Merge remote branch 'origin/master' into nv50-compiler

32 files changed, 1819 insertions, 812 deletions

diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile
index 13152635a6..728bc40a5b 100644
--- a/src/gallium/drivers/r300/Makefile
+++ b/src/gallium/drivers/r300/Makefile
@@ -24,6 +24,7 @@ C_SOURCES = \
 	r300_vs.c \
 	r300_vs_draw.c \
 	r300_texture.c \
+	r300_texture_desc.c \
 	r300_tgsi_to_rc.c \
 	r300_transfer.c
 
diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript
index 552ed4e5be..bf023daaa5 100644
--- a/src/gallium/drivers/r300/SConscript
+++ b/src/gallium/drivers/r300/SConscript
@@ -34,6 +34,7 @@ r300 = env.ConvenienceLibrary(
         'r300_vs.c',
         'r300_vs_draw.c',
         'r300_texture.c',
+        'r300_texture_desc.c',
         'r300_tgsi_to_rc.c',
         'r300_transfer.c',
     ] + r300compiler) + r300compiler
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index 895efaa1c4..47ffc0cb3c 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -21,7 +21,10 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
 #include "r300_context.h"
+#include "r300_emit.h"
+#include "r300_hyperz.h"
 #include "r300_texture.h"
+#include "r300_winsys.h"
 
 #include "util/u_format.h"
 #include "util/u_pack_color.h"
@@ -81,7 +84,7 @@ static void r300_blitter_end(struct r300_context *r300)
 }
 
 static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
-					  const float* rgba)
+                                          const float* rgba)
 {
     union util_color uc;
     util_pack_color(rgba, format, &uc);
@@ -97,29 +100,29 @@ static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
 {
     struct pipe_framebuffer_state *fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
-    struct r300_surface *surf = r300_surface(fb->cbufs[0]);
-    unsigned bpp;
 
     /* Only color clear allowed, and only one colorbuffer. */
     if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
         return FALSE;
 
-    /* The colorbuffer must be point-sampled. */
-    if (surf->base.texture->nr_samples > 1)
-        return FALSE;
-
-    bpp = util_format_get_blocksizebits(surf->base.format);
+    return r300_surface(fb->cbufs[0])->cbzb_allowed;
+}
 
-    /* ZB can only work with the two pixel sizes. */
-    if (bpp != 16 && bpp != 32)
-        return FALSE;
+static uint32_t r300_depth_clear_value(enum pipe_format format,
+                                       double depth, unsigned stencil)
+{
+    switch (format) {
+        case PIPE_FORMAT_Z16_UNORM:
+        case PIPE_FORMAT_X8Z24_UNORM:
+            return util_pack_z(format, depth);
 
-    /* If the midpoint ZB offset is not aligned to 2048, it returns garbage
-     * with certain texture sizes. Macrotiling ensures the alignment. */
-    if (!r300_texture(surf->base.texture)->mip_macrotile[surf->base.level])
-        return FALSE;
+        case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+            return util_pack_z_stencil(format, depth, stencil);
 
-    return TRUE;
+        default:
+            assert(0);
+            return 0;
+    }
 }
 
 /* Clear currently bound buffers. */
@@ -169,8 +172,27 @@ 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_texture *zstex =
+            fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL;
     uint32_t width = fb->width;
     uint32_t height = fb->height;
+    boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
+    uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
+
+    /* Enable fast Z clear.
+     * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
+    if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && has_hyperz) {
+        hyperz_dcv = hyperz->zb_depthclearvalue =
+            r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
+
+        r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG);
+        if (zstex->zmask_mem[fb->zsbuf->level]) {
+            r300->zmask_clear.dirty = TRUE;
+            buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
+        }
+        if (zstex->hiz_mem[fb->zsbuf->level])
+            r300->hiz_clear.dirty = TRUE;
+    }
 
     /* Enable CBZB clear. */
     if (r300_cbzb_clear_allowed(r300, buffers)) {
@@ -187,20 +209,61 @@ static void r300_clear(struct pipe_context* pipe,
     }
 
     /* Clear. */
-    r300_blitter_begin(r300, R300_CLEAR);
-    util_blitter_clear(r300->blitter,
-                       width,
-                       height,
-                       fb->nr_cbufs,
-                       buffers, rgba, depth, stencil);
-    r300_blitter_end(r300);
+    if (buffers) {
+        /* Clear using the blitter. */
+        r300_blitter_begin(r300, R300_CLEAR);
+        util_blitter_clear(r300->blitter,
+                           width,
+                           height,
+                           fb->nr_cbufs,
+                           buffers, rgba, depth, stencil);
+        r300_blitter_end(r300);
+    } else if (r300->zmask_clear.dirty) {
+        /* Just clear zmask and hiz now, this does not use a standard draw
+         * procedure. */
+        unsigned dwords;
+
+        /* Calculate zmask_clear and hiz_clear atom sizes. */
+        r300_update_hyperz_state(r300);
+        dwords = r300->zmask_clear.size +
+                 (r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
+                 r300_get_num_cs_end_dwords(r300);
+
+        /* Reserve CS space. */
+        if (dwords > (r300->cs->ndw - r300->cs->cdw)) {
+            r300->context.flush(&r300->context, 0, NULL);
+        }
+
+        /* Emit clear packets. */
+        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);
+            r300->hiz_clear.dirty = FALSE;
+        }
+    } else {
+        assert(0);
+    }
 
     /* Disable CBZB clear. */
     if (r300->cbzb_clear) {
         r300->cbzb_clear = FALSE;
+        hyperz->zb_depthclearvalue = hyperz_dcv;
         r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
     }
 
+    /* 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 (zstex &&
+        (zstex->zmask_in_use[fb->zsbuf->level] ||
+         zstex->hiz_in_use[fb->zsbuf->level])) {
+        r300->hyperz_state.dirty = TRUE;
+    }
+
     /* XXX this flush "fixes" a hardlock in the cubestorm xscreensaver */
     if (r300->flush_counter == 0)
         pipe->flush(pipe, 0, NULL);
@@ -238,6 +301,33 @@ static void r300_clear_depth_stencil(struct pipe_context *pipe,
     r300_blitter_end(r300);
 }
 
+/* Flush a depth stencil buffer. */
+void r300_flush_depth_stencil(struct pipe_context *pipe,
+                              struct pipe_resource *dst,
+                              struct pipe_subresource subdst,
+                              unsigned zslice)
+{
+    struct r300_context *r300 = r300_context(pipe);
+    struct pipe_surface *dstsurf;
+    struct r300_texture *tex = r300_texture(dst);
+
+    if (!tex->zmask_mem[subdst.level])
+        return;
+    if (!tex->zmask_in_use[subdst.level])
+        return;
+
+    dstsurf = pipe->screen->get_tex_surface(pipe->screen, dst,
+                                            subdst.face, subdst.level, zslice,
+                                            PIPE_BIND_DEPTH_STENCIL);
+    r300->z_decomp_rd = TRUE;
+    r300_blitter_begin(r300, R300_CLEAR_SURFACE);
+    util_blitter_flush_depth_stencil(r300->blitter, dstsurf);
+    r300_blitter_end(r300);
+    r300->z_decomp_rd = FALSE;
+
+    tex->zmask_in_use[subdst.level] = FALSE;
+}
+
 /* Copy a block of pixels from one surface to another using HW. */
 static void r300_hw_copy_region(struct pipe_context* pipe,
                                 struct pipe_resource *dst,
@@ -269,7 +359,7 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
 {
     enum pipe_format old_format = dst->format;
     enum pipe_format new_format = old_format;
-
+    boolean is_depth;
     if (!pipe->screen->is_format_supported(pipe->screen,
                                            old_format, src->target,
                                            src->nr_samples,
@@ -296,6 +386,10 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
         }
     }
 
+    is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
+    if (is_depth) {
+        r300_flush_depth_stencil(pipe, src, subsrc, srcz);
+    }
     if (old_format != new_format) {
         dst->format = new_format;
         src->format = new_format;
diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c
index 21f3b9d261..48c2409211 100644
--- a/src/gallium/drivers/r300/r300_chipset.c
+++ b/src/gallium/drivers/r300/r300_chipset.c
@@ -36,7 +36,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
     caps->num_vert_fpus = 2;
     caps->num_tex_units = 16;
     caps->has_tcl = debug_get_bool_option("RADEON_NO_TCL", FALSE) ? FALSE : TRUE;
-    caps->has_hiz = TRUE;
+    caps->hiz_ram = 0;
     caps->is_r400 = FALSE;
     caps->is_r500 = FALSE;
     caps->high_second_pipe = FALSE;
@@ -49,6 +49,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R300;
             caps->high_second_pipe = TRUE;
             caps->num_vert_fpus = 4;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x4145:
@@ -61,6 +63,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R300;
             caps->high_second_pipe = TRUE;
             caps->num_vert_fpus = 4;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x4150:
@@ -77,8 +81,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x4E54:
         case 0x4E56:
             caps->family = CHIP_FAMILY_RV350;
-            caps->has_hiz = FALSE;
             caps->high_second_pipe = TRUE;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x4148:
@@ -91,12 +95,16 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R350;
             caps->high_second_pipe = TRUE;
             caps->num_vert_fpus = 4;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x4E4A:
             caps->family = CHIP_FAMILY_R360;
             caps->high_second_pipe = TRUE;
             caps->num_vert_fpus = 4;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x5460:
@@ -108,8 +116,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x5B64:
         case 0x5B65:
             caps->family = CHIP_FAMILY_RV370;
-            caps->has_hiz = FALSE;
             caps->high_second_pipe = TRUE;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x3150:
@@ -120,6 +128,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
         case 0x3E54:
             caps->family = CHIP_FAMILY_RV380;
             caps->high_second_pipe = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x4A48:
@@ -135,6 +145,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R420;
             caps->num_vert_fpus = 6;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x5548:
@@ -149,6 +161,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R423;
             caps->num_vert_fpus = 6;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x554C:
@@ -161,6 +175,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R430;
             caps->num_vert_fpus = 6;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x5D4C:
@@ -172,6 +188,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R480;
             caps->num_vert_fpus = 6;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x4B48:
@@ -182,6 +200,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R481;
             caps->num_vert_fpus = 6;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x5E4C:
@@ -199,34 +219,36 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RV410;
             caps->num_vert_fpus = 6;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x5954:
         case 0x5955:
             caps->family = CHIP_FAMILY_RS480;
-            caps->has_hiz = FALSE;
             caps->has_tcl = FALSE;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x5974:
         case 0x5975:
             caps->family = CHIP_FAMILY_RS482;
-            caps->has_hiz = FALSE;
             caps->has_tcl = FALSE;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x5A41:
         case 0x5A42:
             caps->family = CHIP_FAMILY_RS400;
-            caps->has_hiz = FALSE;
             caps->has_tcl = FALSE;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x5A61:
         case 0x5A62:
             caps->family = CHIP_FAMILY_RC410;
-            caps->has_hiz = FALSE;
             caps->has_tcl = FALSE;
+            caps->zmask_ram = RV3xx_ZMASK_SIZE;
             break;
 
         case 0x791E:
@@ -234,6 +256,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RS690;
             caps->has_tcl = FALSE;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x793F:
@@ -242,6 +266,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RS600;
             caps->has_tcl = FALSE;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x796C:
@@ -251,6 +277,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RS740;
             caps->has_tcl = FALSE;
             caps->is_r400 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x7100:
@@ -270,6 +298,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R520;
             caps->num_vert_fpus = 8;
             caps->is_r500 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x7140:
@@ -313,6 +343,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RV515;
             caps->num_vert_fpus = 2;
             caps->is_r500 = TRUE;
+            caps->hiz_ram = R300_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x71C0:
@@ -334,6 +366,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RV530;
             caps->num_vert_fpus = 5;
             caps->is_r500 = TRUE;
+            /*caps->hiz_ram = RV530_HIZ_LIMIT;*/
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x7240:
@@ -354,12 +388,16 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_R580;
             caps->num_vert_fpus = 8;
             caps->is_r500 = TRUE;
+            caps->hiz_ram = RV530_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x7280:
             caps->family = CHIP_FAMILY_RV570;
             caps->num_vert_fpus = 8;
             caps->is_r500 = TRUE;
+            caps->hiz_ram = RV530_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         case 0x7281:
@@ -376,6 +414,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
             caps->family = CHIP_FAMILY_RV560;
             caps->num_vert_fpus = 8;
             caps->is_r500 = TRUE;
+            caps->hiz_ram = RV530_HIZ_LIMIT;
+            caps->zmask_ram = PIPE_ZMASK_SIZE;
             break;
 
         default:
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index 65750f54e7..e7ca642b4f 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -25,6 +25,14 @@
 
 #include "pipe/p_compiler.h"
 
+/* these are sizes in dwords */
+#define R300_HIZ_LIMIT 10240
+#define RV530_HIZ_LIMIT 15360
+
+/* rv3xx have only one pipe */
+#define PIPE_ZMASK_SIZE 4096
+#define RV3xx_ZMASK_SIZE 5120
+
 /* Structure containing all the possible information about a specific Radeon
  * in the R3xx, R4xx, and R5xx families. */
 struct r300_capabilities {
@@ -42,8 +50,10 @@ struct r300_capabilities {
     unsigned num_tex_units;
     /* Whether or not TCL is physically present */
     boolean has_tcl;
-    /* Some chipsets do not have HiZ RAM. */
-    boolean has_hiz;
+    /* Some chipsets do not have HiZ RAM - other have varying amounts . */
+    int hiz_ram;
+    /*  some chipsets have zmask ram per pipe some don't */
+    int zmask_ram;
     /* Whether or not this is RV350 or newer, including all r400 and r500
      * chipsets. The differences compared to the oldest r300 chips are:
      * - Blend LTE/GTE thresholds
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index df90359058..e8b6c4f7af 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -30,6 +30,7 @@
 #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"
@@ -114,6 +115,10 @@ static void r300_destroy_context(struct pipe_context* context)
     u_upload_destroy(r300->upload_vb);
     u_upload_destroy(r300->upload_ib);
 
+    /* setup hyper-z mm */
+    if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+        r300_hyperz_destroy_mm(r300);
+
     translate_cache_destroy(r300->tran.translate_cache);
 
     r300_release_referenced_objects(r300);
@@ -166,6 +171,9 @@ static void r300_setup_atoms(struct r300_context* r300)
     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 has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
+    boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
 
     /* Create the actual atom list.
      *
@@ -188,8 +196,8 @@ static void r300_setup_atoms(struct r300_context* r300)
     R300_INIT_ATOM(gpu_flush, 9);
     R300_INIT_ATOM(aa_state, 4);
     R300_INIT_ATOM(fb_state, 0);
+    R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
     /* ZB (unpipelined), SC. */
-    R300_INIT_ATOM(hyperz_state, 6);
     R300_INIT_ATOM(ztop_state, 2);
     /* ZB, FG. */
     R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
@@ -220,6 +228,13 @@ static void r300_setup_atoms(struct r300_context* r300)
     /* TX. */
     R300_INIT_ATOM(texture_cache_inval, 2);
     R300_INIT_ATOM(textures_state, 0);
+    if (has_hyperz) {
+        /* HiZ Clear */
+        if (has_hiz_ram)
+            R300_INIT_ATOM(hiz_clear, 0);
+        /* zmask clear */
+        R300_INIT_ATOM(zmask_clear, 0);
+    }
     /* ZB (unpipelined), SU. */
     R300_INIT_ATOM(query_start, 4);
 
@@ -282,8 +297,7 @@ static void r300_init_states(struct pipe_context *pipe)
             (struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
     struct r300_invariant_state *invariant =
             (struct r300_invariant_state*)r300->invariant_state.state;
-    struct r300_hyperz_state *hyperz =
-            (struct r300_hyperz_state*)r300->hyperz_state.state;
+
     CB_LOCALS;
 
     pipe->set_blend_color(pipe, &bc);
@@ -351,10 +365,20 @@ static void r300_init_states(struct pipe_context *pipe)
 
     /* Initialize the hyperz state. */
     {
-        BEGIN_CB(&hyperz->cb_begin, r300->hyperz_state.size);
+        struct r300_hyperz_state *hyperz =
+            (struct r300_hyperz_state*)r300->hyperz_state.state;
+        BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
+        OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
+                   R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
         OUT_CB_REG(R300_ZB_BW_CNTL, 0);
         OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
         OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
+
+        if (r300->screen->caps.is_r500 ||
+            (r300->screen->caps.is_rv350 &&
+             r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
+            OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
+        }
         END_CB;
     }
 }
@@ -415,6 +439,10 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
 
     rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
 
+    /* setup hyper-z mm */
+    if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+        r300_hyperz_init_mm(r300);
+
     r300->upload_ib = u_upload_create(&r300->context,
 				      32 * 1024, 16,
 				      PIPE_BIND_INDEX_BUFFER);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index b9c96d5bdd..6fa7f470f9 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -106,13 +106,19 @@ 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;
+    uint32_t zb_zcache_ctlstat;     /* R300_ZB_CACHE_CNTL */
     uint32_t cb_begin;
     uint32_t zb_bw_cntl;            /* R300_ZB_BW_CNTL */
     uint32_t cb_reg1;
     uint32_t zb_depthclearvalue;    /* R300_ZB_DEPTHCLEARVALUE */
     uint32_t cb_reg2;
     uint32_t sc_hyperz;             /* R300_SC_HYPERZ */
+    uint32_t cb_reg3;
+    uint32_t gb_z_peq_config;       /* R300_GB_Z_PEQ_CONFIG: 0x4028 */
 };
 
 struct r300_gpu_flush {
@@ -318,29 +324,39 @@ struct r300_surface {
     uint32_t cbzb_midpoint_offset;  /* DEPTHOFFSET. */
     uint32_t cbzb_pitch;            /* DEPTHPITCH. */
     uint32_t cbzb_format;           /* ZB_FORMAT. */
+
+    /* Whether the CBZB clear is allowed on the surface. */
+    boolean cbzb_allowed;
+
 };
 
-struct r300_texture {
-    /* Parent class */
+struct r300_texture_desc {
+    /* Parent class. */
     struct u_resource b;
 
-    enum r300_buffer_domain domain;
+    /* Buffer tiling.
+     * Macrotiling is specified per-level because small mipmaps cannot
+     * be macrotiled. */
+    enum r300_buffer_tiling microtile;
+    enum r300_buffer_tiling macrotile[R300_MAX_TEXTURE_LEVELS];
 
     /* Offsets into the buffer. */
-    unsigned offset[R300_MAX_TEXTURE_LEVELS];
+    unsigned offset_in_bytes[R300_MAX_TEXTURE_LEVELS];
 
-    /* A pitch for each mip-level */
-    unsigned pitch[R300_MAX_TEXTURE_LEVELS];
+    /* Strides for each mip-level. */
+    unsigned stride_in_pixels[R300_MAX_TEXTURE_LEVELS];
+    unsigned stride_in_bytes[R300_MAX_TEXTURE_LEVELS];
 
-    /* A pitch multiplied by blockwidth as hardware wants
-     * the number of pixels instead of the number of blocks. */
-    unsigned hwpitch[R300_MAX_TEXTURE_LEVELS];
+    /* Size of one zslice or face or 2D image based on the texture target. */
+    unsigned layer_size_in_bytes[R300_MAX_TEXTURE_LEVELS];
 
-    /* Size of one zslice or face based on the texture target */
-    unsigned layer_size[R300_MAX_TEXTURE_LEVELS];
+    /* Total size of this texture, in bytes,
+     * derived from the texture properties. */
+    unsigned size_in_bytes;
 
