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

79 files changed, 1843 insertions, 902 deletions

diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
index e40f546929d..cc16bcfb539 100644
--- a/src/gallium/auxiliary/Makefile
+++ b/src/gallium/auxiliary/Makefile
@@ -111,6 +111,7 @@ C_SOURCES = \
 	util/u_format.c \
 	util/u_format_other.c \
 	util/u_format_s3tc.c \
+	util/u_format_rgtc.c \
 	util/u_format_srgb.c \
 	util/u_format_table.c \
 	util/u_format_tests.c \
diff --git a/src/gallium/auxiliary/SConscript b/src/gallium/auxiliary/SConscript
index 11024d41923..f4ab8a50da4 100644
--- a/src/gallium/auxiliary/SConscript
+++ b/src/gallium/auxiliary/SConscript
@@ -158,6 +158,7 @@ source = [
     'util/u_format.c',
     'util/u_format_other.c',
     'util/u_format_s3tc.c',
+    'util/u_format_rgtc.c',
     'util/u_format_srgb.c',
     'util/u_format_table.c',
     'util/u_format_tests.c',
diff --git a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
index 5ea552f51c1..60f6380c503 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
@@ -874,6 +874,8 @@ draw_install_aapoint_stage(struct draw_context *draw,
 {
    struct aapoint_stage *aapoint;
 
+   pipe->draw = (void *) draw;
+
    /*
     * Create / install AA point drawing / prim stage
     */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index cbb090b2803..6576c74fbde 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -672,6 +672,31 @@ tgsi_exec_machine_bind_shader(
    mach->Processor = parse.FullHeader.Processor.Processor;
    mach->ImmLimit = 0;
 
+   if (mach->Processor == TGSI_PROCESSOR_GEOMETRY &&
+       !mach->UsedGeometryShader) {
+      struct tgsi_exec_vector *inputs =
+         align_malloc(sizeof(struct tgsi_exec_vector) *
+                      TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS,
+                      16);
+      struct tgsi_exec_vector *outputs =
+         align_malloc(sizeof(struct tgsi_exec_vector) *
+                      TGSI_MAX_TOTAL_VERTICES, 16);
+
+      if (!inputs)
+         return;
+      if (!outputs) {
+         align_free(inputs);
+         return;
+      }
+
+      align_free(mach->Inputs);
+      align_free(mach->Outputs);
+
+      mach->Inputs = inputs;
+      mach->Outputs = outputs;
+      mach->UsedGeometryShader = TRUE;
+   }
+
    declarations = (struct tgsi_full_declaration *)
       MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
 
@@ -801,6 +826,11 @@ tgsi_exec_machine_create( void )
    mach->MaxGeometryShaderOutputs = TGSI_MAX_TOTAL_VERTICES;
    mach->Predicates = &mach->Temps[TGSI_EXEC_TEMP_P0];
 
+   mach->Inputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16);
+   mach->Outputs = align_malloc(sizeof(struct tgsi_exec_vector) * PIPE_MAX_ATTRIBS, 16);
+   if (!mach->Inputs || !mach->Outputs)
+      goto fail;
+
    /* Setup constants needed by the SSE2 executor. */
    for( i = 0; i < 4; i++ ) {
       mach->Temps[TGSI_EXEC_TEMP_00000000_I].xyzw[TGSI_EXEC_TEMP_00000000_C].u[i] = 0x00000000;
@@ -824,7 +854,11 @@ tgsi_exec_machine_create( void )
    return mach;
 
 fail:
-   align_free(mach);
+   if (mach) {
+      align_free(mach->Inputs);
+      align_free(mach->Outputs);
+      align_free(mach);
+   }
    return NULL;
 }
 
@@ -836,10 +870,13 @@ tgsi_exec_machine_destroy(struct tgsi_exec_machine *mach)
       if (mach->Instructions)
          FREE(mach->Instructions);
       if (mach->Declarations)
-      FREE(mach->Declarations);
-   }
+         FREE(mach->Declarations);
+
+      align_free(mach->Inputs);
+      align_free(mach->Outputs);
 
-   align_free(mach);
+      align_free(mach);
+   }
 }
 
 static void
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
index 4a423b5bb4e..33f33aa82c7 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.h
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -228,8 +228,8 @@ struct tgsi_exec_machine
 
    float                         ImmArray[TGSI_EXEC_NUM_IMMEDIATES][4];
 
-   struct tgsi_exec_vector       Inputs[TGSI_MAX_PRIM_VERTICES * PIPE_MAX_ATTRIBS];
-   struct tgsi_exec_vector       Outputs[TGSI_MAX_TOTAL_VERTICES];
+   struct tgsi_exec_vector       *Inputs;
+   struct tgsi_exec_vector       *Outputs;
 
    /* System values */
    unsigned                      SysSemanticToIndex[TGSI_SEMANTIC_COUNT];
@@ -309,6 +309,8 @@ struct tgsi_exec_machine
    uint NumDeclarations;
 
    struct tgsi_declaration_resource Resources[PIPE_MAX_SHADER_RESOURCES];
+
+   boolean UsedGeometryShader;
 };
 
 struct tgsi_exec_machine *
diff --git a/src/gallium/auxiliary/util/u_format_pack.py b/src/gallium/auxiliary/util/u_format_pack.py
index 6d0016c0ad8..cc173f808ae 100644
--- a/src/gallium/auxiliary/util/u_format_pack.py
+++ b/src/gallium/auxiliary/util/u_format_pack.py
@@ -632,7 +632,7 @@ def generate_format_fetch(format, dst_channel, dst_native_type, dst_suffix):
 
 
 def is_format_hand_written(format):
-    return format.layout in ('s3tc', 'subsampled', 'other') or format.colorspace == ZS
+    return format.layout in ('s3tc', 'rgtc', 'subsampled', 'other') or format.colorspace == ZS
 
 
 def generate(formats):
diff --git a/src/gallium/auxiliary/util/u_format_rgtc.c b/src/gallium/auxiliary/util/u_format_rgtc.c
new file mode 100644
index 00000000000..6ffcd7e99ef
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_rgtc.c
@@ -0,0 +1,452 @@
+/**************************************************************************
+ *
+ * Copyright (C) 2011 Red Hat Inc.
+ *
+ * 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, sublicense,
+ * 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 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL 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 "u_math.h"
+#include "u_format.h"
+#include "u_format_rgtc.h"
+
+static void u_format_unsigned_encode_rgtc_chan(uint8_t *blkaddr, uint8_t srccolors[4][4],
+					       int numxpixels, int numypixels);
+
+static void u_format_unsigned_fetch_texel_rgtc(unsigned srcRowStride, const uint8_t *pixdata,
+					       unsigned i, unsigned j, uint8_t *value, unsigned comps);
+
+static void u_format_signed_encode_rgtc_chan(int8_t *blkaddr, int8_t srccolors[4][4],
+					     int numxpixels, int numypixels);
+
+static void u_format_signed_fetch_texel_rgtc(unsigned srcRowStride, const int8_t *pixdata,
+					       unsigned i, unsigned j, int8_t *value, unsigned comps);
+
+void
+util_format_rgtc1_unorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   u_format_unsigned_fetch_texel_rgtc(0, src, i, j, dst, 1);
+}
+
+void
+util_format_rgtc1_unorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, comps = 4;
+   unsigned x, y, i, j;
+   unsigned block_size = 8;
+
+   for(y = 0; y < height; y += bh) {
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += bw) {
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+               uint8_t *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*comps;
+	       u_format_unsigned_fetch_texel_rgtc(0, src, i, j, dst, 1);
+	    }
+	 }
+	 src += block_size;
+      }
+      src_row += src_stride;
+   }
+}
+
+void
+util_format_rgtc1_unorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, 
+					 unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, bytes_per_block = 8;
+   unsigned x, y, i, j;
+
+   for(y = 0; y < height; y += bh) {
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp[4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+	       tmp[j][i] = src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4];
+            }
+         }
+         u_format_unsigned_encode_rgtc_chan(dst, tmp, 4, 4);
+         dst += bytes_per_block;
+      }
+      dst_row += dst_stride / sizeof(*dst_row);
+   }
+}
+
+void
+util_format_rgtc1_unorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   unsigned x, y, i, j;
+   int block_size = 8;
+   for(y = 0; y < height; y += 4) {
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += 4) {
+         for(j = 0; j < 4; ++j) {
+            for(i = 0; i < 4; ++i) {
+               float *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+               uint8_t tmp_r;
+               u_format_unsigned_fetch_texel_rgtc(0, src, i, j, &tmp_r, 1);
+               dst[0] = ubyte_to_float(tmp_r);
+               dst[1] = 0.0;
+               dst[2] = 0.0;
+               dst[3] = 1.0;
+            }
+         }
+         src += block_size;
+      }
+      src_row += src_stride;
+   }
+}
+
+void
+util_format_rgtc1_unorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, bytes_per_block = 8;
+   unsigned x, y, i, j;
+
+   for(y = 0; y < height; y += bh) {
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp[4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+	       tmp[j][i] = float_to_ubyte(src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4]);
+            }
+         }
+         u_format_unsigned_encode_rgtc_chan(dst, tmp, 4, 4);
+         dst += bytes_per_block;
+      }
+      dst_row += dst_stride / sizeof(*dst_row);
+   }
+}
+
+void
+util_format_rgtc1_unorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   uint8_t tmp_r;
+   u_format_unsigned_fetch_texel_rgtc(0, src, i, j, &tmp_r, 1);
+   dst[0] = ubyte_to_float(tmp_r);
+   dst[1] = 0.0;
+   dst[2] = 0.0;
+   dst[3] = 1.0;
+}
+
+void
+util_format_rgtc1_snorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   fprintf(stderr,"%s\n", __func__);
+}
+
+void
+util_format_rgtc1_snorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   fprintf(stderr,"%s\n", __func__);
+}
+
+void
+util_format_rgtc1_snorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   fprintf(stderr,"%s\n", __func__);
+}
+
+void
+util_format_rgtc1_snorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, bytes_per_block = 8;
+   unsigned x, y, i, j;
+
+   for(y = 0; y < height; y += bh) {
+      int8_t *dst = (int8_t *)dst_row;
+      for(x = 0; x < width; x += bw) {
+         int8_t tmp[4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+	       tmp[j][i] = float_to_byte_tex(src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4]);
+            }
+         }
+         u_format_signed_encode_rgtc_chan(dst, tmp, 4, 4);
+         dst += bytes_per_block;
+      }
+      dst_row += dst_stride / sizeof(*dst_row);
+   }
+}
+
+void
+util_format_rgtc1_snorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   unsigned x, y, i, j;
+   int block_size = 8;
+   for(y = 0; y < height; y += 4) {
+      const int8_t *src = (int8_t *)src_row;
+      for(x = 0; x < width; x += 4) {
+         for(j = 0; j < 4; ++j) {
+            for(i = 0; i < 4; ++i) {
+               float *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+               int8_t tmp_r;
+               u_format_signed_fetch_texel_rgtc(0, src, i, j, &tmp_r, 1);
+               dst[0] = byte_to_float_tex(tmp_r);
+               dst[1] = 0.0;
+               dst[2] = 0.0;
+               dst[3] = 1.0;
+            }
+         }
+         src += block_size;
+      }
+      src_row += src_stride;
+   }
+}
+
+void
+util_format_rgtc1_snorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   int8_t tmp_r;
+   u_format_signed_fetch_texel_rgtc(0, (int8_t *)src, i, j, &tmp_r, 1);
+   dst[0] = byte_to_float_tex(tmp_r);
+   dst[1] = 0.0;
+   dst[2] = 0.0;
+   dst[3] = 1.0;
+}
+
+
+void
+util_format_rgtc2_unorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   u_format_unsigned_fetch_texel_rgtc(0, src, i, j, dst, 2);
+   u_format_unsigned_fetch_texel_rgtc(0, src + 8, i, j, dst + 1, 2);
+}
+
+void
+util_format_rgtc2_unorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, comps = 4;
+   unsigned x, y, i, j;
+   unsigned block_size = 16;
+
+   for(y = 0; y < height; y += bh) {
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += bw) {
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+               uint8_t *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*comps;
+	       u_format_unsigned_fetch_texel_rgtc(0, src, i, j, dst, 2);
+	       u_format_unsigned_fetch_texel_rgtc(0, src + 8, i, j, dst + 1, 2);
+
+	    }
+	 }
+	 src += block_size;
+      }
+      src_row += src_stride;
+   }
+}
+
+void
+util_format_rgtc2_unorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, bytes_per_block = 16;
+   unsigned x, y, i, j;
+
+   for(y = 0; y < height; y += bh) {
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp_r[4][4];  /* [bh][bw] */
+         uint8_t tmp_g[4][4];  /* [bh][bw] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+	       tmp_r[j][i] = src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4];
+	       tmp_g[j][i] = src_row[((y + j)*src_stride/sizeof(*src_row) + (x + i)*4) + 1];
+            }
+         }
+         u_format_unsigned_encode_rgtc_chan(dst, tmp_r, 4, 4);
+         u_format_unsigned_encode_rgtc_chan(dst + 8, tmp_g, 4, 4);
+         dst += bytes_per_block;
+      }
+      dst_row += dst_stride / sizeof(*dst_row);
+   }
+}
+
+void
+util_format_rgtc2_unorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, bytes_per_block = 16;
+   unsigned x, y, i, j;
+
+   for(y = 0; y < height; y += bh) {
+      uint8_t *dst = dst_row;
+      for(x = 0; x < width; x += bw) {
+         uint8_t tmp_r[4][4];  /* [bh][bw][comps] */
+         uint8_t tmp_g[4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+	       tmp_r[j][i] = float_to_ubyte(src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4]);
+	       tmp_g[j][i] = float_to_ubyte(src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4 + 1]);
+            }
+         }
+         u_format_unsigned_encode_rgtc_chan(dst, tmp_r, 4, 4);
+         u_format_unsigned_encode_rgtc_chan(dst + 8, tmp_g, 4, 4);
+         dst += bytes_per_block;
+      }
+      dst_row += dst_stride / sizeof(*dst_row);
+   }
+}
+
+void
+util_format_rgtc2_unorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   unsigned x, y, i, j;
+   int block_size = 16;
+   for(y = 0; y < height; y += 4) {
+      const uint8_t *src = src_row;
+      for(x = 0; x < width; x += 4) {
+         for(j = 0; j < 4; ++j) {
+            for(i = 0; i < 4; ++i) {
+               float *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+               uint8_t tmp_r, tmp_g;
+               u_format_unsigned_fetch_texel_rgtc(0, src, i, j, &tmp_r, 2);
+               u_format_unsigned_fetch_texel_rgtc(0, src + 8, i, j, &tmp_g, 2);
+               dst[0] = ubyte_to_float(tmp_r);
+               dst[1] = ubyte_to_float(tmp_g);
+               dst[2] = 0.0;
+               dst[3] = 1.0;
+            }
+         }
+         src += block_size;
+      }
+      src_row += src_stride;
+   }
+}
+
+void
+util_format_rgtc2_unorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   uint8_t tmp_r, tmp_g;
+   u_format_unsigned_fetch_texel_rgtc(0, src, i, j, &tmp_r, 2);
+   u_format_unsigned_fetch_texel_rgtc(0, src + 8, i, j, &tmp_g, 2);
+   dst[0] = ubyte_to_float(tmp_r);
+   dst[1] = ubyte_to_float(tmp_g);
+   dst[2] = 0.0;
+   dst[3] = 1.0;
+}
+
+
+void
+util_format_rgtc2_snorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   fprintf(stderr,"%s\n", __func__);
+}
+
+void
+util_format_rgtc2_snorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   fprintf(stderr,"%s\n", __func__);
+}
+
+void
+util_format_rgtc2_snorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   fprintf(stderr,"%s\n", __func__);
+}
+
+void
+util_format_rgtc2_snorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   unsigned x, y, i, j;
+   int block_size = 16;
+   for(y = 0; y < height; y += 4) {
+      const int8_t *src = (int8_t *)src_row;
+      for(x = 0; x < width; x += 4) {
+         for(j = 0; j < 4; ++j) {
+            for(i = 0; i < 4; ++i) {
+               float *dst = dst_row + (y + j)*dst_stride/sizeof(*dst_row) + (x + i)*4;
+               int8_t tmp_r, tmp_g;
+               u_format_signed_fetch_texel_rgtc(0, src, i, j, &tmp_r, 2);
+               u_format_signed_fetch_texel_rgtc(0, src + 8, i, j, &tmp_g, 2);
+               dst[0] = byte_to_float_tex(tmp_r);
+               dst[1] = byte_to_float_tex(tmp_g);
+               dst[2] = 0.0;
+               dst[3] = 1.0;
+            }
+         }
+         src += block_size;
+      }
+      src_row += src_stride;
+   }
+}
+
+void
+util_format_rgtc2_snorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height)
+{
+   const unsigned bw = 4, bh = 4, bytes_per_block = 16;
+   unsigned x, y, i, j;
+
+   for(y = 0; y < height; y += bh) {
+      int8_t *dst = (int8_t *)dst_row;
+      for(x = 0; x < width; x += bw) {
+         int8_t tmp_r[4][4];  /* [bh][bw][comps] */
+         int8_t tmp_g[4][4];  /* [bh][bw][comps] */
+         for(j = 0; j < bh; ++j) {
+            for(i = 0; i < bw; ++i) {
+	       tmp_r[j][i] = float_to_byte_tex(src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4]);
+	       tmp_g[j][i] = float_to_byte_tex(src_row[(y + j)*src_stride/sizeof(*src_row) + (x + i)*4 + 1]);
+            }
+         }
+         u_format_signed_encode_rgtc_chan(dst, tmp_r, 4, 4);
+         u_format_signed_encode_rgtc_chan(dst + 8, tmp_g, 4, 4);
+         dst += bytes_per_block;
+      }
+      dst_row += dst_stride / sizeof(*dst_row);
+   }
+}
+
+void
+util_format_rgtc2_snorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j)
+{
+   int8_t tmp_r, tmp_g;
+   u_format_signed_fetch_texel_rgtc(0, (int8_t *)src, i, j, &tmp_r, 2);
+   u_format_signed_fetch_texel_rgtc(0, (int8_t *)src + 8, i, j, &tmp_g, 2);
+   dst[0] = byte_to_float_tex(tmp_r);
+   dst[1] = byte_to_float_tex(tmp_g);
+   dst[2] = 0.0;
+   dst[3] = 1.0;
+}
+
+
+#define TAG(x) u_format_unsigned_##x
+#define TYPE uint8_t
+#define T_MIN 0
+#define T_MAX 255
+
+#include "../../../mesa/main/texcompress_rgtc_tmp.h"
+
+#undef TYPE
+#undef TAG
+#undef T_MIN
+#undef T_MAX
+
+
+#define TAG(x) u_format_signed_##x
+#define TYPE int8_t
+#define T_MIN (int8_t)-128
+#define T_MAX (int8_t)127
+
+#include "../../../mesa/main/texcompress_rgtc_tmp.h"
+
+#undef TYPE
+#undef TAG
+#undef T_MIN
+#undef T_MAX
diff --git a/src/gallium/auxiliary/util/u_format_rgtc.h b/src/gallium/auxiliary/util/u_format_rgtc.h
new file mode 100644
index 00000000000..3e8636d110c
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_format_rgtc.h
@@ -0,0 +1,108 @@
+/**************************************************************************
+ *
+ * Copyright 2011 Red Hat Inc.
+ * 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.
+ *
+ **************************************************************************/
+
+#ifndef U_FORMAT_RGTC_H_
+#define U_FORMAT_RGTC_H_
+
+void
+util_format_rgtc1_unorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j);
+
+void
+util_format_rgtc1_unorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_unorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_unorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_unorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_unorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j);
+
+
+
+void
+util_format_rgtc1_snorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j);
+
+void
+util_format_rgtc1_snorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_snorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_snorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_snorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc1_snorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j);
+
+
+void
+util_format_rgtc2_unorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j);
+
+void
+util_format_rgtc2_unorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_unorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_unorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_unorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_unorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j);
+
+
+void
+util_format_rgtc2_snorm_fetch_rgba_8unorm(uint8_t *dst, const uint8_t *src, unsigned i, unsigned j);
+
+void
+util_format_rgtc2_snorm_unpack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_snorm_pack_rgba_8unorm(uint8_t *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_snorm_unpack_rgba_float(float *dst_row, unsigned dst_stride, const uint8_t *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_snorm_pack_rgba_float(uint8_t *dst_row, unsigned dst_stride, const float *src_row, unsigned src_stride, unsigned width, unsigned height);
+
+void
+util_format_rgtc2_snorm_fetch_rgba_float(float *dst, const uint8_t *src, unsigned i, unsigned j);
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_format_table.py b/src/gallium/auxiliary/util/u_format_table.py
index 8cc22a56371..7468bc38b32 100755
--- a/src/gallium/auxiliary/util/u_format_table.py
+++ b/src/gallium/auxiliary/util/u_format_table.py
@@ -87,6 +87,7 @@ def write_format_table(formats):
     print
     print '#include "u_format.h"'
     print '#include "u_format_s3tc.h"'
+    print '#include "u_format_rgtc.h"'
     print
     
     u_format_pack.generate(formats)
@@ -132,7 +133,7 @@ def write_format_table(formats):
         if format.colorspace != ZS:
             print "   &util_format_%s_unpack_rgba_8unorm," % format.short_name() 
             print "   &util_format_%s_pack_rgba_8unorm," % format.short_name() 
-            if format.layout == 's3tc':
+            if format.layout == 's3tc' or format.layout == 'rgtc':
                 print "   &util_format_%s_fetch_rgba_8unorm," % format.short_name()
             else:
                 print "   NULL, /* fetch_rgba_8unorm */" 
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
index 30555f92a6d..e3d4c06b6f9 100644
--- a/src/gallium/auxiliary/util/u_math.h
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -475,6 +475,17 @@ float_to_ubyte(float f)
    }
 }
 
+static INLINE float
+byte_to_float_tex(int8_t b)
+{
+   return (b == -128) ? -1.0F : b * 1.0F / 127.0F;
+}
+
+static INLINE int8_t
+float_to_byte_tex(float f)
+{
+   return (int8_t) (127.0F * f);
+}
 
 /**
  * Calc log base 2
diff --git a/src/gallium/auxiliary/util/u_vbuf_mgr.c b/src/gallium/auxiliary/util/u_vbuf_mgr.c
index 3cf8ee0831d..7d157c99ccc 100644
--- a/src/gallium/auxiliary/util/u_vbuf_mgr.c
+++ b/src/gallium/auxiliary/util/u_vbuf_mgr.c
@@ -515,6 +515,7 @@ void u_vbuf_mgr_set_vertex_buffers(struct u_vbuf_mgr *mgrb,
 
 static void u_vbuf_upload_buffers(struct u_vbuf_mgr_priv *mgr,
                                   int min_index, int max_index,
+                                  unsigned instance_count,
                                   boolean *upload_flushed)
 {
    int i, nr = mgr->ve->count;
@@ -530,10 +531,12 @@ static void u_vbuf_upload_buffers(struct u_vbuf_mgr_priv *mgr,
           !uploaded[index]) {
          unsigned first, size;
          boolean flushed;
+         unsigned instance_div = mgr->ve->ve[i].instance_divisor;
 
