Merge remote-tracking branch 'origin/released'

Conflicts:
	xorg-server/glx/glxext.c
	xorg-server/hw/xwin/InitOutput.c

43 files changed, 1228 insertions, 311 deletions

diff --git a/mesalib/configure.ac b/mesalib/configure.ac
index b5250f321..08ea5beb4 100644
--- a/mesalib/configure.ac
+++ b/mesalib/configure.ac
@@ -1030,29 +1030,12 @@ xyesno)
         GL_PC_CFLAGS="$X11_INCLUDES"
 
         # XCB can only be used from pkg-config
-        PKG_CHECK_MODULES([XCB],[x11-xcb xcb-glx])
+        PKG_CHECK_MODULES([XCB],[x11-xcb xcb-glx >= 1.8.1])
         GL_PC_REQ_PRIV="$GL_PC_REQ_PRIV x11-xcb xcb-glx"
         X11_INCLUDES="$X11_INCLUDES $XCB_CFLAGS"
         GL_LIB_DEPS="$GL_LIB_DEPS $XCB_LIBS"
     fi
 
-    # Check to see if the xcb-glx library is new enough to support
-    # GLX_ARB_create_context.  This bit of hackery is necessary until XCB 1.8
-    # is released.
-    save_CPPFLAGS="$CPPFLAGS"
-    save_LDFLAGS="$LDFLAGS"
-    CPPFLAGS="$CPPFLAGS $X11_INCLUDES"
-    LDFLAGS="$LDFLAGS $GL_LIB_DEPS"
-    AC_CHECK_LIB(xcb-glx, xcb_glx_create_context_attribs_arb_checked,
-        [HAVE_XCB_GLX_CREATE_CONTEXT=yes],
-        [HAVE_XCB_GLX_CREATE_CONTEXT=no])
-    CPPFLAGS="$save_CPPFLAGS"
-    LDFLAGS="$save_LDFLAGS"
-
-    if test x$HAVE_XCB_GLX_CREATE_CONTEXT = xyes; then
-        X11_INCLUDES="$X11_INCLUDES -DHAVE_XCB_GLX_CREATE_CONTEXT"
-    fi
-
     # need DRM libs, -lpthread, etc.
     GL_LIB_DEPS="$GL_LIB_DEPS $LIBDRM_LIBS -lm -lpthread $DLOPEN_LIBS"
     GL_PC_LIB_PRIV="-lm -lpthread $DLOPEN_LIBS"
@@ -1061,8 +1044,6 @@ esac
 
 # This is outside the case (above) so that it is invoked even for non-GLX
 # builds.
-AM_CONDITIONAL(HAVE_XCB_GLX_CREATE_CONTEXT,
-    test x$HAVE_XCB_GLX_CREATE_CONTEXT = xyes)
 AM_CONDITIONAL(HAVE_XF86VIDMODE, test "x$HAVE_XF86VIDMODE" = xyes)
 
 GLESv1_CM_LIB_DEPS="$LIBDRM_LIBS -lm -lpthread $DLOPEN_LIBS"
diff --git a/mesalib/docs/GL3.txt b/mesalib/docs/GL3.txt
index 8357289ea..7c6874568 100644
--- a/mesalib/docs/GL3.txt
+++ b/mesalib/docs/GL3.txt
@@ -100,7 +100,7 @@ GL_ARB_texture_buffer_object_rgb32                   not started
 GL_ARB_texture_cube_map_array                        not started
 GL_ARB_texture_gather                                not started
 GL_ARB_transform_feedback2                           DONE
-GL_ARB_transform_feedback3                           not started
+GL_ARB_transform_feedback3                           DONE
 
