[intel] Move over files that will be shared with 965-fbo work.

28 files changed, 9028 insertions, 0 deletions

diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.c b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
new file mode 100644
index 00000000000..8ece6d2760c
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.c
@@ -0,0 +1,250 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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 "intel_batchbuffer.h"
+#include "intel_ioctl.h"
+#include "intel_decode.h"
+#include "intel_reg.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.
+ *
+ */
+
+/* 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 ???
+ */
+
+void
+intel_batchbuffer_reset(struct intel_batchbuffer *batch)
+{
+   struct intel_context *intel = batch->intel;
+
+   if (batch->buf != NULL) {
+      dri_bo_unreference(batch->buf);
+      batch->buf = NULL;
+   }
+
+   batch->buf = dri_bo_alloc(intel->intelScreen->bufmgr, "batchbuffer",
+			     intel->intelScreen->maxBatchSize, 4096,
+			     DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED | DRM_BO_FLAG_CACHED_MAPPED);
+   dri_bo_map(batch->buf, GL_TRUE);
+   batch->map = batch->buf->virtual;
+   batch->size = intel->intelScreen->maxBatchSize;
+   batch->ptr = batch->map;
+}
+
+struct intel_batchbuffer *
+intel_batchbuffer_alloc(struct intel_context *intel)
+{
+   struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
+
+   batch->intel = intel;
+   batch->last_fence = NULL;
+   intel_batchbuffer_reset(batch);
+
+   return batch;
+}
+
+void
+intel_batchbuffer_free(struct intel_batchbuffer *batch)
+{
+   if (batch->last_fence) {
+      dri_fence_wait(batch->last_fence);
+      dri_fence_unreference(batch->last_fence);
+      batch->last_fence = NULL;
+   }
+   if (batch->map) {
+      dri_bo_unmap(batch->buf);
+      batch->map = NULL;
+   }
+   dri_bo_unreference(batch->buf);
+   batch->buf = NULL;
+   free(batch);
+}
+
+
+
+/* TODO: Push this whole function into bufmgr.
+ */
+static void
+do_flush_locked(struct intel_batchbuffer *batch,
+		GLuint used,
+		GLboolean ignore_cliprects, GLboolean allow_unlock)
+{
+   struct intel_context *intel = batch->intel;
+   void *start;
+   GLuint count;
+
+   start = dri_process_relocs(batch->buf, &count);
+
+   batch->map = NULL;
+   batch->ptr = NULL;
+   batch->flags = 0;
+
+   /* 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)) {
+      if (intel->intelScreen->ttm == GL_TRUE) {
+	 intel_exec_ioctl(batch->intel,
+			  used, ignore_cliprects, allow_unlock,
+			  start, count, &batch->last_fence);
+      } else {
+	 intel_batch_ioctl(batch->intel,
+			   batch->buf->offset,
+			   used, ignore_cliprects, allow_unlock);
+      }
+   }
+      
+   dri_post_submit(batch->buf, &batch->last_fence);
+
+   if (intel->numClipRects == 0 && !ignore_cliprects) {
+      if (allow_unlock) {
+	 /* If we are not doing any actual user-visible rendering,
+	  * do a sched_yield to keep the app from pegging the cpu while
+	  * achieving nothing.
+	  */
+         UNLOCK_HARDWARE(intel);
+         sched_yield();
+         LOCK_HARDWARE(intel);
+      }
+      intel->vtbl.lost_hardware(intel);
+   }
+
+   if (INTEL_DEBUG & DEBUG_BATCH) {
+      //      dri_bo_map(batch->buf, GL_FALSE);
+      //      intel_decode(ptr, used / 4, batch->buf->offset,
+      //		   intel->intelScreen->deviceID);
+      //      dri_bo_unmap(batch->buf);
+   }
+}
+
+void
+intel_batchbuffer_flush(struct intel_batchbuffer *batch)
+{
+   struct intel_context *intel = batch->intel;
+   GLuint used = batch->ptr - batch->map;
+   GLboolean was_locked = intel->locked;
+
+   if (used == 0)
+      return;
+
+   /* Add the MI_BATCH_BUFFER_END.  Always add an MI_FLUSH - this is a
+    * performance drain that we would like to avoid.
+    */
+   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;
+   }
+   else {
+      ((int *) batch->ptr)[0] = intel->vtbl.flush_cmd();
+      ((int *) batch->ptr)[1] = MI_BATCH_BUFFER_END;
+      used += 8;
+   }
+
+   /* TODO: Just pass the relocation list and dma buffer up to the
+    * kernel.
+    */
+   if (!was_locked)
+      LOCK_HARDWARE(intel);
+
+   do_flush_locked(batch, used, !(batch->flags & INTEL_BATCH_CLIPRECTS),
+		   GL_FALSE);
+     
+   if (!was_locked)
+      UNLOCK_HARDWARE(intel);
+
+   /* Reset the buffer:
+    */
+   intel_batchbuffer_reset(batch);
+}
+
+void
+intel_batchbuffer_finish(struct intel_batchbuffer *batch)
+{
+   intel_batchbuffer_flush(batch);
+   if (batch->last_fence != NULL)
+      dri_fence_wait(batch->last_fence);
+}
+
+
+/*  This is the only way buffers get added to the validate list.
+ */
+GLboolean
+intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
+                             dri_bo *buffer,
+                             GLuint flags, GLuint delta)
+{
+   dri_emit_reloc(batch->buf, flags, delta, batch->ptr - batch->map, buffer);
+   batch->ptr += 4;
+
+   return GL_TRUE;
+}
+
+void
+intel_batchbuffer_data(struct intel_batchbuffer *batch,
+                       const void *data, GLuint bytes, GLuint flags)
+{
+   assert((bytes & 3) == 0);
+   intel_batchbuffer_require_space(batch, bytes, flags);
+   __memcpy(batch->ptr, data, bytes);
+   batch->ptr += bytes;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_batchbuffer.h b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
new file mode 100644
index 00000000000..b5c7a783a72
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_batchbuffer.h
@@ -0,0 +1,109 @@
+#ifndef INTEL_BATCHBUFFER_H
+#define INTEL_BATCHBUFFER_H
+
+#include "mtypes.h"
+
+#include "dri_bufmgr.h"
+
+struct intel_context;
+
+#define BATCH_SZ 16384
+#define BATCH_RESERVED 16
+
+#define INTEL_BATCH_NO_CLIPRECTS 0x1
+#define INTEL_BATCH_CLIPRECTS    0x2
+
+struct intel_batchbuffer
+{
+   struct intel_context *intel;
+
+   dri_bo *buf;
+   dri_fence *last_fence;
+   GLuint flags;
+
+   GLubyte *map;
+   GLubyte *ptr;
+
+   GLuint size;
+};
+
+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);
+
+void intel_batchbuffer_flush(struct intel_batchbuffer *batch);
+
+void intel_batchbuffer_reset(struct intel_batchbuffer *batch);
+
+
+/* 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,
+                                       dri_bo *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)
+{
+   return (batch->size - BATCH_RESERVED) - (batch->ptr - batch->map);
+}
+
+
+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;
+}
+
+static INLINE void
+intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
+                                GLuint sz, GLuint flags)
+{
+   assert(sz < batch->size - 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/intel/intel_blit.c b/src/mesa/drivers/dri/intel/intel_blit.c
new file mode 100644
index 00000000000..2761136f47f
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_blit.c
@@ -0,0 +1,491 @@
+/**************************************************************************
+ * 
+ * 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 <stdio.h>
+#include <errno.h>
+
+#include "mtypes.h"
+#include "context.h"
+#include "enums.h"
+
+#include "intel_batchbuffer.h"
+#include "intel_blit.h"
+#include "intel_buffers.h"
+#include "intel_context.h"
+#include "intel_fbo.h"
+#include "intel_reg.h"
+#include "intel_regions.h"
+
+#define FILE_DEBUG_FLAG DEBUG_BLIT
+
+/**
+ * Copy the back color buffer to the front color buffer. 
+ * Used for SwapBuffers().
+ */
+void
+intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
+                const drm_clip_rect_t * rect)
+{
+
+   struct intel_context *intel;
+   const intelScreenPrivate *intelScreen;
+
+   DBG("%s\n", __FUNCTION__);
+
+   assert(dPriv);
+
+   intel = intelScreenContext(dPriv->driScreenPriv->private);
+   if (!intel)
+      return;
+
+   intelScreen = intel->intelScreen;
+
+   if (intel->last_swap_fence) {
+      dri_fence_wait(intel->last_swap_fence);
+      dri_fence_unreference(intel->last_swap_fence);
+      intel->last_swap_fence = NULL;
+   }
+   intel->last_swap_fence = intel->first_swap_fence;
+   intel->first_swap_fence = NULL;
+
+   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
+    * should work regardless.
+    */
+   LOCK_HARDWARE(intel);
+
+   if (dPriv && dPriv->numClipRects) {
+      struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+      const struct intel_region *frontRegion
+	 = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT);
+      const struct intel_region *backRegion
+	 = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT);
+      const int nbox = dPriv->numClipRects;
+      const drm_clip_rect_t *pbox = dPriv->pClipRects;
+      const int pitch = frontRegion->pitch;
+      const int cpp = frontRegion->cpp;
+      int BR13, CMD;
+      int i;
+
+      ASSERT(intel_fb);
+      ASSERT(intel_fb->Base.Name == 0);    /* Not a user-created FBO */
+      ASSERT(frontRegion);
+      ASSERT(backRegion);
+      ASSERT(frontRegion->pitch == backRegion->pitch);
+      ASSERT(frontRegion->cpp == backRegion->cpp);
+
+      if (cpp == 2) {
+	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24);
+	 CMD = XY_SRC_COPY_BLT_CMD;
+      }
+      else {
+	 BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
+	 CMD = (XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB);
+      }
+
+      for (i = 0; i < nbox; i++, pbox++) {
+	 drm_clip_rect_t box;
+
+	 if (pbox->x1 > pbox->x2 ||
+	     pbox->y1 > pbox->y2 ||
+	     pbox->x2 > intelScreen->width || pbox->y2 > intelScreen->height)
+	    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, INTEL_BATCH_NO_CLIPRECTS);
+	 OUT_BATCH(CMD);
+	 OUT_BATCH(BR13);
+	 OUT_BATCH((pbox->y1 << 16) | pbox->x1);
+	 OUT_BATCH((pbox->y2 << 16) | pbox->x2);
+
+	 OUT_RELOC(frontRegion->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+		   0);
+	 OUT_BATCH((pbox->y1 << 16) | pbox->x1);
+	 OUT_BATCH(BR13 & 0xffff);
+	 OUT_RELOC(backRegion->buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+		   0);
+
+	 ADVANCE_BATCH();
+      }
+
+      if (intel->first_swap_fence)
+	 dri_fence_unreference(intel->first_swap_fence);
+      intel_batchbuffer_flush(intel->batch);
+      intel->first_swap_fence = intel->batch->last_fence;
+      dri_fence_reference(intel->first_swap_fence);
+   }
+
+   UNLOCK_HARDWARE(intel);
+}
+
+
+
+
+void
+intelEmitFillBlit(struct intel_context *intel,
+                  GLuint cpp,
+                  GLshort dst_pitch,
+                  dri_bo *dst_buffer,
+                  GLuint dst_offset,
+                  GLshort x, GLshort y, GLshort w, GLshort h, GLuint color)
+{
+   GLuint BR13, CMD;
+   BATCH_LOCALS;
+
+   dst_pitch *= cpp;
+
+   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_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB);
+      break;
+   default:
+      return;
+   }
+
+   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+       __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h);
+
+
+   BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
+   OUT_BATCH(CMD);
+   OUT_BATCH(BR13);
+   OUT_BATCH((y << 16) | x);
+   OUT_BATCH(((y + h) << 16) | (x + w));
+   OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, dst_offset);
+   OUT_BATCH(color);
+   ADVANCE_BATCH();
+}
+
+
+static GLuint translate_raster_op(GLenum logicop)
+{
+   switch(logicop) {
+   case GL_CLEAR: return 0x00;
+   case GL_AND: return 0x88;
+   case GL_AND_REVERSE: return 0x44;
+   case GL_COPY: return 0xCC;
+   case GL_AND_INVERTED: return 0x22;
+   case GL_NOOP: return 0xAA;
+   case GL_XOR: return 0x66;
+   case GL_OR: return 0xEE;
+   case GL_NOR: return 0x11;
+   case GL_EQUIV: return 0x99;
+   case GL_INVERT: return 0x55;
+   case GL_OR_REVERSE: return 0xDD;
+   case GL_COPY_INVERTED: return 0x33;
+   case GL_OR_INVERTED: return 0xBB;
+   case GL_NAND: return 0x77;
+   case GL_SET: return 0xFF;
+   default: return 0;
+   }
+}
+
+
+/* Copy BitBlt
+ */
+void
+intelEmitCopyBlit(struct intel_context *intel,
+                  GLuint cpp,
+                  GLshort src_pitch,
+                  dri_bo *src_buffer,
+                  GLuint src_offset,
+                  GLshort dst_pitch,
+                  dri_bo *dst_buffer,
+                  GLuint dst_offset,
+                  GLshort src_x, GLshort src_y,
+                  GLshort dst_x, GLshort dst_y, 
+		  GLshort w, GLshort h,
+		  GLenum logic_op)
+{
+   GLuint CMD, BR13;
+   int dst_y2 = dst_y + h;
+   int dst_x2 = dst_x + w;
+   BATCH_LOCALS;
+
+
+   DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+       __FUNCTION__,
+       src_buffer, src_pitch, src_offset, src_x, src_y,
+       dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
+
+   src_pitch *= cpp;
+   dst_pitch *= cpp;
+
+   switch (cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = (((GLint) dst_pitch) & 0xffff) | 
+	 (translate_raster_op(logic_op) << 16) | (1 << 24);
+      CMD = XY_SRC_COPY_BLT_CMD;
+      break;
+   case 4:
+      BR13 =
+         (((GLint) dst_pitch) & 0xffff) | 
+	 (translate_raster_op(logic_op) << 16) | (1 << 24) | (1 << 25);
+      CMD =
+         (XY_SRC_COPY_BLT_CMD | XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB);
+      break;
+   default:
+      return;
+   }
+
+   if (dst_y2 < dst_y || dst_x2 < dst_x) {
+      return;
+   }
+
+   /* Initial y values don't seem to work with negative pitches.  If
+    * we adjust the offsets manually (below), it seems to work fine.
+    *
+    * On the other hand, if we always adjust, the hardware doesn't
+    * know which blit directions to use, so overlapping copypixels get
+    * the wrong result.
+    */
+   if (dst_pitch > 0 && src_pitch > 0) {
+      BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+      OUT_BATCH(CMD);
+      OUT_BATCH(BR13);
+      OUT_BATCH((dst_y << 16) | dst_x);
+      OUT_BATCH((dst_y2 << 16) | dst_x2);
+      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE, dst_offset);
+      OUT_BATCH((src_y << 16) | src_x);
+      OUT_BATCH(((GLint) src_pitch & 0xffff));
+      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ, src_offset);
+      ADVANCE_BATCH();
+   }
+   else {
+      BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+      OUT_BATCH(CMD);
+      OUT_BATCH(BR13);
+      OUT_BATCH((0 << 16) | dst_x);
+      OUT_BATCH((h << 16) | dst_x2);
+      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                dst_offset + dst_y * dst_pitch);
+      OUT_BATCH((0 << 16) | src_x);
+      OUT_BATCH(((GLint) src_pitch & 0xffff));
+      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                src_offset + src_y * src_pitch);
+      ADVANCE_BATCH();
+   }
+}
+
+
+/**
+ * Use blitting to clear the renderbuffers named by 'flags'.
+ * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferMask field
+ * since that might include software renderbuffers or renderbuffers
+ * which we're clearing with triangles.
+ * \param mask  bitmask of BUFFER_BIT_* values indicating buffers to clear
+ */
+void
+intelClearWithBlit(GLcontext * ctx, GLbitfield mask)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
+   GLuint clear_depth;
+   GLbitfield skipBuffers = 0;
+   BATCH_LOCALS;
+
+   DBG("%s %x\n", __FUNCTION__, mask);
+
+   /*
+    * Compute values for clearing the buffers.
+    */
+   clear_depth = 0;
+   if (mask & BUFFER_BIT_DEPTH) {
+      clear_depth = (GLuint) (fb->_DepthMax * ctx->Depth.Clear);
+   }
+   if (mask & BUFFER_BIT_STENCIL) {
+      clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
+   }
+
+   /* If clearing both depth and stencil, skip BUFFER_BIT_STENCIL in
+    * the loop below.
+    */
+   if ((mask & BUFFER_BIT_DEPTH) && (mask & BUFFER_BIT_STENCIL)) {
+      skipBuffers = BUFFER_BIT_STENCIL;
+   }
+
+   /* XXX Move this flush/lock into the following conditional? */
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+
+   if (intel->numClipRects) {
+      GLint cx, cy, cw, ch;
+      drm_clip_rect_t clear;
+      int i;
+
+      /* Get clear bounds after locking */
+      cx = fb->_Xmin;
+      cy = fb->_Ymin;
+      cw = fb->_Xmax - cx;
+      ch = fb->_Ymax - cy;
+
+      if (fb->Name == 0) {
+         /* clearing a window */
+
+         /* flip top to bottom */
+         clear.x1 = cx + intel->drawX;
+         clear.y1 = intel->driDrawable->y + intel->driDrawable->h - cy - ch;
+         clear.x2 = clear.x1 + cw;
+         clear.y2 = clear.y1 + ch;
+      }
+      else {
+         /* clearing FBO */
+         assert(intel->numClipRects == 1);
+         assert(intel->pClipRects == &intel->fboRect);
+         clear.x1 = cx;
+         clear.y1 = cy;
+         clear.x2 = clear.x1 + cw;
+         clear.y2 = clear.y1 + ch;
+         /* no change to mask */
+      }
+
+      for (i = 0; i < intel->numClipRects; i++) {
+         const drm_clip_rect_t *box = &intel->pClipRects[i];
+         drm_clip_rect_t b;
+         GLuint buf;
+         GLuint clearMask = mask;      /* use copy, since we modify it below */
+         GLboolean all = (cw == fb->Width && ch == fb->Height);
+
+         if (!all) {
+            intel_intersect_cliprects(&b, &clear, box);
+         }
+         else {
+            b = *box;
+         }
+
+         if (b.x1 >= b.x2 || b.y1 >= b.y2)
+            continue;
+
+         if (0)
+            _mesa_printf("clear %d,%d..%d,%d, mask %x\n",
+                         b.x1, b.y1, b.x2, b.y2, mask);
+
+         /* Loop over all renderbuffers */
+         for (buf = 0; buf < BUFFER_COUNT && clearMask; buf++) {
+            const GLbitfield bufBit = 1 << buf;
+            if ((clearMask & bufBit) && !(bufBit & skipBuffers)) {
+               /* OK, clear this renderbuffer */
+               struct intel_region *irb_region =
+		  intel_get_rb_region(fb, buf);
+               dri_bo *write_buffer =
+                  intel_region_buffer(intel->intelScreen, irb_region,
+                                      all ? INTEL_WRITE_FULL :
+                                      INTEL_WRITE_PART);
+
+               GLuint clearVal;
+               GLint pitch, cpp;
+               GLuint BR13, CMD;
+
+               ASSERT(irb_region);
+
+               pitch = irb_region->pitch;
+               cpp = irb_region->cpp;
+
+               DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+                   __FUNCTION__,
+                   irb_region->buffer, (pitch * cpp),
+                   irb_region->draw_offset,
+                   b.x1, b.y1, b.x2 - b.x1, b.y2 - b.y1);
+
+
+               /* Setup the blit command */
+               if (cpp == 4) {
+                  BR13 = (0xF0 << 16) | (pitch * cpp) | (1 << 24) | (1 << 25);
+                  if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
+                     CMD = XY_COLOR_BLT_CMD;
+                     if (clearMask & BUFFER_BIT_DEPTH)
+                        CMD |= XY_BLT_WRITE_RGB;
+                     if (clearMask & BUFFER_BIT_STENCIL)
+                        CMD |= XY_BLT_WRITE_ALPHA;
+                  }
+                  else {
+                     /* clearing RGBA */
+                     CMD = XY_COLOR_BLT_CMD |
+			XY_BLT_WRITE_ALPHA | XY_BLT_WRITE_RGB;
+                  }
+               }
+               else {
+                  ASSERT(cpp == 2 || cpp == 0);
+                  BR13 = (0xF0 << 16) | (pitch * cpp) | (1 << 24);
+                  CMD = XY_COLOR_BLT_CMD;
+               }
+
+               if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
+                  clearVal = clear_depth;
+               }
+               else {
+                  clearVal = (cpp == 4)
+                     ? intel->ClearColor8888 : intel->ClearColor565;
+               }
+               /*
+                  _mesa_debug(ctx, "hardware blit clear buf %d rb id %d\n",
+                  buf, irb->Base.Name);
+                */
+	       intel_wait_flips(intel, INTEL_BATCH_NO_CLIPRECTS);
+
+               BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
+               OUT_BATCH(CMD);
+               OUT_BATCH(BR13);
+               OUT_BATCH((b.y1 << 16) | b.x1);
+               OUT_BATCH((b.y2 << 16) | b.x2);
+               OUT_RELOC(write_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                         irb_region->draw_offset);
+               OUT_BATCH(clearVal);
+               ADVANCE_BATCH();
+               clearMask &= ~bufBit;    /* turn off bit, for faster loop exit */
+            }
+         }
+      }
+      intel_batchbuffer_flush(intel->batch);
+   }
+
+   UNLOCK_HARDWARE(intel);
+}
diff --git a/src/mesa/drivers/dri/intel/intel_blit.h b/src/mesa/drivers/dri/intel/intel_blit.h
new file mode 100644
index 00000000000..a66af863591
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_blit.h
@@ -0,0 +1,62 @@
+/**************************************************************************
+ * 
+ * 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_BLIT_H
+#define INTEL_BLIT_H
+
+#include "intel_context.h"
+#include "intel_ioctl.h"
+#include "dri_bufmgr.h"
+
+extern void intelCopyBuffer(const __DRIdrawablePrivate * dpriv,
+                            const drm_clip_rect_t * rect);
+
+extern void intelClearWithBlit(GLcontext * ctx, GLbitfield mask);
+
+extern void intelEmitCopyBlit(struct intel_context *intel,
+                              GLuint cpp,
+                              GLshort src_pitch,
+                              dri_bo *src_buffer,
+                              GLuint src_offset,
+                              GLshort dst_pitch,
+                              dri_bo *dst_buffer,
+                              GLuint dst_offset,
+                              GLshort srcx, GLshort srcy,
+                              GLshort dstx, GLshort dsty,
+                              GLshort w, GLshort h,
+			      GLenum logicop );
+
+extern void intelEmitFillBlit(struct intel_context *intel,
+                              GLuint cpp,
+                              GLshort dst_pitch,
+                              dri_bo *dst_buffer,
+                              GLuint dst_offset,
+                              GLshort x, GLshort y,
+                              GLshort w, GLshort h, GLuint color);
+
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.c b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
new file mode 100644
index 00000000000..5348822816c
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.c
@@ -0,0 +1,268 @@
+/**************************************************************************
+ * 
+ * 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 "imports.h"
+#include "mtypes.h"
+#include "bufferobj.h"
+
+#include "intel_context.h"
+#include "intel_buffer_objects.h"
+#include "intel_regions.h"
+#include "dri_bufmgr.h"
+
+/** Allocates a new dri_bo to store the data for the buffer object. */
+static void
+intel_bufferobj_alloc_buffer(struct intel_context *intel,
+			     struct intel_buffer_object *intel_obj)
+{
+   intel_obj->buffer = dri_bo_alloc(intel->intelScreen->bufmgr, "bufferobj",
+				    intel_obj->Base.Size, 64,
+				    DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_CACHED | DRM_BO_FLAG_CACHED_MAPPED);
+}
+
+/**
+ * There is some duplication between mesa's bufferobjects and our
+ * bufmgr buffers.  Both have an integer handle and a hashtable to
+ * lookup an opaque structure.  It would be nice if the handles and
+ * internal structure where somehow shared.
+ */
+static struct gl_buffer_object *
+intel_bufferobj_alloc(GLcontext * ctx, GLuint name, GLenum target)
+{
+   struct intel_buffer_object *obj = CALLOC_STRUCT(intel_buffer_object);
+
+   _mesa_initialize_buffer_object(&obj->Base, name, target);
+
+   obj->buffer = NULL;
+
+   return &obj->Base;
+}
+
+/* Break the COW tie to the region.  The region gets to keep the data.
+ */
+void
+intel_bufferobj_release_region(struct intel_context *intel,
+                               struct intel_buffer_object *intel_obj)
+{
+   assert(intel_obj->region->buffer == intel_obj->buffer);
+   intel_obj->region->pbo = NULL;
+   intel_obj->region = NULL;
+
+   dri_bo_unreference(intel_obj->buffer);
+   intel_obj->buffer = NULL;
+}
+
+/* Break the COW tie to the region.  Both the pbo and the region end
+ * up with a copy of the data.
+ */
+void
+intel_bufferobj_cow(struct intel_context *intel,
+                    struct intel_buffer_object *intel_obj)
+{
+   assert(intel_obj->region);
+   intel_region_cow(intel->intelScreen, intel_obj->region);
+}
+
+
+/**
+ * Deallocate/free a vertex/pixel buffer object.
+ * Called via glDeleteBuffersARB().
+ */
+static void
+intel_bufferobj_free(GLcontext * ctx, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+
+   if (intel_obj->region) {
+      intel_bufferobj_release_region(intel, intel_obj);
+   }
+   else if (intel_obj->buffer) {
+      dri_bo_unreference(intel_obj->buffer);
+   }
+
+   _mesa_free(intel_obj);
+}
+
+
+
+/**
+ * Allocate space for and store data in a buffer object.  Any data that was
+ * previously stored in the buffer object is lost.  If data is NULL,
+ * memory will be allocated, but no copy will occur.
+ * Called via glBufferDataARB().
+ */
+static void
+intel_bufferobj_data(GLcontext * ctx,
+                     GLenum target,
+                     GLsizeiptrARB size,
+                     const GLvoid * data,
+                     GLenum usage, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   intel_obj->Base.Size = size;
+   intel_obj->Base.Usage = usage;
+
+   if (intel_obj->region)
+      intel_bufferobj_release_region(intel, intel_obj);
+
+   if (intel_obj->buffer != NULL && intel_obj->buffer->size != size) {
+      dri_bo_unreference(intel_obj->buffer);
+      intel_obj->buffer = NULL;
+   }
+
+   intel_bufferobj_alloc_buffer(intel, intel_obj);
+
+   if (data != NULL)
+      dri_bo_subdata(intel_obj->buffer, 0, size, data);
+}
+
+
+/**
+ * Replace data in a subrange of buffer object.  If the data range
+ * specified by size + offset extends beyond the end of the buffer or
+ * if data is NULL, no copy is performed.
+ * Called via glBufferSubDataARB().
+ */
+static void
+intel_bufferobj_subdata(GLcontext * ctx,
+                        GLenum target,
+                        GLintptrARB offset,
+                        GLsizeiptrARB size,
+                        const GLvoid * data, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+
+   if (intel_obj->region)
+      intel_bufferobj_cow(intel, intel_obj);
+
+   dri_bo_subdata(intel_obj->buffer, offset, size, data);
+}
+
+
+/**
+ * Called via glGetBufferSubDataARB().
+ */
+static void
+intel_bufferobj_get_subdata(GLcontext * ctx,
+                            GLenum target,
+                            GLintptrARB offset,
+                            GLsizeiptrARB size,
+                            GLvoid * data, struct gl_buffer_object *obj)
+{
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+   dri_bo_get_subdata(intel_obj->buffer, offset, size, data);
+}
+
+
+
+/**
+ * Called via glMapBufferARB().
+ */
+static void *
+intel_bufferobj_map(GLcontext * ctx,
+                    GLenum target,
+                    GLenum access, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   /* XXX: Translate access to flags arg below:
+    */
+   assert(intel_obj);
+
+   if (intel_obj->region)
+      intel_bufferobj_cow(intel, intel_obj);
+
+   if (intel_obj->buffer == NULL) {
+      obj->Pointer = NULL;
+      return NULL;
+   }
+
+   dri_bo_map(intel_obj->buffer, GL_TRUE);
+   obj->Pointer = intel_obj->buffer->virtual;
+   return obj->Pointer;
+}
+
+
+/**
+ * Called via glMapBufferARB().
+ */
+static GLboolean
+intel_bufferobj_unmap(GLcontext * ctx,
+                      GLenum target, struct gl_buffer_object *obj)
+{
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+   if (intel_obj->buffer != NULL) {
+      assert(obj->Pointer);
+      dri_bo_unmap(intel_obj->buffer);
+      obj->Pointer = NULL;
+   }
+   return GL_TRUE;
+}
+
+dri_bo *
+intel_bufferobj_buffer(struct intel_context *intel,
+                       struct intel_buffer_object *intel_obj, GLuint flag)
+{
+   if (intel_obj->region) {
+      if (flag == INTEL_WRITE_PART)
+         intel_bufferobj_cow(intel, intel_obj);
+      else if (flag == INTEL_WRITE_FULL) {
+         intel_bufferobj_release_region(intel, intel_obj);
+	 intel_bufferobj_alloc_buffer(intel, intel_obj);
+      }
+   }
+
+   return intel_obj->buffer;
+}
+
+void
+intel_bufferobj_init(struct intel_context *intel)
+{
+   GLcontext *ctx = &intel->ctx;
+
+   ctx->Driver.NewBufferObject = intel_bufferobj_alloc;
+   ctx->Driver.DeleteBuffer = intel_bufferobj_free;
+   ctx->Driver.BufferData = intel_bufferobj_data;
+   ctx->Driver.BufferSubData = intel_bufferobj_subdata;
+   ctx->Driver.GetBufferSubData = intel_bufferobj_get_subdata;
+   ctx->Driver.MapBuffer = intel_bufferobj_map;
+   ctx->Driver.UnmapBuffer = intel_bufferobj_unmap;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_buffer_objects.h b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
new file mode 100644
index 00000000000..db579a8ae40
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_buffer_objects.h
@@ -0,0 +1,86 @@
+ /**************************************************************************
+ * 
+ * Copyright 2005 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_BUFFEROBJ_H
+#define INTEL_BUFFEROBJ_H
+
+#include "mtypes.h"
+
+struct intel_context;
+struct intel_region;
+struct gl_buffer_object;
+
+
+/**
+ * Intel vertex/pixel buffer object, derived from Mesa's gl_buffer_object.
+ */
+struct intel_buffer_object
+{
+   struct gl_buffer_object Base;
+   dri_bo *buffer;     /* the low-level buffer manager's buffer handle */
+
+   struct intel_region *region; /* Is there a zero-copy texture
+                                   associated with this (pixel)
+                                   buffer object? */
+};
+
+
+/* Get the bm buffer associated with a GL bufferobject:
+ */
+dri_bo *intel_bufferobj_buffer(struct intel_context *intel,
+			       struct intel_buffer_object
+			       *obj, GLuint flag);
+
+/* Hook the bufferobject implementation into mesa: 
+ */
+void intel_bufferobj_init(struct intel_context *intel);
+
+
+
+/* Are the obj->Name tests necessary?  Unfortunately yes, mesa
+ * allocates a couple of gl_buffer_object structs statically, and
+ * the Name == 0 test is the only way to identify them and avoid
+ * casting them erroneously to our structs.
