Merge remote branch 'origin/master' into pipe-video

Conflicts:
	src/gallium/drivers/r600/r600_asm.c
	src/gallium/tests/unit/SConscript

22 files changed, 494 insertions, 150 deletions

diff --git a/src/gallium/drivers/i915/TODO b/src/gallium/drivers/i915/TODO
index f4e1423fa59..fba180064c3 100644
--- a/src/gallium/drivers/i915/TODO
+++ b/src/gallium/drivers/i915/TODO
@@ -12,22 +12,19 @@ Random list of problems with i915g:
   unusable :( Upgrading xserver helped here, it doesn't crash anymore. Still
   broken, it doesn't update the viewport/get new buffers.
 
-- Tends to hang the chip after a few minutes of openarena. Looks tiling related,
-  at the last frame rendered has tiling corruption over the complete frame.
-
 - Kills the chip in 3D_PRIMITIVE LINELIST with mesa-demos/fbotexture in
-  wireframe mode.
-
-- Tiling is funny: If unlucky, it renders/samples all black. No clue yet what's
-  going on. Seems to depend on tiny details like whethever the sampler
-  relocation is fenced/unfenced (broken _with_ fenced reloc using tiling bits!).
+  wireframe mode. Changing the cullmode to cw from none mitigates the crash. As
+  does emitting only one line segment (2 indices) per 3D_PRIMITIVE command in
+  the batch.
 
 - Y-tiling is even more fun. i915c doesn't use it, maybe there's a reason?
   Texture sampling from Y-tiled buffers seems to work, though (save above
   problems).
+  RESOLVED: Y-tiling works with the render engine, but not with the blitter.
+  Use u_blitter and hw clears (PRIM3D_CLEAR_RECT).
 
-- Need to validate buffers before usage. Currently do_exec on the batchbuffer
-  can fail with -ENOSPC.
+- src/xvmc/i915_structs.h in xf86-video-intel has a few more bits of various
+  commands defined. Scavenge them and see what's useful.
 
 Other bugs can be found here:
 https://bugs.freedesktop.org/buglist.cgi?bug_status=NEW&bug_status=ASSIGNED&bug_status=REOPENED&component=Drivers/Gallium/i915g
diff --git a/src/gallium/drivers/i915/i915_batch.h b/src/gallium/drivers/i915/i915_batch.h
index 039c8713570..ce2691b2fd7 100644
--- a/src/gallium/drivers/i915/i915_batch.h
+++ b/src/gallium/drivers/i915/i915_batch.h
@@ -37,6 +37,9 @@
 #define OUT_BATCH(dword) \
    i915_winsys_batchbuffer_dword(i915->batch, dword)
 
+#define OUT_BATCH_F(f) \
+   i915_winsys_batchbuffer_float(i915->batch, f)
+
 #define OUT_RELOC(buf, usage, offset) \
    i915_winsys_batchbuffer_reloc(i915->batch, buf, usage, offset, false)
 
diff --git a/src/gallium/drivers/i915/i915_batchbuffer.h b/src/gallium/drivers/i915/i915_batchbuffer.h
index 9df82272604..78554034781 100644
--- a/src/gallium/drivers/i915/i915_batchbuffer.h
+++ b/src/gallium/drivers/i915/i915_batchbuffer.h
@@ -55,6 +55,16 @@ i915_winsys_batchbuffer_dword_unchecked(struct i915_winsys_batchbuffer *batch,
 }
 
 static INLINE void
+i915_winsys_batchbuffer_float(struct i915_winsys_batchbuffer *batch,
+                              float f)
+{
+   union { float f; unsigned int ui; } uif;
+   uif.f = f;
+   assert (i915_winsys_batchbuffer_space(batch) >= 4);
+   i915_winsys_batchbuffer_dword_unchecked(batch, uif.ui);
+}
+
+static INLINE void
 i915_winsys_batchbuffer_dword(struct i915_winsys_batchbuffer *batch,
                               unsigned dword)
 {
diff --git a/src/gallium/drivers/i915/i915_clear.c b/src/gallium/drivers/i915/i915_clear.c
index 6d824a507aa..4a97746e981 100644
--- a/src/gallium/drivers/i915/i915_clear.c
+++ b/src/gallium/drivers/i915/i915_clear.c
@@ -31,17 +31,118 @@
 
 
 #include "util/u_clear.h"
+#include "util/u_format.h"
+#include "util/u_pack_color.h"
 #include "i915_context.h"
+#include "i915_screen.h"
+#include "i915_reg.h"
+#include "i915_batch.h"
+#include "i915_resource.h"
+#include "i915_state.h"
 
+void
+i915_clear_emit(struct pipe_context *pipe, unsigned buffers, const float *rgba,
+                double depth, unsigned stencil,
+                unsigned destx, unsigned desty, unsigned width, unsigned height)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   uint32_t clear_params, clear_color, clear_depth, clear_stencil,
+            clear_color8888, packed_z_stencil;
+   union util_color u_color;
+   float f_depth = depth;
+   struct i915_texture *cbuf_tex, *depth_tex;
+
+   cbuf_tex = depth_tex = NULL;
+   clear_params = 0;
+
+   if (buffers & PIPE_CLEAR_COLOR) {
+      struct pipe_surface *cbuf = i915->framebuffer.cbufs[0];
+
+      clear_params |= CLEARPARAM_WRITE_COLOR;
+      cbuf_tex = i915_texture(cbuf->texture);
+      util_pack_color(rgba, cbuf->format, &u_color);
+      if (util_format_get_blocksize(cbuf_tex->b.b.format) == 4)
+         clear_color = u_color.ui;
+      else
+         clear_color = (u_color.ui & 0xffff) | (u_color.ui << 16);
+
+      util_pack_color(rgba, PIPE_FORMAT_B8G8R8A8_UNORM, &u_color);
+      clear_color8888 = u_color.ui;
+   } else
+      clear_color = clear_color8888 = 0;
+
+   clear_depth = clear_stencil = 0;
+   if (buffers & PIPE_CLEAR_DEPTH) {
+      struct pipe_surface *zbuf = i915->framebuffer.zsbuf;
+
+      clear_params |= CLEARPARAM_WRITE_DEPTH;
+      depth_tex = i915_texture(zbuf->texture);
+      packed_z_stencil = util_pack_z_stencil(depth_tex->b.b.format, depth, stencil);
+
+      if (util_format_get_blocksize(depth_tex->b.b.format) == 4) {
+         /* Avoid read-modify-write if there's no stencil. */
+         if (buffers & PIPE_CLEAR_STENCIL
+               || depth_tex->b.b.format != PIPE_FORMAT_Z24_UNORM_S8_USCALED) {
+            clear_params |= CLEARPARAM_WRITE_STENCIL;
+            clear_stencil = packed_z_stencil & 0xff;
+            clear_depth = packed_z_stencil;
+         } else
+            clear_depth = packed_z_stencil & 0xffffff00;
+      } else {
+         clear_depth = (clear_depth & 0xffff) | (clear_depth << 16);
+      }
+   }
+
+   if (i915->hardware_dirty)
+      i915_emit_hardware_state(i915);
+
+   if (!BEGIN_BATCH(7 + 7)) {
+      FLUSH_BATCH(NULL);
+
+      i915_emit_hardware_state(i915);
+      i915->vbo_flushed = 1;
+
+      assert(BEGIN_BATCH(7 + 7));
+   }
+
+   OUT_BATCH(_3DSTATE_CLEAR_PARAMETERS);
+   OUT_BATCH(clear_params | CLEARPARAM_CLEAR_RECT);
+   OUT_BATCH(clear_color);
+   OUT_BATCH(clear_depth);
+   OUT_BATCH(clear_color8888);
+   OUT_BATCH_F(f_depth);
+   OUT_BATCH(clear_stencil);
+
+   OUT_BATCH(_3DPRIMITIVE | PRIM3D_CLEAR_RECT | 5);
+   OUT_BATCH_F(destx + width);
+   OUT_BATCH_F(desty + height);
+   OUT_BATCH_F(destx);
+   OUT_BATCH_F(desty + height);
+   OUT_BATCH_F(destx);
+   OUT_BATCH_F(desty);
+}
 
