTexObj[unit]->base.bound &= ~(1U << 0);
-      i830->intel.CurrentTexObj[unit] = NULL;
-   }
 
-   return GL_TRUE;
-}
 
-static GLboolean i830UpdateTexUnit( GLcontext *ctx, GLuint unit )
+void i830UpdateTextureState( struct intel_context *intel )
 {
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-
-   if (texUnit->_ReallyEnabled &&
-       INTEL_CONTEXT(ctx)->intelScreen->tex.size < 2048 * 1024)
-      return GL_FALSE;
-
-   switch(texUnit->_ReallyEnabled) {
-   case TEXTURE_1D_BIT:
-   case TEXTURE_2D_BIT:
-      return (enable_tex_common( ctx, unit ) &&
-	      enable_tex_2d( ctx, unit ));
-   case TEXTURE_RECT_BIT:
-      return (enable_tex_common( ctx, unit ) &&
-	      enable_tex_rect( ctx, unit ));
-   case TEXTURE_CUBE_BIT:
-      return (enable_tex_common( ctx, unit ) &&
-	      enable_tex_cube( ctx, unit ));
-   case 0:
-      return disable_tex( ctx, unit );
-   default:
-      return GL_FALSE;
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   GLboolean ok = GL_TRUE;
+   GLuint i;
+
+   for (i = 0 ; i < I830_TEX_UNITS && ok ; i++) {
+      switch (intel->ctx.Texture.Unit[i]._ReallyEnabled) {
+      case TEXTURE_1D_BIT:
+      case TEXTURE_2D_BIT:
+      case TEXTURE_CUBE_BIT:
+	 ok = i830_update_tex_unit( intel, i, TEXCOORDS_ARE_NORMAL );
+	 break;
+      case TEXTURE_RECT_BIT:
+	 ok = i830_update_tex_unit( intel, i, TEXCOORDS_ARE_IN_TEXELUNITS );
+	 break;
+      case 0: 
+	 if (i830->state.active & I830_UPLOAD_TEX(i)) 
+	    I830_ACTIVESTATE(i830, I830_UPLOAD_TEX(i), GL_FALSE);
+	 break;
+      case TEXTURE_3D_BIT:
+      default:
+	 ok = GL_FALSE;
+	 break;
+      }
    }
-}
 
+   FALLBACK( intel, I830_FALLBACK_TEXTURE, !ok );
 
-void i830UpdateTextureState( intelContextPtr intel )
-{
-   i830ContextPtr i830 = I830_CONTEXT(intel);
-   GLcontext *ctx = &intel->ctx;
-   GLboolean ok;
+   if (ok)
+      i830EmitTextureBlend( i830 );
+}
 
-   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
 
-   I830_ACTIVESTATE(i830, I830_UPLOAD_TEX_ALL, GL_FALSE);
 
-   ok = (i830UpdateTexUnit( ctx, 0 ) &&
-	 i830UpdateTexUnit( ctx, 1 ) &&
-	 i830UpdateTexUnit( ctx, 2 ) &&
-	 i830UpdateTexUnit( ctx, 3 ));
 
-   FALLBACK( intel, I830_FALLBACK_TEXTURE, !ok );
 
-   if (ok)
-      i830EmitTextureBlend( i830 );
-}
 
 
 
diff --git a/src/mesa/drivers/dri/i915/i830_vtbl.c b/src/mesa/drivers/dri/i915/i830_vtbl.c
index 9e71b11109..aafe324d6f 100644
--- a/src/mesa/drivers/dri/i915/i830_vtbl.c
+++ b/src/mesa/drivers/dri/i915/i830_vtbl.c
@@ -30,11 +30,12 @@
 #include "i830_reg.h"
 
 #include "intel_batchbuffer.h"
+#include "intel_regions.h"
 
 #include "tnl/t_context.h"
 #include "tnl/t_vertex.h"
 
-static GLboolean i830_check_vertex_size( intelContextPtr intel,
+static GLboolean i830_check_vertex_size( struct intel_context *intel,
 					 GLuint expected );
 
 #define SZ_TO_HW(sz)  ((sz-2)&0x3)
@@ -59,10 +60,10 @@ do {									\
 #define VRTX_TEX_SET_FMT(n, x)          ((x)<<((n)*2))
 #define TEXBIND_SET(n, x) 		((x)<<((n)*4))
 
-static void i830_render_start( intelContextPtr intel )
+static void i830_render_start( struct intel_context *intel )
 {
    GLcontext *ctx = &intel->ctx;
-   i830ContextPtr i830 = I830_CONTEXT(intel);
+   struct i830_context *i830 = i830_context(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    DECLARE_RENDERINPUTS(index_bitset);
@@ -166,6 +167,7 @@ static void i830_render_start( intelContextPtr intel )
        v2 != i830->state.Ctx[I830_CTXREG_VF2] ||
        mcsb1 != i830->state.Ctx[I830_CTXREG_MCSB1] ||
        !RENDERINPUTS_EQUAL( index_bitset, i830->last_index_bitset )) {
+      int k;
     
       I830_STATECHANGE( i830, I830_UPLOAD_CTX );
 
@@ -185,14 +187,15 @@ static void i830_render_start( intelContextPtr intel )
       i830->state.Ctx[I830_CTXREG_MCSB1] = mcsb1;
       RENDERINPUTS_COPY( i830->last_index_bitset, index_bitset );
 
-      assert(i830_check_vertex_size( intel, intel->vertex_size ));
+      k = i830_check_vertex_size( intel, intel->vertex_size );
+      assert(k);
    }
 }
 
-static void i830_reduced_primitive_state( intelContextPtr intel,
+static void i830_reduced_primitive_state( struct intel_context *intel,
 					  GLenum rprim )
 {
-    i830ContextPtr i830 = I830_CONTEXT(intel);
+    struct i830_context *i830 = i830_context(&intel->ctx);
     GLuint st1 = i830->state.Stipple[I830_STPREG_ST1];
 
     st1 &= ~ST1_ENABLE;
@@ -220,10 +223,10 @@ static void i830_reduced_primitive_state( intelContextPtr intel,
 /* Pull apart the vertex format registers and figure out how large a
  * vertex is supposed to be. 
  */
-static GLboolean i830_check_vertex_size( intelContextPtr intel,
+static GLboolean i830_check_vertex_size( struct intel_context *intel,
 					 GLuint expected )
 {
-   i830ContextPtr i830 = I830_CONTEXT(intel);
+   struct i830_context *i830 = i830_context(&intel->ctx);
    int vft0 = i830->current->Ctx[I830_CTXREG_VF];
    int vft1 = i830->current->Ctx[I830_CTXREG_VF2];
    int nrtex = (vft0 & VFT0_TEX_COUNT_MASK) >> VFT0_TEX_COUNT_SHIFT;
@@ -260,16 +263,11 @@ static GLboolean i830_check_vertex_size( intelContextPtr intel,
    return sz == expected;
 }
 
-static void i830_emit_invarient_state( intelContextPtr intel )
+static void i830_emit_invarient_state( struct intel_context *intel )
 {
    BATCH_LOCALS;
 
-   BEGIN_BATCH( 200 );
-
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(0));
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(1));
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(2));
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(3));
+   BEGIN_BATCH(40, 0);
 
    OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
    OUT_BATCH(0);
@@ -333,13 +331,6 @@ static void i830_emit_invarient_state( intelContextPtr intel )
 	     TRI_FAN_PROVOKE_VRTX(2) | 
 	     TRI_STRIP_PROVOKE_VRTX(2));
 
-   OUT_BATCH(_3DSTATE_SCISSOR_ENABLE_CMD | 
-	     DISABLE_SCISSOR_RECT);
-
-   OUT_BATCH(_3DSTATE_SCISSOR_RECT_0_CMD);
-   OUT_BATCH(0);
-   OUT_BATCH(0);
-
    OUT_BATCH(_3DSTATE_VERTEX_TRANSFORM);
    OUT_BATCH(DISABLE_VIEWPORT_TRANSFORM | DISABLE_PERSPECTIVE_DIVIDE);
 
@@ -358,11 +349,13 @@ static void i830_emit_invarient_state( intelContextPtr intel )
 #define emit( intel, state, size )			\
 do {							\
    int k;						\
-   BEGIN_BATCH( size / sizeof(GLuint));			\
-   for (k = 0 ; k < size / sizeof(GLuint) ; k++)	\
+   BEGIN_BATCH(size / sizeof(GLuint), 0);		\
+   for (k = 0 ; k < size / sizeof(GLuint) ; k++) {	\
+      if (0) _mesa_printf("  0x%08x\n", state[k]);		\
       OUT_BATCH(state[k]);				\
+   }							\
    ADVANCE_BATCH();					\
-} while (0);
+} while (0)
 
 static GLuint get_state_size( struct i830_hw_state *state )
 {
@@ -370,6 +363,9 @@ static GLuint get_state_size( struct i830_hw_state *state )
    GLuint sz = 0;
    GLuint i;
 
+   if (dirty & I830_UPLOAD_INVARIENT) 
+      sz += 40 * sizeof(int);
+
    if (dirty & I830_UPLOAD_CTX) 
       sz += sizeof(state->Ctx);
 
@@ -393,81 +389,133 @@ static GLuint get_state_size( struct i830_hw_state *state )
 
 /* Push the state into the sarea and/or texture memory.
  */
-static void i830_emit_state( intelContextPtr intel )
+static void i830_emit_state( struct intel_context *intel )
 {
-   i830ContextPtr i830 = I830_CONTEXT(intel);
+   struct i830_context *i830 = i830_context(&intel->ctx);
    struct i830_hw_state *state = i830->current;
    int i;
    GLuint dirty = state->active & ~state->emitted;
-   GLuint counter = intel->batch.counter;
    BATCH_LOCALS;
 
-   if (intel->batch.space < get_state_size(state)) {
-      intelFlushBatch(intel, GL_TRUE);
-      dirty = state->active & ~state->emitted;
-      counter = intel->batch.counter;
+   /* We don't hold the lock at this point, so want to make sure that
+    * there won't be a buffer wrap.  
+    *
+    * It might be better to talk about explicit places where
+    * scheduling is allowed, rather than assume that it is whenever a
+    * batchbuffer fills up.
+    */
+   intel_batchbuffer_require_space(intel->batch, 
+				   get_state_size(state),
+				   0);
+
+   if (dirty & I830_UPLOAD_INVARIENT) {
+      if (INTEL_DEBUG & DEBUG_STATE) 
+	 fprintf(stderr, "I830_UPLOAD_INVARIENT:\n"); 
+      i830_emit_invarient_state( intel );
    }
 
    if (dirty & I830_UPLOAD_CTX) {
-      if (VERBOSE) fprintf(stderr, "I830_UPLOAD_CTX:\n"); 
+      if (INTEL_DEBUG & DEBUG_STATE) 
+	 fprintf(stderr, "I830_UPLOAD_CTX:\n"); 
       emit( i830, state->Ctx, sizeof(state->Ctx) );
+
    }
 
    if (dirty & I830_UPLOAD_BUFFERS) {
-      if (VERBOSE) fprintf(stderr, "I830_UPLOAD_BUFFERS:\n"); 
-      emit( i830, state->Buffer, sizeof(state->Buffer) );
+      if (INTEL_DEBUG & DEBUG_STATE) 
+	 fprintf(stderr, "I830_UPLOAD_BUFFERS:\n"); 
+      BEGIN_BATCH(I830_DEST_SETUP_SIZE+2, 0);
+      OUT_BATCH(state->Buffer[I830_DESTREG_CBUFADDR0]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_CBUFADDR1]);
+      OUT_RELOC(state->draw_region->buffer, DRM_MM_TT|DRM_MM_WRITE, 0);
+
+      if (state->depth_region) {
+	 OUT_BATCH(state->Buffer[I830_DESTREG_DBUFADDR0]);
+	 OUT_BATCH(state->Buffer[I830_DESTREG_DBUFADDR1]);
+	 OUT_RELOC(state->depth_region->buffer, DRM_MM_TT |DRM_MM_WRITE, 0);
+      }
+
+      OUT_BATCH(state->Buffer[I830_DESTREG_DV0]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_DV1]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SENABLE]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SR0]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SR1]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SR2]);
+      ADVANCE_BATCH();
    }
 
    if (dirty & I830_UPLOAD_STIPPLE) {
-      if (VERBOSE) fprintf(stderr, "I830_UPLOAD_STIPPLE:\n"); 
+      if (INTEL_DEBUG & DEBUG_STATE) 
+	 fprintf(stderr, "I830_UPLOAD_STIPPLE:\n"); 
       emit( i830, state->Stipple, sizeof(state->Stipple) );
    }
 
    for (i = 0; i < I830_TEX_UNITS; i++) {
       if ((dirty & I830_UPLOAD_TEX(i))) { 
- 	 if (VERBOSE) fprintf(stderr, "I830_UPLOAD_TEX(%d):\n", i); 
-	 emit( i830, state->Tex[i], sizeof(state->Tex[i])); 
+ 	 if (INTEL_DEBUG & DEBUG_STATE)
+	    fprintf(stderr, "I830_UPLOAD_TEX(%d):\n", i); 
+
+	 BEGIN_BATCH(I830_TEX_SETUP_SIZE+1, 0);
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_TM0LI]);
+
+	 if (state->tex_buffer[i]) {
+	    OUT_RELOC(state->tex_buffer[i],
+		      DRM_MM_TT|DRM_MM_READ,
+		      state->tex_offset[i] | TM0S0_USE_FENCE);
+	 }
+	 else {
+	    assert(i == 0);
+	    assert(state == &i830->meta);
+	    OUT_BATCH(0);
+	 }
+
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S1]);
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S2]);
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S3]);
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S4]);
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_MCS]);
+	 OUT_BATCH(state->Tex[i][I830_TEXREG_CUBE]);
       } 
 
       if (dirty & I830_UPLOAD_TEXBLEND(i)) {
-	 if (VERBOSE) fprintf(stderr, "I830_UPLOAD_TEXBLEND(%d):\n", i); 
+	 if (INTEL_DEBUG & DEBUG_STATE) 
+	    fprintf(stderr, "I830_UPLOAD_TEXBLEND(%d): %d words\n", i,
+		    state->TexBlendWordsUsed[i]); 
 	 emit( i830, state->TexBlend[i], 
 	       state->TexBlendWordsUsed[i] * 4 );
       }
    }
 
    state->emitted |= dirty;
-   intel->batch.last_emit_state = counter;
-   assert(counter == intel->batch.counter);
 }
 
-static void i830_destroy_context( intelContextPtr intel )
+static void i830_destroy_context( struct intel_context *intel )
 {
    _tnl_free_vertices(&intel->ctx);
 }
 
-static void
-i830_set_color_region(intelContextPtr intel, const intelRegion *region)
+static void i830_set_draw_region( struct intel_context *intel, 
+				  struct intel_region *draw_region,
+				  struct intel_region *depth_region)
 {
-   i830ContextPtr i830 = I830_CONTEXT(intel);
-   I830_STATECHANGE( i830, I830_UPLOAD_BUFFERS );
-   i830->state.Buffer[I830_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i830->state.Buffer[I830_DESTREG_CBUFADDR2] = region->offset;
-}
+   struct i830_context *i830 = i830_context(&intel->ctx);
 
+   intel_region_release(intel, &i830->state.draw_region);
+   intel_region_release(intel, &i830->state.depth_region);
+   intel_region_reference(&i830->state.draw_region, draw_region);
+   intel_region_reference(&i830->state.depth_region, depth_region);
+
+   /* XXX FBO: Need code from i915_set_draw_region() */
 
-static void
-i830_set_z_region(intelContextPtr intel, const intelRegion *region)
-{
-   i830ContextPtr i830 = I830_CONTEXT(intel);
    I830_STATECHANGE( i830, I830_UPLOAD_BUFFERS );
+   I830_STATECHANGE( i830, I830_UPLOAD_BUFFERS );
+   i830->state.Buffer[I830_DESTREG_CBUFADDR1] =
+      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(draw_region->pitch) | BUF_3D_USE_FENCE);
    i830->state.Buffer[I830_DESTREG_DBUFADDR1] =
-      (BUF_3D_ID_DEPTH | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i830->state.Buffer[I830_DESTREG_DBUFADDR2] = region->offset;
+      (BUF_3D_ID_DEPTH | BUF_3D_PITCH(depth_region->pitch) | BUF_3D_USE_FENCE);
 }
 
-
+#if 0
 static void
 i830_update_color_z_regions(intelContextPtr intel,
                             const intelRegion *colorRegion,
@@ -483,45 +531,36 @@ i830_update_color_z_regions(intelContextPtr intel,
       (BUF_3D_ID_DEPTH | BUF_3D_PITCH(depthRegion->pitch) | BUF_3D_USE_FENCE);
    i830->state.Buffer[I830_DESTREG_DBUFADDR2] = depthRegion->offset;
 }
+#endif
 
 
 /* This isn't really handled at the moment.
  */
-static void i830_lost_hardware( intelContextPtr intel )
+static void i830_lost_hardware( struct intel_context *intel )
 {
-   I830_CONTEXT(intel)->state.emitted = 0;
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   i830->state.emitted = 0;
 }
 
 
 
-static void i830_emit_flush( intelContextPtr intel )
+static GLuint i830_flush_cmd( void )
 {
-   BATCH_LOCALS;
-
-   BEGIN_BATCH(2);
-   OUT_BATCH( MI_FLUSH | FLUSH_MAP_CACHE ); 
-   OUT_BATCH( 0 );
-   ADVANCE_BATCH();
+   return MI_FLUSH | FLUSH_MAP_CACHE; 
 }
 
 
 
 
-void i830InitVtbl( i830ContextPtr i830 )
+void i830InitVtbl( struct i830_context *i830 )
 {
-   i830->intel.vtbl.alloc_tex_obj = i830AllocTexObj;
    i830->intel.vtbl.check_vertex_size = i830_check_vertex_size;
-   i830->intel.vtbl.clear_with_tris = i830ClearWithTris;
-   i830->intel.vtbl.rotate_window = i830RotateWindow;
    i830->intel.vtbl.destroy = i830_destroy_context;
-   i830->intel.vtbl.emit_invarient_state = i830_emit_invarient_state;
    i830->intel.vtbl.emit_state = i830_emit_state;
    i830->intel.vtbl.lost_hardware = i830_lost_hardware;
    i830->intel.vtbl.reduced_primitive_state = i830_reduced_primitive_state;
-   i830->intel.vtbl.set_color_region = i830_set_color_region;
-   i830->intel.vtbl.set_z_region = i830_set_z_region;
-   i830->intel.vtbl.update_color_z_regions = i830_update_color_z_regions;
+   i830->intel.vtbl.set_draw_region = i830_set_draw_region;
    i830->intel.vtbl.update_texture_state = i830UpdateTextureState;
-   i830->intel.vtbl.emit_flush = i830_emit_flush;
+   i830->intel.vtbl.flush_cmd = i830_flush_cmd;
    i830->intel.vtbl.render_start = i830_render_start;
 }
diff --git a/src/mesa/drivers/dri/i915/i915_context.c b/src/mesa/drivers/dri/i915/i915_context.c
index 783bbc2ab8..99c4b9c16a 100644
--- a/src/mesa/drivers/dri/i915/i915_context.c
+++ b/src/mesa/drivers/dri/i915/i915_context.c
@@ -41,6 +41,10 @@
 #include "utils.h"
 #include "i915_reg.h"
 
+#include "intel_bufmgr.h"
+#include "intel_regions.h"
+#include "intel_batchbuffer.h"
+
 /***************************************
  * Mesa's Driver Functions
  ***************************************/
@@ -65,7 +69,7 @@ static void i915InvalidateState( GLcontext *ctx, GLuint new_state )
    _ac_InvalidateState( ctx, new_state );
    _tnl_InvalidateState( ctx, new_state );
    _tnl_invalidate_vertex_state( ctx, new_state );
-   INTEL_CONTEXT(ctx)->NewGLState |= new_state;
+   intel_context(ctx)->NewGLState |= new_state;
 
    /* Todo: gather state values under which tracked parameters become
     * invalidated, add callbacks for things like
@@ -99,14 +103,16 @@ GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
 			    void *sharedContextPrivate)
 {
    struct dd_function_table functions;
-   i915ContextPtr i915 = (i915ContextPtr) CALLOC_STRUCT(i915_context);
-   intelContextPtr intel = &i915->intel;
+   struct i915_context *i915 = (struct i915_context *) CALLOC_STRUCT(i915_context);
+   struct intel_context *intel = &i915->intel;
    GLcontext *ctx = &intel->ctx;
-   GLuint i;
 
    if (!i915) return GL_FALSE;
 
+   _mesa_printf( "\ntexmem branch (i915, drop3)\n\n");
+   
    i915InitVtbl( i915 );
+   i915InitMetaFuncs( i915 );
 
    i915InitDriverFunctions( &functions );
 
@@ -120,49 +126,28 @@ GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
    ctx->Const.MaxTextureImageUnits = I915_TEX_UNITS;
    ctx->Const.MaxTextureCoordUnits = I915_TEX_UNITS;
 
-   intel->nr_heaps = 1;
-   intel->texture_heaps[0] = 
-      driCreateTextureHeap( 0, intel,
-			    intel->intelScreen->tex.size,
-			    12,
-			    I830_NR_TEX_REGIONS,
-			    intel->sarea->texList,
-			    (unsigned *) & intel->sarea->texAge,
-			    & intel->swapped,
-			    sizeof( struct i915_texture_object ),
-			    (destroy_texture_object_t *)intelDestroyTexObj );
-
-   /* FIXME: driCalculateMaxTextureLevels assumes that mipmaps are
-    * tightly packed, but they're not in Intel graphics
-    * hardware.
+
+   /* Advertise the full hardware capabilities.  The new memory
+    * manager should cope much better with overload situations:
     */
+   ctx->Const.MaxTextureLevels = 12;
+   ctx->Const.Max3DTextureLevels = 9;
+   ctx->Const.MaxCubeTextureLevels = 12;
+   ctx->Const.MaxTextureRectSize = (1<<11);
    ctx->Const.MaxTextureUnits = I915_TEX_UNITS;
-   i = driQueryOptioni( &intel->intelScreen->optionCache, "allow_large_textures");
-   driCalculateMaxTextureLevels( intel->texture_heaps,
-				 intel->nr_heaps,
-				 &intel->ctx.Const,
-				 4,
-				 11, /* max 2D texture size is 2048x2048 */
-				 8,  /* 3D texture */
-				 11, /* cube texture. */
-				 11, /* rect texture */
-				 12,
-				 GL_FALSE,
-				 i );
 
    /* GL_ARB_fragment_program limits - don't think Mesa actually
     * validates programs against these, and in any case one ARB
     * instruction can translate to more than one HW instruction, so
     * we'll still have to check and fallback each time.
-    */
-   
+    */   
    ctx->Const.FragmentProgram.MaxNativeTemps = I915_MAX_TEMPORARY;
    ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* 8 tex, 2 color, fog */
    ctx->Const.FragmentProgram.MaxNativeParameters = I915_MAX_CONSTANT;
    ctx->Const.FragmentProgram.MaxNativeAluInstructions = I915_MAX_ALU_INSN;
    ctx->Const.FragmentProgram.MaxNativeTexInstructions = I915_MAX_TEX_INSN;
    ctx->Const.FragmentProgram.MaxNativeInstructions = (I915_MAX_ALU_INSN + 
-						I915_MAX_TEX_INSN);
+						       I915_MAX_TEX_INSN);
    ctx->Const.FragmentProgram.MaxNativeTexIndirections = I915_MAX_TEX_INDIRECT;
    ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; /* I don't think we have one */
 
diff --git a/src/mesa/drivers/dri/i915/i915_context.h b/src/mesa/drivers/dri/i915/i915_context.h
index 761a475714..4f10c2309a 100644
--- a/src/mesa/drivers/dri/i915/i915_context.h
+++ b/src/mesa/drivers/dri/i915/i915_context.h
@@ -45,6 +45,8 @@
 #define I915_UPLOAD_PROGRAM          0x8
 #define I915_UPLOAD_CONSTANTS        0x10
 #define I915_UPLOAD_FOG              0x20
+#define I915_UPLOAD_INVARIENT        0x40
+#define I915_UPLOAD_DEFAULTS         0x80
 #define I915_UPLOAD_TEX(i)           (0x00010000<<(i))
 #define I915_UPLOAD_TEX_ALL          (0x00ff0000)
 #define I915_UPLOAD_TEX_0_SHIFT      16
@@ -54,10 +56,8 @@
  */
 #define I915_DESTREG_CBUFADDR0 0
 #define I915_DESTREG_CBUFADDR1 1
-#define I915_DESTREG_CBUFADDR2 2
 #define I915_DESTREG_DBUFADDR0 3
 #define I915_DESTREG_DBUFADDR1 4
-#define I915_DESTREG_DBUFADDR2 5
 #define I915_DESTREG_DV0 6
 #define I915_DESTREG_DV1 7
 #define I915_DESTREG_SENABLE 8
@@ -88,7 +88,6 @@
 #define I915_STPREG_ST1        1
 #define I915_STP_SETUP_SIZE    2
 
-#define I915_TEXREG_MS2        0
 #define I915_TEXREG_MS3        1
 #define I915_TEXREG_MS4        2
 #define I915_TEXREG_SS2        3
@@ -96,6 +95,15 @@
 #define I915_TEXREG_SS4        5
 #define I915_TEX_SETUP_SIZE    6
 
+#define I915_DEFREG_C0    0
+#define I915_DEFREG_C1    1
+#define I915_DEFREG_S0    2
+#define I915_DEFREG_S1    3
+#define I915_DEFREG_Z0    4
+#define I915_DEFREG_Z1    5
+#define I915_DEF_SETUP_SIZE    6
+
+
 #define I915_MAX_CONSTANT      32
 #define I915_CONSTANT_SIZE     (2+(4*I915_MAX_CONSTANT))
 
@@ -165,8 +173,6 @@ struct i915_fragment_program {
    GLuint nr_params;
       
 
-
-
    /* Helpers for i915_texprog.c:
     */
    GLuint src_texture;		/* Reg containing sampled texture color,
@@ -187,13 +193,6 @@ struct i915_fragment_program {
 
 
 
-struct i915_texture_object
-{
-   struct intel_texture_object intel;
-   GLenum lastTarget;
-   GLboolean refs_border_color;
-   GLuint Setup[I915_TEX_SETUP_SIZE];
-};
 
 #define I915_TEX_UNITS 8
 
@@ -203,11 +202,27 @@ struct i915_hw_state {
    GLuint Buffer[I915_DEST_SETUP_SIZE];
    GLuint Stipple[I915_STP_SETUP_SIZE];
    GLuint Fog[I915_FOG_SETUP_SIZE];
+   GLuint Defaults[I915_DEF_SETUP_SIZE];
    GLuint Tex[I915_TEX_UNITS][I915_TEX_SETUP_SIZE];
    GLuint Constant[I915_CONSTANT_SIZE];
    GLuint ConstantSize;
    GLuint Program[I915_PROGRAM_SIZE];
    GLuint ProgramSize;
+   
+   /* Region pointers for relocation: 
+    */
+   struct intel_region *draw_region;
+   struct intel_region *depth_region;
+/*    struct intel_region *tex_region[I915_TEX_UNITS]; */
+
+   /* Regions aren't actually that appropriate here as the memory may
+    * be from a PBO or FBO.  Just use the buffer id.  Will have to do
+    * this for draw and depth for FBO's...
+    */
+   GLuint tex_buffer[I915_TEX_UNITS];
+   GLuint tex_offset[I915_TEX_UNITS];
+   
+
    GLuint active;		/* I915_UPLOAD_* */
    GLuint emitted;		/* I915_UPLOAD_* */
 };
@@ -222,6 +237,8 @@ struct i915_context
 
    GLuint last_ReallyEnabled;
    GLuint vertex_fog;
+   GLuint lodbias_ss2[MAX_TEXTURE_UNITS];
+
 
    struct i915_fragment_program tex_program;
    struct i915_fragment_program *current_program;
@@ -230,24 +247,14 @@ struct i915_context
 };
 
 
-typedef struct i915_context *i915ContextPtr;
-typedef struct i915_texture_object *i915TextureObjectPtr;
-
-#define I915_CONTEXT(ctx)	((i915ContextPtr)(ctx))
-
-
-
 #define I915_STATECHANGE(i915, flag)					\
 do {									\
-   if (0) fprintf(stderr, "I915_STATECHANGE %x in %s\n", flag, __FUNCTION__);	\
    INTEL_FIREVERTICES( &(i915)->intel );					\
    (i915)->state.emitted &= ~(flag);					\
 } while (0)
 
 #define I915_ACTIVESTATE(i915, flag, mode)			\
 do {								\
-   if (0) fprintf(stderr, "I915_ACTIVESTATE %x %d in %s\n",	\
-		  flag, mode, __FUNCTION__);			\
    INTEL_FIREVERTICES( &(i915)->intel );				\
    if (mode)							\
       (i915)->state.active |= (flag);				\
@@ -259,7 +266,13 @@ do {								\
 /*======================================================================
  * i915_vtbl.c
  */
-extern void i915InitVtbl( i915ContextPtr i915 );
+extern void i915InitVtbl( struct i915_context *i915 );
+
+extern void
+i915_state_draw_region(struct intel_context *intel, 
+                       struct i915_hw_state *state,
+                       struct intel_region *color_region,
+                       struct intel_region *depth_region);
 
 
 
@@ -296,7 +309,7 @@ extern GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
 /*======================================================================
  * i915_texprog.c
  */
-extern void i915ValidateTextureProgram( i915ContextPtr i915 );
+extern void i915ValidateTextureProgram( struct i915_context *i915 );
 
 
 /*======================================================================
@@ -310,48 +323,43 @@ extern void i915_print_ureg( const char *msg, GLuint ureg );
  * i915_state.c
  */
 extern void i915InitStateFunctions( struct dd_function_table *functions );
-extern void i915InitState( i915ContextPtr i915 );
-extern void i915_update_fog(GLcontext *ctxx);
+extern void i915InitState( struct i915_context *i915 );
+extern void i915_update_fog( GLcontext *ctx );
 
 
 /*======================================================================
  * i915_tex.c
  */
-extern void i915UpdateTextureState( intelContextPtr intel );
+extern void i915UpdateTextureState( struct intel_context *intel );
 extern void i915InitTextureFuncs( struct dd_function_table *functions );
-extern intelTextureObjectPtr i915AllocTexObj( struct gl_texture_object *texObj );
 
 /*======================================================================
  * i915_metaops.c
  */
-extern GLboolean
-i915TryTextureReadPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *pack,
-			  GLvoid *pixels );
-
-extern GLboolean
-i915TryTextureDrawPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *unpack,
-			  const GLvoid *pixels );
+void i915InitMetaFuncs( struct i915_context *i915 );
 
-extern void 
-i915ClearWithTris( intelContextPtr intel, GLbitfield mask,
-		   GLboolean all, GLint cx, GLint cy, GLint cw, GLint ch);
-
-
-extern void
-i915RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                 GLuint srcBuf);
 
 /*======================================================================
  * i915_fragprog.c
  */
-extern void i915ValidateFragmentProgram( i915ContextPtr i915 );
+extern void i915ValidateFragmentProgram( struct i915_context *i915 );
 extern void i915InitFragProgFuncs( struct dd_function_table *functions );
+
+/*======================================================================
+ * Inline conversion functions.  These are better-typed than the
+ * macros used previously:
+ */
+static INLINE struct i915_context *
+i915_context( GLcontext *ctx )
+{
+   return (struct i915_context *)ctx;
+}
+
+
+
+#define I915_CONTEXT(ctx)	i915_context(ctx)
+
+
 	
 #endif
 
diff --git a/src/mesa/drivers/dri/i915/i915_fragprog.c b/src/mesa/drivers/dri/i915/i915_fragprog.c
index b0cc59c306..9266336a1a 100644
--- a/src/mesa/drivers/dri/i915/i915_fragprog.c
+++ b/src/mesa/drivers/dri/i915/i915_fragprog.c
@@ -806,7 +806,7 @@ static void check_wpos( struct i915_fragment_program *p )
 
 static void translate_program( struct i915_fragment_program *p )
 {
-   i915ContextPtr i915 = I915_CONTEXT(p->ctx);
+   struct i915_context *i915 = I915_CONTEXT(p->ctx);
    
    i915_init_program( i915, p );
    check_wpos( p ); 
@@ -840,7 +840,7 @@ static void i915BindProgram( GLcontext *ctx,
 			    struct gl_program *prog )
 {
    if (target == GL_FRAGMENT_PROGRAM_ARB) {
-      i915ContextPtr i915 = I915_CONTEXT(ctx);
+      struct i915_context *i915 = I915_CONTEXT(ctx);
       struct i915_fragment_program *p = (struct i915_fragment_program *)prog;
 
       if (i915->current_program == p) 
@@ -896,7 +896,7 @@ static void i915DeleteProgram( GLcontext *ctx,
 			      struct gl_program *prog )
 {
    if (prog->Target == GL_FRAGMENT_PROGRAM_ARB) {
-      i915ContextPtr i915 = I915_CONTEXT(ctx);
+      struct i915_context *i915 = I915_CONTEXT(ctx);
       struct i915_fragment_program *p = (struct i915_fragment_program *)prog;
       
       if (i915->current_program == p) 
@@ -940,10 +940,10 @@ static void i915ProgramStringNotify( GLcontext *ctx,
 }
 
 
-void i915ValidateFragmentProgram( i915ContextPtr i915 )
+void i915ValidateFragmentProgram( struct i915_context *i915 )
 {
    GLcontext *ctx = &i915->intel.ctx;
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
 
@@ -1028,6 +1028,7 @@ void i915ValidateFragmentProgram( i915ContextPtr i915 )
 
    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
+      int k;
     
       I915_STATECHANGE( i915, I915_UPLOAD_CTX );
 
@@ -1044,7 +1045,8 @@ void i915ValidateFragmentProgram( i915ContextPtr i915 )
       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
 
-      assert(intel->vtbl.check_vertex_size( intel, intel->vertex_size ));
+      k = intel->vtbl.check_vertex_size( intel, intel->vertex_size );
+      assert(k);
    }
 
    if (!p->params_uptodate) 
diff --git a/src/mesa/drivers/dri/i915/i915_metaops.c b/src/mesa/drivers/dri/i915/i915_metaops.c
index f7b8e5415e..280529903f 100644
--- a/src/mesa/drivers/dri/i915/i915_metaops.c
+++ b/src/mesa/drivers/dri/i915/i915_metaops.c
@@ -34,45 +34,46 @@
 #include "intel_screen.h"
 #include "intel_batchbuffer.h"
 #include "intel_ioctl.h"
+#include "intel_regions.h"
 #include "intel_rotate.h"
 
 #include "i915_context.h"
 #include "i915_reg.h"
 
-/* A large amount of state doesn't need to be uploaded.
+/* We touch almost everything:
  */
-#define ACTIVE (I915_UPLOAD_PROGRAM | 		\
-		I915_UPLOAD_STIPPLE |		\
+#define ACTIVE (I915_UPLOAD_INVARIENT | 	\
 		I915_UPLOAD_CTX |		\
 		I915_UPLOAD_BUFFERS |		\
+		I915_UPLOAD_STIPPLE |		\
+                I915_UPLOAD_PROGRAM | 		\
+                I915_UPLOAD_FOG | 		\
 		I915_UPLOAD_TEX(0))		
 
-#define SET_STATE( i915, STATE )			\
+#define SET_STATE( i915, STATE )		\
 do {						\
    i915->current->emitted &= ~ACTIVE;		\
-   i915->current = &i915->STATE;			\
+   i915->current = &i915->STATE;		\
    i915->current->emitted &= ~ACTIVE;		\
 } while (0)
 
-/* Operations where the 3D engine is decoupled temporarily from the
- * current GL state and used for other purposes than simply rendering
- * incoming triangles.
- */
-static void set_initial_state( i915ContextPtr i915 )
-{
-   memcpy(&i915->meta, &i915->initial, sizeof(i915->meta) );
-   i915->meta.active = ACTIVE;
-   i915->meta.emitted = 0;
-}
-
 
-static void set_no_depth_stencil_write( i915ContextPtr i915 )
+static void meta_no_stencil_write( struct intel_context *intel )
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    /* ctx->Driver.Enable( ctx, GL_STENCIL_TEST, GL_FALSE )
     */
    i915->meta.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_TEST_ENABLE | 
 				       S5_STENCIL_WRITE_ENABLE);
 
+   i915->meta.emitted &= ~I915_UPLOAD_CTX;
+}
+
+static void meta_no_depth_write( struct intel_context *intel )
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
     */
    i915->meta.Ctx[I915_CTXREG_LIS6] &= ~(S6_DEPTH_TEST_ENABLE |
@@ -81,12 +82,33 @@ static void set_no_depth_stencil_write( i915ContextPtr i915 )
    i915->meta.emitted &= ~I915_UPLOAD_CTX;
 }
 
+static void meta_depth_replace( struct intel_context *intel )
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
+   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_TRUE )
+    * ctx->Driver.DepthMask( ctx, GL_TRUE )
+    */
+   i915->meta.Ctx[I915_CTXREG_LIS6] |= (S6_DEPTH_TEST_ENABLE |
+					S6_DEPTH_WRITE_ENABLE);
+
+   /* ctx->Driver.DepthFunc( ctx, GL_REPLACE )
+    */
+   i915->meta.Ctx[I915_CTXREG_LIS6] &= ~S6_DEPTH_TEST_FUNC_MASK;
+   i915->meta.Ctx[I915_CTXREG_LIS6] |= 
+      COMPAREFUNC_ALWAYS << S6_DEPTH_TEST_FUNC_SHIFT;
+
+   i915->meta.emitted &= ~I915_UPLOAD_CTX;
+}
+
+
 /* Set stencil unit to replace always with the reference value.
  */
-static void set_stencil_replace( i915ContextPtr i915,
+static void meta_stencil_replace( struct intel_context *intel,
 				 GLuint s_mask,
 				 GLuint s_clear)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
    GLuint op = STENCILOP_REPLACE;
    GLuint func = COMPAREFUNC_ALWAYS;
 
@@ -95,13 +117,6 @@ static void set_stencil_replace( i915ContextPtr i915,
    i915->meta.Ctx[I915_CTXREG_LIS5] |= (S5_STENCIL_TEST_ENABLE | 
 				      S5_STENCIL_WRITE_ENABLE);
 
-
-   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
-    */
-   i915->meta.Ctx[I915_CTXREG_LIS6] &= ~(S6_DEPTH_TEST_ENABLE |
-				       S6_DEPTH_WRITE_ENABLE);
-
-
    /* ctx->Driver.StencilMask( ctx, s_mask )
     */
    i915->meta.Ctx[I915_CTXREG_STATE4] &= ~MODE4_ENABLE_STENCIL_WRITE_MASK;
@@ -109,7 +124,6 @@ static void set_stencil_replace( i915ContextPtr i915,
    i915->meta.Ctx[I915_CTXREG_STATE4] |= (ENABLE_STENCIL_WRITE_MASK |
 					STENCIL_WRITE_MASK(s_mask));
 
-
    /* ctx->Driver.StencilOp( ctx, GL_REPLACE, GL_REPLACE, GL_REPLACE )
     */
    i915->meta.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_FAIL_MASK |
@@ -138,8 +152,9 @@ static void set_stencil_replace( i915ContextPtr i915,
 }
 
 
-static void set_color_mask( i915ContextPtr i915, GLboolean state )
+static void meta_color_mask( struct intel_context *intel, GLboolean state )
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
    const GLuint mask = (S5_WRITEDISABLE_RED |
 			S5_WRITEDISABLE_GREEN |
 			S5_WRITEDISABLE_BLUE |
@@ -160,6 +175,28 @@ static void set_color_mask( i915ContextPtr i915, GLboolean state )
 
 
 
+static void meta_import_pixel_state( struct intel_context *intel )
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   memcpy(i915->meta.Fog, i915->state.Fog, I915_FOG_SETUP_SIZE * 4);
+   
+   i915->meta.Ctx[I915_CTXREG_LIS5] = i915->state.Ctx[I915_CTXREG_LIS5];
+   i915->meta.Ctx[I915_CTXREG_LIS6] = i915->state.Ctx[I915_CTXREG_LIS6];
+   i915->meta.Ctx[I915_CTXREG_STATE4] = i915->state.Ctx[I915_CTXREG_STATE4];
+   i915->meta.Ctx[I915_CTXREG_BLENDCOLOR1] = i915->state.Ctx[I915_CTXREG_BLENDCOLOR1];
+   i915->meta.Ctx[I915_CTXREG_IAB] = i915->state.Ctx[I915_CTXREG_IAB];
+
+   i915->meta.Buffer[I915_DESTREG_SENABLE] = i915->state.Buffer[I915_DESTREG_SENABLE];
+   i915->meta.Buffer[I915_DESTREG_SR1] = i915->state.Buffer[I915_DESTREG_SR1];
+   i915->meta.Buffer[I915_DESTREG_SR2] = i915->state.Buffer[I915_DESTREG_SR2];
+
+   i915->meta.emitted &= ~I915_UPLOAD_FOG;
+   i915->meta.emitted &= ~I915_UPLOAD_BUFFERS;
+   i915->meta.emitted &= ~I915_UPLOAD_CTX;
+}
+
+
+
 
 #define REG( type, nr ) (((type)<<5)|(nr))
 
@@ -211,8 +248,10 @@ static void set_color_mask( i915ContextPtr i915, GLboolean state )
 
 
 
-static void set_no_texture( i915ContextPtr i915 )
+static void meta_no_texture( struct intel_context *intel )
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    static const GLuint prog[] = {
       _3DSTATE_PIXEL_SHADER_PROGRAM,
 
@@ -241,9 +280,10 @@ static void set_no_texture( i915ContextPtr i915 )
    i915->meta.emitted &= ~I915_UPLOAD_PROGRAM;
 }
 
-
-static void enable_texture_blend_replace( i915ContextPtr i915 )
+static void meta_texture_blend_replace( struct intel_context *intel )
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    static const GLuint prog[] = {
       _3DSTATE_PIXEL_SHADER_PROGRAM,
 
@@ -286,85 +326,126 @@ static void enable_texture_blend_replace( i915ContextPtr i915 )
 /* Set up an arbitary piece of memory as a rectangular texture
  * (including the front or back buffer).
  */
-static void set_tex_rect_source( i915ContextPtr i915,
-				 GLuint offset,
-				 GLuint width, 
-				 GLuint height,
-				 GLuint pitch, /* in bytes! */
-				 GLuint textureFormat )
+static GLboolean meta_tex_rect_source( struct intel_context *intel,
+				       GLuint buffer,
+				       GLuint offset, 
+				       GLuint pitch,
+				       GLuint height,
+				       GLenum format,
+				       GLenum type)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
    GLuint unit = 0;
    GLint numLevels = 1;
    GLuint *state = i915->meta.Tex[0];
+   GLuint textureFormat;
+   GLuint cpp;
 
-#if 0
-   printf("TexRect source offset 0x%x  pitch %d\n", offset, pitch);
-#endif
+   /* A full implementation of this would do the upload through
+    * glTexImage2d, and get all the conversion operations at that
+    * point.  We are restricted, but still at least have access to the
+    * fragment program swizzle.
+    */
+   switch (format) {
+   case GL_BGRA:
+      switch (type) {
+      case GL_UNSIGNED_INT_8_8_8_8_REV:
+      case GL_UNSIGNED_BYTE:
+	 textureFormat = (MAPSURF_32BIT | MT_32BIT_ARGB8888);
+	 cpp = 4;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+   case GL_RGBA:
+      switch (type) {
+      case GL_UNSIGNED_INT_8_8_8_8_REV:
+      case GL_UNSIGNED_BYTE:
+	 textureFormat = (MAPSURF_32BIT | MT_32BIT_ABGR8888);
+	 cpp = 4;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+   case GL_BGR:
+      switch (type) {
+      case GL_UNSIGNED_SHORT_5_6_5_REV:
+	 textureFormat = (MAPSURF_16BIT | MT_16BIT_RGB565);
+	 cpp = 2;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+   case GL_RGB:
+      switch (type) {
+      case GL_UNSIGNED_SHORT_5_6_5:
+	 textureFormat = (MAPSURF_16BIT | MT_16BIT_RGB565);
+	 cpp = 2;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   default:
+      return GL_FALSE;
+   }
 
-/*    fprintf(stderr, "%s: offset: %x w: %d h: %d pitch %d format %x\n", */
-/* 	   __FUNCTION__, offset, width, height, pitch, textureFormat ); */
 