 
 GL 4.1:
@@ -121,7 +121,7 @@ GL_ARB_texture_compression_bptc                      not started
 GL_ARB_compressed_texture_pixel_storage              not started
 GL_ARB_shader_atomic_counters                        not started
 GL_ARB_texture_storage                               DONE (r300, r600, swrast)
-GL_ARB_transform_feedback_instanced                  not started
+GL_ARB_transform_feedback_instanced                  DONE
 GL_ARB_base_instance                                 DONE (nv50, nvc0, r600, radeonsi)
 GL_ARB_shader_image_load_store                       not started
 GL_ARB_conservative_depth                            DONE (softpipe)
diff --git a/mesalib/docs/relnotes-8.1.html b/mesalib/docs/relnotes-8.1.html
index c6980b688..45770d392 100644
--- a/mesalib/docs/relnotes-8.1.html
+++ b/mesalib/docs/relnotes-8.1.html
@@ -37,6 +37,8 @@ Note: some of the new features are only available with certain drivers.
 <li>GL_ARB_debug_output</li>
 <li>GL_ARB_shader_bit_encoding</li>
 <li>GL_ARB_timer_query</li>
+<li>GL_ARB_transform_feedback3</li>
+<li>GL_ARB_transform_feedback_instanced</li>
 <li>GL_EXT_unpack_subimage for ES 2.0</li>
 <li>GL_EXT_read_format_bgra for ES 1.1 and 2.0</li>
 <li>GL_EXT_texture_rg for ES 2.x</li>
diff --git a/mesalib/scons/gallium.py b/mesalib/scons/gallium.py
index 75c713d1c..001a5def1 100644
--- a/mesalib/scons/gallium.py
+++ b/mesalib/scons/gallium.py
@@ -368,7 +368,6 @@ def generate(env):
         ccflags += [
             '-Wall',
             '-Wno-long-long',
-            '-ffast-math',
             '-fmessage-length=0', # be nice to Eclipse
         ]
         cflags += [
@@ -405,7 +404,6 @@ def generate(env):
                 '/GL-', # disable whole program optimization
             ]
         ccflags += [
-            '/fp:fast', # fast floating point 
             '/W3', # warning level
             #'/Wp64', # enable 64 bit porting warnings
             '/wd4996', # disable deprecated POSIX name warnings
@@ -516,7 +514,7 @@ def generate(env):
     createInstallMethods(env)
 
     env.PkgCheckModules('X11', ['x11', 'xext', 'xdamage', 'xfixes'])
-    env.PkgCheckModules('XCB', ['x11-xcb', 'xcb-glx'])
+    env.PkgCheckModules('XCB', ['x11-xcb', 'xcb-glx >= 1.8.1'])
     env.PkgCheckModules('XF86VIDMODE', ['xxf86vm'])
     env.PkgCheckModules('DRM', ['libdrm >= 2.4.24'])
     env.PkgCheckModules('DRM_INTEL', ['libdrm_intel >= 2.4.30'])
diff --git a/mesalib/src/gallium/auxiliary/util/u_blit.c b/mesalib/src/gallium/auxiliary/util/u_blit.c
index 9bd2ef52d..b52f4c092 100644
--- a/mesalib/src/gallium/auxiliary/util/u_blit.c
+++ b/mesalib/src/gallium/auxiliary/util/u_blit.c
@@ -56,9 +56,11 @@ struct blit_state
    struct pipe_context *pipe;
    struct cso_context *cso;
 
-   struct pipe_blend_state blend;
-   struct pipe_depth_stencil_alpha_state depthstencil_keep;
-   struct pipe_depth_stencil_alpha_state depthstencil_write;
+   struct pipe_blend_state blend_write_color, blend_keep_color;
+   struct pipe_depth_stencil_alpha_state dsa_keep_depthstencil;
+   struct pipe_depth_stencil_alpha_state dsa_write_depthstencil;
+   struct pipe_depth_stencil_alpha_state dsa_write_depth;
+   struct pipe_depth_stencil_alpha_state dsa_write_stencil;
    struct pipe_rasterizer_state rasterizer;
    struct pipe_sampler_state sampler;
    struct pipe_viewport_state viewport;
@@ -66,13 +68,17 @@ struct blit_state
    enum pipe_texture_target internal_target;
 
    void *vs;
-   void *fs[TGSI_WRITEMASK_XYZW + 1];
-   void *fs_depth;
+   void *fs[PIPE_MAX_TEXTURE_TYPES][TGSI_WRITEMASK_XYZW + 1];
+   void *fs_depthstencil[PIPE_MAX_TEXTURE_TYPES];
+   void *fs_depth[PIPE_MAX_TEXTURE_TYPES];
+   void *fs_stencil[PIPE_MAX_TEXTURE_TYPES];
 
    struct pipe_resource *vbuf;  /**< quad vertices */
    unsigned vbuf_slot;
 
    float vertices[4][2][4];   /**< vertex/texcoords for quad */
+
+   boolean has_stencil_export;
 };
 
 
@@ -94,24 +100,28 @@ util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
    ctx->cso = cso;
 