 /**
  * Clear the given buffers to the specified values.
  * No masking, no scissor (clear entire buffer).
  */
 void
-i915_clear(struct pipe_context *pipe, unsigned buffers, const float *rgba,
-           double depth, unsigned stencil)
+i915_clear_blitter(struct pipe_context *pipe, unsigned buffers, const float *rgba,
+                   double depth, unsigned stencil)
 {
    util_clear(pipe, &i915_context(pipe)->framebuffer, buffers, rgba, depth,
               stencil);
 }
+
+void
+i915_clear_render(struct pipe_context *pipe, unsigned buffers, const float *rgba,
+                  double depth, unsigned stencil)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   if (i915->dirty)
+      i915_update_derived(i915);
+
+   i915_clear_emit(pipe, buffers, rgba, depth, stencil,
+                   0, 0, i915->framebuffer.width, i915->framebuffer.height);
+}
diff --git a/src/gallium/drivers/i915/i915_context.c b/src/gallium/drivers/i915/i915_context.c
index 84c8cb54436..7a98ef73c1f 100644
--- a/src/gallium/drivers/i915/i915_context.c
+++ b/src/gallium/drivers/i915/i915_context.c
@@ -103,6 +103,9 @@ static void i915_destroy(struct pipe_context *pipe)
    int i;
 
    draw_destroy(i915->draw);
+
+   if (i915->blitter)
+      util_blitter_destroy(i915->blitter);
    
    if(i915->batch)
       i915->iws->batchbuffer_destroy(i915->batch);
@@ -137,7 +140,10 @@ i915_create_context(struct pipe_screen *screen, void *priv)
 
    i915->base.destroy = i915_destroy;
 
-   i915->base.clear = i915_clear;
+   if (i915_screen(screen)->debug.use_blitter)
+      i915->base.clear = i915_clear_blitter;
+   else
+      i915->base.clear = i915_clear_render;
 
    i915->base.draw_vbo = i915_draw_vbo;
 
@@ -145,6 +151,10 @@ i915_create_context(struct pipe_screen *screen, void *priv)
    util_slab_create(&i915->transfer_pool, sizeof(struct pipe_transfer),
                     16, UTIL_SLAB_SINGLETHREADED);
 
+   /* Batch stream debugging is a bit hacked up at the moment:
+    */
+   i915->batch = i915->iws->batchbuffer_create(i915->iws);
+
    /*
     * Create drawing context and plug our rendering stage into it.
     */
@@ -164,15 +174,19 @@ i915_create_context(struct pipe_screen *screen, void *priv)
    draw_install_aaline_stage(i915->draw, &i915->base);
    draw_install_aapoint_stage(i915->draw, &i915->base);
 
+   /* augmented draw pipeline clobbers state functions */
+   i915_init_fixup_state_functions(i915);
+
+   /* Create blitter last - calls state creation functions. */
+   i915->blitter = util_blitter_create(&i915->base);
+   assert(i915->blitter);
+
    i915->dirty = ~0;
    i915->hardware_dirty = ~0;
    i915->immediate_dirty = ~0;
    i915->dynamic_dirty = ~0;
+   i915->static_dirty = ~0;
    i915->flush_dirty = 0;
 
-   /* Batch stream debugging is a bit hacked up at the moment:
-    */
-   i915->batch = i915->iws->batchbuffer_create(i915->iws);
-
    return &i915->base;
 }
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index 1da637d068e..dacf50e870d 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -38,6 +38,7 @@
 #include "tgsi/tgsi_scan.h"
 
 #include "util/u_slab.h"
+#include "util/u_blitter.h"
 
 
 struct i915_winsys;
@@ -185,7 +186,7 @@ struct i915_rasterizer_state {
    unsigned LIS7;
    unsigned sc[1];
 
-   const struct pipe_rasterizer_state *templ;
+   struct pipe_rasterizer_state templ;
 
    union { float f; unsigned u; } ds[2];
 };
@@ -244,14 +245,35 @@ struct i915_context {
 
    struct i915_state current;
    unsigned hardware_dirty;
-   unsigned immediate_dirty;
-   unsigned dynamic_dirty;
-   unsigned flush_dirty;
+   unsigned immediate_dirty : I915_MAX_IMMEDIATE;
+   unsigned dynamic_dirty : I915_MAX_DYNAMIC;
+   unsigned static_dirty : 4;
+   unsigned flush_dirty : 2;
 
    struct i915_winsys_buffer *validation_buffers[2 + 1 + I915_TEX_UNITS];
    int num_validation_buffers;
 
    struct util_slab_mempool transfer_pool;
+
+   /** blitter/hw-clear */
+   struct blitter_context* blitter;
+
+   /** State tracking needed by u_blitter for save/restore. */
+   void *saved_fs;
+   void (*saved_bind_fs_state)(struct pipe_context *pipe, void *shader);
+   void *saved_vs;
+   struct pipe_clip_state saved_clip;
+   struct i915_velems_state *saved_velems;
+   unsigned saved_nr_vertex_buffers;
+   struct pipe_vertex_buffer saved_vertex_buffers[PIPE_MAX_ATTRIBS];
+   unsigned saved_nr_samplers;
+   void *saved_samplers[PIPE_MAX_SAMPLERS];
+   void (*saved_bind_sampler_states)(struct pipe_context *pipe,
+                                     unsigned num, void **sampler);
+   unsigned saved_nr_sampler_views;
+   struct pipe_sampler_view *saved_sampler_views[PIPE_MAX_SAMPLERS];
+   void (*saved_set_sampler_views)(struct pipe_context *pipe,
+                                   unsigned num, struct pipe_sampler_view **views);
 };
 
 /* A flag for each state_tracker state object:
@@ -296,6 +318,12 @@ struct i915_context {
 #define I915_FLUSH_CACHE		1
 #define I915_PIPELINE_FLUSH		2
 