-   state[I915_TEXREG_MS2] = offset;
+   if ((pitch * cpp) & 3) {
+      _mesa_printf("%s: texture is not dword pitch\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+/*    intel_region_release(intel, &i915->meta.tex_region[0]); */
+/*    intel_region_reference(&i915->meta.tex_region[0], region); */
+   i915->meta.tex_buffer[0] = buffer;
+   i915->meta.tex_offset[0] = offset;
+
    state[I915_TEXREG_MS3] = (((height - 1) << MS3_HEIGHT_SHIFT) |
-			    ((width - 1) << MS3_WIDTH_SHIFT) |
-			    textureFormat |
-			    MS3_USE_FENCE_REGS);
+			     ((pitch - 1) << MS3_WIDTH_SHIFT) |
+			     textureFormat |
+			     MS3_USE_FENCE_REGS);
 
-   state[I915_TEXREG_MS4] = ((((pitch / 4) - 1) << MS4_PITCH_SHIFT) | 
-			    ((((numLevels-1) * 4)) << MS4_MAX_LOD_SHIFT));
+   state[I915_TEXREG_MS4] = (((((pitch * cpp) / 4) - 1) << MS4_PITCH_SHIFT) | 
+			     MS4_CUBE_FACE_ENA_MASK |
+			     ((((numLevels-1) * 4)) << MS4_MAX_LOD_SHIFT));
 
    state[I915_TEXREG_SS2] = ((FILTER_NEAREST << SS2_MIN_FILTER_SHIFT) |
-			    (MIPFILTER_NONE << SS2_MIP_FILTER_SHIFT) |
-			    (FILTER_NEAREST << SS2_MAG_FILTER_SHIFT));
+			     (MIPFILTER_NONE << SS2_MIP_FILTER_SHIFT) |
+			     (FILTER_NEAREST << SS2_MAG_FILTER_SHIFT));
+
    state[I915_TEXREG_SS3] = ((TEXCOORDMODE_WRAP << SS3_TCX_ADDR_MODE_SHIFT) |
-			    (TEXCOORDMODE_WRAP << SS3_TCY_ADDR_MODE_SHIFT) |
-			    (TEXCOORDMODE_WRAP << SS3_TCZ_ADDR_MODE_SHIFT) |
-			    (unit<<SS3_TEXTUREMAP_INDEX_SHIFT));
+			     (TEXCOORDMODE_WRAP << SS3_TCY_ADDR_MODE_SHIFT) |
+			     (TEXCOORDMODE_WRAP << SS3_TCZ_ADDR_MODE_SHIFT) |
+			     (unit<<SS3_TEXTUREMAP_INDEX_SHIFT));
 
    state[I915_TEXREG_SS4] = 0;
 
    i915->meta.emitted &= ~I915_UPLOAD_TEX(0);
+   return GL_TRUE;
 }
 
 
-/* Select between front and back draw buffers.
+/**
+ * Set the color and depth drawing region for meta ops.
  */
-static void set_draw_region( i915ContextPtr i915, const intelRegion *region )
+static void meta_draw_region( struct intel_context *intel,
+			      struct intel_region *color_region,
+			      struct intel_region *depth_region )
 {
-#if 0
-   printf("Rotate into region: offset 0x%x  pitch %d\n",
-          region->offset, region->pitch);
-#endif
-   i915->meta.Buffer[I915_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i915->meta.Buffer[I915_DESTREG_CBUFADDR2] = region->offset;
-   i915->meta.emitted &= ~I915_UPLOAD_BUFFERS;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   i915_state_draw_region(intel, &i915->meta, color_region, depth_region);
 }
 
 
-#if 0
-/* Setup an arbitary draw format, useful for targeting texture or agp
- * memory.
- */
-static void set_draw_format( i915ContextPtr i915,
-			     GLuint format,
-			     GLuint depth_format)
+static void set_vertex_format( struct intel_context *intel )
 {
-   i915->meta.Buffer[I915_DESTREG_DV1] = (DSTORG_HORT_BIAS(0x8) | /* .5 */
-					DSTORG_VERT_BIAS(0x8) | /* .5 */
-					format |
-					LOD_PRECLAMP_OGL |
-					TEX_DEFAULT_COLOR_OGL |
-					depth_format);
-
-   i915->meta.emitted &= ~I915_UPLOAD_BUFFERS;
-/*    fprintf(stderr, "%s: DV1: %x\n",  */
-/* 	   __FUNCTION__, i915->meta.Buffer[I915_DESTREG_DV1]); */
-}
-#endif
+   struct i915_context *i915 = i915_context(&intel->ctx);
 
-static void set_vertex_format( i915ContextPtr i915 )
-{
    i915->meta.Ctx[I915_CTXREG_LIS2] = 
       (S2_TEXCOORD_FMT(0, TEXCOORDFMT_2D) |
        S2_TEXCOORD_FMT(1, TEXCOORDFMT_NOT_PRESENT) | 
@@ -379,322 +460,54 @@ static void set_vertex_format( i915ContextPtr i915 )
 
    i915->meta.Ctx[I915_CTXREG_LIS4] |= 
       (S4_VFMT_COLOR |
-       S4_VFMT_SPEC_FOG |
-       S4_VFMT_XYZW);
+       S4_VFMT_XYZ);
 
    i915->meta.emitted &= ~I915_UPLOAD_CTX;
-
 }
 
 
-static void draw_quad(i915ContextPtr i915, 
-		      GLfloat x0, GLfloat x1,
-		      GLfloat y0, GLfloat y1, 
-		      GLubyte red, GLubyte green,
-		      GLubyte blue, GLubyte alpha,
-		      GLfloat s0, GLfloat s1,
-		      GLfloat t0, GLfloat t1 )
-{
-   GLuint vertex_size = 8;
-   GLuint *vb = intelEmitInlinePrimitiveLocked( &i915->intel, 
-						PRIM3D_TRIFAN, 
-						4 * vertex_size,
-						vertex_size );
-   intelVertex tmp;
-   int i;
-
-   if (0)
-      fprintf(stderr, "%s: %f,%f-%f,%f 0x%x%x%x%x %f,%f-%f,%f\n",
-	      __FUNCTION__,
-	      x0,y0,x1,y1,red,green,blue,alpha,s0,t0,s1,t1);
-
-
-   /* initial vertex, left bottom */
-   tmp.v.x = x0;
-   tmp.v.y = y0;
-   tmp.v.z = 1.0;
-   tmp.v.w = 1.0; 
-   tmp.v.color.red = red;
-   tmp.v.color.green = green;
-   tmp.v.color.blue = blue;
-   tmp.v.color.alpha = alpha;
-   tmp.v.specular.red = 0;
-   tmp.v.specular.green = 0;
-   tmp.v.specular.blue = 0;
-   tmp.v.specular.alpha = 0;
-   tmp.v.u0 = s0;
-   tmp.v.v0 = t0;
-
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* right bottom */
-   vb += vertex_size;
-   tmp.v.x = x1;
-   tmp.v.u0 = s1;
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* right top */
-   vb += vertex_size;
-   tmp.v.y = y1;
-   tmp.v.v0 = t1;
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* left top */
-   vb += vertex_size;
-   tmp.v.x = x0;
-   tmp.v.u0 = s0;
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-}
-
-
-static void draw_poly(i915ContextPtr i915, 
-		      GLubyte red, GLubyte green, GLubyte blue, GLubyte alpha,
-                      GLuint numVerts,
-                      /*const*/ GLfloat verts[][2],
-                      /*const*/ GLfloat texcoords[][2])
-{
-   GLuint vertex_size = 8;
-   GLuint *vb = intelEmitInlinePrimitiveLocked( &i915->intel, 
-						PRIM3D_TRIFAN, 
-						numVerts * vertex_size,
-						vertex_size );
-   intelVertex tmp;
-   int i, k;
-
-   /* initial constant vertex fields */
-   tmp.v.z = 1.0;
-   tmp.v.w = 1.0; 
-   tmp.v.color.red = red;
-   tmp.v.color.green = green;
-   tmp.v.color.blue = blue;
-   tmp.v.color.alpha = alpha;
-   tmp.v.specular.red = 0;
-   tmp.v.specular.green = 0;
-   tmp.v.specular.blue = 0;
-   tmp.v.specular.alpha = 0;
-
-   for (k = 0; k < numVerts; k++) {
-      tmp.v.x = verts[k][0];
-      tmp.v.y = verts[k][1];
-      tmp.v.u0 = texcoords[k][0];
-      tmp.v.v0 = texcoords[k][1];
-
-      for (i = 0 ; i < vertex_size ; i++)
-         vb[i] = tmp.ui[i];
-
-      vb += vertex_size;
-   }
-}
-
 
-void 
-i915ClearWithTris(intelContextPtr intel, GLbitfield mask,
-		  GLboolean all,
-		  GLint cx, GLint cy, GLint cw, GLint ch)
+/* Operations where the 3D engine is decoupled temporarily from the
+ * current GL state and used for other purposes than simply rendering
+ * incoming triangles.
+ */
+static void install_meta_state( struct intel_context *intel )
 {
-   i915ContextPtr i915 = I915_CONTEXT( intel );
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   intelScreenPrivate *screen = intel->intelScreen;
-   int x0, y0, x1, y1;
-
-   SET_STATE( i915, meta ); 
-   set_initial_state( i915 ); 
-   set_no_texture( i915 ); 
-   set_vertex_format( i915 ); 
-
-   LOCK_HARDWARE(intel);
-
-   if (!all) {
-      x0 = cx;
-      y0 = cy;
-      x1 = x0 + cw;
-      y1 = y0 + ch;
-   } else {
-      x0 = 0;
-      y0 = 0;
-      x1 = x0 + dPriv->w;
-      y1 = y0 + dPriv->h;
-   }
-
-   /* Don't do any clipping to screen - these are window coordinates.
-    * The active cliprects will be applied as for any other geometry.
-    */
-
-   if (mask & BUFFER_BIT_FRONT_LEFT) { 
-      set_no_depth_stencil_write( i915 );
-      set_color_mask( i915, GL_TRUE );
-      set_draw_region( i915, &screen->front );
-
-      draw_quad(i915, x0, x1, y0, y1,
-		intel->clear_red, intel->clear_green, 
- 		intel->clear_blue, intel->clear_alpha, 
-		0, 0, 0, 0);
-   }
-
-   if (mask & BUFFER_BIT_BACK_LEFT) {
-      set_no_depth_stencil_write( i915 );
-      set_color_mask( i915, GL_TRUE );
-      set_draw_region( i915, &screen->back );
-
-      draw_quad(i915, x0, x1, y0, y1,
-		intel->clear_red, intel->clear_green,
-		intel->clear_blue, intel->clear_alpha,
-		0, 0, 0, 0);
-   }
-
-   if (mask & BUFFER_BIT_STENCIL) {
-      set_stencil_replace( i915, 
-			   intel->ctx.Stencil.WriteMask[0], 
-			   intel->ctx.Stencil.Clear);
-      
-      set_color_mask( i915, GL_FALSE );
-      set_draw_region( i915, &screen->front ); /* could be either? */
-
-      draw_quad( i915, x0, x1, y0, y1, 0, 0, 0, 0, 0, 0, 0, 0 );
-   }
-
-   UNLOCK_HARDWARE(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   memcpy(&i915->meta, &i915->initial, sizeof(i915->meta) );
+   i915->meta.active = ACTIVE;
+   i915->meta.emitted = 0;
 
-   SET_STATE( i915, state );
+   SET_STATE(i915, meta);
+   set_vertex_format(intel);
+   meta_no_texture(intel);
 }
 
-
-/**
- * Copy the window contents named by dPriv to the rotated (or reflected)
- * color buffer.
- * srcBuf is BUFFER_BIT_FRONT_LEFT or BUFFER_BIT_BACK_LEFT to indicate the source.
- */
-void
-i915RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                 GLuint srcBuf)
+static void leave_meta_state( struct intel_context *intel )
 {
-   i915ContextPtr i915 = I915_CONTEXT( intel );
-   intelScreenPrivate *screen = intel->intelScreen;
-   const GLuint cpp = screen->cpp;
-   drm_clip_rect_t fullRect;
-   GLuint textureFormat, srcOffset, srcPitch;
-   const drm_clip_rect_t *clipRects;
-   int numClipRects;
-   int i;
-
-   int xOrig, yOrig;
-   int origNumClipRects;
-   drm_clip_rect_t *origRects;
-
-   /*
-    * set up hardware state
-    */
-   intelFlush( &intel->ctx );
-
-   SET_STATE( i915, meta ); 
-   set_initial_state( i915 ); 
-   set_no_texture( i915 ); 
-   set_vertex_format( i915 ); 
-   set_no_depth_stencil_write( i915 );
-   set_color_mask( i915, GL_TRUE );
-
-   LOCK_HARDWARE(intel);
-
-   /* save current drawing origin and cliprects (restored at end) */
-   xOrig = intel->drawX;
-   yOrig = intel->drawY;
-   origNumClipRects = intel->numClipRects;
-   origRects = intel->pClipRects;
-
-   if (!intel->numClipRects)
-      goto done;
-
-   /*
-    * set drawing origin, cliprects for full-screen access to rotated screen
-    */
-   fullRect.x1 = 0;
-   fullRect.y1 = 0;
-   fullRect.x2 = screen->rotatedWidth;
-   fullRect.y2 = screen->rotatedHeight;
-   intel->drawX = 0;
-   intel->drawY = 0;
-   intel->numClipRects = 1;
-   intel->pClipRects = &fullRect;
-
-   set_draw_region( i915, &screen->rotated );
-
-   if (cpp == 4)
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-   else
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-
-   if (srcBuf == BUFFER_BIT_FRONT_LEFT) {
-      srcPitch = screen->front.pitch;   /* in bytes */
-      srcOffset = screen->front.offset; /* bytes */
-      clipRects = dPriv->pClipRects;
-      numClipRects = dPriv->numClipRects;
-   }
-   else {
-      srcPitch = screen->back.pitch;   /* in bytes */
-      srcOffset = screen->back.offset; /* bytes */
-      clipRects = dPriv->pBackClipRects;
-      numClipRects = dPriv->numBackClipRects;
-   }
-
-   /* set the whole screen up as a texture to avoid alignment issues */
-   set_tex_rect_source(i915,
-                       srcOffset,
-                       screen->width,
-		       screen->height,
-                       srcPitch,
-                       textureFormat);
-
-   enable_texture_blend_replace(i915);
-
-   /*
-    * loop over the source window's cliprects
-    */
-   for (i = 0; i < numClipRects; i++) {
-      int srcX0 = clipRects[i].x1;
-      int srcY0 = clipRects[i].y1;
-      int srcX1 = clipRects[i].x2;
-      int srcY1 = clipRects[i].y2;
-      GLfloat verts[4][2], tex[4][2];
-      int j;
-
-      /* build vertices for four corners of clip rect */
-      verts[0][0] = srcX0;  verts[0][1] = srcY0;
-      verts[1][0] = srcX1;  verts[1][1] = srcY0;
-      verts[2][0] = srcX1;  verts[2][1] = srcY1;
-      verts[3][0] = srcX0;  verts[3][1] = srcY1;
-
-      /* .. and texcoords */
-      tex[0][0] = srcX0;  tex[0][1] = srcY0;
-      tex[1][0] = srcX1;  tex[1][1] = srcY0;
-      tex[2][0] = srcX1;  tex[2][1] = srcY1;
-      tex[3][0] = srcX0;  tex[3][1] = srcY1;
-
-      /* transform coords to rotated screen coords */
-      for (j = 0; j < 4; j++) {
-         matrix23TransformCoordf(&screen->rotMatrix,
-                                 &verts[j][0], &verts[j][1]);
-      }
-
-      /* draw polygon to map source image to dest region */
-      draw_poly(i915, 255, 255, 255, 255, 4, verts, tex);
-
-   } /* cliprect loop */
-
-   intelFlushBatchLocked( intel, GL_FALSE, GL_FALSE, GL_FALSE );
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   intel_region_release(intel, &i915->meta.draw_region);
+   intel_region_release(intel, &i915->meta.depth_region);
+/*    intel_region_release(intel, &i915->meta.tex_region[0]); */
+   SET_STATE(i915, state);
+}
 
- done:
-   /* restore original drawing origin and cliprects */
-   intel->drawX = xOrig;
-   intel->drawY = yOrig;
-   intel->numClipRects = origNumClipRects;
-   intel->pClipRects = origRects;
 
-   UNLOCK_HARDWARE(intel);
 
-   SET_STATE( i915, state );
+void i915InitMetaFuncs( struct i915_context *i915 )
+{
+   i915->intel.vtbl.install_meta_state = install_meta_state;
+   i915->intel.vtbl.leave_meta_state = leave_meta_state;
+   i915->intel.vtbl.meta_no_depth_write = meta_no_depth_write;
+   i915->intel.vtbl.meta_no_stencil_write = meta_no_stencil_write;
+   i915->intel.vtbl.meta_stencil_replace = meta_stencil_replace;
+   i915->intel.vtbl.meta_depth_replace = meta_depth_replace;
+   i915->intel.vtbl.meta_color_mask = meta_color_mask;
+   i915->intel.vtbl.meta_no_texture = meta_no_texture;
+   i915->intel.vtbl.meta_texture_blend_replace = meta_texture_blend_replace;
+   i915->intel.vtbl.meta_tex_rect_source = meta_tex_rect_source;
+   i915->intel.vtbl.meta_draw_region = meta_draw_region;
+   i915->intel.vtbl.meta_import_pixel_state = meta_import_pixel_state;
 }
 
+
diff --git a/src/mesa/drivers/dri/i915/i915_program.c b/src/mesa/drivers/dri/i915/i915_program.c
index 45276fb690..07ac8f09bb 100644
--- a/src/mesa/drivers/dri/i915/i915_program.c
+++ b/src/mesa/drivers/dri/i915/i915_program.c
@@ -144,7 +144,8 @@ GLuint i915_emit_arith( struct i915_fragment_program *p,
    GLuint nr_const = 0;
 
    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
-   assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
+   dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
+   assert(dest);
 
    if (GET_UREG_TYPE(src0) == REG_TYPE_CONST) c[nr_const++] = 0;
    if (GET_UREG_TYPE(src1) == REG_TYPE_CONST) c[nr_const++] = 1;
@@ -202,7 +203,8 @@ GLuint i915_emit_texld( struct i915_fragment_program *p,
 			  GLuint op )
 {
    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
-   assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
+   dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
+   assert(dest);
 
    if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
       p->nr_tex_indirect++;
@@ -358,7 +360,7 @@ void i915_program_error( struct i915_fragment_program *p, const char *msg )
    p->error = 1;
 }
 
-void i915_init_program( i915ContextPtr i915, struct i915_fragment_program *p )
+void i915_init_program( struct i915_context *i915, struct i915_fragment_program *p )
 {
    GLcontext *ctx = &i915->intel.ctx;
    TNLcontext *tnl = TNL_CONTEXT( ctx );
@@ -431,7 +433,7 @@ void i915_fini_program( struct i915_fragment_program *p )
    p->declarations[0] |= program_size + decl_size - 2;
 }
 
-void i915_upload_program( i915ContextPtr i915, struct i915_fragment_program *p )
+void i915_upload_program( struct i915_context *i915, struct i915_fragment_program *p )
 {
    GLuint program_size = p->csr - p->program;
    GLuint decl_size = p->decl - p->declarations;
diff --git a/src/mesa/drivers/dri/i915/i915_program.h b/src/mesa/drivers/dri/i915/i915_program.h
index 8891a17785..a0d2530ed5 100644
--- a/src/mesa/drivers/dri/i915/i915_program.h
+++ b/src/mesa/drivers/dri/i915/i915_program.h
@@ -84,7 +84,7 @@
 
 /* One neat thing about the UREG representation:  
  */
-static __inline int swizzle( int reg, int x, int y, int z, int w )
+static INLINE int swizzle( int reg, int x, int y, int z, int w )
 {
    return ((reg & ~UREG_XYZW_CHANNEL_MASK) |
 	   CHANNEL_SRC( GET_CHANNEL_SRC( reg, x ), 0 ) |
@@ -95,7 +95,7 @@ static __inline int swizzle( int reg, int x, int y, int z, int w )
 
 /* Another neat thing about the UREG representation:  
  */
-static __inline int negate( int reg, int x, int y, int z, int w )
+static INLINE int negate( int reg, int x, int y, int z, int w )
 {
    return reg ^ (((x&1)<<UREG_CHANNEL_X_NEGATE_SHIFT)|
 		 ((y&1)<<UREG_CHANNEL_Y_NEGATE_SHIFT)|
@@ -149,10 +149,10 @@ extern GLuint i915_emit_param4fv( struct i915_fragment_program *p,
 extern void i915_program_error( struct i915_fragment_program *p,
                                 const char *msg );
 
-extern void i915_init_program( i915ContextPtr i915,
+extern void i915_init_program( struct i915_context *i915,
 			      struct i915_fragment_program *p );
 
-extern void i915_upload_program( i915ContextPtr i915, 
+extern void i915_upload_program( struct i915_context *i915, 
 				struct i915_fragment_program *p );
 
 extern void i915_fini_program( struct i915_fragment_program *p );
diff --git a/src/mesa/drivers/dri/i915/i915_reg.h b/src/mesa/drivers/dri/i915/i915_reg.h
index 3ba792a234..f81b1264b7 100644
--- a/src/mesa/drivers/dri/i915/i915_reg.h
+++ b/src/mesa/drivers/dri/i915/i915_reg.h
@@ -435,7 +435,7 @@
 #define LOGICOP_MASK			(0xf<<18)
 #define MODE4_ENABLE_STENCIL_TEST_MASK	((1<<17)|(0xff00))
 #define ENABLE_STENCIL_TEST_MASK	(1<<17)
-#define STENCIL_TEST_MASK(x)		((x)<<8)
+#define STENCIL_TEST_MASK(x)		(((x)&0xff)<<8)
 #define MODE4_ENABLE_STENCIL_WRITE_MASK	((1<<16)|(0x00ff))
 #define ENABLE_STENCIL_WRITE_MASK	(1<<16)
 #define STENCIL_WRITE_MASK(x)		((x)&0xff)
@@ -826,10 +826,14 @@
 #define ST1_ENABLE               (1<<16)
 #define ST1_MASK                 (0xffff)
 
+#define _3DSTATE_DEFAULT_Z          ((0x3<<29)|(0x1d<<24)|(0x98<<16))
+#define _3DSTATE_DEFAULT_DIFFUSE    ((0x3<<29)|(0x1d<<24)|(0x99<<16))
+#define _3DSTATE_DEFAULT_SPECULAR   ((0x3<<29)|(0x1d<<24)|(0x9a<<16))
 
-#define MI_FLUSH           ((0<<29)|(4<<23))
-#define FLUSH_MAP_CACHE    (1<<0)
-#define FLUSH_RENDER_CACHE (1<<1)
+
+#define MI_FLUSH                   ((0<<29)|(4<<23))
+#define FLUSH_MAP_CACHE            (1<<0)
+#define INHIBIT_FLUSH_RENDER_CACHE (1<<2)
 
 
 #endif
diff --git a/src/mesa/drivers/dri/i915/i915_state.c b/src/mesa/drivers/dri/i915/i915_state.c
index bad947f5de..ab262ff735 100644
--- a/src/mesa/drivers/dri/i915/i915_state.c
+++ b/src/mesa/drivers/dri/i915/i915_state.c
@@ -36,6 +36,7 @@
 
 #include "texmem.h"
 
+#include "intel_fbo.h"
 #include "intel_screen.h"
 #include "intel_batchbuffer.h"
 
@@ -48,7 +49,7 @@ static void
 i915StencilFuncSeparate(GLcontext *ctx, GLenum face, GLenum func, GLint ref,
                         GLuint mask)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int test = intel_translate_compare_func( func );
 
    mask = mask & 0xff;
@@ -73,7 +74,7 @@ i915StencilFuncSeparate(GLcontext *ctx, GLenum face, GLenum func, GLint ref,
 static void
 i915StencilMaskSeparate(GLcontext *ctx, GLenum face, GLuint mask)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
 
    if (INTEL_DEBUG&DEBUG_DRI)
       fprintf(stderr, "%s : mask 0x%x\n", __FUNCTION__, mask);
@@ -91,7 +92,7 @@ static void
 i915StencilOpSeparate(GLcontext *ctx, GLenum face, GLenum fail, GLenum zfail,
                       GLenum zpass)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int fop = intel_translate_stencil_op(fail); 
    int dfop = intel_translate_stencil_op(zfail); 
    int dpop = intel_translate_stencil_op(zpass);
@@ -116,7 +117,7 @@ i915StencilOpSeparate(GLcontext *ctx, GLenum face, GLenum fail, GLenum zfail,
 
 static void i915AlphaFunc(GLcontext *ctx, GLenum func, GLfloat ref)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int test = intel_translate_compare_func( func );
    GLubyte refByte;
 
@@ -137,7 +138,7 @@ static void i915AlphaFunc(GLcontext *ctx, GLenum func, GLfloat ref)
  */
 static void i915EvalLogicOpBlendState(GLcontext *ctx)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
 
    I915_STATECHANGE(i915, I915_UPLOAD_CTX);
 
@@ -157,7 +158,7 @@ static void i915EvalLogicOpBlendState(GLcontext *ctx)
 
 static void i915BlendColor(GLcontext *ctx, const GLfloat color[4])
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    GLubyte r, g, b, a;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -194,7 +195,7 @@ static GLuint translate_blend_equation( GLenum mode )
 
 static void i915UpdateBlendState( GLcontext *ctx )
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    GLuint iab = (i915->state.Ctx[I915_CTXREG_IAB] & 
 		 ~(IAB_SRC_FACTOR_MASK |
 		   IAB_DST_FACTOR_MASK |
@@ -261,7 +262,7 @@ static void i915BlendEquationSeparate(GLcontext *ctx, GLenum eqRGB,
 
 static void i915DepthFunc(GLcontext *ctx, GLenum func)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int test = intel_translate_compare_func( func );
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -274,7 +275,7 @@ static void i915DepthFunc(GLcontext *ctx, GLenum func)
 
 static void i915DepthMask(GLcontext *ctx, GLboolean flag)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
 
    if (INTEL_DEBUG&DEBUG_DRI)
       fprintf(stderr, "%s flag (%d)\n", __FUNCTION__, flag);
@@ -295,7 +296,7 @@ static void i915DepthMask(GLcontext *ctx, GLboolean flag)
  */
 static void i915PolygonStipple( GLcontext *ctx, const GLubyte *mask )
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    const GLubyte *m = mask;
    GLubyte p[4];
    int i,j,k;
@@ -348,15 +349,14 @@ static void i915PolygonStipple( GLcontext *ctx, const GLubyte *mask )
 static void i915Scissor(GLcontext *ctx, GLint x, GLint y, 
 			GLsizei w, GLsizei h)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
-   intelScreenPrivate *screen = i915->intel.intelScreen;
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int x1, y1, x2, y2;
 
-   if (!i915->intel.driDrawable)
+   if (!ctx->DrawBuffer)
       return;
 
    x1 = x;
-   y1 = i915->intel.driDrawable->h - (y + h);
+   y1 = ctx->DrawBuffer->Height - (y + h);
    x2 = x + w - 1;
    y2 = y1 + h - 1;
 
@@ -364,16 +364,10 @@ static void i915Scissor(GLcontext *ctx, GLint x, GLint y,
       fprintf(stderr, "[%s] x(%d) y(%d) w(%d) h(%d)\n", __FUNCTION__,
 	      x, y, w, h);
 
-   if (x1 < 0) x1 = 0;
-   if (y1 < 0) y1 = 0;
-   if (x2 < 0) x2 = 0;
-   if (y2 < 0) y2 = 0;
-
-   if (x2 >= screen->width) x2 = screen->width-1;
-   if (y2 >= screen->height) y2 = screen->height-1;
-   if (x1 >= screen->width) x1 = screen->width-1;
-   if (y1 >= screen->height) y1 = screen->height-1;
-
+   x1 = CLAMP(x1, 0, ctx->DrawBuffer->Width - 1);
+   y1 = CLAMP(y1, 0, ctx->DrawBuffer->Height - 1);
+   x2 = CLAMP(x2, 0, ctx->DrawBuffer->Width - 1);
+   y2 = CLAMP(y2, 0, ctx->DrawBuffer->Height - 1);
 
    I915_STATECHANGE(i915, I915_UPLOAD_BUFFERS);
    i915->state.Buffer[I915_DESTREG_SR1] = (y1 << 16) | (x1 & 0xffff);
@@ -382,7 +376,7 @@ static void i915Scissor(GLcontext *ctx, GLint x, GLint y,
 
 static void i915LogicOp(GLcontext *ctx, GLenum opcode)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int tmp = intel_translate_logic_op(opcode);
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -397,11 +391,12 @@ static void i915LogicOp(GLcontext *ctx, GLenum opcode)
 
 static void i915CullFaceFrontFace(GLcontext *ctx, GLenum unused)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    GLuint mode;
 
    if (INTEL_DEBUG&DEBUG_DRI)
-      fprintf(stderr, "%s\n", __FUNCTION__);
+      fprintf(stderr, "%s %d\n", __FUNCTION__, 
+	      ctx->DrawBuffer ? ctx->DrawBuffer->Name : 0);
 
    if (!ctx->Polygon.CullFlag) {
       mode = S4_CULLMODE_NONE;
@@ -409,6 +404,8 @@ static void i915CullFaceFrontFace(GLcontext *ctx, GLenum unused)
    else if (ctx->Polygon.CullFaceMode != GL_FRONT_AND_BACK) {
       mode = S4_CULLMODE_CW;
 
+      if (ctx->DrawBuffer && ctx->DrawBuffer->Name != 0) 
+	 mode ^= (S4_CULLMODE_CW ^ S4_CULLMODE_CCW);
       if (ctx->Polygon.CullFaceMode == GL_FRONT)
 	 mode ^= (S4_CULLMODE_CW ^ S4_CULLMODE_CCW);
       if (ctx->Polygon.FrontFace != GL_CCW)
@@ -425,7 +422,7 @@ static void i915CullFaceFrontFace(GLcontext *ctx, GLenum unused)
 
 static void i915LineWidth( GLcontext *ctx, GLfloat widthf )
 {
-   i915ContextPtr i915 = I915_CONTEXT( ctx );
+   struct i915_context *i915 = I915_CONTEXT( ctx );
    int lis4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_LINE_WIDTH_MASK;
    int width;
 
@@ -444,7 +441,7 @@ static void i915LineWidth( GLcontext *ctx, GLfloat widthf )
 
 static void i915PointSize(GLcontext *ctx, GLfloat size)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    int lis4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_POINT_WIDTH_MASK;
    GLint point_size = (int)size;
 
@@ -469,7 +466,7 @@ static void i915ColorMask(GLcontext *ctx,
 			 GLboolean r, GLboolean g,
 			 GLboolean b, GLboolean a)
 {
-   i915ContextPtr i915 = I915_CONTEXT( ctx );
+   struct i915_context *i915 = I915_CONTEXT( ctx );
    GLuint tmp = i915->state.Ctx[I915_CTXREG_LIS5] & ~S5_WRITEDISABLE_MASK;
 
    if (INTEL_DEBUG&DEBUG_DRI)
@@ -490,7 +487,7 @@ static void update_specular( GLcontext *ctx )
 {
    /* A hack to trigger the rebuild of the fragment program.
     */
-   INTEL_CONTEXT(ctx)->NewGLState |= _NEW_TEXTURE;
+   intel_context(ctx)->NewGLState |= _NEW_TEXTURE;
    I915_CONTEXT(ctx)->tex_program.translated = 0; 
 }
 
@@ -507,7 +504,7 @@ static void i915LightModelfv(GLcontext *ctx, GLenum pname,
 
 static void i915ShadeModel(GLcontext *ctx, GLenum mode)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    I915_STATECHANGE(i915, I915_UPLOAD_CTX);
 
    if (mode == GL_SMOOTH) {
@@ -526,7 +523,7 @@ static void i915ShadeModel(GLcontext *ctx, GLenum mode)
  */
 void i915_update_fog( GLcontext *ctx )
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
    GLenum mode;
    GLboolean enabled;
    GLboolean try_pixel_fog;
@@ -619,7 +616,7 @@ void i915_update_fog( GLcontext *ctx )
 
 static void i915Fogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
 
    switch (pname) {
    case GL_FOG_COORDINATE_SOURCE_EXT: 
@@ -671,7 +668,7 @@ static void i915Hint(GLcontext *ctx, GLenum target, GLenum state)
 
 static void i915Enable(GLcontext *ctx, GLenum cap, GLboolean state)
 {
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
+   struct i915_context *i915 = I915_CONTEXT(ctx);
 
    switch(cap) {
    case GL_TEXTURE_2D:
@@ -699,7 +696,7 @@ static void i915Enable(GLcontext *ctx, GLenum cap, GLboolean state)
 
       /* Logicop doesn't seem to work at 16bpp:
        */
-      if (i915->intel.intelScreen->cpp == 2)
+      if (i915->intel.intelScreen->cpp == 2) /* XXX FBO fix */
 	 FALLBACK( &i915->intel, I915_FALLBACK_LOGICOP, state );
       break;
 
@@ -750,16 +747,24 @@ static void i915Enable(GLcontext *ctx, GLenum cap, GLboolean state)
       break;
 
    case GL_STENCIL_TEST:
-      if (i915->intel.hw_stencil) {
-	 I915_STATECHANGE(i915, I915_UPLOAD_CTX);
-	 if (state)
-	    i915->state.Ctx[I915_CTXREG_LIS5] |= (S5_STENCIL_TEST_ENABLE |
-						S5_STENCIL_WRITE_ENABLE);
-	 else
-	    i915->state.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_TEST_ENABLE | 
-						 S5_STENCIL_WRITE_ENABLE);
-      } else {
-	 FALLBACK( &i915->intel, I915_FALLBACK_STENCIL, state );
+      {
+         GLboolean hw_stencil = GL_FALSE;
+         if (ctx->DrawBuffer) {
+            struct intel_renderbuffer *irbStencil
+               = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_STENCIL);
+            hw_stencil = (irbStencil && irbStencil->region);
+         }
+         if (hw_stencil) {
+            I915_STATECHANGE(i915, I915_UPLOAD_CTX);
+            if (state)
+               i915->state.Ctx[I915_CTXREG_LIS5] |= (S5_STENCIL_TEST_ENABLE |
+                                                   S5_STENCIL_WRITE_ENABLE);
+            else
+               i915->state.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_TEST_ENABLE | 
+                                                    S5_STENCIL_WRITE_ENABLE);
+         } else {
+            FALLBACK( &i915->intel, I915_FALLBACK_STENCIL, state );
+         }
       }
       break;
 
@@ -793,7 +798,7 @@ static void i915Enable(GLcontext *ctx, GLenum cap, GLboolean state)
 }
 
 
-static void i915_init_packets( i915ContextPtr i915 )
+static void i915_init_packets( struct i915_context *i915 )
 {
    intelScreenPrivate *screen = i915->intel.intelScreen;
 
@@ -816,7 +821,7 @@ static void i915_init_packets( i915ContextPtr i915 )
       i915->state.Ctx[I915_CTXREG_LIS4] = 0;
       i915->state.Ctx[I915_CTXREG_LIS5] = 0;
 
-      if (screen->cpp == 2)
+      if (screen->cpp == 2) /* XXX FBO fix */
 	 i915->state.Ctx[I915_CTXREG_LIS5] |= S5_COLOR_DITHER_ENABLE;
 
 
@@ -831,7 +836,6 @@ static void i915_init_packets( i915ContextPtr i915 )
 					   ENABLE_STENCIL_WRITE_MASK |
 					   STENCIL_WRITE_MASK(0xff));
 
-
       i915->state.Ctx[I915_CTXREG_IAB] = (_3DSTATE_INDEPENDENT_ALPHA_BLEND_CMD |
 					IAB_MODIFY_ENABLE |
 					IAB_MODIFY_FUNC |
@@ -866,27 +870,24 @@ static void i915_init_packets( i915ContextPtr i915 )
       I915_STATECHANGE(i915, I915_UPLOAD_BUFFERS);
       /* color buffer offset/stride */
       i915->state.Buffer[I915_DESTREG_CBUFADDR0] = _3DSTATE_BUF_INFO_CMD;
+      /* XXX FBO: remove this?  Also get set in i915_set_draw_region() */
       i915->state.Buffer[I915_DESTREG_CBUFADDR1] = 
 	 (BUF_3D_ID_COLOR_BACK | 
 	  BUF_3D_PITCH(screen->front.pitch) |  /* pitch in bytes */
 	  BUF_3D_USE_FENCE);
-      /*i915->state.Buffer[I915_DESTREG_CBUFADDR2] is the offset */
-
 
-      /* depth/Z buffer offset/stride */
       i915->state.Buffer[I915_DESTREG_DBUFADDR0] = _3DSTATE_BUF_INFO_CMD;
+      /* XXX FBO: remove this?  Also get set in i915_set_draw_region() */
       i915->state.Buffer[I915_DESTREG_DBUFADDR1] = 
 	 (BUF_3D_ID_DEPTH |
 	  BUF_3D_PITCH(screen->depth.pitch) |  /* pitch in bytes */
 	  BUF_3D_USE_FENCE);
-      i915->state.Buffer[I915_DESTREG_DBUFADDR2] = screen->depth.offset;
-
 
       i915->state.Buffer[I915_DESTREG_DV0] = _3DSTATE_DST_BUF_VARS_CMD;
 
-      /* color/depth pixel format */
+      /* XXX FBO: remove this?  Also get set in i915_set_draw_region() */
+#if 0 /* seems we don't need this */
       switch (screen->fbFormat) {
-      case DV_PF_555:
       case DV_PF_565:
 	 i915->state.Buffer[I915_DESTREG_DV1] = (DSTORG_HORT_BIAS(0x8) | /* .5 */
 					       DSTORG_VERT_BIAS(0x8) | /* .5 */
@@ -905,6 +906,8 @@ static void i915_init_packets( i915ContextPtr i915 )
 					       DEPTH_FRMT_24_FIXED_8_OTHER);
 	 break;
       }
+#endif
+
 
       /* scissor */
       i915->state.Buffer[I915_DESTREG_SENABLE] = (_3DSTATE_SCISSOR_ENABLE_CMD |
@@ -915,13 +918,27 @@ static void i915_init_packets( i915ContextPtr i915 )
    }
 
 
+#if 0
+   {
+      I915_STATECHANGE(i915, I915_UPLOAD_DEFAULTS);
+      i915->state.Default[I915_DEFREG_C0] = _3DSTATE_DEFAULT_DIFFUSE;
+      i915->state.Default[I915_DEFREG_C1] = 0;
+      i915->state.Default[I915_DEFREG_S0] = _3DSTATE_DEFAULT_SPECULAR;
+      i915->state.Default[I915_DEFREG_S1] = 0;
+      i915->state.Default[I915_DEFREG_Z0] = _3DSTATE_DEFAULT_Z;
+      i915->state.Default[I915_DEFREG_Z1] = 0;
+   }
+#endif
+
+
    /* These will be emitted every at the head of every buffer, unless
     * we get hardware contexts working.
     */
    i915->state.active = (I915_UPLOAD_PROGRAM | 
-			I915_UPLOAD_STIPPLE | 
-			I915_UPLOAD_CTX | 
-			I915_UPLOAD_BUFFERS);
+			 I915_UPLOAD_STIPPLE | 
+			 I915_UPLOAD_CTX | 
+			 I915_UPLOAD_BUFFERS |
+			 I915_UPLOAD_INVARIENT);
 }
 
 void i915InitStateFunctions( struct dd_function_table *functions )
@@ -951,7 +968,7 @@ void i915InitStateFunctions( struct dd_function_table *functions )
 }
 
 
-void i915InitState( i915ContextPtr i915 )
+void i915InitState( struct i915_context *i915 )
 {
    GLcontext *ctx = &i915->intel.ctx;
 
diff --git a/src/mesa/drivers/dri/i915/i915_tex.c b/src/mesa/drivers/dri/i915/i915_tex.c
index d9609d3193..7f47cf9044 100644
--- a/src/mesa/drivers/dri/i915/i915_tex.c
+++ b/src/mesa/drivers/dri/i915/i915_tex.c
@@ -45,76 +45,10 @@
 
 
 
-
-
-
-/**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-intelTextureObjectPtr i915AllocTexObj( struct gl_texture_object *texObj )
-{
-   i915TextureObjectPtr t = CALLOC_STRUCT( i915_texture_object );
-   if ( !t ) 
-      return NULL;
-
-   texObj->DriverData = t;
-   t->intel.base.tObj = texObj;
-   t->intel.dirty = I915_UPLOAD_TEX_ALL;
-   make_empty_list( &t->intel.base );
-   return &t->intel;
-}
-
-
-static void i915TexParameter( GLcontext *ctx, GLenum target,
-			     struct gl_texture_object *tObj,
-			     GLenum pname, const GLfloat *params )
-{
-   i915TextureObjectPtr t = (i915TextureObjectPtr) tObj->DriverData;
- 
-   switch (pname) {
-   case GL_TEXTURE_MIN_FILTER:
-   case GL_TEXTURE_MAG_FILTER:
-   case GL_TEXTURE_MAX_ANISOTROPY_EXT:
-   case GL_TEXTURE_WRAP_S:
-   case GL_TEXTURE_WRAP_T:
-   case GL_TEXTURE_WRAP_R:
-   case GL_TEXTURE_BORDER_COLOR:
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-
-   case GL_TEXTURE_COMPARE_MODE:
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-   case GL_TEXTURE_COMPARE_FUNC:
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-
-   case GL_TEXTURE_BASE_LEVEL:
-   case GL_TEXTURE_MAX_LEVEL:
-   case GL_TEXTURE_MIN_LOD:
-   case GL_TEXTURE_MAX_LOD:
-      /* The i915 and its successors can do a lot of this without
-       * reloading the textures.  A project for someone?
-       */
-      intelFlush( ctx );
-      driSwapOutTextureObject( (driTextureObject *) t );
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-
-   default:
-      return;
-   }
-}
-
-
 static void i915TexEnv( GLcontext *ctx, GLenum target, 
 			GLenum pname, const GLfloat *param )
 {
-   i915ContextPtr i915 = I915_CONTEXT( ctx );
-   GLuint unit = ctx->Texture.CurrentUnit;
+   struct i915_context *i915 = I915_CONTEXT( ctx );
 
    switch (pname) {
    case GL_TEXTURE_ENV_COLOR: 	/* Should be a tracked param */
@@ -139,13 +73,12 @@ static void i915TexEnv( GLcontext *ctx, GLenum target,
       break;
 
    case GL_TEXTURE_LOD_BIAS: {
-      int b = (int) ((*param) * 16.0);
+      GLuint unit = ctx->Texture.CurrentUnit;
+      GLint b = (int) ((*param) * 16.0);
       if (b > 255) b = 255;
       if (b < -256) b = -256;
       I915_STATECHANGE(i915, I915_UPLOAD_TEX(unit));
-      i915->state.Tex[unit][I915_TEXREG_SS2] &= ~SS2_LOD_BIAS_MASK;
-      i915->state.Tex[unit][I915_TEXREG_SS2] |= 
-	 ((b << SS2_LOD_BIAS_SHIFT) & SS2_LOD_BIAS_MASK);
+      i915->lodbias_ss2[unit] = ((b << SS2_LOD_BIAS_SHIFT) & SS2_LOD_BIAS_MASK);
       break;
    }
 
@@ -156,20 +89,8 @@ static void i915TexEnv( GLcontext *ctx, GLenum target,
 
 
 static void i915BindTexture( GLcontext *ctx, GLenum target,
-			    struct gl_texture_object *texObj )
+                             struct gl_texture_object *texobj )
 {
-   i915TextureObjectPtr tex;
-   
-   if (!texObj->DriverData)
-      i915AllocTexObj( texObj );
-   
-   tex = (i915TextureObjectPtr)texObj->DriverData;
-
-   if (tex->lastTarget != texObj->Target) {
-      tex->intel.dirty = I915_UPLOAD_TEX_ALL;
-      tex->lastTarget = texObj->Target;
-   }
-
    /* Need this if image format changes between bound textures.
     * Could try and shortcircuit by checking for differences in
     * state between incoming and outgoing textures:
@@ -183,5 +104,4 @@ void i915InitTextureFuncs( struct dd_function_table *functions )
 {
    functions->BindTexture = i915BindTexture;
    functions->TexEnv = i915TexEnv;
-   functions->TexParameter = i915TexParameter;
 }
diff --git a/src/mesa/drivers/dri/i915/i915_texprog.c b/src/mesa/drivers/dri/i915/i915_texprog.c
index 4fbce34ba2..8aedc43d94 100644
--- a/src/mesa/drivers/dri/i915/i915_texprog.c
+++ b/src/mesa/drivers/dri/i915/i915_texprog.c
@@ -536,7 +536,7 @@ static void emit_program_fini( struct i915_fragment_program *p )
 }
 