-         if (mgr->ve->ve[i].instance_divisor) {
+         if (instance_div) {
             first = 0;
-            size = vb->buffer->width0;
+            size = vb->stride *
+                   ((instance_count + instance_div - 1) / instance_div);
          } else if (vb->stride) {
             first = vb->stride * min_index;
             size = vb->stride * count;
@@ -581,7 +584,8 @@ void u_vbuf_mgr_draw_begin(struct u_vbuf_mgr *mgrb,
 
    /* Upload user buffers. */
    if (mgr->any_user_vbs) {
-      u_vbuf_upload_buffers(mgr, min_index, max_index, &upload_flushed);
+      u_vbuf_upload_buffers(mgr, min_index, max_index, info->instance_count,
+                            &upload_flushed);
       bufs_updated = TRUE;
    }
 
diff --git a/src/gallium/drivers/i915/i915_batch.h b/src/gallium/drivers/i915/i915_batch.h
index 6e93da76209..039c8713570 100644
--- a/src/gallium/drivers/i915/i915_batch.h
+++ b/src/gallium/drivers/i915/i915_batch.h
@@ -31,8 +31,8 @@
 #include "i915_batchbuffer.h"
 
 
-#define BEGIN_BATCH(dwords, relocs) \
-   (i915_winsys_batchbuffer_check(i915->batch, dwords, relocs))
+#define BEGIN_BATCH(dwords) \
+   (i915_winsys_batchbuffer_check(i915->batch, dwords))
 
 #define OUT_BATCH(dword) \
    i915_winsys_batchbuffer_dword(i915->batch, dword)
diff --git a/src/gallium/drivers/i915/i915_batchbuffer.h b/src/gallium/drivers/i915/i915_batchbuffer.h
index b4a91dabb37..9df82272604 100644
--- a/src/gallium/drivers/i915/i915_batchbuffer.h
+++ b/src/gallium/drivers/i915/i915_batchbuffer.h
@@ -41,11 +41,9 @@ i915_winsys_batchbuffer_space(struct i915_winsys_batchbuffer *batch)
 
 static INLINE boolean
 i915_winsys_batchbuffer_check(struct i915_winsys_batchbuffer *batch,
-                              size_t dwords,
-                              size_t relocs)
+                              size_t dwords)
 {
-   return dwords * 4 <= i915_winsys_batchbuffer_space(batch) &&
-          relocs <= (batch->max_relocs - batch->relocs);
+   return dwords * 4 <= i915_winsys_batchbuffer_space(batch);
 }
 
 static INLINE void
@@ -71,7 +69,7 @@ i915_winsys_batchbuffer_write(struct i915_winsys_batchbuffer *batch,
 {
    assert (i915_winsys_batchbuffer_space(batch) >= size);
 
-   memcpy(data, batch->ptr, size);
+   memcpy(batch->ptr, data, size);
    batch->ptr += size;
 }
 
diff --git a/src/gallium/drivers/i915/i915_blit.c b/src/gallium/drivers/i915/i915_blit.c
index f885417f8ed..baaed3767ff 100644
--- a/src/gallium/drivers/i915/i915_blit.c
+++ b/src/gallium/drivers/i915/i915_blit.c
@@ -71,9 +71,9 @@ i915_fill_blit(struct i915_context *i915,
       return;
    }
 
-   if (!BEGIN_BATCH(6, 1)) {
+   if (!BEGIN_BATCH(6)) {
       FLUSH_BATCH(NULL);
-      assert(BEGIN_BATCH(6, 1));
+      assert(BEGIN_BATCH(6));
    }
    OUT_BATCH(CMD);
    OUT_BATCH(BR13);
@@ -143,9 +143,9 @@ i915_copy_blit(struct i915_context *i915,
     */
    assert (dst_pitch > 0 && src_pitch > 0);
 
-   if (!BEGIN_BATCH(8, 2)) {
+   if (!BEGIN_BATCH(8)) {
       FLUSH_BATCH(NULL);
-      assert(BEGIN_BATCH(8, 2));
+      assert(BEGIN_BATCH(8));
    }
    OUT_BATCH(CMD);
    OUT_BATCH(BR13);
diff --git a/src/gallium/drivers/i915/i915_context.c b/src/gallium/drivers/i915/i915_context.c
index cbf919754e5..84c8cb54436 100644
--- a/src/gallium/drivers/i915/i915_context.c
+++ b/src/gallium/drivers/i915/i915_context.c
@@ -39,7 +39,7 @@
 #include "pipe/p_screen.h"
 
 
-DEBUG_GET_ONCE_BOOL_OPTION(i915_no_vbuf, "I915_NO_VBUF", FALSE);
+DEBUG_GET_ONCE_BOOL_OPTION(i915_no_vbuf, "I915_NO_VBUF", FALSE)
 
 
 /*
diff --git a/src/gallium/drivers/i915/i915_flush.c b/src/gallium/drivers/i915/i915_flush.c
index 22a2c7b2cb4..f2044d661e3 100644
--- a/src/gallium/drivers/i915/i915_flush.c
+++ b/src/gallium/drivers/i915/i915_flush.c
@@ -59,9 +59,9 @@ static void i915_flush_pipe( struct pipe_context *pipe,
       if (flags & PIPE_FLUSH_TEXTURE_CACHE)
 	 flush |= FLUSH_MAP_CACHE;
 
-      if (!BEGIN_BATCH(1, 0)) {
+      if (!BEGIN_BATCH(1)) {
 	 FLUSH_BATCH(NULL);
-	 assert(BEGIN_BATCH(1, 0));
+	 assert(BEGIN_BATCH(1));
       }
       OUT_BATCH( flush );
    }
diff --git a/src/gallium/drivers/i915/i915_prim_emit.c b/src/gallium/drivers/i915/i915_prim_emit.c
index dd997e2cf48..276e33d4b9d 100644
--- a/src/gallium/drivers/i915/i915_prim_emit.c
+++ b/src/gallium/drivers/i915/i915_prim_emit.c
@@ -144,7 +144,7 @@ emit_prim( struct draw_stage *stage,
    vertex_size = i915->current.vertex_info.size * 4; /* in bytes */
    assert(vertex_size >= 12); /* never smaller than 12 bytes */
 
-   if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) {
+   if (!BEGIN_BATCH( 1 + nr * vertex_size / 4)) {
       FLUSH_BATCH(NULL);
 
       /* Make sure state is re-emitted after a flush: 
@@ -152,7 +152,7 @@ emit_prim( struct draw_stage *stage,
       i915_update_derived( i915 );
       i915_emit_hardware_state( i915 );
 
-      if (!BEGIN_BATCH( 1 + nr * vertex_size / 4, 0 )) {
+      if (!BEGIN_BATCH( 1 + nr * vertex_size / 4)) {
 	 assert(0);
 	 return;
       }
diff --git a/src/gallium/drivers/i915/i915_prim_vbuf.c b/src/gallium/drivers/i915/i915_prim_vbuf.c
index 3473c863970..fb4c0516dd8 100644
--- a/src/gallium/drivers/i915/i915_prim_vbuf.c
+++ b/src/gallium/drivers/i915/i915_prim_vbuf.c
@@ -465,7 +465,7 @@ draw_arrays_fallback(struct vbuf_render *render,
    if (i915->hardware_dirty)
       i915_emit_hardware_state(i915);
 
-   if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) {
+   if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) {
       FLUSH_BATCH(NULL);
 
       /* Make sure state is re-emitted after a flush:
@@ -474,7 +474,7 @@ draw_arrays_fallback(struct vbuf_render *render,
       i915_emit_hardware_state(i915);
       i915->vbo_flushed = 1;
 
-      if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) {
+      if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) {
          assert(0);
          goto out;
       }
@@ -514,7 +514,7 @@ i915_vbuf_render_draw_arrays(struct vbuf_render *render,
    if (i915->hardware_dirty)
       i915_emit_hardware_state(i915);
 
-   if (!BEGIN_BATCH(2, 0)) {
+   if (!BEGIN_BATCH(2)) {
       FLUSH_BATCH(NULL);
 
       /* Make sure state is re-emitted after a flush:
@@ -523,7 +523,7 @@ i915_vbuf_render_draw_arrays(struct vbuf_render *render,
       i915_emit_hardware_state(i915);
       i915->vbo_flushed = 1;
 
-      if (!BEGIN_BATCH(2, 0)) {
+      if (!BEGIN_BATCH(2)) {
          assert(0);
          goto out;
       }
@@ -635,7 +635,7 @@ i915_vbuf_render_draw_elements(struct vbuf_render *render,
    if (i915->hardware_dirty)
       i915_emit_hardware_state(i915);
 
-   if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) {
+   if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) {
       FLUSH_BATCH(NULL);
 
       /* Make sure state is re-emitted after a flush: 
@@ -644,7 +644,7 @@ i915_vbuf_render_draw_elements(struct vbuf_render *render,
       i915_emit_hardware_state(i915);
       i915->vbo_flushed = 1;
 
-      if (!BEGIN_BATCH(1 + (nr_indices + 1)/2, 1)) {
+      if (!BEGIN_BATCH(1 + (nr_indices + 1)/2)) {
          assert(0);
          goto out;
       }
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c
index 0323ad940f9..15350c0a5d7 100644
--- a/src/gallium/drivers/i915/i915_state_emit.c
+++ b/src/gallium/drivers/i915/i915_state_emit.c
@@ -40,13 +40,19 @@
 
 struct i915_tracked_hw_state {
    const char *name;
-   void (*validate)(struct i915_context *);
+   void (*validate)(struct i915_context *, unsigned *batch_space);
    void (*emit)(struct i915_context *);
    unsigned dirty, batch_space;
 };
 
 
 static void
+validate_flush(struct i915_context *i915, unsigned *batch_space)
+{
+   *batch_space = i915->flush_dirty ? 1 : 0;
+}
+
+static void
 emit_flush(struct i915_context *i915)
 {
    /* Cache handling is very cheap atm. State handling can request to flushes:
@@ -61,109 +67,343 @@ emit_flush(struct i915_context *i915)
       OUT_BATCH(MI_FLUSH | INHIBIT_FLUSH_RENDER_CACHE);
 }
 
+uint32_t invariant_state[] = {
+   _3DSTATE_AA_CMD | AA_LINE_ECAAR_WIDTH_ENABLE | AA_LINE_ECAAR_WIDTH_1_0 |
+             AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0,
+
+   _3DSTATE_DFLT_DIFFUSE_CMD, 0,
+
+   _3DSTATE_DFLT_SPEC_CMD, 0,
+
+   _3DSTATE_DFLT_Z_CMD, 0,
+
+   _3DSTATE_COORD_SET_BINDINGS |
+             CSB_TCB(0, 0) |
+             CSB_TCB(1, 1) |
+             CSB_TCB(2, 2) |
+             CSB_TCB(3, 3) |
+             CSB_TCB(4, 4) |
+             CSB_TCB(5, 5) |
+             CSB_TCB(6, 6) |
+             CSB_TCB(7, 7),
+
+   _3DSTATE_RASTER_RULES_CMD |
+             ENABLE_POINT_RASTER_RULE |
+             OGL_POINT_RASTER_RULE |
+             ENABLE_LINE_STRIP_PROVOKE_VRTX |
+             ENABLE_TRI_FAN_PROVOKE_VRTX |
+             LINE_STRIP_PROVOKE_VRTX(1) |
+             TRI_FAN_PROVOKE_VRTX(2) |
+             ENABLE_TEXKILL_3D_4D |
+             TEXKILL_4D,
+
+   _3DSTATE_DEPTH_SUBRECT_DISABLE,
+
+   /* disable indirect state for now
+    */
+   _3DSTATE_LOAD_INDIRECT | 0, 0};
+
 static void
-validate_immediate(struct i915_context *i915)
+emit_invariant(struct i915_context *i915)
 {
+   i915_winsys_batchbuffer_write(i915->batch, invariant_state,
+                                 Elements(invariant_state)*sizeof(uint32_t));
+}
+
+static void
+validate_immediate(struct i915_context *i915, unsigned *batch_space)
+{
+   unsigned dirty = (1 << I915_IMMEDIATE_S0 | 1 << I915_IMMEDIATE_S1 |
+                     1 << I915_IMMEDIATE_S2 | 1 << I915_IMMEDIATE_S3 |
+                     1 << I915_IMMEDIATE_S3 | 1 << I915_IMMEDIATE_S4 |
+                     1 << I915_IMMEDIATE_S5 | 1 << I915_IMMEDIATE_S6) &
+                    i915->immediate_dirty;
+
    if (i915->immediate_dirty & (1 << I915_IMMEDIATE_S0))
       i915->validation_buffers[i915->num_validation_buffers++] = i915->vbo;
+
+   *batch_space = 1 + util_bitcount(dirty);
 }
 
 static void
-validate_static(struct i915_context *i915)
+emit_immediate(struct i915_context *i915)
 {
-   if (i915->current.cbuf_bo)
+   /* remove unwatned bits and S7 */
+   unsigned dirty = (1 << I915_IMMEDIATE_S0 | 1 << I915_IMMEDIATE_S1 |
+                     1 << I915_IMMEDIATE_S2 | 1 << I915_IMMEDIATE_S3 |
+                     1 << I915_IMMEDIATE_S3 | 1 << I915_IMMEDIATE_S4 |
+                     1 << I915_IMMEDIATE_S5 | 1 << I915_IMMEDIATE_S6) &
+                    i915->immediate_dirty;
+   int i, num = util_bitcount(dirty);
+   assert(num && num <= I915_MAX_IMMEDIATE);
+
+   OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
+             dirty << 4 | (num - 1));
+
+   if (i915->immediate_dirty & (1 << I915_IMMEDIATE_S0)) {
+      if (i915->vbo)
+         OUT_RELOC(i915->vbo, I915_USAGE_VERTEX,
+                   i915->current.immediate[I915_IMMEDIATE_S0]);
+      else
+         OUT_BATCH(0);
+   }
+
+   for (i = 1; i < I915_MAX_IMMEDIATE; i++) {
+      if (dirty & (1 << i))
+         OUT_BATCH(i915->current.immediate[i]);
+   }
+}
+
+static void
+validate_dynamic(struct i915_context *i915, unsigned *batch_space)
+{
+   *batch_space = util_bitcount(i915->dynamic_dirty & ((1 << I915_MAX_DYNAMIC) - 1));
+}
+
+static void
+emit_dynamic(struct i915_context *i915)
+{
+   int i;
+   for (i = 0; i < I915_MAX_DYNAMIC; i++) {
+      if (i915->dynamic_dirty & (1 << i))
+         OUT_BATCH(i915->current.dynamic[i]);
+   }
+}
+
+static void
+validate_static(struct i915_context *i915, unsigned *batch_space)
+{
+   *batch_space = 2 + 5; /* including DRAW_RECT */
+
+   if (i915->current.cbuf_bo) {
       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->validation_buffers[i915->num_validation_buffers++]
          = i915->current.depth_bo;
+      *batch_space += 3;
+   }
 }
 
 static void
-validate_map(struct i915_context *i915)
+emit_static(struct i915_context *i915)
+{
+   if (i915->current.cbuf_bo) {
+      OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
+      OUT_BATCH(i915->current.cbuf_flags);
+      OUT_RELOC(i915->current.cbuf_bo,
+                I915_USAGE_RENDER,
+                0);
+   }
+
+   /* What happens if no zbuf??
+    */
+   if (i915->current.depth_bo) {
+      OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
+      OUT_BATCH(i915->current.depth_flags);
+      OUT_RELOC(i915->current.depth_bo,
+                I915_USAGE_RENDER,
+                0);
+   }
+
+   {
+      OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD);
+      OUT_BATCH(i915->current.dst_buf_vars);
+   }
+}
+
+static void
+validate_map(struct i915_context *i915, unsigned *batch_space)
 {
    const uint enabled = i915->current.sampler_enable_flags;
    uint unit;
    struct i915_texture *tex;
 
+   *batch_space = i915->current.sampler_enable_nr ?
+     2 + 3*i915->current.sampler_enable_nr : 0;
 
    for (unit = 0; unit < I915_TEX_UNITS; unit++) {
       if (enabled & (1 << unit)) {
-	 tex = i915_texture(i915->fragment_sampler_views[unit]->texture);
-	 i915->validation_buffers[i915->num_validation_buffers++] = tex->buffer;
+         tex = i915_texture(i915->fragment_sampler_views[unit]->texture);
+         i915->validation_buffers[i915->num_validation_buffers++] = tex->buffer;
       }
    }
 }
 
-const static struct i915_tracked_hw_state hw_atoms[] = {
-   { "flush", NULL, emit_flush, I915_HW_FLUSH, 1 },
-   { "immediate", validate_immediate, NULL, I915_HW_IMMEDIATE },
-   { "static", validate_static, NULL, I915_HW_STATIC },
-   { "map", validate_map, NULL, I915_HW_MAP }
-};
+static void
+emit_map(struct i915_context *i915)
+{
+   const uint nr = i915->current.sampler_enable_nr;
+   if (nr) {
+      const uint enabled = i915->current.sampler_enable_flags;
+      uint unit;
+      uint count = 0;
+      OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
+      OUT_BATCH(enabled);
+      for (unit = 0; unit < I915_TEX_UNITS; unit++) {
+         if (enabled & (1 << unit)) {
+            struct i915_texture *texture = i915_texture(i915->fragment_sampler_views[unit]->texture);
+            struct i915_winsys_buffer *buf = texture->buffer;
+            assert(buf);
+
+            count++;
+
+            OUT_RELOC(buf, I915_USAGE_SAMPLER, 0);
+            OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */
+            OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */
+         }
+      }
+      assert(count == nr);
+   }
+}
+
+static void
+validate_sampler(struct i915_context *i915, unsigned *batch_space)
+{
+   *batch_space = i915->current.sampler_enable_nr ?
+     2 + 3*i915->current.sampler_enable_nr : 0;
+}
+
+static void
+emit_sampler(struct i915_context *i915)
+{
+   if (i915->current.sampler_enable_nr) {
+      int i;
+
+      OUT_BATCH( _3DSTATE_SAMPLER_STATE | 
+                 (3 * i915->current.sampler_enable_nr) );
+
+      OUT_BATCH( i915->current.sampler_enable_flags );
+
+      for (i = 0; i < I915_TEX_UNITS; i++) {
+         if (i915->current.sampler_enable_flags & (1<<i)) {
+            OUT_BATCH( i915->current.sampler[i][0] );
+            OUT_BATCH( i915->current.sampler[i][1] );
+            OUT_BATCH( i915->current.sampler[i][2] );
+         }
+      }
+   }
+}
+
+static void
+validate_constants(struct i915_context *i915, unsigned *batch_space)
+{
+   *batch_space = i915->fs->num_constants ?
+      2 + 4*i915->fs->num_constants : 0;
+}
+
+static void
+emit_constants(struct i915_context *i915)
+{
+   /* Collate the user-defined constants with the fragment shader's
+    * immediates according to the constant_flags[] array.
+    */
+   const uint nr = i915->fs->num_constants;
+   if (nr) {
+      uint i;
+
+      OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) );
+      OUT_BATCH((1 << nr) - 1);
+
+      for (i = 0; i < nr; i++) {
+         const uint *c;
+         if (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER) {
+            /* grab user-defined constant */
+            c = (uint *) i915_buffer(i915->constants[PIPE_SHADER_FRAGMENT])->data;
+            c += 4 * i;
+         }
+         else {
+            /* emit program constant */
+            c = (uint *) i915->fs->constants[i];
+         }
+#if 0 /* debug */
+         {
+            float *f = (float *) c;
+            printf("Const %2d: %f %f %f %f %s\n", i, f[0], f[1], f[2], f[3],
+                   (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER
+                    ? "user" : "immediate"));
+         }
+#endif
+         OUT_BATCH(*c++);
+         OUT_BATCH(*c++);
+         OUT_BATCH(*c++);
+         OUT_BATCH(*c++);
+      }
+   }
+}
+
+static void
+validate_program(struct i915_context *i915, unsigned *batch_space)
+{
+   *batch_space = i915->fs->program_len;
+}
+
+static void
+emit_program(struct i915_context *i915)
+{
+      uint i;
+      /* we should always have, at least, a pass-through program */
+      assert(i915->fs->program_len > 0);
+      for (i = 0; i < i915->fs->program_len; i++) {
+         OUT_BATCH(i915->fs->program[i]);
+      }
+}
+
+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);
+}
 
 static boolean
 i915_validate_state(struct i915_context *i915, unsigned *batch_space)
 {
-   int i;
+   unsigned tmp;
 
    i915->num_validation_buffers = 0;
-   *batch_space = 0;
-
-   for (i = 0; i < Elements(hw_atoms); i++)
-      if ((i915->hardware_dirty & hw_atoms[i].dirty) && hw_atoms[i].validate) {
-	 hw_atoms[i].validate(i915);
-	 *batch_space += hw_atoms[i].batch_space;
-      }
+   if (i915->hardware_dirty & I915_HW_INVARIANT)
+      *batch_space = Elements(invariant_state);
+   else
+      *batch_space = 0;
+
+#define VALIDATE_ATOM(atom, hw_dirty) \
+   if (i915->hardware_dirty & hw_dirty) { \
+      validate_##atom(i915, &tmp); \
+      *batch_space += tmp; }
+   VALIDATE_ATOM(flush, I915_HW_FLUSH);
+   VALIDATE_ATOM(immediate, I915_HW_IMMEDIATE);
+   VALIDATE_ATOM(dynamic, I915_HW_DYNAMIC);
+   VALIDATE_ATOM(static, I915_HW_STATIC);
+   VALIDATE_ATOM(map, I915_HW_MAP);
+   VALIDATE_ATOM(sampler, I915_HW_SAMPLER);
+   VALIDATE_ATOM(constants, I915_HW_CONSTANTS);
+   VALIDATE_ATOM(program, I915_HW_PROGRAM);
+#undef VALIDATE_ATOM
 
    if (i915->num_validation_buffers == 0)
       return TRUE;
 
    if (!i915_winsys_validate_buffers(i915->batch, i915->validation_buffers,
-				     i915->num_validation_buffers))
+                                     i915->num_validation_buffers))
       return FALSE;
 
    return TRUE;
 }
 
-static void
-emit_state(struct i915_context *i915)
-{
-   int i;
-
-   for (i = 0; i < Elements(hw_atoms); i++)
-      if ((i915->hardware_dirty & hw_atoms[i].dirty) && hw_atoms[i].emit)
-	 hw_atoms[i].emit(i915);
-}
-
 /* Push the state into the sarea and/or texture memory.
  */
 void
 i915_emit_hardware_state(struct i915_context *i915 )
 {
    unsigned batch_space;
-   /* XXX: there must be an easier way */
-   const unsigned dwords = ( 14 + 
-                             7 + 
-                             I915_MAX_DYNAMIC + 
-                             8 + 
-                             2 + I915_TEX_UNITS*3 + 
-                             2 + I915_TEX_UNITS*3 +
-                             2 + I915_MAX_CONSTANT*4 + 
-#if 0
-                             i915->current.program_len + 
-#else
-                             i915->fs->program_len + 
-#endif
-                             6 
-                           ) * 3/2; /* plus 50% margin */
-   const unsigned relocs = ( I915_TEX_UNITS +
-                             3
-                           ) * 3/2; /* plus 50% margin */
-
    uintptr_t save_ptr;
-   size_t save_relocs;
 
    if (I915_DBG_ON(DBG_ATOMS))
       i915_dump_hardware_dirty(i915, __FUNCTION__);
@@ -173,262 +413,36 @@ i915_emit_hardware_state(struct i915_context *i915 )
       assert(i915_validate_state(i915, &batch_space));
    }
 