-    /* Whether the mipmap level is macrotiled. */
-    enum r300_buffer_tiling mip_macrotile[R300_MAX_TEXTURE_LEVELS];
+    /* Total size of the buffer backing this texture, in bytes.
+     * It must be >= size. */
+    unsigned buffer_size_in_bytes;
 
     /**
      * If non-zero, override the natural texture layout with
@@ -350,16 +366,24 @@ struct r300_texture {
      *
      * \sa r300_texture_get_stride
      */
-    unsigned stride_override;
+    unsigned stride_in_bytes_override;
 
-    /* Total size of this texture, in bytes. */
-    unsigned size;
+    /* Whether this texture has non-power-of-two dimensions.
+     * It can be either a regular texture or a rectangle one. */
+    boolean is_npot;
 
-    /* Whether this texture has non-power-of-two dimensions
-     * or a user-specified pitch.
-     * It can be either a regular texture or a rectangle one.
-     */
-    boolean uses_pitch;
+    /* This flag says that hardware must use the stride for addressing
+     * instead of the width. */
+    boolean uses_stride_addressing;
+
+    /* Whether CBZB fast color clear is allowed on the miplevel. */
+    boolean cbzb_allowed[R300_MAX_TEXTURE_LEVELS];
+};
+
+struct r300_texture {
+    struct r300_texture_desc desc;
+
+    enum r300_buffer_domain domain;
 
     /* Pipe buffer backing this texture. */
     struct r300_winsys_buffer *buffer;
@@ -370,8 +394,11 @@ struct r300_texture {
     /* All bits should be filled in. */
     struct r300_texture_fb_state fb_state;
 
-    /* Buffer tiling */
-    enum r300_buffer_tiling microtile, macrotile;
+    /* hyper-z memory allocs */
+    struct mem_block *hiz_mem[R300_MAX_TEXTURE_LEVELS];
+    struct mem_block *zmask_mem[R300_MAX_TEXTURE_LEVELS];
+    boolean zmask_in_use[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.*/
@@ -498,6 +525,10 @@ struct r300_context {
     struct r300_atom texture_cache_inval;
     /* GPU flush. */
     struct r300_atom gpu_flush;
+    /* HiZ clear */
+    struct r300_atom hiz_clear;
+    /* zmask clear */
+    struct r300_atom zmask_clear;
 
     /* Invariant state. This must be emitted to get the engine started. */
     struct r300_atom invariant_state;
@@ -510,6 +541,8 @@ struct r300_context {
     struct r300_vertex_element_state *velems;
     bool any_user_vbs;
 
+    struct pipe_index_buffer index_buffer;
+
     /* Vertex info for Draw. */
     struct vertex_info vertex_info;
 
@@ -533,8 +566,16 @@ struct r300_context {
     boolean two_sided_color;
     /* Incompatible vertex buffer layout? (misaligned stride or buffer_offset) */
     boolean incompatible_vb_layout;
-
+#define R300_Z_COMPRESS_44 1
+#define RV350_Z_COMPRESS_88 2
+    int z_compression;
     boolean cbzb_clear;
+    boolean z_decomp_rd;
+
+    /* two mem block managers for hiz/zmask ram space */
+    struct mem_block *hiz_mm;
+    struct mem_block *zmask_mm;
+
     /* upload managers */
     struct u_upload_mgr *upload_vb;
     struct u_upload_mgr *upload_ib;
@@ -586,6 +627,12 @@ void r300_init_render_functions(struct r300_context *r300);
 void r300_init_state_functions(struct r300_context* r300);
 void r300_init_resource_functions(struct r300_context* r300);
 
+/* r300_blit.c */
+void r300_flush_depth_stencil(struct pipe_context *pipe,
+                              struct pipe_resource *dst,
+                              struct pipe_subresource subdst,
+                              unsigned zslice);
+
 /* r300_query.c */
 void r300_resume_query(struct r300_context *r300,
                        struct r300_query *query);
@@ -605,7 +652,8 @@ void r300_plug_in_stencil_ref_fallback(struct r300_context *r300);
 /* r300_state.c */
 enum r300_fb_state_change {
     R300_CHANGED_FB_STATE = 0,
-    R300_CHANGED_CBZB_FLAG
+    R300_CHANGED_CBZB_FLAG,
+    R300_CHANGED_ZCLEAR_FLAG
 };
 
 void r300_mark_fb_state_dirty(struct r300_context *r300,
diff --git a/src/gallium/drivers/r300/r300_cs.h b/src/gallium/drivers/r300/r300_cs.h
index 3beb625d43..c194d6a1b0 100644
--- a/src/gallium/drivers/r300/r300_cs.h
+++ b/src/gallium/drivers/r300/r300_cs.h
@@ -136,8 +136,8 @@
 
 #define WRITE_CS_TABLE(values, count) do { \
     CS_DEBUG(assert(cs_count == 0);) \
-    memcpy(cs_copy->ptr + cs_copy->cdw, values, count * 4); \
-    cs_copy->cdw += count; \
+    memcpy(cs_copy->ptr + cs_copy->cdw, (values), (count) * 4); \
+    cs_copy->cdw += (count); \
 } while (0)
 
 #endif /* R300_CS_H */
diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c
index 31d4e14681..c3e157e99a 100644
--- a/src/gallium/drivers/r300/r300_debug.c
+++ b/src/gallium/drivers/r300/r300_debug.c
@@ -38,11 +38,13 @@ static const struct debug_named_value debug_options[] = {
     { "fall", DBG_FALL, "Fallbacks (for debugging)" },
     { "rs", DBG_RS, "Rasterizer (for debugging)" },
     { "fb", DBG_FB, "Framebuffer (for debugging)" },
+    { "cbzb", DBG_CBZB, "Fast color clear info (for debugging)" },
     { "fakeocc", DBG_FAKE_OCC, "Use fake occlusion queries (for debugging)" },
     { "anisohq", DBG_ANISOHQ, "High quality anisotropic filtering (for benchmarking)" },
     { "notiling", DBG_NO_TILING, "Disable tiling (for benchmarking)" },
     { "noimmd", DBG_NO_IMMD, "Disable immediate mode (for benchmarking)" },
     { "stats", DBG_STATS, "Gather statistics" },
+    { "hyperz", DBG_HYPERZ, "HyperZ (for debugging)" },
 
     /* must be last */
     DEBUG_NAMED_VALUE_END
diff --git a/src/gallium/drivers/r300/r300_defines.h b/src/gallium/drivers/r300/r300_defines.h
index d510d80a7b..896aeef395 100644
--- a/src/gallium/drivers/r300/r300_defines.h
+++ b/src/gallium/drivers/r300/r300_defines.h
@@ -36,7 +36,10 @@
 enum r300_buffer_tiling {
     R300_BUFFER_LINEAR = 0,
     R300_BUFFER_TILED,
-    R300_BUFFER_SQUARETILED
+    R300_BUFFER_SQUARETILED,
+
+    R300_BUFFER_UNKNOWN,
+    R300_BUFFER_SELECT_LAYOUT = R300_BUFFER_UNKNOWN
 };
 
 enum r300_buffer_domain { /* bitfield */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 36a26a7871..d0fd45349e 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -25,6 +25,7 @@
 
 #include "util/u_format.h"
 #include "util/u_math.h"
+#include "util/u_mm.h"
 #include "util/u_simple_list.h"
 
 #include "r300_context.h"
@@ -329,6 +330,7 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
     struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
     struct r300_surface* surf;
     unsigned i;
+    boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
     CS_LOCALS(r300);
 
     BEGIN_CS(size);
@@ -364,6 +366,10 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
         OUT_CS_RELOC(surf->buffer, surf->cbzb_pitch, 0, surf->domain);
+
+        DBG(r300, DBG_CBZB,
+            "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
+            surf->cbzb_pitch);
     }
     /* Set up a zbuffer. */
     else if (fb->zsbuf) {
@@ -377,15 +383,32 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
         OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
         OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain);
 
-        /* HiZ RAM. */
-        if (r300->screen->caps.has_hiz) {
-            OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
-            OUT_CS_REG(R300_ZB_HIZ_PITCH, 0);
+        if (has_hyperz) {
+            uint32_t surf_pitch;
+            struct r300_texture *tex;
+            int level = surf->base.level;
+            tex = r300_texture(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);
+                }
+            }
+            /* Z Mask RAM. (compressed zbuffer) */
+            if (tex->zmask_mem[level]) {
+                OUT_CS_REG(R300_ZB_ZMASK_OFFSET, tex->zmask_mem[level]->ofs << 2);
+                OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf_pitch);
+            } else {
+                OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
+                OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0);
+            }
         }
-
-        /* Z Mask RAM. (compressed zbuffer) */
-        OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
-        OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0);
     }
 
     END_CS;
@@ -394,8 +417,12 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
 void r300_emit_hyperz_state(struct r300_context *r300,
                             unsigned size, void *state)
 {
+    struct r300_hyperz_state *z = state;
     CS_LOCALS(r300);
-    WRITE_CS_TABLE(state, size);
+    if (z->flush)
+        WRITE_CS_TABLE(&z->cb_flush_begin, size);
+    else
+        WRITE_CS_TABLE(&z->cb_begin, size - 2);
 }
 
 void r300_emit_hyperz_end(struct r300_context *r300)
@@ -403,9 +430,11 @@ void r300_emit_hyperz_end(struct r300_context *r300)
     struct r300_hyperz_state z =
             *(struct r300_hyperz_state*)r300->hyperz_state.state;
 
+    z.flush = 1;
     z.zb_bw_cntl = 0;
     z.zb_depthclearvalue = 0;
     z.sc_hyperz = R300_SC_HYPERZ_ADJ_2;
+    z.gb_z_peq_config = 0;
 
     r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z);
 }
@@ -907,6 +936,22 @@ void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
             OUT_CS_TABLE(data, 4);
         }
     }
+
+    /* Emit flow control instructions. */
+    if (code->num_fc_ops) {
+
+        OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
+        if (r300screen->caps.is_r500) {
+            OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, code->num_fc_ops * 2);
+            OUT_CS_TABLE(code->fc_op_addrs.r500, code->num_fc_ops * 2);
+        } else {
+            OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, code->num_fc_ops);
+            OUT_CS_TABLE(code->fc_op_addrs.r300, code->num_fc_ops);
+        }
+        OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, code->num_fc_ops);
+        OUT_CS_TABLE(code->fc_loop_index, code->num_fc_ops);
+    }
+
     END_CS;
 }
 
@@ -943,6 +988,111 @@ 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;
+}
+
+static void r300_emit_zmask_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_ZMASK, 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_texture* tex;
+    int i;
+
+    tex = r300_texture(fb->zsbuf->texture);
+
+    offset = tex->hiz_mem[fb->zsbuf->level]->ofs;
+    stride = tex->desc.stride_in_pixels[fb->zsbuf->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;
+
+    /* Mark the current zbuffer's hiz ram as in use. */
+    tex->hiz_in_use[fb->zsbuf->level] = TRUE;
+}
+
+void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
+{
+    struct pipe_framebuffer_state *fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
+    struct r300_screen* r300screen = r300->screen;
+    uint32_t stride, offset = 0;
+    struct r300_texture* tex;
+    uint32_t i, height;
+    int mult, offset_shift;
+
+    tex = r300_texture(fb->zsbuf->texture);
+    stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+
+    offset = tex->zmask_mem[fb->zsbuf->level]->ofs;
+
+    if (r300->z_compression == RV350_Z_COMPRESS_88)
+        mult = 8;
+    else
+        mult = 4;
+
+    height = ALIGN_DIVUP(fb->zsbuf->height, mult);
+
+    offset_shift = 4;
+    offset_shift += (r300screen->caps.num_frag_pipes / 2);
+    stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes);
+
+    /* okay have width in pixels - divide by block width */
+    stride = ALIGN_DIVUP(stride, mult);
+    /* have width in blocks - divide by number of fragment pipes screen width */
+    /* 16 blocks per dword */
+    stride = ALIGN_DIVUP(stride, 16);
+
+    for (i = 0; i < height; i++) {
+        offset = i * stride;
+        offset <<= offset_shift;
+        r300_emit_zmask_line_clear(r300, offset, stride, 0x0);//0xffffffff);
+    }
+
+    /* Mark the current zbuffer's zmask as in use. */
+    tex->zmask_in_use[fb->zsbuf->level] = TRUE;
+}
+
 void r300_emit_ztop_state(struct r300_context* r300,
                           unsigned size, void* state)
 {
@@ -1062,6 +1212,17 @@ unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
     return dwords;
 }
 
+unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
+{
+    unsigned dwords = 0;
+
+    /* Emitted in flush. */
+    dwords += 26; /* emit_query_end */
+    dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
+
+    return dwords;
+}
+
 /* Emit all dirty state. */
 void r300_emit_dirty_state(struct r300_context* r300)
 {
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 5d05039669..bae2525634 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -112,7 +112,11 @@ void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, voi
 void r300_emit_invariant_state(struct r300_context *r300,
                                unsigned size, void *state);
 
+void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state);
+void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state);
+
 unsigned r300_get_num_dirty_dwords(struct r300_context *r300);
+unsigned r300_get_num_cs_end_dwords(struct r300_context *r300);
 
 /* Emit all dirty state. */
 void r300_emit_dirty_state(struct r300_context* r300);
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index ae7b5759e7..fe182b6615 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -43,14 +43,6 @@ static void r300_flush(struct pipe_context* pipe,
     u_upload_flush(r300->upload_vb);
     u_upload_flush(r300->upload_ib);
 
-    /* We probably need to flush Draw, but we may have been called from
-     * within Draw. This feels kludgy, but it might be the best thing.
-     *
-     * Of course, the best thing is to kill Draw with fire. :3 */
-    if (r300->draw && !r300->draw->flushing) {
-        draw_flush(r300->draw);
-    }
-
     if (r300->dirty_hw) {
         r300_emit_hyperz_end(r300);
         r300_emit_query_end(r300);
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index b145ded639..2a0c30620a 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -72,6 +72,11 @@ void r300_shader_read_fs_inputs(struct tgsi_shader_info* info,
                 fs_inputs->wpos = i;
                 break;
 
+            case TGSI_SEMANTIC_FACE:
+                assert(index == 0);
+                fs_inputs->face = i;
+                break;
+
             default:
                 fprintf(stderr, "r300: FP: Unknown input semantic: %i\n",
                         info->input_semantic_name[i]);
@@ -120,6 +125,9 @@ static void allocate_hardware_inputs(
             allocate(mydata, inputs->color[i], reg++);
         }
     }
+    if (inputs->face != ATTR_UNUSED) {
+        allocate(mydata, inputs->face, reg++);
+    }
     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
         if (inputs->generic[i] != ATTR_UNUSED) {
             allocate(mydata, inputs->generic[i], reg++);
@@ -173,7 +181,7 @@ static void get_external_state(
             t = (struct r300_texture*)texstate->sampler_views[i]->base.texture;
 
             /* XXX this should probably take into account STR, not just S. */
-            if (t->uses_pitch) {
+            if (t->desc.is_npot) {
                 switch (s->state.wrap_s) {
                     case PIPE_TEX_WRAP_REPEAT:
                         state->unit[i].wrap_mode = RC_WRAP_REPEAT;
@@ -248,13 +256,18 @@ static void r300_emit_fs_code_to_buffer(
 
         shader->cb_code_size = 19 +
                                ((code->inst_end + 1) * 6) +
-                               imm_count * 7;
+                               imm_count * 7 +
+			       code->int_constant_count * 2;
 
         NEW_CB(shader->cb_code, shader->cb_code_size);
         OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
         OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx);
         OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl);
-        OUT_CB_REG(R500_US_CODE_RANGE,
+        for(i = 0; i < code->int_constant_count; i++){
+		OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4),
+						code->int_constants[i]);
+	}
+	OUT_CB_REG(R500_US_CODE_RANGE,
                    R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
         OUT_CB_REG(R500_US_CODE_OFFSET, 0);
         OUT_CB_REG(R500_US_CODE_ADDR,
@@ -355,13 +368,14 @@ static void r300_translate_fragment_shader(
 {
     struct r300_fragment_program_compiler compiler;
     struct tgsi_to_rc ttr;
-    int wpos;
+    int wpos, face;
     unsigned i;
 
     tgsi_scan_shader(tokens, &shader->info);
     r300_shader_read_fs_inputs(&shader->info, &shader->inputs);
 
     wpos = shader->inputs.wpos;
+    face = shader->inputs.face;
 
     /* Setup the compiler. */
     memset(&compiler, 0, sizeof(compiler));
@@ -378,7 +392,7 @@ static void r300_translate_fragment_shader(
     find_output_registers(&compiler, shader);
 
     if (compiler.Base.Debug) {
-        debug_printf("r300: Initial fragment program\n");
+        DBG(r300, DBG_FP, "r300: Initial fragment program\n");
         tgsi_dump(tokens, 0);
     }
 
@@ -401,6 +415,10 @@ static void r300_translate_fragment_shader(
         rc_transform_fragment_wpos(&compiler.Base, wpos, wpos, TRUE);
     }
 
+    if (face != ATTR_UNUSED) {
+        rc_transform_fragment_face(&compiler.Base, face);
+    }
+
     /* Invoke the compiler */
     r3xx_compile_fragment_program(&compiler);
 
@@ -413,7 +431,7 @@ static void r300_translate_fragment_shader(
     }
 
     if (compiler.Base.Error) {
-        fprintf(stderr, "r300 FP: Compiler Error:\n%sUsing a dummy shader"
+        DBG(r300, DBG_FP, "r300 FP: Compiler Error:\n%sUsing a dummy shader"
                 " instead.\nIf there's an 'unknown opcode' message, please"
                 " file a bug report and attach this log.\n", compiler.Base.ErrorMsg);
 
diff --git a/src/gallium/drivers/r300/r300_hyperz.c b/src/gallium/drivers/r300/r300_hyperz.c
index e952895601..811b5646e1 100644
--- a/src/gallium/drivers/r300/r300_hyperz.c
+++ b/src/gallium/drivers/r300/r300_hyperz.c
@@ -25,21 +25,170 @@
 #include "r300_hyperz.h"
 #include "r300_reg.h"
 #include "r300_fs.h"
+#include "r300_winsys.h"
 
+#include "util/u_format.h"
+#include "util/u_mm.h"
+
+/*
+  HiZ rules - taken from various docs 
+   1. HiZ only works on depth values
+   2. Cannot HiZ if stencil fail or zfail is !KEEP
+   3. on R300/400, HiZ is disabled if depth test is EQUAL
+   4. comparison changes without clears usually mean disabling HiZ
+*/
 /*****************************************************************************/
 /* The HyperZ setup                                                          */
 /*****************************************************************************/
 