+ */
+static INLINE struct intel_buffer_object *
+intel_buffer_object(struct gl_buffer_object *obj)
+{
+   if (obj->Name)
+      return (struct intel_buffer_object *) obj;
+   else
+      return NULL;
+}
+
+/* Helpers for zerocopy image uploads.  See also intel_regions.h:
+ */
+void intel_bufferobj_cow(struct intel_context *intel,
+                         struct intel_buffer_object *intel_obj);
+void intel_bufferobj_release_region(struct intel_context *intel,
+                                    struct intel_buffer_object *intel_obj);
+
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_buffers.c b/src/mesa/drivers/dri/intel/intel_buffers.c
new file mode 100644
index 00000000000..1ae8b5feb4c
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_buffers.c
@@ -0,0 +1,1209 @@
+/**************************************************************************
+ * 
+ * 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 "intel_screen.h"
+#include "intel_context.h"
+#include "intel_blit.h"
+#include "intel_buffers.h"
+#include "intel_depthstencil.h"
+#include "intel_fbo.h"
+#include "intel_tris.h"
+#include "intel_regions.h"
+#include "intel_batchbuffer.h"
+#include "intel_reg.h"
+#include "context.h"
+#include "utils.h"
+#include "drirenderbuffer.h"
+#include "framebuffer.h"
+#include "swrast/swrast.h"
+#include "vblank.h"
+
+
+/* This block can be removed when libdrm >= 2.3.1 is required */
+
+#ifndef DRM_IOCTL_I915_FLIP
+
+#define DRM_VBLANK_FLIP 0x8000000
+
+typedef struct drm_i915_flip {
+   int pipes;
+} drm_i915_flip_t;
+
+#undef DRM_IOCTL_I915_FLIP
+#define DRM_IOCTL_I915_FLIP DRM_IOW(DRM_COMMAND_BASE + DRM_I915_FLIP, \
+				    drm_i915_flip_t)
+
+#endif
+
+
+/**
+ * XXX move this into a new dri/common/cliprects.c file.
+ */
+GLboolean
+intel_intersect_cliprects(drm_clip_rect_t * dst,
+                          const drm_clip_rect_t * a,
+                          const drm_clip_rect_t * b)
+{
+   GLint bx = b->x1;
+   GLint by = b->y1;
+   GLint bw = b->x2 - bx;
+   GLint bh = b->y2 - by;
+
+   if (bx < a->x1)
+      bw -= a->x1 - bx, bx = a->x1;
+   if (by < a->y1)
+      bh -= a->y1 - by, by = a->y1;
+   if (bx + bw > a->x2)
+      bw = a->x2 - bx;
+   if (by + bh > a->y2)
+      bh = a->y2 - by;
+   if (bw <= 0)
+      return GL_FALSE;
+   if (bh <= 0)
+      return GL_FALSE;
+
+   dst->x1 = bx;
+   dst->y1 = by;
+   dst->x2 = bx + bw;
+   dst->y2 = by + bh;
+
+   return GL_TRUE;
+}
+
+/**
+ * Return pointer to current color drawing region, or NULL.
+ */
+struct intel_region *
+intel_drawbuf_region(struct intel_context *intel)
+{
+   struct intel_renderbuffer *irbColor =
+      intel_renderbuffer(intel->ctx.DrawBuffer->_ColorDrawBuffers[0][0]);
+   if (irbColor)
+      return irbColor->region;
+   else
+      return NULL;
+}
+
+/**
+ * Return pointer to current color reading region, or NULL.
+ */
+struct intel_region *
+intel_readbuf_region(struct intel_context *intel)
+{
+   struct intel_renderbuffer *irb
+      = intel_renderbuffer(intel->ctx.ReadBuffer->_ColorReadBuffer);
+   if (irb)
+      return irb->region;
+   else
+      return NULL;
+}
+
+
+
+/**
+ * Update the following fields for rendering to a user-created FBO:
+ *   intel->numClipRects
+ *   intel->pClipRects
+ *   intel->drawX
+ *   intel->drawY
+ */
+static void
+intelSetRenderbufferClipRects(struct intel_context *intel)
+{
+   assert(intel->ctx.DrawBuffer->Width > 0);
+   assert(intel->ctx.DrawBuffer->Height > 0);
+   intel->fboRect.x1 = 0;
+   intel->fboRect.y1 = 0;
+   intel->fboRect.x2 = intel->ctx.DrawBuffer->Width;
+   intel->fboRect.y2 = intel->ctx.DrawBuffer->Height;
+   intel->numClipRects = 1;
+   intel->pClipRects = &intel->fboRect;
+   intel->drawX = 0;
+   intel->drawY = 0;
+}
+
+
+/**
+ * As above, but for rendering to front buffer of a window.
+ * \sa intelSetRenderbufferClipRects
+ */
+static void
+intelSetFrontClipRects(struct intel_context *intel)
+{
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+
+   if (!dPriv)
+      return;
+
+   intel->numClipRects = dPriv->numClipRects;
+   intel->pClipRects = dPriv->pClipRects;
+   intel->drawX = dPriv->x;
+   intel->drawY = dPriv->y;
+}
+
+
+/**
+ * As above, but for rendering to back buffer of a window.
+ */
+static void
+intelSetBackClipRects(struct intel_context *intel)
+{
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+   struct intel_framebuffer *intel_fb;
+
+   if (!dPriv)
+      return;
+
+   intel_fb = dPriv->driverPrivate;
+
+   if (intel_fb->pf_active || dPriv->numBackClipRects == 0) {
+      /* use the front clip rects */
+      intel->numClipRects = dPriv->numClipRects;
+      intel->pClipRects = dPriv->pClipRects;
+      intel->drawX = dPriv->x;
+      intel->drawY = dPriv->y;
+   }
+   else {
+      /* use the back clip rects */
+      intel->numClipRects = dPriv->numBackClipRects;
+      intel->pClipRects = dPriv->pBackClipRects;
+      intel->drawX = dPriv->backX;
+      intel->drawY = dPriv->backY;
+   }
+}
+
+
+/**
+ * This will be called whenever the currently bound window is moved/resized.
+ * XXX: actually, it seems to NOT be called when the window is only moved (BP).
+ */
+void
+intelWindowMoved(struct intel_context *intel)
+{
+   GLcontext *ctx = &intel->ctx;
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+   struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+
+   if (!intel->ctx.DrawBuffer) {
+      /* when would this happen? -BP */
+      intelSetFrontClipRects(intel);
+   }
+   else if (intel->ctx.DrawBuffer->Name != 0) {
+      /* drawing to user-created FBO - do nothing */
+      /* Cliprects would be set from intelDrawBuffer() */
+   }
+   else {
+      /* drawing to a window */
+      switch (intel_fb->Base._ColorDrawBufferMask[0]) {
+      case BUFFER_BIT_FRONT_LEFT:
+         intelSetFrontClipRects(intel);
+         break;
+      case BUFFER_BIT_BACK_LEFT:
+         intelSetBackClipRects(intel);
+         break;
+      default:
+         /* glDrawBuffer(GL_NONE or GL_FRONT_AND_BACK): software fallback */
+         intelSetFrontClipRects(intel);
+      }
+   }
+
+   if (intel->intelScreen->driScrnPriv->ddx_version.minor >= 7) {
+      drmI830Sarea *sarea = intel->sarea;
+      drm_clip_rect_t drw_rect = { .x1 = dPriv->x, .x2 = dPriv->x + dPriv->w,
+				   .y1 = dPriv->y, .y2 = dPriv->y + dPriv->h };
+      drm_clip_rect_t planeA_rect = { .x1 = sarea->planeA_x, .y1 = sarea->planeA_y,
+				     .x2 = sarea->planeA_x + sarea->planeA_w,
+				     .y2 = sarea->planeA_y + sarea->planeA_h };
+      drm_clip_rect_t planeB_rect = { .x1 = sarea->planeB_x, .y1 = sarea->planeB_y,
+				     .x2 = sarea->planeB_x + sarea->planeB_w,
+				     .y2 = sarea->planeB_y + sarea->planeB_h };
+      GLint areaA = driIntersectArea( drw_rect, planeA_rect );
+      GLint areaB = driIntersectArea( drw_rect, planeB_rect );
+      GLuint flags = dPriv->vblFlags;
+      GLboolean pf_active;
+      GLint pf_planes;
+
+      /* Update page flipping info
+       */
+      pf_planes = 0;
+
+      if (areaA > 0)
+	 pf_planes |= 1;
+
+      if (areaB > 0)
+	 pf_planes |= 2;
+
+      intel_fb->pf_current_page = (intel->sarea->pf_current_page >>
+				   (intel_fb->pf_planes & 0x2)) & 0x3;
+
+      intel_fb->pf_num_pages = intel->intelScreen->third.handle ? 3 : 2;
+
+      pf_active = pf_planes && (pf_planes & intel->sarea->pf_active) == pf_planes;
+
+      if (INTEL_DEBUG & DEBUG_LOCK)
+	 if (pf_active != intel_fb->pf_active)
+	    _mesa_printf("%s - Page flipping %sactive\n", __progname,
+			 pf_active ? "" : "in");
+
+      if (pf_active) {
+	 /* Sync pages between planes if flipping on both at the same time */
+	 if (pf_planes == 0x3 && pf_planes != intel_fb->pf_planes &&
+	     (intel->sarea->pf_current_page & 0x3) !=
+	     (((intel->sarea->pf_current_page) >> 2) & 0x3)) {
+	    drm_i915_flip_t flip;
+
+	    if (intel_fb->pf_current_page ==
+		(intel->sarea->pf_current_page & 0x3)) {
+	       /* XXX: This is ugly, but emitting two flips 'in a row' can cause
+		* lockups for unknown reasons.
+		*/
+               intel->sarea->pf_current_page =
+		  intel->sarea->pf_current_page & 0x3;
+	       intel->sarea->pf_current_page |=
+		  ((intel_fb->pf_current_page + intel_fb->pf_num_pages - 1) %
+		   intel_fb->pf_num_pages) << 2;
+
+	       flip.pipes = 0x2;
+	    } else {
+               intel->sarea->pf_current_page =
+		  intel->sarea->pf_current_page & (0x3 << 2);
+	       intel->sarea->pf_current_page |=
+		  (intel_fb->pf_current_page + intel_fb->pf_num_pages - 1) %
+		  intel_fb->pf_num_pages;
+
+	       flip.pipes = 0x1;
+	    }
+
+	    drmCommandWrite(intel->driFd, DRM_I915_FLIP, &flip, sizeof(flip));
+	 }
+
+	 intel_fb->pf_planes = pf_planes;
+      }
+
+      intel_fb->pf_active = pf_active;
+      intel_flip_renderbuffers(intel_fb);
+      intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer);
+
+      /* Update vblank info
+       */
+      if (areaB > areaA || (areaA == areaB && areaB > 0)) {
+	 flags = dPriv->vblFlags | VBLANK_FLAG_SECONDARY;
+      } else {
+	 flags = dPriv->vblFlags & ~VBLANK_FLAG_SECONDARY;
+      }
+
+      /* Check to see if we changed pipes */
+      if (flags != dPriv->vblFlags && dPriv->vblFlags &&
+	  !(dPriv->vblFlags & VBLANK_FLAG_NO_IRQ)) {
+	 int64_t count;
+	 drmVBlank vbl;
+	 int i;
+
+	 /*
+	  * Deal with page flipping
+	  */
+	 vbl.request.type = DRM_VBLANK_ABSOLUTE;
+
+	 if ( dPriv->vblFlags & VBLANK_FLAG_SECONDARY ) {
+	    vbl.request.type |= DRM_VBLANK_SECONDARY;
+	 }
+
+	 for (i = 0; i < intel_fb->pf_num_pages; i++) {
+	    if (!intel_fb->color_rb[i] ||
+		(intel_fb->vbl_waited - intel_fb->color_rb[i]->vbl_pending) <=
+		(1<<23))
+	       continue;
+
+	    vbl.request.sequence = intel_fb->color_rb[i]->vbl_pending;
+	    drmWaitVBlank(intel->driFd, &vbl);
+	 }
+
+	 /*
+	  * Update msc_base from old pipe
+	  */
+	 driDrawableGetMSC32(dPriv->driScreenPriv, dPriv, &count);
+	 dPriv->msc_base = count;
+	 /*
+	  * Then get new vblank_base and vblSeq values
+	  */
+	 dPriv->vblFlags = flags;
+	 driGetCurrentVBlank(dPriv);
+	 dPriv->vblank_base = dPriv->vblSeq;
+
+	 intel_fb->vbl_waited = dPriv->vblSeq;
+
+	 for (i = 0; i < intel_fb->pf_num_pages; i++) {
+	    if (intel_fb->color_rb[i])
+	       intel_fb->color_rb[i]->vbl_pending = intel_fb->vbl_waited;
+	 }
+      }
+   } else {
+      dPriv->vblFlags &= ~VBLANK_FLAG_SECONDARY;
+   }
+
+   /* Update Mesa's notion of window size */
+   driUpdateFramebufferSize(ctx, dPriv);
+   intel_fb->Base.Initialized = GL_TRUE; /* XXX remove someday */
+
+   /* Update hardware scissor */
+   ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
+                       ctx->Scissor.Width, ctx->Scissor.Height);
+
+   /* Re-calculate viewport related state */
+   ctx->Driver.DepthRange( ctx, ctx->Viewport.Near, ctx->Viewport.Far );
+}
+
+
+
+/* A true meta version of this would be very simple and additionally
+ * machine independent.  Maybe we'll get there one day.
+ */
+static void
+intelClearWithTris(struct intel_context *intel, GLbitfield mask)
+{
+   GLcontext *ctx = &intel->ctx;
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
+   drm_clip_rect_t clear;
+
+   if (INTEL_DEBUG & DEBUG_BLIT)
+      _mesa_printf("%s 0x%x\n", __FUNCTION__, mask);
+
+   LOCK_HARDWARE(intel);
+
+   /* XXX FBO: was: intel->driDrawable->numClipRects */
+   if (intel->numClipRects) {
+      GLint cx, cy, cw, ch;
+      GLuint buf;
+
+      intel->vtbl.install_meta_state(intel);
+
+      /* Get clear bounds after locking */
+      cx = fb->_Xmin;
+      cy = fb->_Ymin;
+      ch = fb->_Ymax - cx;
+      cw = fb->_Xmax - cy;
+
+      /* note: regardless of 'all', cx, cy, cw, ch are now correct */
+      clear.x1 = cx;
+      clear.y1 = cy;
+      clear.x2 = cx + cw;
+      clear.y2 = cy + ch;
+
+      /* Back and stencil cliprects are the same.  Try and do both
+       * buffers at once:
+       */
+      if (mask &
+          (BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH)) {
+         struct intel_region *backRegion =
+            intel_get_rb_region(fb, BUFFER_BACK_LEFT);
+         struct intel_region *depthRegion =
+            intel_get_rb_region(fb, BUFFER_DEPTH);
+         const GLuint clearColor = (backRegion && backRegion->cpp == 4)
+            ? intel->ClearColor8888 : intel->ClearColor565;
+
+         intel->vtbl.meta_draw_region(intel, backRegion, depthRegion);
+
+         if (mask & BUFFER_BIT_BACK_LEFT)
+            intel->vtbl.meta_color_mask(intel, GL_TRUE);
+         else
+            intel->vtbl.meta_color_mask(intel, GL_FALSE);
+
+         if (mask & BUFFER_BIT_STENCIL)
+            intel->vtbl.meta_stencil_replace(intel,
+                                             intel->ctx.Stencil.WriteMask[0],
+                                             intel->ctx.Stencil.Clear);
+         else
+            intel->vtbl.meta_no_stencil_write(intel);
+
+         if (mask & BUFFER_BIT_DEPTH)
+            intel->vtbl.meta_depth_replace(intel);
+         else
+            intel->vtbl.meta_no_depth_write(intel);
+
+         /* XXX: Using INTEL_BATCH_NO_CLIPRECTS here is dangerous as the
+          * drawing origin may not be correctly emitted.
+          */
+         intel_meta_draw_quad(intel, clear.x1, clear.x2, clear.y1, clear.y2, intel->ctx.Depth.Clear, clearColor, 0, 0, 0, 0);   /* texcoords */
+
+         mask &=
+            ~(BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH);
+      }
+
+      /* clear the remaining (color) renderbuffers */
+      for (buf = 0; buf < BUFFER_COUNT && mask; buf++) {
+         const GLuint bufBit = 1 << buf;
+         if (mask & bufBit) {
+            struct intel_renderbuffer *irbColor =
+               intel_renderbuffer(fb->Attachment[buf].Renderbuffer);
+            GLuint color = (irbColor->region->cpp == 4)
+               ? intel->ClearColor8888 : intel->ClearColor565;
+
+            ASSERT(irbColor);
+
+            intel->vtbl.meta_no_depth_write(intel);
+            intel->vtbl.meta_no_stencil_write(intel);
+            intel->vtbl.meta_color_mask(intel, GL_TRUE);
+            intel->vtbl.meta_draw_region(intel, irbColor->region, NULL);
+
+            /* XXX: Using INTEL_BATCH_NO_CLIPRECTS here is dangerous as the
+             * drawing origin may not be correctly emitted.
+             */
+            intel_meta_draw_quad(intel, clear.x1, clear.x2, clear.y1, clear.y2, 0,      /* depth clear val */
+                                 color, 0, 0, 0, 0);    /* texcoords */
+
+            mask &= ~bufBit;
+         }
+      }
+
+      intel->vtbl.leave_meta_state(intel);
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(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
+intelRotateWindow(struct intel_context *intel,
+                  __DRIdrawablePrivate * dPriv, GLuint srcBuf)
+{
+   intelScreenPrivate *screen = intel->intelScreen;
+   drm_clip_rect_t fullRect;
+   struct intel_framebuffer *intel_fb;
+   struct intel_region *src;
+   const drm_clip_rect_t *clipRects;
+   int numClipRects;
+   int i;
+   GLenum format, type;
+
+   int xOrig, yOrig;
+   int origNumClipRects;
+   drm_clip_rect_t *origRects;
+
+   /*
+    * set up hardware state
+    */
+   intelFlush(&intel->ctx);
+
+   LOCK_HARDWARE(intel);
+
+   if (!intel->numClipRects) {
+      UNLOCK_HARDWARE(intel);
+      return;
+   }
+
+   intel->vtbl.install_meta_state(intel);
+
+   intel->vtbl.meta_no_depth_write(intel);
+   intel->vtbl.meta_no_stencil_write(intel);
+   intel->vtbl.meta_color_mask(intel, GL_FALSE);
+
+
+   /* save current drawing origin and cliprects (restored at end) */
+   xOrig = intel->drawX;
+   yOrig = intel->drawY;
+   origNumClipRects = intel->numClipRects;
+   origRects = intel->pClipRects;
+
+   /*
+    * 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;
+
+   intel->vtbl.meta_draw_region(intel, screen->rotated_region, NULL);    /* ? */
+
+   intel_fb = dPriv->driverPrivate;
+
+   if ((srcBuf == BUFFER_BIT_BACK_LEFT && !intel_fb->pf_active)) {
+      src = intel_get_rb_region(&intel_fb->Base, BUFFER_BACK_LEFT);
+      clipRects = dPriv->pBackClipRects;
+      numClipRects = dPriv->numBackClipRects;
+   }
+   else {
+      src = intel_get_rb_region(&intel_fb->Base, BUFFER_FRONT_LEFT);
+      clipRects = dPriv->pClipRects;
+      numClipRects = dPriv->numClipRects;
+   }
+
+   if (src->cpp == 4) {
+      format = GL_BGRA;
+      type = GL_UNSIGNED_BYTE;
+   }
+   else {
+      format = GL_BGR;
+      type = GL_UNSIGNED_SHORT_5_6_5_REV;
+   }
+
+   /* set the whole screen up as a texture to avoid alignment issues */
+   intel->vtbl.meta_tex_rect_source(intel,
+                                    src->buffer,
+                                    screen->width,
+                                    screen->height, src->pitch, format, type);
+
+   intel->vtbl.meta_texture_blend_replace(intel);
+
+   /*
+    * 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 */
+      intel_meta_draw_poly(intel, 4, verts, 0, 0, tex);
+
+   }                            /* cliprect loop */
+
+   intel->vtbl.leave_meta_state(intel);
+   intel_batchbuffer_flush(intel->batch);
+
+   /* restore original drawing origin and cliprects */
+   intel->drawX = xOrig;
+   intel->drawY = yOrig;
+   intel->numClipRects = origNumClipRects;
+   intel->pClipRects = origRects;
+
+   UNLOCK_HARDWARE(intel);
+}
+
+
+/**
+ * Called by ctx->Driver.Clear.
+ */
+static void
+intelClear(GLcontext *ctx, GLbitfield mask)
+{
+   struct intel_context *intel = intel_context(ctx);
+   const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask);
+   GLbitfield tri_mask = 0;
+   GLbitfield blit_mask = 0;
+   GLbitfield swrast_mask = 0;
+   struct gl_framebuffer *fb = ctx->DrawBuffer;
+   GLuint i;
+
+   if (0)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   /* HW color buffers (front, back, aux, generic FBO, etc) */
+   if (colorMask == ~0) {
+      /* clear all R,G,B,A */
+      /* XXX FBO: need to check if colorbuffers are software RBOs! */
+      blit_mask |= (mask & BUFFER_BITS_COLOR);
+   }
+   else {
+      /* glColorMask in effect */
+      tri_mask |= (mask & BUFFER_BITS_COLOR);
+   }
+
+   /* HW stencil */
+   if (mask & BUFFER_BIT_STENCIL) {
+      const struct intel_region *stencilRegion
+         = intel_get_rb_region(fb, BUFFER_STENCIL);
+      if (stencilRegion) {
+         /* have hw stencil */
+         if ((ctx->Stencil.WriteMask[0] & 0xff) != 0xff) {
+            /* not clearing all stencil bits, so use triangle clearing */
+            tri_mask |= BUFFER_BIT_STENCIL;
+         }
+         else {
+            /* clearing all stencil bits, use blitting */
+            blit_mask |= BUFFER_BIT_STENCIL;
+         }
+      }
+   }
+
+   /* HW depth */
+   if (mask & BUFFER_BIT_DEPTH) {
+      /* clear depth with whatever method is used for stencil (see above) */
+      if (tri_mask & BUFFER_BIT_STENCIL)
+         tri_mask |= BUFFER_BIT_DEPTH;
+      else
+         blit_mask |= BUFFER_BIT_DEPTH;
+   }
+
+   /* SW fallback clearing */
+   swrast_mask = mask & ~tri_mask & ~blit_mask;
+
+   for (i = 0; i < BUFFER_COUNT; i++) {
+      GLuint bufBit = 1 << i;
+      if ((blit_mask | tri_mask) & bufBit) {
+         if (!fb->Attachment[i].Renderbuffer->ClassID) {
+            blit_mask &= ~bufBit;
+            tri_mask &= ~bufBit;
+            swrast_mask |= bufBit;
+         }
+      }
+   }
+
+
+   intelFlush(ctx);             /* XXX intelClearWithBlit also does this */
+
+   if (blit_mask)
+      intelClearWithBlit(ctx, blit_mask);
+
+   if (tri_mask)
+      intelClearWithTris(intel, tri_mask);
+
+   if (swrast_mask)
+      _swrast_Clear(ctx, swrast_mask);
+}
+
+
+/* Emit wait for pending flips */
+void
+intel_wait_flips(struct intel_context *intel, GLuint batch_flags)
+{
+   struct intel_framebuffer *intel_fb =
+      (struct intel_framebuffer *) intel->ctx.DrawBuffer;
+   struct intel_renderbuffer *intel_rb =
+      intel_get_renderbuffer(&intel_fb->Base,
+			     intel_fb->Base._ColorDrawBufferMask[0] ==
+			     BUFFER_BIT_FRONT_LEFT ? BUFFER_FRONT_LEFT :
+			     BUFFER_BACK_LEFT);
+
+   if (intel_fb->Base.Name == 0 && intel_rb->pf_pending == intel_fb->pf_seq) {
+      GLint pf_planes = intel_fb->pf_planes;
+      BATCH_LOCALS;
+
+      /* Wait for pending flips to take effect */
+      BEGIN_BATCH(2, batch_flags);
+      OUT_BATCH(pf_planes & 0x1 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_A_FLIP)
+		: 0);
+      OUT_BATCH(pf_planes & 0x2 ? (MI_WAIT_FOR_EVENT | MI_WAIT_FOR_PLANE_B_FLIP)
+		: 0);
+      ADVANCE_BATCH();
+
+      intel_rb->pf_pending--;
+   }
+}
+
+
+/* Flip the front & back buffers
+ */
+static GLboolean
+intelPageFlip(const __DRIdrawablePrivate * dPriv)
+{
+   struct intel_context *intel;
+   int ret;
+   struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+
+   if (INTEL_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   intel = (struct intel_context *) dPriv->driContextPriv->driverPrivate;
+
+   if (intel->intelScreen->drmMinor < 9)
+      return GL_FALSE;
+
+   intelFlush(&intel->ctx);
+
+   ret = 0;
+
+   LOCK_HARDWARE(intel);
+
+   if (dPriv->numClipRects && intel_fb->pf_active) {
+      drm_i915_flip_t flip;
+
+      flip.pipes = intel_fb->pf_planes;
+
+      ret = drmCommandWrite(intel->driFd, DRM_I915_FLIP, &flip, sizeof(flip));
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+   if (ret || !intel_fb->pf_active)
+      return GL_FALSE;
+
+   if (!dPriv->numClipRects) {
+      usleep(10000);	/* throttle invisible client 10ms */
+   }
+
+   intel_fb->pf_current_page = (intel->sarea->pf_current_page >>
+				(intel_fb->pf_planes & 0x2)) & 0x3;
+
+   if (dPriv->numClipRects != 0) {
+      intel_get_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT)->pf_pending =
+      intel_get_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT)->pf_pending =
+	 ++intel_fb->pf_seq;
+   }
+
+   intel_flip_renderbuffers(intel_fb);
+   intel_draw_buffer(&intel->ctx, &intel_fb->Base);
+
+   return GL_TRUE;
+}
+
+#if 0
+void
+intelSwapBuffers(__DRIdrawablePrivate * dPriv)
+{
+   if (dPriv->driverPrivate) {
+      const struct gl_framebuffer *fb
+         = (struct gl_framebuffer *) dPriv->driverPrivate;
+      if (fb->Visual.doubleBufferMode) {
+         GET_CURRENT_CONTEXT(ctx);
+         if (ctx && ctx->DrawBuffer == fb) {
+            _mesa_notifySwapBuffers(ctx);       /* flush pending rendering */
+         }
+         if (intel->doPageFlip) {
+            intelPageFlip(dPriv);
+         }
+         else {
+            intelCopyBuffer(dPriv);
+         }
+      }
+   }
+   else {
+      _mesa_problem(NULL,
+                    "dPriv has no gl_framebuffer pointer in intelSwapBuffers");
+   }
+}
+#else
+/* Trunk version:
+ */
+
+static GLboolean
+intelScheduleSwap(__DRIdrawablePrivate * dPriv, GLboolean *missed_target)
+{
+   struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+   unsigned int interval;
+   struct intel_context *intel =
+      intelScreenContext(dPriv->driScreenPriv->private);
+   const intelScreenPrivate *intelScreen = intel->intelScreen;
+   unsigned int target;
+   drm_i915_vblank_swap_t swap;
+   GLboolean ret;
+
+   if (!dPriv->vblFlags ||
+       (dPriv->vblFlags & VBLANK_FLAG_NO_IRQ) ||
+       intelScreen->current_rotation != 0 ||
+       intelScreen->drmMinor < (intel_fb->pf_active ? 9 : 6))
+      return GL_FALSE;
+
+   interval = driGetVBlankInterval(dPriv);
+
+   swap.seqtype = DRM_VBLANK_ABSOLUTE;
+
+   if (dPriv->vblFlags & VBLANK_FLAG_SYNC) {
+      swap.seqtype |= DRM_VBLANK_NEXTONMISS;
+   } else if (interval == 0) {
+      return GL_FALSE;
+   }
+
+   swap.drawable = dPriv->hHWDrawable;
+   target = swap.sequence = dPriv->vblSeq + interval;
+
+   if ( dPriv->vblFlags & VBLANK_FLAG_SECONDARY ) {
+      swap.seqtype |= DRM_VBLANK_SECONDARY;
+   }
+
+   LOCK_HARDWARE(intel);
+
+   intel_batchbuffer_flush(intel->batch);
+
+   if ( intel_fb->pf_active ) {
+      swap.seqtype |= DRM_VBLANK_FLIP;
+
+      intel_fb->pf_current_page = (((intel->sarea->pf_current_page >>
+				     (intel_fb->pf_planes & 0x2)) & 0x3) + 1) %
+				  intel_fb->pf_num_pages;
+   }
+
+   if (!drmCommandWriteRead(intel->driFd, DRM_I915_VBLANK_SWAP, &swap,
+			    sizeof(swap))) {
+      dPriv->vblSeq = swap.sequence;
+      swap.sequence -= target;
+      *missed_target = swap.sequence > 0 && swap.sequence <= (1 << 23);
+
+      intel_get_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT)->vbl_pending =
+	 intel_get_renderbuffer(&intel_fb->Base,
+				BUFFER_FRONT_LEFT)->vbl_pending =
+	 dPriv->vblSeq;
+
+      if (swap.seqtype & DRM_VBLANK_FLIP) {
+	 intel_flip_renderbuffers(intel_fb);
+	 intel_draw_buffer(&intel->ctx, intel->ctx.DrawBuffer);
+      }
+
+      ret = GL_TRUE;
+   } else {
+      if (swap.seqtype & DRM_VBLANK_FLIP) {
+	 intel_fb->pf_current_page = ((intel->sarea->pf_current_page >>
+					(intel_fb->pf_planes & 0x2)) & 0x3) %
+				     intel_fb->pf_num_pages;
+      }
+
+      ret = GL_FALSE;
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+   return ret;
+}
+  
+void
+intelSwapBuffers(__DRIdrawablePrivate * dPriv)
+{
+   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+      GET_CURRENT_CONTEXT(ctx);
+      struct intel_context *intel;
+
+      if (ctx == NULL)
+	 return;
+
+      intel = intel_context(ctx);
+
+      if (ctx->Visual.doubleBufferMode) {
+         intelScreenPrivate *screen = intel->intelScreen;
+	 GLboolean missed_target;
+	 struct intel_framebuffer *intel_fb = dPriv->driverPrivate;
+	 int64_t ust;
+         
+	 _mesa_notifySwapBuffers(ctx);  /* flush pending rendering comands */
+
+         if (screen->current_rotation != 0 ||
+	     !intelScheduleSwap(dPriv, &missed_target)) {
+	    driWaitForVBlank(dPriv, &missed_target);
+
+	    if (screen->current_rotation != 0 || !intelPageFlip(dPriv)) {
+	       intelCopyBuffer(dPriv, NULL);
+	    }
+
+	    if (screen->current_rotation != 0) {
+	       intelRotateWindow(intel, dPriv, BUFFER_BIT_FRONT_LEFT);
+	    }
+	 }
+
+	 intel_fb->swap_count++;
+	 (*dri_interface->getUST) (&ust);
+	 if (missed_target) {
+	    intel_fb->swap_missed_count++;
+	    intel_fb->swap_missed_ust = ust - intel_fb->swap_ust;
+	 }
+
+	 intel_fb->swap_ust = ust;
+      }
+   }
+   else {
+      /* XXX this shouldn't be an error but we can't handle it for now */
+      fprintf(stderr, "%s: drawable has no context!\n", __FUNCTION__);
+   }
+}
+#endif
+
+void
+intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h)
+{
+   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+      struct intel_context *intel =
+         (struct intel_context *) dPriv->driContextPriv->driverPrivate;
+      GLcontext *ctx = &intel->ctx;
+
+      if (ctx->Visual.doubleBufferMode) {
+         drm_clip_rect_t rect;
+         rect.x1 = x + dPriv->x;
+         rect.y1 = (dPriv->h - y - h) + dPriv->y;
+         rect.x2 = rect.x1 + w;
+         rect.y2 = rect.y1 + h;
+         _mesa_notifySwapBuffers(ctx);  /* flush pending rendering comands */
+         intelCopyBuffer(dPriv, &rect);
+      }
+   }
+   else {
+      /* XXX this shouldn't be an error but we can't handle it for now */
+      fprintf(stderr, "%s: drawable has no context!\n", __FUNCTION__);
+   }
+}
+
+
+/**
+ * Update the hardware state for drawing into a window or framebuffer object.