-   if(!BEGIN_BATCH(batch_space + dwords, relocs)) {
+   if(!BEGIN_BATCH(batch_space)) {
       FLUSH_BATCH(NULL);
       assert(i915_validate_state(i915, &batch_space));
-      assert(BEGIN_BATCH(batch_space + dwords, relocs));
+      assert(BEGIN_BATCH(batch_space));
    }
 
    save_ptr = (uintptr_t)i915->batch->ptr;
-   save_relocs = i915->batch->relocs;
-
-   emit_state(i915);
-   /* 14 dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_INVARIANT)
-   {
-      OUT_BATCH(_3DSTATE_AA_CMD |
-                AA_LINE_ECAAR_WIDTH_ENABLE |
-                AA_LINE_ECAAR_WIDTH_1_0 |
-                AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0);
-
-      OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
-      OUT_BATCH(0);
-
-      OUT_BATCH(_3DSTATE_DFLT_SPEC_CMD);
-      OUT_BATCH(0);
-      
-      OUT_BATCH(_3DSTATE_DFLT_Z_CMD);
-      OUT_BATCH(0);
-
-      OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS |
-                CSB_TCB(0, 0) |
-                CSB_TCB(1, 1) |
-                CSB_TCB(2, 2) |
-                CSB_TCB(3, 3) |
-                CSB_TCB(4, 4) | 
-                CSB_TCB(5, 5) | 
-                CSB_TCB(6, 6) | 
-                CSB_TCB(7, 7));
-
-      OUT_BATCH(_3DSTATE_RASTER_RULES_CMD |
-                ENABLE_POINT_RASTER_RULE |
-                OGL_POINT_RASTER_RULE |
-                ENABLE_LINE_STRIP_PROVOKE_VRTX |
-                ENABLE_TRI_FAN_PROVOKE_VRTX |
-                LINE_STRIP_PROVOKE_VRTX(1) |
-                TRI_FAN_PROVOKE_VRTX(2) | 
-                ENABLE_TEXKILL_3D_4D | 
-                TEXKILL_4D);
-
-      OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE);
-
-      /* disable indirect state for now
-       */
-      OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0);
-      OUT_BATCH(0);
-   }
-
-   /* 7 dwords, 1 relocs */
-   if (i915->hardware_dirty & I915_HW_IMMEDIATE)
-   {
-      /* remove unwatned bits and S7 */
-      unsigned dirty = (1 << I915_IMMEDIATE_S0 | 1 << I915_IMMEDIATE_S1 |
-                        1 << I915_IMMEDIATE_S2 | 1 << I915_IMMEDIATE_S3 |
-                        1 << I915_IMMEDIATE_S3 | 1 << I915_IMMEDIATE_S4 |
-                        1 << I915_IMMEDIATE_S5 | 1 << I915_IMMEDIATE_S6) &
-                       i915->immediate_dirty;
-      int i, num = util_bitcount(dirty);
-      assert(num && num <= I915_MAX_IMMEDIATE);
-
-      OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 |
-                dirty << 4 | (num - 1));
-
-      if (i915->immediate_dirty & (1 << I915_IMMEDIATE_S0)) {
-         if (i915->vbo)
-            OUT_RELOC(i915->vbo, I915_USAGE_VERTEX,
-                      i915->current.immediate[I915_IMMEDIATE_S0]);
-         else
-            OUT_BATCH(0);
-      }
-
-      for (i = 1; i < I915_MAX_IMMEDIATE; i++) {
-         if (dirty & (1 << i))
-            OUT_BATCH(i915->current.immediate[i]);
-      }
-   } 
-
-#if 01
-   /* I915_MAX_DYNAMIC dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_DYNAMIC) 
-   {
-      int i;
-      for (i = 0; i < I915_MAX_DYNAMIC; i++) {
-         if (i915->dynamic_dirty & (1 << i))
-            OUT_BATCH(i915->current.dynamic[i]);
-      }
-   }
-#endif
-
-#if 01
-   /* 8 dwords, 2 relocs */
-   if (i915->hardware_dirty & I915_HW_STATIC)
-   {
-      if (i915->current.cbuf_bo) {
-         OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
-         OUT_BATCH(i915->current.cbuf_flags);
-         OUT_RELOC(i915->current.cbuf_bo,
-                   I915_USAGE_RENDER,
-                   0);
-      }
-
-      /* What happens if no zbuf??
-       */
-      if (i915->current.depth_bo) {
-         OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
-         OUT_BATCH(i915->current.depth_flags);
-         OUT_RELOC(i915->current.depth_bo,
-                   I915_USAGE_RENDER,
-                   0);
-      }
-
-      {
-         OUT_BATCH(_3DSTATE_DST_BUF_VARS_CMD);
-         OUT_BATCH(i915->current.dst_buf_vars);
-      }
-   }
-#endif
-
-#if 01
-      /* texture images */
-      /* 2 + I915_TEX_UNITS*3 dwords, I915_TEX_UNITS relocs */
-      if (i915->hardware_dirty & (I915_HW_MAP | I915_HW_SAMPLER))
-      {
-         const uint nr = i915->current.sampler_enable_nr;
-         if (nr) {
-            const uint enabled = i915->current.sampler_enable_flags;
-            uint unit;
-            uint count = 0;
-            OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
-            OUT_BATCH(enabled);
-            for (unit = 0; unit < I915_TEX_UNITS; unit++) {
-               if (enabled & (1 << unit)) {
-                  struct i915_texture *texture = i915_texture(i915->fragment_sampler_views[unit]->texture);
-                  struct i915_winsys_buffer *buf = texture->buffer;
-                  assert(buf);
-
-                  count++;
-
-                  OUT_RELOC(buf, I915_USAGE_SAMPLER, 0);
-                  OUT_BATCH(i915->current.texbuffer[unit][0]); /* MS3 */
-                  OUT_BATCH(i915->current.texbuffer[unit][1]); /* MS4 */
-               }
-            }
-            assert(count == nr);
-         }
-      }
-#endif
-
-#if 01
-   /* samplers */
-   /* 2 + I915_TEX_UNITS*3 dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_SAMPLER) 
-   {
-      if (i915->current.sampler_enable_nr) {
-         int i;
-
-         OUT_BATCH( _3DSTATE_SAMPLER_STATE | 
-                    (3 * i915->current.sampler_enable_nr) );
-
-         OUT_BATCH( i915->current.sampler_enable_flags );
-
-         for (i = 0; i < I915_TEX_UNITS; i++) {
-            if (i915->current.sampler_enable_flags & (1<<i)) {
-               OUT_BATCH( i915->current.sampler[i][0] );
-               OUT_BATCH( i915->current.sampler[i][1] );
-               OUT_BATCH( i915->current.sampler[i][2] );
-            }
-         }
-      }
-   }
-#endif
-
-#if 01
-   /* constants */
-   /* 2 + I915_MAX_CONSTANT*4 dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_CONSTANTS)
-   {
-      /* Collate the user-defined constants with the fragment shader's
-       * immediates according to the constant_flags[] array.
-       */
-      const uint nr = i915->fs->num_constants;
-      if (nr) {
-         uint i;
-
-         OUT_BATCH( _3DSTATE_PIXEL_SHADER_CONSTANTS | (nr * 4) );
-	 OUT_BATCH((1 << nr) - 1);
-
-         for (i = 0; i < nr; i++) {
-            const uint *c;
-            if (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER) {
-               /* grab user-defined constant */
-               c = (uint *) i915_buffer(i915->constants[PIPE_SHADER_FRAGMENT])->data;
-               c += 4 * i;
-            }
-            else {
-               /* emit program constant */
-               c = (uint *) i915->fs->constants[i];
-            }
-#if 0 /* debug */
-            {
-               float *f = (float *) c;
-               printf("Const %2d: %f %f %f %f %s\n", i, f[0], f[1], f[2], f[3],
-                      (i915->fs->constant_flags[i] == I915_CONSTFLAG_USER
-                       ? "user" : "immediate"));
-            }
-#endif
-            OUT_BATCH(*c++);
-            OUT_BATCH(*c++);
-            OUT_BATCH(*c++);
-            OUT_BATCH(*c++);
-         }
-      }
-   }
-#endif
-
-#if 01
-   /* Fragment program */
-   /* i915->current.program_len dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_PROGRAM)
-   {
-      uint i;
-      /* we should always have, at least, a pass-through program */
-      assert(i915->fs->program_len > 0);
-      for (i = 0; i < i915->fs->program_len; i++) {
-         OUT_BATCH(i915->fs->program[i]);
-      }
-   }
-#endif
-
-#if 01
-   /* drawing surface size */
-   /* 6 dwords, 0 relocs */
-   if (i915->hardware_dirty & I915_HW_STATIC)
-   {
-      /* XXX flush only required when the draw_offset changes! */
-      OUT_BATCH(MI_FLUSH | INHIBIT_FLUSH_RENDER_CACHE);
-      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);
-   }
-#endif
 
-   I915_DBG(DBG_EMIT, "%s: used %d dwords, %d relocs\n", __FUNCTION__,
+#define EMIT_ATOM(atom, hw_dirty) \
+   if (i915->hardware_dirty & hw_dirty) \
+      emit_##atom(i915);
+   EMIT_ATOM(flush, I915_HW_FLUSH);
+   EMIT_ATOM(invariant, I915_HW_INVARIANT);
+   EMIT_ATOM(immediate, I915_HW_IMMEDIATE);
+   EMIT_ATOM(dynamic, I915_HW_DYNAMIC);
+   EMIT_ATOM(static, I915_HW_STATIC);
+   EMIT_ATOM(map, I915_HW_MAP);
+   EMIT_ATOM(sampler, I915_HW_SAMPLER);
+   EMIT_ATOM(constants, I915_HW_CONSTANTS);
+   EMIT_ATOM(program, I915_HW_PROGRAM);
+   EMIT_ATOM(draw_rect, I915_HW_STATIC);
+#undef EMIT_ATOM
+
+   I915_DBG(DBG_EMIT, "%s: used %d dwords, %d dwords reserved\n", __FUNCTION__,
             ((uintptr_t)i915->batch->ptr - save_ptr) / 4,
-            i915->batch->relocs - save_relocs);
+            batch_space);
+   assert(((uintptr_t)i915->batch->ptr - save_ptr) / 4 == batch_space);
 
    i915->hardware_dirty = 0;
    i915->immediate_dirty = 0;
    i915->dynamic_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 97044499990..20cd23f8f73 100644
--- a/src/gallium/drivers/i915/i915_state_static.c
+++ b/src/gallium/drivers/i915/i915_state_static.c
@@ -164,7 +164,7 @@ static void update_framebuffer(struct i915_context *i915)
    assert(ret);
    if (i915->current.draw_offset != draw_offset) {
       i915->current.draw_offset = draw_offset;
-      /* XXX: only emit flush on change and not always in emit */
+      i915_set_flush_dirty(i915, I915_PIPELINE_FLUSH);
    }
    i915->current.draw_size = (w - 1 + x) | ((h - 1 + y) << 16);
 
diff --git a/src/gallium/drivers/i915/i915_winsys.h b/src/gallium/drivers/i915/i915_winsys.h
index 4ac2f5b9777..21cfdc9613e 100644
--- a/src/gallium/drivers/i915/i915_winsys.h
+++ b/src/gallium/drivers/i915/i915_winsys.h
@@ -76,7 +76,6 @@ struct i915_winsys_batchbuffer {
    size_t size;
 
    size_t relocs;
-   size_t max_relocs;
    /*@}*/
 };
 
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index 4f86db39926..6391ea7f3be 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -22,7 +22,6 @@
 
 #include "r300_context.h"
 #include "r300_emit.h"
-#include "r300_hyperz.h"
 #include "r300_texture.h"
 #include "r300_winsys.h"
 
@@ -117,6 +116,14 @@ static boolean r300_fast_zclear_allowed(struct r300_context *r300)
     return r300_resource(fb->zsbuf->texture)->tex.zmask_dwords[fb->zsbuf->u.tex.level];
 }
 
+static boolean r300_hiz_clear_allowed(struct r300_context *r300)
+{
+    struct pipe_framebuffer_state *fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
+
+    return r300_resource(fb->zsbuf->texture)->tex.hiz_dwords[fb->zsbuf->u.tex.level];
+}
+
 static uint32_t r300_depth_clear_value(enum pipe_format format,
                                        double depth, unsigned stencil)
 {
@@ -134,6 +141,13 @@ static uint32_t r300_depth_clear_value(enum pipe_format format,
     }
 }
 
+static uint32_t r300_hiz_clear_value(double depth)
+{
+    uint32_t r = (uint32_t)(CLAMP(depth, 0, 1) * 255.5);
+    assert(r <= 255);
+    return r | (r << 8) | (r << 16) | (r << 24);
+}
+
 /* Clear currently bound buffers. */
 static void r300_clear(struct pipe_context* pipe,
                        unsigned buffers,
@@ -190,8 +204,6 @@ static void r300_clear(struct pipe_context* pipe,
         (struct pipe_framebuffer_state*)r300->fb_state.state;
     struct r300_hyperz_state *hyperz =
         (struct r300_hyperz_state*)r300->hyperz_state.state;
-    struct r300_resource *zstex =
-            fb->zsbuf ? r300_resource(fb->zsbuf->texture) : NULL;
     uint32_t width = fb->width;
     uint32_t height = fb->height;
     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
@@ -200,20 +212,18 @@ static void r300_clear(struct pipe_context* pipe,
     /* Enable fast Z clear.
      * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
     if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && can_hyperz) {
-        hyperz_dcv = hyperz->zb_depthclearvalue =
-            r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
-
         if (r300_fast_zclear_allowed(r300)) {
+            hyperz_dcv = hyperz->zb_depthclearvalue =
+                r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
+
             r300_mark_atom_dirty(r300, &r300->zmask_clear);
             buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
         }
 
-        if (zstex->hiz_mem[fb->zsbuf->u.tex.level])
+        if (r300_hiz_clear_allowed(r300)) {
+            r300->hiz_clear_value = r300_hiz_clear_value(depth);
             r300_mark_atom_dirty(r300, &r300->hiz_clear);
-
-        /* XXX Change this to r300_mark_atom_dirty(r300, &r300->hyperz_state);
-         * once hiz offset is constant. */
-        r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
+        }
     }
 
     /* Enable CBZB clear. */
@@ -240,14 +250,14 @@ static void r300_clear(struct pipe_context* pipe,
                            fb->nr_cbufs,
                            buffers, rgba, depth, stencil);
         r300_blitter_end(r300);
-    } else if (r300->zmask_clear.dirty) {
+    } else if (r300->zmask_clear.dirty || r300->hiz_clear.dirty) {
         /* Just clear zmask and hiz now, this does not use the standard draw
          * procedure. */
         unsigned dwords;
 
         /* Calculate zmask_clear and hiz_clear atom sizes. */
         r300_update_hyperz_state(r300);
-        dwords = r300->zmask_clear.size +
+        dwords = (r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
                  (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
                  r300_get_num_cs_end_dwords(r300);
 
@@ -257,9 +267,11 @@ static void r300_clear(struct pipe_context* pipe,
         }
 
         /* Emit clear packets. */
-        r300_emit_zmask_clear(r300, r300->zmask_clear.size,
-                              r300->zmask_clear.state);
-        r300->zmask_clear.dirty = FALSE;
+        if (r300->zmask_clear.dirty) {
+            r300_emit_zmask_clear(r300, r300->zmask_clear.size,
+                                  r300->zmask_clear.state);
+            r300->zmask_clear.dirty = FALSE;
+        }
         if (r300->hiz_clear.dirty) {
             r300_emit_hiz_clear(r300, r300->hiz_clear.size,
                                 r300->hiz_clear.state);
@@ -279,9 +291,8 @@ static void r300_clear(struct pipe_context* pipe,
     /* Enable fastfill and/or hiz.
      *
      * If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
-     * looks if zmask/hiz is in use and enables fastfill accordingly. */
-    if (r300->zmask_in_use ||
-        (zstex && zstex->hiz_in_use[fb->zsbuf->u.tex.level])) {
+     * looks if zmask/hiz is in use and programs hardware accordingly. */
+    if (r300->zmask_in_use || r300->hiz_in_use) {
         r300_mark_atom_dirty(r300, &r300->hyperz_state);
     }
 }
@@ -295,7 +306,7 @@ static void r300_clear_render_target(struct pipe_context *pipe,
 {
     struct r300_context *r300 = r300_context(pipe);
 
-    r300->zmask_locked = TRUE;
+    r300->hyperz_locked = TRUE;
     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 
     r300_blitter_begin(r300, R300_CLEAR_SURFACE);
@@ -303,7 +314,7 @@ static void r300_clear_render_target(struct pipe_context *pipe,
                                      dstx, dsty, width, height);
     r300_blitter_end(r300);
 
-    r300->zmask_locked = FALSE;
+    r300->hyperz_locked = FALSE;
     r300_mark_atom_dirty(r300, &r300->hyperz_state);
 }
 
@@ -320,11 +331,11 @@ static void r300_clear_depth_stencil(struct pipe_context *pipe,
     struct pipe_framebuffer_state *fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
 
-    if (r300->zmask_in_use && !r300->zmask_locked) {
+    if (r300->zmask_in_use && !r300->hyperz_locked) {
         if (fb->zsbuf->texture == dst->texture) {
             r300_decompress_zmask(r300);
         } else {
-            r300->zmask_locked = TRUE;
+            r300->hyperz_locked = TRUE;
             r300_mark_atom_dirty(r300, &r300->hyperz_state);
         }
     }
@@ -334,8 +345,8 @@ static void r300_clear_depth_stencil(struct pipe_context *pipe,
                                      dstx, dsty, width, height);
     r300_blitter_end(r300);
 
-    if (r300->zmask_locked) {
-        r300->zmask_locked = FALSE;
+    if (r300->hyperz_locked) {
+        r300->hyperz_locked = FALSE;
         r300_mark_atom_dirty(r300, &r300->hyperz_state);
     }
 }
@@ -345,7 +356,7 @@ void r300_decompress_zmask(struct r300_context *r300)
     struct pipe_framebuffer_state *fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
 
-    if (!r300->zmask_in_use || r300->zmask_locked)
+    if (!r300->zmask_in_use || r300->hyperz_locked)
         return;
 
     r300->zmask_decompress = TRUE;
@@ -420,12 +431,12 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
             util_format_description(dst->format);
     struct pipe_box box;
 
-    if (r300->zmask_in_use && !r300->zmask_locked) {
+    if (r300->zmask_in_use && !r300->hyperz_locked) {
         if (fb->zsbuf->texture == src ||
             fb->zsbuf->texture == dst) {
             r300_decompress_zmask(r300);
         } else {
-            r300->zmask_locked = TRUE;
+            r300->hyperz_locked = TRUE;
             r300_mark_atom_dirty(r300, &r300->hyperz_state);
         }
     }
@@ -463,7 +474,8 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
     }
 
     /* Handle compressed formats. */
-    if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
+    if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC ||
+        desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
         switch (util_format_get_blocksize(old_dst.format)) {
         case 8:
             /* 1 pixel = 4 bits,
@@ -502,8 +514,8 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
     if (old_dst.format != new_dst.format)
         r300_resource_set_properties(pipe->screen, dst, 0, &old_dst);
 
-    if (r300->zmask_locked) {
-        r300->zmask_locked = FALSE;
+    if (r300->hyperz_locked) {
+        r300->hyperz_locked = FALSE;
         r300_mark_atom_dirty(r300, &r300->hyperz_state);
     }
 }
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index 0be161fa07a..68943d561ba 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -90,8 +90,6 @@ struct r300_capabilities {
     boolean high_second_pipe;
     /* DXTC texture swizzling. */
     boolean dxtc_swizzle;
-    /* Index bias (AKA index offset). */
-    boolean index_bias_supported;
 };
 
 /* Enumerations for legibility and telling which card we're running on. */
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index d422ffe03f8..166d965aa5b 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -30,7 +30,6 @@
 #include "r300_cb.h"
 #include "r300_context.h"
 #include "r300_emit.h"
-#include "r300_hyperz.h"
 #include "r300_screen.h"
 #include "r300_screen_buffer.h"
 #include "r300_winsys.h"
@@ -170,7 +169,6 @@ static boolean r300_setup_atoms(struct r300_context* r300)
     boolean is_rv350 = r300->screen->caps.is_rv350;
     boolean is_r500 = r300->screen->caps.is_r500;
     boolean has_tcl = r300->screen->caps.has_tcl;
-    boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
     boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
     boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
     boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
@@ -203,11 +201,11 @@ static boolean r300_setup_atoms(struct r300_context* r300)
     /* SC. */
     R300_INIT_ATOM(scissor_state, 3);
     /* GB, FG, GA, SU, SC, RB3D. */
-    R300_INIT_ATOM(invariant_state, 18 + (is_rv350 ? 4 : 0));
+    R300_INIT_ATOM(invariant_state, 18 + (is_rv350 ? 4 : 0) + (is_r500 ? 4 : 0));
     /* VAP. */
     R300_INIT_ATOM(viewport_state, 9);
     R300_INIT_ATOM(pvs_flush, 2);
-    R300_INIT_ATOM(vap_invariant_state, 9);
+    R300_INIT_ATOM(vap_invariant_state, is_r500 ? 11 : 9);
     R300_INIT_ATOM(vertex_stream_state, 0);
     R300_INIT_ATOM(vs_state, 0);
     R300_INIT_ATOM(vs_constants, 0);
@@ -216,7 +214,7 @@ static boolean r300_setup_atoms(struct r300_context* r300)
     R300_INIT_ATOM(rs_block_state, 0);
     R300_INIT_ATOM(rs_state, 0);
     /* SC, US. */
-    R300_INIT_ATOM(fb_state_pipelined, 5 + (drm_2_3_0 ? 3 : 0));
+    R300_INIT_ATOM(fb_state_pipelined, 8);
     /* US. */
     R300_INIT_ATOM(fs, 0);
     R300_INIT_ATOM(fs_rc_constant_state, 0);
@@ -227,7 +225,7 @@ static boolean r300_setup_atoms(struct r300_context* r300)
     if (can_hyperz) {
         /* HiZ Clear */
         if (has_hiz_ram)
-            R300_INIT_ATOM(hiz_clear, 0);
+            R300_INIT_ATOM(hiz_clear, 4);
         /* zmask clear */
         R300_INIT_ATOM(zmask_clear, 4);
     }
@@ -331,7 +329,7 @@ static void r300_init_states(struct pipe_context *pipe)
 
     /* Initialize the VAP invariant state. */
     {
-        BEGIN_CB(vap_invariant->cb, 9);
+        BEGIN_CB(vap_invariant->cb, r300->vap_invariant_state.size);
         OUT_CB_REG(VAP_PVS_VTX_TIMEOUT_REG, 0xffff);
         OUT_CB_REG_SEQ(R300_VAP_GB_VERT_CLIP_ADJ, 4);
         OUT_CB_32F(1.0);
@@ -339,6 +337,10 @@ static void r300_init_states(struct pipe_context *pipe)
         OUT_CB_32F(1.0);
         OUT_CB_32F(1.0);
         OUT_CB_REG(R300_VAP_PSC_SGN_NORM_CNTL, R300_SGN_NORM_NO_ZERO);
+
+        if (r300->screen->caps.is_r500) {
+            OUT_CB_REG(R500_VAP_TEX_TO_COLOR_CNTL, 0);
+        }
         END_CB;
     }
 