+static bool r300_get_sc_hz_max(struct r300_context *r300)
+{
+    struct r300_dsa_state *dsa_state = r300->dsa_state.state;
+    int func = dsa_state->z_stencil_control & 0x7;
+    int ret = R300_SC_HYPERZ_MIN;
+
+    if (func >= 4 && func <= 7)
+       ret = R300_SC_HYPERZ_MAX;
+    return ret;
+}
+
+static bool r300_zfunc_same_direction(int func1, int func2)
+{
+    /* func1 is less/lessthan */
+    if (func1 == 1 || func1 == 2)
+        if (func2 == 3 || func2 == 4 || func2 == 5)
+            return FALSE;
+
+    if (func2 == 1 || func2 == 2)
+        if (func1 == 4 || func1 == 5)
+            return FALSE;
+    return TRUE;
+}
+    
+static int r300_get_hiz_min(struct r300_context *r300)
+{
+    struct r300_dsa_state *dsa_state = r300->dsa_state.state;
+    int func = dsa_state->z_stencil_control & 0x7;
+    int ret = R300_HIZ_MIN;
+
+    if (func == 1 || func == 2)
+       ret = R300_HIZ_MAX;
+    return ret;
+}
+
+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;
+}
+
+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;
+
+    /* shader writes depth - no HiZ */
+    if (r300_fragment_shader_writes_depth(r300_fs(r300))) /* (5) */
+        return FALSE;
+
+    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]))
+        return FALSE;
+
+    if (dsa->depth.enabled) {
+        /* if depth func is EQUAL pre-r500 */
+        if (dsa->depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
+            return FALSE;
+        /* if depth func is NOTEQUAL */
+        if (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 & 0x7;
+        if (func != 0 && func != 7)
+            z->current_func = dsa_state->z_stencil_control & 0x7;
+    } else {
+        /* simple don't change */
+        if (!r300_zfunc_same_direction(z->current_func, (dsa_state->z_stencil_control & 0x7))) {
+            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;
+}
+
 static void r300_update_hyperz(struct r300_context* r300)
 {
     struct r300_hyperz_state *z =
         (struct r300_hyperz_state*)r300->hyperz_state.state;
+    struct pipe_framebuffer_state *fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
+    struct r300_texture *zstex =
+            fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL;
+    boolean zmask_in_use = FALSE;
+    boolean hiz_in_use = FALSE;
 
+    z->gb_z_peq_config = 0;
     z->zb_bw_cntl = 0;
     z->sc_hyperz = R300_SC_HYPERZ_ADJ_2;
+    z->flush = 0;
 
-    if (r300->cbzb_clear)
+    if (r300->cbzb_clear) {
         z->zb_bw_cntl |= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY;
+        return;
+    }
+
+    if (!zstex)
+        return;
+
+    if (!r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+        return;
+
+    zmask_in_use = zstex->zmask_in_use[fb->zsbuf->level];
+    hiz_in_use = zstex->hiz_in_use[fb->zsbuf->level];
+
+    /* Z fastfill. */
+    if (zmask_in_use) {
+        z->zb_bw_cntl |= R300_FAST_FILL_ENABLE; /*  | R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE;*/
+    }
+
+    /* Zbuffer compression. */
+    if (zmask_in_use && r300->z_compression) {
+        z->zb_bw_cntl |= R300_RD_COMP_ENABLE;
+        if (r300->z_decomp_rd == false)
+            z->zb_bw_cntl |= R300_WR_COMP_ENABLE;
+    }
+    /* RV350 and up optimizations. */
+    /* The section 10.4.9 in the docs is a lie. */
+    if (r300->z_compression == RV350_Z_COMPRESS_88)
+        z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
+
+    if (hiz_in_use) {
+        bool can_hiz = r300_can_hiz(r300);
+        if (can_hiz) {
+            z->zb_bw_cntl |= R300_HIZ_ENABLE;
+            z->sc_hyperz |= R300_SC_HYPERZ_ENABLE;
+            z->sc_hyperz |= r300_get_sc_hz_max(r300);
+            z->zb_bw_cntl |= r300_get_hiz_min(r300);
+        }
+    }
+
+    /* R500-specific features and optimizations. */
+    if (r300->screen->caps.is_r500) {
+        z->zb_bw_cntl |= R500_HIZ_FP_EXP_BITS_3;
+        z->zb_bw_cntl |=
+                R500_HIZ_EQUAL_REJECT_ENABLE |
+                R500_PEQ_PACKING_ENABLE |
+                R500_COVERED_PTR_MASKING_ENABLE;
+    }
 }
 
 /*****************************************************************************/
@@ -126,15 +275,121 @@ static void r300_update_ztop(struct r300_context* r300)
     } else {
         ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
     }
-
     if (ztop_state->z_buffer_top != old_ztop)
         r300->ztop_state.dirty = TRUE;
 }
 
+#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;
+}
+
+static void r300_update_zmask_clear(struct r300_context *r300)
+{
+    struct pipe_framebuffer_state *fb =
+        (struct pipe_framebuffer_state*)r300->fb_state.state;
+    uint32_t height;
+    int mult;
+
+    if (r300->z_compression == RV350_Z_COMPRESS_88)
+        mult = 8;
+    else
+        mult = 4;
+
+    height = ALIGN_DIVUP(fb->zsbuf->height, mult);
+
+    r300->zmask_clear.size = height * 4;
+}
+
 void r300_update_hyperz_state(struct r300_context* r300)
 {
     r300_update_ztop(r300);
     if (r300->hyperz_state.dirty) {
         r300_update_hyperz(r300);
     }
+
+    if (r300->hiz_clear.dirty) {
+       r300_update_hiz_clear(r300);
+    }
+    if (r300->zmask_clear.dirty) {
+       r300_update_zmask_clear(r300);
+    }
+}
+
+void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf)
+{
+    struct r300_texture *tex;
+    uint32_t zsize, ndw;
+    int level = surf->base.level;
+
+    tex = r300_texture(surf->base.texture);
+
+    if (tex->hiz_mem[level])
+        return;
+
+    zsize = tex->desc.layer_size_in_bytes[level];
+    zsize /= util_format_get_blocksize(tex->desc.b.b.format);
+    ndw = ALIGN_DIVUP(zsize, 64);
+
+    tex->hiz_mem[level] = u_mmAllocMem(r300->hiz_mm, ndw, 0, 0);
+    return;
+}
+
+void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress)
+{
+    int bsize = 256;
+    uint32_t zsize, ndw;
+    int level = surf->base.level;
+    struct r300_texture *tex;
+
+    tex = r300_texture(surf->base.texture);
+
+    /* We currently don't handle decompression for 3D textures and cubemaps
+     * correctly. */
+    if (tex->desc.b.b.target != PIPE_TEXTURE_1D &&
+        tex->desc.b.b.target != PIPE_TEXTURE_2D)
+        return;
+
+    if (tex->zmask_mem[level])
+        return;
+
+    zsize = tex->desc.layer_size_in_bytes[level];
+    zsize /= util_format_get_blocksize(tex->desc.b.b.format);
+
+    /* each zmask dword represents 16 4x4 blocks - which is 256 pixels
+       or 16 8x8 depending on the gb peq flag = 1024 pixels */
+    if (compress == RV350_Z_COMPRESS_88)
+        bsize = 1024;
+
+    ndw = ALIGN_DIVUP(zsize, bsize);
+    tex->zmask_mem[level] = u_mmAllocMem(r300->zmask_mm, ndw, 0, 0);
+    return;
+}
+
+void 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);
+
+    r300->zmask_mm = u_mmInit(0, r300screen->caps.zmask_ram * frag_pipes);
+}
+
+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);
+
+    u_mmDestroy(r300->zmask_mm);
 }
diff --git a/src/gallium/drivers/r300/r300_hyperz.h b/src/gallium/drivers/r300/r300_hyperz.h
index 3df5053b89..09e1ff6625 100644
--- a/src/gallium/drivers/r300/r300_hyperz.h
+++ b/src/gallium/drivers/r300/r300_hyperz.h
@@ -27,4 +27,9 @@ 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);
+void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress);
+
+void 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 2acc1a903e..60d3b600cb 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -496,6 +496,12 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_VAP_GB_HORZ_CLIP_ADJ                   0x2228
 #define R300_VAP_GB_HORZ_DISC_ADJ                   0x222c
 
+#define R300_VAP_PVS_FLOW_CNTL_ADDRS_0      0x2230
+#define R300_PVS_FC_ACT_ADRS(x)             ((x) << 0)
+#define R300_PVS_FC_LOOP_CNT_JMP_INST(x)    ((x) << 8)
+#define R300_PVS_FC_LAST_INST(x)            ((x) << 16)
+#define R300_PVS_FC_RTN_INST(x)             ((x) << 24)
+
 /* gap */
 
 /* Sometimes, END_OF_PKT and 0x2284=0 are the only commands sent between
@@ -514,6 +520,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #       define R300_2288_R300                    0x00750000 /* -- nh */
 #       define R300_2288_RV350                   0x0000FFFF /* -- Vladimir */
 
+#define R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0 0x2290
+#define R300_PVS_FC_LOOP_INIT_VAL(x)        ((x) << 0)
+#define R300_PVS_FC_LOOP_STEP_VAL(x)        ((x) << 8)
+
 /* gap */
 
 /* Addresses are relative to the vertex program instruction area of the
@@ -548,6 +558,9 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R300_VAP_PVS_CODE_CNTL_1	    0x22D8
 #       define R300_PVS_LAST_VTX_SRC_INST_SHIFT  0
 #define R300_VAP_PVS_FLOW_CNTL_OPC          0x22DC
+#define R300_VAP_PVS_FC_OPC_JUMP(x)         (1 << (2 * (x)))
+#define R300_VAP_PVS_FC_OPC_LOOP(x)         (2 << (2 * (x)))
+#define R300_VAP_PVS_FC_OPC_JSR(x)          (3 << (2 * (x)))
 
 /* The entire range from 0x2300 to 0x2AC inclusive seems to be used for
  * immediate vertices
@@ -564,6 +577,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* write 0 to indicate end of packet? */
 #define R300_VAP_VTX_END_OF_PKT             0x24AC
 
+#define R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0   0x2500
+#define R500_PVS_FC_ACT_ADRS(x)             ((x) << 0)
+#define R500_PVS_FC_LOOP_CNT_JMP_INST(x)    ((x) << 16)
+
+#define R500_VAP_PVS_FLOW_CNTL_ADDRS_UW_0   0x2504
+#define R500_PVS_FC_LAST_INST(x)            ((x) << 0)
+#define R500_PVS_FC_RTN_INST(x)             ((x) << 16)
+
 /* gap */
 
 /* These are values from r300_reg/r300_reg.h - they are known to be correct
@@ -3436,6 +3457,7 @@ enum {
 #   define R300_VBPNTR_SIZE1(x)    (((x) >> 2) << 16)
 #   define R300_VBPNTR_STRIDE1(x)  (((x) >> 2) << 24)
 
+#define R300_PACKET3_3D_CLEAR_ZMASK         0x00003200
 #define R300_PACKET3_INDX_BUFFER            0x00003300
 #    define R300_INDX_BUFFER_DST_SHIFT          0
 #    define R300_INDX_BUFFER_SKIP_SHIFT         16
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index bae02135da..86b11ca045 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -186,20 +186,14 @@ enum r300_prepare_flags {
  * \param cs_dwords     The number of dwords to reserve in CS.
  * \param aos_offset    The offset passed to emit_aos.
  * \param index_bias    The index bias to emit.
- * \param end_cs_dwords The number of free dwords which must be available
- *                      at the end of CS after drawing in case the CS space
- *                      management is performed by a draw_* function manually.
- *                      The parameter may be NULL.
  */
 static void r300_prepare_for_rendering(struct r300_context *r300,
                                        enum r300_prepare_flags flags,
                                        struct pipe_resource *index_buffer,
                                        unsigned cs_dwords,
                                        int aos_offset,
-                                       int index_bias,
-                                       unsigned *end_cs_dwords)
+                                       int index_bias)
 {
-    unsigned end_dwords    = 0;
     boolean flushed        = FALSE;
     boolean first_draw     = flags & PREP_FIRST_DRAW;
     boolean emit_aos       = flags & PREP_EMIT_AOS;
@@ -221,11 +215,7 @@ static void r300_prepare_for_rendering(struct r300_context *r300,
             cs_dwords += 7; /* emit_aos_swtcl */
     }
 
-    /* Emitted in flush. */
-    end_dwords += 26; /* emit_query_end */
-    end_dwords += r300->hyperz_state.size; /* emit_hyperz_end */
-
-    cs_dwords += end_dwords;
+    cs_dwords += r300_get_num_cs_end_dwords(r300);
 
     /* Reserve requested CS space. */
     if (cs_dwords > (r300->cs->ndw - r300->cs->cdw)) {
@@ -250,9 +240,6 @@ static void r300_prepare_for_rendering(struct r300_context *r300,
         if (emit_aos_swtcl)
             r300_emit_aos_swtcl(r300, indexed);
     }
-
-    if (end_cs_dwords)
-        *end_cs_dwords = end_dwords;
 }
 
 static boolean immd_is_good_idea(struct r300_context *r300,
@@ -353,7 +340,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     dwords = 9 + count * vertex_size;
 
-    r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL);
+    r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0);
 
     BEGIN_CS(dwords);
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
@@ -533,7 +520,7 @@ static void r300_draw_range_elements(struct pipe_context* pipe,
     /* 15 dwords for emit_draw_elements */
     r300_prepare_for_rendering(r300,
         PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
-        indexBuffer, 15, buffer_offset, indexBias, NULL);
+        indexBuffer, 15, buffer_offset, indexBias);
 
     if (alt_num_verts || count <= 65535) {
         r300_emit_draw_elements(r300, indexBuffer, indexSize,
@@ -552,7 +539,7 @@ static void r300_draw_range_elements(struct pipe_context* pipe,
             if (count) {
                 r300_prepare_for_rendering(r300,
                     PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
-                    indexBuffer, 15, buffer_offset, indexBias, NULL);
+                    indexBuffer, 15, buffer_offset, indexBias);
             }
         } while (count);
     }
@@ -566,19 +553,6 @@ static void r300_draw_range_elements(struct pipe_context* pipe,
     }
 }
 
-/* Simple helpers for context setup. Should probably be moved to util. */
-static void r300_draw_elements(struct pipe_context* pipe,
-                               struct pipe_resource* indexBuffer,
-                               unsigned indexSize, int indexBias, unsigned mode,
-                               unsigned start, unsigned count)
-{
-    struct r300_context *r300 = r300_context(pipe);
-
-    pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
-                              0, r300->vertex_buffer_max_index,
-                              mode, start, count);
-}
-
 static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                              unsigned start, unsigned count)
 {
@@ -610,7 +584,7 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
     } else {
         /* 9 spare dwords for emit_draw_arrays. */
         r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
-                               NULL, 9, start, 0, NULL);
+                               NULL, 9, start, 0);
 
         if (alt_num_verts || count <= 65535) {
             r300_emit_draw_arrays(r300, mode, count);
@@ -626,7 +600,7 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                 if (count) {
                     r300_prepare_for_rendering(r300,
                         PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
-                        start, 0, NULL);
+                        start, 0);
                 }
             } while (count);
         }
@@ -638,111 +612,104 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
     }
 }
 
-/****************************************************************************
- * The rest of this file is for SW TCL rendering only. Please be polite and *
- * keep these functions separated so that they are easier to locate. ~C.    *
- ***************************************************************************/
-
-/* SW TCL arrays, using Draw. */
-static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
-                                   unsigned mode,
-                                   unsigned start,
-                                   unsigned count)
+static void r300_draw_vbo(struct pipe_context* pipe,
+                          const struct pipe_draw_info *info)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
-    int i;
 
-    if (r300->skip_rendering) {
-        return;
+    if (info->indexed && r300->index_buffer.buffer) {
+        unsigned offset;
+
+        assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0);
+        offset = r300->index_buffer.offset / r300->index_buffer.index_size;
+
+        r300_draw_range_elements(pipe,
+                                 r300->index_buffer.buffer,
+                                 r300->index_buffer.index_size,
+                                 info->index_bias,
+                                 info->min_index,
+                                 info->max_index,
+                                 info->mode,
+                                 info->start + offset,
+                                 info->count);
     }
-
-    if (!u_trim_pipe_prim(mode, &count)) {
-        return;
-    }
-
-    r300_update_derived_state(r300);
-
-    for (i = 0; i < r300->vertex_buffer_count; i++) {
-        void* buf = pipe_buffer_map(pipe,
-                                    r300->vertex_buffer[i].buffer,
-                                    PIPE_TRANSFER_READ,
-				    &vb_transfer[i]);
-        draw_set_mapped_vertex_buffer(r300->draw, i, buf);
-    }
-
-    draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL);
-
-    draw_arrays(r300->draw, mode, start, count);
-
-    /* XXX Not sure whether this is the best fix.
-     * It prevents CS from being rejected and weird assertion failures. */
-    draw_flush(r300->draw);
-
-    for (i = 0; i < r300->vertex_buffer_count; i++) {
-        pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
-			  vb_transfer[i]);
-        draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+    else {
+        r300_draw_arrays(pipe,
+                         info->mode,
+                         info->start,
+                         info->count);
     }
 }
 
+/****************************************************************************
+ * The rest of this file is for SW TCL rendering only. Please be polite and *
+ * keep these functions separated so that they are easier to locate. ~C.    *
+ ***************************************************************************/
+
 /* SW TCL elements, using Draw. */
-static void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
-                                           struct pipe_resource* indexBuffer,
-                                           unsigned indexSize,
-                                           int indexBias,
-                                           unsigned minIndex,
-                                           unsigned maxIndex,
-                                           unsigned mode,
-                                           unsigned start,
-                                           unsigned count)
+static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
+                                const struct pipe_draw_info *info)
 {
     struct r300_context* r300 = r300_context(pipe);
     struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
-    struct pipe_transfer *ib_transfer;
+    struct pipe_transfer *ib_transfer = NULL;
+    unsigned count = info->count;
     int i;
-    void* indices;
+    void* indices = NULL;
 
     if (r300->skip_rendering) {
         return;
     }
 
-    if (!u_trim_pipe_prim(mode, &count)) {
+    if (!u_trim_pipe_prim(info->mode, &count)) {
         return;
     }
 
     r300_update_derived_state(r300);
 
     for (i = 0; i < r300->vertex_buffer_count; i++) {
-        void* buf = pipe_buffer_map(pipe,
-                                    r300->vertex_buffer[i].buffer,
-                                    PIPE_TRANSFER_READ,
-				    &vb_transfer[i]);
-        draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+        if (r300->vertex_buffer[i].buffer) {
+            void *buf = pipe_buffer_map(pipe,
+                                  r300->vertex_buffer[i].buffer,
+                                  PIPE_TRANSFER_READ,
+                                  &vb_transfer[i]);
+            draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+        }
     }
 
-    indices = pipe_buffer_map(pipe, indexBuffer,
-                              PIPE_TRANSFER_READ, &ib_transfer);
-    draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias,
-                                         minIndex, maxIndex, indices);
+    if (info->indexed && r300->index_buffer.buffer) {
+        indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
+                                  PIPE_TRANSFER_READ, &ib_transfer);
+        if (indices)
+            indices = (void *) ((char *) indices + r300->index_buffer.offset);
+    }
 
-    draw_arrays(r300->draw, mode, start, count);
+    draw_set_mapped_element_buffer_range(r300->draw, (indices) ?
+                                         r300->index_buffer.index_size : 0,
+                                         info->index_bias,
+                                         info->min_index,
+                                         info->max_index,
+                                         indices);
+
+    draw_arrays(r300->draw, info->mode, info->start, count);
 
     /* XXX Not sure whether this is the best fix.
      * It prevents CS from being rejected and weird assertion failures. */
     draw_flush(r300->draw);
 
     for (i = 0; i < r300->vertex_buffer_count; i++) {
-        pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
-			  vb_transfer[i]);
-        draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+        if (r300->vertex_buffer[i].buffer) {
+            pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
+                              vb_transfer[i]);
+            draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+        }
     }
 