 
-static void i915EmitTextureProgram( i915ContextPtr i915 )
+static void i915EmitTextureProgram( struct i915_context *i915 )
 {
    GLcontext *ctx = &i915->intel.ctx;
    struct i915_fragment_program *p = &i915->tex_program;
@@ -570,9 +570,9 @@ static void i915EmitTextureProgram( i915ContextPtr i915 )
 }
 
 
-void i915ValidateTextureProgram( i915ContextPtr i915 )
+void i915ValidateTextureProgram( struct i915_context *i915 )
 {
-   intelContextPtr intel = &i915->intel;
+   struct intel_context *intel = &i915->intel;
    GLcontext *ctx = &intel->ctx;
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
@@ -642,7 +642,8 @@ void i915ValidateTextureProgram( i915ContextPtr i915 )
     */
    if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
        s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
-    
+      int k;
+
       I915_STATECHANGE( i915, I915_UPLOAD_CTX );
 
       i915->tex_program.translated = 0;
@@ -660,7 +661,8 @@ void i915ValidateTextureProgram( i915ContextPtr i915 )
       i915->state.Ctx[I915_CTXREG_LIS2] = s2;
       i915->state.Ctx[I915_CTXREG_LIS4] = s4;
 
-      assert(intel->vtbl.check_vertex_size( intel, intel->vertex_size ));
+      k = intel->vtbl.check_vertex_size( intel, intel->vertex_size );
+      assert(k);
    }
 
    if (!i915->tex_program.translated ||
diff --git a/src/mesa/drivers/dri/i915/i915_texstate.c b/src/mesa/drivers/dri/i915/i915_texstate.c
index 3b639e7144..af68c9f424 100644
--- a/src/mesa/drivers/dri/i915/i915_texstate.c
+++ b/src/mesa/drivers/dri/i915/i915_texstate.c
@@ -25,501 +25,63 @@
  * 
  **************************************************************************/
 
-#include "glheader.h"
-#include "macros.h"
 #include "mtypes.h"
-#include "simple_list.h"
 #include "enums.h"
 #include "texformat.h"
-#include "texstore.h"
 
-#include "mm.h"
-
-#include "intel_screen.h"
-#include "intel_ioctl.h"
+#include "intel_mipmap_tree.h"
 #include "intel_tex.h"
 
 #include "i915_context.h"
 #include "i915_reg.h"
 
-static GLint initial_offsets[6][2] = { {0,0},
-				       {0,2},
-				       {1,0},
-				       {1,2},
-				       {1,1},
-				       {1,3} };
-
-
-static GLint step_offsets[6][2] = { {0,2},
-				    {0,2},
-				    {-1,2},
-				    {-1,2},
-				    {-1,1},
-				    {-1,1} };
-
-
-#define I915_TEX_UNIT_ENABLED(unit)		(1<<unit)
-
-static void i915LayoutTextureImages( i915ContextPtr i915,
-				     struct gl_texture_object *tObj )
-{
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   i915TextureObjectPtr t = (i915TextureObjectPtr) tObj->DriverData;
-   GLint firstLevel, lastLevel, numLevels;
-   GLint i, total_height, pitch;
-
-   /* Compute which mipmap levels we really want to send to the hardware.
-    */
-   driCalculateTextureFirstLastLevel( (driTextureObject *) t );
-
-   /* Figure out the amount of memory required to hold all the mipmap
-    * levels.  Choose the smallest pitch to accomodate the largest
-    * mipmap:
-    */
-   firstLevel = t->intel.base.firstLevel;
-   lastLevel = t->intel.base.lastLevel;
-   numLevels = lastLevel - firstLevel + 1;
-
-
-
-   /* All images must be loaded at this pitch.  Count the number of
-    * lines required:
-    */
-   switch (tObj->Target) {
-   case GL_TEXTURE_CUBE_MAP: {
-      const GLuint dim = tObj->Image[0][firstLevel]->Width;
-      GLuint face;
-
-      pitch = dim * t->intel.texelBytes;
-      pitch *= 2;		/* double pitch for cube layouts */
-      pitch = (pitch + 3) & ~3;
-      
-      total_height = dim * 4;
-
-      for ( face = 0 ; face < 6 ; face++) {
-	 GLuint x = initial_offsets[face][0] * dim;
-	 GLuint y = initial_offsets[face][1] * dim;
-	 GLuint d = dim;
-	 
-	 t->intel.base.dirty_images[face] = ~0;
-
-	 assert(tObj->Image[face][firstLevel]->Width == dim);
-	 assert(tObj->Image[face][firstLevel]->Height == dim);
-
-	 for (i = 0; i < numLevels; i++) {
-	    t->intel.image[face][i].image = tObj->Image[face][firstLevel + i];
-	    if (!t->intel.image[face][i].image) {
-	       fprintf(stderr, "no image %d %d\n", face, i);
-	       break;		/* can't happen */
-	    }
-	 
-	    t->intel.image[face][i].offset = 
-	       y * pitch + x * t->intel.texelBytes;
-	    t->intel.image[face][i].internalFormat = baseImage->_BaseFormat;
-
-	    d >>= 1;
-	    x += step_offsets[face][0] * d;
-	    y += step_offsets[face][1] * d;
-	 }
-      }
-      break;
-   }
-   case GL_TEXTURE_3D: {
-      GLuint virtual_height;
-      GLuint tmp_numLevels = numLevels;
-      pitch = tObj->Image[0][firstLevel]->Width * t->intel.texelBytes;
-      pitch = (pitch + 3) & ~3;
-      t->intel.base.dirty_images[0] = ~0;
-
-      /* Calculate the size of a single slice.  Hardware demands a
-       * minimum of 8 mipmaps, some of which might ultimately not be
-       * used:
-       */
-      if (tmp_numLevels < 9)
-	 tmp_numLevels = 9;
-
-      virtual_height = tObj->Image[0][firstLevel]->Height;
-
-      for ( total_height = i = 0 ; i < tmp_numLevels ; i++ ) {
-	 t->intel.image[0][i].image = tObj->Image[0][firstLevel + i];
-	 if (t->intel.image[0][i].image) {
-	    t->intel.image[0][i].offset = total_height * pitch;
-	    t->intel.image[0][i].internalFormat = baseImage->_BaseFormat;
-	 }
-
-	 total_height += MAX2(2, virtual_height);
-	 virtual_height >>= 1;
-      }
-
-      t->intel.depth_pitch = total_height * pitch;
-
-      /* Multiply slice size by texture depth for total size.  It's
-       * remarkable how wasteful of memory all the i8x0 texture
-       * layouts are.
-       */
-      total_height *= t->intel.image[0][0].image->Depth;
-      break;
-   }
-   default:
-      pitch = tObj->Image[0][firstLevel]->Width * t->intel.texelBytes;
-      pitch = (pitch + 3) & ~3;
-      t->intel.base.dirty_images[0] = ~0;
-
-      for ( total_height = i = 0 ; i < numLevels ; i++ ) {
-	 t->intel.image[0][i].image = tObj->Image[0][firstLevel + i];
-	 if (!t->intel.image[0][i].image) 
-	    break;
-	 
-	 t->intel.image[0][i].offset = total_height * pitch;
-	 t->intel.image[0][i].internalFormat = baseImage->_BaseFormat;
-	 if (t->intel.image[0][i].image->IsCompressed)
-	 {
-	   if (t->intel.image[0][i].image->Height > 4)
-	     total_height += t->intel.image[0][i].image->Height/4;
-	   else
-	     total_height += 1;
-	 }
-	 else
-	   total_height += MAX2(2, t->intel.image[0][i].image->Height);
-      }
-      break;
-   }
-
-   t->intel.Pitch = pitch;
-   t->intel.base.totalSize = total_height*pitch;
-   t->intel.max_level = numLevels-1;
-}
-
-
-static void i945LayoutTextureImages( i915ContextPtr i915,
-				    struct gl_texture_object *tObj )
-{
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   i915TextureObjectPtr t = (i915TextureObjectPtr) tObj->DriverData;
-   GLint firstLevel, lastLevel, numLevels;
-   GLint i, total_height, pitch, sz, max_offset = 0, offset;
-
-
-   /* Compute which mipmap levels we really want to send to the hardware.
-    */
-   driCalculateTextureFirstLastLevel( (driTextureObject *) t );
-
-   /* Figure out the amount of memory required to hold all the mipmap
-    * levels.  Choose the smallest pitch to accomodate the largest
-    * mipmap:
-    */
-   firstLevel = t->intel.base.firstLevel;
-   lastLevel = t->intel.base.lastLevel;
-   numLevels = lastLevel - firstLevel + 1;
-
-
-
-   /* All images must be loaded at this pitch.  Count the number of
-    * lines required:
-    */
-   switch (tObj->Target) {
-   case GL_TEXTURE_CUBE_MAP: {
-      const GLuint dim = tObj->Image[0][firstLevel]->Width;
-      GLuint face;
-
-      /* Depending on the size of the largest images, pitch can be
-       * determined either by the old-style packing of cubemap faces,
-       * or the final row of 4x4, 2x2 and 1x1 faces below this. 
-       */
-      if (dim > 32) {
-	 pitch = dim * t->intel.texelBytes;
-	 pitch *= 2;		/* double pitch for cube layouts */
-	 pitch = (pitch + 3) & ~3;
-      }
-      else {
-	 pitch = 14 * 8 * t->intel.texelBytes; /* determined by row of
-						* little maps at
-						* bottom */
-      }
-      
-      total_height = dim * 4 + 4;
-
-      for ( face = 0 ; face < 6 ; face++) {
-	 GLuint x = initial_offsets[face][0] * dim;
-	 GLuint y = initial_offsets[face][1] * dim;
-	 GLuint d = dim;
-	 
-	 if (dim == 4 && face >= 4) {
-	    y = total_height - 4;
-	    x = (face - 4) * 8;
-	 }
-	 else if (dim < 4) {
-	    y = total_height - 4;
-	    x = face * 8;
-	 }
-
-	 t->intel.base.dirty_images[face] = ~0;
-
-	 assert(tObj->Image[face][firstLevel]->Width == dim);
-	 assert(tObj->Image[face][firstLevel]->Height == dim);
-
-	 for (i = 0; i < numLevels; i++) {
-
-
-	    t->intel.image[face][i].image = tObj->Image[face][firstLevel + i];
-	    assert(t->intel.image[face][i].image);
-	 
-	    t->intel.image[face][i].offset = 
-	       y * pitch + x * t->intel.texelBytes;
-	    t->intel.image[face][i].internalFormat = baseImage->_BaseFormat;
-
-	    d >>= 1;
-	    
-	    switch (d) {
-	    case 4:
-	       switch (face) {
-	       case FACE_POS_X:
-	       case FACE_NEG_X:
-		  x += step_offsets[face][0] * d;
-		  y += step_offsets[face][1] * d;
-		  break;
-	       case FACE_POS_Y:
-	       case FACE_NEG_Y:
-		  y += 12;
-		  x -= 8;
-		  break;
-	       case FACE_POS_Z:
-	       case FACE_NEG_Z:
-		  y = total_height - 4;
-		  x = (face - 4) * 8;
-		  break;
-	       }
-
-	    case 2:
-	       y = total_height - 4;
-	       x = 16 + face * 8;
-	       break;
-
-	    case 1:
-	       x += 48;
-	       break;
-	       
-	    default:
-	       x += step_offsets[face][0] * d;
-	       y += step_offsets[face][1] * d;
-	       break;
-	    }
-	 }
-      }
-      max_offset = total_height * pitch;
-      break;
-   }
-   case GL_TEXTURE_3D: {
-      GLuint depth_packing = 0, depth_pack_pitch;
-      GLuint tmp_numLevels = numLevels;
-      pitch = tObj->Image[0][firstLevel]->Width * t->intel.texelBytes;
-      pitch = (pitch + 3) & ~3;
-      depth_pack_pitch = pitch;
-      
-      t->intel.base.dirty_images[0] = ~0;
-
-
-      for ( total_height = i = 0 ; i < tmp_numLevels ; i++ ) {
-	 t->intel.image[0][i].image = tObj->Image[0][firstLevel + i];
-	 if (!t->intel.image[0][i].image) 
-	    break;
-
-	 
-	 t->intel.image[0][i].offset = total_height * pitch;
-	 t->intel.image[0][i].internalFormat = baseImage->_BaseFormat;
-	 
-
 
-	 total_height += MAX2(2, t->intel.image[0][i].image->Height) * 
-	    MAX2((t->intel.image[0][i].image->Depth >> depth_packing), 1);
-
-	 /* When alignment dominates, can't increase depth packing?
-	  * Or does pitch grow???  What are the alignment constraints,
-	  * anyway?
-	  */
-	 if (depth_pack_pitch > 4) {
-	    depth_packing++;
-	    depth_pack_pitch <<= 2;
-	 }
-      }
-
-      max_offset = total_height * pitch;
-      break;
-   }
-   default:
-      pitch = tObj->Image[0][firstLevel]->Width * t->intel.texelBytes;
-      pitch = (pitch + 3) & ~3;
-      t->intel.base.dirty_images[0] = ~0;
-      max_offset = 0;
-
-      for ( offset = i = 0 ; i < numLevels ; i++ ) {
-	 t->intel.image[0][i].image = tObj->Image[0][firstLevel + i];
-	 if (!t->intel.image[0][i].image) 
-	    break;
-	 
-	 t->intel.image[0][i].offset = offset;
-	 t->intel.image[0][i].internalFormat = baseImage->_BaseFormat;
-
-	 if (t->intel.image[0][i].image->IsCompressed)
-	    sz = MAX2(1, t->intel.image[0][i].image->Height/4) * pitch;
-	 else
-	    sz = MAX2(2, t->intel.image[0][i].image->Height) * pitch;
-	 
-	 /* Because the images are packed better, the final offset
-	  * might not be the maximal one:
-	  */
-	 max_offset = MAX2(max_offset, offset + sz);
-
-	 /* LPT change: step right after second mipmap.
-	  */
-	 if (i == 1) 
-	    offset += pitch / 2;
-	 else 
-	    offset += sz;
-
-      }
-      break;
-   }
-
-   t->intel.Pitch = pitch;
-   t->intel.base.totalSize = max_offset;
-   t->intel.max_level = numLevels-1;
-}
-
-
-
-
-static void i915SetTexImages( i915ContextPtr i915, 
-			     struct gl_texture_object *tObj )
+static GLuint translate_texture_format( GLuint mesa_format )
 {
-   GLuint textureFormat;
-   i915TextureObjectPtr t = (i915TextureObjectPtr) tObj->DriverData;
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   GLint ss2 = 0;
-
-   switch( baseImage->TexFormat->MesaFormat ) {
+   switch (mesa_format) {
    case MESA_FORMAT_L8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_L8;
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_L8;
    case MESA_FORMAT_I8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_I8;
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_I8;
    case MESA_FORMAT_A8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_A8; 
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_A8; 
    case MESA_FORMAT_AL88:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_AY88;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_AY88;
    case MESA_FORMAT_RGB565:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_RGB565;
    case MESA_FORMAT_ARGB1555:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB1555;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_ARGB1555;
    case MESA_FORMAT_ARGB4444:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB4444;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_ARGB4444;
    case MESA_FORMAT_ARGB8888:
-      t->intel.texelBytes = 4;
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-      break;
-
+      return MAPSURF_32BIT | MT_32BIT_ARGB8888;
    case MESA_FORMAT_YCBCR_REV:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_NORMAL);
-      ss2 |= SS2_COLORSPACE_CONVERSION;
-      break;
-
+      return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
    case MESA_FORMAT_YCBCR:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_SWAPY);
-      ss2 |= SS2_COLORSPACE_CONVERSION;
-      break;
-
+      return (MAPSURF_422 | MT_422_YCRCB_SWAPY);
    case MESA_FORMAT_RGB_FXT1:
    case MESA_FORMAT_RGBA_FXT1:
-     t->intel.texelBytes = 2;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_FXT1);
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_FXT1);
    case MESA_FORMAT_Z16:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_16BIT | MT_16BIT_L16);
-      break;
-
+      return (MAPSURF_16BIT | MT_16BIT_L16);
    case MESA_FORMAT_RGBA_DXT1:
    case MESA_FORMAT_RGB_DXT1:
-     /* 
-      * DXTn pitches are Width/4 * blocksize in bytes 
-      * for DXT1: blocksize=8 so Width/4*8 = Width * 2 
-      * for DXT3/5: blocksize=16 so Width/4*16 = Width * 4
-      */
-     t->intel.texelBytes = 2;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
    case MESA_FORMAT_RGBA_DXT3:
-     t->intel.texelBytes = 4;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
    case MESA_FORMAT_RGBA_DXT5:
-     t->intel.texelBytes = 4;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
-     break;
-
-#if 0
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
    case MESA_FORMAT_Z24_S8:
-      t->intel.texelBytes = 4;
-      textureFormat = (MAPSURF_32BIT | MT_32BIT_xL824);
-      break;
-#endif
-
+      return (MAPSURF_32BIT | MT_32BIT_xL824);
    default:
       fprintf(stderr, "%s: bad image format %x\n", __FUNCTION__,
-	      baseImage->TexFormat->MesaFormat);
+	      mesa_format);
       abort();
+      return 0;
    }
+}
 
 
-   if (i915->intel.intelScreen->deviceID == PCI_CHIP_I945_G ||
-       i915->intel.intelScreen->deviceID == PCI_CHIP_I945_GM)
-      i945LayoutTextureImages( i915, tObj );	 
-   else
-      i915LayoutTextureImages( i915, tObj );
-
-   t->Setup[I915_TEXREG_MS3] = 
-      (((tObj->Image[0][t->intel.base.firstLevel]->Height - 1) << MS3_HEIGHT_SHIFT) |
-       ((tObj->Image[0][t->intel.base.firstLevel]->Width - 1) << MS3_WIDTH_SHIFT) |
-       textureFormat |
-       MS3_USE_FENCE_REGS);
-
-   t->Setup[I915_TEXREG_MS4] = 
-      ((((t->intel.Pitch / 4) - 1) << MS4_PITCH_SHIFT) | 
-       MS4_CUBE_FACE_ENA_MASK |
-       (((t->intel.max_level * 4)) << MS4_MAX_LOD_SHIFT) |
-       ((tObj->Image[0][t->intel.base.firstLevel]->Depth - 1) << MS4_VOLUME_DEPTH_SHIFT));
-
-   t->Setup[I915_TEXREG_SS2] &= ~(SS2_COLORSPACE_CONVERSION);
-   t->Setup[I915_TEXREG_SS2] |= ss2;
-
-   t->intel.dirty = I915_UPLOAD_TEX_ALL;
-
-}
 
 
 /* The i915 (and related graphics cores) do not support GL_CLAMP.  The
@@ -529,391 +91,235 @@ static void i915SetTexImages( i915ContextPtr i915,
 static GLuint translate_wrap_mode( GLenum wrap )
 {
    switch( wrap ) {
-   case GL_REPEAT: return TEXCOORDMODE_WRAP;
-   case GL_CLAMP:  return TEXCOORDMODE_CLAMP_EDGE; /* not quite correct */
-   case GL_CLAMP_TO_EDGE: return TEXCOORDMODE_CLAMP_EDGE;
-   case GL_CLAMP_TO_BORDER: return TEXCOORDMODE_CLAMP_BORDER;
-   case GL_MIRRORED_REPEAT: return TEXCOORDMODE_MIRROR;
-   default: return TEXCOORDMODE_WRAP;
+   case GL_REPEAT: 
+      return TEXCOORDMODE_WRAP;
+   case GL_CLAMP:  
+      return TEXCOORDMODE_CLAMP_EDGE; /* not quite correct */
+   case GL_CLAMP_TO_EDGE: 
+      return TEXCOORDMODE_CLAMP_EDGE;
+   case GL_CLAMP_TO_BORDER: 
+      return TEXCOORDMODE_CLAMP_BORDER;
+   case GL_MIRRORED_REPEAT: 
+      return TEXCOORDMODE_MIRROR;
+   default: 
+      return TEXCOORDMODE_WRAP;
    }
 }
 
 
-/**
- */
-static void i915ImportTexObjState( struct gl_texture_object *texObj )
-{   
-   i915TextureObjectPtr t = (i915TextureObjectPtr)texObj->DriverData;
-   int minFilt = 0, mipFilt = 0, magFilt = 0, shadow = 0;
-
-   if(INTEL_DEBUG&DEBUG_DRI)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   switch (texObj->MinFilter) {
-   case GL_NEAREST:
-      minFilt = FILTER_NEAREST;
-      mipFilt = MIPFILTER_NONE;
-      break;
-   case GL_LINEAR:
-      minFilt = FILTER_LINEAR;
-      mipFilt = MIPFILTER_NONE;
-      break;
-   case GL_NEAREST_MIPMAP_NEAREST:
-      minFilt = FILTER_NEAREST;
-      mipFilt = MIPFILTER_NEAREST;
-      break;
-   case GL_LINEAR_MIPMAP_NEAREST:
-      minFilt = FILTER_LINEAR;
-      mipFilt = MIPFILTER_NEAREST;
-      break;
-   case GL_NEAREST_MIPMAP_LINEAR:
-      minFilt = FILTER_NEAREST;
-      mipFilt = MIPFILTER_LINEAR;
-      break;
-   case GL_LINEAR_MIPMAP_LINEAR:
-      minFilt = FILTER_LINEAR;
-      mipFilt = MIPFILTER_LINEAR;
-      break;
-   default:
-      break;
-   }
-
-   if ( texObj->MaxAnisotropy > 1.0 ) {
-      minFilt = FILTER_ANISOTROPIC; 
-      magFilt = FILTER_ANISOTROPIC;
-   }
-   else {
-      switch (texObj->MagFilter) {
-      case GL_NEAREST:
-	 magFilt = FILTER_NEAREST;
-	 break;
-      case GL_LINEAR:
-	 magFilt = FILTER_LINEAR;
-	 break;
-      default:
-	 break;
-      }  
-   }
-
-   if (texObj->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB && 
-       texObj->Target != GL_TEXTURE_3D) {
-
-      shadow = SS2_SHADOW_ENABLE;
-      shadow |= intel_translate_compare_func( texObj->CompareFunc );
-      
-      minFilt = FILTER_4X4_FLAT;
-      magFilt = FILTER_4X4_FLAT;
-   }
 
-
-   t->Setup[I915_TEXREG_SS2] &= ~(SS2_MIN_FILTER_MASK |
-				 SS2_MIP_FILTER_MASK |
-				 SS2_MAG_FILTER_MASK |
-				 SS2_SHADOW_ENABLE |
-				 SS2_SHADOW_FUNC_MASK);
-   t->Setup[I915_TEXREG_SS2] |= ((minFilt << SS2_MIN_FILTER_SHIFT) |
-				(mipFilt << SS2_MIP_FILTER_SHIFT) |
-				(magFilt << SS2_MAG_FILTER_SHIFT) |
-				shadow);
-
-   {
-      GLuint ss3 = t->Setup[I915_TEXREG_SS3] & ~(SS3_TCX_ADDR_MODE_MASK |
-						SS3_TCY_ADDR_MODE_MASK |
-						SS3_TCZ_ADDR_MODE_MASK);
-      GLenum ws = texObj->WrapS;
-      GLenum wt = texObj->WrapT;
-      GLenum wr = texObj->WrapR;
-      
-      t->refs_border_color = 0;
-
-      if (texObj->Target == GL_TEXTURE_3D &&
-	  (texObj->MinFilter != GL_NEAREST ||
-	   texObj->MagFilter != GL_NEAREST)) {
-	 
-	 /* Try to mimic GL_CLAMP functionality a little better -
-	  * switch to CLAMP_TO_BORDER whenever a non-NEAREST filter is
-	  * in use.  Only do this for 3D textures at the moment --
-	  * doing it universally would fix the conform texbc.c
-	  * failure, though.
-	  */
-	 if (ws == GL_CLAMP) ws = GL_CLAMP_TO_BORDER;
-	 if (wt == GL_CLAMP) wt = GL_CLAMP_TO_BORDER;
-	 if (wr == GL_CLAMP) wr = GL_CLAMP_TO_BORDER;
-
-	 /* 3D textures don't seem to respect the border color.
-	  * Fallback if there's ever a danger that they might refer to
-	  * it.
-	  */
-	 if (ws == GL_CLAMP_TO_BORDER) t->refs_border_color = 1;
-	 if (wt == GL_CLAMP_TO_BORDER) t->refs_border_color = 1;
-	 if (wr == GL_CLAMP_TO_BORDER) t->refs_border_color = 1;
-      }
-
-      ss3 |= translate_wrap_mode(ws) << SS3_TCX_ADDR_MODE_SHIFT;
-      ss3 |= translate_wrap_mode(wt) << SS3_TCY_ADDR_MODE_SHIFT;
-      ss3 |= translate_wrap_mode(wr) << SS3_TCZ_ADDR_MODE_SHIFT;
-   
-      if (ss3 != t->Setup[I915_TEXREG_SS3]) {
-	 t->intel.dirty = I915_UPLOAD_TEX_ALL;
-	 t->Setup[I915_TEXREG_SS3] = ss3;
-      }
-   }
-
-   {   
-      const GLubyte *color = texObj->_BorderChan;
-
-      t->Setup[I915_TEXREG_SS4] = INTEL_PACKCOLOR8888(color[0],color[1],
-						     color[2],color[3]);
-   }
-}
-
-
-
-static void i915_import_tex_unit( i915ContextPtr i915, 
-				 i915TextureObjectPtr t,
-				 GLuint unit )
+/* Recalculate all state from scratch.  Perhaps not the most
+ * efficient, but this has gotten complex enough that we need
+ * something which is understandable and reliable.
+ */
+static GLboolean i915_update_tex_unit( struct intel_context *intel,
+				       GLuint unit, 
+				       GLuint ss3 )
 {
-   GLuint state[I915_TEX_SETUP_SIZE];
+   GLcontext *ctx = &intel->ctx;
+   struct i915_context *i915 = i915_context(ctx);
+   struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
+   struct intel_texture_object *intelObj = intel_texture_object(tObj);
+   struct gl_texture_image *firstImage;
+   GLuint *state = i915->state.Tex[unit];
 
-   if(INTEL_DEBUG&DEBUG_TEXTURE)
-      fprintf(stderr, "%s unit(%d)\n", __FUNCTION__, unit);
-   
-   if (i915->intel.CurrentTexObj[unit]) 
-      i915->intel.CurrentTexObj[unit]->base.bound &= ~(1U << unit);
+   memset(state, 0, sizeof(state));
 
-   i915->intel.CurrentTexObj[unit] = (intelTextureObjectPtr)t;
-   t->intel.base.bound |= (1 << unit);
+/*    intel_region_release(intel, &i915->state.tex_region[unit]); */
 
-   if (t->intel.dirty & I915_UPLOAD_TEX(unit)) {
-      i915ImportTexObjState( t->intel.base.tObj );
-      t->intel.dirty &= ~I915_UPLOAD_TEX(unit);
-   }
+   if (!intel_finalize_mipmap_tree(intel, unit))
+      return GL_FALSE;   
 
-   state[I915_TEXREG_MS2] = t->intel.TextureOffset;
-   state[I915_TEXREG_MS3] = t->Setup[I915_TEXREG_MS3];
-   state[I915_TEXREG_MS4] = t->Setup[I915_TEXREG_MS4];
+   /* Get first image here, since intelObj->firstLevel will get set in
+    * the intel_finalize_mipmap_tree() call above.
+    */
+   firstImage = tObj->Image[0][intelObj->firstLevel];
 
-   state[I915_TEXREG_SS2] = (i915->state.Tex[unit][I915_TEXREG_SS2] &
-			    SS2_LOD_BIAS_MASK);
-   state[I915_TEXREG_SS2] |= (t->Setup[I915_TEXREG_SS2] & ~SS2_LOD_BIAS_MASK);
+/*    intel_region_reference(&i915->state.tex_region[unit], */
+/* 			  intelObj->mt->region); */
 
-   state[I915_TEXREG_SS3] = (i915->state.Tex[unit][I915_TEXREG_SS3] &
-			    SS3_NORMALIZED_COORDS);
-   state[I915_TEXREG_SS3] |= (t->Setup[I915_TEXREG_SS3] &
-			     ~(SS3_NORMALIZED_COORDS|
-			       SS3_TEXTUREMAP_INDEX_MASK));
+   i915->state.tex_buffer[unit] = intelObj->mt->region->buffer;
+   i915->state.tex_offset[unit] = intel_miptree_image_offset(intelObj->mt, 0,
+							     intelObj->firstLevel); 
 
-   state[I915_TEXREG_SS3] |= (unit<<SS3_TEXTUREMAP_INDEX_SHIFT);
 
-   state[I915_TEXREG_SS4] = t->Setup[I915_TEXREG_SS4];
+   state[I915_TEXREG_MS3] = 
+      (((firstImage->Height - 1) << MS3_HEIGHT_SHIFT) |
+       ((firstImage->Width - 1) << MS3_WIDTH_SHIFT) |
+       translate_texture_format( firstImage->TexFormat->MesaFormat ) |
+       MS3_USE_FENCE_REGS);
 
+   state[I915_TEXREG_MS4] = 
+      (((((intelObj->mt->pitch * intelObj->mt->cpp) / 4) - 1) << MS4_PITCH_SHIFT) | 
+       MS4_CUBE_FACE_ENA_MASK |
+       ((((intelObj->lastLevel - intelObj->firstLevel) * 4)) << MS4_MAX_LOD_SHIFT) |
+       ((firstImage->Depth - 1) << MS4_VOLUME_DEPTH_SHIFT));
 
-   if (memcmp(state, i915->state.Tex[unit], sizeof(state)) != 0) {
-      I915_STATECHANGE( i915, I915_UPLOAD_TEX(unit) );
-      memcpy(i915->state.Tex[unit], state, sizeof(state));
-   }
-}
 
+   {
+      GLuint minFilt, mipFilt, magFilt;
 
+      switch (tObj->MinFilter) {
+      case GL_NEAREST:
+	 minFilt = FILTER_NEAREST;
+	 mipFilt = MIPFILTER_NONE;
+	 break;
+      case GL_LINEAR:
+	 minFilt = FILTER_LINEAR;
+	 mipFilt = MIPFILTER_NONE;
+	 break;
+      case GL_NEAREST_MIPMAP_NEAREST:
+	 minFilt = FILTER_NEAREST;
+	 mipFilt = MIPFILTER_NEAREST;
+	 break;
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 minFilt = FILTER_LINEAR;
+	 mipFilt = MIPFILTER_NEAREST;
+	 break;
+      case GL_NEAREST_MIPMAP_LINEAR:
+	 minFilt = FILTER_NEAREST;
+	 mipFilt = MIPFILTER_LINEAR;
+	 break;
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 minFilt = FILTER_LINEAR;
+	 mipFilt = MIPFILTER_LINEAR;
+	 break;
+      default:
+	 return GL_FALSE;
+      }
 
-static GLboolean enable_tex_common( GLcontext *ctx, GLuint unit )
-{
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i915TextureObjectPtr t = (i915TextureObjectPtr)tObj->DriverData;
+      if ( tObj->MaxAnisotropy > 1.0 ) {
+	 minFilt = FILTER_ANISOTROPIC; 
+	 magFilt = FILTER_ANISOTROPIC;
+      }
+      else {
+	 switch (tObj->MagFilter) {
+	 case GL_NEAREST:
+	    magFilt = FILTER_NEAREST;
+	    break;
+	 case GL_LINEAR:
+	    magFilt = FILTER_LINEAR;
+	    break;
+	 default:
+	    return GL_FALSE;
+	 }  
+      }
 
-   if (0) fprintf(stderr, "%s %d\n", __FUNCTION__, unit);
+      state[I915_TEXREG_SS2] = i915->lodbias_ss2[unit];
 
-   if (!(i915->state.active & I915_UPLOAD_TEX(unit))) {
-      I915_ACTIVESTATE(i915, I915_UPLOAD_TEX(unit), GL_TRUE);
-   }
+      /* YUV conversion:
+       */
+      if (firstImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR ||
+	  firstImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR_REV)
+	 state[I915_TEXREG_SS2] |= SS2_COLORSPACE_CONVERSION;
 
-   /* Fallback if there's a texture border */
-   if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 ) {
-      return GL_FALSE;
-   }
+      /* Shadow:
+       */
+      if (tObj->CompareMode == GL_COMPARE_R_TO_TEXTURE_ARB && 
+	  tObj->Target != GL_TEXTURE_3D) {
 
+	 state[I915_TEXREG_SS2] |= 
+	    (SS2_SHADOW_ENABLE |
+	     intel_translate_compare_func( tObj->CompareFunc ));
+      
+	 minFilt = FILTER_4X4_FLAT;
+	 magFilt = FILTER_4X4_FLAT;
+      }
 
-   /* Update state if this is a different texture object to last
-    * time.
-    */
-   if (i915->intel.CurrentTexObj[unit] != &t->intel || 
-       (t->intel.dirty & I915_UPLOAD_TEX(unit))) {
-      i915_import_tex_unit( i915, t, unit);
-      i915->tex_program.translated = 0;
+      state[I915_TEXREG_SS2] |= ((minFilt << SS2_MIN_FILTER_SHIFT) |
+				 (mipFilt << SS2_MIP_FILTER_SHIFT) |
+				 (magFilt << SS2_MAG_FILTER_SHIFT));
    }
 
-   return GL_TRUE;
-}
-
-static GLboolean enable_tex_rect( GLcontext *ctx, GLuint unit )
-{
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i915TextureObjectPtr t = (i915TextureObjectPtr)tObj->DriverData;
-   GLuint ss3 = i915->state.Tex[unit][I915_TEXREG_SS3];
-
-   ss3 &= ~SS3_NORMALIZED_COORDS;
-
-   if (ss3 != i915->state.Tex[unit][I915_TEXREG_SS3]) {
-      I915_STATECHANGE(i915, I915_UPLOAD_TEX(unit));
-      i915->state.Tex[unit][I915_TEXREG_SS3] = ss3;
-   }
+   {
+      GLenum ws = tObj->WrapS;
+      GLenum wt = tObj->WrapT;
+      GLenum wr = tObj->WrapR;
+      
 
-   /* Upload teximages (not pipelined)
-    */
-   if (t->intel.base.dirty_images[0]) {
-      i915SetTexImages( i915, tObj );
-      if (!intelUploadTexImages( &i915->intel, &t->intel, 0 )) {
+      /* 3D textures don't seem to respect the border color.
+       * Fallback if there's ever a danger that they might refer to
+       * it.  
+       * 
+       * Effectively this means fallback on 3D clamp or
+       * clamp_to_border.
+       */
+      if (tObj->Target == GL_TEXTURE_3D &&
+	  (tObj->MinFilter != GL_NEAREST ||
+	   tObj->MagFilter != GL_NEAREST) &&	 
+	  (ws == GL_CLAMP ||
+	   wt == GL_CLAMP ||
+	   wr == GL_CLAMP ||
+	   ws == GL_CLAMP_TO_BORDER ||
+	   wt == GL_CLAMP_TO_BORDER ||
+	   wr == GL_CLAMP_TO_BORDER))
 	 return GL_FALSE;
-      }
-   }
+	     
 
-   return GL_TRUE;
-}
+      state[I915_TEXREG_SS3] = ss3; /* SS3_NORMALIZED_COORDS */
 
+      state[I915_TEXREG_SS3] |= ((translate_wrap_mode(ws) << SS3_TCX_ADDR_MODE_SHIFT) |
+				(translate_wrap_mode(wt) << SS3_TCY_ADDR_MODE_SHIFT) |
+				(translate_wrap_mode(wr) << SS3_TCZ_ADDR_MODE_SHIFT));
 
-static GLboolean enable_tex_2d( GLcontext *ctx, GLuint unit )
-{
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i915TextureObjectPtr t = (i915TextureObjectPtr)tObj->DriverData;
-   GLuint ss3 = i915->state.Tex[unit][I915_TEXREG_SS3];
-
-   ss3 |= SS3_NORMALIZED_COORDS;
-
-   if (ss3 != i915->state.Tex[unit][I915_TEXREG_SS3]) {
-      I915_STATECHANGE(i915, I915_UPLOAD_TEX(unit));
-      i915->state.Tex[unit][I915_TEXREG_SS3] = ss3;
+      state[I915_TEXREG_SS3] |= (unit << SS3_TEXTUREMAP_INDEX_SHIFT);
    }
 
-   /* Upload teximages (not pipelined)
-    */
-   if (t->intel.base.dirty_images[0]) {
-      i915SetTexImages( i915, tObj );
-      if (!intelUploadTexImages( &i915->intel, &t->intel, 0 )) {
-	 return GL_FALSE;
-      }
-   }
 
-   return GL_TRUE;
-}
-
-static GLboolean enable_tex_cube( GLcontext *ctx, GLuint unit )
-{
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i915TextureObjectPtr t = (i915TextureObjectPtr)tObj->DriverData;
-   GLuint ss3 = i915->state.Tex[unit][I915_TEXREG_SS3];
-   GLuint face;
-
-   ss3 |= SS3_NORMALIZED_COORDS;
-
-   if (ss3 != i915->state.Tex[unit][I915_TEXREG_SS3]) {
-      I915_STATECHANGE(i915, I915_UPLOAD_TEX(unit));
-      i915->state.Tex[unit][I915_TEXREG_SS3] = ss3;
-   }
+   state[I915_TEXREG_SS4] = INTEL_PACKCOLOR8888(tObj->_BorderChan[0],
+						tObj->_BorderChan[1],
+						tObj->_BorderChan[2],
+						tObj->_BorderChan[3]);
+   
 
-   /* Upload teximages (not pipelined)
+   I915_ACTIVESTATE(i915, I915_UPLOAD_TEX(unit), GL_TRUE);
+   /* memcmp was already disabled, but definitely won't work as the
+    * region might now change and that wouldn't be detected:
     */
-   if ( t->intel.base.dirty_images[0] || t->intel.base.dirty_images[1] ||
-        t->intel.base.dirty_images[2] || t->intel.base.dirty_images[3] ||
-        t->intel.base.dirty_images[4] || t->intel.base.dirty_images[5] ) {
-      i915SetTexImages( i915, tObj );
-   }
-
-   /* upload (per face) */
-   for (face = 0; face < 6; face++) {
-      if (t->intel.base.dirty_images[face]) {
-	 if (!intelUploadTexImages( &i915->intel, &t->intel, face )) {
-	    return GL_FALSE;
-	 }
-      }
-   }
-
-
-   return GL_TRUE;
-}
+			  I915_STATECHANGE( i915, I915_UPLOAD_TEX(unit) );
 
-static GLboolean enable_tex_3d( GLcontext *ctx, GLuint unit )
-{
-   struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
-   i915TextureObjectPtr t = (i915TextureObjectPtr)tObj->DriverData;
 
-   /* 3D textures on I915 seem to get bogus border colors, hence this
-    * fallback:
-    */
-   if (t->refs_border_color)
-      return GL_FALSE;
+#if 0
+   DBG(TEXTURE, "state[I915_TEXREG_SS2] = 0x%x\n", state[I915_TEXREG_SS2]);
+   DBG(TEXTURE, "state[I915_TEXREG_SS3] = 0x%x\n", state[I915_TEXREG_SS3]);
+   DBG(TEXTURE, "state[I915_TEXREG_SS4] = 0x%x\n", state[I915_TEXREG_SS4]);
+   DBG(TEXTURE, "state[I915_TEXREG_MS2] = 0x%x\n", state[I915_TEXREG_MS2]);
+   DBG(TEXTURE, "state[I915_TEXREG_MS3] = 0x%x\n", state[I915_TEXREG_MS3]);
+   DBG(TEXTURE, "state[I915_TEXREG_MS4] = 0x%x\n", state[I915_TEXREG_MS4]);
+#endif
 
    return GL_TRUE;
 }
 
 
 
- 
-static GLboolean disable_tex( GLcontext *ctx, GLuint unit )
-{
-   i915ContextPtr i915 = I915_CONTEXT(ctx);
-
-   if (i915->state.active & I915_UPLOAD_TEX(unit)) {
-      I915_ACTIVESTATE(i915, I915_UPLOAD_TEX(unit), GL_FALSE);
-   }
-
-   /* The old texture is no longer bound to this texture unit.
-    * Mark it as such.
-    */
-   if ( i915->intel.CurrentTexObj[unit] != NULL ) {
-      i915->intel.CurrentTexObj[unit]->base.bound &= ~(1U << 0);
-      i915->intel.CurrentTexObj[unit] = NULL;
-   }
-
-   return GL_TRUE;
-}
 
-static GLboolean i915UpdateTexUnit( GLcontext *ctx, GLuint unit )
+void i915UpdateTextureState( struct intel_context *intel )
 {
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-
-   if (texUnit->_ReallyEnabled &&
-       INTEL_CONTEXT(ctx)->intelScreen->tex.size < 2048 * 1024)
-      return GL_FALSE;
-
-   switch (texUnit->_ReallyEnabled) {
-   case TEXTURE_1D_BIT:
-   case TEXTURE_2D_BIT:
-      return (enable_tex_2d( ctx, unit ) &&
-	      enable_tex_common( ctx, unit ));
-   case TEXTURE_RECT_BIT:
-      return (enable_tex_rect( ctx, unit ) &&
-	      enable_tex_common( ctx, unit ));
-   case TEXTURE_CUBE_BIT:
-      return (enable_tex_cube( ctx, unit ) &&
-	      enable_tex_common( ctx, unit ));
-   case TEXTURE_3D_BIT:
-       return (enable_tex_2d( ctx, unit ) && 
-	       enable_tex_common( ctx, unit ) &&
-	       enable_tex_3d( ctx, unit)); 
-   case 0:
-      return disable_tex( ctx, unit );
-   default:
-      return GL_FALSE;
-   }
-}
-
-
-void i915UpdateTextureState( intelContextPtr intel )
-{
-   GLcontext *ctx = &intel->ctx;
    GLboolean ok = GL_TRUE;
    GLuint i;
 
    for (i = 0 ; i < I915_TEX_UNITS && ok ; i++) {
-      ok = i915UpdateTexUnit( ctx, i );
+      switch (intel->ctx.Texture.Unit[i]._ReallyEnabled) {
+      case TEXTURE_1D_BIT:
+      case TEXTURE_2D_BIT:
+      case TEXTURE_CUBE_BIT:
+      case TEXTURE_3D_BIT:
+	 ok = i915_update_tex_unit( intel, i, SS3_NORMALIZED_COORDS );
+	 break;
+      case TEXTURE_RECT_BIT:
+	 ok = i915_update_tex_unit( intel, i, 0 );
+	 break;
+      case 0: {
+	 struct i915_context *i915 = i915_context(&intel->ctx);
+	 if (i915->state.active & I915_UPLOAD_TEX(i)) 
+	    I915_ACTIVESTATE(i915, I915_UPLOAD_TEX(i), GL_FALSE);
+	 break;
+      }
+      default:
+	 ok = GL_FALSE;
+	 break;
+      }
    }
 
    FALLBACK( intel, I915_FALLBACK_TEXTURE, !ok );
diff --git a/src/mesa/drivers/dri/i915/i915_vtbl.c b/src/mesa/drivers/dri/i915/i915_vtbl.c
index 25575e5ed2..d3e459a556 100644
--- a/src/mesa/drivers/dri/i915/i915_vtbl.c
+++ b/src/mesa/drivers/dri/i915/i915_vtbl.c
@@ -37,14 +37,16 @@
 #include "tnl/t_vertex.h"
 
 #include "intel_batchbuffer.h"
+#include "intel_tex.h"
+#include "intel_regions.h"
 
 #include "i915_reg.h"
 #include "i915_context.h"
 
-static void i915_render_start( intelContextPtr intel )
+static void i915_render_start( struct intel_context *intel )
 {
    GLcontext *ctx = &intel->ctx;
-   i915ContextPtr i915 = I915_CONTEXT(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
 
    if (ctx->FragmentProgram._Active) 
       i915ValidateFragmentProgram( i915 );
@@ -53,42 +55,42 @@ static void i915_render_start( intelContextPtr intel )
 }
 