+ *
+ * Called by glDrawBuffer, glBindFramebufferEXT, MakeCurrent, and other
+ * places within the driver.
+ *
+ * Basically, this needs to be called any time the current framebuffer
+ * changes, the renderbuffers change, or we need to draw into different
+ * color buffers.
+ */
+void
+intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *colorRegion, *depthRegion = NULL;
+   struct intel_renderbuffer *irbDepth = NULL, *irbStencil = NULL;
+   int front = 0;               /* drawing to front color buffer? */
+
+   if (!fb) {
+      /* this can happen during the initial context initialization */
+      return;
+   }
+
+   /* Do this here, note core Mesa, since this function is called from
+    * many places within the driver.
+    */
+   if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) {
+      /* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */
+      _mesa_update_framebuffer(ctx);
+      /* this updates the DrawBuffer's Width/Height if it's a FBO */
+      _mesa_update_draw_buffer_bounds(ctx);
+   }
+
+   if (fb->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) {
+      /* this may occur when we're called by glBindFrameBuffer() during
+       * the process of someone setting up renderbuffers, etc.
+       */
+      /*_mesa_debug(ctx, "DrawBuffer: incomplete user FBO\n");*/
+      return;
+   }
+
+   if (fb->Name)
+      intel_validate_paired_depth_stencil(ctx, fb);
+
+   /*
+    * How many color buffers are we drawing into?
+    */
+   if (fb->_NumColorDrawBuffers[0] != 1
+#if 0
+       /* XXX FBO temporary - always use software rendering */
+       || 1
+#endif
+      ) {
+      /* writing to 0 or 2 or 4 color buffers */
+      /*_mesa_debug(ctx, "Software rendering\n");*/
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE);
+      front = 1;                /* might not have back color buffer */
+   }
+   else {
+      /* draw to exactly one color buffer */
+      /*_mesa_debug(ctx, "Hardware rendering\n");*/
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE);
+      if (fb->_ColorDrawBufferMask[0] == BUFFER_BIT_FRONT_LEFT) {
+         front = 1;
+      }
+   }
+
+   /*
+    * Get the intel_renderbuffer for the colorbuffer we're drawing into.
+    * And set up cliprects.
+    */
+   if (fb->Name == 0) {
+      /* drawing to window system buffer */
+      if (front) {
+         intelSetFrontClipRects(intel);
+         colorRegion = intel_get_rb_region(fb, BUFFER_FRONT_LEFT);
+      }
+      else {
+         intelSetBackClipRects(intel);
+         colorRegion = intel_get_rb_region(fb, BUFFER_BACK_LEFT);
+      }
+   }
+   else {
+      /* drawing to user-created FBO */
+      struct intel_renderbuffer *irb;
+      intelSetRenderbufferClipRects(intel);
+      irb = intel_renderbuffer(fb->_ColorDrawBuffers[0][0]);
+      colorRegion = (irb && irb->region) ? irb->region : NULL;
+   }
+
+   /* Update culling direction which changes depending on the
+    * orientation of the buffer:
+    */
+   if (ctx->Driver.FrontFace)
+      ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
+   else
+      ctx->NewState |= _NEW_POLYGON;
+
+   if (!colorRegion) {
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE);
+   }
+   else {
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE);
+   }
+
+   /***
+    *** Get depth buffer region and check if we need a software fallback.
+    *** Note that the depth buffer is usually a DEPTH_STENCIL buffer.
+    ***/
+   if (fb->_DepthBuffer && fb->_DepthBuffer->Wrapped) {
+      irbDepth = intel_renderbuffer(fb->_DepthBuffer->Wrapped);
+      if (irbDepth && irbDepth->region) {
+         FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_FALSE);
+         depthRegion = irbDepth->region;
+      }
+      else {
+         FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_TRUE);
+         depthRegion = NULL;
+      }
+   }
+   else {
+      /* not using depth buffer */
+      FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_FALSE);
+      depthRegion = NULL;
+   }
+
+   /***
+    *** Stencil buffer
+    *** This can only be hardware accelerated if we're using a
+    *** combined DEPTH_STENCIL buffer (for now anyway).
+    ***/
+   if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) {
+      irbStencil = intel_renderbuffer(fb->_StencilBuffer->Wrapped);
+      if (irbStencil && irbStencil->region) {
+         ASSERT(irbStencil->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+         FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE);
+         /* need to re-compute stencil hw state */
+         ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
+         if (!depthRegion)
+            depthRegion = irbStencil->region;
+      }
+      else {
+         FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_TRUE);
+      }
+   }
+   else {
+      /* XXX FBO: instead of FALSE, pass ctx->Stencil.Enabled ??? */
+      FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE);
+      /* need to re-compute stencil hw state */
+      ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
+   }
+
+   /*
+    * Update depth test state
+    */
+   if (ctx->Depth.Test && fb->Visual.depthBits > 0) {
+      ctx->Driver.Enable(ctx, GL_DEPTH_TEST, GL_TRUE);
+   }
+   else {
+      ctx->Driver.Enable(ctx, GL_DEPTH_TEST, GL_FALSE);
+   }
+
+   /**
+    ** Release old regions, reference new regions
+    **/
+#if 0                           /* XXX FBO: this seems to be redundant with i915_state_draw_region() */
+   if (intel->draw_region != colorRegion) {
+      intel_region_release(&intel->draw_region);
+      intel_region_reference(&intel->draw_region, colorRegion);
+   }
+   if (intel->intelScreen->depth_region != depthRegion) {
+      intel_region_release(&intel->intelScreen->depth_region);
+      intel_region_reference(&intel->intelScreen->depth_region, depthRegion);
+   }
+#endif
+
+   intel->vtbl.set_draw_region(intel, colorRegion, depthRegion);
+
+   /* update viewport since it depends on window size */
+   ctx->Driver.Viewport(ctx, ctx->Viewport.X, ctx->Viewport.Y,
+                        ctx->Viewport.Width, ctx->Viewport.Height);
+
+   /* Update hardware scissor */
+   ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
+                       ctx->Scissor.Width, ctx->Scissor.Height);
+}
+
+
+static void
+intelDrawBuffer(GLcontext * ctx, GLenum mode)
+{
+   intel_draw_buffer(ctx, ctx->DrawBuffer);
+}
+
+
+static void
+intelReadBuffer(GLcontext * ctx, GLenum mode)
+{
+   if (ctx->ReadBuffer == ctx->DrawBuffer) {
+      /* This will update FBO completeness status.
+       * A framebuffer will be incomplete if the GL_READ_BUFFER setting
+       * refers to a missing renderbuffer.  Calling glReadBuffer can set
+       * that straight and can make the drawing buffer complete.
+       */
+      intel_draw_buffer(ctx, ctx->DrawBuffer);
+   }
+   /* Generally, functions which read pixels (glReadPixels, glCopyPixels, etc)
+    * reference ctx->ReadBuffer and do appropriate state checks.
+    */
+}
+
+
+void
+intelInitBufferFuncs(struct dd_function_table *functions)
+{
+   functions->Clear = intelClear;
+   functions->DrawBuffer = intelDrawBuffer;
+   functions->ReadBuffer = intelReadBuffer;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_buffers.h b/src/mesa/drivers/dri/intel/intel_buffers.h
new file mode 100644
index 00000000000..3b686cb5c18
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_buffers.h
@@ -0,0 +1,59 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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_BUFFERS_H
+#define INTEL_BUFFERS_H
+
+
+struct intel_context;
+struct intel_framebuffer;
+
+
+extern GLboolean
+intel_intersect_cliprects(drm_clip_rect_t * dest,
+                          const drm_clip_rect_t * a,
+                          const drm_clip_rect_t * b);
+
+extern struct intel_region *intel_readbuf_region(struct intel_context *intel);
+
+extern struct intel_region *intel_drawbuf_region(struct intel_context *intel);
+
+extern void intel_wait_flips(struct intel_context *intel, GLuint batch_flags);
+
+extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
+
+extern void intelWindowMoved(struct intel_context *intel);
+
+extern void intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb);
+
+extern void intelInitBufferFuncs(struct dd_function_table *functions);
+
+extern void
+intelRotateWindow(struct intel_context *intel,
+                  __DRIdrawablePrivate * dPriv, GLuint srcBuf);
+
+#endif /* INTEL_BUFFERS_H */
diff --git a/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.c b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.c
new file mode 100644
index 00000000000..32d98860916
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.c
@@ -0,0 +1,833 @@
+/**************************************************************************
+ * 
+ * Copyright © 2007 Red Hat Inc.
+ * Copyright © 2007 Intel Corporation
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * 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 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
+ * THE COPYRIGHT HOLDERS, AUTHORS 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.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ *          Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ *	    Eric Anholt <eric@anholt.net>
+ *	    Dave Airlie <airlied@linux.ie>
+ */
+
+#include <xf86drm.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include "glthread.h"
+#include "errno.h"
+#include "mtypes.h"
+#include "dri_bufmgr.h"
+#include "string.h"
+#include "imports.h"
+
+#include "i915_drm.h"
+
+#include "intel_bufmgr_ttm.h"
+
+#define BUFMGR_DEBUG 0
+
+struct intel_reloc_info
+{
+   GLuint type;
+   GLuint reloc;
+   GLuint delta;                /* not needed? */
+   GLuint index;
+   drm_handle_t handle;
+};
+
+struct intel_bo_node
+{
+    drmMMListHead head;
+    drmBO *buf;
+    struct drm_i915_op_arg bo_arg;
+    unsigned long arg0;
+    unsigned long arg1;
+    void (*destroy)(void *);
+    void *priv;
+};
+
+struct intel_bo_reloc_list
+{
+    drmMMListHead head;
+    drmBO buf;
+    uint32_t *relocs;
+};
+
+struct intel_bo_reloc_node
+{
+    drmMMListHead head;
+    drm_handle_t handle;
+    uint32_t nr_reloc_types;
+    struct intel_bo_reloc_list type_list;
+};
+
+struct intel_bo_list {
+    unsigned numCurrent;
+    drmMMListHead list;
+    void (*destroy)(void *node);
+};
+
+typedef struct _dri_bufmgr_ttm {
+   dri_bufmgr bufmgr;
+
+   int fd;
+   _glthread_Mutex mutex;
+   unsigned int fence_type;
+   unsigned int fence_type_flush;
+
+   uint32_t max_relocs;
+   /** ttm relocation list */
+   struct intel_bo_list list;
+   struct intel_bo_list reloc_list;
+
+} dri_bufmgr_ttm;
+
+typedef struct _dri_bo_ttm {
+   dri_bo bo;
+
+   int refcount;		/* Protected by bufmgr->mutex */
+   drmBO drm_bo;
+   const char *name;
+} dri_bo_ttm;
+
+typedef struct _dri_fence_ttm
+{
+   dri_fence fence;
+
+   int refcount;		/* Protected by bufmgr->mutex */
+   const char *name;
+   drmFence drm_fence;
+} dri_fence_ttm;
+
+
+static void intel_bo_free_list(struct intel_bo_list *list)
+{
+    struct intel_bo_node *node;
+    drmMMListHead *l;
+
+    l = list->list.next;
+    while(l != &list->list) {
+	DRMLISTDEL(l);
+	node = DRMLISTENTRY(struct intel_bo_node, l, head);
+	list->destroy(node);
+	l = list->list.next;
+	list->numCurrent--;
+    }
+}
+
+static void generic_destroy(void *nodep)
+{
+    free(nodep);
+}
+
+static int intel_create_bo_list(int numTarget, struct intel_bo_list *list, void (*destroy)(void *))
+{
+    DRMINITLISTHEAD(&list->list);
+    list->numCurrent = 0;
+    if (destroy)
+        list->destroy = destroy;
+    else
+        list->destroy = generic_destroy;
+    return 0;
+}
+
+
+static struct drm_i915_op_arg *
+intel_setup_validate_list(int fd, struct intel_bo_list *list, struct intel_bo_list *reloc_list, GLuint *count_p)
+{
+    struct intel_bo_node *node;
+    struct intel_bo_reloc_node *rl_node;
+    drmMMListHead *l, *rl;
+    struct drm_i915_op_arg *arg, *first;
+    struct drm_bo_op_req *req;
+    uint64_t *prevNext = NULL;
+    GLuint count = 0;
+
+    first = NULL;
+
+    for (l = list->list.next; l != &list->list; l = l->next) {
+        node = DRMLISTENTRY(struct intel_bo_node, l, head);
+
+        arg = &node->bo_arg;
+        req = &arg->d.req;
+
+        if (!first)
+            first = arg;
+
+	if (prevNext)
+	    *prevNext = (unsigned long) arg;
+
+	memset(arg, 0, sizeof(*arg));
+	prevNext = &arg->next;
+	req->bo_req.handle = node->buf->handle;
+	req->op = drm_bo_validate;
+	req->bo_req.flags = node->arg0;
+	req->bo_req.hint = 0;
+	req->bo_req.mask = node->arg1;
+	req->bo_req.fence_class = 0; /* Backwards compat. */
+	arg->reloc_handle = 0;
+
+	for (rl = reloc_list->list.next; rl != &reloc_list->list; rl = rl->next) {
+	    rl_node = DRMLISTENTRY(struct intel_bo_reloc_node, rl, head);
+
+	    if (rl_node->handle == node->buf->handle) {
+		arg->reloc_handle = rl_node->type_list.buf.handle;
+	    }
+	}
+	count++;
+    }
+
+    if (!first)
+	return 0;
+
+    *count_p = count;
+    return first;
+}
+
+static void intel_free_validate_list(int fd, struct intel_bo_list *list)
+{
+    struct intel_bo_node *node;
+    drmMMListHead *l;
+
+    for (l = list->list.next; l != &list->list; l = l->next) {
+        node = DRMLISTENTRY(struct intel_bo_node, l, head);    
+
+	if (node->destroy)
+	    (*node->destroy)(node->priv);
+
+    }
+}
+
+static void intel_free_reloc_list(int fd, struct intel_bo_list *reloc_list)
+{
+    struct intel_bo_reloc_node *reloc_node;
+    drmMMListHead *rl, *tmp;
+
+    for (rl = reloc_list->list.next, tmp = rl->next; rl != &reloc_list->list; rl = tmp, tmp = rl->next) {
+	reloc_node = DRMLISTENTRY(struct intel_bo_reloc_node, rl, head);
+
+	DRMLISTDEL(rl);
+
+	if (reloc_node->nr_reloc_types > 1) {
+	    
+	    /* TODO */
+	}
+
+	drmBOUnmap(fd, &reloc_node->type_list.buf);
+	drmBOUnreference(fd, &reloc_node->type_list.buf);
+	free(reloc_node);
+    }
+}
+
+static int intel_add_validate_buffer(struct intel_bo_list *list, dri_bo *buf, unsigned flags,
+				     unsigned mask, int *itemLoc, void (*destroy_cb)(void *))
+{
+    struct intel_bo_node *node, *cur;
+    drmMMListHead *l;
+    int count = 0;
+    int ret = 0;
+    drmBO *buf_bo = &((dri_bo_ttm *)buf)->drm_bo;
+    cur = NULL;
+
+    for (l = list->list.next; l != &list->list; l = l->next) {
+	node = DRMLISTENTRY(struct intel_bo_node, l, head);
+	if (node->buf->handle == buf_bo->handle) {
+	    cur = node;
+	    break;
+	}
+	count++;
+    }
+    
+    if (!cur) {
+	cur = drmMalloc(sizeof(*cur));
+	if (!cur) {
+	    return -ENOMEM;
+	}
+	cur->buf = buf_bo;
+	cur->priv = buf;
+	cur->arg0 = flags;
+	cur->arg1 = mask;
+	cur->destroy = destroy_cb;
+	ret = 1;
+
+	DRMLISTADDTAIL(&cur->head, &list->list);
+
+    } else {
+	unsigned memMask = (cur->arg1 | mask) & DRM_BO_MASK_MEM;
+	unsigned memFlags = cur->arg0 & flags & memMask;
+	
+	if (!memFlags) {
+	    return -EINVAL;
+	}
+	if (mask & cur->arg1 & ~DRM_BO_MASK_MEM  & (cur->arg0 ^ flags)) {
+	    return -EINVAL;
+	}
+	cur->arg1 |= mask;
+	cur->arg0 = memFlags | ((cur->arg0 | flags) & 
+				cur->arg1 & ~DRM_BO_MASK_MEM);
+    }
+    *itemLoc = count;
+    return ret;
+}
+
+
+#define RELOC_BUF_SIZE(x) ((I915_RELOC_HEADER + x * I915_RELOC0_STRIDE) * sizeof(uint32_t))
+
+static int intel_create_new_reloc_type_list(int fd, struct intel_bo_reloc_list *cur_type, int max_relocs)
+{
+    int ret;
+    
+    /* should allocate a drmBO here */
+    ret = drmBOCreate(fd, RELOC_BUF_SIZE(max_relocs), 0,
+		      NULL,
+		      DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE | DRM_BO_FLAG_MAPPABLE | DRM_BO_FLAG_CACHED,
+		      0, &cur_type->buf);
+    if (ret)
+	return ret;
+    
+    ret = drmBOMap(fd, &cur_type->buf, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0, (void **)&cur_type->relocs);
+    if (ret)
+	return ret;
+    return 0;
+}
+
+
+static int intel_add_validate_reloc(int fd, struct intel_bo_list *reloc_list, struct intel_reloc_info *reloc_info, uint32_t max_relocs)
+{
+    struct intel_bo_reloc_node *rl_node, *cur;
+    drmMMListHead *rl, *l;
+    int ret = 0;
+    uint32_t *reloc_start;
+    int num_relocs;
+    struct intel_bo_reloc_list *cur_type;
+
+    cur = NULL;
+
+    for (rl = reloc_list->list.next; rl != &reloc_list->list; rl = rl->next) {
+	rl_node = DRMLISTENTRY(struct intel_bo_reloc_node, rl, head);
+	if (rl_node->handle == reloc_info->handle) {
+	    cur = rl_node;
+	    break;
+	}
+    }
+
+    if (!cur) {
+
+	cur = malloc(sizeof(*cur));
+	if (!cur)
+	    return -ENOMEM;
+
+	cur->nr_reloc_types = 1;
+	cur->handle = reloc_info->handle;
+	cur_type = &cur->type_list;
+
+	DRMINITLISTHEAD(&cur->type_list.head);
+	ret = intel_create_new_reloc_type_list(fd, cur_type, max_relocs);
+	if (ret) {
+	    return -1;
+	}
+	DRMLISTADDTAIL(&cur->head, &reloc_list->list);
+
+	cur_type->relocs[0] = 0 | (reloc_info->type << 16);
+	cur_type->relocs[1] = 0; // next reloc buffer handle is 0
+
+    } else {
+	int found = 0;
+	if ((cur->type_list.relocs[0] >> 16) == reloc_info->type) {
+		cur_type = &cur->type_list;
+		found = 1;
+	} else {
+	    for (l = cur->type_list.head.next; l != &cur->type_list.head; l = l->next) {
+	        cur_type = DRMLISTENTRY(struct intel_bo_reloc_list, l, head);
+	        if (((cur_type->relocs[0] >> 16) & 0xffff) == reloc_info->type)
+	    	    found = 1;
+		break;
+	    }
+        }
+	
+	/* didn't find the relocation type */
+	if (!found) {
+	    cur_type = malloc(sizeof(*cur_type));
+	    if (!cur_type) {
+		return -ENOMEM;
+	    }
+	    
+	    ret = intel_create_new_reloc_type_list(fd, cur_type, max_relocs);
+	    DRMLISTADDTAIL(&cur_type->head, &cur->type_list.head);
+
+	    cur_type->relocs[0] = (reloc_info->type << 16);
+	    cur_type->relocs[1] = 0;
+
+	    cur->nr_reloc_types++;
+	}
+    }
+
+    reloc_start = cur_type->relocs;
+
+    num_relocs = (reloc_start[0] & 0xffff);
+
+    reloc_start[num_relocs*I915_RELOC0_STRIDE + I915_RELOC_HEADER] = reloc_info->reloc;
+    reloc_start[num_relocs*I915_RELOC0_STRIDE + I915_RELOC_HEADER+1] = reloc_info->delta;
+    reloc_start[num_relocs*I915_RELOC0_STRIDE + I915_RELOC_HEADER+2] = reloc_info->index;
+    reloc_start[0]++;
+    if (((reloc_start[0] & 0xffff)) > (max_relocs)) {
+	return -ENOMEM;
+    }
+    return 0;
+}
+
+
+#if 0
+int
+driFenceSignaled(DriFenceObject * fence, unsigned type)
+{
+   int signaled;
+   int ret;
+
+   if (fence == NULL)
+      return GL_TRUE;
+
+   _glthread_LOCK_MUTEX(fence->mutex);
+   ret = drmFenceSignaled(bufmgr_ttm->fd, &fence->fence, type, &signaled);
+   _glthread_UNLOCK_MUTEX(fence->mutex);
+   BM_CKFATAL(ret);
+   return signaled;
+}
+#endif
+
+static dri_bo *
+dri_ttm_alloc(dri_bufmgr *bufmgr, const char *name,
+	      unsigned long size, unsigned int alignment,
+	      unsigned int location_mask)
+{
+   dri_bufmgr_ttm *ttm_bufmgr;
+   dri_bo_ttm *ttm_buf;
+   unsigned int pageSize = getpagesize();
+   int ret;
+   unsigned int flags, hint;
+
+   ttm_bufmgr = (dri_bufmgr_ttm *)bufmgr;
+
+   ttm_buf = malloc(sizeof(*ttm_buf));
+   if (!ttm_buf)
+      return NULL;
+
+   /* The mask argument doesn't do anything for us that we want other than
+    * determine which pool (TTM or local) the buffer is allocated into, so just
+    * pass all of the allocation class flags.
+    */
+   flags = location_mask | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE |
+      DRM_BO_FLAG_EXE;
+   /* No hints we want to use. */
+   hint = 0;
+
+   ret = drmBOCreate(ttm_bufmgr->fd, size, alignment / pageSize,
+		     NULL, flags, hint, &ttm_buf->drm_bo);
+   if (ret != 0) {
+      free(ttm_buf);
+      return NULL;
+   }
+   ttm_buf->bo.size = ttm_buf->drm_bo.size;
+   ttm_buf->bo.offset = ttm_buf->drm_bo.offset;
+   ttm_buf->bo.virtual = NULL;
+   ttm_buf->bo.bufmgr = bufmgr;
+   ttm_buf->name = name;
+   ttm_buf->refcount = 1;
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "bo_create: %p (%s)\n", &ttm_buf->bo, ttm_buf->name);
+#endif
+
+   return &ttm_buf->bo;
+}
+
+/* Our TTM backend doesn't allow creation of static buffers, as that requires
+ * privelege for the non-fake case, and the lock in the fake case where we were
+ * working around the X Server not creating buffers and passing handles to us.
+ */
+static dri_bo *
+dri_ttm_alloc_static(dri_bufmgr *bufmgr, const char *name,
+		     unsigned long offset, unsigned long size, void *virtual,
+		     unsigned int location_mask)
+{
+   return NULL;
+}
+
+/** Returns a dri_bo wrapping the given buffer object handle.
+ *
+ * This can be used when one application needs to pass a buffer object
+ * to another.
+ */
+dri_bo *
+intel_ttm_bo_create_from_handle(dri_bufmgr *bufmgr, const char *name,
+			      unsigned int handle)
+{
+   dri_bufmgr_ttm *ttm_bufmgr;
+   dri_bo_ttm *ttm_buf;
+   int ret;
+
+   ttm_bufmgr = (dri_bufmgr_ttm *)bufmgr;
+
+   ttm_buf = malloc(sizeof(*ttm_buf));
+   if (!ttm_buf)
+      return NULL;
+
+   ret = drmBOReference(ttm_bufmgr->fd, handle, &ttm_buf->drm_bo);
+   if (ret != 0) {
+      free(ttm_buf);
+      return NULL;
+   }
+   ttm_buf->bo.size = ttm_buf->drm_bo.size;
+   ttm_buf->bo.offset = ttm_buf->drm_bo.offset;
+   ttm_buf->bo.virtual = NULL;
+   ttm_buf->bo.bufmgr = bufmgr;
+   ttm_buf->name = name;
+   ttm_buf->refcount = 1;
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "bo_create_from_handle: %p %08x (%s)\n", &ttm_buf->bo, handle,
+	   ttm_buf->name);
+#endif
+
+   return &ttm_buf->bo;
+}
+
+static void
+dri_ttm_bo_reference(dri_bo *buf)
+{
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+   dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+
+   _glthread_LOCK_MUTEX(bufmgr_ttm->mutex);
+   ttm_buf->refcount++;
+   _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+}
+
+static void
+dri_ttm_bo_unreference(dri_bo *buf)
+{
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+   dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+
+   if (!buf)
+      return;
+
+   _glthread_LOCK_MUTEX(bufmgr_ttm->mutex);
+   if (--ttm_buf->refcount == 0) {
+      int ret;
+
+      ret = drmBOUnreference(bufmgr_ttm->fd, &ttm_buf->drm_bo);
+      if (ret != 0) {
+	 fprintf(stderr, "drmBOUnreference failed (%s): %s\n", ttm_buf->name,
+		 strerror(-ret));
+      }
+#if BUFMGR_DEBUG
+      fprintf(stderr, "bo_unreference final: %p (%s)\n",
+	      &ttm_buf->bo, ttm_buf->name);
+#endif
+      _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+      free(buf);
+      return;
+   }
+   _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+}
+
+static int
+dri_ttm_bo_map(dri_bo *buf, GLboolean write_enable)
+{
+   dri_bufmgr_ttm *bufmgr_ttm;
+   dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+   unsigned int flags;
+
+   bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+
+   flags = DRM_BO_FLAG_READ;
+   if (write_enable)
+       flags |= DRM_BO_FLAG_WRITE;
+
+   assert(buf->virtual == NULL);
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "bo_map: %p (%s)\n", &ttm_buf->bo, ttm_buf->name);
+#endif
+
+   return drmBOMap(bufmgr_ttm->fd, &ttm_buf->drm_bo, flags, 0, &buf->virtual);
+}
+
+static int
+dri_ttm_bo_unmap(dri_bo *buf)
+{
+   dri_bufmgr_ttm *bufmgr_ttm;
+   dri_bo_ttm *ttm_buf = (dri_bo_ttm *)buf;
+
+   if (buf == NULL)
+      return 0;
+
+   bufmgr_ttm = (dri_bufmgr_ttm *)buf->bufmgr;
+
+   assert(buf->virtual != NULL);
+
+   buf->virtual = NULL;
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "bo_unmap: %p (%s)\n", &ttm_buf->bo, ttm_buf->name);
+#endif
+
+   return drmBOUnmap(bufmgr_ttm->fd, &ttm_buf->drm_bo);
+}
+
+/* Returns a dri_bo wrapping the given buffer object handle.
+ *
+ * This can be used when one application needs to pass a buffer object
+ * to another.
+ */
+dri_fence *
+intel_ttm_fence_create_from_arg(dri_bufmgr *bufmgr, const char *name,
+				drm_fence_arg_t *arg)
+{
+   dri_bufmgr_ttm *ttm_bufmgr;
+   dri_fence_ttm *ttm_fence;
+
+   ttm_bufmgr = (dri_bufmgr_ttm *)bufmgr;
+
+   ttm_fence = malloc(sizeof(*ttm_fence));
+   if (!ttm_fence)
+      return NULL;
+
+   ttm_fence->drm_fence.handle = arg->handle;
+   ttm_fence->drm_fence.fence_class = arg->fence_class;
+   ttm_fence->drm_fence.type = arg->type;
+   ttm_fence->drm_fence.flags = arg->flags;
+   ttm_fence->drm_fence.signaled = 0;
+   ttm_fence->drm_fence.sequence = arg->sequence;
+
+   ttm_fence->fence.bufmgr = bufmgr;
+   ttm_fence->name = name;
+   ttm_fence->refcount = 1;
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "fence_create_from_handle: %p (%s)\n", &ttm_fence->fence,
+	   ttm_fence->name);
+#endif
+
+   return &ttm_fence->fence;
+}
+
+
+static void
+dri_ttm_fence_reference(dri_fence *fence)
+{
+   dri_fence_ttm *fence_ttm = (dri_fence_ttm *)fence;
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)fence->bufmgr;
+
+   _glthread_LOCK_MUTEX(bufmgr_ttm->mutex);
+   ++fence_ttm->refcount;
+   _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+#if BUFMGR_DEBUG
+   fprintf(stderr, "fence_reference: %p (%s)\n", &fence_ttm->fence,
+	   fence_ttm->name);
+#endif
+}
+
+static void
+dri_ttm_fence_unreference(dri_fence *fence)
+{
+   dri_fence_ttm *fence_ttm = (dri_fence_ttm *)fence;
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)fence->bufmgr;
+
+   if (!fence)
+      return;
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "fence_unreference: %p (%s)\n", &fence_ttm->fence,
+	   fence_ttm->name);
+#endif
+   _glthread_LOCK_MUTEX(bufmgr_ttm->mutex);
+   if (--fence_ttm->refcount == 0) {
+      int ret;
+
+      ret = drmFenceUnreference(bufmgr_ttm->fd, &fence_ttm->drm_fence);
+      if (ret != 0) {
+	 fprintf(stderr, "drmFenceUnreference failed (%s): %s\n",
+		 fence_ttm->name, strerror(-ret));
+      }
+
+      _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+      free(fence);
+      return;
+   }
+   _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+}
+
+static void
+dri_ttm_fence_wait(dri_fence *fence)
+{
+   dri_fence_ttm *fence_ttm = (dri_fence_ttm *)fence;
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)fence->bufmgr;
+   int ret;
+
+   _glthread_LOCK_MUTEX(bufmgr_ttm->mutex);
+   ret = drmFenceWait(bufmgr_ttm->fd, 0, &fence_ttm->drm_fence, 0);
+   _glthread_UNLOCK_MUTEX(bufmgr_ttm->mutex);
+   if (ret != 0) {
+      _mesa_printf("%s:%d: Error %d waiting for fence %s.\n",
+		   __FILE__, __LINE__, ret, fence_ttm->name);
+      abort();
+   }
+
+#if BUFMGR_DEBUG
+   fprintf(stderr, "fence_wait: %p (%s)\n", &fence_ttm->fence,
+	   fence_ttm->name);
+#endif
+}
+
+static void
+dri_bufmgr_ttm_destroy(dri_bufmgr *bufmgr)
+{
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)bufmgr;
+
+   intel_bo_free_list(&bufmgr_ttm->list);
+   intel_bo_free_list(&bufmgr_ttm->reloc_list);
+
+   _glthread_DESTROY_MUTEX(bufmgr_ttm->mutex);
+   free(bufmgr);
+}
+
+
+static void intel_dribo_destroy_callback(void *priv)
+{
+   dri_bo *dribo = priv;
+   
+   if (dribo) {
+     dri_bo_unreference(dribo);
+   }
+}
+
+static void
+dri_ttm_emit_reloc(dri_bo *batch_buf, GLuint flags, GLuint delta, GLuint offset,
+		    dri_bo *relocatee)
+{
+   dri_bo_ttm *ttm_buf = (dri_bo_ttm *)batch_buf;
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)batch_buf->bufmgr;
+   int newItem;
+   struct intel_reloc_info reloc;
+   int mask;
+   int ret;
+
+   mask = DRM_BO_MASK_MEM;
+   mask |= flags & (DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE | DRM_BO_FLAG_EXE);
+
+   ret = intel_add_validate_buffer(&bufmgr_ttm->list, relocatee, flags, mask, &newItem, intel_dribo_destroy_callback);
+   if (ret < 0)
+     return;
+
+   if (ret == 1) {
+      dri_bo_reference(relocatee);
+   }
+
+   reloc.type = I915_RELOC_TYPE_0;
+   reloc.reloc = offset;
+   reloc.delta = delta;
+   reloc.index = newItem;
+   reloc.handle = ttm_buf->drm_bo.handle;
+
+   intel_add_validate_reloc(bufmgr_ttm->fd, &bufmgr_ttm->reloc_list, &reloc, bufmgr_ttm->max_relocs);
+   return;
+}
+
+
+static void *
+dri_ttm_process_reloc(dri_bo *batch_buf, GLuint *count)
+{
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)batch_buf->bufmgr;
+   void *ptr;
+   int itemLoc;
+
+   dri_bo_unmap(batch_buf);
+
+   intel_add_validate_buffer(&bufmgr_ttm->list, batch_buf, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE,
+			     DRM_BO_MASK_MEM | DRM_BO_FLAG_EXE, &itemLoc, NULL);
+
+   ptr = intel_setup_validate_list(bufmgr_ttm->fd, &bufmgr_ttm->list, &bufmgr_ttm->reloc_list, count);
+
+   return ptr;
+}
+
+static void
+dri_ttm_post_submit(dri_bo *batch_buf, dri_fence **last_fence)
+{
+   dri_bufmgr_ttm *bufmgr_ttm = (dri_bufmgr_ttm *)batch_buf->bufmgr;
+
+   intel_free_validate_list(bufmgr_ttm->fd, &bufmgr_ttm->list);
+   intel_free_reloc_list(bufmgr_ttm->fd, &bufmgr_ttm->reloc_list);
+
+   intel_bo_free_list(&bufmgr_ttm->list);
+}
+
+/**
+ * Initializes the TTM buffer manager, which uses the kernel to allocate, map,
+ * and manage map buffer objections.