+/* split up static state */
+#define I915_DST_BUF_COLOR              1
+#define I915_DST_BUF_DEPTH              2
+#define I915_DST_VARS                   4
+#define I915_DST_RECT                   8
+
 static INLINE
 void i915_set_flush_dirty(struct i915_context *i915, unsigned flush)
 {
@@ -326,14 +354,20 @@ void i915_emit_hardware_state(struct i915_context *i915 );
 /***********************************************************************
  * i915_clear.c: 
  */
-void i915_clear( struct pipe_context *pipe, unsigned buffers, const float *rgba,
-                 double depth, unsigned stencil);
+void i915_clear_blitter(struct pipe_context *pipe, unsigned buffers, const float *rgba,
+                        double depth, unsigned stencil);
+void i915_clear_render(struct pipe_context *pipe, unsigned buffers, const float *rgba,
+                       double depth, unsigned stencil);
+void i915_clear_emit(struct pipe_context *pipe, unsigned buffers, const float *rgba,
+                     double depth, unsigned stencil,
+                     unsigned destx, unsigned desty, unsigned width, unsigned height);
 
 
 /***********************************************************************
  * 
  */
 void i915_init_state_functions( struct i915_context *i915 );
+void i915_init_fixup_state_functions( struct i915_context *i915 );
 void i915_init_flush_functions( struct i915_context *i915 );
 void i915_init_string_functions( struct i915_context *i915 );
 
diff --git a/src/gallium/drivers/i915/i915_debug.c b/src/gallium/drivers/i915/i915_debug.c
index 1713bf131f2..c4eed473e90 100644
--- a/src/gallium/drivers/i915/i915_debug.c
+++ b/src/gallium/drivers/i915/i915_debug.c
@@ -46,17 +46,18 @@ static const struct debug_named_value debug_options[] = {
 };
 
 unsigned i915_debug = 0;
-boolean i915_tiling = TRUE;
 
 DEBUG_GET_ONCE_FLAGS_OPTION(i915_debug, "I915_DEBUG", debug_options, 0)
 DEBUG_GET_ONCE_BOOL_OPTION(i915_no_tiling, "I915_NO_TILING", FALSE)
 DEBUG_GET_ONCE_BOOL_OPTION(i915_lie, "I915_LIE", FALSE)
+DEBUG_GET_ONCE_BOOL_OPTION(i915_use_blitter, "I915_USE_BLITTER", FALSE)
 
 void i915_debug_init(struct i915_screen *is)
 {
    i915_debug = debug_get_option_i915_debug();
    is->debug.tiling = !debug_get_option_i915_no_tiling();
    is->debug.lie = debug_get_option_i915_lie();
+   is->debug.use_blitter = debug_get_option_i915_use_blitter();
 }
 
 
diff --git a/src/gallium/drivers/i915/i915_flush.c b/src/gallium/drivers/i915/i915_flush.c
index f2044d661e3..b4e81147c4f 100644
--- a/src/gallium/drivers/i915/i915_flush.c
+++ b/src/gallium/drivers/i915/i915_flush.c
@@ -39,34 +39,12 @@
 
 
 static void i915_flush_pipe( struct pipe_context *pipe,
-                             unsigned flags,
                              struct pipe_fence_handle **fence )
 {
    struct i915_context *i915 = i915_context(pipe);
 
    draw_flush(i915->draw);
 
-#if 0
-   /* Do we need to emit an MI_FLUSH command to flush the hardware
-    * caches?
-    */
-   if (flags & (PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_TEXTURE_CACHE)) {
-      unsigned flush = MI_FLUSH;
-      
-      if (!(flags & PIPE_FLUSH_RENDER_CACHE))
-	 flush |= INHIBIT_FLUSH_RENDER_CACHE;
-
-      if (flags & PIPE_FLUSH_TEXTURE_CACHE)
-	 flush |= FLUSH_MAP_CACHE;
-
-      if (!BEGIN_BATCH(1)) {
-	 FLUSH_BATCH(NULL);
-	 assert(BEGIN_BATCH(1));
-      }
-      OUT_BATCH( flush );
-   }
-#endif
-
    if (i915->batch->map == i915->batch->ptr) {
       return;
    }
@@ -96,6 +74,7 @@ void i915_flush(struct i915_context *i915, struct pipe_fence_handle **fence)
    i915->hardware_dirty = ~0;
    i915->immediate_dirty = ~0;
    i915->dynamic_dirty = ~0;
+   i915->static_dirty = ~0;
    /* kernel emits flushes in between batchbuffers */
    i915->flush_dirty = 0;
 }
diff --git a/src/gallium/drivers/i915/i915_prim_emit.c b/src/gallium/drivers/i915/i915_prim_emit.c
index 276e33d4b9d..85656cd7846 100644
--- a/src/gallium/drivers/i915/i915_prim_emit.c
+++ b/src/gallium/drivers/i915/i915_prim_emit.c
@@ -149,7 +149,6 @@ emit_prim( struct draw_stage *stage,
 
       /* Make sure state is re-emitted after a flush: 
        */
-      i915_update_derived( i915 );
       i915_emit_hardware_state( i915 );
 
       if (!BEGIN_BATCH( 1 + nr * vertex_size / 4)) {
diff --git a/src/gallium/drivers/i915/i915_prim_vbuf.c b/src/gallium/drivers/i915/i915_prim_vbuf.c
index fb4c0516dd8..79db3b650eb 100644
--- a/src/gallium/drivers/i915/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915/i915_prim_vbuf.c
@@ -470,7 +470,6 @@ draw_arrays_fallback(struct vbuf_render *render,
 
       /* Make sure state is re-emitted after a flush:
        */
-      i915_update_derived(i915);
       i915_emit_hardware_state(i915);
       i915->vbo_flushed = 1;
 
@@ -519,7 +518,6 @@ i915_vbuf_render_draw_arrays(struct vbuf_render *render,
 
       /* Make sure state is re-emitted after a flush:
        */
-      i915_update_derived(i915);
       i915_emit_hardware_state(i915);
       i915->vbo_flushed = 1;
 
@@ -640,7 +638,6 @@ i915_vbuf_render_draw_elements(struct vbuf_render *render,
 
       /* Make sure state is re-emitted after a flush: 
        */
-      i915_update_derived(i915);
       i915_emit_hardware_state(i915);
       i915->vbo_flushed = 1;
 