 
-static void i915_reduced_primitive_state( intelContextPtr intel,
+static void i915_reduced_primitive_state( struct intel_context *intel,
 					  GLenum rprim )
 {
-    i915ContextPtr i915 = I915_CONTEXT(intel);
-    GLuint st1 = i915->state.Stipple[I915_STPREG_ST1];
-
-    st1 &= ~ST1_ENABLE;
-
-    switch (rprim) {
-    case GL_TRIANGLES:
-       if (intel->ctx.Polygon.StippleFlag &&
-	   intel->hw_stipple)
-	  st1 |= ST1_ENABLE;
-       break;
-    case GL_LINES:
-    case GL_POINTS:
-    default:
-       break;
-    }
-
-    i915->intel.reduced_primitive = rprim;
-
-    if (st1 != i915->state.Stipple[I915_STPREG_ST1]) {
-       I915_STATECHANGE(i915, I915_UPLOAD_STIPPLE);
-       i915->state.Stipple[I915_STPREG_ST1] = st1;
-    }
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   GLuint st1 = i915->state.Stipple[I915_STPREG_ST1];
+
+   st1 &= ~ST1_ENABLE;
+
+   switch (rprim) {
+   case GL_TRIANGLES:
+      if (intel->ctx.Polygon.StippleFlag &&
+	  intel->hw_stipple)
+	 st1 |= ST1_ENABLE;
+      break;
+   case GL_LINES:
+   case GL_POINTS:
+   default:
+      break;
+   }
+
+   i915->intel.reduced_primitive = rprim;
+
+   if (st1 != i915->state.Stipple[I915_STPREG_ST1]) {
+      I915_STATECHANGE(i915, I915_UPLOAD_STIPPLE);
+      i915->state.Stipple[I915_STPREG_ST1] = st1;
+   }
 }
 
 
 /* Pull apart the vertex format registers and figure out how large a
  * vertex is supposed to be. 
  */
-static GLboolean i915_check_vertex_size( intelContextPtr intel,
+static GLboolean i915_check_vertex_size( struct intel_context *intel,
 					 GLuint expected )
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
    int lis2 = i915->current->Ctx[I915_CTXREG_LIS2];
    int lis4 = i915->current->Ctx[I915_CTXREG_LIS4];
    int i, sz = 0;
@@ -132,11 +134,11 @@ static GLboolean i915_check_vertex_size( intelContextPtr intel,
 }
 
 
-static void i915_emit_invarient_state( intelContextPtr intel )
+static void i915_emit_invarient_state( struct intel_context *intel )
 {
    BATCH_LOCALS;
 
-   BEGIN_BATCH( 200 );
+   BEGIN_BATCH( 200, 0 );
 
    OUT_BATCH(_3DSTATE_AA_CMD |
 	     AA_LINE_ECAAR_WIDTH_ENABLE |
@@ -204,14 +206,8 @@ static void i915_emit_invarient_state( intelContextPtr intel )
 }
 
 
-#define emit( intel, state, size )			\
-do {							\
-   int k;						\
-   BEGIN_BATCH( (size) / sizeof(GLuint));		\
-   for (k = 0 ; k < (size) / sizeof(GLuint) ; k++)	\
-      OUT_BATCH((state)[k]);				\
-   ADVANCE_BATCH();					\
-} while (0);
+#define emit(intel, state, size )		     \
+   intel_batchbuffer_data(intel->batch, state, size, 0 )
 
 static GLuint get_dirty( struct i915_hw_state *state )
 {
@@ -268,42 +264,70 @@ static GLuint get_state_size( struct i915_hw_state *state )
 
 /* Push the state into the sarea and/or texture memory.
  */
-static void i915_emit_state( intelContextPtr intel )
+static void i915_emit_state( struct intel_context *intel )
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
    struct i915_hw_state *state = i915->current;
    int i;
    GLuint dirty = get_dirty(state);
-   GLuint counter = intel->batch.counter;
    BATCH_LOCALS;
 
-   if (intel->batch.space < get_state_size(state)) {
-      intelFlushBatch(intel, GL_TRUE);
-      dirty = get_dirty(state);
-      counter = intel->batch.counter;
-   }
+   /* We don't hold the lock at this point, so want to make sure that
+    * there won't be a buffer wrap.  
+    *
+    * It might be better to talk about explicit places where
+    * scheduling is allowed, rather than assume that it is whenever a
+    * batchbuffer fills up.
+    */
+   intel_batchbuffer_require_space(intel->batch, 
+				   get_state_size(state),
+				   0);
 
-   if (VERBOSE) 
+   if (INTEL_DEBUG & DEBUG_STATE) 
       fprintf(stderr, "%s dirty: %x\n", __FUNCTION__, dirty);
 
+   if (dirty & I915_UPLOAD_INVARIENT) {
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_INVARIENT:\n"); 
+      i915_emit_invarient_state( intel );
+   }
+
    if (dirty & I915_UPLOAD_CTX) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_CTX:\n"); 
-      emit( i915, state->Ctx, sizeof(state->Ctx) );
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_CTX:\n"); 
+      emit(intel, state->Ctx, sizeof(state->Ctx) );
    }
 
    if (dirty & I915_UPLOAD_BUFFERS) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_BUFFERS:\n"); 
-      emit( i915, state->Buffer, sizeof(state->Buffer) );
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_BUFFERS:\n"); 
+      BEGIN_BATCH(I915_DEST_SETUP_SIZE+2, 0);
+      OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
+      OUT_RELOC(state->draw_region->buffer, DRM_MM_TT|DRM_MM_WRITE,
+                state->draw_region->draw_offset);
+
+      if (state->depth_region) {
+         OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
+         OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
+         OUT_RELOC(state->depth_region->buffer, DRM_MM_TT|DRM_MM_WRITE,
+                   state->depth_region->draw_offset);
+      }
+
+      OUT_BATCH(state->Buffer[I915_DESTREG_DV0]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_DV1]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SENABLE]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SR0]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SR1]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SR2]);
+      ADVANCE_BATCH();
    }
 
    if (dirty & I915_UPLOAD_STIPPLE) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_STIPPLE:\n"); 
-      emit( i915, state->Stipple, sizeof(state->Stipple) );
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_STIPPLE:\n"); 
+      emit(intel, state->Stipple, sizeof(state->Stipple) );
    }
 
    if (dirty & I915_UPLOAD_FOG) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_FOG:\n"); 
-      emit( i915, state->Fog, sizeof(state->Fog) );
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_FOG:\n"); 
+      emit(intel, state->Fog, sizeof(state->Fog) );
    }
 
    /* Combine all the dirty texture state into a single command to
@@ -316,18 +340,29 @@ static void i915_emit_state( intelContextPtr intel )
 	 if (dirty & I915_UPLOAD_TEX(i)) 
 	    nr++;
 
-      BEGIN_BATCH(2+nr*3);
+      BEGIN_BATCH(2+nr*3, 0);
       OUT_BATCH(_3DSTATE_MAP_STATE | (3*nr));
       OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
       for (i = 0 ; i < I915_TEX_UNITS ; i++)
 	 if (dirty & I915_UPLOAD_TEX(i)) {
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_MS2]);
+
+	    if (state->tex_buffer[i]) {
+	       OUT_RELOC(state->tex_buffer[i],
+			 DRM_MM_TT|DRM_MM_READ,
+			 state->tex_offset[i]);
+	    }
+	    else {
+	       assert(i == 0);
+	       assert(state == &i915->meta);
+	       OUT_BATCH(0);
+	    }
+
 	    OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
 	    OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
 	 }
       ADVANCE_BATCH();
 
-      BEGIN_BATCH(2+nr*3);
+      BEGIN_BATCH(2+nr*3, 0);
       OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3*nr));
       OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
       for (i = 0 ; i < I915_TEX_UNITS ; i++)
@@ -340,114 +375,132 @@ static void i915_emit_state( intelContextPtr intel )
    }
 
    if (dirty & I915_UPLOAD_CONSTANTS) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_CONSTANTS:\n"); 
-      emit( i915, state->Constant, state->ConstantSize * sizeof(GLuint) );
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_CONSTANTS:\n"); 
+      emit(intel, state->Constant, state->ConstantSize * sizeof(GLuint) );
    }
 
    if (dirty & I915_UPLOAD_PROGRAM) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_PROGRAM:\n"); 
+      if (INTEL_DEBUG & DEBUG_STATE) fprintf(stderr, "I915_UPLOAD_PROGRAM:\n"); 
 
       assert((state->Program[0] & 0x1ff)+2 == state->ProgramSize);
       
-      emit( i915, state->Program, state->ProgramSize * sizeof(GLuint) );
-      if (VERBOSE)
+      emit(intel, state->Program, state->ProgramSize * sizeof(GLuint) );
+      if (INTEL_DEBUG & DEBUG_STATE)
 	 i915_disassemble_program( state->Program, state->ProgramSize );
    }
 
    state->emitted |= dirty;
-   intel->batch.last_emit_state = counter;
-   assert(counter == intel->batch.counter);
 }
 
-static void i915_destroy_context( intelContextPtr intel )
+static void i915_destroy_context( struct intel_context *intel )
 {
    _tnl_free_vertices(&intel->ctx);
 }
 
 
 /**
- * Set the color buffer drawing region.
+ * Set the drawing regions for the color and depth/stencil buffers.
+ * This involves setting the pitch, cpp and buffer ID/location.
+ * Also set pixel format for color and Z rendering
+ * Used for setting both regular and meta state.
  */
-static void
-i915_set_color_region( intelContextPtr intel, const intelRegion *region)
+void
+i915_state_draw_region(struct intel_context *intel, 
+                       struct i915_hw_state *state,
+                       struct intel_region *color_region,
+                       struct intel_region *depth_region)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
-   I915_STATECHANGE( i915, I915_UPLOAD_BUFFERS );
-   i915->state.Buffer[I915_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_CBUFADDR2] = region->offset;
-}
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   GLuint value;
 
+   ASSERT(state == &i915->state || state == &i915->meta);
+
+   if (state->draw_region != color_region) {
+      intel_region_release(intel, &state->draw_region);
+      intel_region_reference(&state->draw_region, color_region);
+   }
+   if (state->depth_region != depth_region) {
+      intel_region_release(intel, &state->depth_region);
+      intel_region_reference(&state->depth_region, depth_region);
+   }
+
+   /*
+    * Set stride/cpp values
+    */
+   if (color_region) {
+      state->Buffer[I915_DESTREG_CBUFADDR0] = _3DSTATE_BUF_INFO_CMD;
+      state->Buffer[I915_DESTREG_CBUFADDR1] = 
+         (BUF_3D_ID_COLOR_BACK | 
+          BUF_3D_PITCH(color_region->pitch * color_region->cpp) |
+          BUF_3D_USE_FENCE);
+   }
+
+   if (depth_region) {
+      state->Buffer[I915_DESTREG_DBUFADDR0] = _3DSTATE_BUF_INFO_CMD;
+      state->Buffer[I915_DESTREG_DBUFADDR1] = 
+	 (BUF_3D_ID_DEPTH |
+	  BUF_3D_PITCH(depth_region->pitch * depth_region->cpp) |
+	  BUF_3D_USE_FENCE);
+   }
+
+   /*
+    * Compute/set I915_DESTREG_DV1 value
+    */
+   value = (DSTORG_HORT_BIAS(0x8) | /* .5 */
+            DSTORG_VERT_BIAS(0x8) | /* .5 */
+            LOD_PRECLAMP_OGL |
+            TEX_DEFAULT_COLOR_OGL);
+   if (color_region && color_region->cpp == 4) {
+      value |= DV_PF_8888;
+   }
+   else {
+      value |= (DITHER_FULL_ALWAYS | DV_PF_565);
+   }
+   if (depth_region && depth_region->cpp == 4) {
+      value |= DEPTH_FRMT_24_FIXED_8_OTHER;
+   }
+   else {
+      value |= DEPTH_FRMT_16_FIXED;
+   }
+   state->Buffer[I915_DESTREG_DV1] = value;
 
-/**
- * specify the z-buffer/stencil region
- */
-static void
-i915_set_z_region( intelContextPtr intel, const intelRegion *region)
-{
-   i915ContextPtr i915 = I915_CONTEXT(intel);
    I915_STATECHANGE( i915, I915_UPLOAD_BUFFERS );
-   i915->state.Buffer[I915_DESTREG_DBUFADDR1] =
-      (BUF_3D_ID_DEPTH | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_DBUFADDR2] = region->offset;
 }
 
 
-/**
- * Set both the color and Z/stencil drawing regions.
- * Similar to two previous functions, but don't use I915_STATECHANGE()
- */
 static void
-i915_update_color_z_regions(intelContextPtr intel,
-                            const intelRegion *colorRegion,
-                            const intelRegion *depthRegion)
+i915_set_draw_region(struct intel_context *intel, 
+                     struct intel_region *color_region,
+                     struct intel_region *depth_region)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
-
-   i915->state.Buffer[I915_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(colorRegion->pitch) | BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_CBUFADDR2] = colorRegion->offset;
-
-   i915->state.Buffer[I915_DESTREG_DBUFADDR1] =
-      (BUF_3D_ID_DEPTH |
-       BUF_3D_PITCH(depthRegion->pitch) |  /* pitch in bytes */
-       BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_DBUFADDR2] = depthRegion->offset;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   i915_state_draw_region(intel, &i915->state, color_region, depth_region);
 }
 
 
-static void i915_lost_hardware( intelContextPtr intel )
+
+static void i915_lost_hardware( struct intel_context *intel )
 {
-   I915_CONTEXT(intel)->state.emitted = 0;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   i915->state.emitted = 0;
 }
 
-static void i915_emit_flush( intelContextPtr intel )
+static GLuint i915_flush_cmd( void )
 {
-   BATCH_LOCALS;
-
-   BEGIN_BATCH(2);
-   OUT_BATCH( MI_FLUSH | FLUSH_MAP_CACHE | FLUSH_RENDER_CACHE ); 
-   OUT_BATCH( 0 );
-   ADVANCE_BATCH();
+   return MI_FLUSH | FLUSH_MAP_CACHE;
 }
 
 
-void i915InitVtbl( i915ContextPtr i915 )
+void i915InitVtbl( struct i915_context *i915 )
 {
-   i915->intel.vtbl.alloc_tex_obj = i915AllocTexObj;
    i915->intel.vtbl.check_vertex_size = i915_check_vertex_size;
-   i915->intel.vtbl.clear_with_tris = i915ClearWithTris;
-   i915->intel.vtbl.rotate_window = i915RotateWindow;
    i915->intel.vtbl.destroy = i915_destroy_context;
-   i915->intel.vtbl.emit_invarient_state = i915_emit_invarient_state;
    i915->intel.vtbl.emit_state = i915_emit_state;
    i915->intel.vtbl.lost_hardware = i915_lost_hardware;
    i915->intel.vtbl.reduced_primitive_state = i915_reduced_primitive_state;
    i915->intel.vtbl.render_start = i915_render_start;
-   i915->intel.vtbl.set_color_region = i915_set_color_region;
-   i915->intel.vtbl.set_z_region = i915_set_z_region;
-   i915->intel.vtbl.update_color_z_regions = i915_update_color_z_regions;
+   i915->intel.vtbl.set_draw_region = i915_set_draw_region;
    i915->intel.vtbl.update_texture_state = i915UpdateTextureState;
-   i915->intel.vtbl.emit_flush = i915_emit_flush;
+   i915->intel.vtbl.flush_cmd = i915_flush_cmd;
 }
 
diff --git a/src/mesa/drivers/dri/i915/intel_batchbuffer.c b/src/mesa/drivers/dri/i915/intel_batchbuffer.c
index 0631237277..02863f8b53 100644
--- a/src/mesa/drivers/dri/i915/intel_batchbuffer.c
+++ b/src/mesa/drivers/dri/i915/intel_batchbuffer.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -25,760 +25,270 @@
  * 
  **************************************************************************/
 
-
-#include <stdio.h>
-#include <errno.h>
-
-#include "mtypes.h"
-#include "context.h"
-#include "enums.h"
-#include "vblank.h"
-
-#include "intel_reg.h"
 #include "intel_batchbuffer.h"
-#include "intel_context.h"
-
-
-
-
-/* ================================================================
- * Performance monitoring functions
+#include "intel_ioctl.h"
+#include "intel_bufmgr.h"
+
+/* Relocations in kernel space:
+ *    - pass dma buffer seperately
+ *    - memory manager knows how to patch
+ *    - pass list of dependent buffers
+ *    - pass relocation list
+ *
+ * Either:
+ *    - get back an offset for buffer to fire
+ *    - memory manager knows how to fire buffer
+ *
+ * Really want the buffer to be AGP and pinned.
+ *
  */
 
-static void intel_fill_box( intelContextPtr intel,
-			    GLshort x, GLshort y,
-			    GLshort w, GLshort h,
-			    GLubyte r, GLubyte g, GLubyte b )
+/* Cliprect fence: The highest fence protecting a dma buffer
+ * containing explicit cliprect information.  Like the old drawable
+ * lock but irq-driven.  X server must wait for this fence to expire
+ * before changing cliprects [and then doing sw rendering?].  For
+ * other dma buffers, the scheduler will grab current cliprect info
+ * and mix into buffer.  X server must hold the lock while changing
+ * cliprects???  Make per-drawable.  Need cliprects in shared memory
+ * -- beats storing them with every cmd buffer in the queue.
+ *
+ * ==> X server must wait for this fence to expire before touching the
+ * framebuffer with new cliprects.
+ *
+ * ==> Cliprect-dependent buffers associated with a
+ * cliprect-timestamp.  All of the buffers associated with a timestamp
+ * must go to hardware before any buffer with a newer timestamp.
+ *
+ * ==> Dma should be queued per-drawable for correct X/GL
+ * synchronization.  Or can fences be used for this?
+ *
+ * Applies to: Blit operations, metaops, X server operations -- X
+ * server automatically waits on its own dma to complete before
+ * modifying cliprects ???
+ */				
+
+static void intel_dump_batchbuffer( GLuint offset,
+				    GLuint *ptr,
+				    GLuint count )
 {
-   x += intel->drawX;
-   y += intel->drawY;
-
-   if (x >= 0 && y >= 0 &&
-       x+w < intel->intelScreen->width &&
-       y+h < intel->intelScreen->height)
-      intelEmitFillBlitLocked( intel, 
-			       intel->intelScreen->cpp,
-			       intel->intelScreen->back.pitch,
-			       intel->intelScreen->back.offset,
-			       x, y, w, h,
-			       INTEL_PACKCOLOR(intel->intelScreen->fbFormat,
-					       r,g,b,0xff));
+   int i;
+   fprintf(stderr, "\n\n\nSTART BATCH (%d dwords):\n", count/4);
+   for (i = 0; i < count/4; i += 4) 
+      fprintf(stderr, "0x%x:\t0x%08x 0x%08x 0x%08x 0x%08x\n", 
+	      offset + i*4, ptr[i], ptr[i+1], ptr[i+2], ptr[i+3]);
+   fprintf(stderr, "END BATCH\n\n\n");
 }
 
-static void intel_draw_performance_boxes( intelContextPtr intel )
-{
-   /* Purple box for page flipping
-    */
-   if ( intel->perf_boxes & I830_BOX_FLIP ) 
-      intel_fill_box( intel, 4, 4, 8, 8, 255, 0, 255 );
-
-   /* Red box if we have to wait for idle at any point
-    */
-   if ( intel->perf_boxes & I830_BOX_WAIT ) 
-      intel_fill_box( intel, 16, 4, 8, 8, 255, 0, 0 );
-
-   /* Blue box: lost context?
-    */
-   if ( intel->perf_boxes & I830_BOX_LOST_CONTEXT ) 
-      intel_fill_box( intel, 28, 4, 8, 8, 0, 0, 255 );
-
-   /* Yellow box for texture swaps
-    */
-   if ( intel->perf_boxes & I830_BOX_TEXTURE_LOAD ) 
-      intel_fill_box( intel, 40, 4, 8, 8, 255, 255, 0 );
 
-   /* Green box if hardware never idles (as far as we can tell)
-    */
-   if ( !(intel->perf_boxes & I830_BOX_RING_EMPTY) ) 
-      intel_fill_box( intel, 64, 4, 8, 8, 0, 255, 0 );
-
-
-   /* Draw bars indicating number of buffers allocated 
-    * (not a great measure, easily confused)
-    */
-#if 0
-   if (intel->dma_used) {
-      int bar = intel->dma_used / 10240;
-      if (bar > 100) bar = 100;
-      if (bar < 1) bar = 1;
-      intel_fill_box( intel, 4, 16, bar, 4, 196, 128, 128 );
-      intel->dma_used = 0;
-   }
-#endif
-
-   intel->perf_boxes = 0;
-}
-
-
-
-
-
-
-static int bad_prim_vertex_nr( int primitive, int nr )
+void intel_batchbuffer_reset( struct intel_batchbuffer *batch )
 {
-   switch (primitive & PRIM3D_MASK) {
-   case PRIM3D_POINTLIST:
-      return nr < 1;
-   case PRIM3D_LINELIST:
-      return (nr & 1) || nr == 0;
-   case PRIM3D_LINESTRIP:
-      return nr < 2;
-   case PRIM3D_TRILIST:
-   case PRIM3D_RECTLIST:
-      return nr % 3 || nr == 0;
-   case PRIM3D_POLY:
-   case PRIM3D_TRIFAN:
-   case PRIM3D_TRISTRIP:
-   case PRIM3D_TRISTRIP_RVRSE:
-      return nr < 3;
-   default:
-      return 1;
-   }	
+   bmBufferData(batch->bm,
+		batch->buffer,
+		BATCH_SZ,
+		NULL,
+		0);
+		
+   if (!batch->list) 
+      batch->list = bmNewBufferList();
+
+   drmMMClearBufList(batch->list);
+   batch->list_count = 0;
+   batch->nr_relocs = 0;
+   batch->flags = 0;
+
+   bmAddBuffer( batch->bm,
+	        batch->list,
+		batch->buffer,
+		DRM_MM_TT,
+		NULL,
+		&batch->offset[batch->list_count++]);
+
+   batch->map = bmMapBuffer(batch->bm, batch->buffer, DRM_MM_WRITE);
+   batch->ptr = batch->map;
 }
 
-static void intel_flush_inline_primitive( GLcontext *ctx )
+/*======================================================================
+ * Public functions
+ */
+struct intel_batchbuffer *intel_batchbuffer_alloc( struct intel_context *intel )
 {
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   GLuint used = intel->batch.ptr - intel->prim.start_ptr;
-   GLuint vertcount;
-
-   assert(intel->prim.primitive != ~0);
-
-   if (1) {
-      /* Check vertex size against the vertex we're specifying to
-       * hardware.  If it's wrong, ditch the primitive.
-       */ 
-      if (!intel->vtbl.check_vertex_size( intel, intel->vertex_size )) 
-	 goto do_discard;
-
-      vertcount = (used - 4)/ (intel->vertex_size * 4);
-
-      if (!vertcount)
-	 goto do_discard;
-      
-      if (vertcount * intel->vertex_size * 4 != used - 4) {
-	 fprintf(stderr, "vertex size confusion %d %d\n", used, 
-		 intel->vertex_size * vertcount * 4);
-	 goto do_discard;
-      }
-
-      if (bad_prim_vertex_nr( intel->prim.primitive, vertcount )) {
-	 fprintf(stderr, "bad_prim_vertex_nr %x %d\n", intel->prim.primitive,
-		 vertcount);
-	 goto do_discard;
-      }
-   }
+   struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
 
-   if (used < 8)
-      goto do_discard;
+   batch->intel = intel;
+   batch->bm = intel->bm;
 
-   *(int *)intel->prim.start_ptr = (_3DPRIMITIVE | 
-				    intel->prim.primitive |
-				    (used/4-2));
+   bmGenBuffers(intel->bm, 1, &batch->buffer, BM_BATCHBUFFER);
+   batch->last_fence = bmInitFence(batch->bm);
+   intel_batchbuffer_reset( batch );
+   return batch;
+}
 
-   goto finished;
+void intel_batchbuffer_free( struct intel_batchbuffer *batch )
+{
+   if (batch->map)
+      bmUnmapBuffer(batch->bm, batch->buffer);
    
- do_discard:
-   intel->batch.ptr -= used;
-   intel->batch.space += used;
-   assert(intel->batch.space >= 0);
-
- finished:
-   intel->prim.primitive = ~0;
-   intel->prim.start_ptr = 0;
-   intel->prim.flush = 0;
+   free(batch);
 }
 
-
-/* Emit a primitive referencing vertices in a vertex buffer.
+/* TODO: Push this whole function into bufmgr.
  */
-void intelStartInlinePrimitive( intelContextPtr intel, GLuint prim )
+static void do_flush_locked( struct intel_batchbuffer *batch,
+			     GLuint used,
+			     GLboolean ignore_cliprects,
+			     GLboolean allow_unlock)
 {
-   BATCH_LOCALS;
+   GLuint *ptr;
+   GLuint i;
 
-   if (0)
-      fprintf(stderr, "%s %x\n", __FUNCTION__, prim);
+   bmValidateBufferList( batch->bm, 
+			 batch->list,
+			 DRM_MM_TT );
 
-
-   /* Finish any in-progress primitive:
-    */
-   INTEL_FIREVERTICES( intel );
-   
-   /* Emit outstanding state:
+   /* Apply the relocations.  This nasty map indicates to me that the
+    * whole task should be done internally by the memory manager, and
+    * that dma buffers probably need to be pinned within agp space.
     */
-   intel->vtbl.emit_state( intel );
+   ptr = (GLuint *)bmMapBuffer(batch->bm, batch->buffer, DRM_MM_WRITE);
+
    
-   /* Make sure there is some space in this buffer:
-    */
-   if (intel->vertex_size * 10 * sizeof(GLuint) >= intel->batch.space) {
-      intelFlushBatch(intel, GL_TRUE); 
-      intel->vtbl.emit_state( intel );
+   for (i = 0; i < batch->nr_relocs; i++) {
+      struct buffer_reloc *r = &batch->reloc[i];
+      
+      assert(r->elem < batch->list_count);
+      ptr[r->offset/4] = batch->offset[r->elem] + r->delta;
    }
 
-#if 1
-   if (((unsigned long)intel->batch.ptr) & 0x4) {
-      BEGIN_BATCH(1);
-      OUT_BATCH(0);
-      ADVANCE_BATCH();
-   }
-#endif
+   if (INTEL_DEBUG & DEBUG_DMA)
+      intel_dump_batchbuffer( 0, ptr, used );
 
-   /* Emit a slot which will be filled with the inline primitive
-    * command later.
+   bmUnmapBuffer(batch->bm, batch->buffer);
+   
+   /* Fire the batch buffer, which was uploaded above:
     */
-   BEGIN_BATCH(2);
-   OUT_BATCH( 0 );
-
-   intel->prim.start_ptr = batch_ptr;
-   intel->prim.primitive = prim;
-   intel->prim.flush = intel_flush_inline_primitive;
-   intel->batch.contains_geometry = 1;
-
-   OUT_BATCH( 0 );
-   ADVANCE_BATCH();
-}
-
-
-void intelRestartInlinePrimitive( intelContextPtr intel )
-{
-   GLuint prim = intel->prim.primitive;
 
-   intel_flush_inline_primitive( &intel->ctx );
-   if (1) intelFlushBatch(intel, GL_TRUE); /* GL_TRUE - is critical */
-   intelStartInlinePrimitive( intel, prim );
+#if 1
+   intel_batch_ioctl(batch->intel, 
+		     batch->offset[0],
+		     used,
+		     ignore_cliprects,
+		     allow_unlock);
+#endif
+   batch->last_fence = bmFenceBufferList(batch->bm, batch->list);
+   if (!batch->intel->last_swap_fence_retired) {
+      int retired;
+      drmFence dFence = {0,batch->intel->last_swap_fence};
+ 
+     /*FIXME: Temporary fix for fence ageing
+      *
+      */
+      if (!drmTestFence(batch->intel->driFd, dFence, 0, &retired)) {
+	 batch->intel->last_swap_fence_retired = retired; 
+      }
+   }	  
 }
 
 
 
-void intelWrapInlinePrimitive( intelContextPtr intel )
-{
-   GLuint prim = intel->prim.primitive;
-
-   if (0)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-   intel_flush_inline_primitive( &intel->ctx );
-   intelFlushBatch(intel, GL_TRUE);
-   intelStartInlinePrimitive( intel, prim );
-}
-
-
-/* Emit a primitive with space for inline vertices.
- */
-GLuint *intelEmitInlinePrimitiveLocked(intelContextPtr intel, 
-				       int primitive,
-				       int dwords,
-				       int vertex_size )
+GLuint intel_batchbuffer_flush( struct intel_batchbuffer *batch )
 {
-   GLuint *tmp = 0;
-   BATCH_LOCALS;
+   struct intel_context *intel = batch->intel;
+   GLuint used = batch->ptr - batch->map;
 
-   if (0)
-      fprintf(stderr, "%s 0x%x %d\n", __FUNCTION__, primitive, dwords);
+   if (used == 0) 
+      return batch->last_fence;
 
-   /* Emit outstanding state:
+   /* Add the MI_BATCH_BUFFER_END.  Always add an MI_FLUSH - this is a
+    * performance drain that we would like to avoid.
     */
-   intel->vtbl.emit_state( intel );
-
-   if ((1+dwords)*4 >= intel->batch.space) {
-      intelFlushBatch(intel, GL_TRUE); 
-      intel->vtbl.emit_state( intel );
+   if (used & 4) {
+      ((int *)batch->ptr)[0] = intel->vtbl.flush_cmd();
+      ((int *)batch->ptr)[1] = 0;
+      ((int *)batch->ptr)[2] = MI_BATCH_BUFFER_END;
+      used += 12;
    }
-
-
-   if (1) {
-      int used = dwords * 4;
-      int vertcount;
-
-      /* Check vertex size against the vertex we're specifying to
-       * hardware.  If it's wrong, ditch the primitive.
-       */ 
-      if (!intel->vtbl.check_vertex_size( intel, vertex_size )) 
-	 goto do_discard;
-
-      vertcount = dwords / vertex_size;
-      
-      if (dwords % vertex_size) {
-	 fprintf(stderr, "did not request a whole number of vertices\n");
-	 goto do_discard;
-      }
-
-      if (bad_prim_vertex_nr( primitive, vertcount )) {
-	 fprintf(stderr, "bad_prim_vertex_nr %x %d\n", primitive, vertcount);
-	 goto do_discard;
-      }
-
-      if (used < 8)
-	 goto do_discard;
-   }
-
-   /* Emit 3D_PRIMITIVE commands:
-    */
-   BEGIN_BATCH(1 + dwords);
-   OUT_BATCH( _3DPRIMITIVE | 
-	      primitive |
-	      (dwords-1) );
-
-   tmp = (GLuint *)batch_ptr;
-   batch_ptr += dwords * 4;
-
-   ADVANCE_BATCH();
-
-   intel->batch.contains_geometry = 1;
-
- do_discard:
-   return tmp;
-}
-
-
-static void intelWaitForFrameCompletion( intelContextPtr intel )
-{
-  drm_i915_sarea_t *sarea = (drm_i915_sarea_t *)intel->sarea;
-
-   if (intel->do_irqs) {
-      if (intelGetLastFrame(intel) < sarea->last_dispatch) {
-	 if (!intel->irqsEmitted) {
-	    while (intelGetLastFrame (intel) < sarea->last_dispatch)
-	       ;
-	 }
-	 else {
-	    UNLOCK_HARDWARE( intel ); 
-	    intelWaitIrq( intel, intel->alloc.irq_emitted );	
-	    LOCK_HARDWARE( intel ); 
-	 }
-	 intel->irqsEmitted = 10;
-      }
-
-      if (intel->irqsEmitted) {
-	 intelEmitIrqLocked( intel );
-	 intel->irqsEmitted--;
-      }
-   } 
    else {
-      while (intelGetLastFrame (intel) < sarea->last_dispatch) {
-	 UNLOCK_HARDWARE( intel ); 
-	 if (intel->do_usleeps) 
-	    DO_USLEEP( 1 );
-	 LOCK_HARDWARE( intel ); 
-      }
+      ((int *)batch->ptr)[0] = intel->vtbl.flush_cmd() ; 
+      ((int *)batch->ptr)[1] = MI_BATCH_BUFFER_END;
+      used += 8;
    }
-}
-
-/*
- * Copy the back buffer to the front buffer. 
- */
-void intelCopyBuffer( const __DRIdrawablePrivate *dPriv,
-		      const drm_clip_rect_t	 *rect)
-{
-   intelContextPtr intel;
-   GLboolean   missed_target;
-   int64_t ust;
 
-   if (0)
-      fprintf(stderr, "%s\n", __FUNCTION__);
+   bmUnmapBuffer(batch->bm, batch->buffer);
+   batch->ptr = NULL;
+   batch->map = NULL;
 
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   intel = (intelContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   intelFlush( &intel->ctx );
-   
-   LOCK_HARDWARE( intel );
-   intelWaitForFrameCompletion( intel );
-
-   if (!rect)
-   {
-       UNLOCK_HARDWARE( intel );
-       driWaitForVBlank( dPriv, &intel->vbl_seq, intel->vblank_flags, & missed_target );
-       LOCK_HARDWARE( intel );
-   }
+   /* TODO: Just pass the relocation list and dma buffer up to the
+    * kernel.
+    */
+   if (!intel->locked)
    {
-      const intelScreenPrivate *intelScreen = intel->intelScreen;
-      const __DRIdrawablePrivate *dPriv = intel->driDrawable;
-      const int nbox = dPriv->numClipRects;
-      const drm_clip_rect_t *pbox = dPriv->pClipRects;
-      drm_clip_rect_t box;
-      const int cpp = intelScreen->cpp;
-      const int pitch = intelScreen->front.pitch; /* in bytes */
-      int i;
-      GLuint CMD, BR13;
-      BATCH_LOCALS;
-
-      switch(cpp) {
-      case 2: 
-	 BR13 = (pitch) | (0xCC << 16) | (1<<24);
-	 CMD = XY_SRC_COPY_BLT_CMD;
-	 break;
-      case 4:
-	 BR13 = (pitch) | (0xCC << 16) | (1<<24) | (1<<25);
-	 CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
-		XY_SRC_COPY_BLT_WRITE_RGB);
-	 break;
-      default:
-	 BR13 = (pitch) | (0xCC << 16) | (1<<24);
-	 CMD = XY_SRC_COPY_BLT_CMD;
-	 break;
-      }
-   
-      if (0) 
-	 intel_draw_performance_boxes( intel );
-
-      for (i = 0 ; i < nbox; i++, pbox++) 
-      {
-	 if (pbox->x1 > pbox->x2 ||
-	     pbox->y1 > pbox->y2 ||
-	     pbox->x2 > intelScreen->width ||
-	     pbox->y2 > intelScreen->height) {
-            _mesa_warning(&intel->ctx, "Bad cliprect in intelCopyBuffer()");
-	    continue;
-         }
-
-	 box = *pbox;
-
-	 if (rect)
-	 {
-	     if (rect->x1 > box.x1)
-		 box.x1 = rect->x1;
-	     if (rect->y1 > box.y1)
-		 box.y1 = rect->y1;
-	     if (rect->x2 < box.x2)
-		 box.x2 = rect->x2;
-	     if (rect->y2 < box.y2)
-		 box.y2 = rect->y2;
-
-	     if (box.x1 > box.x2 || box.y1 > box.y2)
-		 continue;
-	 }
-
-	 BEGIN_BATCH( 8);
-	 OUT_BATCH( CMD );
-	 OUT_BATCH( BR13 );
-	 OUT_BATCH( (box.y1 << 16) | box.x1 );
-	 OUT_BATCH( (box.y2 << 16) | box.x2 );
-
-	 if (intel->sarea->pf_current_page == 0) 
-	    OUT_BATCH( intelScreen->front.offset );
-	 else
-	    OUT_BATCH( intelScreen->back.offset );			
-
-	 OUT_BATCH( (box.y1 << 16) | box.x1 );
-	 OUT_BATCH( BR13 & 0xffff );
-
-	 if (intel->sarea->pf_current_page == 0) 
-	    OUT_BATCH( intelScreen->back.offset );			
-	 else
-	    OUT_BATCH( intelScreen->front.offset );
-
-	 ADVANCE_BATCH();
-      }
-   }
-   intelFlushBatchLocked( intel, GL_TRUE, GL_TRUE, GL_TRUE );
-   UNLOCK_HARDWARE( intel );
+      assert(!(batch->flags & INTEL_BATCH_NO_CLIPRECTS));
 
-   if (!rect)
-   {
-       intel->swap_count++;
-       (*dri_interface->getUST)(&ust);
-       if (missed_target) {
-	   intel->swap_missed_count++;
-	   intel->swap_missed_ust = ust -  intel->swap_ust;
-       }
-   
-       intel->swap_ust = ust;
+      LOCK_HARDWARE(intel);
+      do_flush_locked(batch, used, GL_FALSE, GL_TRUE);
+      UNLOCK_HARDWARE(intel);
    }
-}
-
-
-
-
-void intelEmitFillBlitLocked( intelContextPtr intel,
-			      GLuint cpp,
-			      GLshort dst_pitch,  /* in bytes */
-			      GLuint dst_offset,
-			      GLshort x, GLshort y, 
-			      GLshort w, GLshort h,
-			      GLuint color )
-{
-   GLuint BR13, CMD;
-   BATCH_LOCALS;
-
-   switch(cpp) {
-   case 1: 
-   case 2: 
-   case 3: 
-      BR13 = dst_pitch | (0xF0 << 16) | (1<<24);
-      CMD = XY_COLOR_BLT_CMD;
-      break;
-   case 4:
-      BR13 = dst_pitch | (0xF0 << 16) | (1<<24) | (1<<25);
-      CMD = (XY_COLOR_BLT_CMD | XY_COLOR_BLT_WRITE_ALPHA |
-	     XY_COLOR_BLT_WRITE_RGB);
-      break;
-   default:
-      return;
+   else {
+      GLboolean ignore_cliprects = !(batch->flags & INTEL_BATCH_CLIPRECTS);
+      do_flush_locked(batch, used, ignore_cliprects, GL_FALSE);
    }
 
-   BEGIN_BATCH( 6);
-   OUT_BATCH( CMD );
-   OUT_BATCH( BR13 );
-   OUT_BATCH( (y << 16) | x );
-   OUT_BATCH( ((y+h) << 16) | (x+w) );
-   OUT_BATCH( dst_offset );
-   OUT_BATCH( color );
-   ADVANCE_BATCH();
+   /* Reset the buffer:
+    */
+   intel_batchbuffer_reset( batch );
+   return batch->last_fence;
 }
 
-
-/* Copy BitBlt
- */
-void intelEmitCopyBlitLocked( intelContextPtr intel,
-			      GLuint cpp,
-			      GLshort src_pitch,
-			      GLuint  src_offset,
-			      GLshort dst_pitch,
-			      GLuint  dst_offset,
-			      GLshort src_x, GLshort src_y,
-			      GLshort dst_x, GLshort dst_y,
-			      GLshort w, GLshort h )
-{
-   GLuint CMD, BR13;
-   int dst_y2 = dst_y + h;
-   int dst_x2 = dst_x + w;
-   BATCH_LOCALS;
-
-   src_pitch *= cpp;
-   dst_pitch *= cpp;
-
-   switch(cpp) {
-   case 1: 
-   case 2: 
-   case 3: 
-      BR13 = dst_pitch | (0xCC << 16) | (1<<24);
-      CMD = XY_SRC_COPY_BLT_CMD;
-      break;
-   case 4:
-      BR13 = dst_pitch | (0xCC << 16) | (1<<24) | (1<<25);
-      CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
-	     XY_SRC_COPY_BLT_WRITE_RGB);
-      break;
-   default:
-      return;
-   }
-
-   if (dst_y2 < dst_y ||
-       dst_x2 < dst_x) {
-      return;
-   }
-
-   BEGIN_BATCH( 12);
-   OUT_BATCH( CMD );
-   OUT_BATCH( BR13 );
-   OUT_BATCH( (dst_y << 16) | dst_x );
-   OUT_BATCH( (dst_y2 << 16) | dst_x2 );
-   OUT_BATCH( dst_offset );	
-   OUT_BATCH( (src_y << 16) | src_x );
-   OUT_BATCH( src_pitch );
-   OUT_BATCH( src_offset ); 
-   ADVANCE_BATCH();
+void intel_batchbuffer_finish( struct intel_batchbuffer *batch )
+{   
+   bmFinishFence(batch->bm, 
+		 intel_batchbuffer_flush(batch));
 }
+   
 
-
-
-void intelClearWithBlit(GLcontext *ctx, GLbitfield flags, GLboolean all,
-		      GLint cx1, GLint cy1, GLint cw, GLint ch)
+/*  This is the only way buffers get added to the validate list.
+ */
+GLboolean intel_batchbuffer_emit_reloc( struct intel_batchbuffer *batch,
+					GLuint buffer,
+					GLuint flags,
+					GLuint delta )
 {
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   intelScreenPrivate *intelScreen = intel->intelScreen;
-   GLuint clear_depth, clear_color;
-   GLint cx, cy;
-   GLint pitch;
-   GLint cpp = intelScreen->cpp;
-   GLint i;
-   GLuint BR13, CMD, D_CMD;
-   BATCH_LOCALS;
-
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-
-   pitch = intelScreen->front.pitch;
-
-   clear_color = intel->ClearColor;
-   clear_depth = 0;
-
-   if (flags & BUFFER_BIT_DEPTH) {
-      clear_depth = (GLuint)(ctx->Depth.Clear * intel->ClearDepth);
-   }
-
-   if (flags & BUFFER_BIT_STENCIL) {
-      clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
-   }
-
-   switch(cpp) {
-   case 2: 
-      BR13 = (0xF0 << 16) | (pitch) | (1<<24);
-      D_CMD = CMD = XY_COLOR_BLT_CMD;
-      break;
-   case 4:
-      BR13 = (0xF0 << 16) | (pitch) | (1<<24) | (1<<25);
-      CMD = (XY_COLOR_BLT_CMD |
-	     XY_COLOR_BLT_WRITE_ALPHA | 
-	     XY_COLOR_BLT_WRITE_RGB);
-      D_CMD = XY_COLOR_BLT_CMD;
-      if (flags & BUFFER_BIT_DEPTH) D_CMD |= XY_COLOR_BLT_WRITE_RGB;
-      if (flags & BUFFER_BIT_STENCIL) D_CMD |= XY_COLOR_BLT_WRITE_ALPHA;
-      break;
-   default:
-      BR13 = (0xF0 << 16) | (pitch) | (1<<24);
-      D_CMD = CMD = XY_COLOR_BLT_CMD;
-      break;
+   GLuint i;
+
+   assert(batch->nr_relocs <= MAX_RELOCS);
+
+   i = bmScanBufferList(batch->bm, batch->list, buffer);
+   if (i == -1) {
+      i = batch->list_count; 
+      bmAddBuffer(batch->bm,
+		  batch->list,
+		  buffer,
+		  flags,
+		  NULL,
+		  &batch->offset[batch->list_count++]);
    }
 
    {
-      /* flip top to bottom */
-      cy = intel->driDrawable->h-cy1-ch;
-      cx = cx1 + intel->drawX;
-      cy += intel->drawY;
-
-      /* adjust for page flipping */
-      if ( intel->sarea->pf_current_page == 1 ) {
-	 GLuint tmp = flags;
-
-	 flags &= ~(BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_BACK_LEFT);
-	 if ( tmp & BUFFER_BIT_FRONT_LEFT ) flags |= BUFFER_BIT_BACK_LEFT;
-	 if ( tmp & BUFFER_BIT_BACK_LEFT )  flags |= BUFFER_BIT_FRONT_LEFT;
-      }
-
-      for (i = 0 ; i < intel->numClipRects ; i++) 
-      { 	 
-	 drm_clip_rect_t *box = &intel->pClipRects[i];	 
-	 drm_clip_rect_t b;
-
-	 if (!all) {
-	    GLint x = box->x1;
-	    GLint y = box->y1;
-	    GLint w = box->x2 - x;
-	    GLint h = box->y2 - y;
-
-	    if (x < cx) w -= cx - x, x = cx; 
-	    if (y < cy) h -= cy - y, y = cy;
-	    if (x + w > cx + cw) w = cx + cw - x;
-	    if (y + h > cy + ch) h = cy + ch - y;
-	    if (w <= 0) continue;
-	    if (h <= 0) continue;
-
-	    b.x1 = x;
-	    b.y1 = y;
-	    b.x2 = x + w;
-	    b.y2 = y + h;      
-	 } else {
-	    b = *box;
-	 }
-
-
-	 if (b.x1 > b.x2 ||
-	     b.y1 > b.y2 ||
-	     b.x2 > intelScreen->width ||
-	     b.y2 > intelScreen->height)
-	    continue;
-
-	 if ( flags & BUFFER_BIT_FRONT_LEFT ) {	    
-	    BEGIN_BATCH( 6);	    
-	    OUT_BATCH( CMD );
-	    OUT_BATCH( BR13 );
-	    OUT_BATCH( (b.y1 << 16) | b.x1 );
-	    OUT_BATCH( (b.y2 << 16) | b.x2 );
-	    OUT_BATCH( intelScreen->front.offset );
-	    OUT_BATCH( clear_color );
-	    ADVANCE_BATCH();
-	 }
-
-	 if ( flags & BUFFER_BIT_BACK_LEFT ) {
-	    BEGIN_BATCH( 6); 
-	    OUT_BATCH( CMD );
-	    OUT_BATCH( BR13 );
-	    OUT_BATCH( (b.y1 << 16) | b.x1 );
-	    OUT_BATCH( (b.y2 << 16) | b.x2 );
-	    OUT_BATCH( intelScreen->back.offset );
-	    OUT_BATCH( clear_color );
-	    ADVANCE_BATCH();
-	 }
-
-	 if ( flags & (BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH) ) {
-	    BEGIN_BATCH( 6);
-	    OUT_BATCH( D_CMD );
-	    OUT_BATCH( BR13 );
-	    OUT_BATCH( (b.y1 << 16) | b.x1 );
-	    OUT_BATCH( (b.y2 << 16) | b.x2 );
-	    OUT_BATCH( intelScreen->depth.offset );
-	    OUT_BATCH( clear_depth );
-	    ADVANCE_BATCH();
-	 }      
-      }
+      struct buffer_reloc *r = &batch->reloc[batch->nr_relocs++];
+      r->offset = batch->ptr - batch->map;
+      r->delta = delta;
+      r->elem = i;
    }
-   intelFlushBatchLocked( intel, GL_TRUE, GL_FALSE, GL_TRUE );
-   UNLOCK_HARDWARE( intel );
-}
 