@@ -359,6 +361,11 @@ static void r300_init_states(struct pipe_context *pipe)
             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_LTE_THRESHOLD, 0x01010101);
             OUT_CB_REG(R500_RB3D_DISCARD_SRC_PIXEL_GTE_THRESHOLD, 0xFEFEFEFE);
         }
+
+        if (r300->screen->caps.is_r500) {
+            OUT_CB_REG(R500_GA_COLOR_CONTROL_PS3, 0);
+            OUT_CB_REG(R500_SU_TEX_WRAP_PS3, 0);
+        }
         END_CB;
     }
 
@@ -447,16 +454,10 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
 
     /* Render functions must be initialized after blitter. */
     r300_init_render_functions(r300);
+    r300_init_states(&r300->context);
 
     rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
 
-    /* setup hyper-z mm */
-    if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
-        if (!r300_hyperz_init_mm(r300))
-            goto fail;
-
-    r300_init_states(&r300->context);
-
     /* The KIL opcode needs the first texture unit to be enabled
      * on r3xx-r4xx. In order to calm down the CS checker, we bind this
      * dummy texture there. */
@@ -507,10 +508,10 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     }
 
     /* Print driver info. */
-#ifdef NDEBUG
-    if (DBG_ON(r300, DBG_INFO)) {
-#else
+#ifdef DEBUG
     {
+#else
+    if (DBG_ON(r300, DBG_INFO)) {
 #endif
         fprintf(stderr,
                 "r300: DRM version: %d.%d.%d, Name: %s, ID: 0x%04x, GB: %d, Z: %d\n"
@@ -526,7 +527,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
                 rws->get_value(rws, R300_VID_GART_SIZE) >> 20,
                 rws->get_value(rws, R300_VID_VRAM_SIZE) >> 20,
                 rws->get_value(rws, R300_CAN_AACOMPRESS) ? "YES" : "NO",
-                rws->get_value(rws, R300_CAN_HYPERZ) ? "YES" : "NO",
+                rws->get_value(rws, R300_CAN_HYPERZ) &&
+                r300->screen->caps.zmask_ram ? "YES" : "NO",
                 rws->get_value(rws, R300_CAN_HYPERZ) &&
                 r300->screen->caps.hiz_ram ? "YES" : "NO");
     }
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index e9c7d7bf63f..6f2aab69ab1 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -102,7 +102,6 @@ struct r300_dsa_state {
 };
 
 struct r300_hyperz_state {
-    int current_func; /* -1 after a clear before first op */
     int flush;
     /* This is actually a command buffer with named dwords. */
     uint32_t cb_flush_begin;
@@ -220,11 +219,11 @@ struct r300_vertex_stream_state {
 };
 
 struct r300_invariant_state {
-    uint32_t cb[22];
+    uint32_t cb[26];
 };
 
 struct r300_vap_invariant_state {
-    uint32_t cb[9];
+    uint32_t cb[11];
 };
 
 struct r300_viewport_state {
@@ -295,6 +294,8 @@ struct r300_surface {
 
     uint32_t offset;    /* COLOROFFSET or DEPTHOFFSET. */
     uint32_t pitch;     /* COLORPITCH or DEPTHPITCH. */
+    uint32_t pitch_zmask; /* ZMASK_PITCH */
+    uint32_t pitch_hiz;   /* HIZ_PITCH */
     uint32_t format;    /* US_OUT_FMT or ZB_FORMAT. */
 
     /* Parameters dedicated to the CBZB clear. */
@@ -363,8 +364,12 @@ struct r300_texture_desc {
 
     /* Zbuffer compression info for each miplevel. */
     boolean zcomp8x8[R300_MAX_TEXTURE_LEVELS];
-    /* If zero, then disable compression. */
+    /* If zero, then disable Z compression/HiZ. */
     unsigned zmask_dwords[R300_MAX_TEXTURE_LEVELS];
+    unsigned hiz_dwords[R300_MAX_TEXTURE_LEVELS];
+    /* Zmask/HiZ strides for each miplevel. */
+    unsigned zmask_stride_in_pixels[R300_MAX_TEXTURE_LEVELS];
+    unsigned hiz_stride_in_pixels[R300_MAX_TEXTURE_LEVELS];
 };
 
 struct r300_resource
@@ -390,10 +395,6 @@ struct r300_resource
     /* Where the texture starts in the buffer. */
     unsigned tex_offset;
 
-    /* HiZ memory allocations. */
-    struct mem_block *hiz_mem[R300_MAX_TEXTURE_LEVELS];
-    boolean hiz_in_use[R300_MAX_TEXTURE_LEVELS];
-
     /* This is the level tiling flags were last time set for.
      * It's used to prevent redundant tiling-flags changes from happening.*/
     unsigned surface_level;
@@ -412,6 +413,21 @@ struct r300_vertex_element_state {
     struct r300_vertex_stream_state vertex_stream;
 };
 
+enum r300_hiz_func {
+    HIZ_FUNC_NONE,
+
+    /* The function, when determined, is set in stone
+     * until the next HiZ clear. */
+
+    /* MAX is written to the HiZ buffer.
+     * Used for LESS, LEQUAL. */
+    HIZ_FUNC_MAX,
+
+    /* MIN is written to the HiZ buffer.
+     * Used for GREATER, GEQUAL. */
+    HIZ_FUNC_MIN,
+};
+
 struct r300_context {
     /* Parent class */
     struct pipe_context context;
@@ -545,22 +561,25 @@ struct r300_context {
     int sprite_coord_enable;
     /* Whether two-sided color selection is enabled (AKA light_twoside). */
     boolean two_sided_color;
-
+    /* Whether fast color clear is enabled. */
     boolean cbzb_clear;
     /* Whether ZMASK is enabled. */
     boolean zmask_in_use;
     /* Whether ZMASK is being decompressed. */
     boolean zmask_decompress;
-    /* Whether ZMASK is locked, i.e. should be disabled and cannot be taken over. */
-    boolean zmask_locked;
+    /* Whether ZMASK/HIZ is locked, i.e. should be disabled and cannot be taken over. */
+    boolean hyperz_locked;
     /* The zbuffer the ZMASK of which is locked. */
     struct pipe_surface *locked_zbuffer;
+    /* Whether HIZ is enabled. */
+    boolean hiz_in_use;
+    /* HiZ function. Can be either MIN or MAX. */
+    enum r300_hiz_func hiz_func;
+    /* HiZ clear value. */
+    uint32_t hiz_clear_value;
 
     void *dsa_decompress_zmask;
 
-    /* two mem block managers for hiz/zmask ram space */
-    struct mem_block *hiz_mm;
-
     struct u_vbuf_mgr *vbuf_mgr;
 
     struct util_slab_mempool pool_transfers;
@@ -644,6 +663,9 @@ void r300_decompress_zmask(struct r300_context *r300);
 void r300_decompress_zmask_locked_unsafe(struct r300_context *r300);
 void r300_decompress_zmask_locked(struct r300_context *r300);
 
+/* r300_hyperz.c */
+void r300_update_hyperz_state(struct r300_context* r300);
+
 /* r300_query.c */
 void r300_resume_query(struct r300_context *r300,
                        struct r300_query *query);
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index e2e4719ec82..24c82a3efd2 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -425,27 +425,12 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
         OUT_CS_RELOC(surf);
 
         if (can_hyperz) {
-            uint32_t surf_pitch;
-            struct r300_resource *tex;
-            int level = surf->base.u.tex.level;
-            tex = r300_resource(surf->base.texture);
-
-            surf_pitch = surf->pitch & R300_DEPTHPITCH_MASK;
-
             /* HiZ RAM. */
-            if (r300->screen->caps.hiz_ram) {
-                if (tex->hiz_mem[level]) {
-                    OUT_CS_REG(R300_ZB_HIZ_OFFSET, tex->hiz_mem[level]->ofs << 2);
-                    OUT_CS_REG(R300_ZB_HIZ_PITCH, surf_pitch);
-                } else {
-                    OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
-                    OUT_CS_REG(R300_ZB_HIZ_PITCH, 0);
-                }
-            }
-
+            OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
+            OUT_CS_REG(R300_ZB_HIZ_PITCH, surf->pitch_hiz);
             /* Z Mask RAM. (compressed zbuffer) */
             OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
-            OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf_pitch);
+            OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf->pitch_zmask);
         }
     }
 
@@ -484,6 +469,7 @@ void r300_emit_fb_state_pipelined(struct r300_context *r300,
     struct pipe_framebuffer_state* fb =
             (struct pipe_framebuffer_state*)r300->fb_state.state;
     unsigned i, num_cbufs = fb->nr_cbufs;
+    unsigned mspos0, mspos1;
     CS_LOCALS(r300);
 
     /* If we use the multiwrite feature, the colorbuffers 2,3,4 must be
@@ -507,38 +493,36 @@ void r300_emit_fb_state_pipelined(struct r300_context *r300,
     /* Multisampling. Depends on framebuffer sample count.
      * These are pipelined regs and as such cannot be moved
      * to the AA state. */
-    if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) {
-        unsigned mspos0 = 0x66666666;
-        unsigned mspos1 = 0x6666666;
-
-        if (fb->nr_cbufs && fb->cbufs[0]->texture->nr_samples > 1) {
-            /* Subsample placement. These may not be optimal. */
-            switch (fb->cbufs[0]->texture->nr_samples) {
-                case 2:
-                    mspos0 = 0x33996633;
-                    mspos1 = 0x6666663;
-                    break;
-                case 3:
-                    mspos0 = 0x33936933;
-                    mspos1 = 0x6666663;
-                    break;
-                case 4:
-                    mspos0 = 0x33939933;
-                    mspos1 = 0x3966663;
-                    break;
-                case 6:
-                    mspos0 = 0x22a2aa22;
-                    mspos1 = 0x2a65672;
-                    break;
-                default:
-                    debug_printf("r300: Bad number of multisamples!\n");
-            }
-        }
+    mspos0 = 0x66666666;
+    mspos1 = 0x6666666;
 
-        OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2);
-        OUT_CS(mspos0);
-        OUT_CS(mspos1);
+    if (fb->nr_cbufs && fb->cbufs[0]->texture->nr_samples > 1) {
+        /* Subsample placement. These may not be optimal. */
+        switch (fb->cbufs[0]->texture->nr_samples) {
+        case 2:
+            mspos0 = 0x33996633;
+            mspos1 = 0x6666663;
+            break;
+        case 3:
+            mspos0 = 0x33936933;
+            mspos1 = 0x6666663;
+            break;
+        case 4:
+            mspos0 = 0x33939933;
+            mspos1 = 0x3966663;
+            break;
+        case 6:
+            mspos0 = 0x22a2aa22;
+            mspos1 = 0x2a65672;
+            break;
+        default:
+            debug_printf("r300: Bad number of multisamples!\n");
+        }
     }
+
+    OUT_CS_REG_SEQ(R300_GB_MSPOS0, 2);
+    OUT_CS(mspos0);
+    OUT_CS(mspos1);
     END_CS;
 }
 
@@ -1039,56 +1023,26 @@ void r300_emit_viewport_state(struct r300_context* r300,
     END_CS;
 }
 
-static void r300_emit_hiz_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val)
-{
-    CS_LOCALS(r300);
-    BEGIN_CS(4);
-    OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
-    OUT_CS(start);
-    OUT_CS(count);
-    OUT_CS(val);
-    END_CS;
-}
-
-#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
-
 void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
 {
     struct pipe_framebuffer_state *fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
-    struct r300_hyperz_state *z =
-        (struct r300_hyperz_state*)r300->hyperz_state.state;
-    struct r300_screen* r300screen = r300->screen;
-    uint32_t stride, offset = 0, height, offset_shift;
     struct r300_resource* tex;
-    int i;
+    CS_LOCALS(r300);
 
     tex = r300_resource(fb->zsbuf->texture);
 
-    offset = tex->hiz_mem[fb->zsbuf->u.tex.level]->ofs;
-    stride = tex->tex.stride_in_pixels[fb->zsbuf->u.tex.level];
-
-    /* convert from pixels to 4x4 blocks */
-    stride = ALIGN_DIVUP(stride, 4);
-
-    stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes);    
-    /* there are 4 blocks per dwords */
-    stride = ALIGN_DIVUP(stride, 4);
-
-    height = ALIGN_DIVUP(fb->zsbuf->height, 4);
-
-    offset_shift = 2;
-    offset_shift += (r300screen->caps.num_frag_pipes / 2);
-
-    for (i = 0; i < height; i++) {
-        offset = i * stride;
-        offset <<= offset_shift;
-        r300_emit_hiz_line_clear(r300, offset, stride, 0xffffffff);
-    }
-    z->current_func = -1;
+    BEGIN_CS(size);
+    OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
+    OUT_CS(0);
+    OUT_CS(tex->tex.hiz_dwords[fb->zsbuf->u.tex.level]);
+    OUT_CS(r300->hiz_clear_value);
+    END_CS;
 
     /* Mark the current zbuffer's hiz ram as in use. */
-    tex->hiz_in_use[fb->zsbuf->u.tex.level] = TRUE;
+    r300->hiz_in_use = TRUE;
+    r300->hiz_func = HIZ_FUNC_NONE;
+    r300_mark_atom_dirty(r300, &r300->hyperz_state);
 }
 
 void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
@@ -1236,7 +1190,7 @@ unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
     /* Emitted in flush. */
     dwords += 26; /* emit_query_end */
     dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
-    if (r300->screen->caps.index_bias_supported)
+    if (r300->screen->caps.is_r500)
         dwords += 2;
 
     return dwords;
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index c77cc08539d..9c41a1383ce 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -57,7 +57,7 @@ static void r300_flush(struct pipe_context* pipe,
     if (r300->dirty_hw) {
         r300_emit_hyperz_end(r300);
         r300_emit_query_end(r300);
-        if (r300->screen->caps.index_bias_supported)
+        if (r300->screen->caps.is_r500)
             r500_emit_index_bias(r300, 0);
 
         r300->flush_counter++;
diff --git a/src/gallium/drivers/r300/r300_hyperz.c b/src/gallium/drivers/r300/r300_hyperz.c
index 873e0209d42..ecaadf4af8e 100644
--- a/src/gallium/drivers/r300/r300_hyperz.c
+++ b/src/gallium/drivers/r300/r300_hyperz.c
@@ -22,7 +22,6 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
 #include "r300_context.h"
-#include "r300_hyperz.h"
 #include "r300_reg.h"
 #include "r300_fs.h"
 #include "r300_winsys.h"
@@ -41,58 +40,74 @@
 /* The HyperZ setup                                                          */
 /*****************************************************************************/
 
-static bool r300_get_sc_hz_max(struct r300_context *r300)
+static enum r300_hiz_func r300_get_hiz_func(struct r300_context *r300)
 {
-    struct r300_dsa_state *dsa_state = r300->dsa_state.state;
-    int func = dsa_state->z_stencil_control & R300_ZS_MASK;
-    int ret = R300_SC_HYPERZ_MIN;
+    struct r300_dsa_state *dsa = r300->dsa_state.state;
 
-    if (func >= R300_ZS_GEQUAL && func <= R300_ZS_ALWAYS)
-       ret = R300_SC_HYPERZ_MAX;
-    return ret;
+    if (!dsa->dsa.depth.enabled || !dsa->dsa.depth.writemask)
+        return HIZ_FUNC_NONE;
+
+    switch (dsa->dsa.depth.func) {
+    case PIPE_FUNC_NEVER:
+    case PIPE_FUNC_EQUAL:
+    case PIPE_FUNC_NOTEQUAL:
+    case PIPE_FUNC_ALWAYS:
+        return HIZ_FUNC_NONE;
+
+    case PIPE_FUNC_LESS:
+    case PIPE_FUNC_LEQUAL:
+        return HIZ_FUNC_MAX;
+
+    case PIPE_FUNC_GREATER:
+    case PIPE_FUNC_GEQUAL:
+        return HIZ_FUNC_MIN;
+
+    default:
+        assert(0);
+        return HIZ_FUNC_NONE;
+    }
 }
 
-static bool r300_zfunc_same_direction(int func1, int func2)
+/* Return what's used for the depth test (either minimum or maximum). */
+static unsigned r300_get_sc_hz_max(struct r300_context *r300)
 {
-    /* func1 is less/lessthan */
-    if ((func1 == R300_ZS_LESS || func1 == R300_ZS_LEQUAL) &&
-        (func2 == R300_ZS_EQUAL || func2 == R300_ZS_GEQUAL ||
-         func2 == R300_ZS_GREATER))
-            return FALSE;
-
-    /* func1 is greater/greaterthan */
-    if ((func1 == R300_ZS_GEQUAL || func1 == R300_ZS_GREATER) &&
-        (func2 == R300_ZS_LESS || func2 == R300_ZS_LEQUAL))
-            return FALSE;
+    struct r300_dsa_state *dsa = r300->dsa_state.state;
+    unsigned func = dsa->dsa.depth.func;
 
-    return TRUE;
+    return func >= PIPE_FUNC_GREATER ? R300_SC_HYPERZ_MAX : R300_SC_HYPERZ_MIN;
 }
 
-static int r300_get_hiz_min(struct r300_context *r300)
+static boolean r300_is_hiz_func_valid(struct r300_context *r300)
 {
-    struct r300_dsa_state *dsa_state = r300->dsa_state.state;
-    int func = dsa_state->z_stencil_control & R300_ZS_MASK;
-    int ret = R300_HIZ_MIN;
+    struct r300_dsa_state *dsa = r300->dsa_state.state;
+    unsigned func = dsa->dsa.depth.func;
+
+    if (r300->hiz_func == HIZ_FUNC_NONE)
+        return TRUE;
+
+    /* func1 is less/lessthan */
+    if (r300->hiz_func == HIZ_FUNC_MAX &&
+        (func == PIPE_FUNC_GEQUAL || func == PIPE_FUNC_GREATER))
+        return FALSE;
 
-    if (func == R300_ZS_LESS || func == R300_ZS_LEQUAL)
-       ret = R300_HIZ_MAX;
-    return ret;
+    /* func1 is greater/greaterthan */
+    if (r300->hiz_func == HIZ_FUNC_MIN &&
+        (func == PIPE_FUNC_LESS   || func == PIPE_FUNC_LEQUAL))
+        return FALSE;
+
+    return TRUE;
 }
 
 static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s)
 {
-    if (s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
-                       s->zfail_op != PIPE_STENCIL_OP_KEEP))
-        return TRUE;
-    return FALSE;
+    return s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
+                          s->zfail_op != PIPE_STENCIL_OP_KEEP);
 }
 
 static boolean r300_can_hiz(struct r300_context *r300)
 {
-    struct r300_dsa_state *dsa_state = r300->dsa_state.state;
-    struct pipe_depth_stencil_alpha_state *dsa = &dsa_state->dsa;
-    struct r300_screen* r300screen = r300->screen;
-    struct r300_hyperz_state *z = r300->hyperz_state.state;
+    struct r300_dsa_state *dsa = r300->dsa_state.state;
+    struct r300_screen *r300screen = r300->screen;
 
     /* shader writes depth - no HiZ */
     if (r300_fragment_shader_writes_depth(r300_fs(r300))) /* (5) */
@@ -100,34 +115,21 @@ static boolean r300_can_hiz(struct r300_context *r300)
 
     if (r300->query_current)
         return FALSE;
+
     /* if stencil fail/zfail op is not KEEP */
-    if (r300_dsa_stencil_op_not_keep(&dsa->stencil[0]) ||
-        r300_dsa_stencil_op_not_keep(&dsa->stencil[1]))
+    if (r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[0]) ||
+        r300_dsa_stencil_op_not_keep(&dsa->dsa.stencil[1]))
         return FALSE;
 
-    if (dsa->depth.enabled) {
+    if (dsa->dsa.depth.enabled) {
         /* if depth func is EQUAL pre-r500 */
-        if (dsa->depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
+        if (dsa->dsa.depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
             return FALSE;
+
         /* if depth func is NOTEQUAL */
-        if (dsa->depth.func == PIPE_FUNC_NOTEQUAL)
+        if (dsa->dsa.depth.func == PIPE_FUNC_NOTEQUAL)
             return FALSE;
     }
-    /* depth comparison function - if just cleared save and return okay */
-    if (z->current_func == -1) {
-        int func = dsa_state->z_stencil_control & R300_ZS_MASK;
-        if (func != 0 && func != 7)
-            z->current_func = dsa_state->z_stencil_control & R300_ZS_MASK;
-    } else {
-        /* simple don't change */
-        if (!r300_zfunc_same_direction(z->current_func,
-                                       (dsa_state->z_stencil_control & R300_ZS_MASK))) {
-            DBG(r300, DBG_HYPERZ,
-                "z func changed direction - disabling hyper-z %d -> %d\n",
-                z->current_func, dsa_state->z_stencil_control);
-            return FALSE;
-        }
-    }
     return TRUE;
 }
 
@@ -139,7 +141,6 @@ static void r300_update_hyperz(struct r300_context* r300)
         (struct pipe_framebuffer_state*)r300->fb_state.state;
     struct r300_resource *zstex =
             fb->zsbuf ? r300_resource(fb->zsbuf->texture) : NULL;
-    boolean hiz_in_use = FALSE;
 
     z->gb_z_peq_config = 0;
     z->zb_bw_cntl = 0;
@@ -151,16 +152,12 @@ static void r300_update_hyperz(struct r300_context* r300)
         return;
     }
 
-    if (!zstex)
+    if (!zstex ||
+        !r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
         return;
 
-    if (!r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
-        return;
-
-    hiz_in_use = zstex->hiz_in_use[fb->zsbuf->u.tex.level];
-
     /* Zbuffer compression. */
-    if (r300->zmask_in_use && !r300->zmask_locked) {
+    if (r300->zmask_in_use && !r300->hyperz_locked) {
         z->zb_bw_cntl |= R300_FAST_FILL_ENABLE |
                          /*R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE |*/
                          R300_RD_COMP_ENABLE;
@@ -174,16 +171,28 @@ static void r300_update_hyperz(struct r300_context* r300)
         z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
     }
 
-    if (hiz_in_use && r300_can_hiz(r300)) {
-        z->zb_bw_cntl |= R300_HIZ_ENABLE |
-                         r300_get_hiz_min(r300);
-
-        z->sc_hyperz |= R300_SC_HYPERZ_ENABLE |
-                        r300_get_sc_hz_max(r300);
+    /* HiZ. */
+    if (r300->hiz_in_use && !r300->hyperz_locked) {
+        /* Set the HiZ function if needed. */
+        if (r300->hiz_func == HIZ_FUNC_NONE) {
+            r300->hiz_func = r300_get_hiz_func(r300);
+        }
 