-    pipe_buffer_unmap(pipe, indexBuffer,
-		      ib_transfer);
-    draw_set_mapped_element_buffer_range(r300->draw, 0, 0,
-                                         start, start + count - 1,
-                                         NULL);
+    if (ib_transfer) {
+        pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer);
+        draw_set_mapped_element_buffer_range(r300->draw, 0, 0, info->start,
+                info->start + count - 1, NULL);
+    }
 }
 
 /* Object for rendering using Draw. */
@@ -820,6 +787,8 @@ static void* r300_render_map_vertices(struct vbuf_render* render)
                                           PIPE_TRANSFER_WRITE,
 					  &r300render->vbo_transfer);
 
+    assert(r300render->vbo_ptr);
+
     return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
 }
 
@@ -872,7 +841,7 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
     (void) i; (void) ptr;
 
     r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
-                               NULL, dwords, 0, 0, NULL);
+                               NULL, dwords, 0, 0);
 
     DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
 
@@ -925,7 +894,8 @@ static void r300_render_draw_elements(struct vbuf_render* render,
      * indices than it can fit in CS. */
     r300_prepare_for_rendering(r300,
         PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
-        NULL, 256, 0, 0, &end_cs_dwords);
+        NULL, 256, 0, 0);
+    end_cs_dwords = r300_get_num_cs_end_dwords(r300);
 
     while (count) {
         free_dwords = r300->cs->ndw - r300->cs->cdw;
@@ -955,7 +925,8 @@ static void r300_render_draw_elements(struct vbuf_render* render,
         if (count) {
             r300_prepare_for_rendering(r300,
                 PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
-                NULL, 256, 0, 0, &end_cs_dwords);
+                NULL, 256, 0, 0);
+            end_cs_dwords = r300_get_num_cs_end_dwords(r300);
         }
     }
 }
@@ -1049,7 +1020,7 @@ static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
     r300->clip_state.dirty = FALSE;
     r300->viewport_state.dirty = FALSE;
 
-    r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL);
+    r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0);
 
     DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
 
@@ -1141,16 +1112,11 @@ static void r300_resource_resolve(struct pipe_context* pipe,
 
 void r300_init_render_functions(struct r300_context *r300)
 {
-    /* Set generic functions. */
-    r300->context.draw_elements = r300_draw_elements;
-
     /* Set draw functions based on presence of HW TCL. */
     if (r300->screen->caps.has_tcl) {
-        r300->context.draw_arrays = r300_draw_arrays;
-        r300->context.draw_range_elements = r300_draw_range_elements;
+        r300->context.draw_vbo = r300_draw_vbo;
     } else {
-        r300->context.draw_arrays = r300_swtcl_draw_arrays;
-        r300->context.draw_range_elements = r300_swtcl_draw_range_elements;
+        r300->context.draw_vbo = r300_swtcl_draw_vbo;
     }
 
     r300->context.resource_resolve = r300_resource_resolve;
diff --git a/src/gallium/drivers/r300/r300_render_stencilref.c b/src/gallium/drivers/r300/r300_render_stencilref.c
index 9a6b4e12ff..1f035d64a2 100644
--- a/src/gallium/drivers/r300/r300_render_stencilref.c
+++ b/src/gallium/drivers/r300/r300_render_stencilref.c
@@ -34,13 +34,8 @@
 #include "r300_reg.h"
 
 struct r300_stencilref_context {
-    void (*draw_arrays)(struct pipe_context *pipe,
-                        unsigned mode, unsigned start, unsigned count);
-
-    void (*draw_range_elements)(
-        struct pipe_context *pipe, struct pipe_resource *indexBuffer,
-        unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
-        unsigned mode, unsigned start, unsigned count);
+    void (*draw_vbo)(struct pipe_context *pipe,
+                     const struct pipe_draw_info *info);
 
     uint32_t rs_cull_mode;
     uint32_t zb_stencilrefmask;
@@ -105,41 +100,19 @@ static void r300_stencilref_end(struct r300_context *r300)
     r300->dsa_state.dirty = TRUE;
 }
 
-static void r300_stencilref_draw_arrays(struct pipe_context *pipe, unsigned mode,
-                                        unsigned start, unsigned count)
-{
-    struct r300_context *r300 = r300_context(pipe);
-    struct r300_stencilref_context *sr = r300->stencilref_fallback;
-
-    if (!r300_stencilref_needed(r300)) {
-        sr->draw_arrays(pipe, mode, start, count);
-    } else {
-        r300_stencilref_begin(r300);
-        sr->draw_arrays(pipe, mode, start, count);
-        r300_stencilref_switch_side(r300);
-        sr->draw_arrays(pipe, mode, start, count);
-        r300_stencilref_end(r300);
-    }
-}
-
-static void r300_stencilref_draw_range_elements(
-    struct pipe_context *pipe, struct pipe_resource *indexBuffer,
-    unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
-    unsigned mode, unsigned start, unsigned count)
+static void r300_stencilref_draw_vbo(struct pipe_context *pipe,
+                                     const struct pipe_draw_info *info)
 {
     struct r300_context *r300 = r300_context(pipe);
     struct r300_stencilref_context *sr = r300->stencilref_fallback;
 
     if (!r300_stencilref_needed(r300)) {
-        sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
-                                minIndex, maxIndex, mode, start, count);
+        sr->draw_vbo(pipe, info);
     } else {
         r300_stencilref_begin(r300);
-        sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
-                                minIndex, maxIndex, mode, start, count);
+        sr->draw_vbo(pipe, info);
         r300_stencilref_switch_side(r300);
-        sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
-                                minIndex, maxIndex, mode, start, count);
+        sr->draw_vbo(pipe, info);
         r300_stencilref_end(r300);
     }
 }
@@ -148,11 +121,9 @@ void r300_plug_in_stencil_ref_fallback(struct r300_context *r300)
 {
     r300->stencilref_fallback = CALLOC_STRUCT(r300_stencilref_context);
 
-    /* Save original draw functions. */
-    r300->stencilref_fallback->draw_arrays = r300->context.draw_arrays;
-    r300->stencilref_fallback->draw_range_elements = r300->context.draw_range_elements;
+    /* Save original draw function. */
+    r300->stencilref_fallback->draw_vbo = r300->context.draw_vbo;
 
-    /* Override the draw functions. */
-    r300->context.draw_arrays = r300_stencilref_draw_arrays;
-    r300->context.draw_range_elements = r300_stencilref_draw_range_elements;
+    /* Override the draw function. */
+    r300->context.draw_vbo = r300_stencilref_draw_vbo;
 }
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 5a11b98eb6..1e4edcdbc3 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -115,7 +115,6 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
         case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
         case PIPE_CAP_BLEND_EQUATION_SEPARATE:
         case PIPE_CAP_TEXTURE_SWIZZLE:
-        case PIPE_CAP_DEPTH_CLAMP:
             return 1;
 
         /* Unsupported features (boolean caps). */
@@ -124,6 +123,8 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
         case PIPE_CAP_TGSI_CONT_SUPPORTED:
         case PIPE_CAP_INDEP_BLEND_ENABLE:
         case PIPE_CAP_INDEP_BLEND_FUNC:
+        case PIPE_CAP_DEPTH_CLAMP: /* XXX implemented, but breaks Regnum Online */
+        case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
             return 0;
 
         /* Texturing. */
@@ -150,9 +151,6 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
         case PIPE_CAP_MAX_CONST_BUFFER_SIZE:
             return 256;
 
-        case PIPE_CAP_DEPTHSTENCIL_CLEAR_SEPARATE:
-            return 1;
-
         /* Fragment coordinate conventions. */
         case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
         case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
@@ -257,8 +255,6 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
     uint32_t retval = 0;
     boolean is_r500 = r300_screen(screen)->caps.is_r500;
     boolean is_r400 = r300_screen(screen)->caps.is_r400;
-    boolean is_z24 = format == PIPE_FORMAT_X8Z24_UNORM ||
-                     format == PIPE_FORMAT_S8_USCALED_Z24_UNORM;
     boolean is_color2101010 = format == PIPE_FORMAT_R10G10B10A2_UNORM ||
                               format == PIPE_FORMAT_R10G10B10X2_SNORM ||
                               format == PIPE_FORMAT_B10G10R10A2_UNORM ||
@@ -281,11 +277,14 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
         case 3:
         case 4:
         case 6:
+            return FALSE;
+#if 0
             if (usage != PIPE_BIND_RENDER_TARGET ||
                 !util_format_is_rgba8_variant(
                     util_format_description(format))) {
                 return FALSE;
             }
+#endif
             break;
         default:
             return FALSE;
@@ -293,8 +292,6 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
 
     /* Check sampler format support. */
     if ((usage & PIPE_BIND_SAMPLER_VIEW) &&
-        /* Z24 cannot be sampled from on non-r5xx. */
-        (is_r500 || !is_z24) &&
         /* ATI1N is r5xx-only. */
         (is_r500 || !is_ati1n) &&
         /* ATI2N is supported on r4xx-r5xx. */
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index edc494ff6c..13a3320b99 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -90,6 +90,8 @@ r300_winsys_screen(struct pipe_screen *screen) {
 #define DBG_FALL        (1 << 8)
 #define DBG_FB          (1 << 9)
 #define DBG_RS_BLOCK    (1 << 10)
+#define DBG_CBZB        (1 << 11)
+#define DBG_HYPERZ      (1 << 12)
 /* Features. */
 #define DBG_ANISOHQ     (1 << 16)
 #define DBG_NO_TILING   (1 << 17)
diff --git a/src/gallium/drivers/r300/r300_shader_semantics.h b/src/gallium/drivers/r300/r300_shader_semantics.h
index cb7a37033f..4be23e64ce 100644
--- a/src/gallium/drivers/r300/r300_shader_semantics.h
+++ b/src/gallium/drivers/r300/r300_shader_semantics.h
@@ -38,6 +38,7 @@ struct r300_shader_semantics {
     int psize;
     int color[ATTR_COLOR_COUNT];
     int bcolor[ATTR_COLOR_COUNT];
+    int face;
     int generic[ATTR_GENERIC_COUNT];
     int fog;
     int wpos;
@@ -50,6 +51,7 @@ static INLINE void r300_shader_semantics_reset(
 
     info->pos = ATTR_UNUSED;
     info->psize = ATTR_UNUSED;
+    info->face = ATTR_UNUSED;
     info->fog = ATTR_UNUSED;
     info->wpos = ATTR_UNUSED;
 
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index f52265b1c0..239edd98e3 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -25,6 +25,7 @@
 
 #include "util/u_blitter.h"
 #include "util/u_math.h"
+#include "util/u_mm.h"
 #include "util/u_memory.h"
 #include "util/u_pack_color.h"
 
@@ -43,6 +44,7 @@
 #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. */
@@ -446,7 +448,6 @@ static void r300_set_clip_state(struct pipe_context* pipe,
 
         r300->clip_state.dirty = TRUE;
     } else {
-        draw_flush(r300->draw);
         draw_set_clip_state(r300->draw, state);
     }
 }
@@ -473,14 +474,14 @@ static void*
 
     dsa->dsa = *state;
 
-    /* Depth test setup. */
+    /* Depth test setup. - separate write mask depth for decomp flush */
+    if (state->depth.writemask) {
+        dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
+    }
+
     if (state->depth.enabled) {
         dsa->z_buffer_control |= R300_Z_ENABLE;
 
-        if (state->depth.writemask) {
-            dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
-        }
-
         dsa->z_stencil_control |=
             (r300_translate_depth_stencil_function(state->depth.func) <<
                 R300_Z_FUNC_SHIFT);
@@ -593,6 +594,7 @@ static void r300_bind_dsa_state(struct pipe_context* pipe,
 
     UPDATE_STATE(state, r300->dsa_state);
 
+    r300->hyperz_state.dirty = TRUE; /* Will be updated before the emission. */
     r300_dsa_inject_stencilref(r300);
 }
 
@@ -619,7 +621,8 @@ static void r300_tex_set_tiling_flags(struct r300_context *r300,
 {
     /* Check if the macrotile flag needs to be changed.
      * Skip changing the flags otherwise. */
-    if (tex->mip_macrotile[tex->surface_level] != tex->mip_macrotile[level]) {
+    if (tex->desc.macrotile[tex->surface_level] !=
+        tex->desc.macrotile[level]) {
         /* Tiling determines how DRM treats the buffer data.
          * We must flush CS when changing it if the buffer is referenced. */
         if (r300->rws->cs_is_buffer_referenced(r300->cs,
@@ -627,8 +630,8 @@ static void r300_tex_set_tiling_flags(struct r300_context *r300,
             r300->context.flush(&r300->context, 0, NULL);
 
         r300->rws->buffer_set_tiling(r300->rws, tex->buffer,
-                tex->microtile, tex->mip_macrotile[level],
-                tex->pitch[0] * util_format_get_blocksize(tex->b.b.format));
+                tex->desc.microtile, tex->desc.macrotile[level],
+                tex->desc.stride_in_bytes[0]);
 
         tex->surface_level = level;
     }
@@ -670,8 +673,10 @@ static void r300_print_fb_surf_info(struct pipe_surface *surf, unsigned index,
             surf->zslice, surf->face, surf->level,
             util_format_short_name(surf->format),
 
-            rtex->macrotile ? "YES" : " NO", rtex->microtile ? "YES" : " NO",
-            rtex->hwpitch[0], tex->width0, tex->height0, tex->depth0,
+            rtex->desc.macrotile[0] ? "YES" : " NO",
+            rtex->desc.microtile ? "YES" : " NO",
+            rtex->desc.stride_in_pixels[0],
+            tex->width0, tex->height0, tex->depth0,
             tex->last_level, util_format_short_name(tex->format));
 }
 
@@ -679,6 +684,7 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
                               enum r300_fb_state_change change)
 {
     struct pipe_framebuffer_state *state = r300->fb_state.state;
+    boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
 
     /* What is marked as dirty depends on the enum r300_fb_state_change. */
     r300->gpu_flush.dirty = TRUE;
@@ -695,8 +701,11 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
 
     if (r300->cbzb_clear)
         r300->fb_state.size += 10;
-    else if (state->zsbuf)
-        r300->fb_state.size += r300->screen->caps.has_hiz ? 18 : 14;
+    else if (state->zsbuf) {
+        r300->fb_state.size += 10;
+        if (has_hyperz)
+            r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4;
+    }
 
     /* The size of the rest of atoms stays the same. */
 }
@@ -708,8 +717,10 @@ static void
     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 has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
     unsigned max_width, max_height, i;
     uint32_t zbuffer_bpp = 0;
+    int blocksize;
 
     if (r300->screen->caps.is_r500) {
         max_width = max_height = 4096;
@@ -725,10 +736,6 @@ static void
         return;
     }
 
-    if (r300->draw) {
-        draw_flush(r300->draw);
-    }
-
     /* If nr_cbufs is changed from zero to non-zero or vice versa... */
     if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
         r300->blend_state.dirty = TRUE;
@@ -745,20 +752,50 @@ static void
 
     r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
 
-    /* Polygon offset depends on the zbuffer bit depth. */
-    if (state->zsbuf && r300->polygon_offset_enabled) {
-        switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
-            case 2:
-                zbuffer_bpp = 16;
-                break;
-            case 4:
-                zbuffer_bpp = 24;
-                break;
+    r300->z_compression = false;
+    
+    if (state->zsbuf) {
+        blocksize = util_format_get_blocksize(state->zsbuf->texture->format);
+        switch (blocksize) {
+        case 2:
+            zbuffer_bpp = 16;
+            break;
+        case 4:
+            zbuffer_bpp = 24;
+            break;
+        }
+        if (has_hyperz) {
+            struct r300_surface *zs_surf = r300_surface(state->zsbuf);
+            struct r300_texture *tex;
+            int compress = r300->screen->caps.is_rv350 ? RV350_Z_COMPRESS_88 : R300_Z_COMPRESS_44;
+            int level = zs_surf->base.level;
+
+            tex = r300_texture(zs_surf->base.texture);
+
+            /* work out whether we can support hiz on this buffer */
+            r300_hiz_alloc_block(r300, zs_surf);
+        
+            /* work out whether we can support zmask features on this buffer */
+            r300_zmask_alloc_block(r300, zs_surf, compress);
+
+            if (tex->zmask_mem[level]) {
+                /* compression causes hangs on 16-bit */
+                if (zbuffer_bpp == 24)
+                    r300->z_compression = compress;
+            }
+            DBG(r300, DBG_HYPERZ,
+                "hyper-z features: hiz: %d @ %08x z-compression: %d z-fastfill: %d @ %08x\n", tex->hiz_mem[level] ? 1 : 0,
+                tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef,
+                r300->z_compression, tex->zmask_mem[level] ? 1 : 0,
+                tex->zmask_mem[level] ? tex->zmask_mem[level]->ofs : 0xdeadbeef);
         }
 
+        /* Polygon offset depends on the zbuffer bit depth. */
         if (r300->zbuffer_bpp != zbuffer_bpp) {
             r300->zbuffer_bpp = zbuffer_bpp;
-            r300->rs_state.dirty = TRUE;
+
+            if (r300->polygon_offset_enabled)
+                r300->rs_state.dirty = TRUE;
         }
     }
 
@@ -1016,7 +1053,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
 	for (i = 0; i < 8; i++) {
 	    if (state->sprite_coord_enable & (1 << i))
                 stuffing_enable |=
-		    R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2));
+                    R300_GB_TEX_ST << (R300_GB_TEX0_SOURCE_SHIFT + (i*2));
 	}
 
         point_texcoord_left = 0.0f;
@@ -1093,14 +1130,11 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
     boolean last_two_sided_color = r300->two_sided_color;
 
     if (r300->draw && rs) {
-        draw_flush(r300->draw);
         draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
     }
 
     if (rs) {
-        r300->polygon_offset_enabled = (rs->rs.offset_point ||
-                                        rs->rs.offset_line ||
-                                        rs->rs.offset_tri);
+        r300->polygon_offset_enabled = rs->polygon_offset_enable;
         r300->sprite_coord_enable = rs->rs.sprite_coord_enable;
         r300->two_sided_color = rs->rs.light_twoside;
     } else {
@@ -1293,7 +1327,7 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
             /* Set the texrect factor in the fragment shader.
              * Needed for RECT and NPOT fallback. */
             texture = r300_texture(views[i]->texture);
-            if (texture->uses_pitch) {
+            if (texture->desc.is_npot) {
                 r300->fs_rc_constant_state.dirty = TRUE;
             }
 
@@ -1327,6 +1361,7 @@ r300_create_sampler_view(struct pipe_context *pipe,
 {
     struct r300_sampler_view *view = CALLOC_STRUCT(r300_sampler_view);
     struct r300_texture *tex = r300_texture(texture);
+    boolean is_r500 = r300_screen(pipe->screen)->caps.is_r500;
 
     if (view) {
         view->base = *templ;
@@ -1342,8 +1377,9 @@ r300_create_sampler_view(struct pipe_context *pipe,
 
         view->format = tex->tx_format;
         view->format.format1 |= r300_translate_texformat(templ->format,
-                                                         view->swizzle);
-        if (r300_screen(pipe->screen)->caps.is_r500) {
+                                                         view->swizzle,
+                                                         is_r500);
+        if (is_r500) {
             view->format.format2 |= r500_tx_format_msb_bit(templ->format);
         }
     }
@@ -1380,7 +1416,6 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
     r300->viewport = *state;
 
     if (r300->draw) {
-        draw_flush(r300->draw);
         draw_set_viewport_state(r300->draw, state);
         viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT;
         return;
@@ -1481,7 +1516,6 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
 