diff --git a/src/gallium/drivers/i915/i915_reg.h b/src/gallium/drivers/i915/i915_reg.h
index 5e4e80ddf6b..6fe032cdb6e 100644
--- a/src/gallium/drivers/i915/i915_reg.h
+++ b/src/gallium/drivers/i915/i915_reg.h
@@ -148,6 +148,7 @@
 /* p161 */
 #define _3DSTATE_DST_BUF_VARS_CMD	(CMD_3D | (0x1d<<24) | (0x85<<16))
 /* Dword 1 */
+#define CLASSIC_EARLY_DEPTH             (1<<31)
 #define TEX_DEFAULT_COLOR_OGL           (0<<30)
 #define TEX_DEFAULT_COLOR_D3D           (1<<30)
 #define ZR_EARLY_DEPTH                  (1<<29)
diff --git a/src/gallium/drivers/i915/i915_resource.c b/src/gallium/drivers/i915/i915_resource.c
index 499233ceb9b..7f52ba11d61 100644
--- a/src/gallium/drivers/i915/i915_resource.c
+++ b/src/gallium/drivers/i915/i915_resource.c
@@ -31,7 +31,6 @@ i915_resource_from_handle(struct pipe_screen * screen,
 void
 i915_init_resource_functions(struct i915_context *i915 )
 {
-   i915->base.is_resource_referenced = u_default_is_resource_referenced;
    i915->base.get_transfer = u_get_transfer_vtbl;
    i915->base.transfer_map = u_transfer_map_vtbl;
    i915->base.transfer_flush_region = u_transfer_flush_region_vtbl;
diff --git a/src/gallium/drivers/i915/i915_resource_buffer.c b/src/gallium/drivers/i915/i915_resource_buffer.c
index 51482f54fc4..77c03450b3a 100644
--- a/src/gallium/drivers/i915/i915_resource_buffer.c
+++ b/src/gallium/drivers/i915/i915_resource_buffer.c
@@ -123,7 +123,6 @@ struct u_resource_vtbl i915_buffer_vtbl =
 {
    i915_buffer_get_handle,	     /* get_handle */
    i915_buffer_destroy,		     /* resource_destroy */
-   NULL,			     /* is_resource_referenced */
    i915_get_transfer,		     /* get_transfer */
    i915_transfer_destroy,	     /* transfer_destroy */
    i915_buffer_transfer_map,	     /* transfer_map */
diff --git a/src/gallium/drivers/i915/i915_resource_texture.c b/src/gallium/drivers/i915/i915_resource_texture.c
index aad5235a6ad..7816925d230 100644
--- a/src/gallium/drivers/i915/i915_resource_texture.c
+++ b/src/gallium/drivers/i915/i915_resource_texture.c
@@ -184,7 +184,10 @@ i915_texture_tiling(struct i915_screen *is, struct i915_texture *tex)
       /* XXX X-tiling might make sense */
       return I915_TILE_NONE;
 
-   return I915_TILE_X;
+   if (is->debug.use_blitter)
+      return I915_TILE_X;
+   else
+      return I915_TILE_Y;
 }
 
 
@@ -756,6 +759,9 @@ i915_texture_transfer_map(struct pipe_context *pipe,
       assert(box->z == 0);
    offset = i915_texture_offset(tex, transfer->level, box->z);
 
+   /* TODO this is a sledgehammer */
+   pipe->flush(pipe, NULL);
+
    map = iws->buffer_map(iws, tex->buffer,
                          (transfer->usage & PIPE_TRANSFER_WRITE) ? TRUE : FALSE);
    if (map == NULL)
@@ -781,7 +787,6 @@ struct u_resource_vtbl i915_texture_vtbl =
 {
    i915_texture_get_handle,	      /* get_handle */
    i915_texture_destroy,	      /* resource_destroy */
-   NULL,			      /* is_resource_referenced */
    i915_texture_get_transfer,	      /* get_transfer */
    i915_transfer_destroy,	      /* transfer_destroy */
    i915_texture_transfer_map,	      /* transfer_map */
diff --git a/src/gallium/drivers/i915/i915_screen.c b/src/gallium/drivers/i915/i915_screen.c
index 77febbf5012..e62b609eb5a 100644
--- a/src/gallium/drivers/i915/i915_screen.c
+++ b/src/gallium/drivers/i915/i915_screen.c
@@ -105,6 +105,7 @@ i915_get_param(struct pipe_screen *screen, enum pipe_cap cap)
    switch (cap) {
    /* Supported features (boolean caps). */
    case PIPE_CAP_ANISOTROPIC_FILTER:
+   case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
    case PIPE_CAP_NPOT_TEXTURES:
    case PIPE_CAP_PRIMITIVE_RESTART: /* draw module */
    case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
@@ -115,17 +116,18 @@ i915_get_param(struct pipe_screen *screen, enum pipe_cap cap)
    /* Features that should be supported (boolean caps). */
    /* XXX: Just test the code */
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
-      return 0;
+   /* XXX: No code but hw supports it */
+   case PIPE_CAP_POINT_SPRITE:
+      /* Also lie about these when asked to (needed for GLSL / GL 2.0) */
+      return is->debug.lie ? 1 : 0;
 
    /* Unsupported features (boolean caps). */
    case PIPE_CAP_ARRAY_TEXTURES:
    case PIPE_CAP_DEPTH_CLAMP:
-   case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE: /* disable for now */
-   case PIPE_CAP_GLSL:
    case PIPE_CAP_INDEP_BLEND_ENABLE:
    case PIPE_CAP_INDEP_BLEND_FUNC:
-   case PIPE_CAP_INSTANCED_DRAWING: /* draw module? */
-   case PIPE_CAP_POINT_SPRITE:
+   case PIPE_CAP_TGSI_INSTANCEID:
+   case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
    case PIPE_CAP_SHADER_STENCIL_EXPORT:
    case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
    case PIPE_CAP_TEXTURE_SWIZZLE:
@@ -133,6 +135,7 @@ i915_get_param(struct pipe_screen *screen, enum pipe_cap cap)
       return 0;
 
    /* Features we can lie about (boolean caps). */
+   case PIPE_CAP_GLSL:
    case PIPE_CAP_OCCLUSION_QUERY:
       return is->debug.lie ? 1 : 0;
 
@@ -249,8 +252,7 @@ i915_is_format_supported(struct pipe_screen *screen,
                          enum pipe_format format,
                          enum pipe_texture_target target,
                          unsigned sample_count,
-                         unsigned tex_usage,
-                         unsigned geom_flags)
+                         unsigned tex_usage)
 {
    static const enum pipe_format tex_supported[] = {
       PIPE_FORMAT_B8G8R8A8_UNORM,
@@ -317,24 +319,23 @@ i915_fence_reference(struct pipe_screen *screen,
    is->iws->fence_reference(is->iws, ptr, fence);
 }
 
-static int
+static boolean
 i915_fence_signalled(struct pipe_screen *screen,
-                     struct pipe_fence_handle *fence,
-                     unsigned flags)
+                     struct pipe_fence_handle *fence)
 {
    struct i915_screen *is = i915_screen(screen);
 
-   return is->iws->fence_signalled(is->iws, fence);
+   return is->iws->fence_signalled(is->iws, fence) == 0;
 }
 
-static int
+static boolean
 i915_fence_finish(struct pipe_screen *screen,
                   struct pipe_fence_handle *fence,
-                  unsigned flags)
+                  uint64_t timeout)
 {
    struct i915_screen *is = i915_screen(screen);
 
-   return is->iws->fence_finish(is->iws, fence);
+   return is->iws->fence_finish(is->iws, fence) == 0;
 }
 
 
diff --git a/src/gallium/drivers/i915/i915_screen.h b/src/gallium/drivers/i915/i915_screen.h
index 60f0e2971e0..cfc585b5350 100644
--- a/src/gallium/drivers/i915/i915_screen.h
+++ b/src/gallium/drivers/i915/i915_screen.h
@@ -49,6 +49,7 @@ struct i915_screen
    struct {
       boolean tiling;
       boolean lie;
+      boolean use_blitter;
    } debug;
 };
 