-        if (r300->screen->caps.is_r500) {
-            z->zb_bw_cntl |= R500_HIZ_FP_EXP_BITS_3 |
-                             R500_HIZ_EQUAL_REJECT_ENABLE;
+        /* If the depth function is inverted, HiZ must be disabled. */
+        if (!r300_is_hiz_func_valid(r300)) {
+            r300->hiz_in_use = FALSE;
+        } else if (r300_can_hiz(r300)) {
+            /* Setup the HiZ bits. */
+            z->zb_bw_cntl |=
+                R300_HIZ_ENABLE |
+                (r300->hiz_func == HIZ_FUNC_MIN ? R300_HIZ_MIN : R300_HIZ_MAX);
+
+            z->sc_hyperz |= R300_SC_HYPERZ_ENABLE |
+                            r300_get_sc_hz_max(r300);
+
+            if (r300->screen->caps.is_r500) {
+                z->zb_bw_cntl |= R500_HIZ_EQUAL_REJECT_ENABLE;
+            }
         }
     }
 
@@ -282,18 +291,6 @@ static void r300_update_ztop(struct r300_context* r300)
         r300_mark_atom_dirty(r300, &r300->ztop_state);
 }
 
-#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
-
-static void r300_update_hiz_clear(struct r300_context *r300)
-{
-    struct pipe_framebuffer_state *fb =
-        (struct pipe_framebuffer_state*)r300->fb_state.state;
-    uint32_t height;
-
-    height = ALIGN_DIVUP(fb->zsbuf->height, 4);
-    r300->hiz_clear.size = height * 4;
-}
-
 void r300_update_hyperz_state(struct r300_context* r300)
 {
     r300_update_ztop(r300);
@@ -301,51 +298,4 @@ void r300_update_hyperz_state(struct r300_context* r300)
     if (r300->hyperz_state.dirty) {
         r300_update_hyperz(r300);
     }
-
-    if (r300->hiz_clear.dirty) {
-       r300_update_hiz_clear(r300);
-    }
-}
-
-void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf)
-{
-    struct r300_resource *tex;
-    uint32_t zsize, ndw;
-    int level = surf->base.u.tex.level;
-
-    tex = r300_resource(surf->base.texture);
-
-    if (tex->hiz_mem[level])
-        return;
-
-    zsize = tex->tex.layer_size_in_bytes[level];
-    zsize /= util_format_get_blocksize(tex->b.b.b.format);
-    ndw = ALIGN_DIVUP(zsize, 64);
-
-    tex->hiz_mem[level] = u_mmAllocMem(r300->hiz_mm, ndw, 0, 0);
-}
-
-boolean r300_hyperz_init_mm(struct r300_context *r300)
-{
-    struct r300_screen* r300screen = r300->screen;
-    int frag_pipes = r300screen->caps.num_frag_pipes;
-
-    if (r300screen->caps.hiz_ram) {
-      r300->hiz_mm = u_mmInit(0, r300screen->caps.hiz_ram * frag_pipes);
-      if (!r300->hiz_mm) {
-        return FALSE;
-      }
-    }
-
-    return TRUE;
-}
-
-void r300_hyperz_destroy_mm(struct r300_context *r300)
-{
-    struct r300_screen* r300screen = r300->screen;
-
-    if (r300screen->caps.hiz_ram) {
-      u_mmDestroy(r300->hiz_mm);
-      r300->hiz_mm = NULL;
-    }
 }
diff --git a/src/gallium/drivers/r300/r300_hyperz.h b/src/gallium/drivers/r300/r300_hyperz.h
deleted file mode 100644
index d4c8e7c60a9..00000000000
--- a/src/gallium/drivers/r300/r300_hyperz.h
+++ /dev/null
@@ -1,35 +0,0 @@
-/*
- * Copyright 2010 Marek Olšák <[email protected]>
- *
- * 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
- * on 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
- * THE AUTHOR(S) AND/OR THEIR 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 R300_HYPERZ_H
-#define R300_HYPERZ_H
-
-struct r300_context;
-
-void r300_update_hyperz_state(struct r300_context* r300);
-
-void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf);
-
-boolean r300_hyperz_init_mm(struct r300_context *r300);
-void r300_hyperz_destroy_mm(struct r300_context *r300);
-
-#endif
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 1d93dab2ca2..bb30b1ab0be 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -467,6 +467,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
  *
  * See bug #9871. http://bugs.freedesktop.org/attachment.cgi?id=10672&action=view
  */
+#define R500_VAP_TEX_TO_COLOR_CNTL		0x2218
+
 #define R300_VAP_CLIP_CNTL                       0x221C
 #       define R300_VAP_UCP_ENABLE_0             (1 << 0)
 #       define R300_VAP_UCP_ENABLE_1             (1 << 1)
@@ -857,6 +859,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #	define R500_TX_DIRECTION_HORIZONTAL	 (0<<27)
 #	define R500_TX_DIRECTION_VERITCAL	 (1<<27)
 
+#define R500_SU_TEX_WRAP_PS3		    0x4114
+
 /* S Texture Coordinate of Vertex 0 for Point texture stuffing (LLC) */
 #define R300_GA_POINT_S0                              0x4200
 
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index 2ead8667bda..0ec4a225865 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -205,7 +205,7 @@ static boolean r300_reserve_cs_dwords(struct r300_context *r300,
     if (first_draw) {
         cs_dwords += r300_get_num_dirty_dwords(r300);
 
-        if (r300->screen->caps.index_bias_supported)
+        if (r300->screen->caps.is_r500)
             cs_dwords += 2; /* emit_index_offset */
 
         if (emit_vertex_arrays)
@@ -257,7 +257,7 @@ static boolean r300_emit_states(struct r300_context *r300,
         }
 
         r300_emit_dirty_state(r300);
-        if (r300->screen->caps.index_bias_supported) {
+        if (r300->screen->caps.is_r500) {
             if (r300->screen->caps.has_tcl)
                 r500_emit_index_bias(r300, index_bias);
             else
@@ -557,7 +557,7 @@ static void r300_draw_elements_immediate(struct r300_context *r300,
         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
                r300_translate_primitive(mode));
 
-        if (indexBias && !r300->screen->caps.index_bias_supported) {
+        if (indexBias && !r300->screen->caps.is_r500) {
             for (i = 0; i < count-1; i += 2)
                 OUT_CS(((ptr1[i+1] + indexBias) << 16) |
                         (ptr1[i]   + indexBias));
@@ -581,7 +581,7 @@ static void r300_draw_elements_immediate(struct r300_context *r300,
         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
                r300_translate_primitive(mode));
 
-        if (indexBias && !r300->screen->caps.index_bias_supported) {
+        if (indexBias && !r300->screen->caps.is_r500) {
             for (i = 0; i < count-1; i += 2)
                 OUT_CS(((ptr2[i+1] + indexBias) << 16) |
                         (ptr2[i]   + indexBias));
@@ -601,7 +601,7 @@ static void r300_draw_elements_immediate(struct r300_context *r300,
                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
                r300_translate_primitive(mode));
 
-        if (indexBias && !r300->screen->caps.index_bias_supported) {
+        if (indexBias && !r300->screen->caps.is_r500) {
             for (i = 0; i < count; i++)
                 OUT_CS(ptr4[i] + indexBias);
         } else {
@@ -620,13 +620,12 @@ static void r300_draw_elements(struct r300_context *r300, int indexBias,
     unsigned indexSize = r300->index_buffer.index_size;
     struct pipe_resource* orgIndexBuffer = indexBuffer;
     boolean alt_num_verts = r300->screen->caps.is_r500 &&
-                            count > 65536 &&
-                            r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
+                            count > 65536;
     unsigned short_count;
     int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
     uint16_t indices3[3];
 
-    if (indexBias && !r300->screen->caps.index_bias_supported) {
+    if (indexBias && !r300->screen->caps.is_r500) {
         r300_split_index_bias(r300, indexBias, &buffer_offset, &index_offset);
     }
 
@@ -702,8 +701,7 @@ static void r300_draw_arrays(struct r300_context *r300, unsigned mode,
                              unsigned start, unsigned count)
 {
     boolean alt_num_verts = r300->screen->caps.is_r500 &&
-                            count > 65536 &&
-                            r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
+                            count > 65536;
     unsigned short_count;
 
     /* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
@@ -748,6 +746,8 @@ static void r300_draw_vbo(struct pipe_context* pipe,
         return;
     }
 
+    r300_update_derived_state(r300);
+
     /* Start the vbuf manager and update buffers if needed. */
     u_vbuf_mgr_draw_begin(r300->vbuf_mgr, info,
                           &buffers_updated, &uploader_flushed);
@@ -756,8 +756,6 @@ static void r300_draw_vbo(struct pipe_context* pipe,
     }
 
     /* Draw. */
-    r300_update_derived_state(r300);
-
     if (indexed) {
         if (count <= 8 &&
             r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 77a9c6ad86f..52d0247fbfd 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -456,10 +456,6 @@ struct pipe_screen* r300_screen_create(struct r300_winsys_screen *rws)
     if (SCREEN_DBG_ON(r300screen, DBG_NO_HIZ))
         r300screen->caps.hiz_ram = 0;
 
-    r300screen->caps.index_bias_supported =
-            r300screen->caps.is_r500 &&
-            rws->get_value(rws, R300_VID_DRM_2_3_0);
-
     pipe_mutex_init(r300screen->num_contexts_mutex);
 
     util_slab_create(&r300screen->pool_buffers,
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 09f18b3e624..b810f4081c8 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -45,7 +45,6 @@
 #include "r300_texture.h"
 #include "r300_vs.h"
 #include "r300_winsys.h"
-#include "r300_hyperz.h"
 
 /* r300_state: Functions used to intialize state context by translating
  * Gallium state objects into semi-native r300 state objects. */
@@ -707,7 +706,7 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
     else if (state->zsbuf) {
         r300->fb_state.size += 10;
         if (can_hyperz)
-            r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4;
+            r300->fb_state.size += 8;
     }
 
     /* The size of the rest of atoms stays the same. */
@@ -720,7 +719,6 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
     struct r300_context* r300 = r300_context(pipe);
     struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
     struct pipe_framebuffer_state *old_state = r300->fb_state.state;
-    boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
     unsigned max_width, max_height, i;
     uint32_t zbuffer_bpp = 0;
 
@@ -738,28 +736,30 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
         return;
     }
 
-    if (old_state->zsbuf && r300->zmask_in_use && !r300->zmask_locked) {
+    if (old_state->zsbuf && r300->zmask_in_use && !r300->hyperz_locked) {
         /* There is a zmask in use, what are we gonna do? */
         if (state->zsbuf) {
             if (!pipe_surface_equal(old_state->zsbuf, state->zsbuf)) {
                 /* Decompress the currently bound zbuffer before we bind another one. */
                 r300_decompress_zmask(r300);
+                r300->hiz_in_use = FALSE;
             }
         } else {
             /* We don't bind another zbuffer, so lock the current one. */
-            r300->zmask_locked = TRUE;
+            r300->hyperz_locked = TRUE;
             pipe_surface_reference(&r300->locked_zbuffer, old_state->zsbuf);
         }
-    } else if (r300->zmask_locked && r300->locked_zbuffer) {
+    } else if (r300->hyperz_locked && r300->locked_zbuffer) {
         /* We have a locked zbuffer now, what are we gonna do? */
         if (state->zsbuf) {
             if (!pipe_surface_equal(r300->locked_zbuffer, state->zsbuf)) {
                 /* We are binding some other zbuffer, so decompress the locked one,
                  * it gets unlocked automatically. */
                 r300_decompress_zmask_locked_unsafe(r300);
+                r300->hiz_in_use = FALSE;
             } else {
                 /* We are binding the locked zbuffer again, so unlock it. */
-                r300->zmask_locked = FALSE;
+                r300->hyperz_locked = FALSE;
             }
         }
     }
@@ -778,7 +778,7 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
 
     util_copy_framebuffer_state(r300->fb_state.state, state);
 
-    if (!r300->zmask_locked) {
+    if (!r300->hyperz_locked) {
         pipe_surface_reference(&r300->locked_zbuffer, NULL);
     }
 
@@ -794,20 +794,6 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
             break;
         }
 
-        /* Setup Hyper-Z. */
-        if (can_hyperz) {
-            struct r300_surface *zs_surf = r300_surface(state->zsbuf);
-            struct r300_resource *tex = r300_resource(zs_surf->base.texture);
-            int level = zs_surf->base.u.tex.level;
-
-            /* work out whether we can support hiz on this buffer */
-            r300_hiz_alloc_block(r300, zs_surf);
-
-            DBG(r300, DBG_HYPERZ,
-                "hyper-z features: hiz: %d @ %08x\n", tex->hiz_mem[level] ? 1 : 0,
-                tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef);
-        }
-
         /* Polygon offset depends on the zbuffer bit depth. */
         if (r300->zbuffer_bpp != zbuffer_bpp) {
             r300->zbuffer_bpp = zbuffer_bpp;
@@ -818,27 +804,25 @@ r300_set_framebuffer_state(struct pipe_context* pipe,
     }
 
     /* Set up AA config. */
-    if (r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0)) {
-        if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) {
-            aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE;
-
-            switch (state->cbufs[0]->texture->nr_samples) {
-                case 2:
-                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2;
-                    break;
-                case 3:
-                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3;
-                    break;
-                case 4:
-                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4;
-                    break;
-                case 6:
-                    aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6;
-                    break;
-            }
-        } else {
-            aa->aa_config = 0;
+    if (state->nr_cbufs && state->cbufs[0]->texture->nr_samples > 1) {
+        aa->aa_config = R300_GB_AA_CONFIG_AA_ENABLE;
+
+        switch (state->cbufs[0]->texture->nr_samples) {
+        case 2:
+            aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_2;
+            break;
+        case 3:
+            aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_3;
+            break;
+        case 4:
+            aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_4;
+            break;
+        case 6:
+            aa->aa_config |= R300_GB_AA_CONFIG_NUM_AA_SUBSAMPLES_6;
+            break;
         }
+    } else {
+        aa->aa_config = 0;
     }
 
     if (DBG_ON(r300, DBG_FB)) {
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 003fe9a58cd..a1e116f4b61 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -29,7 +29,6 @@
 
 #include "r300_context.h"
 #include "r300_fs.h"
-#include "r300_hyperz.h"
 #include "r300_screen.h"
 #include "r300_shader_semantics.h"
 #include "r300_state_inlines.h"
@@ -642,8 +641,25 @@ static uint32_t r300_get_border_color(enum pipe_format format,
 
     /* Compressed formats. */
     if (util_format_is_compressed(format)) {
-        util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc);
-        return uc.ui;
+        switch (format) {
+        case PIPE_FORMAT_RGTC1_SNORM:
+        case PIPE_FORMAT_RGTC1_UNORM:
+            /* Add 1/32 to round the border color instead of truncating. */
+            /* The Y component is used for the border color. */
+            border_swizzled[1] = border_swizzled[2] + 1.0f/32;
+            util_pack_color(border_swizzled, PIPE_FORMAT_B4G4R4A4_UNORM, &uc);
+            return uc.ui;
+        case PIPE_FORMAT_RGTC2_SNORM:
+            border_swizzled[0] = border_swizzled[2];
+            util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_SNORM, &uc);
+            return uc.ui;
+        case PIPE_FORMAT_RGTC2_UNORM:
+            util_pack_color(border_swizzled, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
+            return uc.ui;
+        default:
+            util_pack_color(border_swizzled, PIPE_FORMAT_R8G8B8A8_UNORM, &uc);
+            return uc.ui;
+        }
     }
 
     switch (desc->channel[0].size) {
@@ -937,7 +953,7 @@ static void r300_decompress_depth_textures(struct r300_context *r300)
                           state->sampler_state_count);
     unsigned i;
 
-    if (!r300->zmask_locked || !r300->locked_zbuffer) {
+    if (!r300->hyperz_locked || !r300->locked_zbuffer) {
         return;
     }
 
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index b97c45ac198..86ad0b8b8e0 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -171,8 +171,16 @@ uint32_t r300_translate_texformat(enum pipe_format format,
             }
     }
 
-    result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view,
-                    util_format_is_compressed(format) && dxtc_swizzle);
+    if (util_format_is_compressed(format) &&
+        dxtc_swizzle &&
+        format != PIPE_FORMAT_RGTC2_UNORM &&
+        format != PIPE_FORMAT_RGTC2_SNORM) {
+        result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view,
+                                            dxtc_swizzle);
+    } else {
+        result |= r300_get_swizzle_combined(desc->swizzle, swizzle_view,
+                                            FALSE);
+    }
 
     /* S3TC formats. */
     if (desc->layout == UTIL_FORMAT_LAYOUT_S3TC) {
@@ -197,41 +205,36 @@ uint32_t r300_translate_texformat(enum pipe_format format,
         }
     }
 
-    /* Add sign. */
-    for (i = 0; i < desc->nr_channels; i++) {
-        if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
-            result |= sign_bit[i];
-        }
-    }
-
-    /* This is truly a special format.
-     * It stores R8G8 and B is computed using sqrt(1 - R^2 - G^2)
-     * in the sampler unit. Also known as D3DFMT_CxV8U8. */
-    if (format == PIPE_FORMAT_R8G8Bx_SNORM) {
-        return R300_TX_FORMAT_CxV8U8 | result;
-    }
-
     /* RGTC formats. */
     if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
         switch (format) {
             case PIPE_FORMAT_RGTC1_SNORM:
-                result |= sign_bit[0];
+                result |= sign_bit[1];
             case PIPE_FORMAT_RGTC1_UNORM:
-                result &= ~(0xfff << 9); /* mask off swizzle */
-                result |= R300_TX_FORMAT_Y << R300_TX_FORMAT_R_SHIFT;
                 return R500_TX_FORMAT_ATI1N | result;
             case PIPE_FORMAT_RGTC2_SNORM:
-                result |= sign_bit[0] | sign_bit[1];
+                result |= sign_bit[2] | sign_bit[3];
             case PIPE_FORMAT_RGTC2_UNORM:
-                result &= ~(0xfff << 9); /* mask off swizzle */
-                result |= R300_TX_FORMAT_Y << R300_TX_FORMAT_R_SHIFT |
-                          R300_TX_FORMAT_X << R300_TX_FORMAT_G_SHIFT;
                 return R400_TX_FORMAT_ATI2N | result;
             default:
                 return ~0; /* Unsupported/unknown. */
         }
     }
 
+    /* This is truly a special format.
+     * It stores R8G8 and B is computed using sqrt(1 - R^2 - G^2)
+     * in the sampler unit. Also known as D3DFMT_CxV8U8. */
+    if (format == PIPE_FORMAT_R8G8Bx_SNORM) {
+        return R300_TX_FORMAT_CxV8U8 | result;
+    }
+
+    /* Add sign. */
+    for (i = 0; i < desc->nr_channels; i++) {
+        if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
+            result |= sign_bit[i];
+        }
+    }
+
     /* See whether the components are of the same size. */
     for (i = 1; i < desc->nr_channels; i++) {
         uniform = uniform && desc->channel[0].size == desc->channel[i].size;
@@ -676,6 +679,8 @@ static void r300_texture_setup_fb_state(struct r300_surface *surf)
                 R300_DEPTHMACROTILE(tex->tex.macrotile[level]) |
                 R300_DEPTHMICROTILE(tex->tex.microtile);
         surf->format = r300_translate_zsformat(surf->base.format);
+        surf->pitch_zmask = tex->tex.zmask_stride_in_pixels[level];
+        surf->pitch_hiz = tex->tex.hiz_stride_in_pixels[level];
     } else {
         surf->pitch =
                 tex->tex.stride_in_pixels[level] |
@@ -713,14 +718,8 @@ static void r300_texture_destroy(struct pipe_screen *screen,
                                  struct pipe_resource* texture)
 {
     struct r300_resource* tex = (struct r300_resource*)texture;
-    int i;
 
     r300_winsys_bo_reference(&tex->buf, NULL);
-    for (i = 0; i < R300_MAX_TEXTURE_LEVELS; i++) {
-        if (tex->hiz_mem[i])
-            u_mmFreeMem(tex->hiz_mem[i]);
-    }
-
     FREE(tex);
 }
 
@@ -868,8 +867,7 @@ struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen,
                 break;
 
             case 2:
-                if (rws->get_value(rws, R300_VID_DRM_2_1_0))
-                    microtile = R300_BUFFER_SQUARETILED;
+                microtile = R300_BUFFER_SQUARETILED;
                 break;
         }
     }
diff --git a/src/gallium/drivers/r300/r300_texture_desc.c b/src/gallium/drivers/r300/r300_texture_desc.c
index 2cfeec7d751..3846fb8b6b3 100644
--- a/src/gallium/drivers/r300/r300_texture_desc.c
+++ b/src/gallium/drivers/r300/r300_texture_desc.c
@@ -207,29 +207,6 @@ static unsigned r300_texture_get_nblocksy(struct r300_resource *tex,
     return util_format_get_nblocksy(tex->b.b.b.format, height);
 }
 
-static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen,
-                                           struct r300_resource *tex)
-{
-    /* The kernels <= 2.6.34-rc4 compute the size of mipmapped 3D textures
-     * incorrectly. This is a workaround to prevent CS from being rejected. */
-
-    unsigned i, size;
-
-    if (!screen->rws->get_value(screen->rws, R300_VID_DRM_2_3_0) &&
-        tex->b.b.b.target == PIPE_TEXTURE_3D &&
-        tex->b.b.b.last_level > 0) {
-        size = 0;
-
-        for (i = 0; i <= tex->b.b.b.last_level; i++) {
-            size += tex->tex.stride_in_bytes[i] *
-                    r300_texture_get_nblocksy(tex, i, FALSE);
-        }
-
-        size *= tex->tex.depth0;
-        tex->tex.size_in_bytes = size;
-    }
-}
-
 /* Get a width in pixels from a stride in bytes. */
 static unsigned stride_to_width(enum pipe_format format,
                                 unsigned stride_in_bytes)
@@ -334,12 +311,17 @@ static void r300_setup_cbzb_flags(struct r300_screen *rscreen,
         tex->tex.cbzb_allowed[i] = first_level_valid && tex->tex.macrotile[i];
 }
 
-#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
+static unsigned r300_pixels_to_dwords(unsigned stride,
+                                      unsigned height,
+                                      unsigned xblock, unsigned yblock)
+{
+    return (align(stride, xblock) * align(height, yblock)) / (xblock * yblock);
+}
 