     } else {
         /* SW TCL. */
-        draw_flush(r300->draw);
         draw_set_vertex_buffers(r300->draw, count, buffers);
     }
 
@@ -1500,6 +1534,23 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
     r300->vertex_buffer_count = count;
 }
 
+static void r300_set_index_buffer(struct pipe_context* pipe,
+                                  const struct pipe_index_buffer *ib)
+{
+    struct r300_context* r300 = r300_context(pipe);
+
+    if (ib) {
+        pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer);
+        memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer));
+    }
+    else {
+        pipe_resource_reference(&r300->index_buffer.buffer, NULL);
+        memset(&r300->index_buffer, 0, sizeof(r300->index_buffer));
+    }
+
+    /* TODO make this more like a state */
+}
+
 /* Initialize the PSC tables. */
 static void r300_vertex_psc(struct r300_vertex_element_state *velems)
 {
@@ -1649,7 +1700,6 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
     r300->velems = velems;
 
     if (r300->draw) {
-        draw_flush(r300->draw);
         draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
         return;
     }
@@ -1701,10 +1751,12 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
     r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
 
     if (r300->screen->caps.has_tcl) {
+        unsigned fc_op_dwords = r300->screen->caps.is_r500 ? 3 : 2;
         r300->vs_state.dirty = TRUE;
         r300->vs_state.size =
                 vs->code.length + 9 +
-                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0);
+                (vs->immediates_count ? vs->immediates_count * 4 + 3 : 0) +
+        (vs->code.num_fc_ops ? vs->code.num_fc_ops * fc_op_dwords + 4 : 0);
 
         if (vs->externals_count) {
             r300->vs_constants.dirty = TRUE;
@@ -1715,7 +1767,6 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
 
         r300->pvs_flush.dirty = TRUE;
     } else {
-        draw_flush(r300->draw);
         draw_bind_vertex_shader(r300->draw,
                 (struct draw_vertex_shader*)vs->draw_vs);
     }
@@ -1847,6 +1898,7 @@ void r300_init_state_functions(struct r300_context* r300)
     r300->context.set_viewport_state = r300_set_viewport_state;
 
     r300->context.set_vertex_buffers = r300_set_vertex_buffers;
+    r300->context.set_index_buffer = r300_set_index_buffer;
 
     r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
     r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 2ef9766578..4a63ed7fc1 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -46,6 +46,11 @@ enum r300_rs_swizzle {
     SWIZ_0001,
 };
 
+enum r300_rs_col_write_type {
+    WRITE_COLOR = 0,
+    WRITE_FACE
+};
+
 static void r300_draw_emit_attrib(struct r300_context* r300,
                                   enum attrib_emit emit,
                                   enum interp_mode interp,
@@ -203,8 +208,10 @@ static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr,
     rs->inst[id] |= R300_RS_INST_COL_ID(id);
 }
 
-static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
+static void r300_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset,
+                              enum r300_rs_col_write_type type)
 {
+    assert(type != WRITE_COLOR);
     rs->inst[id] |= R300_RS_INST_COL_CN_WRITE |
                     R300_RS_INST_COL_ADDR(fp_offset);
 }
@@ -213,19 +220,19 @@ static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr,
                         enum r300_rs_swizzle swiz)
 {
     if (swiz == SWIZ_X001) {
-        rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+        rs->ip[id] |= R300_RS_TEX_PTR(ptr) |
                       R300_RS_SEL_S(R300_RS_SEL_C0) |
                       R300_RS_SEL_T(R300_RS_SEL_K0) |
                       R300_RS_SEL_R(R300_RS_SEL_K0) |
                       R300_RS_SEL_Q(R300_RS_SEL_K1);
     } else if (swiz == SWIZ_XY01) {
-        rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+        rs->ip[id] |= R300_RS_TEX_PTR(ptr) |
                       R300_RS_SEL_S(R300_RS_SEL_C0) |
                       R300_RS_SEL_T(R300_RS_SEL_C1) |
                       R300_RS_SEL_R(R300_RS_SEL_K0) |
                       R300_RS_SEL_Q(R300_RS_SEL_K1);
     } else {
-        rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+        rs->ip[id] |= R300_RS_TEX_PTR(ptr) |
                       R300_RS_SEL_S(R300_RS_SEL_C0) |
                       R300_RS_SEL_T(R300_RS_SEL_C1) |
                       R300_RS_SEL_R(R300_RS_SEL_C2) |
@@ -252,32 +259,36 @@ static void r500_rs_col(struct r300_rs_block* rs, int id, int ptr,
     rs->inst[id] |= R500_RS_INST_COL_ID(id);
 }
 
-static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
+static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset,
+                              enum r300_rs_col_write_type type)
 {
-    rs->inst[id] |= R500_RS_INST_COL_CN_WRITE |
-                    R500_RS_INST_COL_ADDR(fp_offset);
+    if (type == WRITE_FACE)
+        rs->inst[id] |= R500_RS_INST_COL_CN_WRITE_BACKFACE |
+                        R500_RS_INST_COL_ADDR(fp_offset);
+    else
+        rs->inst[id] |= R500_RS_INST_COL_CN_WRITE |
+                        R500_RS_INST_COL_ADDR(fp_offset);
+
 }
 
 static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr,
 			enum r300_rs_swizzle swiz)
 {
-    int rs_tex_comp = ptr*4;
-
     if (swiz == SWIZ_X001) {
-        rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
+        rs->ip[id] |= R500_RS_SEL_S(ptr) |
                       R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
                       R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
                       R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
     } else if (swiz == SWIZ_XY01) {
-        rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
-                      R500_RS_SEL_T(rs_tex_comp + 1) |
+        rs->ip[id] |= R500_RS_SEL_S(ptr) |
+                      R500_RS_SEL_T(ptr + 1) |
                       R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
                       R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
     } else {
-        rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
-                      R500_RS_SEL_T(rs_tex_comp + 1) |
-                      R500_RS_SEL_R(rs_tex_comp + 2) |
-                      R500_RS_SEL_Q(rs_tex_comp + 3);
+        rs->ip[id] |= R500_RS_SEL_S(ptr) |
+                      R500_RS_SEL_T(ptr + 1) |
+                      R500_RS_SEL_R(ptr + 2) |
+                      R500_RS_SEL_Q(ptr + 3);
     }
     rs->inst[id] |= R500_RS_INST_TEX_ID(id);
 }
@@ -305,9 +316,9 @@ static void r300_update_rs_block(struct r300_context *r300)
     struct r300_shader_semantics *vs_outputs = &vs->outputs;
     struct r300_shader_semantics *fs_inputs = &r300_fs(r300)->shader->inputs;
     struct r300_rs_block rs = {0};
-    int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0;
+    int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0, tex_ptr = 0;
     void (*rX00_rs_col)(struct r300_rs_block*, int, int, enum r300_rs_swizzle);
-    void (*rX00_rs_col_write)(struct r300_rs_block*, int, int);
+    void (*rX00_rs_col_write)(struct r300_rs_block*, int, int, enum r300_rs_col_write_type);
     void (*rX00_rs_tex)(struct r300_rs_block*, int, int, enum r300_rs_swizzle);
     void (*rX00_rs_tex_write)(struct r300_rs_block*, int, int);
     boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
@@ -326,6 +337,11 @@ static void r300_update_rs_block(struct r300_context *r300)
         rX00_rs_tex_write = r300_rs_tex_write;
     }
 
+    /* 0x5555 copied from classic, which means:
+     * Select user color 0 for COLOR0 up to COLOR7.
+     * What the hell does that mean? */
+    rs.vap_vtx_state_cntl = 0x5555;
+
     /* The position is always present in VAP. */
     rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS;
     rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
@@ -352,7 +368,7 @@ static void r300_update_rs_block(struct r300_context *r300)
 
             /* Write it to the FS input register if it's needed by the FS. */
             if (fs_inputs->color[i] != ATTR_UNUSED) {
-                rX00_rs_col_write(&rs, col_count, fp_offset);
+                rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_COLOR);
                 fp_offset++;
 
                 DBG(r300, DBG_RS,
@@ -393,12 +409,31 @@ static void r300_update_rs_block(struct r300_context *r300)
                 stream_loc_notcl[loc++] = 6 + tex_count;
 
                 /* Rasterize it. */
-                rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW);
+                rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW);
                 tex_count++;
+                tex_ptr += 4;
             }
         }
     }
 
+    /* gl_FrontFacing.
+     * Note that we can use either the two-sided color selection based on
+     * the front and back vertex shader colors, or gl_FrontFacing,
+     * but not both! It locks up otherwise.
+     *
+     * In Direct3D 9, the two-sided color selection can be used
+     * with shaders 2.0 only, while gl_FrontFacing can be used
+     * with shaders 3.0 only. The hardware apparently hasn't been designed
+     * to support both at the same time. */
+    if (r300->screen->caps.is_r500 && fs_inputs->face != ATTR_UNUSED &&
+        !(any_bcolor_used && r300->two_sided_color)) {
+        rX00_rs_col(&rs, col_count, col_count, SWIZ_XYZW);
+        rX00_rs_col_write(&rs, col_count, fp_offset, WRITE_FACE);
+        fp_offset++;
+        col_count++;
+        DBG(r300, DBG_RS, "r300: Rasterized FACE written to FS.\n");
+    }
+
     /* Rasterize texture coordinates. */
     for (i = 0; i < ATTR_GENERIC_COUNT && tex_count < 8; i++) {
 	bool sprite_coord = !!(r300->sprite_coord_enable & (1 << i));
@@ -412,7 +447,7 @@ static void r300_update_rs_block(struct r300_context *r300)
             }
 
             /* Rasterize it. */
-            rX00_rs_tex(&rs, tex_count, tex_count,
+            rX00_rs_tex(&rs, tex_count, tex_ptr,
 			sprite_coord ? SWIZ_XY01 : SWIZ_XYZW);
 
             /* Write it to the FS input register if it's needed by the FS. */
@@ -429,6 +464,7 @@ static void r300_update_rs_block(struct r300_context *r300)
                     i, sprite_coord ? " (sprite coord)" : "");
             }
             tex_count++;
+            tex_ptr += sprite_coord ? 2 : 4;
         } else {
             /* Skip the FS input register, leave it uninitialized. */
             /* If we try to set it to (0,0,0,1), it will lock up. */
@@ -449,7 +485,7 @@ static void r300_update_rs_block(struct r300_context *r300)
         stream_loc_notcl[loc++] = 6 + tex_count;
 
         /* Rasterize it. */
-        rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_X001);
+        rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_X001);
 
         /* Write it to the FS input register if it's needed by the FS. */
         if (fs_inputs->fog != ATTR_UNUSED) {
@@ -461,6 +497,7 @@ static void r300_update_rs_block(struct r300_context *r300)
             DBG(r300, DBG_RS, "r300: Rasterized fog unused.\n");
         }
         tex_count++;
+        tex_ptr += 4;
     } else {
         /* Skip the FS input register, leave it uninitialized. */
         /* If we try to set it to (0,0,0,1), it will lock up. */
@@ -480,7 +517,7 @@ static void r300_update_rs_block(struct r300_context *r300)
         stream_loc_notcl[loc++] = 6 + tex_count;
 
         /* Rasterize it. */
-        rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW);
+        rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW);
 
         /* Write it to the FS input register. */
         rX00_rs_tex_write(&rs, tex_count, fp_offset);
@@ -489,6 +526,7 @@ static void r300_update_rs_block(struct r300_context *r300)
 
         fp_offset++;
         tex_count++;
+        tex_ptr += 4;
     }
 
     /* Invalidate the rest of the no-TCL (GA) stream locations. */
@@ -507,7 +545,7 @@ static void r300_update_rs_block(struct r300_context *r300)
     DBG(r300, DBG_RS, "r300: --- Rasterizer status ---: colors: %i, "
         "generics: %i.\n", col_count, tex_count);
 
-    rs.count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
+    rs.count = MIN2(tex_ptr, 32) | (col_count << R300_IC_COUNT_SHIFT) |
         R300_HIRES_EN;
 
     count = MAX3(col_count, tex_count, 1);
@@ -528,15 +566,9 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
     struct r300_sampler_state *sampler;
     struct r300_sampler_view *view;
     struct r300_texture *tex;
-    unsigned min_level, max_level, i, size;
+    unsigned min_level, max_level, i, j, size;
     unsigned count = MIN2(state->sampler_view_count,
                           state->sampler_state_count);
-    unsigned char depth_swizzle[4] = {
-        UTIL_FORMAT_SWIZZLE_X,
-        UTIL_FORMAT_SWIZZLE_X,
-        UTIL_FORMAT_SWIZZLE_X,
-        UTIL_FORMAT_SWIZZLE_X
-    };
 
     /* The KIL opcode fix, see below. */
     if (!count && !r300->screen->caps.is_r500)
@@ -563,14 +595,29 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
             /* Assign a texture cache region. */
             texstate->format.format1 |= view->texcache_region;
 
-            /* If compare mode is disabled, the sampler view swizzles
-             * are stored in the format.
-             * Otherwise, swizzles must be applied after the compare mode
-             * in the fragment shader. */
-            if (util_format_is_depth_or_stencil(tex->b.b.format)) {
+            /* Depth textures are kinda special. */
+            if (util_format_is_depth_or_stencil(tex->desc.b.b.format)) {
+                unsigned char depth_swizzle[4];
+
+                if (!r300->screen->caps.is_r500 &&
+                    util_format_get_blocksizebits(tex->desc.b.b.format) == 32) {
+                    /* X24x8 is sampled as Y16X16 on r3xx-r4xx.
+                     * The depth here is at the Y component. */
+                    for (j = 0; j < 4; j++)
+                        depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_Y;
+                } else {
+                    for (j = 0; j < 4; j++)
+                        depth_swizzle[j] = UTIL_FORMAT_SWIZZLE_X;
+                }
+
+                /* If compare mode is disabled, sampler view swizzles
+                 * are stored in the format.
+                 * Otherwise, the swizzles must be applied after the compare
+                 * mode in the fragment shader. */
                 if (sampler->state.compare_mode == PIPE_TEX_COMPARE_NONE) {
                     texstate->format.format1 |=
-                        r300_get_swizzle_combined(depth_swizzle, view->swizzle);
+                        r300_get_swizzle_combined(depth_swizzle,
+                                                  view->swizzle);
                 } else {
                     texstate->format.format1 |=
                         r300_get_swizzle_combined(depth_swizzle, 0);
@@ -578,12 +625,12 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
             }
 
             /* to emulate 1D textures through 2D ones correctly */
-            if (tex->b.b.target == PIPE_TEXTURE_1D) {
+            if (tex->desc.b.b.target == PIPE_TEXTURE_1D) {
                 texstate->filter0 &= ~R300_TX_WRAP_T_MASK;
                 texstate->filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
             }
 
-            if (tex->uses_pitch) {
+            if (tex->desc.is_npot) {
                 /* NPOT textures don't support mip filter, unfortunately.
                  * This prevents incorrect rendering. */
                 texstate->filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK;
@@ -610,7 +657,7 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
                 /* determine min/max levels */
                 /* the MAX_MIP level is the largest (finest) one */
                 max_level = MIN3(sampler->max_lod + view->base.first_level,
-                                 tex->b.b.last_level, view->base.last_level);
+                                 tex->desc.b.b.last_level, view->base.last_level);
                 min_level = MIN2(sampler->min_lod + view->base.first_level,
                                  max_level);
                 texstate->format.format0 |= R300_TX_NUM_LEVELS(max_level);
@@ -665,8 +712,44 @@ static void r300_merge_textures_and_samplers(struct r300_context* r300)
     }
 }
 
+/* We can't use compressed zbuffers as samplers. */
+static void r300_flush_depth_textures(struct r300_context *r300)
+{
+    struct r300_textures_state *state =
+        (struct r300_textures_state*)r300->textures_state.state;
+    unsigned i, level;
+    unsigned count = MIN2(state->sampler_view_count,
+                          state->sampler_state_count);
+
+    if (r300->z_decomp_rd)
+        return;
+
+    for (i = 0; i < count; i++)
+        if (state->sampler_views[i] && state->sampler_states[i]) {
+            struct pipe_resource *tex = state->sampler_views[i]->base.texture;
+
+            if (tex->target == PIPE_TEXTURE_3D ||
+                tex->target == PIPE_TEXTURE_CUBE)
+                continue;
+
+            /* Ignore non-depth textures.
+             * Also ignore reinterpreted depth textures, e.g. resource_copy. */
+            if (!util_format_is_depth_or_stencil(tex->format))
+                continue;
+
+            for (level = 0; level <= tex->last_level; level++)
+                if (r300_texture(tex)->zmask_in_use[level]) {
+                    /* We don't handle 3D textures and cubemaps yet. */
+                    r300_flush_depth_stencil(&r300->context, tex,
+                                             u_subresource(0, level), 0);
+                }
+        }
+}
+
 void r300_update_derived_state(struct r300_context* r300)
 {
+    r300_flush_depth_textures(r300);
+
     if (r300->textures_state.dirty) {
         r300_merge_textures_and_samplers(r300);
     }
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index 5750bc4329..da8eadd3b5 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -26,6 +26,7 @@
 
 #include "r300_context.h"
 #include "r300_reg.h"
+#include "r300_texture_desc.h"
 #include "r300_transfer.h"
 #include "r300_screen.h"
 #include "r300_winsys.h"
@@ -34,14 +35,10 @@
 #include "util/u_format_s3tc.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_mm.h"
 
 #include "pipe/p_screen.h"
 
-enum r300_dim {
-    DIM_WIDTH  = 0,
-    DIM_HEIGHT = 1
-};
-
 unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
                                    const unsigned char *swizzle_view)
 {
@@ -109,7 +106,8 @@ unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
  * The FORMAT specifies how the texture sampler will treat the texture, and
  * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */
 uint32_t r300_translate_texformat(enum pipe_format format,
-                                  const unsigned char *swizzle_view)
+                                  const unsigned char *swizzle_view,
+                                  boolean is_r500)
 {
     uint32_t result = 0;
     const struct util_format_description *desc;
@@ -134,7 +132,10 @@ uint32_t r300_translate_texformat(enum pipe_format format,
                     return R300_TX_FORMAT_X16;
                 case PIPE_FORMAT_X8Z24_UNORM:
                 case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
-                    return R500_TX_FORMAT_Y8X24;
+                    if (is_r500)
+                        return R500_TX_FORMAT_Y8X24;
+                    else
+                        return R300_TX_FORMAT_Y16X16;
                 default:
                     return ~0; /* Unsupported. */
             }
@@ -537,26 +538,27 @@ boolean r300_is_zs_format_supported(enum pipe_format format)
 
 boolean r300_is_sampler_format_supported(enum pipe_format format)
 {
-    return r300_translate_texformat(format, 0) != ~0;
+    return r300_translate_texformat(format, 0, TRUE) != ~0;
 }
 
 static void r300_texture_setup_immutable_state(struct r300_screen* screen,
                                                struct r300_texture* tex)
 {
     struct r300_texture_format_state* f = &tex->tx_format;
-    struct pipe_resource *pt = &tex->b.b;
+    struct pipe_resource *pt = &tex->desc.b.b;
     boolean is_r500 = screen->caps.is_r500;
 
     /* Set sampler state. */
     f->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) |
                  R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff);
 