+ *
+ * \param fd File descriptor of the opened DRM device.
+ * \param fence_type Driver-specific fence type used for fences with no flush.
+ * \param fence_type_flush Driver-specific fence type used for fences with a
+ *	  flush.
+ */
+dri_bufmgr *
+intel_bufmgr_ttm_init(int fd, unsigned int fence_type,
+		      unsigned int fence_type_flush, int batch_size)
+{
+   dri_bufmgr_ttm *bufmgr_ttm;
+
+   bufmgr_ttm = malloc(sizeof(*bufmgr_ttm));
+   bufmgr_ttm->fd = fd;
+   bufmgr_ttm->fence_type = fence_type;
+   bufmgr_ttm->fence_type_flush = fence_type_flush;
+   _glthread_INIT_MUTEX(bufmgr_ttm->mutex);
+
+   /* lets go with one relocation per every four dwords - purely heuristic */
+   bufmgr_ttm->max_relocs = batch_size / sizeof(uint32_t) / 4;
+
+   intel_create_bo_list(10, &bufmgr_ttm->list, NULL);
+   intel_create_bo_list(1, &bufmgr_ttm->reloc_list, NULL);
+
+   bufmgr_ttm->bufmgr.bo_alloc = dri_ttm_alloc;
+   bufmgr_ttm->bufmgr.bo_alloc_static = dri_ttm_alloc_static;
+   bufmgr_ttm->bufmgr.bo_reference = dri_ttm_bo_reference;
+   bufmgr_ttm->bufmgr.bo_unreference = dri_ttm_bo_unreference;
+   bufmgr_ttm->bufmgr.bo_map = dri_ttm_bo_map;
+   bufmgr_ttm->bufmgr.bo_unmap = dri_ttm_bo_unmap;
+   bufmgr_ttm->bufmgr.fence_reference = dri_ttm_fence_reference;
+   bufmgr_ttm->bufmgr.fence_unreference = dri_ttm_fence_unreference;
+   bufmgr_ttm->bufmgr.fence_wait = dri_ttm_fence_wait;
+   bufmgr_ttm->bufmgr.destroy = dri_bufmgr_ttm_destroy;
+   bufmgr_ttm->bufmgr.emit_reloc = dri_ttm_emit_reloc;
+   bufmgr_ttm->bufmgr.process_relocs = dri_ttm_process_reloc;
+   bufmgr_ttm->bufmgr.post_submit = dri_ttm_post_submit;
+   return &bufmgr_ttm->bufmgr;
+}
+
diff --git a/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.h b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.h
new file mode 100644
index 00000000000..0738839cefb
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_bufmgr_ttm.h
@@ -0,0 +1,17 @@
+
+#ifndef INTEL_BUFMGR_TTM_H
+#define INTEL_BUFMGR_TTM_H
+
+#include "dri_bufmgr.h"
+
+extern dri_bo *intel_ttm_bo_create_from_handle(dri_bufmgr *bufmgr, const char *name,
+					       unsigned int handle);
+
+dri_fence *intel_ttm_fence_create_from_arg(dri_bufmgr *bufmgr, const char *name,
+					   drm_fence_arg_t *arg);
+
+
+dri_bufmgr *intel_bufmgr_ttm_init(int fd, unsigned int fence_type,
+				  unsigned int fence_type_flush, int batch_size);
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_depthstencil.c b/src/mesa/drivers/dri/intel/intel_depthstencil.c
new file mode 100644
index 00000000000..d269a85a3c9
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_depthstencil.c
@@ -0,0 +1,282 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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 "imports.h"
+#include "context.h"
+#include "depthstencil.h"
+#include "fbobject.h"
+#include "framebuffer.h"
+#include "hash.h"
+#include "mtypes.h"
+#include "renderbuffer.h"
+
+#include "intel_context.h"
+#include "intel_fbo.h"
+#include "intel_depthstencil.h"
+#include "intel_regions.h"
+
+
+/**
+ * The GL_EXT_framebuffer_object allows the user to create their own
+ * framebuffer objects consisting of color renderbuffers (0 or more),
+ * depth renderbuffers (0 or 1) and stencil renderbuffers (0 or 1).
+ *
+ * The spec considers depth and stencil renderbuffers to be totally independent
+ * buffers.  In reality, most graphics hardware today uses a combined
+ * depth+stencil buffer (one 32-bit pixel = 24 bits of Z + 8 bits of stencil).
+ *
+ * This causes difficulty because the user may create some number of depth
+ * renderbuffers and some number of stencil renderbuffers and bind them
+ * together in framebuffers in any combination.
+ *
+ * This code manages all that.
+ *
+ * 1. Depth renderbuffers are always allocated in hardware as 32bpp
+ *    GL_DEPTH24_STENCIL8 buffers.
+ *
+ * 2. Stencil renderbuffers are initially allocated in software as 8bpp
+ *    GL_STENCIL_INDEX8 buffers.
+ *
+ * 3. Depth and Stencil renderbuffers use the PairedStencil and PairedDepth
+ *    fields (respectively) to indicate if the buffer's currently paired
+ *    with another stencil or depth buffer (respectively).
+ *
+ * 4. When a depth and stencil buffer are initially both attached to the
+ *    current framebuffer, we merge the stencil buffer values into the
+ *    depth buffer (really a depth+stencil buffer).  The then hardware uses
+ *    the combined buffer.
+ *
+ * 5. Whenever a depth or stencil buffer is reallocated (with
+ *    glRenderbufferStorage) we undo the pairing and copy the stencil values
+ *    from the combined depth/stencil buffer back to the stencil-only buffer.
+ *
+ * 6. We also undo the pairing when we find a change in buffer bindings.
+ *
+ * 7. If a framebuffer is only using a depth renderbuffer (no stencil), we
+ *    just use the combined depth/stencil buffer and ignore the stencil values.
+ *
+ * 8. If a framebuffer is only using a stencil renderbuffer (no depth) we have
+ *    to promote the 8bpp software stencil buffer to a 32bpp hardware
+ *    depth+stencil buffer.
+ *
+ */
+
+
+
+static void
+map_regions(GLcontext * ctx,
+            struct intel_renderbuffer *depthRb,
+            struct intel_renderbuffer *stencilRb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   if (depthRb && depthRb->region) {
+      intel_region_map(intel->intelScreen, depthRb->region);
+      depthRb->pfMap = depthRb->region->map;
+      depthRb->pfPitch = depthRb->region->pitch;
+   }
+   if (stencilRb && stencilRb->region) {
+      intel_region_map(intel->intelScreen, stencilRb->region);
+      stencilRb->pfMap = stencilRb->region->map;
+      stencilRb->pfPitch = stencilRb->region->pitch;
+   }
+}
+
+static void
+unmap_regions(GLcontext * ctx,
+              struct intel_renderbuffer *depthRb,
+              struct intel_renderbuffer *stencilRb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   if (depthRb && depthRb->region) {
+      intel_region_unmap(intel->intelScreen, depthRb->region);
+      depthRb->pfMap = NULL;
+      depthRb->pfPitch = 0;
+   }
+   if (stencilRb && stencilRb->region) {
+      intel_region_unmap(intel->intelScreen, stencilRb->region);
+      stencilRb->pfMap = NULL;
+      stencilRb->pfPitch = 0;
+   }
+}
+
+
+
+/**
+ * Undo the pairing/interleaving between depth and stencil buffers.
+ * irb should be a depth/stencil or stencil renderbuffer.
+ */
+void
+intel_unpair_depth_stencil(GLcontext * ctx, struct intel_renderbuffer *irb)
+{
+   if (irb->PairedStencil) {
+      /* irb is a depth/stencil buffer */
+      struct gl_renderbuffer *stencilRb;
+      struct intel_renderbuffer *stencilIrb;
+
+      ASSERT(irb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+
+      stencilRb = _mesa_lookup_renderbuffer(ctx, irb->PairedStencil);
+      stencilIrb = intel_renderbuffer(stencilRb);
+      if (stencilIrb) {
+         /* need to extract stencil values from the depth buffer */
+         ASSERT(stencilIrb->PairedDepth == irb->Base.Name);
+         map_regions(ctx, irb, stencilIrb);
+         _mesa_extract_stencil(ctx, &irb->Base, &stencilIrb->Base);
+         unmap_regions(ctx, irb, stencilIrb);
+         stencilIrb->PairedDepth = 0;
+      }
+      irb->PairedStencil = 0;
+   }
+   else if (irb->PairedDepth) {
+      /* irb is a stencil buffer */
+      struct gl_renderbuffer *depthRb;
+      struct intel_renderbuffer *depthIrb;
+
+      ASSERT(irb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT ||
+             irb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+
+      depthRb = _mesa_lookup_renderbuffer(ctx, irb->PairedDepth);
+      depthIrb = intel_renderbuffer(depthRb);
+      if (depthIrb) {
+         /* need to extract stencil values from the depth buffer */
+         ASSERT(depthIrb->PairedStencil == irb->Base.Name);
+         map_regions(ctx, depthIrb, irb);
+         _mesa_extract_stencil(ctx, &depthIrb->Base, &irb->Base);
+         unmap_regions(ctx, depthIrb, irb);
+         depthIrb->PairedStencil = 0;
+      }
+      irb->PairedDepth = 0;
+   }
+   else {
+      _mesa_problem(ctx, "Problem in undo_depth_stencil_pairing");
+   }
+
+   ASSERT(irb->PairedStencil == 0);
+   ASSERT(irb->PairedDepth == 0);
+}
+
+
+/**
+ * Examine the depth and stencil renderbuffers which are attached to the
+ * framebuffer.  If both depth and stencil are attached, make sure that the
+ * renderbuffers are 'paired' (combined).  If only depth or only stencil is
+ * attached, undo any previous pairing.
+ *
+ * Must be called if NewState & _NEW_BUFFER (when renderbuffer attachments
+ * change, for example).
+ */
+void
+intel_validate_paired_depth_stencil(GLcontext * ctx,
+                                    struct gl_framebuffer *fb)
+{
+   struct intel_renderbuffer *depthRb, *stencilRb;
+
+   depthRb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
+   stencilRb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
+
+   if (depthRb && stencilRb) {
+      if (depthRb == stencilRb) {
+         /* Using a user-created combined depth/stencil buffer.
+          * Nothing to do.
+          */
+         ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_STENCIL_EXT);
+         ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+      }
+      else {
+         /* Separate depth/stencil buffers, need to interleave now */
+         ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_COMPONENT);
+         ASSERT(stencilRb->Base._BaseFormat == GL_STENCIL_INDEX);
+         /* may need to interleave depth/stencil now */
+         if (depthRb->PairedStencil == stencilRb->Base.Name) {
+            /* OK, the depth and stencil buffers are already interleaved */
+            ASSERT(stencilRb->PairedDepth == depthRb->Base.Name);
+         }
+         else {
+            /* need to setup new pairing/interleaving */
+            if (depthRb->PairedStencil) {
+               intel_unpair_depth_stencil(ctx, depthRb);
+            }
+            if (stencilRb->PairedDepth) {
+               intel_unpair_depth_stencil(ctx, stencilRb);
+            }
+
+            ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+            ASSERT(stencilRb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT ||
+                   stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+
+            /* establish new pairing: interleave stencil into depth buffer */
+            map_regions(ctx, depthRb, stencilRb);
+            _mesa_insert_stencil(ctx, &depthRb->Base, &stencilRb->Base);
+            unmap_regions(ctx, depthRb, stencilRb);
+            depthRb->PairedStencil = stencilRb->Base.Name;
+            stencilRb->PairedDepth = depthRb->Base.Name;
+         }
+
+      }
+   }
+   else if (depthRb) {
+      /* Depth buffer but no stencil buffer.
+       * We'll use a GL_DEPTH24_STENCIL8 buffer and ignore the stencil bits.
+       */
+      /* can't assert this until storage is allocated:
+         ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+       */
+      /* intel_undo any previous pairing */
+      if (depthRb->PairedStencil) {
+         intel_unpair_depth_stencil(ctx, depthRb);
+      }
+   }
+   else if (stencilRb) {
+      /* Stencil buffer but no depth buffer.
+       * Since h/w doesn't typically support just 8bpp stencil w/out Z,
+       * we'll use a GL_DEPTH24_STENCIL8 buffer and ignore the depth bits.
+       */
+      /* undo any previous pairing */
+      if (stencilRb->PairedDepth) {
+         intel_unpair_depth_stencil(ctx, stencilRb);
+      }
+      if (stencilRb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT) {
+         /* promote buffer to GL_DEPTH24_STENCIL8 for hw rendering */
+         _mesa_promote_stencil(ctx, &stencilRb->Base);
+         ASSERT(stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+      }
+   }
+
+   /* Finally, update the fb->_DepthBuffer and fb->_StencilBuffer fields */
+   _mesa_update_depth_buffer(ctx, fb, BUFFER_DEPTH);
+   if (depthRb && depthRb->PairedStencil)
+      _mesa_update_stencil_buffer(ctx, fb, BUFFER_DEPTH);
+   else
+      _mesa_update_stencil_buffer(ctx, fb, BUFFER_STENCIL);
+
+
+   /* The hardware should use fb->Attachment[BUFFER_DEPTH].Renderbuffer
+    * first, if present, then fb->Attachment[BUFFER_STENCIL].Renderbuffer
+    * if present.
+    */
+}
diff --git a/src/mesa/drivers/dri/intel/intel_depthstencil.h b/src/mesa/drivers/dri/intel/intel_depthstencil.h
new file mode 100644
index 00000000000..2d3fc48b3a3
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_depthstencil.h
@@ -0,0 +1,14 @@
+
+#ifndef INTEL_DEPTH_STENCIL_H
+#define INTEL_DEPTH_STENCIL_H
+
+
+extern void
+intel_unpair_depth_stencil(GLcontext * ctx, struct intel_renderbuffer *irb);
+
+extern void
+intel_validate_paired_depth_stencil(GLcontext * ctx,
+                                    struct gl_framebuffer *fb);
+
+
+#endif /* INTEL_DEPTH_STENCIL_H */
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.c b/src/mesa/drivers/dri/intel/intel_fbo.c
new file mode 100644
index 00000000000..6f99f401c7c
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_fbo.c
@@ -0,0 +1,687 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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 "imports.h"
+#include "mtypes.h"
+#include "fbobject.h"
+#include "framebuffer.h"
+#include "renderbuffer.h"
+#include "context.h"
+#include "texformat.h"
+#include "texrender.h"
+
+#include "intel_context.h"
+#include "intel_buffers.h"
+#include "intel_depthstencil.h"
+#include "intel_fbo.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
+#include "intel_span.h"
+
+
+#define FILE_DEBUG_FLAG DEBUG_FBO
+
+#define INTEL_RB_CLASS 0x12345678
+
+
+/* XXX FBO: move this to intel_context.h (inlined) */
+/**
+ * Return a gl_renderbuffer ptr casted to intel_renderbuffer.
+ * NULL will be returned if the rb isn't really an intel_renderbuffer.
+ * This is determiend by checking the ClassID.
+ */
+struct intel_renderbuffer *
+intel_renderbuffer(struct gl_renderbuffer *rb)
+{
+   struct intel_renderbuffer *irb = (struct intel_renderbuffer *) rb;
+   if (irb && irb->Base.ClassID == INTEL_RB_CLASS) {
+      /*_mesa_warning(NULL, "Returning non-intel Rb\n");*/
+      return irb;
+   }
+   else
+      return NULL;
+}
+
+
+struct intel_renderbuffer *
+intel_get_renderbuffer(struct gl_framebuffer *fb, GLuint attIndex)
+{
+   return intel_renderbuffer(fb->Attachment[attIndex].Renderbuffer);
+}
+
+
+void
+intel_flip_renderbuffers(struct intel_framebuffer *intel_fb)
+{
+   int current_page = intel_fb->pf_current_page;
+   int next_page = (current_page + 1) % intel_fb->pf_num_pages;
+   struct gl_renderbuffer *tmp_rb;
+
+   /* Exchange renderbuffers if necessary but make sure their reference counts
+    * are preserved.
+    */
+   if (intel_fb->color_rb[current_page] &&
+       intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer !=
+       &intel_fb->color_rb[current_page]->Base) {
+      tmp_rb = NULL;
+      _mesa_reference_renderbuffer(&tmp_rb,
+	 intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer);
+      tmp_rb = &intel_fb->color_rb[current_page]->Base;
+      _mesa_reference_renderbuffer(
+	 &intel_fb->Base.Attachment[BUFFER_FRONT_LEFT].Renderbuffer, tmp_rb);
+      _mesa_reference_renderbuffer(&tmp_rb, NULL);
+   }
+
+   if (intel_fb->color_rb[next_page] &&
+       intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer !=
+       &intel_fb->color_rb[next_page]->Base) {
+      tmp_rb = NULL;
+      _mesa_reference_renderbuffer(&tmp_rb,
+	 intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer);
+      tmp_rb = &intel_fb->color_rb[next_page]->Base;
+      _mesa_reference_renderbuffer(
+	 &intel_fb->Base.Attachment[BUFFER_BACK_LEFT].Renderbuffer, tmp_rb);
+      _mesa_reference_renderbuffer(&tmp_rb, NULL);
+   }
+}
+
+
+struct intel_region *
+intel_get_rb_region(struct gl_framebuffer *fb, GLuint attIndex)
+{
+   struct intel_renderbuffer *irb = intel_get_renderbuffer(fb, attIndex);
+
+   if (irb)
+      return irb->region;
+   else
+      return NULL;
+}
+
+
+
+/**
+ * Create a new framebuffer object.
+ */
+static struct gl_framebuffer *
+intel_new_framebuffer(GLcontext * ctx, GLuint name)
+{
+   /* Only drawable state in intel_framebuffer at this time, just use Mesa's
+    * class
+    */
+   return _mesa_new_framebuffer(ctx, name);
+}
+
+
+static void
+intel_delete_renderbuffer(struct gl_renderbuffer *rb)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);
+
+   ASSERT(irb);
+
+   if (irb->PairedStencil || irb->PairedDepth) {
+      intel_unpair_depth_stencil(ctx, irb);
+   }
+
+   if (intel && irb->region) {
+      intel_region_release(&irb->region);
+   }
+
+   _mesa_free(irb);
+}
+
+
+
+/**
+ * Return a pointer to a specific pixel in a renderbuffer.
+ */
+static void *
+intel_get_pointer(GLcontext * ctx, struct gl_renderbuffer *rb,
+                  GLint x, GLint y)
+{
+   /* By returning NULL we force all software rendering to go through
+    * the span routines.
+    */
+   return NULL;
+}
+
+
+
+/**
+ * Called via glRenderbufferStorageEXT() to set the format and allocate
+ * storage for a user-created renderbuffer.
+ */
+static GLboolean
+intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                                 GLenum internalFormat,
+                                 GLuint width, GLuint height)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);
+   GLboolean softwareBuffer = GL_FALSE;
+   int cpp;
+
+   ASSERT(rb->Name != 0);
+
+   switch (internalFormat) {
+   case GL_R3_G3_B2:
+   case GL_RGB4:
+   case GL_RGB5:
+      rb->_ActualFormat = GL_RGB5;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 5;
+      rb->GreenBits = 6;
+      rb->BlueBits = 5;
+      cpp = 2;
+      break;
+   case GL_RGB:
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+   case GL_RGBA:
+   case GL_RGBA2:
+   case GL_RGBA4:
+   case GL_RGB5_A1:
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      rb->_ActualFormat = GL_RGBA8;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 8;
+      rb->GreenBits = 8;
+      rb->BlueBits = 8;
+      rb->AlphaBits = 8;
+      cpp = 4;
+      break;
+   case GL_STENCIL_INDEX:
+   case GL_STENCIL_INDEX1_EXT:
+   case GL_STENCIL_INDEX4_EXT:
+   case GL_STENCIL_INDEX8_EXT:
+   case GL_STENCIL_INDEX16_EXT:
+      /* alloc a depth+stencil buffer */
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->StencilBits = 8;
+      cpp = 4;
+      break;
+   case GL_DEPTH_COMPONENT16:
+      rb->_ActualFormat = GL_DEPTH_COMPONENT16;
+      rb->DataType = GL_UNSIGNED_SHORT;
+      rb->DepthBits = 16;
+      cpp = 2;
+      break;
+   case GL_DEPTH_COMPONENT:
+   case GL_DEPTH_COMPONENT24:
+   case GL_DEPTH_COMPONENT32:
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->DepthBits = 24;
+      cpp = 4;
+      break;
+   case GL_DEPTH_STENCIL_EXT:
+   case GL_DEPTH24_STENCIL8_EXT:
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->DepthBits = 24;
+      rb->StencilBits = 8;
+      cpp = 4;
+      break;
+   default:
+      _mesa_problem(ctx,
+                    "Unexpected format in intel_alloc_renderbuffer_storage");
+      return GL_FALSE;
+   }
+
+   intelFlush(ctx);
+
+   /* free old region */
+   if (irb->region) {
+      intel_region_release(&irb->region);
+   }
+
+   /* allocate new memory region/renderbuffer */
+   if (softwareBuffer) {
+      return _mesa_soft_renderbuffer_storage(ctx, rb, internalFormat,
+                                             width, height);
+   }
+   else {
+      /* Choose a pitch to match hardware requirements:
+       */
+      GLuint pitch = ((cpp * width + 63) & ~63) / cpp;
+
+      /* alloc hardware renderbuffer */
+      DBG("Allocating %d x %d Intel RBO (pitch %d)\n", width,
+	  height, pitch);
+
+      irb->region = intel_region_alloc(intel->intelScreen, cpp, pitch, height);
+      if (!irb->region)
+         return GL_FALSE;       /* out of memory? */
+
+      ASSERT(irb->region->buffer);
+
+      rb->Width = width;
+      rb->Height = height;
+
+      /* This sets the Get/PutRow/Value functions */
+      intel_set_span_functions(&irb->Base);
+
+      return GL_TRUE;
+   }
+}
+
+
+
+/**
+ * Called for each hardware renderbuffer when a _window_ is resized.
+ * Just update fields.
+ * Not used for user-created renderbuffers!
+ */
+static GLboolean
+intel_alloc_window_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                           GLenum internalFormat, GLuint width, GLuint height)
+{
+   ASSERT(rb->Name == 0);
+   rb->Width = width;
+   rb->Height = height;
+   rb->_ActualFormat = internalFormat;
+
+   return GL_TRUE;
+}
+
+static void
+intel_resize_buffers(GLcontext *ctx, struct gl_framebuffer *fb,
+		     GLuint width, GLuint height)
+{
+   struct intel_framebuffer *intel_fb = (struct intel_framebuffer*)fb;
+   int i;
+
+   _mesa_resize_framebuffer(ctx, fb, width, height);
+
+   fb->Initialized = GL_TRUE; /* XXX remove someday */
+
+   if (fb->Name != 0) {
+      return;
+   }
+
+   /* Make sure all window system renderbuffers are up to date */
+   for (i = 0; i < 3; i++) {
+      struct gl_renderbuffer *rb = &intel_fb->color_rb[i]->Base;
+
+      /* only resize if size is changing */
+      if (rb && (rb->Width != width || rb->Height != height)) {
+	 rb->AllocStorage(ctx, rb, rb->InternalFormat, width, height);
+      }
+   }
+}
+
+static GLboolean
+intel_nop_alloc_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                        GLenum internalFormat, GLuint width, GLuint height)
+{
+   _mesa_problem(ctx, "intel_op_alloc_storage should never be called.");
+   return GL_FALSE;
+}
+
+
+
+/**
+ * Create a new intel_renderbuffer which corresponds to an on-screen window,
+ * not a user-created renderbuffer.
+ * \param width  the screen width
+ * \param height  the screen height
+ */
+struct intel_renderbuffer *
+intel_create_renderbuffer(GLenum intFormat, GLsizei width, GLsizei height,
+                          int offset, int pitch, int cpp, void *map)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   struct intel_renderbuffer *irb;
+   const GLuint name = 0;
+
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&irb->Base, name);
+   irb->Base.ClassID = INTEL_RB_CLASS;
+
+   switch (intFormat) {
+   case GL_RGB5:
+      irb->Base._ActualFormat = GL_RGB5;
+      irb->Base._BaseFormat = GL_RGBA;
+      irb->Base.RedBits = 5;
+      irb->Base.GreenBits = 6;
+      irb->Base.BlueBits = 5;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
+      cpp = 2;
+      break;
+   case GL_RGBA8:
+      irb->Base._ActualFormat = GL_RGBA8;
+      irb->Base._BaseFormat = GL_RGBA;
+      irb->Base.RedBits = 8;
+      irb->Base.GreenBits = 8;
+      irb->Base.BlueBits = 8;
+      irb->Base.AlphaBits = 8;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
+      cpp = 4;
+      break;
+   case GL_STENCIL_INDEX8_EXT:
+      irb->Base._ActualFormat = GL_STENCIL_INDEX8_EXT;
+      irb->Base._BaseFormat = GL_STENCIL_INDEX;
+      irb->Base.StencilBits = 8;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
+      cpp = 1;
+      break;
+   case GL_DEPTH_COMPONENT16:
+      irb->Base._ActualFormat = GL_DEPTH_COMPONENT16;
+      irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      irb->Base.DepthBits = 16;
+      irb->Base.DataType = GL_UNSIGNED_SHORT;
+      cpp = 2;
+      break;
+   case GL_DEPTH_COMPONENT24:
+      irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      irb->Base.DepthBits = 24;
+      irb->Base.DataType = GL_UNSIGNED_INT;
+      cpp = 4;
+      break;
+   case GL_DEPTH24_STENCIL8_EXT:
+      irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      irb->Base._BaseFormat = GL_DEPTH_STENCIL_EXT;
+      irb->Base.DepthBits = 24;
+      irb->Base.StencilBits = 8;
+      irb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT;
+      cpp = 4;
+      break;
+   default:
+      _mesa_problem(NULL,
+                    "Unexpected intFormat in intel_create_renderbuffer");
+      return NULL;
+   }
+
+   irb->Base.InternalFormat = intFormat;
+
+   /* intel-specific methods */
+   irb->Base.Delete = intel_delete_renderbuffer;
+   irb->Base.AllocStorage = intel_alloc_window_storage;
+   irb->Base.GetPointer = intel_get_pointer;
+   /* This sets the Get/PutRow/Value functions */
+   intel_set_span_functions(&irb->Base);
+
+   irb->pfMap = map;
+   irb->pfPitch = pitch / cpp;	/* in pixels */
+
+#if 00
+   irb->region = intel_region_create_static(intel,
+                                            DRM_MM_TT,
+                                            offset, map, cpp, width, height);
+#endif
+
+   return irb;
+}
+
+
+/**
+ * Create a new renderbuffer object.
+ * Typically called via glBindRenderbufferEXT().
+ */
+static struct gl_renderbuffer *
+intel_new_renderbuffer(GLcontext * ctx, GLuint name)
+{
+   /*struct intel_context *intel = intel_context(ctx); */
+   struct intel_renderbuffer *irb;
+
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&irb->Base, name);
+   irb->Base.ClassID = INTEL_RB_CLASS;
+
+   /* intel-specific methods */
+   irb->Base.Delete = intel_delete_renderbuffer;
+   irb->Base.AllocStorage = intel_alloc_renderbuffer_storage;
+   irb->Base.GetPointer = intel_get_pointer;
+   /* span routines set in alloc_storage function */
+
+   return &irb->Base;
+}
+
+
+/**
+ * Called via glBindFramebufferEXT().
+ */
+static void
+intel_bind_framebuffer(GLcontext * ctx, GLenum target,
+                       struct gl_framebuffer *fb, struct gl_framebuffer *fbread)
+{
+   if (target == GL_FRAMEBUFFER_EXT || target == GL_DRAW_FRAMEBUFFER_EXT) {
+      intel_draw_buffer(ctx, fb);
+      /* Integer depth range depends on depth buffer bits */
+      ctx->Driver.DepthRange(ctx, ctx->Viewport.Near, ctx->Viewport.Far);
+   }
+   else {
+      /* don't need to do anything if target == GL_READ_FRAMEBUFFER_EXT */
+   }
+}
+
+
+/**
+ * Called via glFramebufferRenderbufferEXT().
+ */
+static void
+intel_framebuffer_renderbuffer(GLcontext * ctx,
+                               struct gl_framebuffer *fb,
+                               GLenum attachment, struct gl_renderbuffer *rb)
+{
+   DBG("Intel FramebufferRenderbuffer %u %u\n", fb->Name, rb ? rb->Name : 0);
+
+   intelFlush(ctx);
+
+   _mesa_framebuffer_renderbuffer(ctx, fb, attachment, rb);
+   intel_draw_buffer(ctx, fb);
+}
+
+
+/**
+ * When glFramebufferTexture[123]D is called this function sets up the
+ * gl_renderbuffer wrapper around the texture image.
+ * This will have the region info needed for hardware rendering.