+   batch->ptr += 4;
+   return GL_TRUE;
+}
 
 
 
-void intelDestroyBatchBuffer( GLcontext *ctx )
+void intel_batchbuffer_data(struct intel_batchbuffer *batch,
+			    const void *data,
+			    GLuint bytes,
+			    GLuint flags)
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-
-   if (intel->alloc.offset) {
-      intelFreeAGP( intel, intel->alloc.ptr );
-      intel->alloc.ptr = NULL;
-      intel->alloc.offset = 0;
-   }
-   else if (intel->alloc.ptr) {
-      free(intel->alloc.ptr);
-      intel->alloc.ptr = NULL;
-   }
-
-   memset(&intel->batch, 0, sizeof(intel->batch));
+   assert((bytes & 3) == 0);
+   intel_batchbuffer_require_space(batch, bytes, flags);
+   __memcpy(batch->ptr, data, bytes);
+   batch->ptr += bytes;
 }
 
-
-void intelInitBatchBuffer( GLcontext *ctx )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-
-   /* This path isn't really safe with rotate:
-    */
-   if (getenv("INTEL_BATCH") && intel->intelScreen->allow_batchbuffer) {      
-      switch (intel->intelScreen->deviceID) {
-      case PCI_CHIP_I865_G:
-	 /* HW bug?  Seems to crash if batchbuffer crosses 4k boundary.
-	  */
-	 intel->alloc.size = 8 * 1024; 
-	 break;
-      default:
-	 /* This is the smallest amount of memory the kernel deals with.
-	  * We'd ideally like to make this smaller.
-	  */
-	 intel->alloc.size = 1 << intel->intelScreen->logTextureGranularity;
-	 break;
-      }
-
-      intel->alloc.ptr = intelAllocateAGP( intel, intel->alloc.size );
-      if (intel->alloc.ptr)
-	 intel->alloc.offset = 
-	    intelAgpOffsetFromVirtual( intel, intel->alloc.ptr );
-      else
-         intel->alloc.offset = 0; /* OK? */
-   }
-
-   /* The default is now to use a local buffer and pass that to the
-    * kernel.  This is also a fallback if allocation fails on the
-    * above path:
-    */
-   if (!intel->alloc.ptr) {
-      intel->alloc.size = 8 * 1024;
-      intel->alloc.ptr = malloc( intel->alloc.size );
-      intel->alloc.offset = 0;
-   }
-
-   assert(intel->alloc.ptr);
-}
diff --git a/src/mesa/drivers/dri/i915/intel_batchbuffer.h b/src/mesa/drivers/dri/i915/intel_batchbuffer.h
index 577d07137f..87e5cb5e05 100644
--- a/src/mesa/drivers/dri/i915/intel_batchbuffer.h
+++ b/src/mesa/drivers/dri/i915/intel_batchbuffer.h
@@ -1,126 +1,126 @@
-/**************************************************************************
- * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
 #ifndef INTEL_BATCHBUFFER_H
 #define INTEL_BATCHBUFFER_H
 
-#include "intel_context.h"
-#include "intel_ioctl.h"
+#include "mtypes.h"
+#include "intel_bufmgr.h"
 
+struct intel_context;
 
-#define BATCH_LOCALS	GLubyte *batch_ptr;
+#define BATCH_SZ 4096
+#define BATCH_RESERVED 16
 
-/* #define VERBOSE 0 */
-#ifndef VERBOSE
-extern int VERBOSE;
-#endif
+#define MAX_RELOCS 100
 
+#define INTEL_BATCH_NO_CLIPRECTS 0x1
+#define INTEL_BATCH_CLIPRECTS    0x2
 
-#define BEGIN_BATCH(n)							\
-do {									\
-   if (VERBOSE) fprintf(stderr, 					\
-			"BEGIN_BATCH(%ld) in %s, %d dwords free\n",	\
-			((unsigned long)n), __FUNCTION__,		\
-			intel->batch.space/4);				\
-   if (intel->batch.space < (n)*4)					\
-      intelFlushBatch(intel, GL_TRUE);					\
-   if (intel->batch.space == intel->batch.size)	intel->batch.func = __FUNCTION__;			\
-   batch_ptr = intel->batch.ptr;					\
-} while (0)
+struct buffer_reloc {
+   GLuint offset;
+   GLuint elem;			/* elem in buffer list, not buffer id */
+   GLuint delta;		/* not needed? */
+};
+
+struct intel_batchbuffer {
+   struct bufmgr *bm;
+   struct intel_context *intel;
+
+   GLuint buffer;
+   GLuint last_fence;
+   GLuint flags;
+
+   /* In progress:
+    */
+   unsigned long offset[MAX_RELOCS];
+   struct _drmMMBufList *list;   
+   GLuint list_count;
+   GLubyte *map;
+   GLubyte *ptr; 
+
+   struct buffer_reloc reloc[MAX_RELOCS];
+   GLuint nr_relocs;
+};
+
+struct intel_batchbuffer *intel_batchbuffer_alloc( struct intel_context *intel );
+
+void intel_batchbuffer_free( struct intel_batchbuffer *batch );
+
+
+void intel_batchbuffer_finish( struct intel_batchbuffer *batch );
+
+GLuint intel_batchbuffer_flush( struct intel_batchbuffer *batch );
+
+void intel_batchbuffer_reset( struct intel_batchbuffer *batch );
 
-#define OUT_BATCH(n)					\
-do {							\
-   *(GLuint *)batch_ptr = (n);				\
-   if (VERBOSE) fprintf(stderr, " -- %08x at %s/%d\n", (n), __FILE__, __LINE__);	\
-   batch_ptr += 4;					\
-} while (0)
 
-#define ADVANCE_BATCH()						\
-do {								\
-   if (VERBOSE) fprintf(stderr, "ADVANCE_BATCH()\n");		\
-   intel->batch.space -= (batch_ptr - intel->batch.ptr);	\
-   intel->batch.ptr = batch_ptr;				\
-   assert(intel->batch.space >= 0);				\
-} while(0)
-
-extern void intelInitBatchBuffer( GLcontext *ctx );
-extern void intelDestroyBatchBuffer( GLcontext *ctx );
-
-extern void intelStartInlinePrimitive( intelContextPtr intel, GLuint prim );
-extern void intelWrapInlinePrimitive( intelContextPtr intel );
-extern void intelRestartInlinePrimitive( intelContextPtr intel );
-extern GLuint *intelEmitInlinePrimitiveLocked(intelContextPtr intel, 
-					      int primitive, int dwords,
-					      int vertex_size);
-extern void intelCopyBuffer( const __DRIdrawablePrivate *dpriv,
-			     const drm_clip_rect_t	*rect);
-extern void intelClearWithBlit(GLcontext *ctx, GLbitfield mask, GLboolean all,
-			     GLint cx1, GLint cy1, GLint cw, GLint ch);
-
-extern void intelEmitCopyBlitLocked( intelContextPtr intel,
-				     GLuint cpp,
-				     GLshort src_pitch,
-				     GLuint  src_offset,
-				     GLshort dst_pitch,
-				     GLuint  dst_offset,
-				     GLshort srcx, GLshort srcy,
-				     GLshort dstx, GLshort dsty,
-				     GLshort w, GLshort h );
-
-extern void intelEmitFillBlitLocked( intelContextPtr intel,
-				     GLuint cpp,
-				     GLshort dst_pitch,
-				     GLuint dst_offset,
-				     GLshort x, GLshort y, 
-				     GLshort w, GLshort h,
-				     GLuint color );
-
-
-
-
-static __inline GLuint *intelExtendInlinePrimitive( intelContextPtr intel, 
-						GLuint dwords )
+/* Unlike bmBufferData, this currently requires the buffer be mapped.
+ * Consider it a convenience function wrapping multple
+ * intel_buffer_dword() calls.
+ */
+void intel_batchbuffer_data(struct intel_batchbuffer *batch,
+			    const void *data,
+			    GLuint bytes,
+			    GLuint flags);
+
+void intel_batchbuffer_release_space(struct intel_batchbuffer *batch,
+				   GLuint bytes);
+
+GLboolean intel_batchbuffer_emit_reloc( struct intel_batchbuffer *batch,
+					GLuint buffer,
+					GLuint flags,
+					GLuint offset );
+
+/* Inline functions - might actually be better off with these
+ * non-inlined.  Certainly better off switching all command packets to
+ * be passed as structs rather than dwords, but that's a little bit of
+ * work...
+ */
+static INLINE GLuint 
+intel_batchbuffer_space( struct intel_batchbuffer *batch )
 {
-   GLuint sz = dwords * sizeof(GLuint);
-   GLuint *ptr;
+   return (BATCH_SZ - BATCH_RESERVED) - (batch->ptr - batch->map);
+}
 
-   if (intel->batch.space < sz) {
-      intelWrapInlinePrimitive( intel );
-/*       assert(intel->batch.space >= sz); */
-   }
 
-/*    assert(intel->prim.primitive != ~0); */
-   ptr = (GLuint *)intel->batch.ptr;
-   intel->batch.ptr += sz;
-   intel->batch.space -= sz;
+static INLINE void 
+intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch,
+			     GLuint dword)
+{
+   assert(batch->map);
+   assert(intel_batchbuffer_space(batch) >= 4);
+   *(GLuint *)(batch->ptr) = dword;
+   batch->ptr += 4;
+}
 
-   return ptr;
+static INLINE void 
+intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
+				GLuint sz,
+				GLuint flags)
+{
+   assert(sz < BATCH_SZ - 8);
+   if (intel_batchbuffer_space(batch) < sz ||
+       (batch->flags != 0 && flags != 0 && batch->flags != flags))
+      intel_batchbuffer_flush(batch);
+   
+   batch->flags |= flags;
 }
 
+/* Here are the crusty old macros, to be removed:
+ */
+#define BATCH_LOCALS 
+
+#define BEGIN_BATCH(n, flags) do {				\
+   assert(!intel->prim.flush);					\
+   intel_batchbuffer_require_space(intel->batch, (n)*4, flags);	\
+} while (0)
+
+#define OUT_BATCH(d)  intel_batchbuffer_emit_dword(intel->batch, d)
+
+#define OUT_RELOC(buf,flags,delta) do { 				\
+   assert((delta) >= 0);							\
+   intel_batchbuffer_emit_reloc(intel->batch, buf, flags, delta);	\
+} while (0)
+
+#define ADVANCE_BATCH() do { } while(0)
 
 
 #endif
diff --git a/src/mesa/drivers/dri/i915/intel_context.h b/src/mesa/drivers/dri/i915/intel_context.h
index 19213b7bc5..8f36f03166 100644
--- a/src/mesa/drivers/dri/i915/intel_context.h
+++ b/src/mesa/drivers/dri/i915/intel_context.h
@@ -48,88 +48,124 @@
 #define DV_PF_565  (2<<8)
 #define DV_PF_8888 (3<<8)
 
-#define INTEL_CONTEXT(ctx)	((intelContextPtr)(ctx))
+struct intel_region;
+struct intel_context;
 
-typedef struct intel_context intelContext;
-typedef struct intel_context *intelContextPtr;
-typedef struct intel_texture_object *intelTextureObjectPtr;
-
-typedef void (*intel_tri_func)(intelContextPtr, intelVertex *, intelVertex *,
+typedef void (*intel_tri_func)(struct intel_context *, intelVertex *, intelVertex *,
 							  intelVertex *);
-typedef void (*intel_line_func)(intelContextPtr, intelVertex *, intelVertex *);
-typedef void (*intel_point_func)(intelContextPtr, intelVertex *);
+typedef void (*intel_line_func)(struct intel_context *, intelVertex *, intelVertex *);
+typedef void (*intel_point_func)(struct intel_context *, intelVertex *);
 
 #define INTEL_FALLBACK_DRAW_BUFFER	 0x1
 #define INTEL_FALLBACK_READ_BUFFER	 0x2
-#define INTEL_FALLBACK_USER		 0x4
-#define INTEL_FALLBACK_NO_BATCHBUFFER	 0x8
-#define INTEL_FALLBACK_NO_TEXMEM	 0x10
+#define INTEL_FALLBACK_DEPTH_BUFFER      0x4
+#define INTEL_FALLBACK_STENCIL_BUFFER    0x8
+#define INTEL_FALLBACK_USER		 0x10
 #define INTEL_FALLBACK_RENDERMODE	 0x20
 
-extern void intelFallback( intelContextPtr intel, GLuint bit, GLboolean mode );
+extern void intelFallback( struct intel_context *intel, GLuint bit, GLboolean mode );
 #define FALLBACK( intel, bit, mode ) intelFallback( intel, bit, mode )
 
 
-#define INTEL_TEX_MAXLEVELS 10
-
 
 struct intel_texture_object
 {
-   driTextureObject    base;	/* the parent class */
+   struct gl_texture_object base; /* The "parent" object */
+
+   /* The mipmap tree must include at least these levels once
+    * validated:
+    */
+   GLuint firstLevel;
+   GLuint lastLevel;
 
-   GLuint texelBytes;
-   GLuint age;
-   GLuint Pitch;
-   GLuint Height;
-   GLuint TextureOffset;
-   GLubyte *BufAddr;   
+   /* Offset for firstLevel image:
+    */
+   GLuint textureOffset;
 
-   GLuint min_level;
-   GLuint max_level;
-   GLuint depth_pitch;
+   /* On validation any active images held in main memory or in other
+    * regions will be copied to this region and the old storage freed.
+    */
+   struct intel_mipmap_tree *mt;
+};
 
-   struct {
-      const struct gl_texture_image *image;
-      GLuint offset;       /* into BufAddr */
-      GLuint height;
-      GLuint internalFormat;
-   } image[6][INTEL_TEX_MAXLEVELS];
-
-   GLuint dirty;
-   GLuint firstLevel,lastLevel;
+
+
+struct intel_texture_image 
+{
+   struct gl_texture_image base;
+
+   /* These aren't stored in gl_texture_image 
+    */
+   GLuint level;
+   GLuint face;
+
+   /* If intelImage->mt != NULL, image data is stored here.
+    * Else if intelImage->base.Data != NULL, image is stored there.
+    * Else there is no image data.
+    */
+   struct intel_mipmap_tree *mt;
 };
 
 
+#define INTEL_MAX_FIXUP 64
+
 struct intel_context
 {
    GLcontext ctx;		/* the parent class */
 
    struct {
-      void (*destroy)( intelContextPtr intel ); 
-      void (*emit_state)( intelContextPtr intel );
-      void (*emit_invarient_state)( intelContextPtr intel );
-      void (*lost_hardware)( intelContextPtr intel );
-      void (*update_texture_state)( intelContextPtr intel );
-
-      void (*render_start)( intelContextPtr intel );
-      void (*set_color_region)( intelContextPtr intel, const intelRegion *reg );
-      void (*set_z_region)( intelContextPtr intel, const intelRegion *reg );
-      void (*update_color_z_regions)(intelContextPtr intel,
-                                     const intelRegion *colorRegion,
-                                     const intelRegion *depthRegion);
-      void (*emit_flush)( intelContextPtr intel );
-      void (*reduced_primitive_state)( intelContextPtr intel, GLenum rprim );
-
-      GLboolean (*check_vertex_size)( intelContextPtr intel, GLuint expected );
-
-      void (*clear_with_tris)( intelContextPtr intel, GLbitfield mask,
-			       GLboolean all, 
-			       GLint cx, GLint cy, GLint cw, GLint ch);
-
-      void (*rotate_window)( intelContextPtr intel,
-                             __DRIdrawablePrivate *dPriv, GLuint srcBuf);
+      void (*destroy)( struct intel_context *intel ); 
+      void (*emit_state)( struct intel_context *intel );
+      void (*lost_hardware)( struct intel_context *intel );
+      void (*update_texture_state)( struct intel_context *intel );
+
+      void (*render_start)( struct intel_context *intel );
+      void (*set_draw_region)( struct intel_context *intel, 
+			       struct intel_region *draw_region,
+			       struct intel_region *depth_region );
+
+      GLuint (*flush_cmd)( void );
+
+      void (*reduced_primitive_state)( struct intel_context *intel, GLenum rprim );
+
+      GLboolean (*check_vertex_size)( struct intel_context *intel, GLuint expected );
+
+
+      /* Metaops: 
+       */
+      void (*install_meta_state)( struct intel_context *intel );
+      void (*leave_meta_state)( struct intel_context *intel );
 
-      intelTextureObjectPtr (*alloc_tex_obj)( struct gl_texture_object *tObj );
+      void (*meta_draw_region)( struct intel_context *intel,
+				struct intel_region *draw_region,
+				struct intel_region *depth_region );
+
+      void (*meta_color_mask)( struct intel_context *intel,
+			       GLboolean );
+      
+      void (*meta_stencil_replace)( struct intel_context *intel,
+				    GLuint mask,
+				    GLuint clear );
+
+      void (*meta_depth_replace)( struct intel_context *intel );
+
+      void (*meta_texture_blend_replace)( struct intel_context *intel );
+
+      void (*meta_no_stencil_write)( struct intel_context *intel );
+      void (*meta_no_depth_write)( struct intel_context *intel );
+      void (*meta_no_texture)( struct intel_context *intel );
+
+      void (*meta_import_pixel_state)( struct intel_context *intel );
+
+      GLboolean (*meta_tex_rect_source)( struct intel_context *intel,
+					 GLuint buffer,
+					 GLuint offset,
+					 GLuint pitch,
+					 GLuint height,
+					 GLenum format,
+					 GLenum type);
+      void (*rotate_window)( struct intel_context *intel,
+                             __DRIdrawablePrivate *dPriv, GLuint srcBuf);
 
    } vtbl;
 
@@ -137,68 +173,44 @@ struct intel_context
    GLuint Fallback;
    GLuint NewGLState;
    
-   struct {
-      GLuint start_offset;
-      GLint size;
-      GLint space;
-      GLubyte *ptr;
-      GLuint counter;
-      GLuint last_emit_state;
-      GLboolean contains_geometry;
-      const char *func;
-      GLuint last_swap;
-   } batch;
-      
-   struct {
-      void *ptr;
-      GLint size;
-      GLuint offset;
-      GLuint active_buf;
-      GLuint irq_emitted;
-   } alloc;
+   GLuint last_fence;
+   GLuint last_swap_fence;
+   GLboolean last_swap_fence_retired;
+
+   struct intel_batchbuffer *batch;
 
    struct {
+      GLuint id;
       GLuint primitive;
       GLubyte *start_ptr;      
-      void (*flush)( GLcontext * );
+      void (*flush)( struct intel_context * );
    } prim;
 
    GLboolean locked;
+   char *prevLockFile;
+   int prevLockLine;
 
-   GLubyte clear_red;
-   GLubyte clear_green;
-   GLubyte clear_blue;
-   GLubyte clear_alpha;
-   GLuint ClearColor;
-   GLuint ClearDepth;
+   GLuint ClearColor565;
+   GLuint ClearColor8888;
 
+   /* Offsets of fields within the current vertex:
+    */
    GLuint coloroffset;
    GLuint specoffset;
-
-   /* Support for duplicating XYZW as WPOS parameter (crutch for I915).
-    */
    GLuint wpos_offset;
    GLuint wpos_size;
 
    struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
    GLuint vertex_attr_count;
 
-   GLfloat depth_scale;
    GLfloat polygon_offset_scale; /* dependent on depth_scale, bpp */
-   GLuint depth_clear_mask;
-   GLuint stencil_clear_mask;
 
-   GLboolean hw_stencil;
    GLboolean hw_stipple;
    
-   /* Texture object bookkeeping
+   /* AGP memory buffer manager:
     */
-   GLuint                nr_heaps;
-   driTexHeap          * texture_heaps[1];
-   driTextureObject      swapped;
-   GLuint                lastStamp;
+   struct bufmgr *bm;
 
-   struct intel_texture_object *CurrentTexObj[MAX_TEXTURE_UNITS];
 
    /* State for intelvb.c and inteltris.c.
     */
@@ -207,7 +219,14 @@ struct intel_context
    GLenum render_primitive;
    GLenum reduced_primitive;
    GLuint vertex_size;
-   unsigned char *verts;	   /* points to tnl->clipspace.vertex_buf */
+   GLubyte *verts;			/* points to tnl->clipspace.vertex_buf */
+
+
+   struct intel_region *front_region;   /* XXX FBO: obsolete */
+   struct intel_region *rotated_region;   /* XXX FBO: obsolete */
+   struct intel_region *back_region;    /* XXX FBO: obsolete */
+   struct intel_region *draw_region;    /* XXX FBO: rename to color_region */
+   struct intel_region *depth_region;   /**< currently bound depth/Z region */
 
 
    /* Fallback rasterization functions 
@@ -216,17 +235,13 @@ struct intel_context
    intel_line_func draw_line;
    intel_tri_func draw_tri;
 
-   /* Drawing buffer state
+   /* These refer to the current drawing buffer:
     */
-   intelRegion *drawRegion;  /* current drawing buffer */
-   intelRegion *readRegion;  /* current reading buffer */
-
-   int drawX;			/* origin of drawable in draw buffer */
-   int drawY;
-   GLuint numClipRects;		/* cliprects for that buffer */
+   int drawX, drawY;            /**< origin of drawing area within region */
+   GLuint numClipRects;		/**< cliprects for drawing */
    drm_clip_rect_t *pClipRects;
+   drm_clip_rect_t fboRect;     /**< cliprect for FBO rendering */
 
-   int dirtyAge;
    int perf_boxes;
 
    GLuint do_usleeps;
@@ -234,10 +249,6 @@ struct intel_context
    GLuint irqsEmitted;
    drm_i915_irq_wait_t iw;
 
-   GLboolean scissor;
-   drm_clip_rect_t draw_rect;
-   drm_clip_rect_t scissor_rect;
-
    drm_context_t hHWContext;
    drmLock *driHwLock;
    int driFd;
@@ -246,6 +257,8 @@ struct intel_context
    __DRIscreenPrivate *driScreen;
    intelScreenPrivate *intelScreen; 
    drmI830Sarea *sarea; 
+   
+   GLuint lastStamp;
 
    /**
     * Configuration cache
@@ -265,22 +278,20 @@ struct intel_context
 };
 
 
-#define DEBUG_LOCKING	1
+#define DEBUG_LOCKING 1
 
 #if DEBUG_LOCKING
-extern char *prevLockFile;
-extern int prevLockLine;
 
 #define DEBUG_LOCK()							\
    do {									\
-      prevLockFile = (__FILE__);					\
-      prevLockLine = (__LINE__);					\
+      intel->prevLockFile = (__FILE__);					\
+      intel->prevLockLine = (__LINE__);					\
    } while (0)
 
 #define DEBUG_RESET()							\
    do {									\
-      prevLockFile = 0;							\
-      prevLockLine = 0;							\
+      intel->prevLockFile = 0;						\
+      intel->prevLockLine = 0;						\
    } while (0)
 
 /* Slightly less broken way of detecting recursive locking in a
@@ -299,7 +310,8 @@ extern int prevLockLine;
 	   (DRM_LOCK_HELD | intel->hHWContext) ) {			\
 	 fprintf( stderr,						\
 		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
-		  prevLockFile, prevLockLine, __FILE__, __LINE__ );	\
+		  intel->prevLockFile, intel->prevLockLine,		\
+		  __FILE__, __LINE__ );					\
 	 abort();							\
       }									\
    } while (0)
@@ -312,7 +324,7 @@ extern int prevLockLine;
 
 #endif
 
-
+extern _glthread_Mutex lockMutex;
 
 
 /* Lock the hardware and validate our state.  
@@ -320,13 +332,14 @@ extern int prevLockLine;
 #define LOCK_HARDWARE( intel )				\
 do {							\
     char __ret=0;					\
+    _glthread_LOCK_MUTEX(lockMutex);                    \
     DEBUG_CHECK_LOCK();					\
     assert(!(intel)->locked);				\
     DRM_CAS((intel)->driHwLock, (intel)->hHWContext,	\
         (DRM_LOCK_HELD|(intel)->hHWContext), __ret);	\
     if (__ret)						\
         intelGetLock( (intel), 0 );			\
-      DEBUG_LOCK();					\
+     DEBUG_LOCK();					\
     (intel)->locked = 1;				\
 }while (0)
  
@@ -342,6 +355,7 @@ do {									\
    }									\
    DRM_UNLOCK((intel)->driFd, (intel)->driHwLock, (intel)->hHWContext);	\
    DEBUG_RESET();							\
+   _glthread_UNLOCK_MUTEX(lockMutex);                                   \
 } while (0)
 
 
@@ -350,8 +364,7 @@ do {									\
 
 #define INTEL_FIREVERTICES(intel)		\
 do {						\
-   if ((intel)->prim.flush)			\
-      (intel)->prim.flush(&(intel)->ctx);		\
+   assert(!(intel)->prim.flush);		\
 } while (0)
 
 /* ================================================================
@@ -372,20 +385,14 @@ do {						\
   ((a<<24) | (r<<16) | (g<<8) | b)
 
 
-#define INTEL_PACKCOLOR(format, r,  g,  b, a)		\
-(format == DV_PF_555 ? INTEL_PACKCOLOR1555(r,g,b,a) :	\
- (format == DV_PF_565 ? INTEL_PACKCOLOR565(r,g,b) :	\
-  (format == DV_PF_8888 ? INTEL_PACKCOLOR8888(r,g,b,a) :	\
-   0)))
-
-
 
 /* ================================================================
  * From linux kernel i386 header files, copes with odd sizes better
  * than COPY_DWORDS would:
+ * XXX Put this in src/mesa/main/imports.h ???
  */
 #if defined(i386) || defined(__i386__)
-static __inline__ void * __memcpy(void * to, const void * from, size_t n)
+static INLINE void * __memcpy(void * to, const void * from, size_t n)
 {
    int d0, d1, d2;
    __asm__ __volatile__(
@@ -447,21 +454,19 @@ extern int INTEL_DEBUG;
  * intel_context.c:
  */
 
-extern void intelInitDriverFunctions( struct dd_function_table *functions );
-
-extern GLboolean intelInitContext( intelContextPtr intel, 
+extern GLboolean intelInitContext( struct intel_context *intel, 
 				   const __GLcontextModes *mesaVis,
 				   __DRIcontextPrivate *driContextPriv,
 				   void *sharedContextPrivate,
 				   struct dd_function_table *functions );
 
-extern void intelGetLock(intelContextPtr intel, GLuint flags);
-extern void intelSetBackClipRects(intelContextPtr intel);
-extern void intelSetFrontClipRects(intelContextPtr intel);
-extern void intelWindowMoved( intelContextPtr intel );
+extern void intelGetLock(struct intel_context *intel, GLuint flags);
 
 extern void intelInitState( GLcontext *ctx );
-extern const GLubyte *intelGetString( GLcontext *ctx, GLenum name );
+extern void intelFinish( GLcontext *ctx );
+extern void intelFlush( GLcontext *ctx );
+
+extern void intelInitDriverFunctions( struct dd_function_table *functions );
 
 
 /* ================================================================
@@ -521,6 +526,8 @@ extern void intelInitStateFuncs( struct dd_function_table *functions );
 #define BLENDFACT_INV_CONST_ALPHA	0x0f
 #define BLENDFACT_MASK          	0x0f
 
+#define MI_BATCH_BUFFER_END 	(0xA<<23)
+
 
 extern int intel_translate_compare_func( GLenum func );
 extern int intel_translate_stencil_op( GLenum op );
@@ -528,19 +535,26 @@ extern int intel_translate_blend_factor( GLenum factor );
 extern int intel_translate_logic_op( GLenum opcode );
 
 
-/* ================================================================
- * intel_ioctl.c:
+/*======================================================================
+ * Inline conversion functions.  
+ * These are better-typed than the macros used previously:
  */
-extern void intel_dump_batchbuffer( long offset,
-				    int *ptr,
-				    int count );
+static INLINE struct intel_context *intel_context( GLcontext *ctx )
+{
+   return (struct intel_context *)ctx;
+}
 
+static INLINE struct intel_texture_object *intel_texture_object( struct gl_texture_object *obj )
+{
+   return (struct intel_texture_object *)obj;
+}
 
-/* ================================================================
- * intel_pixel.c:
- */	
-extern void intelInitPixelFuncs( struct dd_function_table *functions );
+static INLINE struct intel_texture_image *intel_texture_image( struct gl_texture_image *img )
+{
+   return (struct intel_texture_image *)img;
+}
 
+extern struct intel_renderbuffer *intel_renderbuffer( struct gl_renderbuffer *rb );
 
 
 #endif
diff --git a/src/mesa/drivers/dri/i915/intel_ioctl.c b/src/mesa/drivers/dri/i915/intel_ioctl.c
index d853036766..535b1b461d 100644
--- a/src/mesa/drivers/dri/i915/intel_ioctl.c
+++ b/src/mesa/drivers/dri/i915/intel_ioctl.c
@@ -38,22 +38,12 @@
 #include "intel_context.h"
 #include "intel_ioctl.h"
 #include "intel_batchbuffer.h"
+#include "intel_blit.h"
+#include "intel_regions.h"
 #include "drm.h"
+#include "intel_bufmgr.h"
 
-u_int32_t intelGetLastFrame (intelContextPtr intel) 
-{
-   int ret;
-   u_int32_t frame;
-   drm_i915_getparam_t gp;
-   
-   gp.param = I915_PARAM_LAST_DISPATCH;
-   gp.value = (int *)&frame;
-   ret = drmCommandWriteRead( intel->driFd, DRM_I915_GETPARAM,
-			      &gp, sizeof(gp) );
-   return frame;
-}
-
-int intelEmitIrqLocked( intelContextPtr intel )
+int intelEmitIrqLocked( struct intel_context *intel )
 {
    drmI830IrqEmit ie;
    int ret, seq;
@@ -62,21 +52,21 @@ int intelEmitIrqLocked( intelContextPtr intel )
 	  (DRM_LOCK_HELD|intel->hHWContext));
 
    ie.irq_seq = &seq;
-	 
+
    ret = drmCommandWriteRead( intel->driFd, DRM_I830_IRQ_EMIT, 
 			      &ie, sizeof(ie) );
    if ( ret ) {
       fprintf( stderr, "%s: drmI830IrqEmit: %d\n", __FUNCTION__, ret );
       exit(1);
-   }
-   
+   }   
+
    if (0)
       fprintf(stderr, "%s -->  %d\n", __FUNCTION__, seq );
 
    return seq;
 }
 
-void intelWaitIrq( intelContextPtr intel, int seq )
+void intelWaitIrq( struct intel_context *intel, int seq )
 {
    int ret;
       
@@ -91,570 +81,69 @@ void intelWaitIrq( intelContextPtr intel, int seq )
 
    if ( ret ) {
       fprintf( stderr, "%s: drmI830IrqWait: %d\n", __FUNCTION__, ret );
-      if (0)
-	 intel_dump_batchbuffer( intel->alloc.offset,
-				 intel->alloc.ptr,
-				 intel->alloc.size );
       exit(1);
    }
 }
 
 
-
-static void age_intel( intelContextPtr intel, int age )
-{
-   GLuint i;
-
-   for (i = 0 ; i < MAX_TEXTURE_UNITS ; i++)
-      if (intel->CurrentTexObj[i]) 
-	 intel->CurrentTexObj[i]->age = age;
-}
-
-void intel_dump_batchbuffer( long offset,
-			     int *ptr,
-			     int count )
-{
-   int i;
-   fprintf(stderr, "\n\n\nSTART BATCH (%d dwords):\n", count);
-   for (i = 0; i < count/4; i += 4) 
-      fprintf(stderr, "\t0x%x: 0x%08x 0x%08x 0x%08x 0x%08x\n", 
-	      (unsigned int)offset + i*4, ptr[i], ptr[i+1], ptr[i+2], ptr[i+3]);
-   fprintf(stderr, "END BATCH\n\n\n");
-}
-
-void intelRefillBatchLocked( intelContextPtr intel, GLboolean allow_unlock )
-{
-   GLuint last_irq = intel->alloc.irq_emitted;
-   GLuint half = intel->alloc.size / 2;
-   GLuint buf = (intel->alloc.active_buf ^= 1);
-
-   intel->alloc.irq_emitted = intelEmitIrqLocked( intel );
-
-   if (last_irq) {
-      if (allow_unlock) UNLOCK_HARDWARE( intel ); 
-      intelWaitIrq( intel, last_irq );
-      if (allow_unlock) LOCK_HARDWARE( intel ); 
-   }
-
-   if (0)
-      fprintf(stderr, "%s: now using half %d\n", __FUNCTION__, buf);
-
-   intel->batch.start_offset = intel->alloc.offset + buf * half;
-   intel->batch.ptr = (unsigned char *)intel->alloc.ptr + buf * half;
-   intel->batch.size = half - 8;
-   intel->batch.space = half - 8;
-   assert(intel->batch.space >= 0);
-}
-
-#define MI_BATCH_BUFFER_END 	(0xA<<23)
-
-
-void intelFlushBatchLocked( intelContextPtr intel, 
-			    GLboolean ignore_cliprects,
-			    GLboolean refill,
-			    GLboolean allow_unlock)
+void intel_batch_ioctl( struct intel_context *intel, 
+			GLuint start_offset,
+			GLuint used,
+			GLboolean ignore_cliprects,
+			GLboolean allow_unlock)
 {
    drmI830BatchBuffer batch;
 
    assert(intel->locked);
+   assert(used);
 
    if (0)
-      fprintf(stderr, "%s used %d of %d offset %x..%x refill %d (started in %s)\n",
+      fprintf(stderr, "%s used %d offset %x..%x ignore_cliprects %d\n",
 	      __FUNCTION__, 
-	      (intel->batch.size - intel->batch.space), 
-	      intel->batch.size,
-	      intel->batch.start_offset,
-	      intel->batch.start_offset + 
-	      (intel->batch.size - intel->batch.space), 
-	      refill,
-	      intel->batch.func);
+	      used, 
+	      start_offset,
+	      start_offset + used,
+	      ignore_cliprects);
 
    /* Throw away non-effective packets.  Won't work once we have
     * hardware contexts which would preserve statechanges beyond a
     * single buffer.
     */
    if (intel->numClipRects == 0 && !ignore_cliprects) {
-      
-      /* Without this yeild, an application with no cliprects can hog
-       * the hardware.  Without unlocking, the effect is much worse -
-       * effectively a lock-out of other contexts.
-       */
       if (allow_unlock) {
-	 UNLOCK_HARDWARE( intel );
+	 UNLOCK_HARDWARE(intel);
 	 sched_yield();
-	 LOCK_HARDWARE( intel );
+	 LOCK_HARDWARE(intel);
       }
-
-      /* Note that any state thought to have been emitted actually
-       * hasn't:
-       */
-      intel->batch.ptr -= (intel->batch.size - intel->batch.space);
-      intel->batch.space = intel->batch.size;
       intel->vtbl.lost_hardware( intel ); 
+      return;
    }
 
-   if (intel->batch.space != intel->batch.size) {
-
-      if (intel->sarea->ctxOwner != intel->hHWContext) {
-	 intel->perf_boxes |= I830_BOX_LOST_CONTEXT;
-	 intel->sarea->ctxOwner = intel->hHWContext;
-      }
-
-      batch.start = intel->batch.start_offset;
-      batch.used = intel->batch.size - intel->batch.space;
-      batch.cliprects = intel->pClipRects;
-      batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects;
-      batch.DR1 = 0;
-      batch.DR4 = ((((GLuint)intel->drawX) & 0xffff) | 
-		   (((GLuint)intel->drawY) << 16));
-      
-      if (intel->alloc.offset) {
-	 if ((batch.used & 0x4) == 0) {
-	    ((int *)intel->batch.ptr)[0] = 0;
-	    ((int *)intel->batch.ptr)[1] = MI_BATCH_BUFFER_END;
-	    batch.used += 0x8;
-	    intel->batch.ptr += 0x8;
-	 }
-	 else {
-	    ((int *)intel->batch.ptr)[0] = MI_BATCH_BUFFER_END;
-	    batch.used += 0x4;
-	    intel->batch.ptr += 0x4;
-	 }      
-      }
-
-      if (0)
- 	 intel_dump_batchbuffer( batch.start,
-				 (int *)(intel->batch.ptr - batch.used),
-				 batch.used );
-
-      intel->batch.start_offset += batch.used;
-      intel->batch.size -= batch.used;
-
-      if (intel->batch.size < 8) {
-	 refill = GL_TRUE;
-	 intel->batch.space = intel->batch.size = 0;
-      }
-      else {
-	 intel->batch.size -= 8;
-	 intel->batch.space = intel->batch.size;
-      }
-
-
-      assert(intel->batch.space >= 0);
-      assert(batch.start >= intel->alloc.offset);
-      assert(batch.start < intel->alloc.offset + intel->alloc.size);
-      assert(batch.start + batch.used > intel->alloc.offset);
-      assert(batch.start + batch.used <= 
-	     intel->alloc.offset + intel->alloc.size);
-
-
-      if (intel->alloc.offset) {
-	 if (drmCommandWrite (intel->driFd, DRM_I830_BATCHBUFFER, &batch, 
-			      sizeof(batch))) {
-	    fprintf(stderr, "DRM_I830_BATCHBUFFER: %d\n",  -errno);
-	    UNLOCK_HARDWARE(intel);
-	    exit(1);
-	 }
-      } else {
-	 drmI830CmdBuffer cmd;
-	 cmd.buf = (char *)intel->alloc.ptr + batch.start;
-	 cmd.sz = batch.used;
-	 cmd.DR1 = batch.DR1;
-	 cmd.DR4 = batch.DR4;
-	 cmd.num_cliprects = batch.num_cliprects;
-	 cmd.cliprects = batch.cliprects;
-	 
-	 if (drmCommandWrite (intel->driFd, DRM_I830_CMDBUFFER, &cmd, 
-			      sizeof(cmd))) {
-	    fprintf(stderr, "DRM_I830_CMDBUFFER: %d\n",  -errno);
-	    UNLOCK_HARDWARE(intel);
-	    exit(1);
-	 }
-      }	 
-
+   batch.start = start_offset;
+   batch.used = used;
+   batch.cliprects = intel->pClipRects;
+   batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects;
+   batch.DR1 = 0;
+   batch.DR4 = ((((GLuint)intel->drawX) & 0xffff) | 
+		(((GLuint)intel->drawY) << 16));
       