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index 58bbbd1de2c..1b57c5776f2 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -287,6 +287,17 @@ i915_create_sampler_state(struct pipe_context *pipe,
    return cso;
 }
 
+static void i915_fixup_bind_sampler_states(struct pipe_context *pipe,
+                                           unsigned num, void **sampler)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   i915->saved_nr_samplers = num;
+   memcpy(&i915->saved_samplers, sampler, sizeof(void *) * num);
+
+   i915->saved_bind_sampler_states(pipe, num, sampler);
+}
+
 static void i915_bind_sampler_states(struct pipe_context *pipe,
                                      unsigned num, void **sampler)
 {
@@ -466,6 +477,17 @@ i915_create_fs_state(struct pipe_context *pipe,
 }
 
 static void
+i915_fixup_bind_fs_state(struct pipe_context *pipe, void *shader)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   draw_flush(i915->draw);
+
+   i915->saved_fs = shader;
+
+   i915->saved_bind_fs_state(pipe, shader);
+}
+
+static void
 i915_bind_fs_state(struct pipe_context *pipe, void *shader)
 {
    struct i915_context *i915 = i915_context(pipe);
@@ -505,6 +527,8 @@ static void i915_bind_vs_state(struct pipe_context *pipe, void *shader)
 {
    struct i915_context *i915 = i915_context(pipe);
 
+   i915->saved_vs = shader;
+
    /* just pass-through to draw module */
    draw_bind_vertex_shader(i915->draw, (struct draw_vertex_shader *) shader);
 
@@ -571,6 +595,27 @@ static void i915_set_constant_buffer(struct pipe_context *pipe,
 }
 
 
+static void
+i915_fixup_set_fragment_sampler_views(struct pipe_context *pipe,
+                                      unsigned num,
+                                      struct pipe_sampler_view **views)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   int i;
+
+   for (i = 0; i < num; i++)
+      pipe_sampler_view_reference(&i915->saved_sampler_views[i],
+                                  views[i]);
+
+   for (i = num; i < i915->saved_nr_sampler_views; i++)
+      pipe_sampler_view_reference(&i915->saved_sampler_views[i],
+                                  NULL);
+
+   i915->saved_nr_sampler_views = num;
+
+   i915->saved_set_sampler_views(pipe, num, views);
+}
+
 static void i915_set_fragment_sampler_views(struct pipe_context *pipe,
                                             unsigned num,
                                             struct pipe_sampler_view **views)
@@ -642,7 +687,8 @@ static void i915_set_framebuffer_state(struct pipe_context *pipe,
    i915->framebuffer.height = fb->height;
    i915->framebuffer.nr_cbufs = fb->nr_cbufs;
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
-      pipe_surface_reference(&i915->framebuffer.cbufs[i], fb->cbufs[i]);
+      pipe_surface_reference(&i915->framebuffer.cbufs[i],
+                             i < fb->nr_cbufs ? fb->cbufs[i] : NULL);
    }
    pipe_surface_reference(&i915->framebuffer.zsbuf, fb->zsbuf);
 
@@ -657,6 +703,8 @@ static void i915_set_clip_state( struct pipe_context *pipe,
    struct i915_context *i915 = i915_context(pipe);
    draw_flush(i915->draw);
 
+   i915->saved_clip = *clip;
+
    draw_set_clip_state(i915->draw, clip);
 
    i915->dirty |= I915_NEW_CLIP;
@@ -687,7 +735,7 @@ i915_create_rasterizer_state(struct pipe_context *pipe,
 {
    struct i915_rasterizer_state *cso = CALLOC_STRUCT( i915_rasterizer_state );
 
-   cso->templ = rasterizer;
+   cso->templ = *rasterizer;
    cso->color_interp = rasterizer->flatshade ? INTERP_CONSTANT : INTERP_LINEAR;
    cso->light_twoside = rasterizer->light_twoside;
    cso->ds[0].u = _3DSTATE_DEPTH_OFFSET_SCALE;
@@ -758,7 +806,7 @@ static void i915_bind_rasterizer_state( struct pipe_context *pipe,
 
    /* pass-through to draw module */
    draw_set_rasterizer_state(i915->draw,
-                           (i915->rasterizer ? i915->rasterizer->templ : NULL),
+                           (i915->rasterizer ? &(i915->rasterizer->templ) : NULL),
                            raster);
 
    i915->dirty |= I915_NEW_RASTERIZER;
@@ -778,6 +826,9 @@ static void i915_set_vertex_buffers(struct pipe_context *pipe,
    struct draw_context *draw = i915->draw;
    int i;
 
+   util_copy_vertex_buffers(i915->saved_vertex_buffers,
+                            &i915->saved_nr_vertex_buffers,
+                            buffers, count);
 #if 0
    /* XXX doesn't look like this is needed */
    /* unmap old */
@@ -818,6 +869,8 @@ i915_bind_vertex_elements_state(struct pipe_context *pipe,
    struct i915_context *i915 = i915_context(pipe);
    struct i915_velems_state *i915_velems = (struct i915_velems_state *) velems;
 
+   i915->saved_velems = velems;
+
    /* pass-through to draw module */
    if (i915_velems) {
       draw_set_vertex_elements(i915->draw,
@@ -896,3 +949,14 @@ i915_init_state_functions( struct i915_context *i915 )
    i915->base.set_index_buffer = i915_set_index_buffer;
    i915->base.redefine_user_buffer = u_default_redefine_user_buffer;
 }
+
+void
+i915_init_fixup_state_functions( struct i915_context *i915 )
+{
+   i915->saved_bind_fs_state = i915->base.bind_fs_state;
+   i915->base.bind_fs_state = i915_fixup_bind_fs_state;
+   i915->saved_bind_sampler_states = i915->base.bind_fragment_sampler_states;
+   i915->base.bind_fragment_sampler_states = i915_fixup_bind_sampler_states;
+   i915->saved_set_sampler_views = i915->base.set_fragment_sampler_views;
+   i915->base.set_fragment_sampler_views = i915_fixup_set_fragment_sampler_views;
+}
diff --git a/src/gallium/drivers/i915/i915_state.h b/src/gallium/drivers/i915/i915_state.h
index b4074dc35ba..3f4e40294e6 100644
--- a/src/gallium/drivers/i915/i915_state.h
+++ b/src/gallium/drivers/i915/i915_state.h
@@ -48,6 +48,7 @@ extern struct i915_tracked_state i915_hw_immediate;
 extern struct i915_tracked_state i915_hw_dynamic;
 extern struct i915_tracked_state i915_hw_fs;
 extern struct i915_tracked_state i915_hw_framebuffer;
+extern struct i915_tracked_state i915_hw_dst_buf_vars;
 extern struct i915_tracked_state i915_hw_constants;
 