+ */
+static struct intel_renderbuffer *
+intel_wrap_texture(GLcontext * ctx, struct gl_texture_image *texImage)
+{
+   const GLuint name = ~0;      /* not significant, but distinct for debugging */
+   struct intel_renderbuffer *irb;
+
+   /* make an intel_renderbuffer to wrap the texture image */
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glFramebufferTexture");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&irb->Base, name);
+   irb->Base.ClassID = INTEL_RB_CLASS;
+
+   if (texImage->TexFormat == &_mesa_texformat_argb8888) {
+      irb->Base._ActualFormat = GL_RGBA8;
+      irb->Base._BaseFormat = GL_RGBA;
+      DBG("Render to RGBA8 texture OK\n");
+   }
+   else if (texImage->TexFormat == &_mesa_texformat_rgb565) {
+      irb->Base._ActualFormat = GL_RGB5;
+      irb->Base._BaseFormat = GL_RGB;
+      DBG("Render to RGB5 texture OK\n");
+   }
+   else if (texImage->TexFormat == &_mesa_texformat_z16) {
+      irb->Base._ActualFormat = GL_DEPTH_COMPONENT16;
+      irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      DBG("Render to DEPTH16 texture OK\n");
+   }
+   else {
+      DBG("Render to texture BAD FORMAT %d\n",
+	  texImage->TexFormat->MesaFormat);
+      _mesa_free(irb);
+      return NULL;
+   }
+
+   irb->Base.InternalFormat = irb->Base._ActualFormat;
+   irb->Base.Width = texImage->Width;
+   irb->Base.Height = texImage->Height;
+   irb->Base.DataType = GL_UNSIGNED_BYTE;       /* FBO XXX fix */
+   irb->Base.RedBits = texImage->TexFormat->RedBits;
+   irb->Base.GreenBits = texImage->TexFormat->GreenBits;
+   irb->Base.BlueBits = texImage->TexFormat->BlueBits;
+   irb->Base.AlphaBits = texImage->TexFormat->AlphaBits;
+   irb->Base.DepthBits = texImage->TexFormat->DepthBits;
+
+   irb->Base.Delete = intel_delete_renderbuffer;
+   irb->Base.AllocStorage = intel_nop_alloc_storage;
+   intel_set_span_functions(&irb->Base);
+
+   irb->RenderToTexture = GL_TRUE;
+
+   return irb;
+}
+
+
+/**
+ * Called by glFramebufferTexture[123]DEXT() (and other places) to
+ * prepare for rendering into texture memory.  This might be called
+ * many times to choose different texture levels, cube faces, etc
+ * before intel_finish_render_texture() is ever called.
+ */
+static void
+intel_render_texture(GLcontext * ctx,
+                     struct gl_framebuffer *fb,
+                     struct gl_renderbuffer_attachment *att)
+{
+   struct gl_texture_image *newImage
+      = att->Texture->Image[att->CubeMapFace][att->TextureLevel];
+   struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer);
+   struct intel_texture_image *intel_image;
+   GLuint imageOffset;
+
+   (void) fb;
+
+   ASSERT(newImage);
+
+   if (!irb) {
+      irb = intel_wrap_texture(ctx, newImage);
+      if (irb) {
+         /* bind the wrapper to the attachment point */
+         _mesa_reference_renderbuffer(&att->Renderbuffer, &irb->Base);
+      }
+      else {
+         /* fallback to software rendering */
+         _mesa_render_texture(ctx, fb, att);
+         return;
+      }
+   }
+
+   DBG("Begin render texture tid %x tex=%u w=%d h=%d refcount=%d\n",
+       _glthread_GetID(),
+       att->Texture->Name, newImage->Width, newImage->Height,
+       irb->Base.RefCount);
+
+   /* point the renderbufer's region to the texture image region */
+   intel_image = intel_texture_image(newImage);
+   if (irb->region != intel_image->mt->region) {
+      if (irb->region)
+	 intel_region_release(&irb->region);
+      intel_region_reference(&irb->region, intel_image->mt->region);
+   }
+
+   /* compute offset of the particular 2D image within the texture region */
+   imageOffset = intel_miptree_image_offset(intel_image->mt,
+                                            att->CubeMapFace,
+                                            att->TextureLevel);
+
+   if (att->Texture->Target == GL_TEXTURE_3D) {
+      const GLuint *offsets = intel_miptree_depth_offsets(intel_image->mt,
+                                                          att->TextureLevel);
+      imageOffset += offsets[att->Zoffset];
+   }
+
+   /* store that offset in the region */
+   intel_image->mt->region->draw_offset = imageOffset;
+
+   /* update drawing region, etc */
+   intel_draw_buffer(ctx, fb);
+}
+
+
+/**
+ * Called by Mesa when rendering to a texture is done.
+ */
+static void
+intel_finish_render_texture(GLcontext * ctx,
+                            struct gl_renderbuffer_attachment *att)
+{
+   struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer);
+
+   DBG("End render texture (tid %x) tex %u\n", _glthread_GetID(), att->Texture->Name);
+
+   if (irb) {
+      /* just release the region */
+      intel_region_release(&irb->region);
+   }
+   else if (att->Renderbuffer) {
+      /* software fallback */
+      _mesa_finish_render_texture(ctx, att);
+      /* XXX FBO: Need to unmap the buffer (or in intelSpanRenderStart???) */
+   }
+}
+
+
+/**
+ * Do one-time context initializations related to GL_EXT_framebuffer_object.
+ * Hook in device driver functions.
+ */
+void
+intel_fbo_init(struct intel_context *intel)
+{
+   intel->ctx.Driver.NewFramebuffer = intel_new_framebuffer;
+   intel->ctx.Driver.NewRenderbuffer = intel_new_renderbuffer;
+   intel->ctx.Driver.BindFramebuffer = intel_bind_framebuffer;
+   intel->ctx.Driver.FramebufferRenderbuffer = intel_framebuffer_renderbuffer;
+   intel->ctx.Driver.RenderTexture = intel_render_texture;
+   intel->ctx.Driver.FinishRenderTexture = intel_finish_render_texture;
+   intel->ctx.Driver.ResizeBuffers = intel_resize_buffers;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_fbo.h b/src/mesa/drivers/dri/intel/intel_fbo.h
new file mode 100644
index 00000000000..f9a11d02e30
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_fbo.h
@@ -0,0 +1,113 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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_FBO_H
+#define INTEL_FBO_H
+
+
+struct intel_context;
+struct intel_region;
+
+/**
+ * Intel framebuffer, derived from gl_framebuffer.
+ */
+struct intel_framebuffer
+{
+   struct gl_framebuffer Base;
+
+   struct intel_renderbuffer *color_rb[3];
+
+   /* Drawable page flipping state */
+   GLboolean pf_active;
+   GLuint pf_seq;
+   GLint pf_planes;
+   GLint pf_current_page;
+   GLint pf_num_pages;
+
+   /* VBI
+    */
+   GLuint vbl_waited;
+
+   int64_t swap_ust;
+   int64_t swap_missed_ust;
+
+   GLuint swap_count;
+   GLuint swap_missed_count;
+};
+
+
+/**
+ * Intel renderbuffer, derived from gl_renderbuffer.
+ * Note: The PairedDepth and PairedStencil fields use renderbuffer IDs,
+ * not pointers because in some circumstances a deleted renderbuffer could
+ * result in a dangling pointer here.
+ */
+struct intel_renderbuffer
+{
+   struct gl_renderbuffer Base;
+   struct intel_region *region;
+   void *pfMap;                 /* possibly paged flipped map pointer */
+   GLuint pfPitch;              /* possibly paged flipped pitch */
+   GLboolean RenderToTexture;   /* RTT? */
+
+   GLuint PairedDepth;   /**< only used if this is a depth renderbuffer */
+   GLuint PairedStencil; /**< only used if this is a stencil renderbuffer */
+
+   GLuint pf_pending;  /**< sequence number of pending flip */
+
+   GLuint vbl_pending;   /**< vblank sequence number of pending flip */
+};
+
+
+extern struct intel_renderbuffer *intel_create_renderbuffer(GLenum intFormat,
+                                                            GLsizei width,
+                                                            GLsizei height,
+                                                            int offset,
+                                                            int pitch,
+                                                            int cpp,
+                                                            void *map);
+
+
+extern void intel_fbo_init(struct intel_context *intel);
+
+
+/* XXX make inline or macro */
+extern struct intel_renderbuffer *intel_get_renderbuffer(struct gl_framebuffer
+                                                         *fb,
+                                                         GLuint attIndex);
+
+extern void intel_flip_renderbuffers(struct intel_framebuffer *intel_fb);
+
+
+/* XXX make inline or macro */
+extern struct intel_region *intel_get_rb_region(struct gl_framebuffer *fb,
+                                                GLuint attIndex);
+
+
+
+
+#endif /* INTEL_FBO_H */
diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.c b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
new file mode 100644
index 00000000000..2c167a9ab7b
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.c
@@ -0,0 +1,388 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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 "intel_context.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
+#include "enums.h"
+
+#define FILE_DEBUG_FLAG DEBUG_MIPTREE
+
+static GLenum
+target_to_target(GLenum target)
+{
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB:
+      return GL_TEXTURE_CUBE_MAP_ARB;
+   default:
+      return target;
+   }
+}
+
+struct intel_mipmap_tree *
+intel_miptree_create(struct intel_context *intel,
+                     GLenum target,
+                     GLenum internal_format,
+                     GLuint first_level,
+                     GLuint last_level,
+                     GLuint width0,
+                     GLuint height0,
+                     GLuint depth0, GLuint cpp, GLuint compress_byte)
+{
+   GLboolean ok;
+   struct intel_mipmap_tree *mt = calloc(sizeof(*mt), 1);
+
+   DBG("%s target %s format %s level %d..%d\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(target),
+       _mesa_lookup_enum_by_nr(internal_format), first_level, last_level);
+
+   mt->target = target_to_target(target);
+   mt->internal_format = internal_format;
+   mt->first_level = first_level;
+   mt->last_level = last_level;
+   mt->width0 = width0;
+   mt->height0 = height0;
+   mt->depth0 = depth0;
+   mt->cpp = compress_byte ? compress_byte : cpp;
+   mt->compressed = compress_byte ? 1 : 0;
+   mt->refcount = 1; 
+
+   switch (intel->intelScreen->deviceID) {
+   case PCI_CHIP_I945_G:
+   case PCI_CHIP_I945_GM:
+   case PCI_CHIP_I945_GME:
+   case PCI_CHIP_G33_G:
+   case PCI_CHIP_Q33_G:
+   case PCI_CHIP_Q35_G:
+      ok = i945_miptree_layout(mt);
+      break;
+   case PCI_CHIP_I915_G:
+   case PCI_CHIP_I915_GM:
+   case PCI_CHIP_I830_M:
+   case PCI_CHIP_I855_GM:
+   case PCI_CHIP_I865_G:
+   default:
+      /* All the i830 chips and the i915 use this layout:
+       */
+      ok = i915_miptree_layout(mt);
+      break;
+   }
+
+   if (ok) {
+      if (!mt->compressed) {
+	 int align;
+
+	 if (intel->intelScreen->ttm) {
+	    /* XXX: Align pitch to multiple of 64 bytes for now to allow
+	     * render-to-texture to work in all cases. This should probably be
+	     * replaced at some point by some scheme to only do this when really
+	     * necessary.
+	     */
+	    align = 63;
+	 } else {
+	    align = 3;
+	 }
+
+	 mt->pitch = (mt->pitch * cpp + align) & ~align;
+
+	 /* XXX: At least the i915 seems very upset when the pitch is a multiple
+	  * of 1024 and sometimes 512 bytes - performance can drop by several
+	  * times. Go to the next multiple of the required alignment for now.
+	  */
+	 if (!(mt->pitch & 511))
+	    mt->pitch += align + 1;
+
+	 mt->pitch /= cpp;
+      }
+
+      mt->region = intel_region_alloc(intel->intelScreen,
+                                      mt->cpp, mt->pitch, mt->total_height);
+   }
+
+   if (!mt->region) {
+      free(mt);
+      return NULL;
+   }
+
+   return mt;
+}
+
+
+void
+intel_miptree_reference(struct intel_mipmap_tree **dst,
+                        struct intel_mipmap_tree *src)
+{
+   src->refcount++;
+   *dst = src;
+   DBG("%s %p refcount now %d\n", __FUNCTION__, src, src->refcount);
+}
+
+void
+intel_miptree_release(struct intel_context *intel,
+                      struct intel_mipmap_tree **mt)
+{
+   if (!*mt)
+      return;
+
+   DBG("%s %p refcount will be %d\n", __FUNCTION__, *mt, (*mt)->refcount - 1);
+   if (--(*mt)->refcount <= 0) {
+      GLuint i;
+
+      DBG("%s deleting %p\n", __FUNCTION__, *mt);
+
+      intel_region_release(&((*mt)->region));
+
+      for (i = 0; i < MAX_TEXTURE_LEVELS; i++)
+         if ((*mt)->level[i].image_offset)
+            free((*mt)->level[i].image_offset);
+
+      free(*mt);
+   }
+   *mt = NULL;
+}
+
+
+
+
+/* Can the image be pulled into a unified mipmap tree.  This mirrors
+ * the completeness test in a lot of ways.
+ *
+ * Not sure whether I want to pass gl_texture_image here.
+ */
+GLboolean
+intel_miptree_match_image(struct intel_mipmap_tree *mt,
+                          struct gl_texture_image *image,
+                          GLuint face, GLuint level)
+{
+   /* Images with borders are never pulled into mipmap trees. 
+    */
+   if (image->Border) 
+      return GL_FALSE;
+
+   if (image->InternalFormat != mt->internal_format ||
+       image->IsCompressed != mt->compressed)
+      return GL_FALSE;
+
+   /* Test image dimensions against the base level image adjusted for
+    * minification.  This will also catch images not present in the
+    * tree, changed targets, etc.
+    */
+   if (image->Width != mt->level[level].width ||
+       image->Height != mt->level[level].height ||
+       image->Depth != mt->level[level].depth)
+      return GL_FALSE;
+
+   return GL_TRUE;
+}
+
+
+void
+intel_miptree_set_level_info(struct intel_mipmap_tree *mt,
+                             GLuint level,
+                             GLuint nr_images,
+                             GLuint x, GLuint y, GLuint w, GLuint h, GLuint d)
+{
+
+   mt->level[level].width = w;
+   mt->level[level].height = h;
+   mt->level[level].depth = d;
+   mt->level[level].level_offset = (x + y * mt->pitch) * mt->cpp;
+   mt->level[level].nr_images = nr_images;
+
+   DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
+       level, w, h, d, x, y, mt->level[level].level_offset);
+
+   /* Not sure when this would happen, but anyway: 
+    */
+   if (mt->level[level].image_offset) {
+      free(mt->level[level].image_offset);
+      mt->level[level].image_offset = NULL;
+   }
+
+   assert(nr_images);
+
+   mt->level[level].image_offset = malloc(nr_images * sizeof(GLuint));
+   mt->level[level].image_offset[0] = 0;
+}
+
+
+
+void
+intel_miptree_set_image_offset(struct intel_mipmap_tree *mt,
+                               GLuint level, GLuint img, GLuint x, GLuint y)
+{
+   if (img == 0 && level == 0)
+      assert(x == 0 && y == 0);
+
+   assert(img < mt->level[level].nr_images);
+
+   mt->level[level].image_offset[img] = (x + y * mt->pitch);
+
+   DBG("%s level %d img %d pos %d,%d image_offset %x\n",
+       __FUNCTION__, level, img, x, y, mt->level[level].image_offset[img]);
+}
+
+
+/* Although we use the image_offset[] array to store relative offsets
+ * to cube faces, Mesa doesn't know anything about this and expects
+ * each cube face to be treated as a separate image.
+ *
+ * These functions present that view to mesa:
+ */
+const GLuint *
+intel_miptree_depth_offsets(struct intel_mipmap_tree *mt, GLuint level)
+{
+   static const GLuint zero = 0;
+
+   if (mt->target != GL_TEXTURE_3D || mt->level[level].nr_images == 1)
+      return &zero;
+   else
+      return mt->level[level].image_offset;
+}
+
+
+GLuint
+intel_miptree_image_offset(struct intel_mipmap_tree * mt,
+                           GLuint face, GLuint level)
+{
+   if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
+      return (mt->level[level].level_offset +
+              mt->level[level].image_offset[face] * mt->cpp);
+   else
+      return mt->level[level].level_offset;
+}
+
+
+
+/**
+ * Map a teximage in a mipmap tree.
+ * \param row_stride  returns row stride in bytes
+ * \param image_stride  returns image stride in bytes (for 3D textures).
+ * \return address of mapping
+ */
+GLubyte *
+intel_miptree_image_map(struct intel_context * intel,
+                        struct intel_mipmap_tree * mt,
+                        GLuint face,
+                        GLuint level,
+                        GLuint * row_stride, GLuint * image_offsets)
+{
+   DBG("%s \n", __FUNCTION__);
+
+   if (row_stride)
+      *row_stride = mt->pitch * mt->cpp;
+
+   if (image_offsets)
+      memcpy(image_offsets, mt->level[level].image_offset,
+             mt->level[level].depth * sizeof(GLuint));
+
+   return (intel_region_map(intel->intelScreen, mt->region) +
+           intel_miptree_image_offset(mt, face, level));
+}
+
+void
+intel_miptree_image_unmap(struct intel_context *intel,
+                          struct intel_mipmap_tree *mt)
+{
+   DBG("%s\n", __FUNCTION__);
+   intel_region_unmap(intel->intelScreen, mt->region);
+}
+
+
+
+/* Upload data for a particular image.
+ */
+void
+intel_miptree_image_data(struct intel_context *intel,
+                         struct intel_mipmap_tree *dst,
+                         GLuint face,
+                         GLuint level,
+                         void *src,
+                         GLuint src_row_pitch, GLuint src_image_pitch)
+{
+   GLuint depth = dst->level[level].depth;
+   GLuint dst_offset = intel_miptree_image_offset(dst, face, level);
+   const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level);
+   GLuint i;
+   GLuint height = 0;
+
+   DBG("%s\n", __FUNCTION__);
+   for (i = 0; i < depth; i++) {
+      height = dst->level[level].height;
+      if(dst->compressed)
+	 height /= 4;
+      intel_region_data(intel->intelScreen, dst->region,
+                        dst_offset + dst_depth_offset[i], /* dst_offset */
+                        0, 0,                             /* dstx, dsty */
+                        src,
+                        src_row_pitch,
+                        0, 0,                             /* source x, y */
+                        dst->level[level].width, height); /* width, height */
+
+      src += src_image_pitch * dst->cpp;
+   }
+}
+
+extern GLuint intel_compressed_alignment(GLenum);
+/* Copy mipmap image between trees
+ */
+void
+intel_miptree_image_copy(struct intel_context *intel,
+                         struct intel_mipmap_tree *dst,
+                         GLuint face, GLuint level,
+                         struct intel_mipmap_tree *src)
+{
+   GLuint width = src->level[level].width;
+   GLuint height = src->level[level].height;
+   GLuint depth = src->level[level].depth;
+   GLuint dst_offset = intel_miptree_image_offset(dst, face, level);
+   GLuint src_offset = intel_miptree_image_offset(src, face, level);
+   const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level);
+   const GLuint *src_depth_offset = intel_miptree_depth_offsets(src, level);
+   GLuint i;
+
+   if (dst->compressed) {
+       GLuint alignment = intel_compressed_alignment(dst->internal_format);
+       height = (height + 3) / 4;
+       width = ((width + alignment - 1) & ~(alignment - 1));
+   }
+
+   for (i = 0; i < depth; i++) {
+      intel_region_copy(intel->intelScreen,
+                        dst->region, dst_offset + dst_depth_offset[i],
+                        0,
+                        0,
+                        src->region, src_offset + src_depth_offset[i],
+                        0, 0, width, height);
+   }
+
+}
diff --git a/src/mesa/drivers/dri/intel/intel_mipmap_tree.h b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h
new file mode 100644
index 00000000000..ecdb7be244f
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_mipmap_tree.h
@@ -0,0 +1,198 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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_MIPMAP_TREE_H
+#define INTEL_MIPMAP_TREE_H
+
+#include "intel_regions.h"
+
+/* A layer on top of the intel_regions code which adds:
+ *
+ * - Code to size and layout a region to hold a set of mipmaps.
+ * - Query to determine if a new image fits in an existing tree.
+ * - More refcounting 
+ *     - maybe able to remove refcounting from intel_region?
+ * - ?
+ *
+ * The fixed mipmap layout of intel hardware where one offset
+ * specifies the position of all images in a mipmap hierachy
+ * complicates the implementation of GL texture image commands,
+ * compared to hardware where each image is specified with an
+ * independent offset.
+ *
+ * In an ideal world, each texture object would be associated with a
+ * single bufmgr buffer or 2d intel_region, and all the images within
+ * the texture object would slot into the tree as they arrive.  The
+ * reality can be a little messier, as images can arrive from the user
+ * with sizes that don't fit in the existing tree, or in an order
+ * where the tree layout cannot be guessed immediately.  
+ * 
+ * This structure encodes an idealized mipmap tree.  The GL image
+ * commands build these where possible, otherwise store the images in
+ * temporary system buffers.
+ */
+
+
+/**
+ * Describes the location of each texture image within a texture region.
+ */
+struct intel_mipmap_level
+{
+   GLuint level_offset;
+   GLuint width;
+   GLuint height;
+   GLuint depth;
+   GLuint nr_images;
+
+   /* Explicitly store the offset of each image for each cube face or
+    * depth value.  Pretty much have to accept that hardware formats
+    * are going to be so diverse that there is no unified way to
+    * compute the offsets of depth/cube images within a mipmap level,
+    * so have to store them as a lookup table:
+    */
+   GLuint *image_offset;
+};
+
+struct intel_mipmap_tree
+{
+   /* Effectively the key:
+    */
+   GLenum target;
+   GLenum internal_format;
+
+   GLuint first_level;
+   GLuint last_level;
+
+   GLuint width0, height0, depth0; /**< Level zero image dimensions */
+   GLuint cpp;
+   GLboolean compressed;
+
+   /* Derived from the above:
+    */
+   GLuint pitch;
+   GLuint depth_pitch;          /* per-image on i945? */
+   GLuint total_height;
+
+   /* Includes image offset tables:
+    */
+   struct intel_mipmap_level level[MAX_TEXTURE_LEVELS];
+
+   /* The data is held here:
+    */
+   struct intel_region *region;
+
+   /* These are also refcounted:
+    */
+   GLuint refcount;
+};
+
+
+
+struct intel_mipmap_tree *intel_miptree_create(struct intel_context *intel,
+                                               GLenum target,
+                                               GLenum internal_format,
+                                               GLuint first_level,
+                                               GLuint last_level,
+                                               GLuint width0,
+                                               GLuint height0,
+                                               GLuint depth0,
+                                               GLuint cpp,
+                                               GLuint compress_byte);
+
+void intel_miptree_reference(struct intel_mipmap_tree **dst,
+                             struct intel_mipmap_tree *src);
+
+void intel_miptree_release(struct intel_context *intel,
+                           struct intel_mipmap_tree **mt);
+
+/* Check if an image fits an existing mipmap tree layout
+ */
+GLboolean intel_miptree_match_image(struct intel_mipmap_tree *mt,
+                                    struct gl_texture_image *image,
+                                    GLuint face, GLuint level);
+
+/* Return a pointer to an image within a tree.  Return image stride as
+ * well.
+ */
+GLubyte *intel_miptree_image_map(struct intel_context *intel,
+                                 struct intel_mipmap_tree *mt,
+                                 GLuint face,
+                                 GLuint level,
+                                 GLuint * row_stride, GLuint * image_stride);
+
+void intel_miptree_image_unmap(struct intel_context *intel,
+                               struct intel_mipmap_tree *mt);
+
+
+/* Return the linear offset of an image relative to the start of the
+ * tree:
+ */
+GLuint intel_miptree_image_offset(struct intel_mipmap_tree *mt,
+                                  GLuint face, GLuint level);
+
+/* Return pointers to each 2d slice within an image.  Indexed by depth
+ * value.
+ */
+const GLuint *intel_miptree_depth_offsets(struct intel_mipmap_tree *mt,
+                                          GLuint level);
+
+
+void intel_miptree_set_level_info(struct intel_mipmap_tree *mt,
+                                  GLuint level,
+                                  GLuint nr_images,
+                                  GLuint x, GLuint y,
+                                  GLuint w, GLuint h, GLuint d);
+
+void intel_miptree_set_image_offset(struct intel_mipmap_tree *mt,
+                                    GLuint level,
+                                    GLuint img, GLuint x, GLuint y);
+
+
+/* Upload an image into a tree
+ */
+void intel_miptree_image_data(struct intel_context *intel,
+                              struct intel_mipmap_tree *dst,
+                              GLuint face,
+                              GLuint level,
+                              void *src,
+                              GLuint src_row_pitch, GLuint src_image_pitch);
+
+/* Copy an image between two trees
+ */
+void intel_miptree_image_copy(struct intel_context *intel,
+                              struct intel_mipmap_tree *dst,
+                              GLuint face, GLuint level,
+                              struct intel_mipmap_tree *src);
+
+/* i915_mipmap_tree.c:
+ */
+GLboolean i915_miptree_layout(struct intel_mipmap_tree *mt);
+GLboolean i945_miptree_layout(struct intel_mipmap_tree *mt);
+
+
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_regions.c b/src/mesa/drivers/dri/intel/intel_regions.c
new file mode 100644
index 00000000000..7799fdbb116
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_regions.c
@@ -0,0 +1,483 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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.
+ * 
+ **************************************************************************/
+
+/* Provide additional functionality on top of bufmgr buffers:
+ *   - 2d semantics and blit operations
+ *   - refcounting of buffers for multiple images in a buffer.
+ *   - refcounting of buffer mappings.
+ *   - some logic for moving the buffers to the best memory pools for
+ *     given operations.
+ *
+ * Most of this is to make it easier to implement the fixed-layout
+ * mipmap tree required by intel hardware in the face of GL's
+ * programming interface where each image can be specifed in random
+ * order and it isn't clear what layout the tree should have until the
+ * last moment.
+ */
+
+#include "intel_context.h"
+#include "intel_regions.h"
+#include "intel_blit.h"
+#include "intel_buffer_objects.h"
+#include "dri_bufmgr.h"
+#include "intel_bufmgr_ttm.h"
+#include "intel_batchbuffer.h"
+
+#define FILE_DEBUG_FLAG DEBUG_REGION
+
+void
+intel_region_idle(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   DBG("%s\n", __FUNCTION__);
+   /* XXX: Using this function is likely bogus -- it ought to only have been
+    * used before a map, anyway, but leave this cheap implementation of it
+    * for now.
+    */
+   if (region && region->buffer) {
+      /* Mapping it for read will ensure that any acceleration to the region
+       * would have landed already.
+       */
+      dri_bo_map(region->buffer, GL_TRUE);
+      dri_bo_unmap(region->buffer);
+   }
+}
+
+/* XXX: Thread safety?
+ */
+GLubyte *
+intel_region_map(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   DBG("%s\n", __FUNCTION__);
+   if (!region->map_refcount++) {
+      if (region->pbo)
+         intel_region_cow(intelScreen, region);
+
+      dri_bo_map(region->buffer, GL_TRUE);
+      region->map = region->buffer->virtual;
+   }
+
+   return region->map;
+}
+
+void
+intel_region_unmap(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   DBG("%s\n", __FUNCTION__);
+   if (!--region->map_refcount) {
+      dri_bo_unmap(region->buffer);
+      region->map = NULL;
+   }
+}
+
+struct intel_region *
+intel_region_alloc(intelScreenPrivate *intelScreen,
+                   GLuint cpp, GLuint pitch, GLuint height)
+{
+   struct intel_region *region = calloc(sizeof(*region), 1);
+
+   DBG("%s\n", __FUNCTION__);
+
+   region->cpp = cpp;
+   region->pitch = pitch;
+   region->height = height;     /* needed? */
+   region->refcount = 1;
+
+   region->buffer = dri_bo_alloc(intelScreen->bufmgr, "region",
+				 pitch * cpp * height, 64, DRM_BO_FLAG_MEM_TT);
+   return region;
+}
+
+void
+intel_region_reference(struct intel_region **dst, struct intel_region *src)
+{
+   assert(*dst == NULL);
+   if (src) {
+      src->refcount++;
+      *dst = src;
+   }
+}
+
+void
+intel_region_release(struct intel_region **region)
+{
+   if (!*region)
+      return;
+
+   DBG("%s %d\n", __FUNCTION__, (*region)->refcount - 1);
+
+   ASSERT((*region)->refcount > 0);
+   (*region)->refcount--;
+
+   if ((*region)->refcount == 0) {
+      assert((*region)->map_refcount == 0);
+
+      if ((*region)->pbo)
+	 (*region)->pbo->region = NULL;
+      (*region)->pbo = NULL;
+      dri_bo_unreference((*region)->buffer);
+      free(*region);
+   }
+   *region = NULL;
+}
+
+
+struct intel_region *
+intel_region_create_static(intelScreenPrivate *intelScreen,
+                           GLuint mem_type,
+			   unsigned int bo_handle,
+                           GLuint offset,
+                           void *virtual,
+                           GLuint cpp, GLuint pitch, GLuint height)
+{
+   struct intel_region *region = calloc(sizeof(*region), 1);
+   DBG("%s\n", __FUNCTION__);
+
+   region->cpp = cpp;
+   region->pitch = pitch;
+   region->height = height;     /* needed? */
+   region->refcount = 1;
+
+   if (intelScreen->ttm) {
+      assert(bo_handle != -1);
+      region->buffer = intel_ttm_bo_create_from_handle(intelScreen->bufmgr,
+						     "static region",
+						     bo_handle);
+   } else {
+      region->buffer = dri_bo_alloc_static(intelScreen->bufmgr,
+					   "static region",
+					   offset, pitch * cpp * height,
+					   virtual,
+					   DRM_BO_FLAG_MEM_TT);
+   }
+
+   return region;
+}
+
+
+
+void
+intel_region_update_static(intelScreenPrivate *intelScreen,
+			   struct intel_region *region,
+                           GLuint mem_type,
+			   unsigned int bo_handle,
+                           GLuint offset,
+                           void *virtual,
+                           GLuint cpp, GLuint pitch, GLuint height)
+{
+   DBG("%s\n", __FUNCTION__);
+
+   region->cpp = cpp;
+   region->pitch = pitch;
+   region->height = height;     /* needed? */
+
+   /*
+    * We use a "shared" buffer type to indicate buffers created and
+    * shared by others.
+    */
+
+   dri_bo_unreference(region->buffer);
+   if (intelScreen->ttm) {
+      assert(bo_handle != -1);
+      region->buffer = intel_ttm_bo_create_from_handle(intelScreen->bufmgr,
+						     "static region",
+						     bo_handle);
+   } else {
+      region->buffer = dri_bo_alloc_static(intelScreen->bufmgr,
+					   "static region",
+					   offset, pitch * cpp * height,
+					   virtual,
+					   DRM_BO_FLAG_MEM_TT);
+   }
+}
+
+
+
+/*
+ * XXX Move this into core Mesa?