-      age_intel(intel, intel->sarea->last_enqueue);
-
-      /* FIXME: use hardware contexts to avoid 'losing' hardware after
-       * each buffer flush.
-       */
-      if (intel->batch.contains_geometry) 
-	 assert(intel->batch.last_emit_state == intel->batch.counter);
-
-      intel->batch.counter++;
-      intel->batch.contains_geometry = 0;
-      intel->batch.func = 0;
-      intel->vtbl.lost_hardware( intel );
-   }
-
-   if (refill)
-      intelRefillBatchLocked( intel, allow_unlock );
-}
-
-void intelFlushBatch( intelContextPtr intel, GLboolean refill )
-{
-   if (intel->locked) {
-      intelFlushBatchLocked( intel, GL_FALSE, refill, GL_FALSE );
-   } 
-   else {
-      LOCK_HARDWARE(intel);
-      intelFlushBatchLocked( intel, GL_FALSE, refill, GL_TRUE );
-      UNLOCK_HARDWARE(intel);
-   }
-}
-
-
-void intelWaitForIdle( intelContextPtr intel )
-{   
-   if (0)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   intel->vtbl.emit_flush( intel );
-   intelFlushBatch( intel, GL_TRUE );
-
-   /* Use an irq to wait for dma idle -- Need to track lost contexts
-    * to shortcircuit consecutive calls to this function:
-    */
-   intelWaitIrq( intel, intel->alloc.irq_emitted );
-   intel->alloc.irq_emitted = 0;
-}
-
-
-/**
- * Check if we need to rotate/warp the front color buffer to the
- * rotated screen.  We generally need to do this when we get a glFlush
- * or glFinish after drawing to the front color buffer.
- */
-static void
-intelCheckFrontRotate(GLcontext *ctx)
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   if (intel->ctx.DrawBuffer->_ColorDrawBufferMask[0] == BUFFER_BIT_FRONT_LEFT) {
-      intelScreenPrivate *screen = intel->intelScreen;
-      if (screen->current_rotation != 0) {
-         __DRIdrawablePrivate *dPriv = intel->driDrawable;
-         intelRotateWindow(intel, dPriv, BUFFER_BIT_FRONT_LEFT);
-      }
-   }
-}
-
-
-/**
- * NOT directly called via glFlush.
- */
-void intelFlush( GLcontext *ctx )
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-
-   if (intel->Fallback)
-      _swrast_flush( ctx );
-
-   INTEL_FIREVERTICES( intel );
-
-   if (intel->batch.size != intel->batch.space)
-      intelFlushBatch( intel, GL_FALSE );
-}
-
-
-/**
- * Called via glFlush.
- */
-void intelglFlush( GLcontext *ctx )
-{
-   intelFlush(ctx);
-   intelCheckFrontRotate(ctx);
-}
-
-
-void intelFinish( GLcontext *ctx  ) 
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   intelFlush( ctx );
-   intelWaitForIdle( intel );
-   intelCheckFrontRotate(ctx);
-}
-
-
-void intelClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
-		GLint cx, GLint cy, GLint cw, GLint ch)
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   const GLuint colorMask = *((GLuint *) &ctx->Color.ColorMask);
-   GLbitfield tri_mask = 0;
-   GLbitfield blit_mask = 0;
-   GLbitfield swrast_mask = 0;
-
-   if (0)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Take care of cliprects, which are handled differently for
-    * clears, etc.
-    */
-   intelFlush( &intel->ctx );
-
-   if (mask & BUFFER_BIT_FRONT_LEFT) {
-      if (colorMask == ~0) {
-	 blit_mask |= BUFFER_BIT_FRONT_LEFT;
-      } 
-      else {
-	 tri_mask |= BUFFER_BIT_FRONT_LEFT;
-      }
-   }
-
-   if (mask & BUFFER_BIT_BACK_LEFT) {
-      if (colorMask == ~0) {
-	 blit_mask |= BUFFER_BIT_BACK_LEFT;
-      } 
-      else {
-	 tri_mask |= BUFFER_BIT_BACK_LEFT;
-      }
-   }
-
-   if (mask & BUFFER_BIT_DEPTH) {
-      blit_mask |= BUFFER_BIT_DEPTH;
-   }
-
-   if (mask & BUFFER_BIT_STENCIL) {
-      if (!intel->hw_stencil) {
-	 swrast_mask |= BUFFER_BIT_STENCIL;
-      }
-      else if ((ctx->Stencil.WriteMask[0] & 0xff) != 0xff) {
-	 tri_mask |= BUFFER_BIT_STENCIL;
-      } 
-      else {
-	 blit_mask |= BUFFER_BIT_STENCIL;
-      }
-   }
-
-   swrast_mask |= (mask & BUFFER_BIT_ACCUM);
-
-   if (blit_mask) 
-      intelClearWithBlit( ctx, blit_mask, all, cx, cy, cw, ch );
-
-   if (tri_mask) 
-      intel->vtbl.clear_with_tris( intel, tri_mask, all, cx, cy, cw, ch);
-
-   if (swrast_mask)
-      _swrast_Clear( ctx, swrast_mask, all, cx, cy, cw, ch );
-}
-
-
-void
-intelRotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                  GLuint srcBuffer)
-{
-   if (intel->vtbl.rotate_window) {
-      intel->vtbl.rotate_window(intel, dPriv, srcBuffer);
-   }
-}
-
-
-void *intelAllocateAGP( intelContextPtr intel, GLsizei size )
-{
-   int region_offset;
-   drmI830MemAlloc alloc;
-   int ret;
-
-   if (0)
-      fprintf(stderr, "%s: %d bytes\n", __FUNCTION__, size);
-
-   alloc.region = I830_MEM_REGION_AGP;
-   alloc.alignment = 0;
-   alloc.size = size;
-   alloc.region_offset = &region_offset;
-
-   LOCK_HARDWARE(intel);
-
-   /* Make sure the global heap is initialized
-    */
-   if (intel->texture_heaps[0])
-      driAgeTextures( intel->texture_heaps[0] );
-
-
-   ret = drmCommandWriteRead( intel->driFd,
-			      DRM_I830_ALLOC,
-			      &alloc, sizeof(alloc));
-   
-   if (ret) {
-      fprintf(stderr, "%s: DRM_I830_ALLOC ret %d\n", __FUNCTION__, ret);
+   if (INTEL_DEBUG & DEBUG_DMA)
+      fprintf(stderr, "%s: 0x%x..0x%x DR4: %x cliprects: %d\n",
+	      __FUNCTION__, 
+	      batch.start, 
+	      batch.start + batch.used * 4,
+	      batch.DR4, batch.num_cliprects);
+#if 1
+   if (drmCommandWrite (intel->driFd, DRM_I830_BATCHBUFFER, &batch, 
+			sizeof(batch))) {
+      fprintf(stderr, "DRM_I830_BATCHBUFFER: %d\n",  -errno);
       UNLOCK_HARDWARE(intel);
-      return NULL;
-   }
-   
-   if (0)
-      fprintf(stderr, "%s: allocated %d bytes\n", __FUNCTION__, size);
-
-   /* Need to propogate this information (agp memory in use) to our
-    * local texture lru.  The kernel has already updated the global
-    * lru.  An alternative would have been to allocate memory the
-    * usual way and then notify the kernel to pin the allocation.
-    */
-   if (intel->texture_heaps[0])
-      driAgeTextures( intel->texture_heaps[0] );
-
-   UNLOCK_HARDWARE(intel);   
-
-   return (void *)((char *)intel->intelScreen->tex.map + region_offset);
-}
-
-void intelFreeAGP( intelContextPtr intel, void *pointer )
-{
-   int region_offset;
-   drmI830MemFree memfree;
-   int ret;
-
-   region_offset = (char *)pointer - (char *)intel->intelScreen->tex.map;
-
-   if (region_offset < 0 || 
-       region_offset > intel->intelScreen->tex.size) {
-      fprintf(stderr, "offset %d outside range 0..%d\n", region_offset,
-	      intel->intelScreen->tex.size);
-      return;
-   }
-
-   memfree.region = I830_MEM_REGION_AGP;
-   memfree.region_offset = region_offset;
-   
-   ret = drmCommandWrite( intel->driFd,
-			  DRM_I830_FREE,
-			  &memfree, sizeof(memfree));
-   
-   if (ret) 
-      fprintf(stderr, "%s: DRM_I830_FREE ret %d\n", __FUNCTION__, ret);
-}
-
-/* This version of AllocateMemoryMESA allocates only agp memory, and
- * only does so after the point at which the driver has been
- * initialized.
- *
- * Theoretically a valid context isn't required.  However, in this
- * implementation, it is, as I'm using the hardware lock to protect
- * the kernel data structures, and the current context to get the
- * device fd.
- */
-void *intelAllocateMemoryMESA(__DRInativeDisplay *dpy, int scrn,
-			      GLsizei size, GLfloat readfreq,
-			      GLfloat writefreq, GLfloat priority)
-{
-   GET_CURRENT_CONTEXT(ctx);
-
-   if (INTEL_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s sz %d %f/%f/%f\n", __FUNCTION__, size, readfreq, 
-	      writefreq, priority);
-
-   if (getenv("INTEL_NO_ALLOC"))
-      return NULL;
-   
-   if (!ctx || INTEL_CONTEXT(ctx) == 0) 
-      return NULL;
-   
-   return intelAllocateAGP( INTEL_CONTEXT(ctx), size );
-}
-
-
-/* Called via glXFreeMemoryMESA() */
-void intelFreeMemoryMESA(__DRInativeDisplay *dpy, int scrn, GLvoid *pointer)
-{
-   GET_CURRENT_CONTEXT(ctx);
-   if (INTEL_DEBUG & DEBUG_IOCTL) 
-      fprintf(stderr, "%s %p\n", __FUNCTION__, pointer);
-
-   if (!ctx || INTEL_CONTEXT(ctx) == 0) {
-      fprintf(stderr, "%s: no context\n", __FUNCTION__);
-      return;
-   }
-
-   intelFreeAGP( INTEL_CONTEXT(ctx), pointer );
-}
-
-/* Called via glXGetMemoryOffsetMESA() 
- *
- * Returns offset of pointer from the start of agp aperture.
- */
-GLuint intelGetMemoryOffsetMESA(__DRInativeDisplay *dpy, int scrn, 
-				const GLvoid *pointer)
-{
-   GET_CURRENT_CONTEXT(ctx);
-   intelContextPtr intel;
-
-   if (!ctx || !(intel = INTEL_CONTEXT(ctx)) ) {
-      fprintf(stderr, "%s: no context\n", __FUNCTION__);
-      return ~0;
-   }
-
-   if (!intelIsAgpMemory( intel, pointer, 0 ))
-      return ~0;
-
-   return intelAgpOffsetFromVirtual( intel, pointer );
-}
-
-
-GLboolean intelIsAgpMemory( intelContextPtr intel, const GLvoid *pointer,
-			   GLint size )
-{
-   int offset = (char *)pointer - (char *)intel->intelScreen->tex.map;
-   int valid = (size >= 0 &&
-		offset >= 0 &&
-		offset + size < intel->intelScreen->tex.size);
-
-   if (INTEL_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "intelIsAgpMemory( %p ) : %d\n", pointer, valid );
-   
-   return valid;
-}
-
-
-GLuint intelAgpOffsetFromVirtual( intelContextPtr intel, const GLvoid *pointer )
-{
-   int offset = (char *)pointer - (char *)intel->intelScreen->tex.map;
-
-   if (offset < 0 || offset > intel->intelScreen->tex.size)
-      return ~0;
-   else
-      return intel->intelScreen->tex.offset + offset;
-}
-
-
-
-
-
-/* Flip the front & back buffes
- */
-void intelPageFlip( const __DRIdrawablePrivate *dPriv )
-{
-#if 0
-   intelContextPtr intel;
-   int tmp, ret;
-
-   if (INTEL_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   intel = (intelContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-
-   if (dPriv->pClipRects) {
-      *(drm_clip_rect_t *)intel->sarea->boxes = dPriv->pClipRects[0];
-      intel->sarea->nbox = 1;
-   }
-
-   ret = drmCommandNone(intel->driFd, DRM_I830_FLIP); 
-   if (ret) {
-      fprintf(stderr, "%s: %d\n", __FUNCTION__, ret);
-      UNLOCK_HARDWARE( intel );
       exit(1);
    }
-
-   tmp = intel->sarea->last_enqueue;
-   intelRefillBatchLocked( intel );
-   UNLOCK_HARDWARE( intel );
-
-
-   intelSetDrawBuffer( &intel->ctx, intel->ctx.Color.DriverDrawBuffer );
 #endif
+      
+   /* FIXME: use hardware contexts to avoid 'losing' hardware after
+    * each buffer flush.
+    */
+   intel->vtbl.lost_hardware( intel );
 }
diff --git a/src/mesa/drivers/dri/i915/intel_ioctl.h b/src/mesa/drivers/dri/i915/intel_ioctl.h
index 099a7e1a44..d1e7f9ade6 100644
--- a/src/mesa/drivers/dri/i915/intel_ioctl.h
+++ b/src/mesa/drivers/dri/i915/intel_ioctl.h
@@ -30,44 +30,12 @@
 
 #include "intel_context.h"
 
-extern void intelWaitAgeLocked( intelContextPtr intel, int age, GLboolean unlock );
-
-extern void intelClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
-		       GLint cx, GLint cy, GLint cw, GLint ch);
-
-extern void intelPageFlip( const __DRIdrawablePrivate *dpriv );
-
-extern void intelRotateWindow(intelContextPtr intel,
-                              __DRIdrawablePrivate *dPriv, GLuint srcBuffer);
-
-extern void intelWaitForIdle( intelContextPtr intel );
-extern void intelFlushBatch( intelContextPtr intel, GLboolean refill );
-extern void intelFlushBatchLocked( intelContextPtr intel,
-				   GLboolean ignore_cliprects,
-				   GLboolean refill,
-				   GLboolean allow_unlock);
-extern void intelRefillBatchLocked( intelContextPtr intel, GLboolean allow_unlock );
-extern void intelFinish( GLcontext *ctx );
-extern void intelFlush( GLcontext *ctx );
-extern void intelglFlush( GLcontext *ctx );
-
-extern void *intelAllocateAGP( intelContextPtr intel, GLsizei size );
-extern void intelFreeAGP( intelContextPtr intel, void *pointer );
-
-extern void *intelAllocateMemoryMESA( __DRInativeDisplay *dpy, int scrn, 
-				      GLsizei size, GLfloat readfreq,
-				      GLfloat writefreq, GLfloat priority );
-
-extern void intelFreeMemoryMESA( __DRInativeDisplay *dpy, int scrn, 
-				 GLvoid *pointer );
-
-extern GLuint intelGetMemoryOffsetMESA( __DRInativeDisplay *dpy, int scrn, const GLvoid *pointer );
-extern GLboolean intelIsAgpMemory( intelContextPtr intel, const GLvoid *pointer,
-				  GLint size );
-
-extern GLuint intelAgpOffsetFromVirtual( intelContextPtr intel, const GLvoid *p );
-
-extern void intelWaitIrq( intelContextPtr intel, int seq );
-extern u_int32_t intelGetLastFrame (intelContextPtr intel);
-extern int intelEmitIrqLocked( intelContextPtr intel );
+void intelWaitIrq( struct intel_context *intel, int seq );
+int intelEmitIrqLocked( struct intel_context *intel );
+
+void intel_batch_ioctl( struct intel_context *intel, 
+			GLuint start_offset,
+			GLuint used,
+			GLboolean ignore_cliprects,
+			GLboolean allow_unlock);
 #endif
diff --git a/src/mesa/drivers/dri/i915/intel_pixel.c b/src/mesa/drivers/dri/i915/intel_pixel.c
index 2af38541d9..9b2d7d9d2e 100644
--- a/src/mesa/drivers/dri/i915/intel_pixel.c
+++ b/src/mesa/drivers/dri/i915/intel_pixel.c
@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -12,7 +12,7 @@
  * the following conditions:
  * 
  * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
+ * next paragraph) shall be included in all copies or substantial portionsalloc
  * of the Software.
  * 
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
@@ -25,487 +25,98 @@
  * 
  **************************************************************************/
 
-#include "glheader.h"
 #include "enums.h"
-#include "mtypes.h"
-#include "macros.h"
+#include "state.h"
 #include "swrast/swrast.h"
 
-#include "intel_screen.h"
 #include "intel_context.h"
-#include "intel_ioctl.h"
-#include "intel_batchbuffer.h"
-
-
-
-static GLboolean
-check_color( const GLcontext *ctx, GLenum type, GLenum format,
-	     const struct gl_pixelstore_attrib *packing,
-	     const void *pixels, GLint sz, GLint pitch )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   GLuint cpp = intel->intelScreen->cpp;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (	(pitch & 63) ||
-	ctx->_ImageTransferState ||
-	packing->SwapBytes ||
-	packing->LsbFirst) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: failed 1\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   if ( type == GL_UNSIGNED_INT_8_8_8_8_REV && 
-	cpp == 4 && 
-	format == GL_BGRA ) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: passed 2\n", __FUNCTION__);
-      return GL_TRUE;
-   }
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s: failed\n", __FUNCTION__);
-
-   return GL_FALSE;
-}
-
-static GLboolean
-check_color_per_fragment_ops( const GLcontext *ctx )
-{
-   int result;
-   result = (!(     ctx->Color.AlphaEnabled || 
-		    ctx->Depth.Test ||
-		    ctx->Fog.Enabled ||
-		    ctx->Scissor.Enabled ||
-		    ctx->Stencil.Enabled ||
-		    !ctx->Color.ColorMask[0] ||
-		    !ctx->Color.ColorMask[1] ||
-		    !ctx->Color.ColorMask[2] ||
-		    !ctx->Color.ColorMask[3] ||
-		    ctx->Color.ColorLogicOpEnabled ||
-		    ctx->Texture._EnabledUnits
-           ) &&
-	   ctx->Current.RasterPosValid);
-   
-   return result;
-}
+#include "intel_pixel.h"
+#include "intel_regions.h"
 
 
 /**
- * Clip the given rectangle against the buffer's bounds (including scissor).
- * \param size returns the 
- * \return GL_TRUE if any pixels remain, GL_FALSE if totally clipped.
- *
- * XXX Replace this with _mesa_clip_drawpixels() and _mesa_clip_readpixels()
- * from Mesa 6.4.  We shouldn't apply scissor for ReadPixels.
+ * Check if any fragment operations are in effect which might effect
+ * glDraw/CopyPixels.
  */
-static GLboolean
-clip_pixelrect( const GLcontext *ctx,
-		const GLframebuffer *buffer,
-		GLint *x, GLint *y,
-		GLsizei *width, GLsizei *height)
+GLboolean intel_check_blit_fragment_ops( GLcontext *ctx )
 {
-   /* left clipping */
-   if (*x < buffer->_Xmin) {
-      *width -= (buffer->_Xmin - *x);
-      *x = buffer->_Xmin;
-   }
-
-   /* right clipping */
-   if (*x + *width > buffer->_Xmax)
-      *width -= (*x + *width - buffer->_Xmax - 1);
-
-   if (*width <= 0)
-      return GL_FALSE;
-
-   /* bottom clipping */
-   if (*y < buffer->_Ymin) {
-      *height -= (buffer->_Ymin - *y);
-      *y = buffer->_Ymin;
-   }
-
-   /* top clipping */
-   if (*y + *height > buffer->_Ymax)
-      *height -= (*y + *height - buffer->_Ymax - 1);
-
-   if (*height <= 0)
-      return GL_FALSE;
-
-   return GL_TRUE;
-}
+   if (ctx->NewState)
+      _mesa_update_state(ctx);
 
-
-/**
- * Compute intersection of a clipping rectangle and pixel rectangle,
- * returning results in x/y/w/hOut vars.
- * \return GL_TRUE if there's intersection, GL_FALSE if disjoint.
- */
-static INLINE GLboolean
-intersect_region(const drm_clip_rect_t *box,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLint *xOut, GLint *yOut, GLint *wOut, GLint *hOut)
-{
-   GLint bx = box->x1;
-   GLint by = box->y1;
-   GLint bw = box->x2 - bx;
-   GLint bh = box->y2 - by;
-
-   if (bx < x) bw -= x - bx, bx = x;
-   if (by < y) bh -= y - by, by = y;
-   if (bx + bw > x + width) bw = x + width - bx;
-   if (by + bh > y + height) bh = y + height - by;
-
-   *xOut = bx;
-   *yOut = by;
-   *wOut = bw;
-   *hOut = bh;
-
-   if (bw <= 0) return GL_FALSE;
-   if (bh <= 0) return GL_FALSE;
-
-   return GL_TRUE;
-}
-
-
-
-static GLboolean
-intelTryReadPixels( GLcontext *ctx,
-		  GLint x, GLint y, GLsizei width, GLsizei height,
-		  GLenum format, GLenum type,
-		  const struct gl_pixelstore_attrib *pack,
-		  GLvoid *pixels )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   GLint size = 0; /* not really used */
-   GLint pitch = pack->RowLength ? pack->RowLength : width;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Only accelerate reading to agp buffers.
-    */
-   if ( !intelIsAgpMemory(intel, pixels, 
-			pitch * height * intel->intelScreen->cpp ) ) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: dest not agp\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   /* Need GL_PACK_INVERT_MESA to cope with upsidedown results from
-    * blitter:
-    */
-   if (!pack->Invert) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: MESA_PACK_INVERT not set\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   if (!check_color(ctx, type, format, pack, pixels, size, pitch))
-      return GL_FALSE;
-
-   switch ( intel->intelScreen->cpp ) {
-   case 4:
-      break;
-   default:
-      return GL_FALSE;
-   }
-
-
-   /* Although the blits go on the command buffer, need to do this and
-    * fire with lock held to guarentee cliprects and drawing offset are
-    * correct.
-    *
-    * This is an unusual situation however, as the code which flushes
-    * a full command buffer expects to be called unlocked.  As a
-    * workaround, immediately flush the buffer on aquiring the lock.
+   /* XXX Note: Scissor could be done with the blitter:
     */
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-   {
-      __DRIdrawablePrivate *dPriv = intel->driDrawable;
-      int nbox = dPriv->numClipRects;
-      int src_offset = intel->readRegion->offset;
-      int src_pitch = intel->intelScreen->front.pitch;
-      int dst_offset = intelAgpOffsetFromVirtual( intel, pixels);
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      int i;
-
-      assert(dst_offset != ~0);  /* should have been caught above */
-
-      if (!clip_pixelrect(ctx, ctx->ReadBuffer, &x, &y, &width, &height)) {
-	 UNLOCK_HARDWARE( intel );
-	 if (INTEL_DEBUG & DEBUG_PIXEL)
-	    fprintf(stderr, "%s totally clipped -- nothing to do\n",
-		    __FUNCTION__);
-	 return GL_TRUE;
-      }
-
-      /* convert to screen coords (y=0=top) */
-      y = dPriv->h - y - height;
-      x += dPriv->x;
-      y += dPriv->y;
-
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "readpixel blit src_pitch %d dst_pitch %d\n",
-		 src_pitch, pitch);
-
-      /* We don't really have to do window clipping for readpixels.
-       * The OpenGL spec says that pixels read from outside the
-       * visible window region (pixel ownership) have undefined value.
-       */
-      for (i = 0 ; i < nbox ; i++)
-      {
-         GLint bx, by, bw, bh;
-         if (intersect_region(box+i, x, y, width, height,
-                              &bx, &by, &bw, &bh)) {
-            intelEmitCopyBlitLocked( intel,
-                                     intel->intelScreen->cpp,
-                                     src_pitch, src_offset,
-                                     pitch, dst_offset,
-                                     bx, by,
-                                     bx - x, by - y,
-                                     bw, bh );
-         }
-      }
-   }
-   UNLOCK_HARDWARE( intel );
-   intelFinish( &intel->ctx );
-
-   return GL_TRUE;
+   return !(ctx->_ImageTransferState ||
+	    ctx->Color.AlphaEnabled || 
+	    ctx->Depth.Test ||
+	    ctx->Fog.Enabled ||
+	    ctx->Scissor.Enabled ||
+	    ctx->Stencil.Enabled ||
+	    !ctx->Color.ColorMask[0] ||
+	    !ctx->Color.ColorMask[1] ||
+	    !ctx->Color.ColorMask[2] ||
+	    !ctx->Color.ColorMask[3] ||
+	    ctx->Color.ColorLogicOpEnabled ||
+	    ctx->Texture._EnabledUnits ||
+	    ctx->FragmentProgram._Enabled);
 }
 
-static void
-intelReadPixels( GLcontext *ctx,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLenum format, GLenum type,
-		 const struct gl_pixelstore_attrib *pack,
-		 GLvoid *pixels )
-{
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!intelTryReadPixels( ctx, x, y, width, height, format, type, pack, 
-                            pixels))
-      _swrast_ReadPixels( ctx, x, y, width, height, format, type, pack, 
-			  pixels);
-}
 
-
-
-
-static void do_draw_pix( GLcontext *ctx,
-			 GLint x, GLint y, GLsizei width, GLsizei height,
-			 GLint pitch,
-			 const void *pixels,
-			 GLuint dest )
+GLboolean intel_check_meta_tex_fragment_ops( GLcontext *ctx )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   drm_clip_rect_t *box = dPriv->pClipRects;
-   int nbox = dPriv->numClipRects;
-   int i;
-   int src_offset = intelAgpOffsetFromVirtual( intel, pixels);
-   int src_pitch = pitch;
-
-   assert(src_offset != ~0);  /* should be caught earlier */
+   if (ctx->NewState)
+      _mesa_update_state(ctx);
 
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-   if (ctx->DrawBuffer)
-   {
-      y -= height;			/* cope with pixel zoom */
-   
-      if (!clip_pixelrect(ctx, ctx->DrawBuffer,
-			  &x, &y, &width, &height)) {
-	 UNLOCK_HARDWARE( intel );
-	 return;
-      }
-
-      y = dPriv->h - y - height; 	/* convert from gl to hardware coords */
-      x += dPriv->x;
-      y += dPriv->y;
-
-      for (i = 0 ; i < nbox ; i++ )
-      {
-	 GLint bx, by, bw, bh;
-	 if (intersect_region(box + i, x, y, width, height,
-			      &bx, &by, &bw, &bh)) {
-            intelEmitCopyBlitLocked( intel,
-                                     intel->intelScreen->cpp,
-                                     src_pitch, src_offset,
-                                     intel->intelScreen->front.pitch,
-                                     intel->drawRegion->offset,
-                                     bx - x, by - y,
-                                     bx, by,
-                                     bw, bh );
-         }
-      }
-   }
-   UNLOCK_HARDWARE( intel );
-   intelFinish( &intel->ctx );
+   /* Some of _ImageTransferState (scale, bias) could be done with
+    * fragment programs on i915.
+    */
+   return !(ctx->_ImageTransferState ||
+	    ctx->Fog.Enabled ||	/* not done yet */
+	    ctx->Texture._EnabledUnits ||
+	    ctx->FragmentProgram._Enabled);
 }
 
-
-
-static GLboolean
-intelTryDrawPixels( GLcontext *ctx,
-		  GLint x, GLint y, GLsizei width, GLsizei height,
-		  GLenum format, GLenum type,
-		  const struct gl_pixelstore_attrib *unpack,
-		  const GLvoid *pixels )
+/* The intel_region struct doesn't really do enough to capture the
+ * format of the pixels in the region.  For now this code assumes that
+ * the region is a display surface and hence is either ARGB8888 or
+ * RGB565.
+ * XXX FBO: If we'd pass in the intel_renderbuffer instead of region, we'd
+ * know the buffer's pixel format.
+ *
+ * \param format  as given to glDraw/ReadPixels
+ * \param type  as given to glDraw/ReadPixels
+ */
+GLboolean intel_check_blit_format( struct intel_region *region,
+				   GLenum format, GLenum type )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   GLint pitch = unpack->RowLength ? unpack->RowLength : width;
-   GLuint dest;
-   GLuint cpp = intel->intelScreen->cpp;
-   GLint size = width * pitch * cpp;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   switch (format) {
-   case GL_RGB:
-   case GL_RGBA:
-   case GL_BGRA:
-      dest = intel->drawRegion->offset;
-
-      /* Planemask doesn't have full support in blits.
-       */
-      if (!ctx->Color.ColorMask[RCOMP] ||
-	  !ctx->Color.ColorMask[GCOMP] ||
-	  !ctx->Color.ColorMask[BCOMP] ||
-	  !ctx->Color.ColorMask[ACOMP]) {
-	 if (INTEL_DEBUG & DEBUG_PIXEL)
-	    fprintf(stderr, "%s: planemask\n", __FUNCTION__);
-	 return GL_FALSE;	
-      }
-
-      /* Can't do conversions on agp reads/draws. 
-       */
-      if ( !intelIsAgpMemory( intel, pixels, size ) ) {
-	 if (INTEL_DEBUG & DEBUG_PIXEL)
-	    fprintf(stderr, "%s: not agp memory\n", __FUNCTION__);
-	 return GL_FALSE;
-      }
-
-      if (!check_color(ctx, type, format, unpack, pixels, size, pitch)) {
-	 return GL_FALSE;
-      }
-      if (!check_color_per_fragment_ops(ctx)) {
-	 return GL_FALSE;
-      }
-
-      if (ctx->Pixel.ZoomX != 1.0F ||
-	  ctx->Pixel.ZoomY != -1.0F)
-	 return GL_FALSE;
-      break;
-
-   default:
-      return GL_FALSE;
-   }
-
-   if ( intelIsAgpMemory(intel, pixels, size) )
-   {
-      do_draw_pix( ctx, x, y, width, height, pitch, pixels, dest );
+   if (region->cpp == 4 &&
+       (type == GL_UNSIGNED_INT_8_8_8_8_REV ||
+	type == GL_UNSIGNED_BYTE) &&        
+       format == GL_BGRA ) {
       return GL_TRUE;
    }
-   else if (0)
-   {
-      /* Pixels is in regular memory -- get dma buffers and perform
-       * upload through them.  No point doing this for regular uploads
-       * but once we remove some of the restrictions above (colormask,
-       * pixelformat conversion, zoom?, etc), this could be a win.
-       */
+   
+   if (region->cpp == 2 && 
+       type == GL_UNSIGNED_SHORT_5_6_5_REV && 
+       format == GL_BGR ) {
+      return GL_TRUE;
    }
-   else
-      return GL_FALSE;
 
-   return GL_FALSE;
-}
-
-static void
-intelDrawPixels( GLcontext *ctx,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLenum format, GLenum type,
-		 const struct gl_pixelstore_attrib *unpack,
-		 const GLvoid *pixels )
-{
    if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!intelTryDrawPixels( ctx, x, y, width, height, format, type,
-                            unpack, pixels ))
-      _swrast_DrawPixels( ctx, x, y, width, height, format, type,
-			  unpack, pixels );
-}
-
-
-
+      fprintf(stderr, "%s: bad format for blit (cpp %d, type %s format %s)\n", 
+	      __FUNCTION__, region->cpp, 
+	      _mesa_lookup_enum_by_nr(type),
+	      _mesa_lookup_enum_by_nr(format));
 
-/**
- * Implement glCopyPixels for the front color buffer (or back buffer Pixmap)
- * for the color buffer.  Don't support zooming, pixel transfer, etc.
- * We do support copying from one window to another, ala glXMakeCurrentRead.
- */
-static void
-intelCopyPixels( GLcontext *ctx,
-		 GLint srcx, GLint srcy, GLsizei width, GLsizei height,
-		 GLint destx, GLint desty, GLenum type )
-{
-#if 0
-   const XMesaContext xmesa = XMESA_CONTEXT(ctx);
-   const SWcontext *swrast = SWRAST_CONTEXT( ctx );
-   XMesaDisplay *dpy = xmesa->xm_visual->display;
-   const XMesaDrawable drawBuffer = xmesa->xm_draw_buffer->buffer;
-   const XMesaDrawable readBuffer = xmesa->xm_read_buffer->buffer;
-   const XMesaGC gc = xmesa->xm_draw_buffer->gc;
-
-   ASSERT(dpy);
-   ASSERT(gc);
-
-   if (drawBuffer &&  /* buffer != 0 means it's a Window or Pixmap */
-       readBuffer &&
-       type == GL_COLOR &&
-       (swrast->_RasterMask & ~CLIP_BIT) == 0 && /* no blend, z-test, etc */
-       ctx->_ImageTransferState == 0 &&  /* no color tables, scale/bias, etc */
-       ctx->Pixel.ZoomX == 1.0 &&        /* no zooming */
-       ctx->Pixel.ZoomY == 1.0) {
-      /* Note: we don't do any special clipping work here.  We could,
-       * but X will do it for us.
-       */
-      srcy = FLIP(xmesa->xm_read_buffer, srcy) - height + 1;
-      desty = FLIP(xmesa->xm_draw_buffer, desty) - height + 1;
-      XCopyArea(dpy, readBuffer, drawBuffer, gc,
-                srcx, srcy, width, height, destx, desty);
-   }
-#else
-   _swrast_CopyPixels(ctx, srcx, srcy, width, height, destx, desty, type );
-#endif
+   return GL_FALSE;
 }
 
 
-
-
 void intelInitPixelFuncs( struct dd_function_table *functions )
 {
-   /* Pixel path fallbacks.
-    */
    functions->Accum = _swrast_Accum;
    functions->Bitmap = _swrast_Bitmap;
    functions->CopyPixels = intelCopyPixels;
-
-   if (!getenv("INTEL_NO_BLITS")) {
-      functions->ReadPixels = intelReadPixels;  
-      functions->DrawPixels = intelDrawPixels; 
-   }
-   else {
-      functions->ReadPixels = _swrast_ReadPixels;
-      functions->DrawPixels = _swrast_DrawPixels;
-   }
+   functions->ReadPixels = intelReadPixels;  
+   functions->DrawPixels = intelDrawPixels; 
 }
+
diff --git a/src/mesa/drivers/dri/i915/intel_render.c b/src/mesa/drivers/dri/i915/intel_render.c
index d9438ba0fd..075b453848 100644
--- a/src/mesa/drivers/dri/i915/intel_render.c
+++ b/src/mesa/drivers/dri/i915/intel_render.c
@@ -106,24 +106,29 @@ static const int scale_prim[GL_POLYGON+1] = {
 };
 
 
-static void intelDmaPrimitive( intelContextPtr intel, GLenum prim )
+static void intelDmaPrimitive( struct intel_context *intel, GLenum prim )
 {
    if (0) fprintf(stderr, "%s %s\n", __FUNCTION__, _mesa_lookup_enum_by_nr(prim));
    INTEL_FIREVERTICES(intel);
    intel->vtbl.reduced_primitive_state( intel, reduced_prim[prim] );
-   intelStartInlinePrimitive( intel, hw_prim[prim] );
+   intelStartInlinePrimitive( intel, hw_prim[prim], INTEL_BATCH_CLIPRECTS );
 }
 
 
-#define LOCAL_VARS intelContextPtr intel = INTEL_CONTEXT(ctx)
+#define LOCAL_VARS struct intel_context *intel = intel_context(ctx)
 #define INIT( prim ) 				\
 do {						\
    intelDmaPrimitive( intel, prim );		\
 } while (0)
-#define FLUSH()  INTEL_FIREVERTICES( intel )
+
+#define FLUSH()  				\
+do {						\
+   if (intel->prim.flush) 			\
+      intel->prim.flush(intel);			\
+} while (0)
 
 #define GET_SUBSEQUENT_VB_MAX_VERTS() \
-  (((intel->alloc.size / 2) - 1500) / (intel->vertex_size*4))
+  ((BATCH_SZ - 1500) / (intel->vertex_size*4))
 #define GET_CURRENT_VB_MAX_VERTS() GET_SUBSEQUENT_VB_MAX_VERTS()
 
 #define ALLOC_VERTS( nr ) \
@@ -142,7 +147,7 @@ do {						\
 
 /* Heuristic to choose between the two render paths:  
  */
-static GLboolean choose_render( intelContextPtr intel,
+static GLboolean choose_render( struct intel_context *intel,
 				struct vertex_buffer *VB )
 {
    int vertsz = intel->vertex_size;
@@ -194,7 +199,7 @@ static GLboolean choose_render( intelContextPtr intel,
 static GLboolean intel_run_render( GLcontext *ctx, 
 				 struct tnl_pipeline_stage *stage )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    GLuint i;
@@ -225,6 +230,9 @@ static GLboolean intel_run_render( GLcontext *ctx,
    }
       
    tnl->Driver.Render.Finish( ctx );
+   
+   if (intel->prim.flush)
+      intel->prim.flush(intel);
 
    return GL_FALSE;     /* finished the pipe */
 }
diff --git a/src/mesa/drivers/dri/i915/intel_screen.c b/src/mesa/drivers/dri/i915/intel_screen.c
index a87de17304..6e42c62cc4 100644
--- a/src/mesa/drivers/dri/i915/intel_screen.c
+++ b/src/mesa/drivers/dri/i915/intel_screen.c
@@ -38,10 +38,12 @@
 
 #include "intel_screen.h"
 
+#include "intel_buffers.h"
 #include "intel_tex.h"
 #include "intel_span.h"
 #include "intel_tris.h"
 #include "intel_ioctl.h"
+#include "intel_fbo.h"
 
 #include "i830_dri.h"
 
@@ -283,16 +285,13 @@ static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
    intelScreen->cpp = gDRIPriv->cpp;
 
    switch (gDRIPriv->bitsPerPixel) {
-   case 15: intelScreen->fbFormat = DV_PF_555; break;
    case 16: intelScreen->fbFormat = DV_PF_565; break;
    case 32: intelScreen->fbFormat = DV_PF_8888; break;
+   default: exit(1); break;
    }
 			 
    intelUpdateScreenFromSAREA(intelScreen, sarea);
 
-   if (0)
-      intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv);
-
    if (!intelMapScreenRegions(sPriv)) {
       fprintf(stderr,"\nERROR!  mapping regions\n");
       _mesa_free(intelScreen);
@@ -300,6 +299,28 @@ static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
       return GL_FALSE;
    }
 
+#if 0
+
+   /*
+    * FIXME: Remove this code and its references.
+    */
+
+   intelScreen->tex.offset = gDRIPriv->textureOffset;
+   intelScreen->logTextureGranularity = gDRIPriv->logTextureGranularity;
+   intelScreen->tex.handle = gDRIPriv->textures;
+   intelScreen->tex.size = gDRIPriv->textureSize;
+
+#else
+   intelScreen->tex.offset = 0;
+   intelScreen->logTextureGranularity = 0;
+   intelScreen->tex.handle = 0;
+   intelScreen->tex.size = 0;
+#endif
+		 
+   intelScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+   
+   if (1) intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv);
+
    intelScreen->drmMinor = sPriv->drmMinor;
 
    /* Determine if IRQs are active? */
@@ -340,14 +361,8 @@ static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
       (*glx_enable_extension)( psc, "GLX_MESA_swap_control" );
       (*glx_enable_extension)( psc, "GLX_MESA_swap_frame_usage" );
       (*glx_enable_extension)( psc, "GLX_SGI_make_current_read" );
-      (*glx_enable_extension)( psc, "GLX_MESA_allocate_memory" );
-      (*glx_enable_extension)( psc, "GLX_MESA_copy_sub_buffer" );
    }
    
-   sPriv->psc->allocateMemory = (void *) intelAllocateMemoryMESA;
-   sPriv->psc->freeMemory     = (void *) intelFreeMemoryMESA;
-   sPriv->psc->memoryOffset   = (void *) intelGetMemoryOffsetMESA;
-
    return GL_TRUE;
 }
 		
@@ -362,6 +377,9 @@ static void intelDestroyScreen(__DRIscreenPrivate *sPriv)
 }
 
 
+/**
+ * This is called when we need to set up GL rendering to a new X window.
+ */
 static GLboolean intelCreateBuffer( __DRIscreenPrivate *driScrnPriv,
 				    __DRIdrawablePrivate *driDrawPriv,
 				    const __GLcontextModes *mesaVis,
@@ -374,70 +392,71 @@ static GLboolean intelCreateBuffer( __DRIscreenPrivate *driScrnPriv,
    } else {
       GLboolean swStencil = (mesaVis->stencilBits > 0 && 
 			     mesaVis->depthBits != 24);
+      GLenum rgbFormat = (mesaVis->redBits == 5 ? GL_RGB5 : GL_RGBA8);
 
       struct gl_framebuffer *fb = _mesa_create_framebuffer(mesaVis);
 
+      /* setup the hardware-based renderbuffers */
       {
-         driRenderbuffer *frontRb
-            = driNewRenderbuffer(GL_RGBA,
-                                 screen->front.map,
-                                 screen->cpp,
-                                 screen->front.offset, screen->front.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(frontRb, mesaVis);
+         struct intel_renderbuffer *frontRb
+            = intel_create_renderbuffer(rgbFormat,
+                                        screen->width, screen->height,
+                                        screen->front.offset,
+                                        screen->front.pitch,
+                                        screen->cpp,
+                                        screen->front.map);
+         intel_set_span_functions(&frontRb->Base);
          _mesa_add_renderbuffer(fb, BUFFER_FRONT_LEFT, &frontRb->Base);
       }
 
       if (mesaVis->doubleBufferMode) {
-         driRenderbuffer *backRb
-            = driNewRenderbuffer(GL_RGBA,
-                                 screen->back.map,
-                                 screen->cpp,
-                                 screen->back.offset, screen->back.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(backRb, mesaVis);
+         struct intel_renderbuffer *backRb
+            = intel_create_renderbuffer(rgbFormat,
+                                        screen->width, screen->height,
+                                        screen->back.offset,
+                                        screen->back.pitch,
+                                        screen->cpp,
+                                        screen->back.map);
+         intel_set_span_functions(&backRb->Base);
          _mesa_add_renderbuffer(fb, BUFFER_BACK_LEFT, &backRb->Base);
       }
 
-      if (mesaVis->depthBits == 16) {
-         driRenderbuffer *depthRb
-            = driNewRenderbuffer(GL_DEPTH_COMPONENT16,
-                                 screen->depth.map,
-                                 screen->cpp,
-                                 screen->depth.offset, screen->depth.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(depthRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
+      if (mesaVis->depthBits == 24 && mesaVis->stencilBits == 8) {
+         /* combined depth/stencil buffer */
+         struct intel_renderbuffer *depthStencilRb
+            = intel_create_renderbuffer(
+                                        GL_DEPTH24_STENCIL8_EXT,
+                                        screen->width, screen->height,
+                                        screen->depth.offset,
+                                        screen->depth.pitch,
+                                        screen->cpp, /* 4! */
+                                        screen->depth.map);
+         intel_set_span_functions(&depthStencilRb->Base);
+         /* note: bind RB to two attachment points */
+         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthStencilRb->Base);
+         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &depthStencilRb->Base);
       }
-      else if (mesaVis->depthBits == 24) {
-         driRenderbuffer *depthRb
-            = driNewRenderbuffer(GL_DEPTH_COMPONENT24,
-                                 screen->depth.map,
-                                 screen->cpp,
-                                 screen->depth.offset, screen->depth.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(depthRb, mesaVis);
+      else if (mesaVis->depthBits == 16) {
+         /* just 16-bit depth buffer, no hw stencil */
+         struct intel_renderbuffer *depthRb
+            = intel_create_renderbuffer(GL_DEPTH_COMPONENT16,
+                                        screen->width, screen->height,
+                                        screen->depth.offset,
+                                        screen->depth.pitch,
+                                        screen->cpp, /* 2! */
+                                        screen->depth.map);
+         intel_set_span_functions(&depthRb->Base);
          _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
       }
 
-      if (mesaVis->stencilBits > 0 && !swStencil) {
-         driRenderbuffer *stencilRb
-            = driNewRenderbuffer(GL_STENCIL_INDEX8_EXT,
-                                 screen->depth.map,
-                                 screen->cpp,
-                                 screen->depth.offset, screen->depth.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(stencilRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &stencilRb->Base);
-      }
-
+      /* now add any/all software-based renderbuffers we may need */
       _mesa_add_soft_renderbuffers(fb,
-                                   GL_FALSE, /* color */
-                                   GL_FALSE, /* depth */
+                                   GL_FALSE, /* never sw color */
+                                   GL_FALSE, /* never sw depth */
                                    swStencil,
                                    mesaVis->accumRedBits > 0,
-                                   GL_FALSE, /* alpha */
-                                   GL_FALSE /* aux */);
+                                   GL_FALSE, /* never sw alpha */
+                                   GL_FALSE  /* never sw aux */);
       driDrawPriv->driverPrivate = (void *) fb;
 
       return (driDrawPriv->driverPrivate != NULL);
@@ -456,7 +475,7 @@ static void intelDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
 static int
 intelGetSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
 {
-   intelContextPtr intel;
+   struct intel_context *intel;
 
    if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL)
 	|| (dPriv->driContextPriv->driverPrivate == NULL)
@@ -500,6 +519,8 @@ static GLboolean intelCreateContext( const __GLcontextModes *mesaVis,
    intelScreenPrivate *intelScreen = (intelScreenPrivate *)sPriv->private;
 
    switch (intelScreen->deviceID) {
+      /* Don't deal with i830 until texture work complete:
+       */
    case PCI_CHIP_845_G:
    case PCI_CHIP_I830_M:
    case PCI_CHIP_I855_GM:
@@ -646,7 +667,7 @@ void * __driCreateNewScreen_20050727( __DRInativeDisplay *dpy, int scrn, __DRIsc
    __DRIscreenPrivate *psp;
    static const __DRIversion ddx_expected = { 1, 5, 0 };
    static const __DRIversion dri_expected = { 4, 0, 0 };
-   static const __DRIversion drm_expected = { 1, 4, 0 };
+   static const __DRIversion drm_expected = { 1, 5, 1 };
 
    dri_interface = interface;
 
diff --git a/src/mesa/drivers/dri/i915/intel_screen.h b/src/mesa/drivers/dri/i915/intel_screen.h
index 24cfd9bf8b..c72d1054e6 100644
--- a/src/mesa/drivers/dri/i915/intel_screen.h
+++ b/src/mesa/drivers/dri/i915/intel_screen.h
@@ -29,13 +29,15 @@
 #define _INTEL_INIT_H_
 
 #include <sys/time.h>
-#include "xmlconfig.h"
 #include "dri_util.h"
 #include "intel_rotate.h"
 #include "i830_common.h"
+#include "xmlconfig.h"
 
 
-/* This roughly corresponds to a gl_renderbuffer (Mesa 6.4) */
+/* XXX: change name or eliminate to avoid conflict with "struct
+ * intel_region"!!!
+ */
 typedef struct {
    drm_handle_t handle;
    drmSize size;        /* region size in bytes */
@@ -58,7 +60,8 @@ typedef struct
    int mem;         /* unused */
    
    int cpp;         /* for front and back buffers */
-   int fbFormat;
+/*    int bitsPerPixel;   */
+   int fbFormat;  /* XXX FBO: this is obsolete - remove after i830 updates */
 
    int logTextureGranularity;
    
diff --git a/src/mesa/drivers/dri/i915/intel_span.c b/src/mesa/drivers/dri/i915/intel_span.c
index c3ffc4b2ac..84f5cc037d 100644
--- a/src/mesa/drivers/dri/i915/intel_span.c
+++ b/src/mesa/drivers/dri/i915/intel_span.c
@@ -30,194 +30,340 @@
 #include "mtypes.h"
 #include "colormac.h"
 
+#include "intel_fbo.h"
 #include "intel_screen.h"
-
 #include "intel_span.h"
+#include "intel_regions.h"
 #include "intel_ioctl.h"
+#include "intel_tex.h"
+
 #include "swrast/swrast.h"
 
+/*
+  break intelWriteRGBASpan_ARGB8888
+*/
 
+#undef DBG
 #define DBG 0
 
-#define LOCAL_VARS						\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);			\
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;		\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   GLuint pitch = drb->pitch;					\
-   GLuint height = dPriv->h;					\
-   char *buf = (char *) drb->Base.Data +			\
-			dPriv->x * drb->cpp +			\
-			dPriv->y * pitch;			\
-   GLushort p;							\
-   (void) buf; (void) p
-
-#define LOCAL_DEPTH_VARS					\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);			\
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;		\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   GLuint pitch = drb->pitch;					\
-   GLuint height = dPriv->h;					\
-   char *buf = (char *) drb->Base.Data +			\
-			dPriv->x * drb->cpp +			\
-			dPriv->y * pitch
+#define LOCAL_VARS							\
+   struct intel_context *intel = intel_context(ctx);			\
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);		\
+   const GLint yScale = irb->RenderToTexture ? 1 : -1;			\
+   const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;	\
+   GLubyte *buf = (GLubyte *) irb->pfMap				\
+      + (intel->drawY * irb->pfPitch + intel->drawX) * irb->region->cpp;\
+   GLuint p;								\
+   assert(irb->pfMap);\
+   (void) p;
+
+/* XXX FBO: this is identical to the macro in spantmp2.h except we get
+ * the cliprect info from the context, not the driDrawable.
+ * Move this into spantmp2.h someday.
+ */
+#define HW_CLIPLOOP()							\
+   do {									\
+      int _nc = intel->numClipRects;					\
+      while ( _nc-- ) {							\
+	 int minx = intel->pClipRects[_nc].x1 - intel->drawX;		\
+	 int miny = intel->pClipRects[_nc].y1 - intel->drawY;		\
+	 int maxx = intel->pClipRects[_nc].x2 - intel->drawX;		\
+	 int maxy = intel->pClipRects[_nc].y2 - intel->drawY;
+
 
-#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS 
 
-#define INIT_MONO_PIXEL(p,color)\
-	 p = INTEL_PACKCOLOR565(color[0],color[1],color[2])
 