 void i915_update_derived(struct i915_context *i915);
diff --git a/src/gallium/drivers/i915/i915_state_derived.c b/src/gallium/drivers/i915/i915_state_derived.c
index 1d4026a214e..59ac2f7292a 100644
--- a/src/gallium/drivers/i915/i915_state_derived.c
+++ b/src/gallium/drivers/i915/i915_state_derived.c
@@ -165,6 +165,7 @@ static struct i915_tracked_state *atoms[] = {
    &i915_hw_dynamic,
    &i915_hw_fs,
    &i915_hw_framebuffer,
+   &i915_hw_dst_buf_vars,
    &i915_hw_constants,
    NULL,
 };
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c
index 15350c0a5d7..0155cd83510 100644
--- a/src/gallium/drivers/i915/i915_state_emit.c
+++ b/src/gallium/drivers/i915/i915_state_emit.c
@@ -119,7 +119,7 @@ validate_immediate(struct i915_context *i915, unsigned *batch_space)
                      1 << I915_IMMEDIATE_S5 | 1 << I915_IMMEDIATE_S6) &
                     i915->immediate_dirty;
 
-   if (i915->immediate_dirty & (1 << I915_IMMEDIATE_S0))
+   if (i915->immediate_dirty & (1 << I915_IMMEDIATE_S0) && i915->vbo)
       i915->validation_buffers[i915->num_validation_buffers++] = i915->vbo;
 
    *batch_space = 1 + util_bitcount(dirty);
@@ -173,25 +173,31 @@ emit_dynamic(struct i915_context *i915)
 static void
 validate_static(struct i915_context *i915, unsigned *batch_space)
 {
-   *batch_space = 2 + 5; /* including DRAW_RECT */
+   *batch_space = 0;
 
-   if (i915->current.cbuf_bo) {
+   if (i915->current.cbuf_bo && (i915->static_dirty & I915_DST_BUF_COLOR)) {
       i915->validation_buffers[i915->num_validation_buffers++]
          = i915->current.cbuf_bo;
       *batch_space += 3;
    }
 
-   if (i915->current.depth_bo) {
+   if (i915->current.depth_bo && (i915->static_dirty & I915_DST_BUF_DEPTH)) {
       i915->validation_buffers[i915->num_validation_buffers++]
          = i915->current.depth_bo;
       *batch_space += 3;
    }
+
+   if (i915->static_dirty & I915_DST_VARS)
+      *batch_space += 2;
+
+   if (i915->static_dirty & I915_DST_RECT)
+      *batch_space += 5;
 }
 
 static void
 emit_static(struct i915_context *i915)
 {
-   if (i915->current.cbuf_bo) {
+   if (i915->current.cbuf_bo && (i915->static_dirty & I915_DST_BUF_COLOR)) {
       OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
       OUT_BATCH(i915->current.cbuf_flags);
       OUT_RELOC(i915->current.cbuf_bo,
@@ -201,7 +207,7 @@ emit_static(struct i915_context *i915)
 
    /* What happens if no zbuf??
     */
-   if (i915->current.depth_bo) {
+   if (i915->current.depth_bo && (i915->static_dirty & I915_DST_BUF_DEPTH)) {
       OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
       OUT_BATCH(i915->current.depth_flags);
       OUT_RELOC(i915->current.depth_bo,
@@ -209,7 +215,7 @@ emit_static(struct i915_context *i915)
                 0);
    }
 
-   {
+   if (i915->static_dirty & I915_DST_VARS) {
       OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD);
       OUT_BATCH(i915->current.dst_buf_vars);
    }
@@ -273,7 +279,7 @@ emit_sampler(struct i915_context *i915)
    if (i915->current.sampler_enable_nr) {
       int i;
 
-      OUT_BATCH( _3DSTATE_SAMPLER_STATE | 
+      OUT_BATCH( _3DSTATE_SAMPLER_STATE |
                  (3 * i915->current.sampler_enable_nr) );
 
       OUT_BATCH( i915->current.sampler_enable_flags );
@@ -355,11 +361,13 @@ emit_program(struct i915_context *i915)
 static void
 emit_draw_rect(struct i915_context *i915)
 {
-   OUT_BATCH(_3DSTATE_DRAW_RECT_CMD);
-   OUT_BATCH(DRAW_RECT_DIS_DEPTH_OFS);
-   OUT_BATCH(i915->current.draw_offset);
-   OUT_BATCH(i915->current.draw_size);
-   OUT_BATCH(i915->current.draw_offset);
+   if (i915->static_dirty & I915_DST_RECT) {
+      OUT_BATCH(_3DSTATE_DRAW_RECT_CMD);
+      OUT_BATCH(DRAW_RECT_DIS_DEPTH_OFS);
+      OUT_BATCH(i915->current.draw_offset);
+      OUT_BATCH(i915->current.draw_size);
+      OUT_BATCH(i915->current.draw_offset);
+   }
 }
 
 static boolean
@@ -405,6 +413,8 @@ i915_emit_hardware_state(struct i915_context *i915 )
    unsigned batch_space;
    uintptr_t save_ptr;
 
+   assert(i915->dirty == 0);
+
    if (I915_DBG_ON(DBG_ATOMS))
       i915_dump_hardware_dirty(i915, __FUNCTION__);
 
@@ -444,5 +454,6 @@ i915_emit_hardware_state(struct i915_context *i915 )
    i915->hardware_dirty = 0;
    i915->immediate_dirty = 0;
    i915->dynamic_dirty = 0;
+   i915->static_dirty = 0;
    i915->flush_dirty = 0;
 }
diff --git a/src/gallium/drivers/i915/i915_state_static.c b/src/gallium/drivers/i915/i915_state_static.c
index 20cd23f8f73..2865298318c 100644
--- a/src/gallium/drivers/i915/i915_state_static.c
+++ b/src/gallium/drivers/i915/i915_state_static.c
@@ -28,6 +28,7 @@
 #include "i915_context.h"
 #include "i915_state.h"
 #include "i915_resource.h"
+#include "i915_screen.h"
 
 
 
@@ -78,38 +79,13 @@ buf_3d_tiling_bits(enum i915_winsys_buffer_tile tiling)
    return tiling_bits;
 }
 
-/**
- * Examine framebuffer state to determine width, height.
- */
-static boolean
-framebuffer_size(const struct pipe_framebuffer_state *fb,
-                 uint *width, uint *height)
-{
-   if (fb->cbufs[0]) {
-      *width = fb->cbufs[0]->width;
-      *height = fb->cbufs[0]->height;
-      return TRUE;
-   }
-   else if (fb->zsbuf) {
-      *width = fb->zsbuf->width;
-      *height = fb->zsbuf->height;
-      return TRUE;
-   }
-   else {
-      *width = *height = 0;
-      return FALSE;
-   }
-}
-
 static void update_framebuffer(struct i915_context *i915)
 {
    struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0];
    struct pipe_surface *depth_surface = i915->framebuffer.zsbuf;
-   unsigned cformat, zformat;
-   unsigned x, y, w, h;
+   unsigned x, y;
    int layer;
-   uint32_t draw_offset;
-   boolean ret;
+   uint32_t draw_offset, draw_size;
 
    if (cbuf_surface) {
       struct i915_texture *tex = i915_texture(cbuf_surface->texture);
@@ -119,7 +95,6 @@ static void update_framebuffer(struct i915_context *i915)
       i915->current.cbuf_flags = BUF_3D_ID_COLOR_BACK |
                                  BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
                                  buf_3d_tiling_bits(tex->tiling);
-      cformat = cbuf_surface->format;
 
       layer = cbuf_surface->u.tex.first_layer;
 