-static void r300_setup_zmask_flags(struct r300_screen *screen,
-                                   struct r300_resource *tex)
+static void r300_setup_hyperz_properties(struct r300_screen *screen,
+                                         struct r300_resource *tex)
 {
-    /* The tile size of 1 DWORD is:
+    /* The tile size of 1 DWORD in ZMASK RAM is:
      *
      * GPU    Pipes    4x4 mode   8x8 mode
      * ------------------------------------------
@@ -348,8 +330,31 @@ static void r300_setup_zmask_flags(struct r300_screen *screen,
      * RV530  1P/2Z    32x16      64x32
      *        1P/1Z    16x16      32x32
      */
-    static unsigned num_blocks_x_per_dw[4] = {4, 8, 12, 8};
-    static unsigned num_blocks_y_per_dw[4] = {4, 4,  4, 8};
+    static unsigned zmask_blocks_x_per_dw[4] = {4, 8, 12, 8};
+    static unsigned zmask_blocks_y_per_dw[4] = {4, 4,  4, 8};
+
+    /* In HIZ RAM, one dword is always 8x8 pixels (each byte is 4x4 pixels),
+     * but the blocks have very weird ordering.
+     *
+     * With 2 pipes and an image of size 8xY, where Y >= 1,
+     * clearing 4 dwords clears blocks like this:
+     *
+     *    01012323
+     *
+     * where numbers correspond to dword indices. The blocks are interleaved
+     * in the X direction, so the alignment must be 4x1 blocks (32x8 pixels).
+     *
+     * With 4 pipes and an image of size 8xY, where Y >= 4,
+     * clearing 8 dwords clears blocks like this:
+     *    01012323
+     *    45456767
+     *    01012323
+     *    45456767
+     * where numbers correspond to dword indices. The blocks are interleaved
+     * in both directions, so the alignment must be 4x4 blocks (32x32 pixels)
+     */
+    static unsigned hiz_align_x[4] = {8, 32, 48, 32};
+    static unsigned hiz_align_y[4] = {8, 8, 8, 32};
 
     if (util_format_is_depth_or_stencil(tex->b.b.b.format) &&
         util_format_get_blocksizebits(tex->b.b.b.format) == 32 &&
@@ -363,30 +368,49 @@ static void r300_setup_zmask_flags(struct r300_screen *screen,
         }
 
         for (i = 0; i <= tex->b.b.b.last_level; i++) {
-            unsigned numdw, compsize;
+            unsigned zcomp_numdw, zcompsize, hiz_numdw, stride, height;
+
+            stride = align(tex->tex.stride_in_pixels[i], 16);
+            height = u_minify(tex->b.b.b.height0, i);
 
             /* The 8x8 compression mode needs macrotiling. */
-            compsize = screen->caps.z_compress == R300_ZCOMP_8X8 &&
+            zcompsize = screen->caps.z_compress == R300_ZCOMP_8X8 &&
                        tex->tex.macrotile[i] &&
                        tex->b.b.b.nr_samples <= 1 ? 8 : 4;
 
-            /* Get the zbuffer size (with the aligned width and height). */
-            numdw = align(tex->tex.stride_in_pixels[i],
-                          num_blocks_x_per_dw[pipes-1] * compsize) *
-                    align(u_minify(tex->b.b.b.height0, i),
-                          num_blocks_y_per_dw[pipes-1] * compsize);
+            /* Get the ZMASK buffer size in dwords. */
+            zcomp_numdw = r300_pixels_to_dwords(stride, height,
+                                zmask_blocks_x_per_dw[pipes-1] * zcompsize,
+                                zmask_blocks_y_per_dw[pipes-1] * zcompsize);
 
-            /* Convert pixels -> dwords. */
-            numdw = ALIGN_DIVUP(numdw, num_blocks_x_per_dw[pipes-1] * compsize *
-                                       num_blocks_y_per_dw[pipes-1] * compsize);
+            /* Check whether we have enough ZMASK memory. */
+            if (util_format_get_blocksizebits(tex->b.b.b.format) == 32 &&
+                zcomp_numdw <= screen->caps.zmask_ram * pipes) {
+                tex->tex.zmask_dwords[i] = zcomp_numdw;
+                tex->tex.zcomp8x8[i] = zcompsize == 8;
 
-            /* Check that we have enough ZMASK memory. */
-            if (numdw <= screen->caps.zmask_ram * pipes) {
-                tex->tex.zmask_dwords[i] = numdw;
-                tex->tex.zcomp8x8[i] = compsize == 8;
+                tex->tex.zmask_stride_in_pixels[i] =
+                    align(stride, zmask_blocks_x_per_dw[pipes-1] * zcompsize);
             } else {
                 tex->tex.zmask_dwords[i] = 0;
                 tex->tex.zcomp8x8[i] = FALSE;
+                tex->tex.zmask_stride_in_pixels[i] = 0;
+            }
+
+            /* Now setup HIZ. */
+            stride = align(stride, hiz_align_x[pipes-1]);
+            height = align(height, hiz_align_y[pipes-1]);
+
+            /* Get the HIZ buffer size in dwords. */
+            hiz_numdw = (stride * height) / (8*8 * pipes);
+
+            /* Check whether we have enough HIZ memory. */
+            if (hiz_numdw <= screen->caps.hiz_ram * pipes) {
+                tex->tex.hiz_dwords[i] = hiz_numdw;
+                tex->tex.hiz_stride_in_pixels[i] = stride;
+            } else {
+                tex->tex.hiz_dwords[i] = 0;
+                tex->tex.hiz_stride_in_pixels[i] = 0;
             }
         }
     }
@@ -395,7 +419,6 @@ static void r300_setup_zmask_flags(struct r300_screen *screen,
 static void r300_setup_tiling(struct r300_screen *screen,
                               struct r300_resource *tex)
 {
-    struct r300_winsys_screen *rws = screen->rws;
     enum pipe_format format = tex->b.b.b.format;
     boolean rv350_mode = screen->caps.family >= CHIP_FAMILY_R350;
     boolean is_zb = util_format_is_depth_or_stencil(format);
@@ -422,9 +445,7 @@ static void r300_setup_tiling(struct r300_screen *screen,
             break;
 
         case 2:
-            if (rws->get_value(rws, R300_VID_DRM_2_1_0)) {
-                tex->tex.microtile = R300_BUFFER_SQUARETILED;
-            }
+            tex->tex.microtile = R300_BUFFER_SQUARETILED;
             break;
     }
 
@@ -494,8 +515,7 @@ boolean r300_texture_desc_init(struct r300_screen *rscreen,
         r300_setup_miptree(rscreen, tex, FALSE);
     }
 
-    r300_texture_3d_fix_mipmapping(rscreen, tex);
-    r300_setup_zmask_flags(rscreen, tex);
+    r300_setup_hyperz_properties(rscreen, tex);
 
     if (tex->buf_size) {
         /* Make sure the buffer we got is large enough. */
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index d5c73585c81..c0b66899f8b 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -60,10 +60,8 @@ enum r300_value_id {
     R300_VID_DRM_PATCHLEVEL,
 
     /* These should probably go away: */
-    R300_VID_DRM_2_1_0,  /* Square tiling. */
-    R300_VID_DRM_2_3_0,  /* R500 VAP regs, MSPOS regs, fixed tex3D size checking */
     R300_VID_DRM_2_6_0,  /* Hyper-Z, GB_Z_PEQ_CONFIG on rv350->r4xx, R500 FG_ALPHA_VALUE */
-    R300_VID_DRM_2_8_0,  /* R500 US_FORMAT regs, R500 ARGB2101010 colorbuffer, CMask */
+    R300_VID_DRM_2_8_0,  /* R500 US_FORMAT regs, R500 ARGB2101010 colorbuffer, CMask, R16F/RG16F */
 
     R300_CAN_HYPERZ,     /* ZMask + HiZ */
     R300_CAN_AACOMPRESS, /* CMask */
diff --git a/src/gallium/drivers/r600/eg_state_inlines.h b/src/gallium/drivers/r600/eg_state_inlines.h
index b5fcc7106fe..cae3888051b 100644
--- a/src/gallium/drivers/r600/eg_state_inlines.h
+++ b/src/gallium/drivers/r600/eg_state_inlines.h
@@ -503,9 +503,9 @@ static INLINE uint32_t r600_translate_colorformat(enum pipe_format format)
 	}
 }
 
-static INLINE boolean r600_is_sampler_format_supported(enum pipe_format format)
+static INLINE boolean r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
 {
-	return r600_translate_texformat(format, NULL, NULL, NULL) != ~0;
+	return r600_translate_texformat(screen, format, NULL, NULL, NULL) != ~0;
 }
 
 static INLINE boolean r600_is_colorbuffer_format_supported(enum pipe_format format)
diff --git a/src/gallium/drivers/r600/evergreen_state.c b/src/gallium/drivers/r600/evergreen_state.c
index 3efdbaba0c3..4206b4a201d 100644
--- a/src/gallium/drivers/r600/evergreen_state.c
+++ b/src/gallium/drivers/r600/evergreen_state.c
@@ -305,11 +305,16 @@ static void *evergreen_create_sampler_state(struct pipe_context *ctx,
 {
 	struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
 	union util_color uc;
+	uint32_t coord_trunc = 0;
 
 	if (rstate == NULL) {
 		return NULL;
 	}
 
+	if ((state->mag_img_filter == PIPE_TEX_FILTER_NEAREST) ||
+	    (state->min_img_filter == PIPE_TEX_FILTER_NEAREST))
+		coord_trunc = 1;
+
 	rstate->id = R600_PIPE_STATE_SAMPLER;
 	util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
 	r600_pipe_state_add_reg(rstate, R_03C000_SQ_TEX_SAMPLER_WORD0_0,
@@ -328,6 +333,7 @@ static void *evergreen_create_sampler_state(struct pipe_context *ctx,
 			0xFFFFFFFF, NULL);
 	r600_pipe_state_add_reg(rstate, R_03C008_SQ_TEX_SAMPLER_WORD2_0,
 				S_03C008_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 8)) |
+				S_03C008_MC_COORD_TRUNCATE(coord_trunc) |
 				S_03C008_TYPE(1),
 				0xFFFFFFFF, NULL);
 
@@ -370,7 +376,7 @@ static struct pipe_sampler_view *evergreen_create_sampler_view(struct pipe_conte
 	swizzle[1] = state->swizzle_g;
 	swizzle[2] = state->swizzle_b;
 	swizzle[3] = state->swizzle_a;
-	format = r600_translate_texformat(state->format,
+	format = r600_translate_texformat(ctx->screen, state->format,
 					  swizzle,
 					  &word4, &yuv_format);
 	if (format == ~0) {
diff --git a/src/gallium/drivers/r600/evergreend.h b/src/gallium/drivers/r600/evergreend.h
index f0a1ee0cd02..c51a163bd06 100644
--- a/src/gallium/drivers/r600/evergreend.h
+++ b/src/gallium/drivers/r600/evergreend.h
@@ -108,8 +108,9 @@
 #define PKT3_IT_OPCODE_S(x)             (((x) & 0xFF) << 8)
 #define PKT3_IT_OPCODE_G(x)             (((x) >> 8) & 0xFF)
 #define PKT3_IT_OPCODE_C                0xFFFF00FF
+#define PKT3_PREDICATE(x)               (((x) >> 0) & 0x1)
 #define PKT0(index, count) (PKT_TYPE_S(0) | PKT0_BASE_INDEX_S(index) | PKT_COUNT_S(count))
-#define PKT3(op, count) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count))
+#define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count) | PKT3_PREDICATE(predicate))
 
 /* Registers */
 #define R_008C00_SQ_CONFIG                           0x00008C00
diff --git a/src/gallium/drivers/r600/r600.h b/src/gallium/drivers/r600/r600.h
index 1b76f0098dd..0b7d6f70968 100644
--- a/src/gallium/drivers/r600/r600.h
+++ b/src/gallium/drivers/r600/r600.h
@@ -114,6 +114,8 @@ enum radeon_family r600_get_family(struct radeon *rw);
 enum chip_class r600_get_family_class(struct radeon *radeon);
 struct r600_tiling_info *r600_get_tiling_info(struct radeon *radeon);
 unsigned r600_get_clock_crystal_freq(struct radeon *radeon);
+unsigned r600_get_minor_version(struct radeon *radeon);
+unsigned r600_get_num_backends(struct radeon *radeon);
 
 /* r600_bo.c */
 struct r600_bo;
@@ -251,6 +253,7 @@ struct r600_context {
 	unsigned		num_query_running;
 	struct list_head	fenced_bo;
 	unsigned                max_db; /* for OQ */
+	boolean                 predicate_drawing;
 };
 
 struct r600_draw {
@@ -283,6 +286,8 @@ void r600_query_begin(struct r600_context *ctx, struct r600_query *query);
 void r600_query_end(struct r600_context *ctx, struct r600_query *query);
 void r600_context_queries_suspend(struct r600_context *ctx);
 void r600_context_queries_resume(struct r600_context *ctx);
+void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation,
+			    int flag_wait);
 
 int evergreen_context_init(struct r600_context *ctx, struct radeon *radeon);
 void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *draw);
diff --git a/src/gallium/drivers/r600/r600_asm.c b/src/gallium/drivers/r600/r600_asm.c
index 3e478382801..d7a7928386f 100644
--- a/src/gallium/drivers/r600/r600_asm.c
+++ b/src/gallium/drivers/r600/r600_asm.c
@@ -3061,7 +3061,7 @@ int r600_vertex_elements_build_fetch_shader(struct r600_pipe_context *rctx, stru
 		vtx.fetch_type = elements[i].instance_divisor ? 1 : 0;
 		vtx.src_gpr = elements[i].instance_divisor > 1 ? i + 1 : 0;
 		vtx.src_sel_x = elements[i].instance_divisor ? 3 : 0;
-		vtx.mega_fetch_count = 16;
+		vtx.mega_fetch_count = 0x1F;
 		vtx.dst_gpr = i + 1;
 		vtx.dst_sel_x = desc->swizzle[0];
 		vtx.dst_sel_y = desc->swizzle[1];
diff --git a/src/gallium/drivers/r600/r600_pipe.c b/src/gallium/drivers/r600/r600_pipe.c
index 3fd6668f718..651f994a502 100644
--- a/src/gallium/drivers/r600/r600_pipe.c
+++ b/src/gallium/drivers/r600/r600_pipe.c
@@ -431,7 +431,7 @@ static boolean r600_is_format_supported(struct pipe_screen* screen,
 		return FALSE;
 
 	if ((usage & PIPE_BIND_SAMPLER_VIEW) &&
-	    r600_is_sampler_format_supported(format)) {
+	    r600_is_sampler_format_supported(screen, format)) {
 		retval |= PIPE_BIND_SAMPLER_VIEW;
 	}
 
diff --git a/src/gallium/drivers/r600/r600_pipe.h b/src/gallium/drivers/r600/r600_pipe.h
index 8dc1f4ad5c3..5f701d87e8f 100644
--- a/src/gallium/drivers/r600/r600_pipe.h
+++ b/src/gallium/drivers/r600/r600_pipe.h
@@ -227,7 +227,7 @@ int r600_conv_pipe_prim(unsigned pprim, unsigned *prim);
 /* r600_texture.c */
 void r600_init_screen_texture_functions(struct pipe_screen *screen);
 void r600_init_surface_functions(struct r600_pipe_context *r600);
-uint32_t r600_translate_texformat(enum pipe_format format,
+uint32_t r600_translate_texformat(struct pipe_screen *screen, enum pipe_format format,
 				  const unsigned char *swizzle_view,
 				  uint32_t *word4_p, uint32_t *yuv_format_p);
 unsigned r600_texture_get_offset(struct r600_resource_texture *rtex,
diff --git a/src/gallium/drivers/r600/r600_query.c b/src/gallium/drivers/r600/r600_query.c
index 726668260cc..343403f92f3 100644
--- a/src/gallium/drivers/r600/r600_query.c
+++ b/src/gallium/drivers/r600/r600_query.c
@@ -21,6 +21,7 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE.
  */
 #include "r600_pipe.h"
+#include "r600d.h"
 
 static struct pipe_query *r600_create_query(struct pipe_context *ctx, unsigned query_type)
 {
@@ -66,6 +67,30 @@ static boolean r600_get_query_result(struct pipe_context *ctx,
 	return r600_context_query_result(&rctx->ctx, (struct r600_query *)query, wait, vresult);
 }
 
+static void r600_render_condition(struct pipe_context *ctx,
+				  struct pipe_query *query,
+				  uint mode)
+{
+	struct r600_pipe_context *rctx = (struct r600_pipe_context *)ctx;
+	struct r600_query *rquery = (struct r600_query *)query;
+	int wait_flag = 0;
+
+	if (!query) {
+		rctx->ctx.predicate_drawing = false;
+		r600_query_predication(&rctx->ctx, NULL, PREDICATION_OP_CLEAR, 1);
+		return;
+	}
+
+	if (mode == PIPE_RENDER_COND_WAIT ||
+	    mode == PIPE_RENDER_COND_BY_REGION_WAIT) {
+		wait_flag = 1;
+	}
+
+	rctx->ctx.predicate_drawing = true;
+	r600_query_predication(&rctx->ctx, rquery, PREDICATION_OP_ZPASS, wait_flag);
+	
+}
+
 void r600_init_query_functions(struct r600_pipe_context *rctx)
 {
 	rctx->context.create_query = r600_create_query;
@@ -73,4 +98,7 @@ void r600_init_query_functions(struct r600_pipe_context *rctx)
 	rctx->context.begin_query = r600_begin_query;
 	rctx->context.end_query = r600_end_query;
 	rctx->context.get_query_result = r600_get_query_result;
+
+	if (r600_get_num_backends(rctx->screen->radeon) > 0)
+	    rctx->context.render_condition = r600_render_condition;
 }
diff --git a/src/gallium/drivers/r600/r600_state.c b/src/gallium/drivers/r600/r600_state.c
index 576067ae81e..1be641798f7 100644
--- a/src/gallium/drivers/r600/r600_state.c
+++ b/src/gallium/drivers/r600/r600_state.c
@@ -358,11 +358,16 @@ static void *r600_create_sampler_state(struct pipe_context *ctx,
 {
 	struct r600_pipe_state *rstate = CALLOC_STRUCT(r600_pipe_state);
 	union util_color uc;
+	uint32_t coord_trunc = 0;
 
 	if (rstate == NULL) {
 		return NULL;
 	}
 
+	if ((state->mag_img_filter == PIPE_TEX_FILTER_NEAREST) ||
+	    (state->min_img_filter == PIPE_TEX_FILTER_NEAREST))
+		coord_trunc = 1;
+
 	rstate->id = R600_PIPE_STATE_SAMPLER;
 	util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
 	r600_pipe_state_add_reg(rstate, R_03C000_SQ_TEX_SAMPLER_WORD0_0,
@@ -379,7 +384,9 @@ static void *r600_create_sampler_state(struct pipe_context *ctx,
 			S_03C004_MIN_LOD(S_FIXED(CLAMP(state->min_lod, 0, 15), 6)) |
 			S_03C004_MAX_LOD(S_FIXED(CLAMP(state->max_lod, 0, 15), 6)) |
 			S_03C004_LOD_BIAS(S_FIXED(CLAMP(state->lod_bias, -16, 16), 6)), 0xFFFFFFFF, NULL);
-	r600_pipe_state_add_reg(rstate, R_03C008_SQ_TEX_SAMPLER_WORD2_0, S_03C008_TYPE(1), 0xFFFFFFFF, NULL);
+	r600_pipe_state_add_reg(rstate, R_03C008_SQ_TEX_SAMPLER_WORD2_0,
+				S_03C008_MC_COORD_TRUNCATE(coord_trunc) |
+				S_03C008_TYPE(1), 0xFFFFFFFF, NULL);
 	if (uc.ui) {
 		r600_pipe_state_add_reg(rstate, R_00A400_TD_PS_SAMPLER0_BORDER_RED, fui(state->border_color[0]), 0xFFFFFFFF, NULL);
 		r600_pipe_state_add_reg(rstate, R_00A404_TD_PS_SAMPLER0_BORDER_GREEN, fui(state->border_color[1]), 0xFFFFFFFF, NULL);
@@ -420,7 +427,7 @@ static struct pipe_sampler_view *r600_create_sampler_view(struct pipe_context *c
 	swizzle[1] = state->swizzle_g;
 	swizzle[2] = state->swizzle_b;
 	swizzle[3] = state->swizzle_a;
-	format = r600_translate_texformat(state->format,
+	format = r600_translate_texformat(ctx->screen, state->format,
 					  swizzle,
 					  &word4, &yuv_format);
 	if (format == ~0) {
diff --git a/src/gallium/drivers/r600/r600_state_inlines.h b/src/gallium/drivers/r600/r600_state_inlines.h
index 3dd54f45202..9e6ae295239 100644
--- a/src/gallium/drivers/r600/r600_state_inlines.h
+++ b/src/gallium/drivers/r600/r600_state_inlines.h
@@ -500,9 +500,9 @@ static INLINE uint32_t r600_translate_colorformat(enum pipe_format format)
 	}
 }
 
-static INLINE boolean r600_is_sampler_format_supported(enum pipe_format format)
+static INLINE boolean r600_is_sampler_format_supported(struct pipe_screen *screen, enum pipe_format format)
 {
-	return r600_translate_texformat(format, NULL, NULL, NULL) != ~0;
+	return r600_translate_texformat(screen, format, NULL, NULL, NULL) != ~0;
 }
 
 static INLINE boolean r600_is_colorbuffer_format_supported(enum pipe_format format)
diff --git a/src/gallium/drivers/r600/r600_texture.c b/src/gallium/drivers/r600/r600_texture.c
index 048d0b61e3b..7d3d022d973 100644
--- a/src/gallium/drivers/r600/r600_texture.c
+++ b/src/gallium/drivers/r600/r600_texture.c
@@ -226,7 +226,7 @@ static void r600_texture_set_array_mode(struct pipe_screen *screen,
 
 		w = mip_minify(ptex->width0, level);
 		h = mip_minify(ptex->height0, level);
-		if (w < tile_width || h < tile_height)
+		if (w <= tile_width || h <= tile_height)
 			rtex->array_mode[level] = V_0280A0_ARRAY_1D_TILED_THIN1;
 		else
 			rtex->array_mode[level] = array_mode;
@@ -422,8 +422,13 @@ struct pipe_resource *r600_texture_create(struct pipe_screen *screen,
 
 	/* Would like some magic "get_bool_option_once" routine.
 	 */
-	if (force_tiling == -1)
-		force_tiling = debug_get_bool_option("R600_FORCE_TILING", FALSE);
+	if (force_tiling == -1) {
+		struct r600_screen *rscreen = (struct r600_screen *)screen;
+		if (r600_get_minor_version(rscreen->radeon) >= 9)
+			force_tiling = debug_get_bool_option("R600_TILING", TRUE);
+		else
+			force_tiling = debug_get_bool_option("R600_TILING", FALSE);
+	}
 
 	if (force_tiling && permit_hardware_blit(screen, templ)) {
 		if (!(templ->flags & R600_RESOURCE_FLAG_TRANSFER) &&
@@ -813,7 +818,8 @@ static unsigned r600_get_swizzle_combined(const unsigned char *swizzle_format,
 }
 
 /* texture format translate */
-uint32_t r600_translate_texformat(enum pipe_format format,
+uint32_t r600_translate_texformat(struct pipe_screen *screen,
+				  enum pipe_format format,
 				  const unsigned char *swizzle_view,
 				  uint32_t *word4_p, uint32_t *yuv_format_p)
 {
@@ -879,8 +885,13 @@ uint32_t r600_translate_texformat(enum pipe_format format,
 		break;
 	}
 
-	if (r600_enable_s3tc == -1)
-		r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+	if (r600_enable_s3tc == -1) {
+		struct r600_screen *rscreen = (struct r600_screen *)screen;
+		if (r600_get_minor_version(rscreen->radeon) >= 9)
+			r600_enable_s3tc = 1;
+		else
+			r600_enable_s3tc = debug_get_bool_option("R600_ENABLE_S3TC", FALSE);
+	}
 
 	if (desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
 		if (!r600_enable_s3tc)
diff --git a/src/gallium/drivers/r600/r600d.h b/src/gallium/drivers/r600/r600d.h
index e8558c49a7c..df70e2889e2 100644
--- a/src/gallium/drivers/r600/r600d.h
+++ b/src/gallium/drivers/r600/r600d.h
@@ -67,6 +67,10 @@
 #define PKT3_SET_CTL_CONST                     0x6F
 #define PKT3_SURFACE_BASE_UPDATE               0x73
 