+ */
+static void
+_mesa_copy_rect(GLubyte * dst,
+                GLuint cpp,
+                GLuint dst_pitch,
+                GLuint dst_x,
+                GLuint dst_y,
+                GLuint width,
+                GLuint height,
+                const GLubyte * src,
+                GLuint src_pitch, GLuint src_x, GLuint src_y)
+{
+   GLuint i;
+
+   dst_pitch *= cpp;
+   src_pitch *= cpp;
+   dst += dst_x * cpp;
+   src += src_x * cpp;
+   dst += dst_y * dst_pitch;
+   src += src_y * dst_pitch;
+   width *= cpp;
+
+   if (width == dst_pitch && width == src_pitch)
+      memcpy(dst, src, height * width);
+   else {
+      for (i = 0; i < height; i++) {
+         memcpy(dst, src, width);
+         dst += dst_pitch;
+         src += src_pitch;
+      }
+   }
+}
+
+
+/* Upload data to a rectangular sub-region.  Lots of choices how to do this:
+ *
+ * - memcpy by span to current destination
+ * - upload data as new buffer and blit
+ *
+ * Currently always memcpy.
+ */
+void
+intel_region_data(intelScreenPrivate *intelScreen,
+                  struct intel_region *dst,
+                  GLuint dst_offset,
+                  GLuint dstx, GLuint dsty,
+                  const void *src, GLuint src_pitch,
+                  GLuint srcx, GLuint srcy, GLuint width, GLuint height)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intel == NULL)
+      return;
+
+   if (dst->pbo) {
+      if (dstx == 0 &&
+          dsty == 0 && width == dst->pitch && height == dst->height)
+         intel_region_release_pbo(intelScreen, dst);
+      else
+         intel_region_cow(intelScreen, dst);
+   }
+
+
+   LOCK_HARDWARE(intel);
+
+   _mesa_copy_rect(intel_region_map(intelScreen, dst) + dst_offset,
+                   dst->cpp,
+                   dst->pitch,
+                   dstx, dsty, width, height, src, src_pitch, srcx, srcy);
+
+   intel_region_unmap(intelScreen, dst);
+
+   UNLOCK_HARDWARE(intel);
+
+}
+
+/* Copy rectangular sub-regions. Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+void
+intel_region_copy(intelScreenPrivate *intelScreen,
+                  struct intel_region *dst,
+                  GLuint dst_offset,
+                  GLuint dstx, GLuint dsty,
+                  struct intel_region *src,
+                  GLuint src_offset,
+                  GLuint srcx, GLuint srcy, GLuint width, GLuint height)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intel == NULL)
+      return;
+
+   if (dst->pbo) {
+      if (dstx == 0 &&
+          dsty == 0 && width == dst->pitch && height == dst->height)
+         intel_region_release_pbo(intelScreen, dst);
+      else
+         intel_region_cow(intelScreen, dst);
+   }
+
+   assert(src->cpp == dst->cpp);
+
+   intelEmitCopyBlit(intel,
+                     dst->cpp,
+                     src->pitch, src->buffer, src_offset,
+                     dst->pitch, dst->buffer, dst_offset,
+                     srcx, srcy, dstx, dsty, width, height,
+		     GL_COPY);
+}
+
+/* Fill a rectangular sub-region.  Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+void
+intel_region_fill(intelScreenPrivate *intelScreen,
+                  struct intel_region *dst,
+                  GLuint dst_offset,
+                  GLuint dstx, GLuint dsty,
+                  GLuint width, GLuint height, GLuint color)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intel == NULL)
+      return;   
+
+   if (dst->pbo) {
+      if (dstx == 0 &&
+          dsty == 0 && width == dst->pitch && height == dst->height)
+         intel_region_release_pbo(intelScreen, dst);
+      else
+         intel_region_cow(intelScreen, dst);
+   }
+
+   intelEmitFillBlit(intel,
+                     dst->cpp,
+                     dst->pitch, dst->buffer, dst_offset,
+                     dstx, dsty, width, height, color);
+}
+
+/* Attach to a pbo, discarding our data.  Effectively zero-copy upload
+ * the pbo's data.
+ */
+void
+intel_region_attach_pbo(intelScreenPrivate *intelScreen,
+                        struct intel_region *region,
+                        struct intel_buffer_object *pbo)
+{
+   if (region->pbo == pbo)
+      return;
+
+   /* If there is already a pbo attached, break the cow tie now.
+    * Don't call intel_region_release_pbo() as that would
+    * unnecessarily allocate a new buffer we would have to immediately
+    * discard.
+    */
+   if (region->pbo) {
+      region->pbo->region = NULL;
+      region->pbo = NULL;
+   }
+
+   if (region->buffer) {
+      dri_bo_unreference(region->buffer);
+      region->buffer = NULL;
+   }
+
+   region->pbo = pbo;
+   region->pbo->region = region;
+   dri_bo_reference(pbo->buffer);
+   region->buffer = pbo->buffer;
+}
+
+
+/* Break the COW tie to the pbo and allocate a new buffer.
+ * The pbo gets to keep the data.
+ */
+void
+intel_region_release_pbo(intelScreenPrivate *intelScreen,
+                         struct intel_region *region)
+{
+   assert(region->buffer == region->pbo->buffer);
+   region->pbo->region = NULL;
+   region->pbo = NULL;
+   dri_bo_unreference(region->buffer);
+   region->buffer = NULL;
+
+   region->buffer = dri_bo_alloc(intelScreen->bufmgr, "region",
+				 region->pitch * region->cpp * region->height,
+				 64, DRM_BO_FLAG_MEM_TT);
+}
+
+/* Break the COW tie to the pbo.  Both the pbo and the region end up
+ * with a copy of the data.
+ */
+void
+intel_region_cow(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+   struct intel_buffer_object *pbo = region->pbo;
+
+   if (intel == NULL)
+      return;
+
+   intel_region_release_pbo(intelScreen, region);
+
+   assert(region->cpp * region->pitch * region->height == pbo->Base.Size);
+
+   DBG("%s (%d bytes)\n", __FUNCTION__, pbo->Base.Size);
+
+   /* Now blit from the texture buffer to the new buffer: 
+    */
+
+   intel_batchbuffer_flush(intel->batch);
+
+   if (!intel->locked) {
+      LOCK_HARDWARE(intel);
+      intelEmitCopyBlit(intel,
+			region->cpp,
+			region->pitch,
+			region->buffer, 0,
+			region->pitch,
+			pbo->buffer, 0,
+			0, 0, 0, 0, 
+			region->pitch, region->height,
+			GL_COPY);
+      
+      intel_batchbuffer_flush(intel->batch);
+      UNLOCK_HARDWARE(intel);
+   }
+   else {
+      intelEmitCopyBlit(intel,
+			region->cpp,
+			region->pitch,
+			region->buffer, 0,
+			region->pitch,
+			pbo->buffer, 0,
+			0, 0, 0, 0, 
+			region->pitch, region->height,
+			GL_COPY);
+      
+      intel_batchbuffer_flush(intel->batch);
+   }
+}
+
+dri_bo *
+intel_region_buffer(intelScreenPrivate *intelScreen,
+                    struct intel_region *region, GLuint flag)
+{
+   if (region->pbo) {
+      if (flag == INTEL_WRITE_PART)
+         intel_region_cow(intelScreen, region);
+      else if (flag == INTEL_WRITE_FULL)
+         intel_region_release_pbo(intelScreen, region);
+   }
+
+   return region->buffer;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_regions.h b/src/mesa/drivers/dri/intel/intel_regions.h
new file mode 100644
index 00000000000..42d7b177117
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_regions.h
@@ -0,0 +1,143 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 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_REGIONS_H
+#define INTEL_REGIONS_H
+
+#include "mtypes.h"
+#include "intel_screen.h"
+
+struct intel_context;
+struct intel_buffer_object;
+
+/**
+ * A layer on top of the bufmgr buffers that adds a few useful things:
+ *
+ * - Refcounting for local buffer references.
+ * - Refcounting for buffer maps
+ * - Buffer dimensions - pitch and height.
+ * - Blitter commands for copying 2D regions between buffers. (really???)
+ */
+struct intel_region
+{
+   dri_bo *buffer;  /**< buffer manager's buffer */
+   GLuint refcount; /**< Reference count for region */
+   GLuint cpp;      /**< bytes per pixel */
+   GLuint pitch;    /**< in pixels */
+   GLuint height;   /**< in pixels */
+   GLubyte *map;    /**< only non-NULL when region is actually mapped */
+   GLuint map_refcount;  /**< Reference count for mapping */
+
+   GLuint draw_offset; /**< Offset of drawing address within the region */
+
+   struct intel_buffer_object *pbo;     /* zero-copy uploads */
+};
+
+
+/* Allocate a refcounted region.  Pointers to regions should only be
+ * copied by calling intel_reference_region().
+ */
+struct intel_region *intel_region_alloc(intelScreenPrivate *intelScreen,
+                                        GLuint cpp,
+                                        GLuint pitch, GLuint height);
+
+void intel_region_reference(struct intel_region **dst,
+                            struct intel_region *src);
+
+void intel_region_release(struct intel_region **ib);
+
+extern struct intel_region 
+*intel_region_create_static(intelScreenPrivate *intelScreen,
+			    GLuint mem_type,
+			    unsigned int bo_handle,
+			    GLuint offset,
+			    void *virtual,
+			    GLuint cpp,
+			    GLuint pitch, GLuint height);
+extern void 
+intel_region_update_static(intelScreenPrivate *intelScreen,
+			   struct intel_region *region,
+			   GLuint mem_type,
+			   unsigned int bo_handle,
+			   GLuint offset,
+			   void *virtual,
+			   GLuint cpp, GLuint pitch, GLuint height);
+
+
+void intel_region_idle(intelScreenPrivate *intelScreen,
+		       struct intel_region *ib);
+
+/* Map/unmap regions.  This is refcounted also: 
+ */
+GLubyte *intel_region_map(intelScreenPrivate *intelScreen,
+                          struct intel_region *ib);
+
+void intel_region_unmap(intelScreenPrivate *intelScreen, struct intel_region *ib);
+
+
+/* Upload data to a rectangular sub-region
+ */
+void intel_region_data(intelScreenPrivate *intelScreen,
+                       struct intel_region *dest,
+                       GLuint dest_offset,
+                       GLuint destx, GLuint desty,
+                       const void *src, GLuint src_stride,
+                       GLuint srcx, GLuint srcy, GLuint width, GLuint height);
+
+/* Copy rectangular sub-regions
+ */
+void intel_region_copy(intelScreenPrivate *intelScreen,
+                       struct intel_region *dest,
+                       GLuint dest_offset,
+                       GLuint destx, GLuint desty,
+                       struct intel_region *src,
+                       GLuint src_offset,
+                       GLuint srcx, GLuint srcy, GLuint width, GLuint height);
+
+/* Fill a rectangular sub-region
+ */
+void intel_region_fill(intelScreenPrivate *intelScreen,
+                       struct intel_region *dest,
+                       GLuint dest_offset,
+                       GLuint destx, GLuint desty,
+                       GLuint width, GLuint height, GLuint color);
+
+/* Helpers for zerocopy uploads, particularly texture image uploads:
+ */
+void intel_region_attach_pbo(intelScreenPrivate *intelScreen,
+                             struct intel_region *region,
+                             struct intel_buffer_object *pbo);
+void intel_region_release_pbo(intelScreenPrivate *intelScreen,
+                              struct intel_region *region);
+void intel_region_cow(intelScreenPrivate *intelScreen,
+                      struct intel_region *region);
+
+dri_bo *intel_region_buffer(intelScreenPrivate *intelScreen,
+			    struct intel_region *region,
+			    GLuint flag);
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_screen.c b/src/mesa/drivers/dri/intel/intel_screen.c
new file mode 100644
index 00000000000..25f5efa7bcf
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_screen.c
@@ -0,0 +1,945 @@
+/**************************************************************************
+ * 
+ * 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 "context.h"
+#include "framebuffer.h"
+#include "matrix.h"
+#include "renderbuffer.h"
+#include "simple_list.h"
+#include "utils.h"
+#include "vblank.h"
+#include "xmlpool.h"
+
+
+#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"
+#include "dri_bufmgr.h"
+#include "intel_regions.h"
+#include "intel_batchbuffer.h"
+
+#include "intel_bufmgr_ttm.h"
+
+PUBLIC const char __driConfigOptions[] =
+   DRI_CONF_BEGIN DRI_CONF_SECTION_PERFORMANCE
+   DRI_CONF_FTHROTTLE_MODE(DRI_CONF_FTHROTTLE_IRQS)
+   DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0)
+   DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY
+   DRI_CONF_FORCE_S3TC_ENABLE(false)
+   DRI_CONF_ALLOW_LARGE_TEXTURES(1)
+   DRI_CONF_SECTION_END DRI_CONF_END;
+     const GLuint __driNConfigOptions = 4;
+
+#ifdef USE_NEW_INTERFACE
+     static PFNGLXCREATECONTEXTMODES create_context_modes = NULL;
+#endif /*USE_NEW_INTERFACE */
+
+     extern const struct dri_extension card_extensions[];
+     extern const struct dri_extension ttm_extensions[];
+
+/**
+ * Map all the memory regions described by the screen.
+ * \return GL_TRUE if success, GL_FALSE if error.
+ */
+GLboolean
+intelMapScreenRegions(__DRIscreenPrivate * sPriv)
+{
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
+
+   if (intelScreen->front.handle) {
+      if (drmMap(sPriv->fd,
+                 intelScreen->front.handle,
+                 intelScreen->front.size,
+                 (drmAddress *) & intelScreen->front.map) != 0) {
+         _mesa_problem(NULL, "drmMap(frontbuffer) failed!");
+         return GL_FALSE;
+      }
+   }
+   else {
+      _mesa_warning(NULL, "no front buffer handle in intelMapScreenRegions!");
+   }
+
+   if (0)
+      _mesa_printf("Back 0x%08x ", intelScreen->back.handle);
+   if (drmMap(sPriv->fd,
+              intelScreen->back.handle,
+              intelScreen->back.size,
+              (drmAddress *) & intelScreen->back.map) != 0) {
+      intelUnmapScreenRegions(intelScreen);
+      return GL_FALSE;
+   }
+
+   if (intelScreen->third.handle) {
+      if (0)
+	 _mesa_printf("Third 0x%08x ", intelScreen->third.handle);
+      if (drmMap(sPriv->fd,
+		 intelScreen->third.handle,
+		 intelScreen->third.size,
+		 (drmAddress *) & intelScreen->third.map) != 0) {
+	 intelUnmapScreenRegions(intelScreen);
+	 return GL_FALSE;
+      }
+   }
+
+   if (0)
+      _mesa_printf("Depth 0x%08x ", intelScreen->depth.handle);
+   if (drmMap(sPriv->fd,
+              intelScreen->depth.handle,
+              intelScreen->depth.size,
+              (drmAddress *) & intelScreen->depth.map) != 0) {
+      intelUnmapScreenRegions(intelScreen);
+      return GL_FALSE;
+   }
+
+   if (0)
+      _mesa_printf("TEX 0x%08x ", intelScreen->tex.handle);
+   if (intelScreen->tex.size != 0) {
+      if (drmMap(sPriv->fd,
+		 intelScreen->tex.handle,
+		 intelScreen->tex.size,
+		 (drmAddress *) & intelScreen->tex.map) != 0) {
+	 intelUnmapScreenRegions(intelScreen);
+	 return GL_FALSE;
+      }
+   }
+
+   if (0)
+      printf("Mappings:  front: %p  back: %p  third: %p  depth: %p  tex: %p\n",
+             intelScreen->front.map,
+             intelScreen->back.map, intelScreen->third.map,
+             intelScreen->depth.map, intelScreen->tex.map);
+   return GL_TRUE;
+}
+
+/** Driver-specific fence emit implementation for the fake memory manager. */
+static unsigned int
+intel_fence_emit(void *private)
+{
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *)private;
+   unsigned int fence;
+
+   /* XXX: Need to emit a flush, if we haven't already (at least with the
+    * current batchbuffer implementation, we have).
+    */
+
+   fence = intelEmitIrqLocked(intelScreen);
+
+   return fence;
+}
+
+/** Driver-specific fence wait implementation for the fake memory manager. */
+static int
+intel_fence_wait(void *private, unsigned int cookie)
+{
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *)private;
+
+   intelWaitIrq(intelScreen, cookie);
+
+   return 0;
+}
+
+static struct intel_region *
+intel_recreate_static(intelScreenPrivate *intelScreen,
+		      struct intel_region *region,
+		      intelRegion *region_desc,
+		      GLuint mem_type)
+{
+  if (region) {
+    intel_region_update_static(intelScreen, region, mem_type,
+			       region_desc->bo_handle, region_desc->offset,
+			       region_desc->map, intelScreen->cpp,
+			       region_desc->pitch / intelScreen->cpp,
+			       intelScreen->height);
+  } else {
+    region = intel_region_create_static(intelScreen, mem_type,
+					region_desc->bo_handle,
+					region_desc->offset,
+					region_desc->map, intelScreen->cpp,
+					region_desc->pitch / intelScreen->cpp,
+					intelScreen->height);
+  }
+
+  assert(region->buffer != NULL);
+
+  return region;
+}
+    
+
+/* Create intel_region structs to describe the static front,back,depth
+ * buffers created by the xserver. 
+ *
+ * Although FBO's mean we now no longer use these as render targets in
+ * all circumstances, they won't go away until the back and depth
+ * buffers become private, and the front and rotated buffers will
+ * remain even then.
+ *
+ * Note that these don't allocate video memory, just describe
+ * allocations alread made by the X server.
+ */
+static void
+intel_recreate_static_regions(intelScreenPrivate *intelScreen)
+{
+   intelScreen->front_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->front_region,
+			    &intelScreen->front,
+			    DRM_BO_FLAG_MEM_TT);
+
+   /* The rotated region is only used for old DDXes that didn't handle rotation
+\    * on their own.
+    */
+   if (intelScreen->driScrnPriv->ddx_version.minor < 8) {
+      intelScreen->rotated_region =
+	 intel_recreate_static(intelScreen,
+			       intelScreen->rotated_region,
+			       &intelScreen->rotated,
+			       DRM_BO_FLAG_MEM_TT);
+   }
+
+   intelScreen->back_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->back_region,
+			    &intelScreen->back,
+			    DRM_BO_FLAG_MEM_TT);
+
+   if (intelScreen->third.handle) {
+      intelScreen->third_region =
+	 intel_recreate_static(intelScreen,
+			       intelScreen->third_region,
+			       &intelScreen->third,
+			       DRM_BO_FLAG_MEM_TT);
+   }
+
+   /* Still assumes front.cpp == depth.cpp.  We can kill this when we move to
+    * private buffers.
+    */
+   intelScreen->depth_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->depth_region,
+			    &intelScreen->depth,
+			    DRM_BO_FLAG_MEM_TT);
+}
+
+/**
+ * Use the information in the sarea to update the screen parameters
+ * related to screen rotation. Needs to be called locked.
+ */
+void
+intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea)
+{
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
+
+   intelUnmapScreenRegions(intelScreen);
+   intelUpdateScreenFromSAREA(intelScreen, sarea);
+   if (!intelMapScreenRegions(sPriv)) {
+      fprintf(stderr, "ERROR Remapping screen regions!!!\n");
+   }
+   intel_recreate_static_regions(intelScreen);
+}
+
+
+void
+intelUnmapScreenRegions(intelScreenPrivate * intelScreen)
+{
+#define REALLY_UNMAP 1
+   if (intelScreen->front.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->front.map, intelScreen->front.size) != 0)
+         printf("drmUnmap front failed!\n");
+#endif
+      intelScreen->front.map = NULL;
+   }
+   if (intelScreen->back.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->back.map, intelScreen->back.size) != 0)
+         printf("drmUnmap back failed!\n");
+#endif
+      intelScreen->back.map = NULL;
+   }
+   if (intelScreen->third.map) {
+#if REALLY_UNMAP
+      if (drmUnmap(intelScreen->third.map, intelScreen->third.size) != 0)
+         printf("drmUnmap third failed!\n");
+#endif
+      intelScreen->third.map = NULL;
+   }
+   if (intelScreen->depth.map) {
+#if REALLY_UNMAP
+      drmUnmap(intelScreen->depth.map, intelScreen->depth.size);
+      intelScreen->depth.map = NULL;
+#endif
+   }
+   if (intelScreen->tex.map) {
+#if REALLY_UNMAP
+      drmUnmap(intelScreen->tex.map, intelScreen->tex.size);
+      intelScreen->tex.map = NULL;
+#endif
+   }
+}
+
+
+static void
+intelPrintDRIInfo(intelScreenPrivate * intelScreen,
+                  __DRIscreenPrivate * sPriv, I830DRIPtr gDRIPriv)
+{
+   fprintf(stderr, "*** Front size:   0x%x  offset: 0x%x  pitch: %d\n",
+           intelScreen->front.size, intelScreen->front.offset,
+           intelScreen->front.pitch);
+   fprintf(stderr, "*** Back size:    0x%x  offset: 0x%x  pitch: %d\n",
+           intelScreen->back.size, intelScreen->back.offset,
+           intelScreen->back.pitch);
+   fprintf(stderr, "*** Depth size:   0x%x  offset: 0x%x  pitch: %d\n",
+           intelScreen->depth.size, intelScreen->depth.offset,
+           intelScreen->depth.pitch);
+   fprintf(stderr, "*** Rotated size: 0x%x  offset: 0x%x  pitch: %d\n",
+           intelScreen->rotated.size, intelScreen->rotated.offset,
+           intelScreen->rotated.pitch);
+   fprintf(stderr, "*** Texture size: 0x%x  offset: 0x%x\n",
+           intelScreen->tex.size, intelScreen->tex.offset);
+   fprintf(stderr, "*** Memory : 0x%x\n", gDRIPriv->mem);
+}
+
+
+static void
+intelPrintSAREA(const drmI830Sarea * sarea)
+{
+   fprintf(stderr, "SAREA: sarea width %d  height %d\n", sarea->width,
+           sarea->height);
+   fprintf(stderr, "SAREA: pitch: %d\n", sarea->pitch);
+   fprintf(stderr,
+           "SAREA: front offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->front_offset, sarea->front_size,
+           (unsigned) sarea->front_handle);
+   fprintf(stderr,
+           "SAREA: back  offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->back_offset, sarea->back_size,
+           (unsigned) sarea->back_handle);
+   fprintf(stderr, "SAREA: depth offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->depth_offset, sarea->depth_size,
+           (unsigned) sarea->depth_handle);
+   fprintf(stderr, "SAREA: tex   offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
+           sarea->tex_offset, sarea->tex_size, (unsigned) sarea->tex_handle);
+   fprintf(stderr, "SAREA: rotation: %d\n", sarea->rotation);
+   fprintf(stderr,
+           "SAREA: rotated offset: 0x%08x  size: 0x%x\n",
+           sarea->rotated_offset, sarea->rotated_size);
+   fprintf(stderr, "SAREA: rotated pitch: %d\n", sarea->rotated_pitch);
+}
+
+
+/**
+ * A number of the screen parameters are obtained/computed from
+ * information in the SAREA.  This function updates those parameters.
+ */
+void
+intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen,
+                           drmI830Sarea * sarea)
+{
+   intelScreen->width = sarea->width;
+   intelScreen->height = sarea->height;
+
+   intelScreen->front.offset = sarea->front_offset;
+   intelScreen->front.pitch = sarea->pitch * intelScreen->cpp;
+   intelScreen->front.handle = sarea->front_handle;
+   intelScreen->front.size = sarea->front_size;
+
+   intelScreen->back.offset = sarea->back_offset;
+   intelScreen->back.pitch = sarea->pitch * intelScreen->cpp;
+   intelScreen->back.handle = sarea->back_handle;
+   intelScreen->back.size = sarea->back_size;
+
+   if (intelScreen->driScrnPriv->ddx_version.minor >= 8) {
+      intelScreen->third.offset = sarea->third_offset;
+      intelScreen->third.pitch = sarea->pitch * intelScreen->cpp;
+      intelScreen->third.handle = sarea->third_handle;
+      intelScreen->third.size = sarea->third_size;
+   }
+
+   intelScreen->depth.offset = sarea->depth_offset;
+   intelScreen->depth.pitch = sarea->pitch * intelScreen->cpp;
+   intelScreen->depth.handle = sarea->depth_handle;
+   intelScreen->depth.size = sarea->depth_size;
+
+   if (intelScreen->driScrnPriv->ddx_version.minor >= 9) {
+      intelScreen->front.bo_handle = sarea->front_bo_handle;
+      intelScreen->back.bo_handle = sarea->back_bo_handle;
+      intelScreen->third.bo_handle = sarea->third_bo_handle;
+      intelScreen->depth.bo_handle = sarea->depth_bo_handle;
+   } else {
+      intelScreen->front.bo_handle = -1;
+      intelScreen->back.bo_handle = -1;
+      intelScreen->third.bo_handle = -1;
+      intelScreen->depth.bo_handle = -1;
+   }
+
+   intelScreen->tex.offset = sarea->tex_offset;
+   intelScreen->logTextureGranularity = sarea->log_tex_granularity;
+   intelScreen->tex.handle = sarea->tex_handle;
+   intelScreen->tex.size = sarea->tex_size;
+
+   intelScreen->rotated.offset = sarea->rotated_offset;
+   intelScreen->rotated.pitch = sarea->rotated_pitch * intelScreen->cpp;
+   intelScreen->rotated.size = sarea->rotated_size;
+   intelScreen->current_rotation = sarea->rotation;
+   matrix23Rotate(&intelScreen->rotMatrix,
+                  sarea->width, sarea->height, sarea->rotation);
+   intelScreen->rotatedWidth = sarea->virtualX;
+   intelScreen->rotatedHeight = sarea->virtualY;
+
+   if (0)
+      intelPrintSAREA(sarea);
+}
+
+static const __DRItexOffsetExtension intelTexOffsetExtension = {
+   { __DRI_TEX_OFFSET },
+   intelSetTexOffset,
+};
+
+static const __DRIextension *intelExtensions[] = {
+    &driReadDrawableExtension,
+    &driCopySubBufferExtension.base,
+    &driSwapControlExtension.base,
+    &driFrameTrackingExtension.base,
+    &driMediaStreamCounterExtension.base,
+    &intelTexOffsetExtension.base,
+    NULL
+};
+
+
+static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
+{
+   intelScreenPrivate *intelScreen;
+   I830DRIPtr gDRIPriv = (I830DRIPtr) sPriv->pDevPriv;
+   drmI830Sarea *sarea;
+
+   if (sPriv->devPrivSize != sizeof(I830DRIRec)) {
+      fprintf(stderr,
+              "\nERROR!  sizeof(I830DRIRec) does not match passed size from device driver\n");
+      return GL_FALSE;
+   }
+
+   /* Allocate the private area */
+   intelScreen = (intelScreenPrivate *) CALLOC(sizeof(intelScreenPrivate));
+   if (!intelScreen) {
+      fprintf(stderr, "\nERROR!  Allocating private area failed\n");
+      return GL_FALSE;
+   }
+   /* parse information in __driConfigOptions */
+   driParseOptionInfo(&intelScreen->optionCache,
+                      __driConfigOptions, __driNConfigOptions);
+
+   intelScreen->driScrnPriv = sPriv;
+   sPriv->private = (void *) intelScreen;
+   intelScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+   sarea = (drmI830Sarea *)
+      (((GLubyte *) sPriv->pSAREA) + intelScreen->sarea_priv_offset);
+
+   intelScreen->deviceID = gDRIPriv->deviceID;
+   if (intelScreen->deviceID == PCI_CHIP_I865_G)
+      intelScreen->maxBatchSize = 4096;
+   else
+      intelScreen->maxBatchSize = BATCH_SZ;
+
+   intelScreen->mem = gDRIPriv->mem;
+   intelScreen->cpp = gDRIPriv->cpp;
+
+   switch (gDRIPriv->bitsPerPixel) {
+   case 16:
+      intelScreen->fbFormat = DV_PF_565;
+      break;
+   case 32:
+      intelScreen->fbFormat = DV_PF_8888;
+      break;
+   default:
+      exit(1);
+      break;
+   }
+
+   intelUpdateScreenFromSAREA(intelScreen, sarea);
+
+   if (!intelMapScreenRegions(sPriv)) {
+      fprintf(stderr, "\nERROR!  mapping regions\n");
+      _mesa_free(intelScreen);
+      sPriv->private = NULL;
+      return GL_FALSE;
+   }
+
+   intelScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+
+   if (0)
+      intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv);
+
+   intelScreen->drmMinor = sPriv->drm_version.minor;
+
+   /* Determine if IRQs are active? */
+   {
+      int ret;
+      drmI830GetParam gp;
+
+      gp.param = I830_PARAM_IRQ_ACTIVE;
+      gp.value = &intelScreen->irq_active;
+
+      ret = drmCommandWriteRead(sPriv->fd, DRM_I830_GETPARAM,
+                                &gp, sizeof(gp));
+      if (ret) {
+         fprintf(stderr, "drmI830GetParam: %d\n", ret);
+         return GL_FALSE;
+      }
+   }
+
+   /* Determine if batchbuffers are allowed */
+   {
+      int ret;
+      drmI830GetParam gp;
+
+      gp.param = I830_PARAM_ALLOW_BATCHBUFFER;
+      gp.value = &intelScreen->allow_batchbuffer;
+
+      ret = drmCommandWriteRead(sPriv->fd, DRM_I830_GETPARAM,
+                                &gp, sizeof(gp));
+      if (ret) {
+         fprintf(stderr, "drmI830GetParam: (%d) %d\n", gp.param, ret);
+         return GL_FALSE;
+      }
+   }
+
+   sPriv->extensions = intelExtensions;
+
+   /* If we've got a new enough DDX that's initializing TTM and giving us
+    * object handles for the shared buffers, use that.