-#define Y_FLIP(_y) (height - _y - 1)
+#define Y_FLIP(_y) ((_y) * yScale + yBias)
 
 #define HW_LOCK()
 
 #define HW_UNLOCK()
 
-/* 16 bit, 565 rgb color spanline and pixel functions
+/* 16 bit, RGB565 color spanline and pixel functions
  */
-#define WRITE_RGBA( _x, _y, r, g, b, a )				\
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = ( (((int)r & 0xf8) << 8) |	\
-		                             (((int)g & 0xfc) << 3) |	\
-		                             (((int)b & 0xf8) >> 3))
-#define WRITE_PIXEL( _x, _y, p )  \
-   *(GLushort *)(buf + _x*2 + _y*pitch) = p
-
-#define READ_RGBA( rgba, _x, _y )				\
-do {								\
-   GLushort p = *(GLushort *)(buf + _x*2 + _y*pitch);		\
-   rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
-   rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
-   rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
-   rgba[3] = 255;						\
-} while(0)
-
-#define TAG(x) intel##x##_565
-#include "spantmp.h"
-
-/* 15 bit, 555 rgb color spanline and pixel functions
- */
-#define WRITE_RGBA( _x, _y, r, g, b, a )			\
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = (((r & 0xf8) << 7) |	\
-		                            ((g & 0xf8) << 3) |	\
-                         		    ((b & 0xf8) >> 3))
-
-#define WRITE_PIXEL( _x, _y, p )  \
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = p
-
-#define READ_RGBA( rgba, _x, _y )				\
-do {								\
-   GLushort p = *(GLushort *)(buf + _x*2 + _y*pitch);		\
-   rgba[0] = (p >> 7) & 0xf8;					\
-   rgba[1] = (p >> 3) & 0xf8;					\
-   rgba[2] = (p << 3) & 0xf8;					\
-   rgba[3] = 255;						\
-} while(0)
-
-#define TAG(x) intel##x##_555
-#include "spantmp.h"
-
-/* 16 bit depthbuffer functions.
+#define SPANTMP_PIXEL_FMT GL_RGB
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
+
+#define TAG(x)    intel##x##_RGB565
+#define TAG2(x,y) intel##x##_RGB565##y
+#define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 2)
+#include "spantmp2.h"
+
+/* 32 bit, ARGB8888 color spanline and pixel functions
  */
-#define WRITE_DEPTH( _x, _y, d ) \
-   *(GLushort *)(buf + (_x)*2 + (_y)*pitch)  = d;
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
 
-#define READ_DEPTH( d, _x, _y )	\
-   d = *(GLushort *)(buf + (_x)*2 + (_y)*pitch);	 
+#define TAG(x)    intel##x##_ARGB8888
+#define TAG2(x,y) intel##x##_ARGB8888##y
+#define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 4)
+#include "spantmp2.h"
 
 
-#define TAG(x) intel##x##_z16
-#include "depthtmp.h"
+#define LOCAL_DEPTH_VARS						\
+   struct intel_context *intel = intel_context(ctx);			\
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);		\
+   const GLuint pitch = irb->pfPitch/***XXX region->pitch*/; /* in pixels */ \
+   const GLint yScale = irb->RenderToTexture ? 1 : -1;			\
+   const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;	\
+   char *buf = (char *) irb->pfMap/*XXX use region->map*/ +             \
+      (intel->drawY * pitch + intel->drawX) * irb->region->cpp;
 
 
-#undef LOCAL_VARS
-#define LOCAL_VARS						\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);			\
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;		\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   GLuint pitch = drb->pitch;					\
-   GLuint height = dPriv->h;					\
-   char *buf = (char *)drb->Base.Data +				\
-			dPriv->x * drb->cpp +			\
-			dPriv->y * pitch;			\
-   GLuint p;							\
-   (void) buf; (void) p
-
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p,color)\
-	 p = INTEL_PACKCOLOR8888(color[0],color[1],color[2],color[3])
-
-/* 32 bit, 8888 argb color spanline and pixel functions
- */
-#define WRITE_RGBA(_x, _y, r, g, b, a)			\
-    *(GLuint *)(buf + _x*4 + _y*pitch) = ((r << 16) |	\
-					  (g << 8)  |	\
-					  (b << 0)  |	\
-					  (a << 24) )
+#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS 
 
-#define WRITE_PIXEL(_x, _y, p)			\
-    *(GLuint *)(buf + _x*4 + _y*pitch) = p
+/**
+ ** 16-bit depthbuffer functions.
+ **/
+#define WRITE_DEPTH( _x, _y, d ) \
+   ((GLushort *)buf)[(_x) + (_y) * pitch] = d;
 
+#define READ_DEPTH( d, _x, _y )	\
+   d = ((GLushort *)buf)[(_x) + (_y) * pitch];
 
-#define READ_RGBA(rgba, _x, _y)					\
-    do {							\
-	GLuint p = *(GLuint *)(buf + _x*4 + _y*pitch);		\
-	rgba[0] = (p >> 16) & 0xff;				\
-	rgba[1] = (p >> 8)  & 0xff;				\
-	rgba[2] = (p >> 0)  & 0xff;				\
-	rgba[3] = (p >> 24) & 0xff;				\
-    } while (0)
 
-#define TAG(x) intel##x##_8888
-#include "spantmp.h"
+#define TAG(x) intel##x##_z16
+#include "depthtmp.h"
 
 
-/* 24/8 bit interleaved depth/stencil functions
- */
-#define WRITE_DEPTH( _x, _y, d ) {			\
-   GLuint tmp = *(GLuint *)(buf + (_x)*4 + (_y)*pitch);	\
-   tmp &= 0xff000000;					\
-   tmp |= (d) & 0xffffff;				\
-   *(GLuint *)(buf + (_x)*4 + (_y)*pitch) = tmp;		\
+/**
+ ** 24/8-bit interleaved depth/stencil functions
+ ** Note: we're actually reading back combined depth+stencil values.
+ ** The wrappers in main/depthstencil.c are used to extract the depth
+ ** and stencil values.
+ **/
+/* Change ZZZS -> SZZZ */
+#define WRITE_DEPTH( _x, _y, d ) {				\
+   GLuint tmp = ((d) >> 8) | ((d) << 24);			\
+   ((GLuint *)buf)[(_x) + (_y) * pitch] = tmp;			\
 }
 
-#define READ_DEPTH( d, _x, _y )		\
-   d = *(GLuint *)(buf + (_x)*4 + (_y)*pitch) & 0xffffff;
-
+/* Change SZZZ -> ZZZS */
+#define READ_DEPTH( d, _x, _y ) {				\
+   GLuint tmp = ((GLuint *)buf)[(_x) + (_y) * pitch];		\
+   d = (tmp << 8) | (tmp >> 24);				\
+}
 
 #define TAG(x) intel##x##_z24_s8
 #include "depthtmp.h"
 
-#define WRITE_STENCIL( _x, _y, d ) {			\
-   GLuint tmp = *(GLuint *)(buf + (_x)*4 + (_y)*pitch);	\
-   tmp &= 0xffffff;					\
-   tmp |= ((d)<<24);					\
-   *(GLuint *)(buf + (_x)*4 + (_y)*pitch) = tmp;		\
+
+/**
+ ** 8-bit stencil function (XXX FBO: This is obsolete)
+ **/
+#define WRITE_STENCIL( _x, _y, d ) {				\
+   GLuint tmp = ((GLuint *)buf)[(_x) + (_y) * pitch];		\
+   tmp &= 0xffffff;						\
+   tmp |= ((d) << 24);						\
+   ((GLuint *) buf)[(_x) + (_y) * pitch] = tmp;			\
 }
 
-#define READ_STENCIL( d, _x, _y )			\
-   d = *(GLuint *)(buf + (_x)*4 + (_y)*pitch) >> 24;
+#define READ_STENCIL( d, _x, _y )				\
+   d = ((GLuint *)buf)[(_x) + (_y) * pitch] >> 24;
 
 #define TAG(x) intel##x##_z24_s8
 #include "stenciltmp.h"
 
 
-/* Move locking out to get reasonable span performance.
+
+/**
+ * Map or unmap all the renderbuffers which we may need during
+ * software rendering.
+ * XXX in the future, we could probably convey extra information to
+ * reduce the number of mappings needed.  I.e. if doing a glReadPixels
+ * from the depth buffer, we really only need one mapping.
+ *
+ * XXX Rewrite this function someday.
+ * We can probably just loop over all the renderbuffer attachments,
+ * map/unmap all of them, and not worry about the _ColorDrawBuffers
+ * _ColorReadBuffer, _DepthBuffer or _StencilBuffer fields.
+ */
+static void
+intel_map_unmap_buffers(struct intel_context *intel, GLboolean map)
+{
+   GLcontext *ctx = &intel->ctx;
+   GLuint i, j;
+   struct intel_renderbuffer *irb;
+
+   /* color draw buffers */
+   for (i = 0; i < ctx->Const.MaxDrawBuffers; i++) {
+      for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers[i]; j++) {
+         struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[i][j];
+         irb = intel_renderbuffer(rb);
+         if (irb) {
+            /* this is a user-created intel_renderbuffer */
+            if (irb->region) {
+               if (map)
+                  intel_region_map(intel, irb->region);
+               else
+                  intel_region_unmap(intel, irb->region);
+            }
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+      }
+   }
+
+   /* check for render to textures */
+   for (i = 0; i < BUFFER_COUNT; i++) {
+      struct gl_renderbuffer_attachment *att = ctx->DrawBuffer->Attachment + i;
+      struct gl_texture_object *tex = att->Texture;
+      if (tex) {
+         /* render to texture */
+         ASSERT(att->Renderbuffer);
+         if (map) {
+            struct gl_texture_image *texImg;
+            texImg = tex->Image[att->CubeMapFace][att->TextureLevel];
+            intel_tex_map_images(intel, intel_texture_object(tex));
+         }
+         else {
+            intel_tex_unmap_images(intel, intel_texture_object(tex));
+         }
+      }
+   }
+
+   /* color read buffers */
+   irb = intel_renderbuffer(ctx->ReadBuffer->_ColorReadBuffer);
+   if (irb && irb->region) {
+      if (map)
+         intel_region_map(intel, irb->region);
+      else
+         intel_region_unmap(intel, irb->region);
+      irb->pfMap = irb->region->map;
+      irb->pfPitch = irb->region->pitch;
+   }
+
+   /* Account for front/back color page flipping.
+    * The span routines use the pfMap and pfPitch fields which will
+    * swap the front/back region map/pitch if we're page flipped.
+    * Do this after mapping, above, so the map field is valid.
+    */
+#if 0
+   if (map && ctx->DrawBuffer->Name == 0) {
+      struct intel_renderbuffer *irbFront
+         = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_FRONT_LEFT);
+      struct intel_renderbuffer *irbBack
+         = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_BACK_LEFT);
+      if (irbBack) {
+         /* double buffered */
+         if (intel->sarea->pf_current_page == 0) {
+            irbFront->pfMap = irbFront->region->map;
+            irbFront->pfPitch = irbFront->region->pitch;
+            irbBack->pfMap = irbBack->region->map;
+            irbBack->pfPitch = irbBack->region->pitch;
+         }
+         else {
+            irbFront->pfMap = irbBack->region->map;
+            irbFront->pfPitch = irbBack->region->pitch;
+            irbBack->pfMap = irbFront->region->map;
+            irbBack->pfPitch = irbFront->region->pitch;
+         }
+      }
+   }
+#endif
+
+   /* depth buffer (Note wrapper!) */
+   if (ctx->DrawBuffer->_DepthBuffer) {
+      irb = intel_renderbuffer(ctx->DrawBuffer->_DepthBuffer->Wrapped);
+      if (irb && irb->region && irb->Base.Name != 0) {
+         if (map) {
+            intel_region_map(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel, irb->region);
+            irb->pfMap = NULL;
+            irb->pfPitch = 0;
+         }
+      }
+   }
+
+   /* stencil buffer (Note wrapper!) */
+   if (ctx->DrawBuffer->_StencilBuffer) {
+      irb = intel_renderbuffer(ctx->DrawBuffer->_StencilBuffer->Wrapped);
+      if (irb && irb->region && irb->Base.Name != 0) {
+         if (map) {
+            intel_region_map(intel, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel, irb->region);
+            irb->pfMap = NULL;
+            irb->pfPitch = 0;
+         }
+      }
+   }
+}
+
+
+
+/**
+ * Prepare for softare rendering.  Map current read/draw framebuffers'
+ * renderbuffes and all currently bound texture objects.
+ *
+ * Old note: Moved locking out to get reasonable span performance.
  */
 void intelSpanRenderStart( GLcontext *ctx )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
+   GLuint i;
 
    intelFlush(&intel->ctx);
    LOCK_HARDWARE(intel);
-   intelWaitForIdle(intel);
+
+#if 0
+   /* Just map the framebuffer and all textures.  Bufmgr code will
+    * take care of waiting on the necessary fences:
+    */
+   intel_region_map(intel, intel->front_region);
+   intel_region_map(intel, intel->back_region);
+   intel_region_map(intel, intel->depth_region);
+#endif
+
+   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled) {
+	 struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
+	 intel_tex_map_images(intel, intel_texture_object(texObj));
+      }
+   }
+
+   intel_map_unmap_buffers(intel, GL_TRUE);
 }
 
+/**
+ * Called when done softare rendering.  Unmap the buffers we mapped in
+ * the above function.
+ */
 void intelSpanRenderFinish( GLcontext *ctx )
 {
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
+   struct intel_context *intel = intel_context( ctx );
+   GLuint i;
+
    _swrast_flush( ctx );
+
+   /* Now unmap the framebuffer:
+    */
+#if 0
+   intel_region_unmap(intel, intel->front_region);
+   intel_region_unmap(intel, intel->back_region);
+   intel_region_unmap(intel, intel->depth_region);
+#endif
+
+   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled) {
+	 struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
+	 intel_tex_unmap_images(intel, intel_texture_object(texObj));
+      }
+   }
+
+   intel_map_unmap_buffers(intel, GL_FALSE);
+
    UNLOCK_HARDWARE( intel );
 }
 
+
 void intelInitSpanFuncs( GLcontext *ctx )
 {
    struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
@@ -227,32 +373,31 @@ void intelInitSpanFuncs( GLcontext *ctx )
 
 
 /**
- * Plug in the Get/Put routines for the given driRenderbuffer.
+ * Plug in appropriate span read/write functions for the given renderbuffer.
+ * These are used for the software fallbacks.
  */
 void
-intelSetSpanFunctions(driRenderbuffer *drb, const GLvisual *vis)
+intel_set_span_functions(struct gl_renderbuffer *rb)
 {
-   if (drb->Base.InternalFormat == GL_RGBA) {
-      if (vis->redBits == 5 && vis->greenBits == 5 && vis->blueBits == 5) {
-         intelInitPointers_555(&drb->Base);
-      }
-      else if (vis->redBits == 5 && vis->greenBits == 6 && vis->blueBits == 5) {
-         intelInitPointers_565(&drb->Base);
-      }
-      else {
-         assert(vis->redBits == 8);
-         assert(vis->greenBits == 8);
-         assert(vis->blueBits == 8);
-         intelInitPointers_8888(&drb->Base);
-      }
+   if (rb->_ActualFormat == GL_RGB5) {
+      /* 565 RGB */
+      intelInitPointers_RGB565(rb);
+   }
+   else if (rb->_ActualFormat == GL_RGBA8) {
+      /* 8888 RGBA */
+      intelInitPointers_ARGB8888(rb);
+   }
+   else if (rb->_ActualFormat == GL_DEPTH_COMPONENT16) {
+      intelInitDepthPointers_z16(rb);
    }
-   else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) {
-      intelInitDepthPointers_z16(&drb->Base);
+   else if (rb->_ActualFormat == GL_DEPTH_COMPONENT24 || /* XXX FBO remove */
+            rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT) {
+      intelInitDepthPointers_z24_s8(rb);
    }
-   else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) {
-      intelInitDepthPointers_z24_s8(&drb->Base);
+   else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) { /* XXX FBO remove */
+      intelInitStencilPointers_z24_s8(rb);
    }
-   else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) {
-      intelInitStencilPointers_z24_s8(&drb->Base);
+   else {
+      _mesa_problem(NULL, "Unexpected _ActualFormat in intelSetSpanFunctions");
    }
 }
diff --git a/src/mesa/drivers/dri/i915/intel_span.h b/src/mesa/drivers/dri/i915/intel_span.h
index 2d4f8589d0..98dd57b06a 100644
--- a/src/mesa/drivers/dri/i915/intel_span.h
+++ b/src/mesa/drivers/dri/i915/intel_span.h
@@ -28,14 +28,12 @@
 #ifndef _INTEL_SPAN_H
 #define _INTEL_SPAN_H
 
-#include "drirenderbuffer.h"
-
 extern void intelInitSpanFuncs( GLcontext *ctx );
 
 extern void intelSpanRenderFinish( GLcontext *ctx );
 extern void intelSpanRenderStart( GLcontext *ctx );
 
 extern void
-intelSetSpanFunctions(driRenderbuffer *rb, const GLvisual *vis);
+intel_set_span_functions(struct gl_renderbuffer *rb);
 
 #endif
diff --git a/src/mesa/drivers/dri/i915/intel_state.c b/src/mesa/drivers/dri/i915/intel_state.c
index e5988a5ed6..772bec4971 100644
--- a/src/mesa/drivers/dri/i915/intel_state.c
+++ b/src/mesa/drivers/dri/i915/intel_state.c
@@ -30,10 +30,13 @@
 #include "context.h"
 #include "macros.h"
 #include "enums.h"
+#include "colormac.h"
 #include "dd.h"
 
 #include "intel_screen.h"
 #include "intel_context.h"
+#include "intel_fbo.h"
+#include "intel_regions.h"
 #include "swrast/swrast.h"
 
 int intel_translate_compare_func( GLenum func )
@@ -164,87 +167,67 @@ int intel_translate_logic_op( GLenum opcode )
    }
 }
 
-static void intelDrawBuffer(GLcontext *ctx, GLenum mode )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   int front = 0;
- 
-   if (!ctx->DrawBuffer)
-      return;
-
-   switch ( ctx->DrawBuffer->_ColorDrawBufferMask[0] ) {
-   case BUFFER_BIT_FRONT_LEFT:
-      front = 1;
-      FALLBACK( intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE );
-      break;
-   case BUFFER_BIT_BACK_LEFT:
-      front = 0;
-      FALLBACK( intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE );
-      break;
-   default:
-      FALLBACK( intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE );
-      return;
-   }
-
-   if ( intel->sarea->pf_current_page == 1 ) 
-      front ^= 1;
-   
-   intelSetFrontClipRects( intel );
-
-   if (front) {
-      intel->drawRegion = &intel->intelScreen->front;
-      intel->readRegion = &intel->intelScreen->front;
-   } else {
-      intel->drawRegion = &intel->intelScreen->back;
-      intel->readRegion = &intel->intelScreen->back;
-   }
-
-   intel->vtbl.set_color_region( intel, intel->drawRegion );
-}
 
-static void intelReadBuffer( GLcontext *ctx, GLenum mode )
+static void intelClearColor(GLcontext *ctx, const GLfloat color[4])
 {
-   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
-}
+   struct intel_context *intel = intel_context(ctx);
+   GLubyte clear[4];
 
+   CLAMPED_FLOAT_TO_UBYTE(clear[0], color[0]);
+   CLAMPED_FLOAT_TO_UBYTE(clear[1], color[1]);
+   CLAMPED_FLOAT_TO_UBYTE(clear[2], color[2]);
+   CLAMPED_FLOAT_TO_UBYTE(clear[3], color[3]);
 
-static void intelClearColor(GLcontext *ctx, const GLfloat color[4])
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   intelScreenPrivate *screen = intel->intelScreen;
-
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_red, color[0]);
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_green, color[1]);
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_blue, color[2]);
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_alpha, color[3]);
-
-   intel->ClearColor = INTEL_PACKCOLOR(screen->fbFormat,
-				       intel->clear_red, 
-				       intel->clear_green, 
-				       intel->clear_blue, 
-				       intel->clear_alpha);
+   /* compute both 32 and 16-bit clear values */
+   intel->ClearColor8888 = INTEL_PACKCOLOR8888(clear[0], clear[1],
+                                               clear[2], clear[3]);
+   intel->ClearColor565 = INTEL_PACKCOLOR565(clear[0], clear[1], clear[2]);
 }
 
 
+/**
+ * Update the viewport transformation matrix.  Depends on:
+ *  - viewport pos/size
+ *  - depthrange
+ *  - window pos/size or FBO size
+ */
 static void intelCalcViewport( GLcontext *ctx )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    const GLfloat *v = ctx->Viewport._WindowMap.m;
+   const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
    GLfloat *m = intel->ViewportMatrix.m;
-   GLint h = 0;
+   GLfloat yScale, yBias;
 
-   if (intel->driDrawable) 
-      h = intel->driDrawable->h + SUBPIXEL_Y;
+   if (ctx->DrawBuffer->Name) {
+      /* User created FBO */
+      struct intel_renderbuffer *irb
+         = intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[0][0]);
+      if (irb && !irb->RenderToTexture) {
+         /* y=0=top */
+         yScale = -1.0;
+         yBias = irb->Base.Height;
+      }
+      else {
+         /* y=0=bottom */
+         yScale = 1.0;
+         yBias = 0.0;
+      }
+   }
+   else {
+      /* window buffer, y=0=top */
+      yScale = -1.0;
+      yBias = (intel->driDrawable) ? intel->driDrawable->h : 0.0F;
+   }
 
-   /* See also intel_translate_vertex.  SUBPIXEL adjustments can be done
-    * via state vars, too.
-    */
-   m[MAT_SX] =   v[MAT_SX];
-   m[MAT_TX] =   v[MAT_TX] + SUBPIXEL_X;
-   m[MAT_SY] = - v[MAT_SY];
-   m[MAT_TY] = - v[MAT_TY] + h;
-   m[MAT_SZ] =   v[MAT_SZ] * intel->depth_scale;
-   m[MAT_TZ] =   v[MAT_TZ] * intel->depth_scale;
+   m[MAT_SX] = v[MAT_SX];
+   m[MAT_TX] = v[MAT_TX] + SUBPIXEL_X;
+
+   m[MAT_SY] = v[MAT_SY] * yScale;
+   m[MAT_TY] = v[MAT_TY] * yScale + yBias + SUBPIXEL_Y;
+
+   m[MAT_SZ] = v[MAT_SZ] * depthScale;
+   m[MAT_TZ] = v[MAT_TZ] * depthScale;
 }
 
 static void intelViewport( GLcontext *ctx,
@@ -264,18 +247,112 @@ static void intelDepthRange( GLcontext *ctx,
  */
 static void intelRenderMode( GLcontext *ctx, GLenum mode )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    FALLBACK( intel, INTEL_FALLBACK_RENDERMODE, (mode != GL_RENDER) );
 }
 
 
 void intelInitStateFuncs( struct dd_function_table *functions )
 {
-   functions->DrawBuffer = intelDrawBuffer;
-   functions->ReadBuffer = intelReadBuffer;
    functions->RenderMode = intelRenderMode;
    functions->Viewport = intelViewport;
    functions->DepthRange = intelDepthRange;
    functions->ClearColor = intelClearColor;
 }
 
+
+
+
+void intelInitState( GLcontext *ctx )
+{
+   /* Mesa should do this for us:
+    */
+   ctx->Driver.AlphaFunc( ctx, 
+			  ctx->Color.AlphaFunc,
+			  ctx->Color.AlphaRef);
+
+   ctx->Driver.BlendColor( ctx,
+			   ctx->Color.BlendColor );
+
+   ctx->Driver.BlendEquationSeparate( ctx, 
+				      ctx->Color.BlendEquationRGB,
+				      ctx->Color.BlendEquationA);
+
+   ctx->Driver.BlendFuncSeparate( ctx,
+				  ctx->Color.BlendSrcRGB,
+				  ctx->Color.BlendDstRGB,
+				  ctx->Color.BlendSrcA,
+				  ctx->Color.BlendDstA);
+
+   ctx->Driver.ColorMask( ctx, 
+			  ctx->Color.ColorMask[RCOMP],
+			  ctx->Color.ColorMask[GCOMP],
+			  ctx->Color.ColorMask[BCOMP],
+			  ctx->Color.ColorMask[ACOMP]);
+
+   ctx->Driver.CullFace( ctx, ctx->Polygon.CullFaceMode );
+   ctx->Driver.DepthFunc( ctx, ctx->Depth.Func );
+   ctx->Driver.DepthMask( ctx, ctx->Depth.Mask );
+
+   ctx->Driver.Enable( ctx, GL_ALPHA_TEST, ctx->Color.AlphaEnabled );
+   ctx->Driver.Enable( ctx, GL_BLEND, ctx->Color.BlendEnabled );
+   ctx->Driver.Enable( ctx, GL_COLOR_LOGIC_OP, ctx->Color.ColorLogicOpEnabled );
+   ctx->Driver.Enable( ctx, GL_COLOR_SUM, ctx->Fog.ColorSumEnabled );
+   ctx->Driver.Enable( ctx, GL_CULL_FACE, ctx->Polygon.CullFlag );
+   ctx->Driver.Enable( ctx, GL_DEPTH_TEST, ctx->Depth.Test );
+   ctx->Driver.Enable( ctx, GL_DITHER, ctx->Color.DitherFlag );
+   ctx->Driver.Enable( ctx, GL_FOG, ctx->Fog.Enabled );
+   ctx->Driver.Enable( ctx, GL_LIGHTING, ctx->Light.Enabled );
+   ctx->Driver.Enable( ctx, GL_LINE_SMOOTH, ctx->Line.SmoothFlag );
+   ctx->Driver.Enable( ctx, GL_POLYGON_STIPPLE, ctx->Polygon.StippleFlag );
+   ctx->Driver.Enable( ctx, GL_SCISSOR_TEST, ctx->Scissor.Enabled );
+   ctx->Driver.Enable( ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled );
+   ctx->Driver.Enable( ctx, GL_TEXTURE_1D, GL_FALSE );
+   ctx->Driver.Enable( ctx, GL_TEXTURE_2D, GL_FALSE );
+   ctx->Driver.Enable( ctx, GL_TEXTURE_RECTANGLE_NV, GL_FALSE );
+   ctx->Driver.Enable( ctx, GL_TEXTURE_3D, GL_FALSE );
+   ctx->Driver.Enable( ctx, GL_TEXTURE_CUBE_MAP, GL_FALSE );
+
+   ctx->Driver.Fogfv( ctx, GL_FOG_COLOR, ctx->Fog.Color );
+   ctx->Driver.Fogfv( ctx, GL_FOG_MODE, 0 );
+   ctx->Driver.Fogfv( ctx, GL_FOG_DENSITY, &ctx->Fog.Density );
+   ctx->Driver.Fogfv( ctx, GL_FOG_START, &ctx->Fog.Start );
+   ctx->Driver.Fogfv( ctx, GL_FOG_END, &ctx->Fog.End );
+
+   ctx->Driver.FrontFace( ctx, ctx->Polygon.FrontFace );
+
+   {
+      GLfloat f = (GLfloat)ctx->Light.Model.ColorControl;
+      ctx->Driver.LightModelfv( ctx, GL_LIGHT_MODEL_COLOR_CONTROL, &f );
+   }
+
+   ctx->Driver.LineWidth( ctx, ctx->Line.Width );
+   ctx->Driver.LogicOpcode( ctx, ctx->Color.LogicOp );
+   ctx->Driver.PointSize( ctx, ctx->Point.Size );
+   ctx->Driver.PolygonStipple( ctx, (const GLubyte *)ctx->PolygonStipple );
+   ctx->Driver.Scissor( ctx, ctx->Scissor.X, ctx->Scissor.Y,
+			ctx->Scissor.Width, ctx->Scissor.Height );
+   ctx->Driver.ShadeModel( ctx, ctx->Light.ShadeModel );
+   ctx->Driver.StencilFuncSeparate( ctx, GL_FRONT,
+                                    ctx->Stencil.Function[0],
+                                    ctx->Stencil.Ref[0],
+                                    ctx->Stencil.ValueMask[0] );
+   ctx->Driver.StencilFuncSeparate( ctx, GL_BACK,
+                                    ctx->Stencil.Function[1],
+                                    ctx->Stencil.Ref[1],
+                                    ctx->Stencil.ValueMask[1] );
+   ctx->Driver.StencilMaskSeparate( ctx, GL_FRONT, ctx->Stencil.WriteMask[0] );
+   ctx->Driver.StencilMaskSeparate( ctx, GL_BACK, ctx->Stencil.WriteMask[1] );
+   ctx->Driver.StencilOpSeparate( ctx, GL_FRONT,
+                                  ctx->Stencil.FailFunc[0],
+                                  ctx->Stencil.ZFailFunc[0],
+                                  ctx->Stencil.ZPassFunc[0]);
+   ctx->Driver.StencilOpSeparate( ctx, GL_BACK,
+                                  ctx->Stencil.FailFunc[1],
+                                  ctx->Stencil.ZFailFunc[1],
+                                  ctx->Stencil.ZPassFunc[1]);
+
+
+   /* XXX this isn't really needed */
+   ctx->Driver.DrawBuffer( ctx, ctx->Color.DrawBuffer[0] );
+}
diff --git a/src/mesa/drivers/dri/i915/intel_tex.c b/src/mesa/drivers/dri/i915/intel_tex.c
index 6012d3e799..a945609023 100644
--- a/src/mesa/drivers/dri/i915/intel_tex.c
+++ b/src/mesa/drivers/dri/i915/intel_tex.c
@@ -1,871 +1,163 @@
-/**************************************************************************
- * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-#include "glheader.h"
-#include "mtypes.h"
-#include "imports.h"
-#include "macros.h"
-#include "simple_list.h"
-#include "enums.h"
-#include "image.h"
-#include "texstore.h"
-#include "texformat.h"
-#include "teximage.h"
-#include "texmem.h"
 #include "texobj.h"
-#include "swrast/swrast.h"
-
-#include "mm.h"
-
-#include "intel_screen.h"
-#include "intel_batchbuffer.h"
 #include "intel_context.h"
+#include "intel_mipmap_tree.h"
 #include "intel_tex.h"
-#include "intel_ioctl.h"
-
-
 
-static GLboolean
-intelValidateClientStorage( intelContextPtr intel, GLenum target,
-			    GLint internalFormat,
-			    GLint srcWidth, GLint srcHeight, 
-			    GLenum format, GLenum type,  const void *pixels,
-			    const struct gl_pixelstore_attrib *packing,
-			    struct gl_texture_object *texObj,
-			    struct gl_texture_image *texImage)
 
+static GLboolean intelIsTextureResident(GLcontext *ctx,
+                                      struct gl_texture_object *texObj)
 {
-   GLcontext *ctx = &intel->ctx;
-   int texelBytes;
-
-   if (0)
-      fprintf(stderr, "intformat %s format %s type %s\n",
-	      _mesa_lookup_enum_by_nr( internalFormat ),
-	      _mesa_lookup_enum_by_nr( format ),
-	      _mesa_lookup_enum_by_nr( type ));
-
-   if (!ctx->Unpack.ClientStorage)
-      return 0;
-
-   if (ctx->_ImageTransferState ||
-       texImage->IsCompressed ||
-       texObj->GenerateMipmap)
-      return 0;
-
-
-   /* This list is incomplete
-    */
-   switch ( internalFormat ) {
-   case GL_RGBA:
-      if ( format == GL_BGRA && type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
-	 texImage->TexFormat = &_mesa_texformat_argb8888;
-	 texelBytes = 4;
-      }
-      else
-	 return 0;
-      break;
-
-   case GL_RGB:
-      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
-	 texImage->TexFormat = &_mesa_texformat_rgb565;
-	 texelBytes = 2;
-      }
-      else
-	 return 0;
-      break;
-
-   case GL_YCBCR_MESA:
-      if ( format == GL_YCBCR_MESA && 
-	   type == GL_UNSIGNED_SHORT_8_8_REV_APPLE ) {
-	 texImage->TexFormat = &_mesa_texformat_ycbcr_rev;
-	 texelBytes = 2;
-      }
-      else if ( format == GL_YCBCR_MESA && 
-		(type == GL_UNSIGNED_SHORT_8_8_APPLE || 
-		 type == GL_UNSIGNED_BYTE)) {
-	 texImage->TexFormat = &_mesa_texformat_ycbcr;
-	 texelBytes = 2;
-      }
-      else
-	 return 0;
-      break;
-      
-	 
-   default:
-      return 0;
-   }
-
-   /* Could deal with these packing issues, but currently don't:
-    */
-   if (packing->SkipPixels || 
-       packing->SkipRows || 
-       packing->SwapBytes ||
-       packing->LsbFirst) {
-      return 0;
-   }
-
-   {      
-      GLint srcRowStride = _mesa_image_row_stride(packing, srcWidth,
-						  format, type);
-
-      
-      if (0)
-	 fprintf(stderr, "%s: srcRowStride %d/%x\n", 
-		 __FUNCTION__, srcRowStride, srcRowStride);
-
-      /* Could check this later in upload, pitch restrictions could be
-       * relaxed, but would need to store the image pitch somewhere,
-       * as packing details might change before image is uploaded:
-       */
-      if (!intelIsAgpMemory( intel, pixels, srcHeight * srcRowStride ) ||
-	  (srcRowStride & 63))
-	 return 0;
-
-
-      /* Have validated that _mesa_transfer_teximage would be a straight
-       * memcpy at this point.  NOTE: future calls to TexSubImage will
-       * overwrite the client data.  This is explicitly mentioned in the
-       * extension spec.
-       */
-      texImage->Data = (void *)pixels;
-      texImage->IsClientData = GL_TRUE;
-      texImage->RowStride = srcRowStride / texelBytes;
-      return 1;
-   }
+#if 0
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_object *intelObj = intel_texture_object(texObj);
+   
+   return 
+      intelObj->mt && 
+      intelObj->mt->region && 
+      intel_is_region_resident(intel, intelObj->mt->region);
+#endif
+   return 1;
 }
 
- 
-
-static void intelTexImage1D( GLcontext *ctx, GLenum target, GLint level,
-			    GLint internalFormat,
-			    GLint width, GLint border,
-			    GLenum format, GLenum type, const GLvoid *pixels,
-			    const struct gl_pixelstore_attrib *packing,
-			    struct gl_texture_object *texObj,
-			    struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
 
-   assert(t);
-   intelFlush( ctx );
-   driSwapOutTextureObject( t );
 
-   texImage->IsClientData = GL_FALSE;
-
-   _mesa_store_teximage1d( ctx, target, level, internalFormat,
-			   width, border, format, type,
-			   pixels, packing, texObj, texImage );
-
-   t->dirty_images[0] |= (1 << level);
-}
-
-static void intelTexSubImage1D( GLcontext *ctx, 
-			       GLenum target,
-			       GLint level,	
-			       GLint xoffset,
-				GLsizei width,
-			       GLenum format, GLenum type,
-			       const GLvoid *pixels,
-			       const struct gl_pixelstore_attrib *packing,
-			       struct gl_texture_object *texObj,
-			       struct gl_texture_image *texImage )
+static struct gl_texture_image *intelNewTextureImage( GLcontext *ctx )
 {
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   assert(t);
-   intelFlush( ctx );
-   driSwapOutTextureObject( t );
-
-   _mesa_store_texsubimage1d(ctx, target, level, xoffset, width, 
-			     format, type, pixels, packing, texObj,
-			     texImage);
+   (void) ctx;
+   return (struct gl_texture_image *)CALLOC_STRUCT(intel_texture_image);
 }
 
 
-/* Handles 2D, CUBE, RECT:
- */
-static void intelTexImage2D( GLcontext *ctx, GLenum target, GLint level,
-			    GLint internalFormat,
-			    GLint width, GLint height, GLint border,
-			    GLenum format, GLenum type, const GLvoid *pixels,
-			    const struct gl_pixelstore_attrib *packing,
-			    struct gl_texture_object *texObj,
-			    struct gl_texture_image *texImage )
+static struct gl_texture_object *intelNewTextureObject( GLcontext *ctx, 
+							GLuint name, 
+							GLenum target )
 {
-   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;
-   }
+   struct intel_texture_object *obj = CALLOC_STRUCT(intel_texture_object);
 
-   assert(t);
-   intelFlush( ctx );
-   driSwapOutTextureObject( t );
-   texImage->IsClientData = GL_FALSE;
+   _mesa_initialize_texture_object(&obj->base, name, target);
 
-   if (intelValidateClientStorage( INTEL_CONTEXT(ctx), target, 
-				   internalFormat, 
-				   width, height, 
-				   format, type, pixels, 
-				   packing, texObj, texImage)) {
-      if (INTEL_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "%s: Using client storage\n", __FUNCTION__); 
-   }
-   else {
-      _mesa_store_teximage2d( ctx, target, level, internalFormat,
-			      width, height, border, format, type,
-			      pixels, packing, texObj, texImage );
-
-      t->dirty_images[face] |= (1 << level);
-   }
+   return &obj->base;
 }
 
-static void intelTexSubImage2D( GLcontext *ctx, 
-			       GLenum target,
-			       GLint level,	
-			       GLint xoffset, GLint yoffset,
-			       GLsizei width, GLsizei height,
-			       GLenum format, GLenum type,
-			       const GLvoid *pixels,
-			       const struct gl_pixelstore_attrib *packing,
-			       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 (texImage->IsClientData &&
-       (char *)pixels == (char *)texImage->Data + 
-       ((xoffset + yoffset * texImage->RowStride) * 
-	texImage->TexFormat->TexelBytes)) {
-
-      /* Notification only - no upload required */
-   }
-   else {
-      assert( t ); /* this _should_ be true */
-      intelFlush( ctx );
-      driSwapOutTextureObject( t );
-
-      _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width, 
-				height, format, type, pixels, packing, texObj,
-				texImage);
 
-      t->dirty_images[face] |= (1 << level);
-   }
-}
-
-static void intelCompressedTexImage2D( 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 )
+static void intelFreeTextureImageData( GLcontext *ctx, 
+				     struct gl_texture_image *texImage )
 {
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-   GLuint face;
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
 
-   /* 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 (intelImage->mt) {
+      intel_miptree_release(intel, &intelImage->mt);
    }
-
-   assert(t);
-   intelFlush( ctx );
    
-   driSwapOutTextureObject( t );
-   texImage->IsClientData = GL_FALSE;
-
-   if (INTEL_DEBUG & DEBUG_TEXTURE)
-     fprintf(stderr, "%s: Using normal storage\n", __FUNCTION__); 
-   
-   _mesa_store_compressed_teximage2d(ctx, target, level, internalFormat, width,
-				     height, border, imageSize, data, texObj, texImage);
-   
-   t->dirty_images[face] |= (1 << level);
-}
-
-
-static void intelCompressedTexSubImage2D( 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;
+   if (texImage->Data) {
+      free(texImage->Data);
+      texImage->Data = NULL;
    }
-
-   assert( t ); /* this _should_ be true */
-   intelFlush( ctx );
-   driSwapOutTextureObject( t );
-   
-   _mesa_store_compressed_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
-					height, format, imageSize, data, texObj, texImage);
-   
-   t->dirty_images[face] |= (1 << level);
-}
-
-
-static void intelTexImage3D( GLcontext *ctx, GLenum target, GLint level,
-                            GLint internalFormat,
-                            GLint width, GLint height, GLint depth,
-                            GLint border,
-                            GLenum format, GLenum type, const GLvoid *pixels,
-                            const struct gl_pixelstore_attrib *packing,
-                            struct gl_texture_object *texObj,
-                            struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   assert(t);
-   driSwapOutTextureObject( t );
-   texImage->IsClientData = GL_FALSE;
-
-   _mesa_store_teximage3d(ctx, target, level, internalFormat,
-			  width, height, depth, border,
-			  format, type, pixels,
-			  &ctx->Unpack, texObj, texImage);
-   
-   t->dirty_images[0] |= (1 << level);
-}
-
-
-static void
-intelTexSubImage3D( GLcontext *ctx, GLenum target, GLint level,
-                   GLint xoffset, GLint yoffset, GLint zoffset,
-                   GLsizei width, GLsizei height, GLsizei depth,
-                   GLenum format, GLenum type,
-                   const GLvoid *pixels,
-                   const struct gl_pixelstore_attrib *packing,
-                   struct gl_texture_object *texObj,
-                   struct gl_texture_image *texImage )
-{
-   driTextureObject * t = (driTextureObject *) texObj->DriverData;
-
-   assert( t ); /* this _should_ be true */
-   driSwapOutTextureObject( t );
-
-   _mesa_store_texsubimage3d(ctx, target, level, xoffset, yoffset, zoffset,
-                             width, height, depth,
-                             format, type, pixels, packing, texObj, texImage);
-
-   t->dirty_images[0] |= (1 << level);
 }
 
 
-
-
-static void intelDeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj )
+#ifndef __x86_64__
+static unsigned
+fastrdtsc(void)
 {
-   driTextureObject * t = (driTextureObject *) tObj->DriverData;
+    unsigned eax;
+    __asm__ volatile ("\t"
+	"pushl  %%ebx\n\t"
+	"cpuid\n\t" ".byte 0x0f, 0x31\n\t" "popl %%ebx\n":"=a" (eax)
+	:"0"(0)
+	:"ecx", "edx", "cc");
 
-   if ( t != NULL ) {
-      intelFlush( ctx );
-      driDestroyTextureObject( t );
-   }
-   
-   /* Free mipmap images and the texture object itself */
-   _mesa_delete_texture_object(ctx, tObj);
+    return eax;
 }
-
-
-static const struct gl_texture_format *
-intelChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
-			 GLenum format, GLenum type )
+#else
+static unsigned
+fastrdtsc(void)
 {
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   const GLboolean do32bpt = ( intel->intelScreen->cpp == 4 &&
-			       intel->intelScreen->tex.size > 4*1024*1024);
-
-   switch ( internalFormat ) {
-   case 4:
-   case GL_RGBA:
-   case GL_COMPRESSED_RGBA:
-      if ( format == GL_BGRA ) {
-	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
-	    return &_mesa_texformat_argb8888;
-	 }
-         else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
-            return &_mesa_texformat_argb4444;
-	 }
-         else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
-	    return &_mesa_texformat_argb1555;
-	 }
-      }
-      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444;
-
-   case 3:
-   case GL_RGB:
-   case GL_COMPRESSED_RGB:
-      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
-	 return &_mesa_texformat_rgb565;
-      }
-      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_rgb565;
-
-   case GL_RGBA8:
-   case GL_RGB10_A2:
-   case GL_RGBA12:
-   case GL_RGBA16:
-      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444;
-
-   case GL_RGBA4:
-   case GL_RGBA2:
-      return &_mesa_texformat_argb4444;
-
-   case GL_RGB5_A1:
-      return &_mesa_texformat_argb1555;
-
-   case GL_RGB8:
-   case GL_RGB10:
-   case GL_RGB12:
-   case GL_RGB16:
-      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_rgb565;
-
-   case GL_RGB5:
-   case GL_RGB4:
-   case GL_R3_G3_B2:
-      return &_mesa_texformat_rgb565;
-
-   case GL_ALPHA:
-   case GL_ALPHA4:
-   case GL_ALPHA8:
-   case GL_ALPHA12:
-   case GL_ALPHA16:
-   case GL_COMPRESSED_ALPHA:
-      return &_mesa_texformat_a8;
-
-   case 1:
-   case GL_LUMINANCE:
-   case GL_LUMINANCE4:
-   case GL_LUMINANCE8:
-   case GL_LUMINANCE12:
-   case GL_LUMINANCE16:
-   case GL_COMPRESSED_LUMINANCE:
-      return &_mesa_texformat_l8;
-
-   case 2:
-   case GL_LUMINANCE_ALPHA:
-   case GL_LUMINANCE4_ALPHA4:
-   case GL_LUMINANCE6_ALPHA2:
-   case GL_LUMINANCE8_ALPHA8:
-   case GL_LUMINANCE12_ALPHA4:
-   case GL_LUMINANCE12_ALPHA12:
-   case GL_LUMINANCE16_ALPHA16:
-   case GL_COMPRESSED_LUMINANCE_ALPHA:
-      return &_mesa_texformat_al88;
+    unsigned eax;
+    __asm__ volatile ("\t"
+	"cpuid\n\t" ".byte 0x0f, 0x31\n\t" :"=a" (eax)
+	:"0"(0)
+		      :"ecx", "edx", "ebx", "cc");
 