+#define PREDICATION_OP_CLEAR 0x0
+#define PREDICATION_OP_ZPASS 0x1
+#define PREDICATION_OP_PRIMCOUNT 0x2
+
 #define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
 #define PKT_TYPE_G(x)                   (((x) >> 30) & 0x3)
 #define PKT_TYPE_C                      0x3FFFFFFF
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index 603e1de7982..a06817c5735 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -249,11 +249,6 @@ softpipe_is_format_supported( struct pipe_screen *screen,
       return util_format_s3tc_enabled;
    }
 
-   /* u_format doesn't implement RGTC yet */
-   if (format_desc->layout == UTIL_FORMAT_LAYOUT_RGTC) {
-	return FALSE;
-   }
-
    /*
     * Everything else should be supported by u_format.
     */
diff --git a/src/gallium/drivers/softpipe/sp_state_shader.c b/src/gallium/drivers/softpipe/sp_state_shader.c
index 66ddc565722..3dec5de3cc4 100644
--- a/src/gallium/drivers/softpipe/sp_state_shader.c
+++ b/src/gallium/drivers/softpipe/sp_state_shader.c
@@ -91,8 +91,6 @@ softpipe_bind_fs_state(struct pipe_context *pipe, void *fs)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   draw_flush(softpipe->draw);
-
    if (softpipe->fs == fs)
       return;
 
diff --git a/src/gallium/drivers/svga/svga_resource_buffer_upload.c b/src/gallium/drivers/svga/svga_resource_buffer_upload.c
index 76a3803224a..b7d54605e66 100644
--- a/src/gallium/drivers/svga/svga_resource_buffer_upload.c
+++ b/src/gallium/drivers/svga/svga_resource_buffer_upload.c
@@ -40,7 +40,7 @@
 #include "svga_debug.h"
 
 
-#define MAX_DMA_SIZE (8 * 1024 * 1024)
+#define MAX_DMA_SIZE (4 * 1024 * 1024)
 
 
 /**
diff --git a/src/gallium/drivers/svga/svga_state_constants.c b/src/gallium/drivers/svga/svga_state_constants.c
index daf1024fd02..6c3275e74c0 100644
--- a/src/gallium/drivers/svga/svga_state_constants.c
+++ b/src/gallium/drivers/svga/svga_state_constants.c
@@ -40,9 +40,12 @@
 
 /* Convert from PIPE_SHADER_* to SVGA3D_SHADERTYPE_*
  */
-static int svga_shader_type( int unit )
+static int svga_shader_type( int shader )
 {
-   return unit + 1;
+   assert(PIPE_SHADER_VERTEX + 1 == SVGA3D_SHADERTYPE_VS);
+   assert(PIPE_SHADER_FRAGMENT + 1 == SVGA3D_SHADERTYPE_PS);
+   assert(shader <= PIPE_SHADER_FRAGMENT);
+   return shader + 1;
 }
 
 
diff --git a/src/gallium/state_trackers/dri/drm/dri2.c b/src/gallium/state_trackers/dri/drm/dri2.c
index 0181588096d..a5c94b4166b 100644
--- a/src/gallium/state_trackers/dri/drm/dri2.c
+++ b/src/gallium/state_trackers/dri/drm/dri2.c
@@ -317,6 +317,7 @@ dri2_allocate_buffer(__DRIscreen *sPriv,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    buffer->resource =
       screen->base.screen->resource_create(screen->base.screen, &templ);
@@ -510,6 +511,7 @@ dri2_create_image(__DRIscreen *_screen,
    templ.width0 = width;
    templ.height0 = height;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    img->texture = screen->base.screen->resource_create(screen->base.screen, &templ);
    if (!img->texture) {
diff --git a/src/gallium/state_trackers/egl/Makefile b/src/gallium/state_trackers/egl/Makefile
index 98167cceeb3..68643214060 100644
--- a/src/gallium/state_trackers/egl/Makefile
+++ b/src/gallium/state_trackers/egl/Makefile
@@ -26,6 +26,7 @@ x11_OBJECTS = $(x11_SOURCES:.c=.o)
 wayland_INCLUDES = \
 	-I$(TOP)/src/gallium/winsys \
 	-I$(TOP)/src/egl/wayland \
+	-I$(TOP)/src/egl/wayland/wayland-drm \
 	$(shell pkg-config --cflags-only-I libdrm wayland-client)
 
 wayland_SOURCES = $(wildcard wayland/*.c)
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d.c b/src/gallium/state_trackers/egl/common/egl_g3d.c
index 6107df48822..e60a56074dd 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d.c
@@ -455,9 +455,6 @@ egl_g3d_terminate(_EGLDriver *drv, _EGLDisplay *dpy)
 
    _eglReleaseDisplayResources(drv, dpy);
 
-   if (gdpy->pipe)
-      gdpy->pipe->destroy(gdpy->pipe);
-
    if (dpy->Configs) {
       _eglDestroyArray(dpy->Configs, egl_g3d_free_config);
       dpy->Configs = NULL;
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d.h b/src/gallium/state_trackers/egl/common/egl_g3d.h
index 9873fee6ec2..301db3128ff 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d.h
+++ b/src/gallium/state_trackers/egl/common/egl_g3d.h
@@ -56,7 +56,6 @@ struct egl_g3d_display {
 
    const struct egl_g3d_loader *loader;
    struct st_manager *smapi;
-   struct pipe_context *pipe;
 };
 
 struct egl_g3d_context {
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d_api.c b/src/gallium/state_trackers/egl/common/egl_g3d_api.c
index c9f94a3c498..2068256dff2 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d_api.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d_api.c
@@ -640,6 +640,7 @@ egl_g3d_copy_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
    _EGLContext *ctx = _eglGetCurrentContext();
    struct native_surface *nsurf;
    struct pipe_resource *ptex;
+   struct pipe_context *pipe;
 
    if (!gsurf->render_texture)
       return EGL_TRUE;
@@ -655,22 +656,18 @@ egl_g3d_copy_buffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
             PIPE_FLUSH_RENDER_CACHE | PIPE_FLUSH_FRAME, NULL);
    }
 
-   /* create a pipe context to copy surfaces */
-   if (!gdpy->pipe) {
-      gdpy->pipe =
-         gdpy->native->screen->context_create(gdpy->native->screen, NULL);
-      if (!gdpy->pipe)
-         return EGL_FALSE;
-   }
+   pipe = ndpy_get_copy_context(gdpy->native);
+   if (!pipe)
+      return EGL_FALSE;
 
    ptex = get_pipe_resource(gdpy->native, nsurf, NATIVE_ATTACHMENT_FRONT_LEFT);
    if (ptex) {
       struct pipe_box src_box;
 
       u_box_origin_2d(ptex->width0, ptex->height0, &src_box);
-      gdpy->pipe->resource_copy_region(gdpy->pipe, ptex, 0, 0, 0, 0,
+      pipe->resource_copy_region(pipe, ptex, 0, 0, 0, 0,
             gsurf->render_texture, 0, &src_box);
-      gdpy->pipe->flush(gdpy->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
+      pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
       nsurf->present(nsurf, NATIVE_ATTACHMENT_FRONT_LEFT, FALSE, 0);
 
       pipe_resource_reference(&ptex, NULL);
diff --git a/src/gallium/state_trackers/egl/common/egl_g3d_image.c b/src/gallium/state_trackers/egl/common/egl_g3d_image.c
index 81ce7ab45c7..e1c83168b3a 100644
--- a/src/gallium/state_trackers/egl/common/egl_g3d_image.c
+++ b/src/gallium/state_trackers/egl/common/egl_g3d_image.c
@@ -111,6 +111,7 @@ egl_g3d_create_drm_buffer(_EGLDisplay *dpy, _EGLImage *img,
    templ.width0 = attrs.Width;
    templ.height0 = attrs.Height;
    templ.depth0 = 1;
+   templ.array_size = 1;
 
    /*
     * XXX fix apps (e.g. wayland) and pipe drivers (e.g. i915) and remove the
diff --git a/src/gallium/state_trackers/egl/common/native.h b/src/gallium/state_trackers/egl/common/native.h
index 42889075a39..9246f8c32a4 100644
--- a/src/gallium/state_trackers/egl/common/native.h
+++ b/src/gallium/state_trackers/egl/common/native.h
@@ -142,6 +142,11 @@ struct native_display {
    struct pipe_screen *screen;
 
    /**
+    * Context used for copy operations.
+    */
+   struct pipe_context *pipe;
+
+   /**
     * Available for caller's use.
     */
    void *user_data;
@@ -223,6 +228,29 @@ native_attachment_mask_test(uint mask, enum native_attachment att)
    return !!(mask & (1 << att));
 }
 
+/**
+ * Get the display copy context
+ */
+static INLINE struct pipe_context *
+ndpy_get_copy_context(struct native_display *ndpy)
+{
+   if (!ndpy->pipe)
+      ndpy->pipe = ndpy->screen->context_create(ndpy->screen, NULL);
+   return ndpy->pipe;
+}
+
+/**
+ * Free display screen and context resources
+ */
+static INLINE void
+ndpy_uninit(struct native_display *ndpy)
+{
+   if (ndpy->pipe)
+      ndpy->pipe->destroy(ndpy->pipe);
+   if (ndpy->screen)
+      ndpy->screen->destroy(ndpy->screen);
+}
+
 struct native_platform {
    const char *name;
 
diff --git a/src/gallium/state_trackers/egl/common/native_helper.c b/src/gallium/state_trackers/egl/common/native_helper.c
index a9d8f32e007..be6713d03a8 100644
--- a/src/gallium/state_trackers/egl/common/native_helper.c
+++ b/src/gallium/state_trackers/egl/common/native_helper.c
@@ -3,6 +3,7 @@
  * Version:  7.9
  *
  * Copyright (C) 2010 LunarG Inc.
+ * Copyright (C) 2011 VMware Inc. 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"),
@@ -24,6 +25,7 @@
  *
  * Authors:
  *    Chia-I Wu <[email protected]>
+ *    Thomas Hellstrom <[email protected]>
  */
 
 #include "util/u_inlines.h"
@@ -34,6 +36,14 @@
 
 #include "native_helper.h"
 
+/**
+ * Number of swap fences and mask
+ */
+
+#define EGL_SWAP_FENCES_MAX 4
+#define EGL_SWAP_FENCES_MASK 3
+#define EGL_SWAP_FENCES_DEFAULT 1
+
 struct resource_surface {
    struct pipe_screen *screen;
    enum pipe_format format;
@@ -42,6 +52,15 @@ struct resource_surface {
    struct pipe_resource *resources[NUM_NATIVE_ATTACHMENTS];
    uint resource_mask;
    uint width, height;
+
+   /**
+    * Swap fences.
+    */
+   struct pipe_fence_handle *swap_fences[EGL_SWAP_FENCES_MAX];
+   unsigned int cur_fences;
+   unsigned int head;
+   unsigned int tail;
+   unsigned int desired_fences;
 };
 
 struct resource_surface *
@@ -49,11 +68,16 @@ resource_surface_create(struct pipe_screen *screen,
                         enum pipe_format format, uint bind)
 {
    struct resource_surface *rsurf = CALLOC_STRUCT(resource_surface);
+   char *swap_fences = getenv("EGL_THROTTLE_FENCES");
 
    if (rsurf) {
       rsurf->screen = screen;
       rsurf->format = format;
       rsurf->bind = bind;
+      rsurf->desired_fences = (swap_fences) ? atoi(swap_fences) :
+	 EGL_SWAP_FENCES_DEFAULT;
+      if (rsurf->desired_fences > EGL_SWAP_FENCES_MAX)
+	 rsurf->desired_fences = EGL_SWAP_FENCES_MAX;
    }
 
    return rsurf;
@@ -225,3 +249,121 @@ resource_surface_present(struct resource_surface *rsurf,
 
    return TRUE;
 }
+
+/**
+ * Schedule a copy swap from the back to the front buffer using the
+ * native display's copy context.
+ */
+boolean
+resource_surface_copy_swap(struct resource_surface *rsurf,
+			   struct native_display *ndpy)
+{
+   struct pipe_resource *ftex;
+   struct pipe_resource *btex;
+   struct pipe_context *pipe;
+   struct pipe_box src_box;
+   boolean ret = FALSE;
+
+   pipe = ndpy_get_copy_context(ndpy);
+   if (!pipe)
+      return FALSE;
+
+   ftex = resource_surface_get_single_resource(rsurf,
+					       NATIVE_ATTACHMENT_FRONT_LEFT);
+   if (!ftex)
+      goto out_no_ftex;
+   btex = resource_surface_get_single_resource(rsurf,
+					       NATIVE_ATTACHMENT_BACK_LEFT);
+   if (!btex)
+      goto out_no_btex;
+
+   u_box_origin_2d(ftex->width0, ftex->height0, &src_box);
+   pipe->resource_copy_region(pipe, ftex, 0, 0, 0, 0,
+			      btex, 0, &src_box);
+   ret = TRUE;
+
+ out_no_ftex:
+   pipe_resource_reference(&btex, NULL);
+ out_no_btex:
+   pipe_resource_reference(&ftex, NULL);
+
+   return ret;
+}
+
+static struct pipe_fence_handle *
+swap_fences_pop_front(struct resource_surface *rsurf)
+{
+   struct pipe_screen *screen = rsurf->screen;
+   struct pipe_fence_handle *fence = NULL;
+
+   if (rsurf->desired_fences == 0)
+      return NULL;
+
+   if (rsurf->cur_fences >= rsurf->desired_fences) {
+      screen->fence_reference(screen, &fence, rsurf->swap_fences[rsurf->tail]);
+      screen->fence_reference(screen, &rsurf->swap_fences[rsurf->tail++], NULL);
+      rsurf->tail &= EGL_SWAP_FENCES_MASK;
+      --rsurf->cur_fences;
+   }
+   return fence;
+}
+
+static void
+swap_fences_push_back(struct resource_surface *rsurf,
+		      struct pipe_fence_handle *fence)
+{
+   struct pipe_screen *screen = rsurf->screen;
+
+   if (!fence || rsurf->desired_fences == 0)
+      return;
+
+   while(rsurf->cur_fences == rsurf->desired_fences)
+      swap_fences_pop_front(rsurf);
+
+   rsurf->cur_fences++;
+   screen->fence_reference(screen, &rsurf->swap_fences[rsurf->head++],
+			   fence);
+   rsurf->head &= EGL_SWAP_FENCES_MASK;
+}
+
+boolean
+resource_surface_throttle(struct resource_surface *rsurf)
+{
+   struct pipe_screen *screen = rsurf->screen;
+   struct pipe_fence_handle *fence = swap_fences_pop_front(rsurf);
+
+   if (fence) {
+      (void) screen->fence_finish(screen, fence, 0);
+      screen->fence_reference(screen, &fence, NULL);
+      return TRUE;
+   }
+
+   return FALSE;
+}
+
+boolean
+resource_surface_flush(struct resource_surface *rsurf,
+		       struct native_display *ndpy)
+{
+   struct pipe_fence_handle *fence = NULL;
+   struct pipe_screen *screen = rsurf->screen;
+   struct pipe_context *pipe= ndpy_get_copy_context(ndpy);
+
+   if (!pipe)
+      return FALSE;
+
+   pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, &fence);
+   if (fence == NULL)
+      return FALSE;
+
+   swap_fences_push_back(rsurf, fence);
+   screen->fence_reference(screen, &fence, NULL);
+
+   return TRUE;
+}
+
+void
+resource_surface_wait(struct resource_surface *rsurf)
+{
+   while (resource_surface_throttle(rsurf));
+}
diff --git a/src/gallium/state_trackers/egl/common/native_helper.h b/src/gallium/state_trackers/egl/common/native_helper.h
index 03995de19ae..39564a04365 100644
--- a/src/gallium/state_trackers/egl/common/native_helper.h
+++ b/src/gallium/state_trackers/egl/common/native_helper.h
@@ -3,6 +3,7 @@
  * Version:  7.9
  *
  * Copyright (C) 2010 LunarG Inc.
+ * Copyright (C) 2011 VMware Inc. 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"),
@@ -24,6 +25,7 @@
  *
  * Authors:
  *    Chia-I Wu <[email protected]>
+ *    Thomas Hellstrom <[email protected]>
  */
 
 #include "native.h"
@@ -74,3 +76,32 @@ boolean
 resource_surface_present(struct resource_surface *rsurf,
                          enum native_attachment which,
                          void *winsys_drawable_handle);
+
+/**
+ * Perform a gallium copy blit between the back left and front left
+ * surfaces. Needs to be followed by a call to resource_surface_flush.
+ */
+boolean
+resource_surface_copy_swap(struct resource_surface *rsurf,
+			   struct native_display *ndpy);
+
+/**
+ * Throttle on outstanding rendering using the copy context. For example
+ * copy swaps.
+ */
+boolean
+resource_surface_throttle(struct resource_surface *rsurf);
+
+/**
+ * Flush pending rendering using the copy context. This function saves a
+ * marker for upcoming throttles.
+ */
+boolean
+resource_surface_flush(struct resource_surface *rsurf,
+		       struct native_display *ndpy);
+/**
+ * Wait for all rendering using the copy context to be complete. Frees all
+ * throttle markers saved using resource_surface_flush.
+ */
+void
+resource_surface_wait(struct resource_surface *rsurf);
diff --git a/src/gallium/state_trackers/egl/drm/modeset.c b/src/gallium/state_trackers/egl/drm/modeset.c
index 0cc06caa2a1..3fff9540905 100644
--- a/src/gallium/state_trackers/egl/drm/modeset.c
+++ b/src/gallium/state_trackers/egl/drm/modeset.c
@@ -3,6 +3,7 @@
  * Version:  7.9
  *
  * Copyright (C) 2010 LunarG Inc.
+ * Copyright (C) 2011 VMware Inc. 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"),
@@ -24,6 +25,7 @@
  *
  * Authors:
  *    Chia-I Wu <[email protected]>
+ *    Thomas Hellstrom <[email protected]>
  */
 
 #include "util/u_memory.h"
@@ -131,6 +133,25 @@ drm_surface_flush_frontbuffer(struct native_surface *nsurf)
 }
 
 static boolean
+drm_surface_copy_swap(struct native_surface *nsurf)
+{
+   struct drm_surface *drmsurf = drm_surface(nsurf);
+   struct drm_display *drmdpy = drmsurf->drmdpy;
+
+   (void) resource_surface_throttle(drmsurf->rsurf);
+   if (!resource_surface_copy_swap(drmsurf->rsurf, &drmdpy->base))
+      return FALSE;
+
+   (void) resource_surface_flush(drmsurf->rsurf, &drmdpy->base);
+   if (!drm_surface_flush_frontbuffer(nsurf))
+      return FALSE;
+
+   drmsurf->sequence_number++;
+
+   return TRUE;
+}
+
+static boolean
 drm_surface_swap_buffers(struct native_surface *nsurf)
 {
    struct drm_surface *drmsurf = drm_surface(nsurf);
@@ -139,17 +160,21 @@ drm_surface_swap_buffers(struct native_surface *nsurf)
    struct drm_framebuffer tmp_fb;
    int err;
 
+   if (!drmsurf->have_pageflip)
+      return drm_surface_copy_swap(nsurf);
+
    if (!drmsurf->back_fb.buffer_id) {
       if (!drm_surface_init_framebuffers(&drmsurf->base, TRUE))
          return FALSE;
    }
 
    if (drmsurf->is_shown && drmcrtc->crtc) {
-      err = drmModeSetCrtc(drmdpy->fd, drmcrtc->crtc->crtc_id,
-            drmsurf->back_fb.buffer_id, drmcrtc->crtc->x, drmcrtc->crtc->y,
-            drmcrtc->connectors, drmcrtc->num_connectors, &drmcrtc->crtc->mode);
-      if (err)
-         return FALSE;
+      err = drmModePageFlip(drmdpy->fd, drmcrtc->crtc->crtc_id,
+			    drmsurf->back_fb.buffer_id, 0, NULL);
+      if (err) {
+	 drmsurf->have_pageflip = FALSE;
+         return drm_surface_copy_swap(nsurf);
+      }
    }
 
    /* swap the buffers */
@@ -175,7 +200,7 @@ drm_surface_present(struct native_surface *nsurf,
 {
    boolean ret;
 
-   if (preserve || swap_interval)
+   if (swap_interval)
       return FALSE;
 
    switch (natt) {
@@ -183,7 +208,10 @@ drm_surface_present(struct native_surface *nsurf,
       ret = drm_surface_flush_frontbuffer(nsurf);
       break;
    case NATIVE_ATTACHMENT_BACK_LEFT:
-      ret = drm_surface_swap_buffers(nsurf);
+      if (preserve)
+	 ret = drm_surface_copy_swap(nsurf);
+      else
+	 ret = drm_surface_swap_buffers(nsurf);
       break;
    default:
       ret = FALSE;
@@ -196,7 +224,9 @@ drm_surface_present(struct native_surface *nsurf,
 static void
 drm_surface_wait(struct native_surface *nsurf)
 {
-   /* no-op */
+   struct drm_surface *drmsurf = drm_surface(nsurf);
+
+   resource_surface_wait(drmsurf->rsurf);
 }
 
 static void
@@ -204,6 +234,7 @@ drm_surface_destroy(struct native_surface *nsurf)
 {
    struct drm_surface *drmsurf = drm_surface(nsurf);
 
+   resource_surface_wait(drmsurf->rsurf);
    if (drmsurf->current_crtc.crtc)
          drmModeFreeCrtc(drmsurf->current_crtc.crtc);
 
@@ -236,6 +267,7 @@ drm_display_create_surface(struct native_display *ndpy,
    drmsurf->color_format = drmconf->base.color_format;
    drmsurf->width = width;
    drmsurf->height = height;
+   drmsurf->have_pageflip = TRUE;
 
    drmsurf->rsurf = resource_surface_create(drmdpy->base.screen,
          drmsurf->color_format,
diff --git a/src/gallium/state_trackers/egl/drm/native_drm.c b/src/gallium/state_trackers/egl/drm/native_drm.c
index 6932f30a6a4..cdbb304c5ee 100644
--- a/src/gallium/state_trackers/egl/drm/native_drm.c
+++ b/src/gallium/state_trackers/egl/drm/native_drm.c
@@ -124,8 +124,7 @@ drm_display_destroy(struct native_display *ndpy)
 
    drm_display_fini_modeset(&drmdpy->base);
 
-   if (drmdpy->base.screen)
-      drmdpy->base.screen->destroy(drmdpy->base.screen);
+   ndpy_uninit(ndpy);
 
    if (drmdpy->fd >= 0)
       close(drmdpy->fd);
diff --git a/src/gallium/state_trackers/egl/drm/native_drm.h b/src/gallium/state_trackers/egl/drm/native_drm.h
index 03c4fe01dc1..7da9b45f23e 100644
--- a/src/gallium/state_trackers/egl/drm/native_drm.h
+++ b/src/gallium/state_trackers/egl/drm/native_drm.h
@@ -91,6 +91,8 @@ struct drm_surface {
 
    boolean is_shown;
    struct drm_crtc current_crtc;
+
+   boolean have_pageflip;
 };
 
 struct drm_connector {
diff --git a/src/gallium/state_trackers/egl/fbdev/native_fbdev.c b/src/gallium/state_trackers/egl/fbdev/native_fbdev.c
index a1e91ba701c..4a96039ac39 100644
--- a/src/gallium/state_trackers/egl/fbdev/native_fbdev.c
+++ b/src/gallium/state_trackers/egl/fbdev/native_fbdev.c
@@ -320,7 +320,7 @@ fbdev_display_destroy(struct native_display *ndpy)
 {
    struct fbdev_display *fbdpy = fbdev_display(ndpy);
 