-    if (tex->uses_pitch) {
+    if (tex->desc.uses_stride_addressing) {
         /* rectangles love this */
         f->format0 |= R300_TX_PITCH_EN;
-        f->format2 = (tex->hwpitch[0] - 1) & 0x1fff;
+        f->format2 = (tex->desc.stride_in_pixels[0] - 1) & 0x1fff;
     } else {
-        /* power of two textures (3D, mipmaps, and no pitch) */
+        /* Power of two textures (3D, mipmaps, and no pitch),
+         * also NPOT textures with a width being POT. */
         f->format0 |= R300_TX_DEPTH(util_logbase2(pt->depth0) & 0xf);
     }
 
@@ -579,8 +581,8 @@ static void r300_texture_setup_immutable_state(struct r300_screen* screen,
         }
     }
 
-    f->tile_config = R300_TXO_MACRO_TILE(tex->macrotile) |
-                     R300_TXO_MICRO_TILE(tex->microtile);
+    f->tile_config = R300_TXO_MACRO_TILE(tex->desc.macrotile[0]) |
+                     R300_TXO_MICRO_TILE(tex->desc.microtile);
 }
 
 static void r300_texture_setup_fb_state(struct r300_screen* screen,
@@ -589,23 +591,23 @@ static void r300_texture_setup_fb_state(struct r300_screen* screen,
     unsigned i;
 
     /* Set framebuffer state. */
-    if (util_format_is_depth_or_stencil(tex->b.b.format)) {
-        for (i = 0; i <= tex->b.b.last_level; i++) {
+    if (util_format_is_depth_or_stencil(tex->desc.b.b.format)) {
+        for (i = 0; i <= tex->desc.b.b.last_level; i++) {
             tex->fb_state.pitch[i] =
-                tex->hwpitch[i] |
-                R300_DEPTHMACROTILE(tex->mip_macrotile[i]) |
-                R300_DEPTHMICROTILE(tex->microtile);
+                tex->desc.stride_in_pixels[i] |
+                R300_DEPTHMACROTILE(tex->desc.macrotile[i]) |
+                R300_DEPTHMICROTILE(tex->desc.microtile);
         }
-        tex->fb_state.format = r300_translate_zsformat(tex->b.b.format);
+        tex->fb_state.format = r300_translate_zsformat(tex->desc.b.b.format);
     } else {
-        for (i = 0; i <= tex->b.b.last_level; i++) {
+        for (i = 0; i <= tex->desc.b.b.last_level; i++) {
             tex->fb_state.pitch[i] =
-                tex->hwpitch[i] |
-                r300_translate_colorformat(tex->b.b.format) |
-                R300_COLOR_TILE(tex->mip_macrotile[i]) |
-                R300_COLOR_MICROTILE(tex->microtile);
+                tex->desc.stride_in_pixels[i] |
+                r300_translate_colorformat(tex->desc.b.b.format) |
+                R300_COLOR_TILE(tex->desc.macrotile[i]) |
+                R300_COLOR_MICROTILE(tex->desc.microtile);
         }
-        tex->fb_state.format = r300_translate_out_fmt(tex->b.b.format);
+        tex->fb_state.format = r300_translate_out_fmt(tex->desc.b.b.format);
     }
 }
 
@@ -625,282 +627,6 @@ void r300_texture_reinterpret_format(struct pipe_screen *screen,
     r300_texture_setup_fb_state(r300_screen(screen), r300_texture(tex));
 }
 
-unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
-                                 unsigned zslice, unsigned face)
-{
-    unsigned offset = tex->offset[level];
-
-    switch (tex->b.b.target) {
-        case PIPE_TEXTURE_3D:
-            assert(face == 0);
-            return offset + zslice * tex->layer_size[level];
-
-        case PIPE_TEXTURE_CUBE:
-            assert(zslice == 0);
-            return offset + face * tex->layer_size[level];
-
-        default:
-            assert(zslice == 0 && face == 0);
-            return offset;
-    }
-}
-
-/* Returns the number of pixels that the texture should be aligned to
- * in the given dimension. */
-static unsigned r300_get_pixel_alignment(struct r300_texture *tex,
-                                         enum r300_buffer_tiling macrotile,
-                                         enum r300_dim dim)
-{
-    static const unsigned table[2][5][3][2] =
-    {
-        {
-    /* Macro: linear    linear    linear
-       Micro: linear    tiled  square-tiled */
-            {{ 32, 1}, { 8,  4}, { 0,  0}}, /*   8 bits per pixel */
-            {{ 16, 1}, { 8,  2}, { 4,  4}}, /*  16 bits per pixel */
-            {{  8, 1}, { 4,  2}, { 0,  0}}, /*  32 bits per pixel */
-            {{  4, 1}, { 0,  0}, { 2,  2}}, /*  64 bits per pixel */
-            {{  2, 1}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
-        },
-        {
-    /* Macro: tiled     tiled     tiled
-       Micro: linear    tiled  square-tiled */
-            {{256, 8}, {64, 32}, { 0,  0}}, /*   8 bits per pixel */
-            {{128, 8}, {64, 16}, {32, 32}}, /*  16 bits per pixel */
-            {{ 64, 8}, {32, 16}, { 0,  0}}, /*  32 bits per pixel */
-            {{ 32, 8}, { 0,  0}, {16, 16}}, /*  64 bits per pixel */
-            {{ 16, 8}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
-        }
-    };
-    static const unsigned aa_block[2] = {4, 8};
-    unsigned res = 0;
-    unsigned pixsize = util_format_get_blocksize(tex->b.b.format);
-
-    assert(macrotile <= R300_BUFFER_TILED);
-    assert(tex->microtile <= R300_BUFFER_SQUARETILED);
-    assert(pixsize <= 16);
-    assert(dim <= DIM_HEIGHT);
-
-    if (tex->b.b.nr_samples > 1) {
-        /* Multisampled textures have their own alignment scheme. */
-        if (pixsize == 4)
-            res = aa_block[dim];
-    } else {
-        /* Standard alignment. */
-        res = table[macrotile][util_logbase2(pixsize)][tex->microtile][dim];
-    }
-
-    assert(res);
-    return res;
-}
-
-/* Return true if macrotiling should be enabled on the miplevel. */
-static boolean r300_texture_macro_switch(struct r300_texture *tex,
-                                         unsigned level,
-                                         boolean rv350_mode,
-                                         enum r300_dim dim)
-{
-    unsigned tile, texdim;
-
-    tile = r300_get_pixel_alignment(tex, R300_BUFFER_TILED, dim);
-    if (dim == DIM_WIDTH) {
-        texdim = u_minify(tex->b.b.width0, level);
-    } else {
-        texdim = u_minify(tex->b.b.height0, level);
-    }
-
-    /* See TX_FILTER1_n.MACRO_SWITCH. */
-    if (rv350_mode) {
-        return texdim >= tile;
-    } else {
-        return texdim > tile;
-    }
-}
-
-/**
- * Return the stride, in bytes, of the texture images of the given texture
- * at the given level.
- */
-unsigned r300_texture_get_stride(struct r300_screen* screen,
-                                 struct r300_texture* tex, unsigned level)
-{
-    unsigned tile_width, width, stride;
-
-    if (tex->stride_override)
-        return tex->stride_override;
-
-    /* Check the level. */
-    if (level > tex->b.b.last_level) {
-        SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n",
-                   __FUNCTION__, level, tex->b.b.last_level);
-        return 0;
-    }
-
-    width = u_minify(tex->b.b.width0, level);
-
-    if (util_format_is_plain(tex->b.b.format)) {
-        tile_width = r300_get_pixel_alignment(tex, tex->mip_macrotile[level],
-                                              DIM_WIDTH);
-        width = align(width, tile_width);
-
-        stride = util_format_get_stride(tex->b.b.format, width);
-
-        /* Some IGPs need a minimum stride of 64 bytes, hmm...
-         * This doesn't seem to apply to tiled textures, according to r300c. */
-        if (!tex->microtile && !tex->mip_macrotile[level] &&
-            (screen->caps.family == CHIP_FAMILY_RS600 ||
-             screen->caps.family == CHIP_FAMILY_RS690 ||
-             screen->caps.family == CHIP_FAMILY_RS740)) {
-            return stride < 64 ? 64 : stride;
-        }
-
-        /* The alignment to 32 bytes is sort of implied by the layout... */
-        return stride;
-    } else {
-        return align(util_format_get_stride(tex->b.b.format, width), 32);
-    }
-}
-
-static unsigned r300_texture_get_nblocksy(struct r300_texture* tex,
-                                          unsigned level)
-{
-    unsigned height, tile_height;
-
-    height = u_minify(tex->b.b.height0, level);
-
-    if (util_format_is_plain(tex->b.b.format)) {
-        tile_height = r300_get_pixel_alignment(tex, tex->mip_macrotile[level],
-                                               DIM_HEIGHT);
-        height = align(height, tile_height);
-
-        /* This is needed for the kernel checker, unfortunately. */
-        height = util_next_power_of_two(height);
-    }
-
-    return util_format_get_nblocksy(tex->b.b.format, height);
-}
-
-static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen,
-                                           struct r300_texture *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.target == PIPE_TEXTURE_3D &&
-        tex->b.b.last_level > 0) {
-        size = 0;
-
-        for (i = 0; i <= tex->b.b.last_level; i++) {
-            size += r300_texture_get_stride(screen, tex, i) *
-                    r300_texture_get_nblocksy(tex, i);
-        }
-
-        size *= tex->b.b.depth0;
-        tex->size = size;
-    }
-}
-
-static void r300_setup_miptree(struct r300_screen* screen,
-                               struct r300_texture* tex)
-{
-    struct pipe_resource* base = &tex->b.b;
-    unsigned stride, size, layer_size, nblocksy, i;
-    boolean rv350_mode = screen->caps.is_rv350;
-
-    SCREEN_DBG(screen, DBG_TEXALLOC,
-        "r300: Making miptree for texture, format %s\n",
-        util_format_short_name(base->format));
-
-    for (i = 0; i <= base->last_level; i++) {
-        /* Let's see if this miplevel can be macrotiled. */
-        tex->mip_macrotile[i] =
-            (tex->macrotile == R300_BUFFER_TILED &&
-             r300_texture_macro_switch(tex, i, rv350_mode, DIM_WIDTH) &&
-             r300_texture_macro_switch(tex, i, rv350_mode, DIM_HEIGHT)) ?
-             R300_BUFFER_TILED : R300_BUFFER_LINEAR;
-
-        stride = r300_texture_get_stride(screen, tex, i);
-        nblocksy = r300_texture_get_nblocksy(tex, i);
-        layer_size = stride * nblocksy;
-
-        if (base->nr_samples) {
-            layer_size *= base->nr_samples;
-        }
-
-        if (base->target == PIPE_TEXTURE_CUBE)
-            size = layer_size * 6;
-        else
-            size = layer_size * u_minify(base->depth0, i);
-
-        tex->offset[i] = tex->size;
-        tex->size = tex->offset[i] + size;
-        tex->layer_size[i] = layer_size;
-        tex->pitch[i] = stride / util_format_get_blocksize(base->format);
-        tex->hwpitch[i] =
-                tex->pitch[i] * util_format_get_blockwidth(base->format);
-
-        SCREEN_DBG(screen, DBG_TEXALLOC, "r300: Texture miptree: Level %d "
-                "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n",
-                i, u_minify(base->width0, i), u_minify(base->height0, i),
-                u_minify(base->depth0, i), stride, tex->size,
-                tex->mip_macrotile[i] ? "TRUE" : "FALSE");
-    }
-}
-
-static void r300_setup_flags(struct r300_texture* tex)
-{
-    tex->uses_pitch = !util_is_power_of_two(tex->b.b.width0) ||
-                      !util_is_power_of_two(tex->b.b.height0) ||
-                      tex->stride_override;
-}
-
-static void r300_setup_tiling(struct pipe_screen *screen,
-                              struct r300_texture *tex)
-{
-    struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys;
-    enum pipe_format format = tex->b.b.format;
-    boolean rv350_mode = r300_screen(screen)->caps.is_rv350;
-    boolean is_zb = util_format_is_depth_or_stencil(format);
-    boolean dbg_no_tiling = SCREEN_DBG_ON(r300_screen(screen), DBG_NO_TILING);
-
-    if (!util_format_is_plain(format)) {
-        return;
-    }
-
-    /* If height == 1, disable microtiling except for zbuffer. */
-    if (!is_zb && (tex->b.b.height0 == 1 || dbg_no_tiling)) {
-        return;
-    }
-
-    /* Set microtiling. */
-    switch (util_format_get_blocksize(format)) {
-        case 1:
-        case 4:
-            tex->microtile = R300_BUFFER_TILED;
-            break;
-
-        case 2:
-        case 8:
-            if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
-                tex->microtile = R300_BUFFER_SQUARETILED;
-            }
-            break;
-    }
-
-    if (dbg_no_tiling) {
-        return;
-    }
-
-    /* Set macrotiling. */
-    if (r300_texture_macro_switch(tex, 0, rv350_mode, DIM_WIDTH) &&
-        r300_texture_macro_switch(tex, 0, rv350_mode, DIM_HEIGHT)) {
-        tex->macrotile = R300_BUFFER_TILED;
-    }
-}
-
 static unsigned r300_texture_is_referenced(struct pipe_context *context,
 					 struct pipe_resource *texture,
 					 unsigned face, unsigned level)
@@ -920,8 +646,16 @@ static void r300_texture_destroy(struct pipe_screen *screen,
 {
     struct r300_texture* tex = (struct r300_texture*)texture;
     struct r300_winsys_screen *rws = (struct r300_winsys_screen *)texture->screen->winsys;
+    int i;
 
     rws->buffer_reference(rws, &tex->buffer, NULL);
+    for (i = 0; i < R300_MAX_TEXTURE_LEVELS; i++) {
+        if (tex->hiz_mem[i])
+            u_mmFreeMem(tex->hiz_mem[i]);
+        if (tex->zmask_mem[i])
+            u_mmFreeMem(tex->zmask_mem[i]);
+    }
+
     FREE(tex);
 }
 
@@ -937,11 +671,10 @@ static boolean r300_texture_get_handle(struct pipe_screen* screen,
     }
 
     return rws->buffer_get_handle(rws, tex->buffer,
-                    r300_texture_get_stride(r300_screen(screen), tex, 0),
-                    whandle);
+                                  tex->desc.stride_in_bytes[0], whandle);
 }
 
-struct u_resource_vtbl r300_texture_vtbl = 
+struct u_resource_vtbl r300_texture_vtbl =
 {
    r300_texture_get_handle,	      /* get_handle */
    r300_texture_destroy,	      /* resource_destroy */
@@ -954,17 +687,69 @@ struct u_resource_vtbl r300_texture_vtbl =
    u_default_transfer_inline_write    /* transfer_inline_write */
 };
 
-/* Create a new texture. */
-struct pipe_resource* r300_texture_create(struct pipe_screen* screen,
-                                          const struct pipe_resource* base)
+/* The common texture constructor. */
+static struct r300_texture*
+r300_texture_create_object(struct r300_screen *rscreen,
+                           const struct pipe_resource *base,
+                           enum r300_buffer_tiling microtile,
+                           enum r300_buffer_tiling macrotile,
+                           unsigned stride_in_bytes_override,
+                           unsigned max_buffer_size,
+                           struct r300_winsys_buffer *buffer)
 {
-    struct r300_texture* tex = CALLOC_STRUCT(r300_texture);
-    struct r300_screen* rscreen = r300_screen(screen);
-    struct r300_winsys_screen *rws = (struct r300_winsys_screen *)screen->winsys;
-
+    struct r300_winsys_screen *rws = rscreen->rws;
+    struct r300_texture *tex = CALLOC_STRUCT(r300_texture);
     if (!tex) {
+        if (buffer)
+            rws->buffer_reference(rws, &buffer, NULL);
+        return NULL;
+    }
+
+    /* Initialize the descriptor. */
+    if (!r300_texture_desc_init(rscreen, &tex->desc, base,
+                                microtile, macrotile,
+                                stride_in_bytes_override,
+                                max_buffer_size)) {
+        if (buffer)
+            rws->buffer_reference(rws, &buffer, NULL);
+        FREE(tex);
         return NULL;
     }
+    /* Initialize the hardware state. */
+    r300_texture_setup_immutable_state(rscreen, tex);
+    r300_texture_setup_fb_state(rscreen, tex);
+
+    tex->desc.b.vtbl = &r300_texture_vtbl;
+    pipe_reference_init(&tex->desc.b.b.reference, 1);
+    tex->domain = base->flags & R300_RESOURCE_FLAG_TRANSFER ?
+                  R300_DOMAIN_GTT :
+                  R300_DOMAIN_VRAM | R300_DOMAIN_GTT;
+    tex->buffer = buffer;
+
+    /* Create the backing buffer if needed. */
+    if (!tex->buffer) {
+        tex->buffer = rws->buffer_create(rws, tex->desc.size_in_bytes, 2048,
+                                         base->bind, base->usage, tex->domain);
+
+        if (!tex->buffer) {
+            FREE(tex);
+            return NULL;
+        }
+    }
+
+    rws->buffer_set_tiling(rws, tex->buffer,
+            tex->desc.microtile, tex->desc.macrotile[0],
+            tex->desc.stride_in_bytes[0]);
+
+    return tex;
+}
+
+/* Create a new texture. */
+struct pipe_resource *r300_texture_create(struct pipe_screen *screen,
+                                          const struct pipe_resource *base)
+{
+    struct r300_screen *rscreen = r300_screen(screen);
+    enum r300_buffer_tiling microtile, macrotile;
 
     /* Refuse to create a texture with size 0. */
     if (!base->width0 ||
@@ -974,58 +759,70 @@ struct pipe_resource* r300_texture_create(struct pipe_screen* screen,
         fprintf(stderr, "r300: texture_create: "
                 "Got invalid texture dimensions: %ix%ix%i\n",
                 base->width0, base->height0, base->depth0);
-        FREE(tex);
         return NULL;
     }
 
-    tex->b.b = *base;
-    tex->b.vtbl = &r300_texture_vtbl;
-    pipe_reference_init(&tex->b.b.reference, 1);
-    tex->b.b.screen = screen;
+    if ((base->flags & R300_RESOURCE_FLAG_TRANSFER) ||
+        (base->bind & PIPE_BIND_SCANOUT)) {
+        microtile = R300_BUFFER_LINEAR;
+        macrotile = R300_BUFFER_LINEAR;
+    } else {
+        microtile = R300_BUFFER_SELECT_LAYOUT;
+        macrotile = R300_BUFFER_SELECT_LAYOUT;
+    }
+
+    return (struct pipe_resource*)
+           r300_texture_create_object(rscreen, base, microtile, macrotile,
+                                      0, 0, NULL);
+}
+
+struct pipe_resource *r300_texture_from_handle(struct pipe_screen *screen,
+                                               const struct pipe_resource *base,
+                                               struct winsys_handle *whandle)
+{
+    struct r300_winsys_screen *rws = (struct r300_winsys_screen*)screen->winsys;
+    struct r300_screen *rscreen = r300_screen(screen);
+    struct r300_winsys_buffer *buffer;
+    enum r300_buffer_tiling microtile, macrotile;
+    unsigned stride, size;
 
-    r300_setup_flags(tex);
-    if (!(base->flags & R300_RESOURCE_FLAG_TRANSFER) &&
-        !(base->bind & PIPE_BIND_SCANOUT)) {
-        r300_setup_tiling(screen, tex);
+    /* Support only 2D textures without mipmaps */
+    if (base->target != PIPE_TEXTURE_2D ||
+        base->depth0 != 1 ||
+        base->last_level != 0) {
+        return NULL;
     }
-    r300_setup_miptree(rscreen, tex);
-    r300_texture_3d_fix_mipmapping(rscreen, tex);
-    r300_texture_setup_immutable_state(rscreen, tex);
-    r300_texture_setup_fb_state(rscreen, tex);
 