-   case GL_INTENSITY:
-   case GL_INTENSITY4:
-   case GL_INTENSITY8:
-   case GL_INTENSITY12:
-   case GL_INTENSITY16:
-   case GL_COMPRESSED_INTENSITY:
-      return &_mesa_texformat_i8;
-
-   case GL_YCBCR_MESA:
-      if (type == GL_UNSIGNED_SHORT_8_8_MESA ||
-	  type == GL_UNSIGNED_BYTE)
-         return &_mesa_texformat_ycbcr;
-      else
-         return &_mesa_texformat_ycbcr_rev;
-
-   case GL_COMPRESSED_RGB_FXT1_3DFX:
-     return &_mesa_texformat_rgb_fxt1;
-   case GL_COMPRESSED_RGBA_FXT1_3DFX:
-     return &_mesa_texformat_rgba_fxt1;
-
-   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;
-
-   case GL_DEPTH_COMPONENT:
-   case GL_DEPTH_COMPONENT16:
-   case GL_DEPTH_COMPONENT24:
-   case GL_DEPTH_COMPONENT32:
-      return &_mesa_texformat_z16;
-
-   default:
-      fprintf(stderr, "unexpected texture format %s in %s\n", 
-	      _mesa_lookup_enum_by_nr(internalFormat),
-	      __FUNCTION__);
-      return NULL;
-   }
-
-   return NULL; /* never get here */
+    return eax;
 }
+#endif
 
-
-
-void intelDestroyTexObj(intelContextPtr intel, intelTextureObjectPtr t)
+static unsigned
+time_diff(unsigned t, unsigned t2)
 {
-   unsigned   i;
-
-   if ( intel == NULL ) 
-      return;
-
-   if ( t->age > intel->dirtyAge )
-      intel->dirtyAge = t->age;
-
-   for ( i = 0 ; i < MAX_TEXTURE_UNITS ; i++ ) {
-      if ( t == intel->CurrentTexObj[ i ] ) 
-	 intel->CurrentTexObj[ i ] = NULL;
-   }
+    return ((t < t2) ? t2 - t : 0xFFFFFFFFU - (t - t2 - 1));
 }
 
 
-
-/* Upload an image from mesa's internal copy.  Image may be 1D, 2D or
- * 3D.  Cubemaps are expanded elsewhere.
+/* The system memcpy (at least on ubuntu 5.10) has problems copying
+ * to agp (writecombined) memory from a source which isn't 64-byte
+ * aligned - there is a 4x performance falloff.
+ *
+ * The x86 __memcpy is immune to this but is slightly slower
+ * (10%-ish) than the system memcpy.
+ *
+ * The sse_memcpy seems to have a slight cliff at 64/32 bytes, but
+ * isn't much faster than x86_memcpy for agp copies.
+ * 
+ * TODO: switch dynamically.
  */
-static void intelUploadTexImage( intelContextPtr intel,
-				 intelTextureObjectPtr t,
-				 const struct gl_texture_image *image,
-				 const GLuint offset )
+static void *do_memcpy( void *dest, const void *src, size_t n )
 {
-
-   if (!image || !image->Data) 
-      return;
-
-   if (image->Depth == 1 && image->IsClientData) {
-      if (INTEL_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "Blit uploading\n");
-
-      /* Do it with a blit.
-       */
-      intelEmitCopyBlitLocked( intel,
-			       image->TexFormat->TexelBytes,
-			       image->RowStride, /* ? */
-			       intelGetMemoryOffsetMESA( NULL, 0, image->Data ),
-			       t->Pitch / image->TexFormat->TexelBytes,
-			       intelGetMemoryOffsetMESA( NULL, 0, t->BufAddr + offset ),
-			       0, 0,
-			       0, 0,
-			       image->Width,
-			       image->Height);
-   }
-   else if (image->IsCompressed) {
-      GLuint row_len = image->Width * 2;
-      GLubyte *dst = (GLubyte *)(t->BufAddr + offset);
-      GLubyte *src = (GLubyte *)image->Data;
-      GLuint j;
-
-      if (INTEL_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, 
-		 "Upload image %dx%dx%d offset %xm row_len %x "
-		 "pitch %x depth_pitch %x\n",
-		 image->Width, image->Height, image->Depth, offset,
-		 row_len, t->Pitch, t->depth_pitch);
-
-      switch (image->InternalFormat) {
-	case GL_COMPRESSED_RGB_FXT1_3DFX:
-	case GL_COMPRESSED_RGBA_FXT1_3DFX:
-	case GL_RGB_S3TC:
-	case GL_RGB4_S3TC:
-	case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
-	case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
-	  for (j = 0 ; j < image->Height/4 ; j++, dst += (t->Pitch)) {
-	    __memcpy(dst, src, row_len );
-	    src += row_len;
-	  }
-	  break;
-	case GL_RGBA_S3TC:
-	case GL_RGBA4_S3TC:
-	case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
-	case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
-	  for (j = 0 ; j < image->Height/4 ; j++, dst += (t->Pitch)) {
-	    __memcpy(dst, src, (image->Width*4) );
-	    src += image->Width*4;
-	  }
-	  break;
-	default:
-	  fprintf(stderr,"Internal Compressed format not supported %d\n", image->InternalFormat);
-	  break;
-      }
-   }
-   /* Time for another vtbl entry:
-    */
-   else if (intel->intelScreen->deviceID == PCI_CHIP_I945_G ||
-            intel->intelScreen->deviceID == PCI_CHIP_I945_GM) {
-      GLuint row_len = image->Width * image->TexFormat->TexelBytes;
-      GLubyte *dst = (GLubyte *)(t->BufAddr + offset);
-      GLubyte *src = (GLubyte *)image->Data;
-      GLuint d, j;
-
-      if (INTEL_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, 
-		 "Upload image %dx%dx%d offset %xm row_len %x "
-		 "pitch %x depth_pitch %x\n",
-		 image->Width, image->Height, image->Depth, offset,
-		 row_len, t->Pitch, t->depth_pitch);
-
-      if (row_len == t->Pitch) {
-	 memcpy( dst, src, row_len * image->Height * image->Depth );
-      }
-      else { 
-	 GLuint x = 0, y = 0;
-
-	 for (d = 0 ; d < image->Depth ; d++) {
-	    GLubyte *dst0 = dst + x + y * t->Pitch;
-
-	    for (j = 0 ; j < image->Height ; j++) {
-	       __memcpy(dst0, src, row_len );
-	       src += row_len;
-	       dst0 += t->Pitch;
-	    }
-
-	    x += MIN2(4, row_len); /* Guess: 4 byte minimum alignment */
-	    if (x > t->Pitch) {
-	       x = 0;
-	       y += image->Height;
-	    }
-	 }
-      }
-
-   }
-   else {
-      GLuint row_len = image->Width * image->TexFormat->TexelBytes;
-      GLubyte *dst = (GLubyte *)(t->BufAddr + offset);
-      GLubyte *src = (GLubyte *)image->Data;
-      GLuint d, j;
-
-      if (INTEL_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, 
-		 "Upload image %dx%dx%d offset %xm row_len %x "
-		 "pitch %x depth_pitch %x\n",
-		 image->Width, image->Height, image->Depth, offset,
-		 row_len, t->Pitch, t->depth_pitch);
-
-      if (row_len == t->Pitch) {
-	 for (d = 0; d < image->Depth; d++) {
-	    memcpy( dst, src, t->Pitch * image->Height );
-	    dst += t->depth_pitch;
-	    src += row_len * image->Height;
-	 }
-      }
-      else { 
-	 for (d = 0 ; d < image->Depth ; d++) {
-	    for (j = 0 ; j < image->Height ; j++) {
-	       __memcpy(dst, src, row_len );
-	       src += row_len;
-	       dst += t->Pitch;
-	    }
-
-	    dst += t->depth_pitch - (t->Pitch * image->Height);
-	 }
-      }
+   if ( (((unsigned)src) & 63) ||
+	(((unsigned)dest) & 63)) {
+      return  __memcpy(dest, src, n);	
    }
+   else
+      return memcpy(dest, src, n);
 }
 
 
-
-int intelUploadTexImages( intelContextPtr intel, 
-			  intelTextureObjectPtr t,
-			  GLuint face)
+static void *timed_memcpy( void *dest, const void *src, size_t n )
 {
-   const int numLevels = t->base.lastLevel - t->base.firstLevel + 1;
-   const struct gl_texture_image *firstImage = t->image[face][t->base.firstLevel].image;
-   int pitch = firstImage->RowStride * firstImage->TexFormat->TexelBytes;
+   void *ret;
+   unsigned t1, t2;
+   double rate;
 
-   /* Can we texture out of the existing client data? */
-   if ( numLevels == 1 &&
-	firstImage->IsClientData &&
-	(pitch & 3) == 0) {
+   if ( (((unsigned)src) & 63) ||
+	(((unsigned)dest) & 63)) 
+      _mesa_printf("Warning - non-aligned texture copy!\n");
 
-      if (INTEL_DEBUG & DEBUG_TEXTURE)
-	 fprintf(stderr, "AGP texturing from client memory\n");
-
-      t->TextureOffset = intelAgpOffsetFromVirtual( intel, firstImage->Data );
-      t->BufAddr = 0;
-      t->dirty = ~0;
-      return GL_TRUE;
-   }
-   else {
-      if (INTEL_DEBUG & DEBUG_TEXTURE) 
-	 fprintf(stderr, "Uploading client data to agp\n");
+   t1 = fastrdtsc();
+   ret =  do_memcpy(dest, src, n);	
+   t2 = fastrdtsc();
 
-      INTEL_FIREVERTICES( intel );
-      LOCK_HARDWARE( intel );
-
-      if ( t->base.memBlock == NULL ) {
-	 int heap;
-
-	 heap = driAllocateTexture( intel->texture_heaps, intel->nr_heaps,
-				    (driTextureObject *) t );
-	 if ( heap == -1 ) {
-	    UNLOCK_HARDWARE( intel );
-	    return GL_FALSE;
-	 }
-
-	 /* Set the base offset of the texture image */
-	 t->BufAddr = (GLubyte *) (intel->intelScreen->tex.map + 
-				   t->base.memBlock->ofs);
-	 t->TextureOffset = intel->intelScreen->tex.offset + t->base.memBlock->ofs;
-	 t->dirty = ~0;
-      }
-
-
-      /* Let the world know we've used this memory recently.
-       */
-      driUpdateTextureLRU( (driTextureObject *) t );
-
-
-      /* Upload any images that are new */
-      if (t->base.dirty_images[face]) {
-	 int i;
-
- 	 intelWaitForIdle( intel );
-	    
-	 for (i = 0 ; i < numLevels ; i++) { 
-	    int level = i + t->base.firstLevel;
-
-	    if (t->base.dirty_images[face] & (1<<level)) {
-
-	       const struct gl_texture_image *image = t->image[face][i].image;
-	       GLuint offset = t->image[face][i].offset;
-
-     	       if (INTEL_DEBUG & DEBUG_TEXTURE)
-	          fprintf(stderr, "upload level %d, offset %x\n", 
-			  level, offset);
-
-	       intelUploadTexImage( intel, t, image, offset );
-	    }
-	 }
-	 t->base.dirty_images[face] = 0;
-	 intel->perf_boxes |= I830_BOX_TEXTURE_LOAD;
-      }
-      
-      UNLOCK_HARDWARE( intel );
-      return GL_TRUE;
-   }
+   rate = time_diff(t1, t2);
+   rate /= (double) n;
+   _mesa_printf("timed_memcpy: %u %u --> %f clocks/byte\n", t1, t2, rate); 
+   return ret;
 }
 
-/**
- * Allocate a new texture object.
- * Called via ctx->Driver.NewTextureObject.
- * Note: this function will be called during context creation to
- * allocate the default texture objects.
- * Note: we could use containment here to 'derive' the driver-specific
- * texture object from the core mesa gl_texture_object.  Not done at this time.
- */
-static struct gl_texture_object *
-intelNewTextureObject( GLcontext *ctx, GLuint name, GLenum target )
-{
-   struct gl_texture_object *obj = _mesa_new_texture_object(ctx, name, target);
-   INTEL_CONTEXT(ctx)->vtbl.alloc_tex_obj( obj );
-   return obj;
-}
 
-
-void intelInitTextureFuncs( struct dd_function_table *functions )
+void intelInitTextureFuncs(struct dd_function_table * functions)
 {
-   functions->NewTextureObject          = intelNewTextureObject;
-   functions->ChooseTextureFormat       = intelChooseTextureFormat;
-   functions->TexImage1D                = intelTexImage1D;
-   functions->TexImage2D                = intelTexImage2D;
-   functions->TexImage3D                = intelTexImage3D;
-   functions->TexSubImage1D             = intelTexSubImage1D;
-   functions->TexSubImage2D             = intelTexSubImage2D;
-   functions->TexSubImage3D             = intelTexSubImage3D;
-   functions->CopyTexImage1D            = _swrast_copy_teximage1d;
-   functions->CopyTexImage2D            = _swrast_copy_teximage2d;
-   functions->CopyTexSubImage1D         = _swrast_copy_texsubimage1d;
-   functions->CopyTexSubImage2D         = _swrast_copy_texsubimage2d;
-   functions->CopyTexSubImage3D         = _swrast_copy_texsubimage3d;
-   functions->DeleteTexture             = intelDeleteTexture;
-   functions->UpdateTexturePalette      = NULL;
-   functions->IsTextureResident         = driIsTextureResident;
-   functions->TestProxyTexImage         = _mesa_test_proxy_teximage;
-   functions->DeleteTexture             = intelDeleteTexture;
-   functions->CompressedTexImage2D      = intelCompressedTexImage2D;
-   functions->CompressedTexSubImage2D   = intelCompressedTexSubImage2D;
+   functions->ChooseTextureFormat = intelChooseTextureFormat;
+   functions->TexImage1D = intelTexImage1D;
+   functions->TexImage2D = intelTexImage2D;
+   functions->TexImage3D = intelTexImage3D;
+   functions->TexSubImage1D = intelTexSubImage1D;
+   functions->TexSubImage2D = intelTexSubImage2D;
+   functions->TexSubImage3D = intelTexSubImage3D;
+   functions->CopyTexImage1D = intelCopyTexImage1D;
+   functions->CopyTexImage2D = intelCopyTexImage2D;
+   functions->CopyTexSubImage1D = intelCopyTexSubImage1D;
+   functions->CopyTexSubImage2D = intelCopyTexSubImage2D;
+   functions->GetTexImage = intelGetTexImage;
+   functions->NewTextureObject = intelNewTextureObject;
+   functions->NewTextureImage = intelNewTextureImage;
+   functions->DeleteTexture = _mesa_delete_texture_object;
+   functions->FreeTexImageData = intelFreeTextureImageData;
+   functions->UpdateTexturePalette = 0;
+   functions->IsTextureResident = intelIsTextureResident;
+
+   if (INTEL_DEBUG & DEBUG_BUFMGR)
+      functions->TextureMemCpy = timed_memcpy;
+   else
+      functions->TextureMemCpy = do_memcpy;
 }
diff --git a/src/mesa/drivers/dri/i915/intel_tex.h b/src/mesa/drivers/dri/i915/intel_tex.h
index 9b7e550232..2880070dab 100644
--- a/src/mesa/drivers/dri/i915/intel_tex.h
+++ b/src/mesa/drivers/dri/i915/intel_tex.h
@@ -35,11 +35,101 @@
 
 void intelInitTextureFuncs( struct dd_function_table *functions );
 
-void intelDestroyTexObj( intelContextPtr intel, intelTextureObjectPtr t );
-int intelUploadTexImages( intelContextPtr intel, intelTextureObjectPtr t,
-			  GLuint face );
+const struct gl_texture_format *
+intelChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
+                          GLenum format, GLenum type );
+
+
+void intelTexImage3D(GLcontext *ctx, 
+		     GLenum target, GLint level,
+		     GLint internalFormat,
+		     GLint width, GLint height, GLint depth,
+		     GLint border,
+		     GLenum format, GLenum type, const void *pixels,
+		     const struct gl_pixelstore_attrib *packing,
+		     struct gl_texture_object *texObj,
+		     struct gl_texture_image *texImage);
+
+void intelTexSubImage3D(GLcontext *ctx,
+			GLenum target,
+			GLint level,
+			GLint xoffset, GLint yoffset, GLint zoffset,
+			GLsizei width, GLsizei height, GLsizei depth,
+			GLenum format, GLenum type,
+			const GLvoid *pixels,
+			const struct gl_pixelstore_attrib *packing,
+			struct gl_texture_object *texObj,
+			struct gl_texture_image *texImage);
+
+void intelTexImage2D(GLcontext *ctx, 
+		     GLenum target, GLint level,
+		     GLint internalFormat,
+		     GLint width, GLint height, GLint border,
+		     GLenum format, GLenum type, const void *pixels,
+		     const struct gl_pixelstore_attrib *packing,
+		     struct gl_texture_object *texObj,
+		     struct gl_texture_image *texImage);
+
+void intelTexSubImage2D(GLcontext *ctx,
+			GLenum target,
+			GLint level,
+			GLint xoffset, GLint yoffset,
+			GLsizei width, GLsizei height,
+			GLenum format, GLenum type,
+			const GLvoid *pixels,
+			const struct gl_pixelstore_attrib *packing,
+			struct gl_texture_object *texObj,
+			struct gl_texture_image *texImage);
+
+void intelTexImage1D(GLcontext *ctx, 
+		     GLenum target, GLint level,
+		     GLint internalFormat,
+		     GLint width, GLint border,
+		     GLenum format, GLenum type, const void *pixels,
+		     const struct gl_pixelstore_attrib *packing,
+		     struct gl_texture_object *texObj,
+		     struct gl_texture_image *texImage);
+
+void intelTexSubImage1D(GLcontext *ctx,
+			GLenum target,
+			GLint level,
+			GLint xoffset,
+			GLsizei width,
+			GLenum format, GLenum type,
+			const GLvoid *pixels,
+			const struct gl_pixelstore_attrib *packing,
+			struct gl_texture_object *texObj,
+			struct gl_texture_image *texImage);
+
+void intelCopyTexImage1D( GLcontext *ctx, GLenum target, GLint level,
+			  GLenum internalFormat,
+			  GLint x, GLint y, GLsizei width,
+			  GLint border );
+
+void intelCopyTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+			  GLenum internalFormat,
+			  GLint x, GLint y, GLsizei width, GLsizei height,
+			  GLint border );
+
+void intelCopyTexSubImage1D( GLcontext *ctx, GLenum target, GLint level,
+			     GLint xoffset,
+			     GLint x, GLint y, GLsizei width );
+
+void intelCopyTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+			     GLint xoffset, GLint yoffset,
+			     GLint x, GLint y, GLsizei width, GLsizei height );
+
+void intelGetTexImage( GLcontext *ctx, GLenum target, GLint level,
+                       GLenum format, GLenum type, GLvoid *pixels,
+                       struct gl_texture_object *texObj,
+                       struct gl_texture_image *texImage );
+
+GLuint intel_finalize_mipmap_tree( struct intel_context *intel, GLuint unit );
+
+void intel_tex_map_images( struct intel_context *intel,
+			   struct intel_texture_object *intelObj );
+
+void intel_tex_unmap_images( struct intel_context *intel,
+			     struct intel_texture_object *intelObj );
 
-GLboolean 
-intel_driReinitTextureHeap( driTexHeap *heap,
-			    unsigned size );
 #endif
diff --git a/src/mesa/drivers/dri/i915/intel_tris.c b/src/mesa/drivers/dri/i915/intel_tris.c
index cf1673b429..695f2f78e8 100644
--- a/src/mesa/drivers/dri/i915/intel_tris.c
+++ b/src/mesa/drivers/dri/i915/intel_tris.c
@@ -38,14 +38,95 @@
 #include "tnl/t_vertex.h"
 
 #include "intel_screen.h"
+#include "intel_context.h"
 #include "intel_tris.h"
 #include "intel_batchbuffer.h"
 #include "intel_reg.h"
 #include "intel_span.h"
+#include "intel_tex.h"
 
 static void intelRenderPrimitive( GLcontext *ctx, GLenum prim );
 static void intelRasterPrimitive( GLcontext *ctx, GLenum rprim, GLuint hwprim );
 
+/*
+ */
+static void intel_flush_inline_primitive( struct intel_context *intel )
+{
+   GLuint used = intel->batch->ptr - intel->prim.start_ptr;
+
+   assert(intel->prim.primitive != ~0);
+
+   if (used < 8)
+      goto do_discard;
+
+   *(int *)intel->prim.start_ptr = (_3DPRIMITIVE | 
+				    intel->prim.primitive |
+				    (used/4-2));
+
+   goto finished;
+   
+ do_discard:
+   intel->batch->ptr -= used;
+
+ finished:
+   intel->prim.primitive = ~0;
+   intel->prim.start_ptr = 0;
+   intel->prim.flush = 0;
+}
+
+
+/* Emit a primitive referencing vertices in a vertex buffer.
+ */
+void intelStartInlinePrimitive( struct intel_context *intel, 
+				GLuint prim, 
+				GLuint batch_flags )
+{
+   BATCH_LOCALS;
+   
+   /* Emit a slot which will be filled with the inline primitive
+    * command later.
+    */
+   BEGIN_BATCH(2, batch_flags);
+   OUT_BATCH( 0 );
+
+   intel->prim.start_ptr = intel->batch->ptr;
+   intel->prim.primitive = prim;
+   intel->prim.flush = intel_flush_inline_primitive;
+
+   OUT_BATCH( 0 );
+   ADVANCE_BATCH();
+}
+
+
+void intelWrapInlinePrimitive( struct intel_context *intel )
+{
+   GLuint prim = intel->prim.primitive;
+   GLuint batchflags = intel->batch->flags;
+
+   intel_flush_inline_primitive(intel);
+   intel_batchbuffer_flush(intel->batch);
+
+   intel->vtbl.emit_state( intel );
+   intelStartInlinePrimitive( intel, prim, batchflags ); /* ??? */
+}
+
+GLuint *intelExtendInlinePrimitive( struct intel_context *intel, 
+				    GLuint dwords )
+{
+   GLuint sz = dwords * sizeof(GLuint);
+   GLuint *ptr;
+
+   if (intel_batchbuffer_space(intel->batch) < sz) 
+      intelWrapInlinePrimitive( intel );
+
+   ptr = (GLuint *)intel->batch->ptr;
+   intel->batch->ptr += sz;
+
+   return ptr;
+}
+
+
+
 /***********************************************************************
  *                    Emit primitives as inline vertices               *
  ***********************************************************************/
@@ -63,22 +144,18 @@ do {								\
 #else
 #define COPY_DWORDS( j, vb, vertsize, v )	\
 do {						\
-   if (0) fprintf(stderr, "\n");	\
    for ( j = 0 ; j < vertsize ; j++ ) {		\
-      if (0) fprintf(stderr, "   -- v(%d): %x/%f\n",j,	\
-	      ((GLuint *)v)[j],			\
-	      ((GLfloat *)v)[j]);		\
       vb[j] = ((GLuint *)v)[j];			\
    }						\
    vb += vertsize;				\
 } while (0)
 #endif
 
-static void __inline__ intel_draw_quad( intelContextPtr intel,
-					intelVertexPtr v0,
-					intelVertexPtr v1,
-					intelVertexPtr v2,
-					intelVertexPtr v3 )
+static void intel_draw_quad( struct intel_context *intel,
+			     intelVertexPtr v0,
+			     intelVertexPtr v1,
+			     intelVertexPtr v2,
+			     intelVertexPtr v3 )
 {
    GLuint vertsize = intel->vertex_size;
    GLuint *vb = intelExtendInlinePrimitive( intel, 6 * vertsize );
@@ -92,10 +169,10 @@ static void __inline__ intel_draw_quad( intelContextPtr intel,
    COPY_DWORDS( j, vb, vertsize, v3 );
 }
 
-static void __inline__ intel_draw_triangle( intelContextPtr intel,
-					    intelVertexPtr v0,
-					    intelVertexPtr v1,
-					    intelVertexPtr v2 )
+static void intel_draw_triangle( struct intel_context *intel,
+				 intelVertexPtr v0,
+				 intelVertexPtr v1,
+				 intelVertexPtr v2 )
 {
    GLuint vertsize = intel->vertex_size;
    GLuint *vb = intelExtendInlinePrimitive( intel, 3 * vertsize );
@@ -107,9 +184,9 @@ static void __inline__ intel_draw_triangle( intelContextPtr intel,
 }
 
 
-static __inline__ void intel_draw_line( intelContextPtr intel,
-					intelVertexPtr v0,
-					intelVertexPtr v1 )
+static void intel_draw_line( struct intel_context *intel,
+			     intelVertexPtr v0,
+			     intelVertexPtr v1 )
 {
    GLuint vertsize = intel->vertex_size;
    GLuint *vb = intelExtendInlinePrimitive( intel, 2 * vertsize );
@@ -120,8 +197,8 @@ static __inline__ void intel_draw_line( intelContextPtr intel,
 }
 
 
-static __inline__ void intel_draw_point( intelContextPtr intel,
-					 intelVertexPtr v0 )
+static void intel_draw_point( struct intel_context *intel,
+			      intelVertexPtr v0 )
 {
    GLuint vertsize = intel->vertex_size;
    GLuint *vb = intelExtendInlinePrimitive( intel, vertsize );
@@ -140,7 +217,7 @@ static __inline__ void intel_draw_point( intelContextPtr intel,
  *                Fixup for ARB_point_parameters                       *
  ***********************************************************************/
 
-static void intel_atten_point( intelContextPtr intel, intelVertexPtr v0 )
+static void intel_atten_point( struct intel_context *intel, intelVertexPtr v0 )
 {
    GLcontext *ctx = &intel->ctx;
    GLfloat psz[4], col[4], restore_psz, restore_alpha;
@@ -189,7 +266,7 @@ static void intel_atten_point( intelContextPtr intel, intelVertexPtr v0 )
 
 
 
-static void intel_wpos_triangle( intelContextPtr intel,
+static void intel_wpos_triangle( struct intel_context *intel,
 				 intelVertexPtr v0,
 				 intelVertexPtr v1,
 				 intelVertexPtr v2 )
@@ -205,7 +282,7 @@ static void intel_wpos_triangle( intelContextPtr intel,
 }
 
 
-static void intel_wpos_line( intelContextPtr intel,
+static void intel_wpos_line( struct intel_context *intel,
 			     intelVertexPtr v0,
 			     intelVertexPtr v1 )
 {
@@ -219,7 +296,7 @@ static void intel_wpos_line( intelContextPtr intel,
 }
 
 
-static void intel_wpos_point( intelContextPtr intel,
+static void intel_wpos_point( struct intel_context *intel,
 			      intelVertexPtr v0 )
 {
    GLuint offset = intel->wpos_offset;
@@ -349,7 +426,7 @@ do {							\
 #define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx]
 
 #define LOCAL_VARS(n)							\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);				\
+   struct intel_context *intel = intel_context(ctx);				\
    GLuint color[n], spec[n];						\
    GLuint coloroffset = intel->coloroffset;		\
    GLboolean specoffset = intel->specoffset;			\
@@ -481,7 +558,7 @@ static void init_rast_tab( void )
  * primitives.
  */
 static void
-intel_fallback_tri( intelContextPtr intel,
+intel_fallback_tri( struct intel_context *intel,
 		   intelVertex *v0,
 		   intelVertex *v1,
 		   intelVertex *v2 )
@@ -491,6 +568,9 @@ intel_fallback_tri( intelContextPtr intel,
 
    if (0)
       fprintf(stderr, "\n%s\n", __FUNCTION__);
+   
+   if (intel->prim.flush)
+      intel->prim.flush(intel);
 
    _swsetup_Translate( ctx, v0, &v[0] );
    _swsetup_Translate( ctx, v1, &v[1] );
@@ -502,7 +582,7 @@ intel_fallback_tri( intelContextPtr intel,
 
 
 static void
-intel_fallback_line( intelContextPtr intel,
+intel_fallback_line( struct intel_context *intel,
 		    intelVertex *v0,
 		    intelVertex *v1 )
 {
@@ -512,6 +592,9 @@ intel_fallback_line( intelContextPtr intel,
    if (0)
       fprintf(stderr, "\n%s\n", __FUNCTION__);
 
+   if (intel->prim.flush)
+      intel->prim.flush(intel);
+
    _swsetup_Translate( ctx, v0, &v[0] );
    _swsetup_Translate( ctx, v1, &v[1] );
    intelSpanRenderStart( ctx );
@@ -520,24 +603,6 @@ intel_fallback_line( intelContextPtr intel,
 }
 
 
-static void
-intel_fallback_point( intelContextPtr intel,
-		     intelVertex *v0 )
-{
-   GLcontext *ctx = &intel->ctx;
-   SWvertex v[1];
-
-   if (0)
-      fprintf(stderr, "\n%s\n", __FUNCTION__);
-
-   _swsetup_Translate( ctx, v0, &v[0] );
-   intelSpanRenderStart( ctx );
-   _swrast_Point( ctx, &v[0] );
-   intelSpanRenderFinish( ctx );
-}
-
-
-
 /**********************************************************************/
 /*               Render unclipped begin/end objects                   */
 /**********************************************************************/
@@ -552,7 +617,7 @@ intel_fallback_point( intelContextPtr intel,
 #define INIT(x) intelRenderPrimitive( ctx, x )
 #undef LOCAL_VARS
 #define LOCAL_VARS						\
-    intelContextPtr intel = INTEL_CONTEXT(ctx);			\
+    struct intel_context *intel = intel_context(ctx);			\
     GLubyte *vertptr = (GLubyte *)intel->verts;			\
     const GLuint vertsize = intel->vertex_size;       	\
     const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
@@ -578,7 +643,7 @@ intel_fallback_point( intelContextPtr intel,
 static void intelRenderClippedPoly( GLcontext *ctx, const GLuint *elts,
 				   GLuint n )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
    GLuint prim = intel->render_primitive;
@@ -609,7 +674,7 @@ static void intelRenderClippedLine( GLcontext *ctx, GLuint ii, GLuint jj )
 static void intelFastRenderClippedPoly( GLcontext *ctx, const GLuint *elts,
 				       GLuint n )
 {
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
+   struct intel_context *intel = intel_context( ctx );
    const GLuint vertsize = intel->vertex_size;
    GLuint *vb = intelExtendInlinePrimitive( intel, (n-2) * 3 * vertsize );
    GLubyte *vertptr = (GLubyte *)intel->verts;
@@ -630,17 +695,13 @@ static void intelFastRenderClippedPoly( GLcontext *ctx, const GLuint *elts,
 
 
 
-#define POINT_FALLBACK (0)
-#define LINE_FALLBACK (DD_LINE_STIPPLE)
-#define TRI_FALLBACK (0)
-#define ANY_FALLBACK_FLAGS (POINT_FALLBACK|LINE_FALLBACK|TRI_FALLBACK|\
-                            DD_TRI_STIPPLE|DD_POINT_ATTEN)
-#define ANY_RASTER_FLAGS (DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET|DD_TRI_UNFILLED)
+#define ANY_FALLBACK_FLAGS (DD_LINE_STIPPLE | DD_TRI_STIPPLE | DD_POINT_ATTEN)
+#define ANY_RASTER_FLAGS (DD_TRI_LIGHT_TWOSIDE | DD_TRI_OFFSET | DD_TRI_UNFILLED)
 
 void intelChooseRenderState(GLcontext *ctx)
 {
    TNLcontext *tnl = TNL_CONTEXT(ctx);
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    GLuint flags = ctx->_TriangleCaps;
    const struct gl_fragment_program *fprog = ctx->FragmentProgram._Current;
    GLboolean have_wpos = (fprog && (fprog->Base.InputsRead & FRAG_BIT_WPOS));
@@ -676,15 +737,9 @@ void intelChooseRenderState(GLcontext *ctx)
        */
       if (flags & ANY_FALLBACK_FLAGS)
       {
-	 if (flags & POINT_FALLBACK)
-	    intel->draw_point = intel_fallback_point;
-
-	 if (flags & LINE_FALLBACK)
+	 if (flags & DD_LINE_STIPPLE)
 	    intel->draw_line = intel_fallback_line;
 
-	 if (flags & TRI_FALLBACK)
-	    intel->draw_tri = intel_fallback_tri;
-
 	 if ((flags & DD_TRI_STIPPLE) && !intel->hw_stipple) 
 	    intel->draw_tri = intel_fallback_tri;
 
@@ -740,7 +795,7 @@ static const GLenum reduced_prim[GL_POLYGON+1] = {
 
 static void intelRunPipeline( GLcontext *ctx )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
 
    if (intel->NewGLState) {
       if (intel->NewGLState & _NEW_TEXTURE) {
@@ -760,13 +815,21 @@ static void intelRunPipeline( GLcontext *ctx )
 
 static void intelRenderStart( GLcontext *ctx )
 {
-   INTEL_CONTEXT(ctx)->vtbl.render_start( INTEL_CONTEXT(ctx) );
+   struct intel_context *intel = intel_context(ctx);
+
+   intel->vtbl.render_start( intel_context(ctx) );
+   intel->vtbl.emit_state( intel );
 }
 
 static void intelRenderFinish( GLcontext *ctx )
 {
-   if (INTEL_CONTEXT(ctx)->RenderIndex & INTEL_FALLBACK_BIT)
+   struct intel_context *intel = intel_context(ctx);
+
+   if (intel->RenderIndex & INTEL_FALLBACK_BIT)
       _swrast_flush( ctx );
+
+   if (intel->prim.flush)
+      intel->prim.flush(intel);
 }
 
 
@@ -777,7 +840,7 @@ static void intelRenderFinish( GLcontext *ctx )
   */
 static void intelRasterPrimitive( GLcontext *ctx, GLenum rprim, GLuint hwprim )
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
 
    if (0)
       fprintf(stderr, "%s %s %x\n", __FUNCTION__, 
@@ -787,111 +850,179 @@ static void intelRasterPrimitive( GLcontext *ctx, GLenum rprim, GLuint hwprim )
     
    /* Start a new primitive.  Arrange to have it flushed later on.
     */
-   if (hwprim != intel->prim.primitive) 
-      intelStartInlinePrimitive( intel, hwprim );
+   if (hwprim != intel->prim.primitive) {
+      if (intel->prim.flush)
+	 intel->prim.flush(intel);
+
+      intelStartInlinePrimitive( intel, hwprim, INTEL_BATCH_CLIPRECTS );
+   }
 }
 
 
  /* 
   */
- static void intelRenderPrimitive( GLcontext *ctx, GLenum prim )
- {
-    intelContextPtr intel = INTEL_CONTEXT(ctx);
+static void intelRenderPrimitive( GLcontext *ctx, GLenum prim )
+{
+   struct intel_context *intel = intel_context(ctx);
 
-    if (0)
-       fprintf(stderr, "%s %s\n", __FUNCTION__, _mesa_lookup_enum_by_nr(prim));
+   if (0)
+      fprintf(stderr, "%s %s\n", __FUNCTION__, _mesa_lookup_enum_by_nr(prim));
 
-    /* Let some clipping routines know which primitive they're dealing
-     * with.
-     */
-    intel->render_primitive = prim;
+   /* Let some clipping routines know which primitive they're dealing
+    * with.
+    */
+   intel->render_primitive = prim;
 
-    /* Shortcircuit this when called from t_dd_rendertmp.h for unfilled
-     * triangles.  The rasterized primitive will always be reset by
-     * lower level functions in that case, potentially pingponging the
-     * state:
-     */
-    if (reduced_prim[prim] == GL_TRIANGLES && 
-	(ctx->_TriangleCaps & DD_TRI_UNFILLED))
-       return;
+   /* Shortcircuit this when called from t_dd_rendertmp.h for unfilled
+    * triangles.  The rasterized primitive will always be reset by
+    * lower level functions in that case, potentially pingponging the
+    * state:
+    */
+   if (reduced_prim[prim] == GL_TRIANGLES && 
+       (ctx->_TriangleCaps & DD_TRI_UNFILLED))
+      return;
 
-    /* Set some primitive-dependent state and Start? a new primitive.
-     */
-    intelRasterPrimitive( ctx, reduced_prim[prim], hw_prim[prim] );
- }
+   /* Set some primitive-dependent state and Start? a new primitive.
+    */
+   intelRasterPrimitive( ctx, reduced_prim[prim], hw_prim[prim] );
+}
 
 
  /**********************************************************************/
  /*           Transition to/from hardware rasterization.               */
  /**********************************************************************/
 
- static char *fallbackStrings[] = {
-    "Texture",
-    "Draw buffer",
-    "Read buffer",
-    "Color mask",
-    "Render mode",
-    "Stencil",
-    "Stipple",
-    "User disable"
- };
-
-
- static char *getFallbackString(GLuint bit)
- {
-    int i = 0;
-    while (bit > 1) {
-       i++;
-       bit >>= 1;
-    }
-    return fallbackStrings[i];
- }
-
-
-
- void intelFallback( intelContextPtr intel, GLuint bit, GLboolean mode )
- {
-    GLcontext *ctx = &intel->ctx;
-    TNLcontext *tnl = TNL_CONTEXT(ctx);
-    GLuint oldfallback = intel->Fallback;
-
-    if (mode) {
-       intel->Fallback |= bit;
-       if (oldfallback == 0) {
-	  intelFlush(ctx);
-	  if (INTEL_DEBUG & DEBUG_FALLBACKS) 
-	     fprintf(stderr, "ENTER FALLBACK %x: %s\n",
-		     bit, getFallbackString( bit ));
-	  _swsetup_Wakeup( ctx );
-	  intel->RenderIndex = ~0;
-       }
-    }
-    else {
-       intel->Fallback &= ~bit;
-       if (oldfallback == bit) {
-	  _swrast_flush( ctx );
-	  if (INTEL_DEBUG & DEBUG_FALLBACKS) 
-	     fprintf(stderr, "LEAVE FALLBACK %s\n", getFallbackString( bit ));
-	  tnl->Driver.Render.Start = intelRenderStart;
-	  tnl->Driver.Render.PrimitiveNotify = intelRenderPrimitive;
-	  tnl->Driver.Render.Finish = intelRenderFinish;
-	  tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
-	  tnl->Driver.Render.CopyPV = _tnl_copy_pv;
-	  tnl->Driver.Render.Interp = _tnl_interp;
-
-	  _tnl_invalidate_vertex_state( ctx, ~0 );
-	  _tnl_invalidate_vertices( ctx, ~0 );
-	  _tnl_install_attrs( ctx, 
-			      intel->vertex_attrs, 
-			      intel->vertex_attr_count,
-			      intel->ViewportMatrix.m, 0 ); 
-
-	  intel->NewGLState |= _INTEL_NEW_RENDERSTATE;
-       }
-    }
- }
+static char *fallbackStrings[] = {
+   "Texture",
+   "Draw buffer",
+   "Read buffer",
+   "Color mask",
+   "Render mode",
+   "Stencil",
+   "Stipple",
+   "User disable"
+};
 
 
+static char *getFallbackString(GLuint bit)
+{
+   int i = 0;
+   while (bit > 1) {
+      i++;
+      bit >>= 1;
+   }
+   return fallbackStrings[i];
+}
+
+
+
+void intelFallback( struct intel_context *intel, GLuint bit, GLboolean mode )
+{
+   GLcontext *ctx = &intel->ctx;
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint oldfallback = intel->Fallback;
+
+   if (mode) {
+      intel->Fallback |= bit;
+      if (oldfallback == 0) {
+	 intelFlush(ctx);
+	 if (INTEL_DEBUG & DEBUG_FALLBACKS) 
+	    fprintf(stderr, "ENTER FALLBACK %x: %s\n",
+		    bit, getFallbackString( bit ));
+	 _swsetup_Wakeup( ctx );
+	 intel->RenderIndex = ~0;
+      }
+   }
+   else {
+      intel->Fallback &= ~bit;
+      if (oldfallback == bit) {
+	 _swrast_flush( ctx );
+	 if (INTEL_DEBUG & DEBUG_FALLBACKS) 
+	    fprintf(stderr, "LEAVE FALLBACK %s\n", getFallbackString( bit ));
+	 tnl->Driver.Render.Start = intelRenderStart;
+	 tnl->Driver.Render.PrimitiveNotify = intelRenderPrimitive;
+	 tnl->Driver.Render.Finish = intelRenderFinish;
+	 tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
+	 tnl->Driver.Render.CopyPV = _tnl_copy_pv;
+	 tnl->Driver.Render.Interp = _tnl_interp;
+
+	 _tnl_invalidate_vertex_state( ctx, ~0 );
+	 _tnl_invalidate_vertices( ctx, ~0 );
+	 _tnl_install_attrs( ctx, 
+			     intel->vertex_attrs, 
+			     intel->vertex_attr_count,
+			     intel->ViewportMatrix.m, 0 ); 
+
+	 intel->NewGLState |= _INTEL_NEW_RENDERSTATE;
+      }
+   }
+}
+
+union fi { 
+   GLfloat f; 
+   GLint i; 
+};
+
+
+/**********************************************************************/
+/*             Used only with the metaops callbacks.                  */
+/**********************************************************************/
+void intel_meta_draw_poly(struct intel_context *intel, 
+			  GLuint n,
+			  GLfloat xy[][2],
+			  GLfloat z,
+			  GLuint color,
+			  GLfloat tex[][2])
+{
+   union fi *vb;
+   GLint i;
+
+   intel->vtbl.emit_state( intel );
+
+   /* All 3d primitives should be emitted with INTEL_BATCH_CLIPRECTS,
+    * otherwise the drawing origin (DR4) might not be set correctly.
+    */
+   intelStartInlinePrimitive( intel, PRIM3D_TRIFAN, INTEL_BATCH_CLIPRECTS );  
+   vb = (union fi *)intelExtendInlinePrimitive( intel, n * 6 );
+
+   for (i = 0; i < n; i++) {
+      vb[0].f = xy[i][0];
+      vb[1].f = xy[i][1];
+      vb[2].f = z;
+      vb[3].i = color;
+      vb[4].f = tex[i][0];
+      vb[5].f = tex[i][1];
+      vb += 6;
+   }
+
+   if (intel->prim.flush)
+      intel->prim.flush(intel);
+}
+
+void intel_meta_draw_quad(struct intel_context *intel, 
+			  GLfloat x0, GLfloat x1,
+			  GLfloat y0, GLfloat y1, 
+			  GLfloat z,
+			  GLuint color,
+			  GLfloat s0, GLfloat s1,
+			  GLfloat t0, GLfloat t1)
+{
+   GLfloat xy[4][2];
+   GLfloat tex[4][2];
+
+   xy[0][0] = x0; xy[0][1] = y0;
+   xy[1][0] = x1; xy[1][1] = y0;
+   xy[2][0] = x1; xy[2][1] = y1;
+   xy[3][0] = x0; xy[3][1] = y1;
+
+   tex[0][0] = s0; tex[0][1] = t0;
+   tex[1][0] = s1; tex[1][1] = t0;
+   tex[2][0] = s1; tex[2][1] = t1;
+   tex[3][0] = s0; tex[3][1] = t1;
+
+   intel_meta_draw_poly(intel, 4, xy, z, color, tex);
+}
+
 
 
 /**********************************************************************/
diff --git a/src/mesa/drivers/dri/i915/intel_tris.h b/src/mesa/drivers/dri/i915/intel_tris.h
index d7e382fdb3..b4c2f40708 100644
--- a/src/mesa/drivers/dri/i915/intel_tris.h
+++ b/src/mesa/drivers/dri/i915/intel_tris.h
@@ -30,6 +30,8 @@
 
 #include "mtypes.h"
 
+
+
 #define _INTEL_NEW_RENDERSTATE (_DD_NEW_LINE_STIPPLE |		\
 			       _DD_NEW_TRI_UNFILLED |		\
 			       _DD_NEW_TRI_LIGHT_TWOSIDE |	\
@@ -40,7 +42,30 @@
 
 extern void intelInitTriFuncs( GLcontext *ctx );
 
-extern void intelPrintRenderState( const char *msg, GLuint state );
 extern void intelChooseRenderState( GLcontext *ctx );
 
+extern void intelStartInlinePrimitive( struct intel_context *intel, GLuint prim, GLuint flags );
+extern void intelWrapInlinePrimitive( struct intel_context *intel );
+
+GLuint *intelExtendInlinePrimitive( struct intel_context *intel, 
+				    GLuint dwords );
+
+
+void intel_meta_draw_quad(struct intel_context *intel, 
+			  GLfloat x0, GLfloat x1,
+			  GLfloat y0, GLfloat y1, 
+			  GLfloat z,
+			  GLuint color,
+			  GLfloat s0, GLfloat s1,
+			  GLfloat t0, GLfloat t1);
+
+void intel_meta_draw_poly(struct intel_context *intel, 
+			  GLuint n,
+			  GLfloat xy[][2],
+			  GLfloat z,
+			  GLuint color,
+			  GLfloat tex[][2]);
+
+
+
 #endif