-   fbdpy->base.screen->destroy(fbdpy->base.screen);
+   ndpy_uninit(&fbdpy->base);
    close(fbdpy->fd);
    FREE(fbdpy);
 }
diff --git a/src/gallium/state_trackers/egl/gdi/native_gdi.c b/src/gallium/state_trackers/egl/gdi/native_gdi.c
index 3cc4aefa937..3c2475f84d8 100644
--- a/src/gallium/state_trackers/egl/gdi/native_gdi.c
+++ b/src/gallium/state_trackers/egl/gdi/native_gdi.c
@@ -363,7 +363,7 @@ gdi_display_destroy(struct native_display *ndpy)
    if (gdpy->configs)
       FREE(gdpy->configs);
 
-   gdpy->base.screen->destroy(gdpy->base.screen);
+   ndpy_uninit(ndpy);
 
    FREE(gdpy);
 }
diff --git a/src/gallium/state_trackers/egl/wayland/native_wayland.c b/src/gallium/state_trackers/egl/wayland/native_wayland.c
index d4d5f9c2ebb..068c3cd7c8e 100644
--- a/src/gallium/state_trackers/egl/wayland/native_wayland.c
+++ b/src/gallium/state_trackers/egl/wayland/native_wayland.c
@@ -41,6 +41,7 @@
 #include "radeon/drm/radeon_drm_public.h"
 
 #include <wayland-client.h>
+#include "wayland-drm-client-protocol.h"
 #include "wayland-egl-priv.h"
 
 #include <xf86drm.h>
@@ -137,8 +138,7 @@ wayland_display_destroy(struct native_display *ndpy)
    if (display->config)
       FREE(display->config);
 
-   if (display->base.screen)
-      display->base.screen->destroy(display->base.screen);
+   ndpy_uninit(ndpy);
 
    FREE(display);
 }
diff --git a/src/gallium/state_trackers/egl/x11/native_ximage.c b/src/gallium/state_trackers/egl/x11/native_ximage.c
index d4f4dd04df4..8e32c6ff0c4 100644
--- a/src/gallium/state_trackers/egl/x11/native_ximage.c
+++ b/src/gallium/state_trackers/egl/x11/native_ximage.c
@@ -476,7 +476,7 @@ ximage_display_destroy(struct native_display *ndpy)
    if (xdpy->configs)
       FREE(xdpy->configs);
 
-   xdpy->base.screen->destroy(xdpy->base.screen);
+   ndpy_uninit(ndpy);
 
    x11_screen_destroy(xdpy->xscr);
    if (xdpy->own_dpy)
diff --git a/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c b/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c
index 7cc5af89639..03aa1b1537a 100644
--- a/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c
+++ b/src/gallium/winsys/i915/drm/i915_drm_batchbuffer.c
@@ -63,7 +63,6 @@ i915_drm_batchbuffer_create(struct i915_winsys *iws)
    batch->base.size = 0;
 
    batch->base.relocs = 0;
-   batch->base.max_relocs = 300;/*INTEL_DEFAULT_RELOCS;*/
 
    batch->base.iws = iws;
 
@@ -104,8 +103,6 @@ i915_drm_batchbuffer_reloc(struct i915_winsys_batchbuffer *ibatch,
    unsigned offset;
    int ret = 0;
 
-   assert(batch->base.relocs < batch->base.max_relocs);
-
    switch (usage) {
    case I915_USAGE_SAMPLER:
       write_domain = 0;
diff --git a/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c b/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c
index 3d0c1fa6224..3bf54011d9e 100644
--- a/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c
+++ b/src/gallium/winsys/i915/sw/i915_sw_batchbuffer.c
@@ -49,7 +49,6 @@ i915_sw_batchbuffer_create(struct i915_winsys *iws)
    batch->base.size = 0;
 
    batch->base.relocs = 0;
-   batch->base.max_relocs = 300;/*INTEL_DEFAULT_RELOCS;*/
 
    batch->base.iws = iws;
 
@@ -75,8 +74,6 @@ i915_sw_batchbuffer_reloc(struct i915_winsys_batchbuffer *ibatch,
    struct i915_sw_batchbuffer *batch = i915_sw_batchbuffer(ibatch);
    int ret = 0;
 
-   assert(batch->base.relocs < batch->base.max_relocs);
-
    if (usage == I915_USAGE_SAMPLER) {
 
    } else if (usage == I915_USAGE_RENDER) {
diff --git a/src/gallium/winsys/r600/drm/evergreen_hw_context.c b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
index aa4035a302b..66398afa698 100644
--- a/src/gallium/winsys/r600/drm/evergreen_hw_context.c
+++ b/src/gallium/winsys/r600/drm/evergreen_hw_context.c
@@ -834,25 +834,25 @@ void evergreen_context_draw(struct r600_context *ctx, const struct r600_draw *dr
 	}
 
 	/* draw packet */
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_index_type;
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_instances;
 	if (draw->indices) {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3);
+	        ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing);
 		ctx->pm4[ctx->pm4_cdwords++] = draw->indices_bo_offset + r600_bo_offset(draw->indices);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices;
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator;
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 		r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], draw->indices);
 	} else {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing);
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices;
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator;
 	}
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
 
 	/* flush color buffer */
diff --git a/src/gallium/winsys/r600/drm/r600_drm.c b/src/gallium/winsys/r600/drm/r600_drm.c
index f5cd48d39c6..cd0aa318be5 100644
--- a/src/gallium/winsys/r600/drm/r600_drm.c
+++ b/src/gallium/winsys/r600/drm/r600_drm.c
@@ -45,6 +45,10 @@
 #define RADEON_INFO_CLOCK_CRYSTAL_FREQ 0x9
 #endif
 
+#ifndef RADEON_INFO_NUM_BACKENDS
+#define RADEON_INFO_NUM_BACKENDS 0xa
+#endif
+
 enum radeon_family r600_get_family(struct radeon *r600)
 {
 	return r600->family;
@@ -65,6 +69,17 @@ unsigned r600_get_clock_crystal_freq(struct radeon *radeon)
 	return radeon->clock_crystal_freq;
 }
 
+unsigned r600_get_num_backends(struct radeon *radeon)
+{
+	return radeon->num_backends;
+}
+
+unsigned r600_get_minor_version(struct radeon *radeon)
+{
+	return radeon->minor_version;
+}
+
+
 static int radeon_get_device(struct radeon *radeon)
 {
 	struct drm_radeon_info info = {};
@@ -195,6 +210,26 @@ static int radeon_get_clock_crystal_freq(struct radeon *radeon)
 	return 0;
 }
 
+
+static int radeon_get_num_backends(struct radeon *radeon)
+{
+	struct drm_radeon_info info;
+	uint32_t num_backends;
+	int r;
+
+	radeon->device = 0;
+	info.request = RADEON_INFO_NUM_BACKENDS;
+	info.value = (uintptr_t)&num_backends;
+	r = drmCommandWriteRead(radeon->fd, DRM_RADEON_INFO, &info,
+			sizeof(struct drm_radeon_info));
+	if (r)
+		return r;
+
+	radeon->num_backends = num_backends;
+	return 0;
+}
+
+
 static int radeon_init_fence(struct radeon *radeon)
 {
 	radeon->fence = 1;
@@ -211,6 +246,7 @@ static struct radeon *radeon_new(int fd, unsigned device)
 {
 	struct radeon *radeon;
 	int r;
+	drmVersionPtr version;
 
 	radeon = calloc(1, sizeof(*radeon));
 	if (radeon == NULL) {
@@ -219,13 +255,27 @@ static struct radeon *radeon_new(int fd, unsigned device)
 	radeon->fd = fd;
 	radeon->device = device;
 	radeon->refcount = 1;
-	if (fd >= 0) {
-		r = radeon_get_device(radeon);
-		if (r) {
-			fprintf(stderr, "Failed to get device id\n");
-			return radeon_decref(radeon);
-		}
+
+	version = drmGetVersion(radeon->fd);
+	if (version->version_major != 2) {
+		fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is "
+			"only compatible with 2.x.x\n", __FUNCTION__,
+			version->version_major, version->version_minor,
+			version->version_patchlevel);
+		drmFreeVersion(version);
+		exit(1);
 	}
+
+	radeon->minor_version = version->version_minor;
+
+	drmFreeVersion(version);
+
+	r = radeon_get_device(radeon);
+	if (r) {
+		fprintf(stderr, "Failed to get device id\n");
+		return radeon_decref(radeon);
+	}
+
 	radeon->family = radeon_family_from_device(radeon->device);
 	if (radeon->family == CHIP_UNKNOWN) {
 		fprintf(stderr, "Unknown chipset 0x%04X\n", radeon->device);
@@ -278,6 +328,9 @@ static struct radeon *radeon_new(int fd, unsigned device)
 	/* get the GPU counter frequency, failure is non fatal */
 	radeon_get_clock_crystal_freq(radeon);
 
+	if (radeon->minor_version >= 9)
+		radeon_get_num_backends(radeon);
+
 	radeon->bomgr = r600_bomgr_create(radeon, 1000000);
 	if (radeon->bomgr == NULL) {
 		return NULL;
diff --git a/src/gallium/winsys/r600/drm/r600_hw_context.c b/src/gallium/winsys/r600/drm/r600_hw_context.c
index f170640407d..6b3baa2c909 100644
--- a/src/gallium/winsys/r600/drm/r600_hw_context.c
+++ b/src/gallium/winsys/r600/drm/r600_hw_context.c
@@ -107,7 +107,7 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg,
 
 		/* initialize block */
 		block->start_offset = reg[i].offset;
-		block->pm4[block->pm4_ndwords++] = PKT3(reg[i].opcode, n);
+		block->pm4[block->pm4_ndwords++] = PKT3(reg[i].opcode, n, 0);
 		block->pm4[block->pm4_ndwords++] = (block->start_offset - reg[i].offset_base) >> 2;
 		block->reg = &block->pm4[block->pm4_ndwords];
 		block->pm4_ndwords += n;
@@ -119,7 +119,7 @@ int r600_context_add_block(struct r600_context *ctx, const struct r600_reg *reg,
 				block->nbo++;
 				assert(block->nbo < R600_BLOCK_MAX_BO);
 				block->pm4_bo_index[j] = block->nbo;
-				block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0);
+				block->pm4[block->pm4_ndwords++] = PKT3(PKT3_NOP, 0, 0);
 				block->pm4[block->pm4_ndwords++] = 0x00000000;
 				block->reloc[block->nbo].flush_flags = reg[i+j].flush_flags;
 				block->reloc[block->nbo].flush_mask = reg[i+j].flush_mask;
@@ -771,12 +771,12 @@ void r600_context_bo_flush(struct r600_context *ctx, unsigned flush_flags,
 		bo->last_flush &= flush_mask;
 		return;
 	}
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SURFACE_SYNC, 3, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = flush_flags;
 	ctx->pm4[ctx->pm4_cdwords++] = (bo->size + 255) >> 8;
 	ctx->pm4[ctx->pm4_cdwords++] = 0x00000000;
 	ctx->pm4[ctx->pm4_cdwords++] = 0x0000000A;
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = bo->reloc_id;
 	bo->last_flush = (bo->last_flush | flush_flags) & flush_mask;
 }
@@ -1048,25 +1048,25 @@ void r600_context_draw(struct r600_context *ctx, const struct r600_draw *draw)
 	}
 
 	/* draw packet */
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_INDEX_TYPE, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_index_type;
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NUM_INSTANCES, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_instances;
 	if (draw->indices) {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX, 3, ctx->predicate_drawing);
 		ctx->pm4[ctx->pm4_cdwords++] = draw->indices_bo_offset + r600_bo_offset(draw->indices);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices;
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator;
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, ctx->predicate_drawing);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 		r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], draw->indices);
 	} else {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_DRAW_INDEX_AUTO, 1, ctx->predicate_drawing);
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_num_indices;
 		ctx->pm4[ctx->pm4_cdwords++] = draw->vgt_draw_initiator;
 	}
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, ctx->predicate_drawing);
 	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_EVENT) | EVENT_INDEX(0);
 
 	/* flush color buffer */
@@ -1101,15 +1101,15 @@ void r600_context_flush(struct r600_context *ctx)
 	r600_context_queries_suspend(ctx);
 
 	/* emit fence */
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 0, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_PS_PARTIAL_FLUSH) | EVENT_INDEX(4);
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	ctx->pm4[ctx->pm4_cdwords++] = (1 << 29) | (0 << 24);
 	ctx->pm4[ctx->pm4_cdwords++] = ctx->radeon->fence;
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], ctx->radeon->fence_bo);
 
@@ -1282,6 +1282,7 @@ static boolean r600_query_result(struct r600_context *ctx, struct r600_query *qu
 void r600_query_begin(struct r600_context *ctx, struct r600_query *query)
 {
 	unsigned required_space;
+	int num_backends = r600_get_num_backends(ctx->radeon);
 
 	/* query request needs 6/8 dwords for begin + 6/8 dwords for end */
 	if (query->type == PIPE_QUERY_TIME_ELAPSED)
@@ -1300,21 +1301,39 @@ void r600_query_begin(struct r600_context *ctx, struct r600_query *query)
 		r600_query_result(ctx, query, TRUE);
 	}
 
+	if (query->type == PIPE_QUERY_OCCLUSION_COUNTER &&
+	    num_backends > 0 && num_backends < ctx->max_db) {
+		/* as per info on ZPASS the driver must set the unusued DB top bits */
+		u32 *results;
+		int i;
+
+		results = r600_bo_map(ctx->radeon, query->buffer, PB_USAGE_DONTBLOCK | PB_USAGE_CPU_WRITE, NULL);
+		if (results) {
+			memset(results + (query->num_results * 4), 0, ctx->max_db * 4 * 4);
+			
+			for (i = num_backends; i < ctx->max_db; i++) {
+				results[(i * 4)+1] = 0x80000000;
+				results[(i * 4)+3] = 0x80000000;
+			}
+			r600_bo_unmap(ctx->radeon, query->buffer);
+		}
+	}
+	
 	/* emit begin query */
 	if (query->type == PIPE_QUERY_TIME_ELAPSED) {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
 		ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
 		ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + r600_bo_offset(query->buffer);
 		ctx->pm4[ctx->pm4_cdwords++] = (3 << 29);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 	} else {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
 		ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
 		ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + r600_bo_offset(query->buffer);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 	}
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer);
 
@@ -1327,19 +1346,19 @@ void r600_query_end(struct r600_context *ctx, struct r600_query *query)
 {
 	/* emit begin query */
 	if (query->type == PIPE_QUERY_TIME_ELAPSED) {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE_EOP, 4, 0);
 		ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_CACHE_FLUSH_AND_INV_TS_EVENT) | EVENT_INDEX(5);
 		ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + 8 + r600_bo_offset(query->buffer);
 		ctx->pm4[ctx->pm4_cdwords++] = (3 << 29);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 	} else {
-		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2);
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_EVENT_WRITE, 2, 0);
 		ctx->pm4[ctx->pm4_cdwords++] = EVENT_TYPE(EVENT_TYPE_ZPASS_DONE) | EVENT_INDEX(1);
 		ctx->pm4[ctx->pm4_cdwords++] = query->num_results*4 + 8 + r600_bo_offset(query->buffer);
 		ctx->pm4[ctx->pm4_cdwords++] = 0;
 	}
-	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0);
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
 	ctx->pm4[ctx->pm4_cdwords++] = 0;
 	r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer);
 
@@ -1349,6 +1368,28 @@ void r600_query_end(struct r600_context *ctx, struct r600_query *query)
 	ctx->num_query_running--;
 }
 
+void r600_query_predication(struct r600_context *ctx, struct r600_query *query, int operation,
+			    int flag_wait)
+{
+	ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_SET_PREDICATION, 1, 0);
+
+	if (operation == PREDICATION_OP_CLEAR) {
+		ctx->pm4[ctx->pm4_cdwords++] = 0;
+		ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(PREDICATION_OP_CLEAR);
+	} else {
+		int results_base = query->num_results - (4 * ctx->max_db);
+
+		if (results_base < 0)
+			results_base = 0;
+
+		ctx->pm4[ctx->pm4_cdwords++] = results_base*4 + r600_bo_offset(query->buffer);
+		ctx->pm4[ctx->pm4_cdwords++] = PRED_OP(operation) | (flag_wait ? PREDICATION_HINT_WAIT : PREDICATION_HINT_NOWAIT_DRAW) | PREDICATION_DRAW_VISIBLE;
+		ctx->pm4[ctx->pm4_cdwords++] = PKT3(PKT3_NOP, 0, 0);
+		ctx->pm4[ctx->pm4_cdwords++] = 0;
+		r600_context_bo_reloc(ctx, &ctx->pm4[ctx->pm4_cdwords - 1], query->buffer);
+	}
+}
+
 struct r600_query *r600_context_query_create(struct r600_context *ctx, unsigned query_type)
 {
 	struct r600_query *query;
diff --git a/src/gallium/winsys/r600/drm/r600_priv.h b/src/gallium/winsys/r600/drm/r600_priv.h
index 2d91cd97d68..41c5ee02c38 100644
--- a/src/gallium/winsys/r600/drm/r600_priv.h
+++ b/src/gallium/winsys/r600/drm/r600_priv.h
@@ -50,6 +50,8 @@ struct radeon {
 	unsigned			*cfence;
 	struct r600_bo			*fence_bo;
 	unsigned			clock_crystal_freq;
+	unsigned			num_backends;
+	unsigned                        minor_version;
 };
 
 struct r600_reg {
diff --git a/src/gallium/winsys/r600/drm/r600d.h b/src/gallium/winsys/r600/drm/r600d.h
index 1c1ac76fe69..cb12865ff01 100644
--- a/src/gallium/winsys/r600/drm/r600d.h
+++ b/src/gallium/winsys/r600/drm/r600d.h
@@ -105,6 +105,18 @@
 		 * 5 - TS events
 		 */
 
+#define PREDICATION_OP_CLEAR 0x0
+#define PREDICATION_OP_ZPASS 0x1
+#define PREDICATION_OP_PRIMCOUNT 0x2
+
+#define PRED_OP(x) ((x) << 16)
+
+#define PREDICATION_HINT_WAIT (0 << 12)
+#define PREDICATION_HINT_NOWAIT_DRAW (1 << 12)
+
+#define PREDICATION_DRAW_NOT_VISIBLE (0 << 8)
+#define PREDICATION_DRAW_VISIBLE (1 << 8)
+
 #define PKT_TYPE_S(x)                   (((x) & 0x3) << 30)
 #define PKT_TYPE_G(x)                   (((x) >> 30) & 0x3)
 #define PKT_TYPE_C                      0x3FFFFFFF
@@ -117,8 +129,9 @@
 #define PKT3_IT_OPCODE_S(x)             (((x) & 0xFF) << 8)
 #define PKT3_IT_OPCODE_G(x)             (((x) >> 8) & 0xFF)
 #define PKT3_IT_OPCODE_C                0xFFFF00FF
+#define PKT3_PRED_S(x)               (((x) >> 0) & 0x1)
 #define PKT0(index, count) (PKT_TYPE_S(0) | PKT0_BASE_INDEX_S(index) | PKT_COUNT_S(count))
-#define PKT3(op, count) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count))
+#define PKT3(op, count, predicate) (PKT_TYPE_S(3) | PKT3_IT_OPCODE_S(op) | PKT_COUNT_S(count) | PKT3_PRED_S(predicate))
 
 /* Registers */
 #define R_0280A0_CB_COLOR0_INFO                      0x0280A0
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_common.c b/src/gallium/winsys/radeon/drm/radeon_drm_common.c
index 4676c2a1ea9..72c2ff11125 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_common.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_common.c
@@ -96,10 +96,13 @@ static void do_ioctls(struct radeon_drm_winsys *winsys)
      * we don't actually use the info for anything yet. */
 
     version = drmGetVersion(winsys->fd);
-    if (version->version_major != 2) {
+    if (version->version_major != 2 ||
+        version->version_minor < 3) {
         fprintf(stderr, "%s: DRM version is %d.%d.%d but this driver is "
-                "only compatible with 2.x.x\n", __FUNCTION__,
-                version->version_major, version->version_minor,
+                "only compatible with 2.3.x (kernel 2.6.34) and later.\n",
+                __FUNCTION__,
+                version->version_major,
+                version->version_minor,
                 version->version_patchlevel);
         drmFreeVersion(version);
         exit(1);
@@ -157,6 +160,8 @@ static void do_ioctls(struct radeon_drm_winsys *winsys)
     winsys->vram_size = gem_info.vram_size;
 
     drmFreeVersion(version);
+
+    winsys->num_cpus = sysconf(_SC_NPROCESSORS_ONLN);
 }
 
 static void radeon_winsys_destroy(struct r300_winsys_screen *rws)
@@ -190,10 +195,6 @@ static uint32_t radeon_get_value(struct r300_winsys_screen *rws,
         return ws->drm_minor;
     case R300_VID_DRM_PATCHLEVEL:
         return ws->drm_patchlevel;
-    case R300_VID_DRM_2_1_0:
-        return ws->drm_major*100 + ws->drm_minor >= 201;
-    case R300_VID_DRM_2_3_0:
-        return ws->drm_major*100 + ws->drm_minor >= 203;
     case R300_VID_DRM_2_6_0:
         return ws->drm_major*100 + ws->drm_minor >= 206;
     case R300_VID_DRM_2_8_0:
diff --git a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
index a38b01048b2..9ee800f5950 100644
--- a/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
+++ b/src/gallium/winsys/radeon/drm/radeon_drm_cs.c
@@ -384,7 +384,7 @@ void radeon_drm_cs_flush(struct r300_winsys_cs *rcs)
         for (i = 0; i < crelocs; i++)
             p_atomic_inc(&cs->csc->relocs_bo[i]->num_active_ioctls);
 
-        if (debug_get_option_thread()) {
+        if (cs->ws->num_cpus > 1 && debug_get_option_thread()) {
             cs->thread = pipe_thread_create(radeon_drm_cs_emit_ioctl, cs->csc);
             assert(cs->thread);
         } else {
diff --git a/src/gallium/winsys/radeon/drm/radeon_winsys.h b/src/gallium/winsys/radeon/drm/radeon_winsys.h
index f8a89abcfe4..9ecbb074572 100644
--- a/src/gallium/winsys/radeon/drm/radeon_winsys.h
+++ b/src/gallium/winsys/radeon/drm/radeon_winsys.h
@@ -46,6 +46,7 @@ struct radeon_drm_winsys {
     uint32_t z_pipes;       /* Z pipe count (rv530 only) */
     uint32_t gart_size;     /* GART size. */
     uint32_t vram_size;     /* VRAM size. */
+    uint32_t num_cpus;      /* Number of CPUs. */
 
     unsigned drm_major;
     unsigned drm_minor;