+    */
+   intelScreen->ttm = GL_FALSE;
+   if (getenv("INTEL_NO_TTM") == NULL &&
+       intelScreen->driScrnPriv->ddx_version.minor >= 9 &&
+       intelScreen->drmMinor >= 11 &&
+       intelScreen->front.bo_handle != -1) {
+      intelScreen->bufmgr = intel_bufmgr_ttm_init(sPriv->fd,
+						  DRM_FENCE_TYPE_EXE,
+						  DRM_FENCE_TYPE_EXE |
+						  DRM_I915_FENCE_TYPE_RW,
+						  BATCH_SZ);
+      if (intelScreen->bufmgr != NULL)
+	 intelScreen->ttm = GL_TRUE;
+   }
+   /* Otherwise, use the classic buffer manager. */
+   if (intelScreen->bufmgr == NULL) {
+      if (intelScreen->tex.size == 0) {
+	 fprintf(stderr, "[%s:%u] Error initializing buffer manager.\n",
+		 __func__, __LINE__);
+	 return GL_FALSE;
+      }
+      fprintf(stderr, "[%s:%u] Failed to init TTM buffer manager, falling back"
+	      " to classic.\n", __func__, __LINE__);
+      intelScreen->bufmgr = dri_bufmgr_fake_init(intelScreen->tex.offset,
+						 intelScreen->tex.map,
+						 intelScreen->tex.size,
+						 intel_fence_emit,
+						 intel_fence_wait,
+						 intelScreen);
+   }
+
+   intel_recreate_static_regions(intelScreen);
+
+   return GL_TRUE;
+}
+
+
+static void
+intelDestroyScreen(__DRIscreenPrivate * sPriv)
+{
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
+
+   intelUnmapScreenRegions(intelScreen);
+
+   dri_bufmgr_destroy(intelScreen->bufmgr);
+   FREE(intelScreen);
+   sPriv->private = NULL;
+}
+
+
+/**
+ * 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, GLboolean isPixmap)
+{
+   intelScreenPrivate *screen = (intelScreenPrivate *) driScrnPriv->private;
+
+   if (isPixmap) {
+      return GL_FALSE;          /* not implemented */
+   }
+   else {
+      GLboolean swStencil = (mesaVis->stencilBits > 0 &&
+                             mesaVis->depthBits != 24);
+      GLenum rgbFormat = (mesaVis->redBits == 5 ? GL_RGB5 : GL_RGBA8);
+
+      struct intel_framebuffer *intel_fb = CALLOC_STRUCT(intel_framebuffer);
+
+      if (!intel_fb)
+	 return GL_FALSE;
+
+      _mesa_initialize_framebuffer(&intel_fb->Base, mesaVis);
+
+      /* setup the hardware-based renderbuffers */
+      {
+         intel_fb->color_rb[0]
+            = intel_create_renderbuffer(rgbFormat,
+                                        screen->width, screen->height,
+                                        screen->front.offset,
+                                        screen->front.pitch,
+                                        screen->cpp,
+                                        screen->front.map);
+         intel_set_span_functions(&intel_fb->color_rb[0]->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_FRONT_LEFT,
+				&intel_fb->color_rb[0]->Base);
+      }
+
+      if (mesaVis->doubleBufferMode) {
+         intel_fb->color_rb[1]
+            = intel_create_renderbuffer(rgbFormat,
+                                        screen->width, screen->height,
+                                        screen->back.offset,
+                                        screen->back.pitch,
+                                        screen->cpp,
+                                        screen->back.map);
+         intel_set_span_functions(&intel_fb->color_rb[1]->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_BACK_LEFT,
+				&intel_fb->color_rb[1]->Base);
+
+	 if (screen->third.handle) {
+	    struct gl_renderbuffer *tmp_rb = NULL;
+
+	    intel_fb->color_rb[2]
+	       = intel_create_renderbuffer(rgbFormat,
+					   screen->width, screen->height,
+					   screen->third.offset,
+					   screen->third.pitch,
+					   screen->cpp,
+					   screen->third.map);
+	    intel_set_span_functions(&intel_fb->color_rb[2]->Base);
+	    _mesa_reference_renderbuffer(&tmp_rb, &intel_fb->color_rb[2]->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(&intel_fb->Base, BUFFER_DEPTH,
+				&depthStencilRb->Base);
+         _mesa_add_renderbuffer(&intel_fb->Base, BUFFER_STENCIL,
+				&depthStencilRb->Base);
+      }
+      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(&intel_fb->Base, BUFFER_DEPTH, &depthRb->Base);
+      }
+
+      /* now add any/all software-based renderbuffers we may need */
+      _mesa_add_soft_renderbuffers(&intel_fb->Base,
+                                   GL_FALSE, /* never sw color */
+                                   GL_FALSE, /* never sw depth */
+                                   swStencil, mesaVis->accumRedBits > 0,
+                                   GL_FALSE, /* never sw alpha */
+                                   GL_FALSE  /* never sw aux */ );
+      driDrawPriv->driverPrivate = (void *) intel_fb;
+
+      return GL_TRUE;
+   }
+}
+
+static void
+intelDestroyBuffer(__DRIdrawablePrivate * driDrawPriv)
+{
+   _mesa_unreference_framebuffer((GLframebuffer **)(&(driDrawPriv->driverPrivate)));
+}
+
+
+/**
+ * Get information about previous buffer swaps.
+ */
+static int
+intelGetSwapInfo(__DRIdrawablePrivate * dPriv, __DRIswapInfo * sInfo)
+{
+   struct intel_framebuffer *intel_fb;
+
+   if ((dPriv == NULL) || (dPriv->driverPrivate == NULL)
+       || (sInfo == NULL)) {
+      return -1;
+   }
+
+   intel_fb = dPriv->driverPrivate;
+   sInfo->swap_count = intel_fb->swap_count;
+   sInfo->swap_ust = intel_fb->swap_ust;
+   sInfo->swap_missed_count = intel_fb->swap_missed_count;
+
+   sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0)
+      ? driCalculateSwapUsage(dPriv, 0, intel_fb->swap_missed_ust)
+      : 0.0;
+
+   return 0;
+}
+
+
+/* There are probably better ways to do this, such as an
+ * init-designated function to register chipids and createcontext
+ * functions.
+ */
+extern GLboolean i830CreateContext(const __GLcontextModes * mesaVis,
+                                   __DRIcontextPrivate * driContextPriv,
+                                   void *sharedContextPrivate);
+
+extern GLboolean i915CreateContext(const __GLcontextModes * mesaVis,
+                                   __DRIcontextPrivate * driContextPriv,
+                                   void *sharedContextPrivate);
+
+
+
+
+static GLboolean
+intelCreateContext(const __GLcontextModes * mesaVis,
+                   __DRIcontextPrivate * driContextPriv,
+                   void *sharedContextPrivate)
+{
+   __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
+   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:
+   case PCI_CHIP_I865_G:
+      return i830CreateContext(mesaVis, driContextPriv, sharedContextPrivate);
+
+   case PCI_CHIP_I915_G:
+   case PCI_CHIP_I915_GM:
+   case PCI_CHIP_I945_G:
+   case PCI_CHIP_I945_GM:
+   case PCI_CHIP_I945_GME:
+   case PCI_CHIP_G33_G:
+   case PCI_CHIP_Q35_G:
+   case PCI_CHIP_Q33_G:
+      return i915CreateContext(mesaVis, driContextPriv, sharedContextPrivate);
+
+   default:
+      fprintf(stderr, "Unrecognized deviceID %x\n", intelScreen->deviceID);
+      return GL_FALSE;
+   }
+}
+
+
+static const struct __DriverAPIRec intelAPI = {
+   .DestroyScreen = intelDestroyScreen,
+   .CreateContext = intelCreateContext,
+   .DestroyContext = intelDestroyContext,
+   .CreateBuffer = intelCreateBuffer,
+   .DestroyBuffer = intelDestroyBuffer,
+   .SwapBuffers = intelSwapBuffers,
+   .MakeCurrent = intelMakeCurrent,
+   .UnbindContext = intelUnbindContext,
+   .GetSwapInfo = intelGetSwapInfo,
+   .GetMSC = driGetMSC32,
+   .GetDrawableMSC = driDrawableGetMSC32,
+   .WaitForMSC = driWaitForMSC32,
+   .WaitForSBC = NULL,
+   .SwapBuffersMSC = NULL,
+   .CopySubBuffer = intelCopySubBuffer,
+   .setTexOffset = intelSetTexOffset,
+};
+
+
+static __GLcontextModes *
+intelFillInModes(unsigned pixel_bits, unsigned depth_bits,
+                 unsigned stencil_bits, GLboolean have_back_buffer)
+{
+   __GLcontextModes *modes;
+   __GLcontextModes *m;
+   unsigned num_modes;
+   unsigned depth_buffer_factor;
+   unsigned back_buffer_factor;
+   GLenum fb_format;
+   GLenum fb_type;
+
+   /* GLX_SWAP_COPY_OML is only supported because the Intel driver doesn't
+    * support pageflipping at all.
+    */
+   static const GLenum back_buffer_modes[] = {
+      GLX_NONE, GLX_SWAP_UNDEFINED_OML, GLX_SWAP_COPY_OML
+   };
+
+   u_int8_t depth_bits_array[3];
+   u_int8_t stencil_bits_array[3];
+
+
+   depth_bits_array[0] = 0;
+   depth_bits_array[1] = depth_bits;
+   depth_bits_array[2] = depth_bits;
+
+   /* Just like with the accumulation buffer, always provide some modes
+    * with a stencil buffer.  It will be a sw fallback, but some apps won't
+    * care about that.
+    */
+   stencil_bits_array[0] = 0;
+   stencil_bits_array[1] = 0;
+   if (depth_bits == 24)
+      stencil_bits_array[1] = (stencil_bits == 0) ? 8 : stencil_bits;
+
+   stencil_bits_array[2] = (stencil_bits == 0) ? 8 : stencil_bits;
+
+   depth_buffer_factor = ((depth_bits != 0) || (stencil_bits != 0)) ? 3 : 1;
+   back_buffer_factor = (have_back_buffer) ? 3 : 1;
+
+   num_modes = depth_buffer_factor * back_buffer_factor * 4;
+
+   if (pixel_bits == 16) {
+      fb_format = GL_RGB;
+      fb_type = GL_UNSIGNED_SHORT_5_6_5;
+   }
+   else {
+      fb_format = GL_BGRA;
+      fb_type = GL_UNSIGNED_INT_8_8_8_8_REV;
+   }
+
+   modes =
+      (*dri_interface->createContextModes) (num_modes,
+                                            sizeof(__GLcontextModes));
+   m = modes;
+   if (!driFillInModes(&m, fb_format, fb_type,
+                       depth_bits_array, stencil_bits_array,
+                       depth_buffer_factor, back_buffer_modes,
+                       back_buffer_factor, GLX_TRUE_COLOR)) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
+   }
+   if (!driFillInModes(&m, fb_format, fb_type,
+                       depth_bits_array, stencil_bits_array,
+                       depth_buffer_factor, back_buffer_modes,
+                       back_buffer_factor, GLX_DIRECT_COLOR)) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
+   }
+
+   /* Mark the visual as slow if there are "fake" stencil bits.
+    */
+   for (m = modes; m != NULL; m = m->next) {
+      if ((m->stencilBits != 0) && (m->stencilBits != stencil_bits)) {
+         m->visualRating = GLX_SLOW_CONFIG;
+      }
+   }
+
+   return modes;
+}
+
+
+/**
+ * This is the driver specific part of the createNewScreen entry point.
+ * 
+ * \todo maybe fold this into intelInitDriver
+ *
+ * \return the __GLcontextModes supported by this driver
+ */
+PUBLIC __GLcontextModes *__driDriverInitScreen(__DRIscreenPrivate *psp)
+{
+   static const __DRIversion ddx_expected = { 1, 5, 0 };
+   static const __DRIversion dri_expected = { 4, 0, 0 };
+   static const __DRIversion drm_expected = { 1, 5, 0 };
+   I830DRIPtr dri_priv = (I830DRIPtr) psp->pDevPriv;
+
+   psp->DriverAPI = intelAPI;
+
+   if (!driCheckDriDdxDrmVersions2("i915",
+                                   &psp->dri_version, &dri_expected,
+                                   &psp->ddx_version, &ddx_expected,
+                                   &psp->drm_version, &drm_expected)) {
+      return NULL;
+   }
+
+   /* Calling driInitExtensions here, with a NULL context pointer,
+    * does not actually enable the extensions.  It just makes sure
+    * that all the dispatch offsets for all the extensions that
+    * *might* be enables are known.  This is needed because the
+    * dispatch offsets need to be known when _mesa_context_create is
+    * called, but we can't enable the extensions until we have a
+    * context pointer.
+    *
+    * Hello chicken.  Hello egg.  How are you two today?
+    */
+   driInitExtensions(NULL, card_extensions, GL_FALSE);
+   driInitExtensions(NULL, ttm_extensions, GL_FALSE);
+
+   if (!intelInitDriver(psp))
+       return NULL;
+
+   return intelFillInModes(dri_priv->cpp * 8,
+			   (dri_priv->cpp == 2) ? 16 : 24,
+			   (dri_priv->cpp == 2) ? 0  : 8, 1);
+}
+
+struct intel_context *intelScreenContext(intelScreenPrivate *intelScreen)
+{
+  /*
+   * This should probably change to have the screen allocate a dummy
+   * context at screen creation. For now just use the current context.
+   */
+
+  GET_CURRENT_CONTEXT(ctx);
+  if (ctx == NULL) {
+     _mesa_problem(NULL, "No current context in intelScreenContext\n");
+     return NULL;
+  }
+  return intel_context(ctx);
+}
+
diff --git a/src/mesa/drivers/dri/intel/intel_screen.h b/src/mesa/drivers/dri/intel/intel_screen.h
new file mode 100644
index 00000000000..aa0ef2c5090
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_screen.h
@@ -0,0 +1,134 @@
+/**************************************************************************
+ * 
+ * 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_INIT_H_
+#define _INTEL_INIT_H_
+
+#include <sys/time.h>
+#include "dri_util.h"
+#include "intel_rotate.h"
+#include "i830_common.h"
+#include "xmlconfig.h"
+#include "dri_bufmgr.h"
+
+/* XXX: change name or eliminate to avoid conflict with "struct
+ * intel_region"!!!
+ */
+typedef struct
+{
+   drm_handle_t handle;
+   drmSize size;                /* region size in bytes */
+   char *map;                   /* memory map */
+   int offset;                  /* from start of video mem, in bytes */
+   int pitch;                   /* row stride, in bytes */
+   unsigned int bo_handle;	/* buffer object id if available, or -1 */
+} intelRegion;
+
+typedef struct
+{
+   intelRegion front;
+   intelRegion back;
+   intelRegion third;
+   intelRegion rotated;
+   intelRegion depth;
+   intelRegion tex;
+
+   struct intel_region *front_region;
+   struct intel_region *back_region;
+   struct intel_region *third_region;
+   struct intel_region *depth_region;
+   struct intel_region *rotated_region;
+
+   int deviceID;
+   int width;
+   int height;
+   int mem;                     /* unused */
+
+   int cpp;                     /* for front and back buffers */
+/*    int bitsPerPixel;   */
+   int fbFormat;                /* XXX FBO: this is obsolete - remove after i830 updates */
+
+   int logTextureGranularity;
+
+   __DRIscreenPrivate *driScrnPriv;
+   unsigned int sarea_priv_offset;
+
+   int drmMinor;
+
+   int irq_active;
+   int allow_batchbuffer;
+
+   struct matrix23 rotMatrix;
+
+   int current_rotation;        /* 0, 90, 180 or 270 */
+   int rotatedWidth, rotatedHeight;
+
+   /**
+   * Configuration cache with default values for all contexts
+   */
+   driOptionCache optionCache;
+
+   dri_bufmgr *bufmgr;
+   unsigned int maxBatchSize;
+
+   /**
+    * This value indicates that the kernel memory manager is being used
+    * instead of the fake client-side memory manager.
+    */
+   GLboolean ttm;
+} intelScreenPrivate;
+
+
+
+extern GLboolean intelMapScreenRegions(__DRIscreenPrivate * sPriv);
+
+extern void intelUnmapScreenRegions(intelScreenPrivate * intelScreen);
+
+extern void
+intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen,
+                           drmI830Sarea * sarea);
+
+extern void intelDestroyContext(__DRIcontextPrivate * driContextPriv);
+
+extern GLboolean intelUnbindContext(__DRIcontextPrivate * driContextPriv);
+
+extern GLboolean
+intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
+                 __DRIdrawablePrivate * driDrawPriv,
+                 __DRIdrawablePrivate * driReadPriv);
+
+extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
+
+extern void
+intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h);
+
+extern struct intel_context *intelScreenContext(intelScreenPrivate *intelScreen);
+
+extern void
+intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea);
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_span.c b/src/mesa/drivers/dri/intel/intel_span.c
new file mode 100644
index 00000000000..d1f8ef06beb
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_span.c
@@ -0,0 +1,409 @@
+/**************************************************************************
+ * 
+ * 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 "macros.h"
+#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							\
+   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 Y_FLIP(_y) ((_y) * yScale + yBias)
+
+#define HW_LOCK()
+
+#define HW_UNLOCK()
+
+/* 16 bit, RGB565 color spanline and pixel 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 SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
+
+#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 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;
+
+
+#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
+
+/**
+ ** 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 TAG(x) intel##x##_z16
+#include "depthtmp.h"
+
+
+/**
+ ** 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;			\
+}
+
+/* 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"
+
+
+/**
+ ** 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) + (_y) * pitch] >> 24;
+
+#define TAG(x) intel##x##_z24_s8
+#include "stenciltmp.h"
+
+
+
+/**
+ * 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->intelScreen, irb->region);
+               else
+                  intel_region_unmap(intel->intelScreen, 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->intelScreen, irb->region);
+      else
+         intel_region_unmap(intel->intelScreen, 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->intelScreen, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel->intelScreen, 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->intelScreen, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel->intelScreen, 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)
+{
+   struct intel_context *intel = intel_context(ctx);
+   GLuint i;
+
+   intelFinish(&intel->ctx);
+   LOCK_HARDWARE(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->intelScreen, intel->front_region);
+   intel_region_map(intel->intelScreen, intel->back_region);
+   intel_region_map(intel->intelScreen, intel->intelScreen->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)
+{
+   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->intelScreen->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);
+   swdd->SpanRenderStart = intelSpanRenderStart;
+   swdd->SpanRenderFinish = intelSpanRenderFinish;
+}
+
+
+/**
+ * Plug in appropriate span read/write functions for the given renderbuffer.
+ * These are used for the software fallbacks.
+ */
+void
+intel_set_span_functions(struct gl_renderbuffer *rb)
+{
+   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 (rb->_ActualFormat == GL_DEPTH_COMPONENT24 ||        /* XXX FBO remove */
+            rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT) {
+      intelInitDepthPointers_z24_s8(rb);
+   }
+   else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) {       /* XXX FBO remove */
+      intelInitStencilPointers_z24_s8(rb);
+   }
+   else {
+      _mesa_problem(NULL,
+                    "Unexpected _ActualFormat in intelSetSpanFunctions");
+   }
+}
diff --git a/src/mesa/drivers/dri/intel/intel_span.h b/src/mesa/drivers/dri/intel/intel_span.h
new file mode 100644
index 00000000000..5201f6d6c6e
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_span.h
@@ -0,0 +1,38 @@
+/**************************************************************************
+ * 
+ * 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_SPAN_H
+#define _INTEL_SPAN_H
+
+extern void intelInitSpanFuncs(GLcontext * ctx);
+
+extern void intelSpanRenderFinish(GLcontext * ctx);
+extern void intelSpanRenderStart(GLcontext * ctx);
+
+extern void intel_set_span_functions(struct gl_renderbuffer *rb);
+
+#endif
diff --git a/src/mesa/drivers/dri/intel/intel_tex.c b/src/mesa/drivers/dri/intel/intel_tex.c
new file mode 100644
index 00000000000..b08dee43bc1
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_tex.c
@@ -0,0 +1,192 @@
+#include "texobj.h"
+#include "intel_context.h"
+#include "intel_mipmap_tree.h"
+#include "intel_tex.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+static GLboolean
+intelIsTextureResident(GLcontext * ctx, struct gl_texture_object *texObj)
+{
+#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 struct gl_texture_image *
+intelNewTextureImage(GLcontext * ctx)
+{
+   DBG("%s\n", __FUNCTION__);
+   (void) ctx;
+   return (struct gl_texture_image *) CALLOC_STRUCT(intel_texture_image);
+}
+
+
+static struct gl_texture_object *
+intelNewTextureObject(GLcontext * ctx, GLuint name, GLenum target)
+{
+   struct intel_texture_object *obj = CALLOC_STRUCT(intel_texture_object);
+
+   DBG("%s\n", __FUNCTION__);
+   _mesa_initialize_texture_object(&obj->base, name, target);
+
+   return &obj->base;
+}
+
+static void 
+intelDeleteTextureObject(GLcontext *ctx,
+			 struct gl_texture_object *texObj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_object *intelObj = intel_texture_object(texObj);
+
+   if (intelObj->mt)
+      intel_miptree_release(intel, &intelObj->mt);
+
+   _mesa_delete_texture_object(ctx, texObj);
+}
+
+
+static void
+intelFreeTextureImageData(GLcontext * ctx, struct gl_texture_image *texImage)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intelImage->mt) {
+      intel_miptree_release(intel, &intelImage->mt);
+   }
+
+   if (texImage->Data) {
+      free(texImage->Data);
+      texImage->Data = NULL;
+   }
+}
+
+
+/* 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 *
+do_memcpy(void *dest, const void *src, size_t n)
+{
+   if ((((unsigned) src) & 63) || (((unsigned) dest) & 63)) {
+      return __memcpy(dest, src, n);
+   }
+   else
+      return memcpy(dest, src, n);
+}
+
+
+#if DO_DEBUG
+
+#ifndef __x86_64__
+static unsigned
+fastrdtsc(void)
+{
+   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");
+
+   return eax;
+}
+#else
+static unsigned
+fastrdtsc(void)
+{
+   unsigned eax;
+   __asm__ volatile ("\t" "cpuid\n\t" ".byte 0x0f, 0x31\n\t":"=a" (eax)
+                     :"0"(0)
+                     :"ecx", "edx", "ebx", "cc");
+
+   return eax;
+}
+#endif
+
+static unsigned
+time_diff(unsigned t, unsigned t2)
+{
+   return ((t < t2) ? t2 - t : 0xFFFFFFFFU - (t - t2 - 1));
+}
+
+
+static void *
+timed_memcpy(void *dest, const void *src, size_t n)
+{
+   void *ret;
+   unsigned t1, t2;
+   double rate;
+
+   if ((((unsigned) src) & 63) || (((unsigned) dest) & 63))
+      _mesa_printf("Warning - non-aligned texture copy!\n");
+
+   t1 = fastrdtsc();
+   ret = do_memcpy(dest, src, n);
+   t2 = fastrdtsc();
+
+   rate = time_diff(t1, t2);
+   rate /= (double) n;
+   _mesa_printf("timed_memcpy: %u %u --> %f clocks/byte\n", t1, t2, rate);
+   return ret;
+}
+#endif /* DO_DEBUG */
+
+
+void
+intelInitTextureFuncs(struct dd_function_table *functions)
+{
+   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;
+
+   /* compressed texture functions */
+   functions->CompressedTexImage2D = intelCompressedTexImage2D;
+   functions->GetCompressedTexImage = intelGetCompressedTexImage;
+
+   functions->NewTextureObject = intelNewTextureObject;
+   functions->NewTextureImage = intelNewTextureImage;
+   functions->DeleteTexture = intelDeleteTextureObject;
+   functions->FreeTexImageData = intelFreeTextureImageData;
+   functions->UpdateTexturePalette = 0;
+   functions->IsTextureResident = intelIsTextureResident;
+
+#if DO_DEBUG
+   if (INTEL_DEBUG & DEBUG_BUFMGR)
+      functions->TextureMemCpy = timed_memcpy;
+   else
+#endif
+      functions->TextureMemCpy = do_memcpy;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_tex_copy.c b/src/mesa/drivers/dri/intel/intel_tex_copy.c
new file mode 100644
index 00000000000..b85a25642a2
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_tex_copy.c
@@ -0,0 +1,302 @@
+/**************************************************************************
+ * 
+ * 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 "mtypes.h"
+#include "enums.h"
+#include "image.h"
+#include "teximage.h"
+#include "swrast/swrast.h"
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_batchbuffer.h"
+#include "intel_buffers.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
+#include "intel_fbo.h"
+#include "intel_tex.h"
+#include "intel_blit.h"
+#include "intel_pixel.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+/**
+ * Get the intel_region which is the source for any glCopyTex[Sub]Image call.
+ *
+ * Do the best we can using the blitter.  A future project is to use
+ * the texture engine and fragment programs for these copies.
+ */
+static const struct intel_region *
+get_teximage_source(struct intel_context *intel, GLenum internalFormat)
+{
+   struct intel_renderbuffer *irb;
+
+   DBG("%s %s\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(internalFormat));
+
+   switch (internalFormat) {
+   case GL_DEPTH_COMPONENT:
+   case GL_DEPTH_COMPONENT16_ARB:
+      irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH);
+      if (irb && irb->region && irb->region->cpp == 2)
+         return irb->region;
+      return NULL;
+   case GL_DEPTH24_STENCIL8_EXT:
+   case GL_DEPTH_STENCIL_EXT:
+      irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH);
+      if (irb && irb->region && irb->region->cpp == 4)
+         return irb->region;
+      return NULL;
+   case GL_RGBA:
+   case GL_RGBA8:
+      return intel_readbuf_region(intel);
+   case GL_RGB:
+      if (intel->intelScreen->cpp == 2)
+         return intel_readbuf_region(intel);
+      return NULL;
+   default:
+      return NULL;
+   }
+}
+
+
+static GLboolean
+do_copy_texsubimage(struct intel_context *intel,
+                    struct intel_texture_image *intelImage,
+                    GLenum internalFormat,
+                    GLint dstx, GLint dsty,
+                    GLint x, GLint y, GLsizei width, GLsizei height)
+{
+   GLcontext *ctx = &intel->ctx;
+   const struct intel_region *src =
+      get_teximage_source(intel, internalFormat);
+
+   if (!intelImage->mt || !src) {
+      DBG("%s fail %p %p\n", __FUNCTION__, intelImage->mt, src);
+      return GL_FALSE;
+   }
+
+   intelFlush(ctx);
+   LOCK_HARDWARE(intel);
+   {
+      GLuint image_offset = intel_miptree_image_offset(intelImage->mt,
+                                                       intelImage->face,
+                                                       intelImage->level);
+      const GLint orig_x = x;
+      const GLint orig_y = y;
+      const struct gl_framebuffer *fb = ctx->DrawBuffer;
+
+      if (_mesa_clip_to_region(fb->_Xmin, fb->_Ymin, fb->_Xmax, fb->_Ymax,
+                               &x, &y, &width, &height)) {
+         /* Update dst for clipped src.  Need to also clip the source rect.
+          */
+         dstx += x - orig_x;
+         dsty += y - orig_y;
+
+         if (ctx->ReadBuffer->Name == 0) {
+            /* reading from a window, adjust x, y */
+            __DRIdrawablePrivate *dPriv = intel->driDrawable;
+            GLuint window_y;
+            /* window_y = position of window on screen if y=0=bottom */
+            window_y = intel->intelScreen->height - (dPriv->y + dPriv->h);
+            y = window_y + y;
+            x += dPriv->x;
+         }
+         else {
+            /* reading from a FBO */
+            /* invert Y */
+            y = ctx->ReadBuffer->Height - y - 1;
+         }
+
+
+         /* A bit of fiddling to get the blitter to work with -ve
+          * pitches.  But we get a nice inverted blit this way, so it's
+          * worth it:
+          */
+         intelEmitCopyBlit(intel,
+                           intelImage->mt->cpp,
+                           -src->pitch,
+                           src->buffer,
+                           src->height * src->pitch * src->cpp,
+                           intelImage->mt->pitch,
+                           intelImage->mt->region->buffer,
+                           image_offset,
+                           x, y + height, dstx, dsty, width, height,
+			   GL_COPY); /* ? */
+
+         intel_batchbuffer_flush(intel->batch);
+      }
+   }
+
+
+   UNLOCK_HARDWARE(intel);
+
+#if 0
+   /* GL_SGIS_generate_mipmap -- this can be accelerated now.
+    * XXX Add a ctx->Driver.GenerateMipmaps() function?
+    */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      intel_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+#endif
+
+   return GL_TRUE;
+}
+
+
+
+
+
+void
+intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level,
+                    GLenum internalFormat,
+                    GLint x, GLint y, GLsizei width, GLint border)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_object *texObj =
+      _mesa_select_tex_object(ctx, texUnit, target);
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texObj, target, level);
+
+   if (border)
+      goto fail;
+
+   /* Setup or redefine the texture object, mipmap tree and texture
+    * image.  Don't populate yet.  
+    */
+   ctx->Driver.TexImage1D(ctx, target, level, internalFormat,
+                          width, border,
+                          GL_RGBA, CHAN_TYPE, NULL,
+                          &ctx->DefaultPacking, texObj, texImage);
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat, 0, 0, x, y, width, 1))
+      goto fail;
+
+   return;
+
+ fail:
+   _swrast_copy_teximage1d(ctx, target, level, internalFormat, x, y,
+                           width, border);
+}
+
+void
+intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level,
+                    GLenum internalFormat,
+                    GLint x, GLint y, GLsizei width, GLsizei height,
+                    GLint border)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_object *texObj =
+      _mesa_select_tex_object(ctx, texUnit, target);
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texObj, target, level);
+
+   if (border)
+      goto fail;
+
+   /* Setup or redefine the texture object, mipmap tree and texture
+    * image.  Don't populate yet.  
+    */
+   ctx->Driver.TexImage2D(ctx, target, level, internalFormat,
+                          width, height, border,
+                          GL_RGBA, CHAN_TYPE, NULL,
+                          &ctx->DefaultPacking, texObj, texImage);
+
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat, 0, 0, x, y, width, height))
+      goto fail;
+
+   return;
+
+ fail:
+   _swrast_copy_teximage2d(ctx, target, level, internalFormat, x, y,
+                           width, height, border);
+}
+
+
+void
+intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
+                       GLint xoffset, GLint x, GLint y, GLsizei width)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_object *texObj =
+      _mesa_select_tex_object(ctx, texUnit, target);
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texObj, target, level);
+   GLenum internalFormat = texImage->InternalFormat;
+
+   /* XXX need to check <border> as in above function? */
+
+   /* Need to check texture is compatible with source format. 
+    */
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat, xoffset, 0, x, y, width, 1)) {
+      _swrast_copy_texsubimage1d(ctx, target, level, xoffset, x, y, width);
+   }
+}
+
+
+
+void
+intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
+                       GLint xoffset, GLint yoffset,
+                       GLint x, GLint y, GLsizei width, GLsizei height)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_object *texObj =
+      _mesa_select_tex_object(ctx, texUnit, target);
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texObj, target, level);
+   GLenum internalFormat = texImage->InternalFormat;
+
+
+   /* Need to check texture is compatible with source format. 
+    */
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat,
+                            xoffset, yoffset, x, y, width, height)) {
+
+      DBG("%s - fallback to swrast\n", __FUNCTION__);
+
+      _swrast_copy_texsubimage2d(ctx, target, level,
+                                 xoffset, yoffset, x, y, width, height);
+   }
+}
diff --git a/src/mesa/drivers/dri/intel/intel_tex_format.c b/src/mesa/drivers/dri/intel/intel_tex_format.c
new file mode 100644
index 00000000000..6e058dff69f
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_tex_format.c
@@ -0,0 +1,172 @@
+#include "intel_context.h"
+#include "intel_tex.h"
+#include "texformat.h"
+#include "enums.h"
+
+/* It works out that this function is fine for all the supported
+ * hardware.  However, there is still a need to map the formats onto
+ * hardware descriptors.
+ */
+/* Note that the i915 can actually support many more formats than
+ * these if we take the step of simply swizzling the colors
+ * immediately after sampling...
+ */
+const struct gl_texture_format *
+intelChooseTextureFormat(GLcontext * ctx, GLint internalFormat,
+                         GLenum format, GLenum type)
+{
+   struct intel_context *intel = intel_context(ctx);
+   const GLboolean do32bpt = (intel->intelScreen->cpp == 4);
+
+   switch (internalFormat) {
+   case 4:
+   case GL_RGBA:
+   case GL_COMPRESSED_RGBA:
+      if (format == GL_BGRA) {
+         if (type == GL_UNSIGNED_BYTE || 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 &_mesa_texformat_argb8888;
+
+   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;
+
+   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;
+
+   case GL_DEPTH_STENCIL_EXT:
+   case GL_DEPTH24_STENCIL8_EXT:
+      return &_mesa_texformat_z24_s8;
+
+   default:
+      fprintf(stderr, "unexpected texture format %s in %s\n",
+              _mesa_lookup_enum_by_nr(internalFormat), __FUNCTION__);
+      return NULL;
+   }
+
+   return NULL;                 /* never get here */
+}
+
+int intel_compressed_num_bytes(GLuint mesaFormat)
+{
+   int bytes = 0;
+   switch(mesaFormat) {
+     
+   case MESA_FORMAT_RGB_FXT1:
+   case MESA_FORMAT_RGBA_FXT1:
+   case MESA_FORMAT_RGB_DXT1:
+   case MESA_FORMAT_RGBA_DXT1:
+     bytes = 2;
+     break;
+     
+   case MESA_FORMAT_RGBA_DXT3:
+   case MESA_FORMAT_RGBA_DXT5:
+     bytes = 4;
+   default:
+     break;
+   }
+   
+   return bytes;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_tex_image.c b/src/mesa/drivers/dri/intel/intel_tex_image.c
new file mode 100644
index 00000000000..197cf35ebe6
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_tex_image.c
@@ -0,0 +1,690 @@
+
+#include <stdlib.h>
+#include <stdio.h>
+
+#include "glheader.h"
+#include "macros.h"
+#include "mtypes.h"
+#include "enums.h"
+#include "colortab.h"
+#include "convolve.h"
+#include "context.h"
+#include "simple_list.h"
+#include "texcompress.h"
+#include "texformat.h"
+#include "texobj.h"
+#include "texstore.h"
+
+#include "intel_context.h"
+#include "intel_mipmap_tree.h"
+#include "intel_buffer_objects.h"
+#include "intel_batchbuffer.h"
+#include "intel_tex.h"
+#include "intel_ioctl.h"
+#include "intel_blit.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+/* Functions to store texture images.  Where possible, mipmap_tree's
+ * will be created or further instantiated with image data, otherwise
+ * images will be stored in malloc'd memory.  A validation step is
+ * required to pull those images into a mipmap tree, or otherwise
+ * decide a fallback is required.