-    SCREEN_DBG(rscreen, DBG_TEX,
-               "r300: texture_create: Macro: %s, Micro: %s, Pitch: %i, "
-               "Dim: %ix%ix%i, LastLevel: %i, Size: %i, Format: %s\n",
-               tex->macrotile ? "YES" : " NO",
-               tex->microtile ? "YES" : " NO",
-               tex->hwpitch[0],
-               base->width0, base->height0, base->depth0, base->last_level,
-               tex->size,
-               util_format_short_name(base->format));
+    buffer = rws->buffer_from_handle(rws, whandle, &stride, &size);
+    if (!buffer)
+        return NULL;
 
-    tex->domain = base->flags & R300_RESOURCE_FLAG_TRANSFER ?
-                  R300_DOMAIN_GTT :
-                  R300_DOMAIN_VRAM | R300_DOMAIN_GTT;
+    rws->buffer_get_tiling(rws, buffer, &microtile, &macrotile);
 
-    tex->buffer = rws->buffer_create(rws, tex->size, 2048, base->bind,
-                                     base->usage, tex->domain);
+    /* Enforce a microtiled zbuffer. */
+    if (util_format_is_depth_or_stencil(base->format) &&
+        microtile == R300_BUFFER_LINEAR) {
+        switch (util_format_get_blocksize(base->format)) {
+            case 4:
+                microtile = R300_BUFFER_TILED;
+                break;
 
-    if (!tex->buffer) {
-	FREE(tex);
-	return NULL;
+            case 2:
+                if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT))
+                    microtile = R300_BUFFER_SQUARETILED;
+                break;
+        }
     }
 
-    rws->buffer_set_tiling(rws, tex->buffer,
-            tex->microtile, tex->macrotile,
-            tex->pitch[0] * util_format_get_blocksize(tex->b.b.format));
-
-    return (struct pipe_resource*)tex;
+    return (struct pipe_resource*)
+           r300_texture_create_object(rscreen, base, microtile, macrotile,
+                                      stride, size, buffer);
 }
 
 /* Not required to implement u_resource_vtbl, consider moving to another file:
  */
 struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
-					  struct pipe_resource* texture,
+                                          struct pipe_resource* texture,
 					  unsigned face,
 					  unsigned level,
 					  unsigned zslice,
@@ -1035,7 +832,7 @@ struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
     struct r300_surface* surface = CALLOC_STRUCT(r300_surface);
 
     if (surface) {
-        uint32_t stride, offset, tile_height;
+        uint32_t offset, tile_height;
 
         pipe_reference_init(&surface->base.reference, 1);
         pipe_resource_reference(&surface->base.texture, texture);
@@ -1054,23 +851,29 @@ struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
         if (surface->domain & R300_DOMAIN_VRAM)
             surface->domain &= ~R300_DOMAIN_GTT;
 
-        surface->offset = r300_texture_get_offset(tex, level, zslice, face);
+        surface->offset = r300_texture_get_offset(&tex->desc,
+                                                  level, zslice, face);
         surface->pitch = tex->fb_state.pitch[level];
         surface->format = tex->fb_state.format;
 
         /* Parameters for the CBZB clear. */
+        surface->cbzb_allowed = tex->desc.cbzb_allowed[level];
         surface->cbzb_width = align(surface->base.width, 64);
 
         /* Height must be aligned to the size of a tile. */
-        tile_height = r300_get_pixel_alignment(tex, tex->mip_macrotile[level],
+        tile_height = r300_get_pixel_alignment(tex->desc.b.b.format,
+                                               tex->desc.b.b.nr_samples,
+                                               tex->desc.microtile,
+                                               tex->desc.macrotile[level],
                                                DIM_HEIGHT);
+
         surface->cbzb_height = align((surface->base.height + 1) / 2,
                                      tile_height);
 
         /* Offset must be aligned to 2K and must point at the beginning
          * of a scanline. */
-        stride = r300_texture_get_stride(r300_screen(screen), tex, level);
-        offset = surface->offset + stride * surface->cbzb_height;
+        offset = surface->offset +
+                 tex->desc.stride_in_bytes[level] * surface->cbzb_height;
         surface->cbzb_midpoint_offset = offset & ~2047;
 
         surface->cbzb_pitch = surface->pitch & 0x1ffffc;
@@ -1080,11 +883,11 @@ struct pipe_surface* r300_get_tex_surface(struct pipe_screen* screen,
         else
             surface->cbzb_format = R300_DEPTHFORMAT_16BIT_INT_Z;
 
-        SCREEN_DBG(r300_screen(screen), DBG_TEX,
+        SCREEN_DBG(r300_screen(screen), DBG_CBZB,
                    "CBZB Dim: %ix%i, Misalignment: %i, Macro: %s\n",
                    surface->cbzb_width, surface->cbzb_height,
                    offset & 2047,
-                   tex->mip_macrotile[level] ? "YES" : " NO");
+                   tex->desc.macrotile[level] ? "YES" : " NO");
     }
 
     return &surface->base;
@@ -1097,88 +900,3 @@ void r300_tex_surface_destroy(struct pipe_surface* s)
     pipe_resource_reference(&s->texture, NULL);
     FREE(s);
 }
-
-struct pipe_resource*
-r300_texture_from_handle(struct pipe_screen* screen,
-			  const struct pipe_resource* base,
-			  struct winsys_handle *whandle)
-{
-    struct r300_winsys_screen *rws = (struct r300_winsys_screen*)screen->winsys;
-    struct r300_screen* rscreen = r300_screen(screen);
-    struct r300_winsys_buffer *buffer;
-    struct r300_texture* tex;
-    unsigned stride;
-    boolean override_zb_flags;
-
-    /* Support only 2D textures without mipmaps */
-    if (base->target != PIPE_TEXTURE_2D ||
-        base->depth0 != 1 ||
-        base->last_level != 0) {
-        return NULL;
-    }
-
-    buffer = rws->buffer_from_handle(rws, whandle, &stride);
-    if (!buffer) {
-        return NULL;
-    }
-
-    tex = CALLOC_STRUCT(r300_texture);
-    if (!tex) {
-        return NULL;
-    }
-
-    tex->b.b = *base;
-    tex->b.vtbl = &r300_texture_vtbl;
-    pipe_reference_init(&tex->b.b.reference, 1);
-    tex->b.b.screen = screen;
-    tex->domain = R300_DOMAIN_VRAM;
-
-    tex->stride_override = stride;
-
-    /* one ref already taken */
-    tex->buffer = buffer;
-
-    rws->buffer_get_tiling(rws, buffer, &tex->microtile, &tex->macrotile);
-    r300_setup_flags(tex);
-    SCREEN_DBG(rscreen, DBG_TEX,
-               "r300: texture_from_handle: Macro: %s, Micro: %s, "
-               "Pitch: % 4i, Dim: %ix%i, Format: %s\n",
-               tex->macrotile ? "YES" : " NO",
-               tex->microtile ? "YES" : " NO",
-               stride / util_format_get_blocksize(base->format),
-               base->width0, base->height0,
-               util_format_short_name(base->format));
-
-    /* Enforce microtiled zbuffer. */
-    override_zb_flags = util_format_is_depth_or_stencil(base->format) &&
-                        tex->microtile == R300_BUFFER_LINEAR;
-
-    if (override_zb_flags) {
-        switch (util_format_get_blocksize(base->format)) {
-            case 4:
-                tex->microtile = R300_BUFFER_TILED;
-                break;
-
-            case 2:
-                if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
-                    tex->microtile = R300_BUFFER_SQUARETILED;
-                    break;
-                }
-                /* Pass through. */
-
-            default:
-                override_zb_flags = FALSE;
-        }
-    }
-
-    r300_setup_miptree(rscreen, tex);
-    r300_texture_setup_immutable_state(rscreen, tex);
-    r300_texture_setup_fb_state(rscreen, tex);
-
-    if (override_zb_flags) {
-        rws->buffer_set_tiling(rws, tex->buffer,
-                tex->microtile, tex->macrotile,
-                tex->pitch[0] * util_format_get_blocksize(tex->b.b.format));
-    }
-    return (struct pipe_resource*)tex;
-}
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 99e7694254..a4524320fd 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -35,16 +35,11 @@ unsigned r300_get_swizzle_combined(const unsigned char *swizzle_format,
                                    const unsigned char *swizzle_view);
 
 uint32_t r300_translate_texformat(enum pipe_format format,
-                                  const unsigned char *swizzle_view);
+                                  const unsigned char *swizzle_view,
+                                  boolean is_r500);
 
 uint32_t r500_tx_format_msb_bit(enum pipe_format format);
 
-unsigned r300_texture_get_stride(struct r300_screen* screen,
-                                 struct r300_texture* tex, unsigned level);
-
-unsigned r300_texture_get_offset(struct r300_texture* tex, unsigned level,
-                                 unsigned zslice, unsigned face);
-
 void r300_texture_reinterpret_format(struct pipe_screen *screen,
                                      struct pipe_resource *tex,
                                      enum pipe_format new_format);
diff --git a/src/gallium/drivers/r300/r300_texture_desc.c b/src/gallium/drivers/r300/r300_texture_desc.c
new file mode 100644
index 0000000000..5d690e8c33
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_texture_desc.c
@@ -0,0 +1,479 @@
+/*
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
+ *
+ * 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. */
+
+#include "r300_texture_desc.h"
+
+#include "r300_context.h"
+#include "r300_winsys.h"
+
+#include "util/u_format.h"
+
+/* Returns the number of pixels that the texture should be aligned to
+ * in the given dimension. */
+unsigned r300_get_pixel_alignment(enum pipe_format format,
+                                  unsigned num_samples,
+                                  enum r300_buffer_tiling microtile,
+                                  enum r300_buffer_tiling macrotile,
+                                  enum r300_dim dim)
+{
+    static const unsigned table[2][5][3][2] =
+    {
+        {
+    /* Macro: linear    linear    linear
+       Micro: linear    tiled  square-tiled */
+            {{ 32, 1}, { 8,  4}, { 0,  0}}, /*   8 bits per pixel */
+            {{ 16, 1}, { 8,  2}, { 4,  4}}, /*  16 bits per pixel */
+            {{  8, 1}, { 4,  2}, { 0,  0}}, /*  32 bits per pixel */
+            {{  4, 1}, { 0,  0}, { 2,  2}}, /*  64 bits per pixel */
+            {{  2, 1}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
+        },
+        {
+    /* Macro: tiled     tiled     tiled
+       Micro: linear    tiled  square-tiled */
+            {{256, 8}, {64, 32}, { 0,  0}}, /*   8 bits per pixel */
+            {{128, 8}, {64, 16}, {32, 32}}, /*  16 bits per pixel */
+            {{ 64, 8}, {32, 16}, { 0,  0}}, /*  32 bits per pixel */
+            {{ 32, 8}, { 0,  0}, {16, 16}}, /*  64 bits per pixel */
+            {{ 16, 8}, { 0,  0}, { 0,  0}}  /* 128 bits per pixel */
+        }
+    };
+    static const unsigned aa_block[2] = {4, 8};
+    unsigned tile = 0;
+    unsigned pixsize = util_format_get_blocksize(format);
+
+    assert(macrotile <= R300_BUFFER_TILED);
+    assert(microtile <= R300_BUFFER_SQUARETILED);
+    assert(pixsize <= 16);
+    assert(dim <= DIM_HEIGHT);
+
+    if (num_samples > 1) {
+        /* Multisampled textures have their own alignment scheme. */
+        if (pixsize == 4)
+            tile = aa_block[dim];
+        /* XXX FP16 AA. */
+    } else {
+        /* Standard alignment. */
+        tile = table[macrotile][util_logbase2(pixsize)][microtile][dim];
+    }
+
+    assert(tile);
+    return tile;
+}
+
+/* Return true if macrotiling should be enabled on the miplevel. */
+static boolean r300_texture_macro_switch(struct r300_texture_desc *desc,
+                                         unsigned level,
+                                         boolean rv350_mode,
+                                         enum r300_dim dim)
+{
+    unsigned tile, texdim;
+
+    tile = r300_get_pixel_alignment(desc->b.b.format, desc->b.b.nr_samples,
+                                    desc->microtile, R300_BUFFER_TILED, dim);
+    if (dim == DIM_WIDTH) {
+        texdim = u_minify(desc->b.b.width0, level);
+    } else {
+        texdim = u_minify(desc->b.b.height0, level);
+    }
+
+    /* See TX_FILTER1_n.MACRO_SWITCH. */
+    if (rv350_mode) {
+        return texdim >= tile;
+    } else {
+        return texdim > tile;
+    }
+}
+
+/**
+ * Return the stride, in bytes, of the texture image of the given texture
+ * at the given level.
+ */
+static unsigned r300_texture_get_stride(struct r300_screen *screen,
+                                        struct r300_texture_desc *desc,
+                                        unsigned level)
+{
+    unsigned tile_width, width, stride;
+
+    if (desc->stride_in_bytes_override)
+        return desc->stride_in_bytes_override;
+
+    /* Check the level. */
+    if (level > desc->b.b.last_level) {
+        SCREEN_DBG(screen, DBG_TEX, "%s: level (%u) > last_level (%u)\n",
+                   __FUNCTION__, level, desc->b.b.last_level);
+        return 0;
+    }
+
+    width = u_minify(desc->b.b.width0, level);
+
+    if (util_format_is_plain(desc->b.b.format)) {
+        tile_width = r300_get_pixel_alignment(desc->b.b.format,
+                                              desc->b.b.nr_samples,
+                                              desc->microtile,
+                                              desc->macrotile[level],
+                                              DIM_WIDTH);
+        width = align(width, tile_width);
+
+        stride = util_format_get_stride(desc->b.b.format, width);
+
+        /* Some IGPs need a minimum stride of 64 bytes, hmm... */
+        if (!desc->macrotile[level] &&
+            (screen->caps.family == CHIP_FAMILY_RS600 ||
+             screen->caps.family == CHIP_FAMILY_RS690 ||
+             screen->caps.family == CHIP_FAMILY_RS740)) {
+            unsigned min_stride;
+
+            if (desc->microtile) {
+                unsigned tile_height =
+                        r300_get_pixel_alignment(desc->b.b.format,
+                                                 desc->b.b.nr_samples,
+                                                 desc->microtile,
+                                                 desc->macrotile[level],
+                                                 DIM_HEIGHT);
+
+                min_stride = 64 / tile_height;
+            } else {
+                min_stride = 64;
+            }
+
+            return stride < min_stride ? min_stride : stride;
+        }
+
+        /* The alignment to 32 bytes is sort of implied by the layout... */
+        return stride;
+    } else {
+        return align(util_format_get_stride(desc->b.b.format, width), 32);
+    }
+}
+
+static unsigned r300_texture_get_nblocksy(struct r300_texture_desc *desc,
+                                          unsigned level,
+                                          boolean *out_aligned_for_cbzb)
+{
+    unsigned height, tile_height;
+
+    height = u_minify(desc->b.b.height0, level);
+
+    if (util_format_is_plain(desc->b.b.format)) {
+        tile_height = r300_get_pixel_alignment(desc->b.b.format,
+                                               desc->b.b.nr_samples,
+                                               desc->microtile,
+                                               desc->macrotile[level],
+                                               DIM_HEIGHT);
+        height = align(height, tile_height);
+
+        /* This is needed for the kernel checker, unfortunately. */
+        if ((desc->b.b.target != PIPE_TEXTURE_1D &&
+             desc->b.b.target != PIPE_TEXTURE_2D) ||
+            desc->b.b.last_level != 0) {
+            height = util_next_power_of_two(height);
+        }
+
+        /* See if the CBZB clear can be used on the buffer,
+         * taking the texture size into account. */
+        if (out_aligned_for_cbzb) {
+            if (desc->macrotile[level]) {
+                /* When clearing, the layer (width*height) is horizontally split
+                 * into two, and the upper and lower halves are cleared by the CB
+                 * and ZB units, respectively. Therefore, the number of macrotiles
+                 * in the Y direction must be even. */
+
+                /* Align the height so that there is an even number of macrotiles.
+                 * Do so for 3 or more macrotiles in the Y direction. */
+                if (level == 0 && desc->b.b.last_level == 0 &&
+                    (desc->b.b.target == PIPE_TEXTURE_1D ||
+                     desc->b.b.target == PIPE_TEXTURE_2D) &&
+                    height >= tile_height * 3) {
+                    height = align(height, tile_height * 2);
+                }
+
+                *out_aligned_for_cbzb = height % (tile_height * 2) == 0;
+            } else {
+                *out_aligned_for_cbzb = FALSE;
+            }
+        }
+    }
+
+    return util_format_get_nblocksy(desc->b.b.format, height);
+}
+
+static void r300_texture_3d_fix_mipmapping(struct r300_screen *screen,
+                                           struct r300_texture_desc *desc)
+{
+    /* 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) &&
+        desc->b.b.target == PIPE_TEXTURE_3D &&
+        desc->b.b.last_level > 0) {
+        size = 0;
+
+        for (i = 0; i <= desc->b.b.last_level; i++) {
+            size += desc->stride_in_bytes[i] *
+                    r300_texture_get_nblocksy(desc, i, FALSE);
+        }
+
+        size *= desc->b.b.depth0;
+        desc->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)
+{
+    return (stride_in_bytes / util_format_get_blocksize(format)) *
+            util_format_get_blockwidth(format);
+}
+
+static void r300_setup_miptree(struct r300_screen *screen,
+                               struct r300_texture_desc *desc,
+                               boolean align_for_cbzb)
+{
+    struct pipe_resource *base = &desc->b.b;
+    unsigned stride, size, layer_size, nblocksy, i;
+    boolean rv350_mode = screen->caps.is_rv350;
+    boolean aligned_for_cbzb;
+
+    desc->size_in_bytes = 0;
+
+    SCREEN_DBG(screen, DBG_TEXALLOC,
+        "r300: Making miptree for texture, format %s\n",
+        util_format_short_name(base->format));
+
+    for (i = 0; i <= base->last_level; i++) {
+        /* Let's see if this miplevel can be macrotiled. */
+        desc->macrotile[i] =
+            (desc->macrotile[0] == R300_BUFFER_TILED &&
+             r300_texture_macro_switch(desc, i, rv350_mode, DIM_WIDTH) &&
+             r300_texture_macro_switch(desc, i, rv350_mode, DIM_HEIGHT)) ?
+             R300_BUFFER_TILED : R300_BUFFER_LINEAR;
+
+        stride = r300_texture_get_stride(screen, desc, i);
+
+        /* Compute the number of blocks in Y, see if the CBZB clear can be
+         * used on the texture. */
+        aligned_for_cbzb = FALSE;
+        if (align_for_cbzb && desc->cbzb_allowed[i])
+            nblocksy = r300_texture_get_nblocksy(desc, i, &aligned_for_cbzb);
+        else
+            nblocksy = r300_texture_get_nblocksy(desc, i, NULL);
+
+        layer_size = stride * nblocksy;
+
+        if (base->nr_samples) {
+            layer_size *= base->nr_samples;
+        }
+
+        if (base->target == PIPE_TEXTURE_CUBE)
+            size = layer_size * 6;
+        else
+            size = layer_size * u_minify(base->depth0, i);
+
+        desc->offset_in_bytes[i] = desc->size_in_bytes;
+        desc->size_in_bytes = desc->offset_in_bytes[i] + size;
+        desc->layer_size_in_bytes[i] = layer_size;
+        desc->stride_in_bytes[i] = stride;
+        desc->stride_in_pixels[i] = stride_to_width(desc->b.b.format, stride);
+        desc->cbzb_allowed[i] = desc->cbzb_allowed[i] && aligned_for_cbzb;
+
+        SCREEN_DBG(screen, DBG_TEXALLOC, "r300: Texture miptree: Level %d "
+                "(%dx%dx%d px, pitch %d bytes) %d bytes total, macrotiled %s\n",
+                i, u_minify(base->width0, i), u_minify(base->height0, i),
+                u_minify(base->depth0, i), stride, desc->size_in_bytes,
+                desc->macrotile[i] ? "TRUE" : "FALSE");
+    }
+}
+
+static void r300_setup_flags(struct r300_texture_desc *desc)
+{
+    desc->uses_stride_addressing =
+        !util_is_power_of_two(desc->b.b.width0) ||
+        !util_is_power_of_two(desc->b.b.height0) ||
+        (desc->stride_in_bytes_override &&
+         stride_to_width(desc->b.b.format,
+                         desc->stride_in_bytes_override) != desc->b.b.width0);
+
+    desc->is_npot =
+        desc->uses_stride_addressing ||
+        !util_is_power_of_two(desc->b.b.height0);
+}
+
+static void r300_setup_cbzb_flags(struct r300_screen *rscreen,
+                                  struct r300_texture_desc *desc)
+{
+    unsigned i, bpp;
+    boolean first_level_valid;
+
+    bpp = util_format_get_blocksizebits(desc->b.b.format);
+
+    /* 1) The texture must be point-sampled,
+     * 2) The depth must be 16 or 32 bits.
+     * 3) If the midpoint ZB offset is not aligned to 2048, it returns garbage
+     *    with certain texture sizes. Macrotiling ensures the alignment. */
+    first_level_valid = desc->b.b.nr_samples <= 1 &&
+                       (bpp == 16 || bpp == 32) &&
+                       desc->macrotile[0];
+
+    for (i = 0; i <= desc->b.b.last_level; i++)
+        desc->cbzb_allowed[i] = first_level_valid && desc->macrotile[i];
+}
+
+static void r300_setup_tiling(struct r300_screen *screen,
+                              struct r300_texture_desc *desc)
+{
+    struct r300_winsys_screen *rws = screen->rws;
+    enum pipe_format format = desc->b.b.format;
+    boolean rv350_mode = screen->caps.is_rv350;
+    boolean is_zb = util_format_is_depth_or_stencil(format);
+    boolean dbg_no_tiling = SCREEN_DBG_ON(screen, DBG_NO_TILING);
+
+    if (!util_format_is_plain(format)) {
+        return;
+    }
+
+    /* If height == 1, disable microtiling except for zbuffer. */
+    if (!is_zb && (desc->b.b.height0 == 1 || dbg_no_tiling)) {
+        return;
+    }
+
+    /* Set microtiling. */
+    switch (util_format_get_blocksize(format)) {
+        case 1:
+        case 4:
+            desc->microtile = R300_BUFFER_TILED;
+            break;
+
+        case 2:
+        case 8:
+            if (rws->get_value(rws, R300_VID_SQUARE_TILING_SUPPORT)) {
+                desc->microtile = R300_BUFFER_SQUARETILED;
+            }
+            break;
+    }
+
+    if (dbg_no_tiling) {
+        return;
+    }
+
+    /* Set macrotiling. */
+    if (r300_texture_macro_switch(desc, 0, rv350_mode, DIM_WIDTH) &&
+        r300_texture_macro_switch(desc, 0, rv350_mode, DIM_HEIGHT)) {
+        desc->macrotile[0] = R300_BUFFER_TILED;
+    }
+}
+
+static void r300_tex_print_info(struct r300_screen *rscreen,
+                                struct r300_texture_desc *desc,
+                                const char *func)
+{
+    fprintf(stderr,
+            "r300: %s: Macro: %s, Micro: %s, Pitch: %i, Dim: %ix%ix%i, "
+            "LastLevel: %i, Size: %i, Format: %s\n",
+            func,
+            desc->macrotile[0] ? "YES" : " NO",
+            desc->microtile ? "YES" : " NO",
+            desc->stride_in_pixels[0],
+            desc->b.b.width0, desc->b.b.height0, desc->b.b.depth0,
+            desc->b.b.last_level, desc->size_in_bytes,
+            util_format_short_name(desc->b.b.format));
+}
+
+boolean r300_texture_desc_init(struct r300_screen *rscreen,
+                               struct r300_texture_desc *desc,
+                               const struct pipe_resource *base,
+                               enum r300_buffer_tiling microtile,
+                               enum r300_buffer_tiling macrotile,
+                               unsigned stride_in_bytes_override,
+                               unsigned max_buffer_size)
+{
+    desc->b.b = *base;
+    desc->b.b.screen = &rscreen->screen;
+
+    desc->stride_in_bytes_override = stride_in_bytes_override;
+
+    if (microtile == R300_BUFFER_SELECT_LAYOUT ||
+        macrotile == R300_BUFFER_SELECT_LAYOUT) {
+        r300_setup_tiling(rscreen, desc);
+    } else {
+        desc->microtile = microtile;
+        desc->macrotile[0] = macrotile;
+        assert(desc->b.b.last_level == 0);
+    }
+
+    r300_setup_flags(desc);
+    r300_setup_cbzb_flags(rscreen, desc);
+
+    /* Setup the miptree description. */
+    r300_setup_miptree(rscreen, desc, TRUE);
+    /* If the required buffer size is larger the given max size,
+     * try again without the alignment for the CBZB clear. */
+    if (max_buffer_size && desc->size_in_bytes > max_buffer_size) {
+        r300_setup_miptree(rscreen, desc, FALSE);
+    }
+
+    r300_texture_3d_fix_mipmapping(rscreen, desc);
+
+    if (max_buffer_size) {
+        /* Make sure the buffer we got is large enough. */
+        if (desc->size_in_bytes > max_buffer_size) {
+            fprintf(stderr, "r300: texture_from_handle: The buffer is not "
+                            "large enough. Got: %i, Need: %i, Info:\n",
+                            max_buffer_size, desc->size_in_bytes);
+            r300_tex_print_info(rscreen, desc, "texture_from_handle");
+            return FALSE;
+        }
+
+        desc->buffer_size_in_bytes = max_buffer_size;
+    } else {
+        desc->buffer_size_in_bytes = desc->size_in_bytes;
+    }
+
+    if (SCREEN_DBG_ON(rscreen, DBG_TEX))
+        r300_tex_print_info(rscreen, desc, "texture_from_handle");
+
+    return TRUE;
+}
+
+unsigned r300_texture_get_offset(struct r300_texture_desc *desc,
+                                 unsigned level, unsigned zslice,
+                                 unsigned face)
+{
+    unsigned offset = desc->offset_in_bytes[level];
+
+    switch (desc->b.b.target) {
+        case PIPE_TEXTURE_3D:
+            assert(face == 0);
+            return offset + zslice * desc->layer_size_in_bytes[level];
+
+        case PIPE_TEXTURE_CUBE:
+            assert(zslice == 0);
+            return offset + face * desc->layer_size_in_bytes[level];
+
+        default:
+            assert(zslice == 0 && face == 0);
+            return offset;
+    }
+}
diff --git a/src/gallium/drivers/r300/r300_texture_desc.h b/src/gallium/drivers/r300/r300_texture_desc.h
new file mode 100644
index 0000000000..95de66f654
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_texture_desc.h
@@ -0,0 +1,57 @@
+/*
+ * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
+ * Copyright 2010 Marek Olšák <maraeo@gmail.com>
+ *
+ * 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_TEXTURE_DESC_H
+#define R300_TEXTURE_DESC_H
+
+#include "r300_defines.h"
+
+struct pipe_resource;
+struct r300_screen;
+struct r300_texture_desc;
+struct r300_texture;
+
+enum r300_dim {
+    DIM_WIDTH  = 0,
+    DIM_HEIGHT = 1
+};
+
+unsigned r300_get_pixel_alignment(enum pipe_format format,
+                                  unsigned num_samples,
+                                  enum r300_buffer_tiling microtile,
+                                  enum r300_buffer_tiling macrotile,
+                                  enum r300_dim dim);
+
+boolean r300_texture_desc_init(struct r300_screen *rscreen,
+                               struct r300_texture_desc *desc,
+                               const struct pipe_resource *base,
+                               enum r300_buffer_tiling microtile,
+                               enum r300_buffer_tiling macrotile,
+                               unsigned stride_in_bytes_override,
+                               unsigned max_buffer_size);
+
+unsigned r300_texture_get_offset(struct r300_texture_desc *desc,
+                                 unsigned level, unsigned zslice,
+                                 unsigned face);
+
+#endif
diff --git a/src/gallium/drivers/r300/r300_tgsi_to_rc.c b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
index 51b2c55550..dd697b9c37 100644
--- a/src/gallium/drivers/r300/r300_tgsi_to_rc.c
+++ b/src/gallium/drivers/r300/r300_tgsi_to_rc.c
@@ -126,7 +126,7 @@ static unsigned translate_opcode(unsigned opcode)
      /* case TGSI_OPCODE_SAD: return RC_OPCODE_SAD; */
      /* case TGSI_OPCODE_TXF: return RC_OPCODE_TXF; */
      /* case TGSI_OPCODE_TXQ: return RC_OPCODE_TXQ; */
-     /* case TGSI_OPCODE_CONT: return RC_OPCODE_CONT; */
+        case TGSI_OPCODE_CONT: return RC_OPCODE_CONT;
      /* case TGSI_OPCODE_EMIT: return RC_OPCODE_EMIT; */
      /* case TGSI_OPCODE_ENDPRIM: return RC_OPCODE_ENDPRIM; */
      /* case TGSI_OPCODE_BGNLOOP2: return RC_OPCODE_BGNLOOP2; */
diff --git a/src/gallium/drivers/r300/r300_transfer.c b/src/gallium/drivers/r300/r300_transfer.c
index 3cc4c8c958..e9333b35ef 100644
--- a/src/gallium/drivers/r300/r300_transfer.c
+++ b/src/gallium/drivers/r300/r300_transfer.c
@@ -22,7 +22,7 @@
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
 #include "r300_transfer.h"
-#include "r300_texture.h"
+#include "r300_texture_desc.h"
 #include "r300_screen_buffer.h"
 