@@ -127,10 +102,9 @@ static void update_framebuffer(struct i915_context *i915)
       y = tex->image_offset[cbuf_surface->u.tex.level][layer].nblocksy;
    } else {
       i915->current.cbuf_bo = NULL;
-      cformat = PIPE_FORMAT_B8G8R8A8_UNORM; /* arbitrary */
       x = y = 0;
    }
-   cformat = translate_format(cformat);
+   i915->static_dirty |= I915_DST_BUF_COLOR;
 
    /* What happens if no zbuf??
     */
@@ -145,28 +119,23 @@ static void update_framebuffer(struct i915_context *i915)
       i915->current.depth_flags = BUF_3D_ID_DEPTH |
                                   BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
                                   buf_3d_tiling_bits(tex->tiling);
-      zformat = translate_depth_format(depth_surface->format);
-   } else {
+   } else
       i915->current.depth_bo = NULL;
-      zformat = 0;
-   }
-
-   i915->current.dst_buf_vars = DSTORG_HORT_BIAS(0x8) | /* .5 */
-                                DSTORG_VERT_BIAS(0x8) | /* .5 */
-                                LOD_PRECLAMP_OGL |
-                                TEX_DEFAULT_COLOR_OGL |
-                                cformat |
-                                zformat;
+   i915->static_dirty |= I915_DST_BUF_DEPTH;
 
    /* drawing rect calculations */
    draw_offset = x | (y << 16);
-   ret = framebuffer_size(&i915->framebuffer, &w, &h);
-   assert(ret);
+   draw_size = (i915->framebuffer.width - 1 + x) |
+               ((i915->framebuffer.height - 1 + y) << 16);
    if (i915->current.draw_offset != draw_offset) {
       i915->current.draw_offset = draw_offset;
       i915_set_flush_dirty(i915, I915_PIPELINE_FLUSH);
+      i915->static_dirty |= I915_DST_RECT;
+   }
+   if (i915->current.draw_size != draw_size) {
+      i915->current.draw_size = draw_size;
+      i915->static_dirty |= I915_DST_RECT;
    }
-   i915->current.draw_size = (w - 1 + x) | ((h - 1 + y) << 16);
 
    i915->hardware_dirty |= I915_HW_STATIC;
 
@@ -179,3 +148,52 @@ struct i915_tracked_state i915_hw_framebuffer = {
    update_framebuffer,
    I915_NEW_FRAMEBUFFER
 };
+
+static void update_dst_buf_vars(struct i915_context *i915)
+{
+   struct pipe_surface *cbuf_surface = i915->framebuffer.cbufs[0];
+   struct pipe_surface *depth_surface = i915->framebuffer.zsbuf;
+   uint32_t dst_buf_vars, cformat, zformat;
+   uint32_t early_z = 0;
+
+   if (cbuf_surface)
+      cformat = cbuf_surface->format;
+   else
+      cformat = PIPE_FORMAT_B8G8R8A8_UNORM; /* arbitrary */
+   cformat = translate_format(cformat);
+
+   if (depth_surface) {
+      struct i915_texture *tex = i915_texture(depth_surface->texture);
+      struct i915_screen *is = i915_screen(i915->base.screen);
+
+      zformat = translate_depth_format(depth_surface->format);
+
+      if (is->is_i945 && tex->tiling != I915_TILE_NONE
+            && !i915->fs->info.writes_z)
+         early_z = CLASSIC_EARLY_DEPTH;
+   } else
+      zformat = 0;
+
+   dst_buf_vars = DSTORG_HORT_BIAS(0x8) | /* .5 */
+                  DSTORG_VERT_BIAS(0x8) | /* .5 */
+                  LOD_PRECLAMP_OGL |
+                  TEX_DEFAULT_COLOR_OGL |
+                  cformat |
+                  zformat |
+                  early_z;
+
+   if (i915->current.dst_buf_vars != dst_buf_vars) {
+      if (early_z != (i915->current.dst_buf_vars & CLASSIC_EARLY_DEPTH))
+         i915_set_flush_dirty(i915, I915_PIPELINE_FLUSH);
+
+      i915->current.dst_buf_vars = dst_buf_vars;
+      i915->static_dirty |= I915_DST_VARS;
+      i915->hardware_dirty |= I915_HW_STATIC;
+   }
+}
+
+struct i915_tracked_state i915_hw_dst_buf_vars = {
+   "dst buf vars",
+   update_dst_buf_vars,
+   I915_NEW_FRAMEBUFFER | I915_NEW_FS
+};
diff --git a/src/gallium/drivers/i915/i915_surface.c b/src/gallium/drivers/i915/i915_surface.c
index becc6e93c2d..d02c420f6c2 100644
--- a/src/gallium/drivers/i915/i915_surface.c
+++ b/src/gallium/drivers/i915/i915_surface.c
@@ -27,6 +27,7 @@
 
 #include "i915_surface.h"
 #include "i915_resource.h"
+#include "i915_state.h"
 #include "i915_blit.h"
 #include "i915_reg.h"
 #include "i915_screen.h"
@@ -37,16 +38,119 @@
 #include "util/u_memory.h"
 #include "util/u_pack_color.h"
 