+ */
+
+
+static int
+logbase2(int n)
+{
+   GLint i = 1;
+   GLint log2 = 0;
+
+   while (n > i) {
+      i *= 2;
+      log2++;
+   }
+
+   return log2;
+}
+
+
+/* Otherwise, store it in memory if (Border != 0) or (any dimension ==
+ * 1).
+ *    
+ * Otherwise, if max_level >= level >= min_level, create tree with
+ * space for textures from min_level down to max_level.
+ *
+ * Otherwise, create tree with space for textures from (level
+ * 0)..(1x1).  Consider pruning this tree at a validation if the
+ * saving is worth it.
+ */
+static void
+guess_and_alloc_mipmap_tree(struct intel_context *intel,
+                            struct intel_texture_object *intelObj,
+                            struct intel_texture_image *intelImage)
+{
+   GLuint firstLevel;
+   GLuint lastLevel;
+   GLuint width = intelImage->base.Width;
+   GLuint height = intelImage->base.Height;
+   GLuint depth = intelImage->base.Depth;
+   GLuint l2width, l2height, l2depth;
+   GLuint i, comp_byte = 0;
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intelImage->base.Border)
+      return;
+
+   if (intelImage->level > intelObj->base.BaseLevel &&
+       (intelImage->base.Width == 1 ||
+        (intelObj->base.Target != GL_TEXTURE_1D &&
+         intelImage->base.Height == 1) ||
+        (intelObj->base.Target == GL_TEXTURE_3D &&
+         intelImage->base.Depth == 1)))
+      return;
+
+   /* If this image disrespects BaseLevel, allocate from level zero.
+    * Usually BaseLevel == 0, so it's unlikely to happen.
+    */
+   if (intelImage->level < intelObj->base.BaseLevel)
+      firstLevel = 0;
+   else
+      firstLevel = intelObj->base.BaseLevel;
+
+
+   /* Figure out image dimensions at start level. 
+    */
+   for (i = intelImage->level; i > firstLevel; i--) {
+      width <<= 1;
+      if (height != 1)
+         height <<= 1;
+      if (depth != 1)
+         depth <<= 1;
+   }
+
+   /* Guess a reasonable value for lastLevel.  This is probably going
+    * to be wrong fairly often and might mean that we have to look at
+    * resizable buffers, or require that buffers implement lazy
+    * pagetable arrangements.
+    */
+   if ((intelObj->base.MinFilter == GL_NEAREST ||
+        intelObj->base.MinFilter == GL_LINEAR) &&
+       intelImage->level == firstLevel) {
+      lastLevel = firstLevel;
+   }
+   else {
+      l2width = logbase2(width);
+      l2height = logbase2(height);
+      l2depth = logbase2(depth);
+      lastLevel = firstLevel + MAX2(MAX2(l2width, l2height), l2depth);
+   }
+
+   assert(!intelObj->mt);
+   if (intelImage->base.IsCompressed)
+      comp_byte = intel_compressed_num_bytes(intelImage->base.TexFormat->MesaFormat);
+   intelObj->mt = intel_miptree_create(intel,
+                                       intelObj->base.Target,
+                                       intelImage->base.InternalFormat,
+                                       firstLevel,
+                                       lastLevel,
+                                       width,
+                                       height,
+                                       depth,
+                                       intelImage->base.TexFormat->TexelBytes,
+                                       comp_byte);
+
+   DBG("%s - success\n", __FUNCTION__);
+}
+
+
+
+
+static GLuint
+target_to_face(GLenum target)
+{
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB:
+      return ((GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X);
+   default:
+      return 0;
+   }
+}
+
+/* There are actually quite a few combinations this will work for,
+ * more than what I've listed here.
+ */
+static GLboolean
+check_pbo_format(GLint internalFormat,
+                 GLenum format, GLenum type,
+                 const struct gl_texture_format *mesa_format)
+{
+   switch (internalFormat) {
+   case 4:
+   case GL_RGBA:
+      return (format == GL_BGRA &&
+              (type == GL_UNSIGNED_BYTE ||
+               type == GL_UNSIGNED_INT_8_8_8_8_REV) &&
+              mesa_format == &_mesa_texformat_argb8888);
+   case 3:
+   case GL_RGB:
+      return (format == GL_RGB &&
+              type == GL_UNSIGNED_SHORT_5_6_5 &&
+              mesa_format == &_mesa_texformat_rgb565);
+   case GL_YCBCR_MESA:
+      return (type == GL_UNSIGNED_SHORT_8_8_MESA || type == GL_UNSIGNED_BYTE);
+   default:
+      return GL_FALSE;
+   }
+}
+
+
+/* XXX: Do this for TexSubImage also:
+ */
+static GLboolean
+try_pbo_upload(struct intel_context *intel,
+               struct intel_texture_image *intelImage,
+               const struct gl_pixelstore_attrib *unpack,
+               GLint internalFormat,
+               GLint width, GLint height,
+               GLenum format, GLenum type, const void *pixels)
+{
+   struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
+   GLuint src_offset, src_stride;
+   GLuint dst_offset, dst_stride;
+
+   if (!pbo ||
+       intel->ctx._ImageTransferState ||
+       unpack->SkipPixels || unpack->SkipRows) {
+      _mesa_printf("%s: failure 1\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   src_offset = (GLuint) pixels;
+
+   if (unpack->RowLength > 0)
+      src_stride = unpack->RowLength;
+   else
+      src_stride = width;
+
+   dst_offset = intel_miptree_image_offset(intelImage->mt,
+                                           intelImage->face,
+                                           intelImage->level);
+
+   dst_stride = intelImage->mt->pitch;
+
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+   {
+      dri_bo *src_buffer = intel_bufferobj_buffer(intel, pbo, INTEL_READ);
+      dri_bo *dst_buffer = intel_region_buffer(intel->intelScreen,
+					       intelImage->mt->region,
+					       INTEL_WRITE_FULL);
+
+
+      intelEmitCopyBlit(intel,
+                        intelImage->mt->cpp,
+                        src_stride, src_buffer, src_offset,
+                        dst_stride, dst_buffer, dst_offset,
+                        0, 0, 0, 0, width, height,
+			GL_COPY);
+
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(intel);
+
+   return GL_TRUE;
+}
+
+
+
+static GLboolean
+try_pbo_zcopy(struct intel_context *intel,
+              struct intel_texture_image *intelImage,
+              const struct gl_pixelstore_attrib *unpack,
+              GLint internalFormat,
+              GLint width, GLint height,
+              GLenum format, GLenum type, const void *pixels)
+{
+   struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
+   GLuint src_offset, src_stride;
+   GLuint dst_offset, dst_stride;
+
+   if (!pbo ||
+       intel->ctx._ImageTransferState ||
+       unpack->SkipPixels || unpack->SkipRows) {
+      _mesa_printf("%s: failure 1\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   src_offset = (GLuint) pixels;
+
+   if (unpack->RowLength > 0)
+      src_stride = unpack->RowLength;
+   else
+      src_stride = width;
+
+   dst_offset = intel_miptree_image_offset(intelImage->mt,
+                                           intelImage->face,
+                                           intelImage->level);
+
+   dst_stride = intelImage->mt->pitch;
+
+   if (src_stride != dst_stride || dst_offset != 0 || src_offset != 0) {
+      _mesa_printf("%s: failure 2\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   intel_region_attach_pbo(intel->intelScreen, intelImage->mt->region, pbo);
+
+   return GL_TRUE;
+}
+
+
+
+
+
+
+static void
+intelTexImage(GLcontext * ctx,
+              GLint dims,
+              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 *unpack,
+              struct gl_texture_object *texObj,
+              struct gl_texture_image *texImage, GLsizei imageSize, int compressed)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_object *intelObj = intel_texture_object(texObj);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+   GLint postConvWidth = width;
+   GLint postConvHeight = height;
+   GLint texelBytes, sizeInBytes;
+   GLuint dstRowStride;
+
+
+   DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(target), level, width, height, depth, border);
+
+   intelFlush(ctx);
+
+   intelImage->face = target_to_face(target);
+   intelImage->level = level;
+
+   if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) {
+      _mesa_adjust_image_for_convolution(ctx, dims, &postConvWidth,
+                                         &postConvHeight);
+   }
+
+   /* choose the texture format */
+   texImage->TexFormat = intelChooseTextureFormat(ctx, internalFormat,
+                                                  format, type);
+
+   _mesa_set_fetch_functions(texImage, dims);
+
+   if (texImage->TexFormat->TexelBytes == 0) {
+      /* must be a compressed format */
+      texelBytes = 0;
+      texImage->IsCompressed = GL_TRUE;
+      texImage->CompressedSize =
+	 ctx->Driver.CompressedTextureSize(ctx, texImage->Width,
+					   texImage->Height, texImage->Depth,
+					   texImage->TexFormat->MesaFormat);
+   } else {
+      texelBytes = texImage->TexFormat->TexelBytes;
+      
+      /* Minimum pitch of 32 bytes */
+      if (postConvWidth * texelBytes < 32) {
+	 postConvWidth = 32 / texelBytes;
+	 texImage->RowStride = postConvWidth;
+      }
+      
+      assert(texImage->RowStride == postConvWidth);
+   }
+
+   /* Release the reference to a potentially orphaned buffer.   
+    * Release any old malloced memory.
+    */
+   if (intelImage->mt) {
+      intel_miptree_release(intel, &intelImage->mt);
+      assert(!texImage->Data);
+   }
+   else if (texImage->Data) {
+      _mesa_align_free(texImage->Data);
+   }
+
+   /* If this is the only texture image in the tree, could call
+    * bmBufferData with NULL data to free the old block and avoid
+    * waiting on any outstanding fences.
+    */
+   if (intelObj->mt &&
+       intelObj->mt->first_level == level &&
+       intelObj->mt->last_level == level &&
+       intelObj->mt->target != GL_TEXTURE_CUBE_MAP_ARB &&
+       !intel_miptree_match_image(intelObj->mt, &intelImage->base,
+                                  intelImage->face, intelImage->level)) {
+
+      DBG("release it\n");
+      intel_miptree_release(intel, &intelObj->mt);
+      assert(!intelObj->mt);
+   }
+
+   if (!intelObj->mt) {
+      guess_and_alloc_mipmap_tree(intel, intelObj, intelImage);
+      if (!intelObj->mt) {
+	 DBG("guess_and_alloc_mipmap_tree: failed\n");
+      }
+   }
+
+   assert(!intelImage->mt);
+
+   if (intelObj->mt &&
+       intel_miptree_match_image(intelObj->mt, &intelImage->base,
+                                 intelImage->face, intelImage->level)) {
+
+      intel_miptree_reference(&intelImage->mt, intelObj->mt);
+      assert(intelImage->mt);
+   }
+
+   if (!intelImage->mt)
+      DBG("XXX: Image did not fit into tree - storing in local memory!\n");
+
+   /* PBO fastpaths:
+    */
+   if (dims <= 2 &&
+       intelImage->mt &&
+       intel_buffer_object(unpack->BufferObj) &&
+       check_pbo_format(internalFormat, format,
+                        type, intelImage->base.TexFormat)) {
+
+      DBG("trying pbo upload\n");
+
+      /* Attempt to texture directly from PBO data (zero copy upload).
+       *
+       * Currently disable as it can lead to worse as well as better
+       * performance (in particular when intel_region_cow() is
+       * required).
+       */
+      if (intelObj->mt == intelImage->mt &&
+          intelObj->mt->first_level == level &&
+          intelObj->mt->last_level == level) {
+
+         if (try_pbo_zcopy(intel, intelImage, unpack,
+                           internalFormat,
+                           width, height, format, type, pixels)) {
+
+            DBG("pbo zcopy upload succeeded\n");
+            return;
+         }
+      }
+
+
+      /* Otherwise, attempt to use the blitter for PBO image uploads.
+       */
+      if (try_pbo_upload(intel, intelImage, unpack,
+                         internalFormat,
+                         width, height, format, type, pixels)) {
+         DBG("pbo upload succeeded\n");
+         return;
+      }
+
+      DBG("pbo upload failed\n");
+   }
+
+
+
+   /* intelCopyTexImage calls this function with pixels == NULL, with
+    * the expectation that the mipmap tree will be set up but nothing
+    * more will be done.  This is where those calls return:
+    */
+   if (compressed) {
+      pixels = _mesa_validate_pbo_compressed_teximage(ctx, imageSize, pixels,
+						      unpack,
+						      "glCompressedTexImage");
+   } else {
+      pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, 1,
+					   format, type,
+					   pixels, unpack, "glTexImage");
+   }
+   if (!pixels)
+      return;
+
+
+   if (intelImage->mt)
+      intel_region_idle(intel->intelScreen, intelImage->mt->region);
+
+   LOCK_HARDWARE(intel);
+
+   if (intelImage->mt) {
+      texImage->Data = intel_miptree_image_map(intel,
+                                               intelImage->mt,
+                                               intelImage->face,
+                                               intelImage->level,
+                                               &dstRowStride,
+                                               intelImage->base.ImageOffsets);
+   }
+   else {
+      /* Allocate regular memory and store the image there temporarily.   */
+      if (texImage->IsCompressed) {
+         sizeInBytes = texImage->CompressedSize;
+         dstRowStride =
+            _mesa_compressed_row_stride(texImage->TexFormat->MesaFormat, width);
+         assert(dims != 3);
+      }
+      else {
+         dstRowStride = postConvWidth * texelBytes;
+         sizeInBytes = depth * dstRowStride * postConvHeight;
+      }
+
+      texImage->Data = malloc(sizeInBytes);
+   }
+
+   DBG("Upload image %dx%dx%d row_len %x "
+       "pitch %x\n",
+       width, height, depth, width * texelBytes, dstRowStride);
+
+   /* Copy data.  Would like to know when it's ok for us to eg. use
+    * the blitter to copy.  Or, use the hardware to do the format
+    * conversion and copy:
+    */
+   if (compressed) {
+     memcpy(texImage->Data, pixels, imageSize);
+   } else if (!texImage->TexFormat->StoreImage(ctx, dims, 
+					       texImage->_BaseFormat, 
+					       texImage->TexFormat, 
+					       texImage->Data, 0, 0, 0, /* dstX/Y/Zoffset */
+					       dstRowStride,
+					       texImage->ImageOffsets,
+					       width, height, depth,
+					       format, type, pixels, unpack)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
+   }
+
+   _mesa_unmap_teximage_pbo(ctx, unpack);
+
+   if (intelImage->mt) {
+      intel_miptree_image_unmap(intel, intelImage->mt);
+      texImage->Data = NULL;
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+#if 0
+   /* GL_SGIS_generate_mipmap -- this can be accelerated now.
+    */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      intel_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+#endif
+}
+
+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 *unpack,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+   intelTexImage(ctx, 3, target, level,
+                 internalFormat, width, height, depth, border,
+                 format, type, pixels, unpack, texObj, texImage, 0, 0);
+}
+
+
+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 *unpack,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+   intelTexImage(ctx, 2, target, level,
+                 internalFormat, width, height, 1, border,
+                 format, type, pixels, unpack, texObj, texImage, 0, 0);
+}
+
+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 *unpack,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+   intelTexImage(ctx, 1, target, level,
+                 internalFormat, width, 1, 1, border,
+                 format, type, pixels, unpack, texObj, texImage, 0, 0);
+}
+
+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 )
+{
+   intelTexImage(ctx, 2, target, level,
+		 internalFormat, width, height, 1, border,
+		 0, 0, data, &ctx->Unpack, texObj, texImage, imageSize, 1);
+}
+
+/**
+ * Need to map texture image into memory before copying image data,
+ * then unmap it.
+ */
+static void
+intel_get_tex_image(GLcontext * ctx, GLenum target, GLint level,
+		    GLenum format, GLenum type, GLvoid * pixels,
+		    struct gl_texture_object *texObj,
+		    struct gl_texture_image *texImage, int compressed)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+
+   /* Map */
+   if (intelImage->mt) {
+      /* Image is stored in hardware format in a buffer managed by the
+       * kernel.  Need to explicitly map and unmap it.
+       */
+      intelImage->base.Data =
+         intel_miptree_image_map(intel,
+                                 intelImage->mt,
+                                 intelImage->face,
+                                 intelImage->level,
+                                 &intelImage->base.RowStride,
+                                 intelImage->base.ImageOffsets);
+      intelImage->base.RowStride /= intelImage->mt->cpp;
+   }
+   else {
+      /* Otherwise, the image should actually be stored in
+       * intelImage->base.Data.  This is pretty confusing for
+       * everybody, I'd much prefer to separate the two functions of
+       * texImage->Data - storage for texture images in main memory
+       * and access (ie mappings) of images.  In other words, we'd
+       * create a new texImage->Map field and leave Data simply for
+       * storage.
+       */
+      assert(intelImage->base.Data);
+   }
+
+
+   if (compressed) {
+      _mesa_get_compressed_teximage(ctx, target, level, pixels,
+				    texObj, texImage);
+   } else {
+      _mesa_get_teximage(ctx, target, level, format, type, pixels,
+			 texObj, texImage);
+   }
+     
+
+   /* Unmap */
+   if (intelImage->mt) {
+      intel_miptree_image_unmap(intel, intelImage->mt);
+      intelImage->base.Data = NULL;
+   }
+}
+
+void
+intelGetTexImage(GLcontext * ctx, GLenum target, GLint level,
+                 GLenum format, GLenum type, GLvoid * pixels,
+                 struct gl_texture_object *texObj,
+                 struct gl_texture_image *texImage)
+{
+   intel_get_tex_image(ctx, target, level, format, type, pixels,
+		       texObj, texImage, 0);
+
+
+}
+
+void
+intelGetCompressedTexImage(GLcontext *ctx, GLenum target, GLint level,
+			   GLvoid *pixels,
+			   const struct gl_texture_object *texObj,
+			   const struct gl_texture_image *texImage)
+{
+   intel_get_tex_image(ctx, target, level, 0, 0, pixels,
+		       texObj, texImage, 1);
+
+}
+
+void
+intelSetTexOffset(__DRIcontext *pDRICtx, GLint texname,
+		  unsigned long long offset, GLint depth, GLuint pitch)
+{
+   struct intel_context *intel = (struct intel_context*)
+      ((__DRIcontextPrivate*)pDRICtx->private)->driverPrivate;
+   struct gl_texture_object *tObj = _mesa_lookup_texture(&intel->ctx, texname);
+   struct intel_texture_object *intelObj = intel_texture_object(tObj);
+
+   if (!intelObj)
+      return;
+
+   if (intelObj->mt)
+      intel_miptree_release(intel, &intelObj->mt);
+
+   intelObj->imageOverride = GL_TRUE;
+   intelObj->depthOverride = depth;
+   intelObj->pitchOverride = pitch;
+
+   if (offset)
+      intelObj->textureOffset = offset;
+}
diff --git a/src/mesa/drivers/dri/intel/intel_tex_subimage.c b/src/mesa/drivers/dri/intel/intel_tex_subimage.c
new file mode 100644
index 00000000000..3935787806b
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_tex_subimage.c
@@ -0,0 +1,182 @@
+
+/**************************************************************************
+ * 
+ * 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 "mtypes.h"
+#include "texobj.h"
+#include "texstore.h"
+#include "enums.h"
+
+#include "intel_context.h"
+#include "intel_tex.h"
+#include "intel_mipmap_tree.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+static void
+intelTexSubimage(GLcontext * ctx,
+                 GLint dims,
+                 GLenum target, GLint level,
+                 GLint xoffset, GLint yoffset, GLint zoffset,
+                 GLint width, GLint height, GLint depth,
+                 GLenum format, GLenum type, const void *pixels,
+                 const struct gl_pixelstore_attrib *packing,
+                 struct gl_texture_object *texObj,
+                 struct gl_texture_image *texImage)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+   GLuint dstRowStride;
+
+   DBG("%s target %s level %d offset %d,%d %dx%d\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(target),
+       level, xoffset, yoffset, width, height);
+
+   intelFlush(ctx);
+
+   pixels =
+      _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, format,
+                                  type, pixels, packing, "glTexSubImage2D");
+   if (!pixels)
+      return;
+
+   if (intelImage->mt)
+      intel_region_idle(intel->intelScreen, intelImage->mt->region);
+
+   LOCK_HARDWARE(intel);
+
+   /* Map buffer if necessary.  Need to lock to prevent other contexts
+    * from uploading the buffer under us.
+    */
+   if (intelImage->mt) 
+      texImage->Data = intel_miptree_image_map(intel,
+                                               intelImage->mt,
+                                               intelImage->face,
+                                               intelImage->level,
+                                               &dstRowStride,
+                                               texImage->ImageOffsets);
+
+   assert(dstRowStride);
+
+   if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat,
+                                        texImage->TexFormat,
+                                        texImage->Data,
+                                        xoffset, yoffset, zoffset,
+                                        dstRowStride,
+                                        texImage->ImageOffsets,
+                                        width, height, depth,
+                                        format, type, pixels, packing)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage");
+   }
+
+#if 0
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+#endif
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
+
+   if (intelImage->mt) {
+      intel_miptree_image_unmap(intel, intelImage->mt);
+      texImage->Data = NULL;
+   }
+
+   UNLOCK_HARDWARE(intel);
+}
+
+
+
+
+
+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)
+{
+
+   intelTexSubimage(ctx, 3,
+                    target, level,
+                    xoffset, yoffset, zoffset,
+                    width, height, depth,
+                    format, type, pixels, packing, texObj, 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)
+{
+
+   intelTexSubimage(ctx, 2,
+                    target, level,
+                    xoffset, yoffset, 0,
+                    width, height, 1,
+                    format, type, pixels, packing, texObj, 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)
+{
+   intelTexSubimage(ctx, 1,
+                    target, level,
+                    xoffset, 0, 0,
+                    width, 1, 1,
+                    format, type, pixels, packing, texObj, texImage);
+
+}
diff --git a/src/mesa/drivers/dri/intel/intel_tex_validate.c b/src/mesa/drivers/dri/intel/intel_tex_validate.c
new file mode 100644
index 00000000000..af18c26d55c
--- /dev/null
+++ b/src/mesa/drivers/dri/intel/intel_tex_validate.c
@@ -0,0 +1,272 @@
+#include "mtypes.h"
+#include "macros.h"
+
+#include "intel_context.h"
+#include "intel_batchbuffer.h"
+#include "intel_mipmap_tree.h"
+#include "intel_tex.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+/**
+ * Compute which mipmap levels that really need to be sent 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.
+ */
+static void
+intel_calculate_first_last_level(struct intel_texture_object *intelObj)
+{
+   struct gl_texture_object *tObj = &intelObj->base;
+   const struct gl_texture_image *const baseImage =
+      tObj->Image[0][tObj->BaseLevel];
+
+   /* These must be signed values.  MinLod and MaxLod can be negative numbers,
+    * and having firstLevel and lastLevel as signed prevents the need for
+    * extra sign checks.
+    */
+   int firstLevel;
+   int lastLevel;
+
+   /* Yes, this looks overly complicated, but it's all needed.
+    */
+   switch (tObj->Target) {
+   case GL_TEXTURE_1D:
+   case GL_TEXTURE_2D:
+   case GL_TEXTURE_3D:
+   case GL_TEXTURE_CUBE_MAP:
+      if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) {
+         /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
+          */
+         firstLevel = lastLevel = tObj->BaseLevel;
+      }
+      else {
+         firstLevel = tObj->BaseLevel + (GLint) (tObj->MinLod + 0.5);
+         firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+         lastLevel = tObj->BaseLevel + (GLint) (tObj->MaxLod + 0.5);
+         lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+         lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
+         lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+         lastLevel = MAX2(firstLevel, lastLevel);       /* need at least one level */
+      }
+      break;
+   case GL_TEXTURE_RECTANGLE_NV:
+   case GL_TEXTURE_4D_SGIS:
+      firstLevel = lastLevel = 0;
+      break;
+   default:
+      return;
+   }
+
+   /* save these values */
+   intelObj->firstLevel = firstLevel;
+   intelObj->lastLevel = lastLevel;
+}
+
+static void
+copy_image_data_to_tree(struct intel_context *intel,
+                        struct intel_texture_object *intelObj,
+                        struct intel_texture_image *intelImage)
+{
+   if (intelImage->mt) {
+      /* Copy potentially with the blitter:
+       */
+      intel_miptree_image_copy(intel,
+                               intelObj->mt,
+                               intelImage->face,
+                               intelImage->level, intelImage->mt);
+
+      intel_miptree_release(intel, &intelImage->mt);
+   }
+   else {
+      assert(intelImage->base.Data != NULL);
+
+      /* More straightforward upload.  
+       */
+      intel_miptree_image_data(intel,
+                               intelObj->mt,
+                               intelImage->face,
+                               intelImage->level,
+                               intelImage->base.Data,
+                               intelImage->base.RowStride,
+                               intelImage->base.RowStride *
+                               intelImage->base.Height);
+      _mesa_align_free(intelImage->base.Data);
+      intelImage->base.Data = NULL;
+   }
+
+   intel_miptree_reference(&intelImage->mt, intelObj->mt);
+}
+
+
+/*  
+ */
+GLuint
+intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit)
+{
+   struct gl_texture_object *tObj = intel->ctx.Texture.Unit[unit]._Current;
+   struct intel_texture_object *intelObj = intel_texture_object(tObj);
+   int comp_byte = 0;
+   int cpp;
+
+   GLuint face, i;
+   GLuint nr_faces = 0;
+   struct intel_texture_image *firstImage;
+
+   GLboolean need_flush = GL_FALSE;
+
+   /* We know/require this is true by now: 
+    */
+   assert(intelObj->base._Complete);
+
+   /* What levels must the tree include at a minimum?
+    */
+   intel_calculate_first_last_level(intelObj);
+   firstImage =
+      intel_texture_image(intelObj->base.Image[0][intelObj->firstLevel]);
+
+   /* Fallback case:
+    */
+   if (firstImage->base.Border) {
+      if (intelObj->mt) {
+         intel_miptree_release(intel, &intelObj->mt);
+      }
+      return GL_FALSE;
+   }
+
+
+   /* If both firstImage and intelObj have a tree which can contain
+    * all active images, favour firstImage.  Note that because of the
+    * completeness requirement, we know that the image dimensions
+    * will match.
+    */
+   if (firstImage->mt &&
+       firstImage->mt != intelObj->mt &&
+       firstImage->mt->first_level <= intelObj->firstLevel &&
+       firstImage->mt->last_level >= intelObj->lastLevel) {
+
+      if (intelObj->mt)
+         intel_miptree_release(intel, &intelObj->mt);
+
+      intel_miptree_reference(&intelObj->mt, firstImage->mt);
+   }
+
+   if (firstImage->base.IsCompressed) {
+      comp_byte = intel_compressed_num_bytes(firstImage->base.TexFormat->MesaFormat);
+      cpp = comp_byte;
+   }
+   else cpp = firstImage->base.TexFormat->TexelBytes;
+
+   /* Check tree can hold all active levels.  Check tree matches
+    * target, imageFormat, etc.
+    * 
+    * XXX: For some layouts (eg i945?), the test might have to be
+    * first_level == firstLevel, as the tree isn't valid except at the
+    * original start level.  Hope to get around this by
+    * programming minLod, maxLod, baseLevel into the hardware and
+    * leaving the tree alone.
+    */
+   if (intelObj->mt &&
+       (intelObj->mt->target != intelObj->base.Target ||
+	intelObj->mt->internal_format != firstImage->base.InternalFormat ||
+	intelObj->mt->first_level != intelObj->firstLevel ||
+	intelObj->mt->last_level != intelObj->lastLevel ||
+	intelObj->mt->width0 != firstImage->base.Width ||
+	intelObj->mt->height0 != firstImage->base.Height ||
+	intelObj->mt->depth0 != firstImage->base.Depth ||
+	intelObj->mt->cpp != cpp ||
+	intelObj->mt->compressed != firstImage->base.IsCompressed)) {
+      intel_miptree_release(intel, &intelObj->mt);
+   }
+
+
+   /* May need to create a new tree:
+    */
+   if (!intelObj->mt) {
+      intelObj->mt = intel_miptree_create(intel,
+                                          intelObj->base.Target,
+                                          firstImage->base.InternalFormat,
+                                          intelObj->firstLevel,
+                                          intelObj->lastLevel,
+                                          firstImage->base.Width,
+                                          firstImage->base.Height,
+                                          firstImage->base.Depth,
+                                          cpp,
+                                          comp_byte);
+   }
+
+   /* Pull in any images not in the object's tree:
+    */
+   nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   for (face = 0; face < nr_faces; face++) {
+      for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) {
+         struct intel_texture_image *intelImage =
+            intel_texture_image(intelObj->base.Image[face][i]);
+
+         /* Need to import images in main memory or held in other trees.
+          */
+         if (intelObj->mt != intelImage->mt) {
+            copy_image_data_to_tree(intel, intelObj, intelImage);
+	    need_flush = GL_TRUE;
+         }
+      }
+   }
+
+   if (need_flush)
+      intel_batchbuffer_flush(intel->batch);
+
+   return GL_TRUE;
+}
+
+
+
+void
+intel_tex_map_images(struct intel_context *intel,
+                     struct intel_texture_object *intelObj)
+{
+   GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   GLuint face, i;
+
+   DBG("%s\n", __FUNCTION__);
+
+   for (face = 0; face < nr_faces; face++) {
+      for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) {
+         struct intel_texture_image *intelImage =
+            intel_texture_image(intelObj->base.Image[face][i]);
+
+         if (intelImage->mt) {
+            intelImage->base.Data =
+               intel_miptree_image_map(intel,
+                                       intelImage->mt,
+                                       intelImage->face,
+                                       intelImage->level,
+                                       &intelImage->base.RowStride,
+                                       intelImage->base.ImageOffsets);
+            /* convert stride to texels, not bytes */
+            intelImage->base.RowStride /= intelImage->mt->cpp;
+/*             intelImage->base.ImageStride /= intelImage->mt->cpp; */
+         }
+      }
+   }
+}
+
+
+
+void
+intel_tex_unmap_images(struct intel_context *intel,
+                       struct intel_texture_object *intelObj)
+{
+   GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   GLuint face, i;
+
+   for (face = 0; face < nr_faces; face++) {
+      for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) {
+         struct intel_texture_image *intelImage =
+            intel_texture_image(intelObj->base.Image[face][i]);
+
+         if (intelImage->mt) {
+            intel_miptree_image_unmap(intel, intelImage->mt);
+            intelImage->base.Data = NULL;
+         }
+      }
+   }
+}