 #include "util/u_memory.h"
@@ -35,8 +35,8 @@ struct r300_transfer {
     /* Offset from start of buffer. */
     unsigned offset;
 
-    /* Detiled texture. */
-    struct r300_texture *detiled_texture;
+    /* Linear texture. */
+    struct r300_texture *linear_texture;
 };
 
 /* Convenience cast wrapper. */
@@ -57,7 +57,7 @@ static void r300_copy_from_tiled_texture(struct pipe_context *ctx,
     subdst.face = 0;
     subdst.level = 0;
 
-    ctx->resource_copy_region(ctx, &r300transfer->detiled_texture->b.b, subdst,
+    ctx->resource_copy_region(ctx, &r300transfer->linear_texture->desc.b.b, subdst,
 			      0, 0, 0,
 			      tex, transfer->sr,
 			      transfer->box.x, transfer->box.y, transfer->box.z,
@@ -77,7 +77,7 @@ static void r300_copy_into_tiled_texture(struct pipe_context *ctx,
 
     ctx->resource_copy_region(ctx, tex, transfer->sr,
 			      transfer->box.x, transfer->box.y, transfer->box.z,
-			      &r300transfer->detiled_texture->b.b, subsrc,
+                              &r300transfer->linear_texture->desc.b.b, subsrc,
 			      0, 0, 0,
 			      transfer->box.width, transfer->box.height);
 
@@ -93,7 +93,6 @@ r300_texture_get_transfer(struct pipe_context *ctx,
 {
     struct r300_context *r300 = r300_context(ctx);
     struct r300_texture *tex = r300_texture(texture);
-    struct r300_screen *r300screen = r300_screen(ctx->screen);
     struct r300_transfer *trans;
     struct pipe_resource base;
     boolean referenced_cs, referenced_hw, blittable;
@@ -124,7 +123,7 @@ r300_texture_get_transfer(struct pipe_context *ctx,
         /* If the texture is tiled, we must create a temporary detiled texture
          * for this transfer.
          * Also make write transfers pipelined. */
-        if (tex->microtile || tex->macrotile ||
+        if (tex->desc.microtile || tex->desc.macrotile[sr.level] ||
             ((referenced_hw & !(usage & PIPE_TRANSFER_READ)) && blittable)) {
             base.target = PIPE_TEXTURE_2D;
             base.format = texture->format;
@@ -149,23 +148,23 @@ r300_texture_get_transfer(struct pipe_context *ctx,
             }
 
             /* Create the temporary texture. */
-            trans->detiled_texture = r300_texture(
+            trans->linear_texture = r300_texture(
                ctx->screen->resource_create(ctx->screen,
                                             &base));
 
-            if (!trans->detiled_texture) {
+            if (!trans->linear_texture) {
                 /* Oh crap, the thing can't create the texture.
                  * Let's flush and try again. */
                 ctx->flush(ctx, 0, NULL);
 
-                trans->detiled_texture = r300_texture(
+                trans->linear_texture = r300_texture(
                    ctx->screen->resource_create(ctx->screen,
                                                 &base));
 
-                if (!trans->detiled_texture) {
+                if (!trans->linear_texture) {
                     /* For linear textures, it's safe to fallback to
                      * an unpipelined transfer. */
-                    if (!tex->microtile && !tex->macrotile) {
+                    if (!tex->desc.microtile && !tex->desc.macrotile[sr.level]) {
                         goto unpipelined;
                     }
 
@@ -177,8 +176,8 @@ r300_texture_get_transfer(struct pipe_context *ctx,
                 }
             }
 
-            assert(!trans->detiled_texture->microtile &&
-                   !trans->detiled_texture->macrotile);
+            assert(!trans->linear_texture->desc.microtile &&
+                   !trans->linear_texture->desc.macrotile[0]);
 
             /* Set the stride.
 	     *
@@ -188,7 +187,7 @@ r300_texture_get_transfer(struct pipe_context *ctx,
 	     * right thing internally.
 	     */
             trans->transfer.stride =
-                r300_texture_get_stride(r300screen, trans->detiled_texture, 0);
+                    trans->linear_texture->desc.stride_in_bytes[0];
 
             if (usage & PIPE_TRANSFER_READ) {
                 /* We cannot map a tiled texture directly because the data is
@@ -203,9 +202,9 @@ r300_texture_get_transfer(struct pipe_context *ctx,
 
     unpipelined:
         /* Unpipelined transfer. */
-        trans->transfer.stride =
-                r300_texture_get_stride(r300screen, tex, sr.level);
-        trans->offset = r300_texture_get_offset(tex, sr.level, box->z, sr.face);
+        trans->transfer.stride = tex->desc.stride_in_bytes[sr.level];
+        trans->offset = r300_texture_get_offset(&tex->desc,
+                                                sr.level, box->z, sr.face);
 
         if (referenced_cs)
             ctx->flush(ctx, PIPE_FLUSH_RENDER_CACHE, NULL);
@@ -219,13 +218,13 @@ void r300_texture_transfer_destroy(struct pipe_context *ctx,
 {
     struct r300_transfer *r300transfer = r300_transfer(trans);
 
-    if (r300transfer->detiled_texture) {
+    if (r300transfer->linear_texture) {
         if (trans->usage & PIPE_TRANSFER_WRITE) {
             r300_copy_into_tiled_texture(ctx, r300transfer);
         }
 
         pipe_resource_reference(
-            (struct pipe_resource**)&r300transfer->detiled_texture, NULL);
+            (struct pipe_resource**)&r300transfer->linear_texture, NULL);
     }
     pipe_resource_reference(&trans->resource, NULL);
     FREE(trans);
@@ -239,13 +238,13 @@ void* r300_texture_transfer_map(struct pipe_context *ctx,
     struct r300_transfer *r300transfer = r300_transfer(transfer);
     struct r300_texture *tex = r300_texture(transfer->resource);
     char *map;
-    enum pipe_format format = tex->b.b.format;
+    enum pipe_format format = tex->desc.b.b.format;
 
-    if (r300transfer->detiled_texture) {
+    if (r300transfer->linear_texture) {
         /* The detiled texture is of the same size as the region being mapped
          * (no offset needed). */
         return rws->buffer_map(rws,
-                               r300transfer->detiled_texture->buffer,
+                               r300transfer->linear_texture->buffer,
                                r300->cs,
                                transfer->usage);
     } else {
@@ -270,8 +269,8 @@ void r300_texture_transfer_unmap(struct pipe_context *ctx,
     struct r300_transfer *r300transfer = r300_transfer(transfer);
     struct r300_texture *tex = r300_texture(transfer->resource);
 
-    if (r300transfer->detiled_texture) {
-	rws->buffer_unmap(rws, r300transfer->detiled_texture->buffer);
+    if (r300transfer->linear_texture) {
+        rws->buffer_unmap(rws, r300transfer->linear_texture->buffer);
     } else {
         rws->buffer_unmap(rws, tex->buffer);
     }
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index b25c786d6b..54c8de1241 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -207,7 +207,7 @@ void r300_translate_vertex_shader(struct r300_context *r300,
     compiler.Base.max_temp_regs = 32;
 
     if (compiler.Base.Debug) {
-        debug_printf("r300: Initial vertex program\n");
+        DBG(r300, DBG_VP, "r300: Initial vertex program\n");
         tgsi_dump(vs->state.tokens, 0);
     }
 
@@ -227,8 +227,7 @@ void r300_translate_vertex_shader(struct r300_context *r300,
     /* Invoke the compiler */
     r3xx_compile_vertex_program(&compiler);
     if (compiler.Base.Error) {
-        /* XXX We should fallback using Draw. */
-        fprintf(stderr, "r300 VP: Compiler error:\n%sUsing a dummy shader"
+        DBG(r300, DBG_VP, "r300 VP: Compiler error:\n%sUsing a dummy shader"
                 " instead.\nIf there's an 'unknown opcode' message, please"
                 " file a bug report and attach this log.\n", compiler.Base.ErrorMsg);
 
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 7e115c2d62..187780750f 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -49,6 +49,8 @@ enum r300_value_id {
     R300_VID_Z_PIPES,
     R300_VID_SQUARE_TILING_SUPPORT,
     R300_VID_DRM_2_3_0,
+    R300_VID_DRM_2_6_0,
+    R300_CAN_HYPERZ,
 };
 
 enum r300_reference_domain { /* bitfield */
@@ -184,12 +186,13 @@ struct r300_winsys_screen {
      * \param ws        The winsys this function is called from.
      * \param whandle   A winsys handle pointer as was received from a state
      *                  tracker.
-     * \param stride    A pointer to the stride return variable.
-     *                  The stride is in bytes.
+     * \param stride    The returned buffer stride in bytes.
+     * \param size      The returned buffer size.
      */
     struct r300_winsys_buffer *(*buffer_from_handle)(struct r300_winsys_screen *ws,
                                                      struct winsys_handle *whandle,
-                                                     unsigned *stride);
+                                                     unsigned *stride,
+                                                     unsigned *size);
 
     /**
      * Get a winsys handle from a winsys buffer. The internal structure