+/*
+ * surface functions using the render engine
+ */
+
+static void
+i915_surface_copy_render(struct pipe_context *pipe,
+                         struct pipe_resource *dst, unsigned dst_level,
+                         unsigned dstx, unsigned dsty, unsigned dstz,
+                         struct pipe_resource *src, unsigned src_level,
+                         const struct pipe_box *src_box)
+{
+   struct i915_context *i915 = i915_context(pipe);
+
+   util_blitter_save_blend(i915->blitter, (void *)i915->blend);
+   util_blitter_save_depth_stencil_alpha(i915->blitter, (void *)i915->depth_stencil);
+   util_blitter_save_stencil_ref(i915->blitter, &i915->stencil_ref);
+   util_blitter_save_rasterizer(i915->blitter, (void *)i915->rasterizer);
+   util_blitter_save_fragment_shader(i915->blitter, i915->saved_fs);
+   util_blitter_save_vertex_shader(i915->blitter, i915->saved_vs);
+   util_blitter_save_viewport(i915->blitter, &i915->viewport);
+   util_blitter_save_clip(i915->blitter, &i915->saved_clip);
+   util_blitter_save_vertex_elements(i915->blitter, i915->saved_velems);
+   util_blitter_save_vertex_buffers(i915->blitter, i915->saved_nr_vertex_buffers,
+                                    i915->saved_vertex_buffers);
+
+   util_blitter_save_framebuffer(i915->blitter, &i915->framebuffer);
+
+   util_blitter_save_fragment_sampler_states(i915->blitter,
+                                             i915->saved_nr_samplers,
+                                             i915->saved_samplers);
+   util_blitter_save_fragment_sampler_views(i915->blitter,
+                                            i915->saved_nr_sampler_views,
+                                            i915->saved_sampler_views);
+
+   util_blitter_copy_region(i915->blitter, dst, dst_level, dstx, dsty, dstz,
+                            src, src_level, src_box, TRUE);
+}
+
+static void
+i915_clear_render_target_render(struct pipe_context *pipe,
+                                struct pipe_surface *dst,
+                                const float *rgba,
+                                unsigned dstx, unsigned dsty,
+                                unsigned width, unsigned height)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   struct pipe_framebuffer_state fb_state;
+
+   util_blitter_save_framebuffer(i915->blitter, &i915->framebuffer);
+
+   fb_state.width = dst->width;
+   fb_state.height = dst->height;
+   fb_state.nr_cbufs = 1;
+   fb_state.cbufs[0] = dst;
+   fb_state.zsbuf = NULL;
+   pipe->set_framebuffer_state(pipe, &fb_state);
+
+   if (i915->dirty)
+      i915_update_derived(i915);
+
+   i915_clear_emit(pipe, PIPE_CLEAR_COLOR, rgba, 0.0, 0x0,
+                   dstx, dsty, width, height);
+
+   pipe->set_framebuffer_state(pipe, &i915->blitter->saved_fb_state);
+   util_unreference_framebuffer_state(&i915->blitter->saved_fb_state);
+   i915->blitter->saved_fb_state.nr_cbufs = ~0;
+}
+
+static void
+i915_clear_depth_stencil_render(struct pipe_context *pipe,
+                                struct pipe_surface *dst,
+                                unsigned clear_flags,
+                                double depth,
+                                unsigned stencil,
+                                unsigned dstx, unsigned dsty,
+                                unsigned width, unsigned height)
+{
+   struct i915_context *i915 = i915_context(pipe);
+   struct pipe_framebuffer_state fb_state;
+
+   util_blitter_save_framebuffer(i915->blitter, &i915->framebuffer);
+
+   fb_state.width = dst->width;
+   fb_state.height = dst->height;
+   fb_state.nr_cbufs = 0;
+   fb_state.zsbuf = dst;
+   pipe->set_framebuffer_state(pipe, &fb_state);
+
+   if (i915->dirty)
+      i915_update_derived(i915);
+
+   i915_clear_emit(pipe, clear_flags & PIPE_CLEAR_DEPTHSTENCIL,
+                   NULL, depth, stencil,
+                   dstx, dsty, width, height);
+
+   pipe->set_framebuffer_state(pipe, &i915->blitter->saved_fb_state);
+   util_unreference_framebuffer_state(&i915->blitter->saved_fb_state);
+   i915->blitter->saved_fb_state.nr_cbufs = ~0;
+}
+
+/*
+ * surface functions using the blitter
+ */
 
 /* Assumes all values are within bounds -- no checking at this level -
  * do it higher up if required.
  */
 static void
-i915_surface_copy(struct pipe_context *pipe,
-                  struct pipe_resource *dst, unsigned dst_level,
-                  unsigned dstx, unsigned dsty, unsigned dstz,
-                  struct pipe_resource *src, unsigned src_level,
-                  const struct pipe_box *src_box)
+i915_surface_copy_blitter(struct pipe_context *pipe,
+                          struct pipe_resource *dst, unsigned dst_level,
+                          unsigned dstx, unsigned dsty, unsigned dstz,
+                          struct pipe_resource *src, unsigned src_level,
+                          const struct pipe_box *src_box)
 {
    struct i915_texture *dst_tex = i915_texture(dst);
    struct i915_texture *src_tex = i915_texture(src);
@@ -66,7 +170,6 @@ i915_surface_copy(struct pipe_context *pipe,
       assert(src_box->z == 0);
    src_offset = i915_texture_offset(src_tex, src_level, src_box->z);
 
-   assert( dst != src );
    assert( util_format_get_blocksize(dpt->format) == util_format_get_blocksize(spt->format) );
    assert( util_format_get_blockwidth(dpt->format) == util_format_get_blockwidth(spt->format) );
    assert( util_format_get_blockheight(dpt->format) == util_format_get_blockheight(spt->format) );
@@ -81,13 +184,12 @@ i915_surface_copy(struct pipe_context *pipe,
                    (short) src_box->width, (short) src_box->height );
 }
 
-
 static void
-i915_clear_render_target(struct pipe_context *pipe,
-                         struct pipe_surface *dst,
-                         const float *rgba,
-                         unsigned dstx, unsigned dsty,
-                         unsigned width, unsigned height)
+i915_clear_render_target_blitter(struct pipe_context *pipe,
+                                 struct pipe_surface *dst,
+                                 const float *rgba,
+                                 unsigned dstx, unsigned dsty,
+                                 unsigned width, unsigned height)
 {
    struct i915_texture *tex = i915_texture(dst->texture);
    struct pipe_resource *pt = &tex->b.b;
@@ -109,13 +211,13 @@ i915_clear_render_target(struct pipe_context *pipe,
 }
 
 static void
-i915_clear_depth_stencil(struct pipe_context *pipe,
-                         struct pipe_surface *dst,
-                         unsigned clear_flags,
-                         double depth,
-                         unsigned stencil,
-                         unsigned dstx, unsigned dsty,
-                         unsigned width, unsigned height)
+i915_clear_depth_stencil_blitter(struct pipe_context *pipe,
+                                 struct pipe_surface *dst,
+                                 unsigned clear_flags,
+                                 double depth,
+                                 unsigned stencil,
+                                 unsigned dstx, unsigned dsty,
+                                 unsigned width, unsigned height)
 {
    struct i915_texture *tex = i915_texture(dst->texture);
    struct pipe_resource *pt = &tex->b.b;
@@ -193,9 +295,15 @@ i915_surface_destroy(struct pipe_context *ctx,
 void
 i915_init_surface_functions(struct i915_context *i915)
 {
-   i915->base.resource_copy_region = i915_surface_copy;
-   i915->base.clear_render_target = i915_clear_render_target;
-   i915->base.clear_depth_stencil = i915_clear_depth_stencil;
+   if (i915_screen(i915->base.screen)->debug.use_blitter) {
+      i915->base.resource_copy_region = i915_surface_copy_blitter;
+      i915->base.clear_render_target = i915_clear_render_target_blitter;
+      i915->base.clear_depth_stencil = i915_clear_depth_stencil_blitter;
+   } else {
+      i915->base.resource_copy_region = i915_surface_copy_render;
+      i915->base.clear_render_target = i915_clear_render_target_render;
+      i915->base.clear_depth_stencil = i915_clear_depth_stencil_render;
+   }
    i915->base.create_surface = i915_create_surface;
    i915->base.surface_destroy = i915_surface_destroy;
 }