    /* disabled blending/masking */
-   memset(&ctx->blend, 0, sizeof(ctx->blend));
-   ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
-
-   /* no-op depth/stencil/alpha */
-   memset(&ctx->depthstencil_keep, 0, sizeof(ctx->depthstencil_keep));
-   memset(&ctx->depthstencil_write, 0, sizeof(ctx->depthstencil_write));
-   ctx->depthstencil_write.depth.enabled = 1;
-   ctx->depthstencil_write.depth.writemask = 1;
-   ctx->depthstencil_write.depth.func = PIPE_FUNC_ALWAYS;
+   ctx->blend_write_color.rt[0].colormask = PIPE_MASK_RGBA;
+
+   /* depth stencil states */
+   ctx->dsa_write_depth.depth.enabled = 1;
+   ctx->dsa_write_depth.depth.writemask = 1;
+   ctx->dsa_write_depth.depth.func = PIPE_FUNC_ALWAYS;
+   ctx->dsa_write_stencil.stencil[0].enabled = 1;
+   ctx->dsa_write_stencil.stencil[0].func = PIPE_FUNC_ALWAYS;
+   ctx->dsa_write_stencil.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
+   ctx->dsa_write_stencil.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
+   ctx->dsa_write_stencil.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
+   ctx->dsa_write_stencil.stencil[0].valuemask = 0xff;
+   ctx->dsa_write_stencil.stencil[0].writemask = 0xff;
+   ctx->dsa_write_depthstencil.depth = ctx->dsa_write_depth.depth;
+   ctx->dsa_write_depthstencil.stencil[0] = ctx->dsa_write_stencil.stencil[0];
 
    /* rasterizer */
-   memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
    ctx->rasterizer.cull_face = PIPE_FACE_NONE;
    ctx->rasterizer.gl_rasterization_rules = 1;
    ctx->rasterizer.depth_clip = 1;
 
    /* samplers */
-   memset(&ctx->sampler, 0, sizeof(ctx->sampler));
    ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
    ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
    ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
@@ -120,7 +130,6 @@ util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
    ctx->sampler.mag_img_filter = 0; /* set later */
 
    /* vertex elements state */
-   memset(&ctx->velem[0], 0, sizeof(ctx->velem[0]) * 2);
    for (i = 0; i < 2; i++) {
       ctx->velem[i].src_offset = i * 4 * sizeof(float);
       ctx->velem[i].instance_divisor = 0;
@@ -142,6 +151,9 @@ util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
    else
       ctx->internal_target = PIPE_TEXTURE_RECT;
 
+   ctx->has_stencil_export =
+      pipe->screen->get_param(pipe->screen, PIPE_CAP_SHADER_STENCIL_EXPORT);
+
    return ctx;
 }
 
@@ -153,17 +165,29 @@ void
 util_destroy_blit(struct blit_state *ctx)
 {
    struct pipe_context *pipe = ctx->pipe;
-   unsigned i;
+   unsigned i, j;
 
    if (ctx->vs)
       pipe->delete_vs_state(pipe, ctx->vs);
 
-   for (i = 0; i < Elements(ctx->fs); i++)
-      if (ctx->fs[i])
-         pipe->delete_fs_state(pipe, ctx->fs[i]);
+   for (i = 0; i < Elements(ctx->fs); i++) {
+      for (j = 0; j < Elements(ctx->fs[i]); j++) {
+         if (ctx->fs[i][j])
+            pipe->delete_fs_state(pipe, ctx->fs[i][j]);
+      }
+   }
 
-   if (ctx->fs_depth)
-      pipe->delete_fs_state(pipe, ctx->fs_depth);
+   for (i = 0; i < PIPE_MAX_TEXTURE_TYPES; i++) {
+      if (ctx->fs_depthstencil[i]) {
+         pipe->delete_fs_state(pipe, ctx->fs_depthstencil[i]);
+      }
+      if (ctx->fs_depth[i]) {
+         pipe->delete_fs_state(pipe, ctx->fs_depth[i]);
+      }
+      if (ctx->fs_stencil[i]) {
+         pipe->delete_fs_state(pipe, ctx->fs_stencil[i]);
+      }
+   }
 
    pipe_resource_reference(&ctx->vbuf, NULL);
 
@@ -175,30 +199,76 @@ util_destroy_blit(struct blit_state *ctx)
  * Helper function to set the fragment shaders.
  */
 static INLINE void
-set_fragment_shader(struct blit_state *ctx, uint writemask)
+set_fragment_shader(struct blit_state *ctx, uint writemask,
+                    enum pipe_texture_target pipe_tex)
 {
-   if (!ctx->fs[writemask])
-      ctx->fs[writemask] =
-         util_make_fragment_tex_shader_writemask(ctx->pipe, TGSI_TEXTURE_2D,
+   if (!ctx->fs[pipe_tex][writemask]) {
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(pipe_tex);
+
+      ctx->fs[pipe_tex][writemask] =
+         util_make_fragment_tex_shader_writemask(ctx->pipe, tgsi_tex,
                                                  TGSI_INTERPOLATE_LINEAR,
                                                  writemask);
+   }
 
-   cso_set_fragment_shader_handle(ctx->cso, ctx->fs[writemask]);
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs[pipe_tex][writemask]);
 }
 
 
 /**
- * Helper function to set the depthwrite shader.
+ * Helper function to set the shader which writes depth and stencil.
  */
 static INLINE void
-set_depth_fragment_shader(struct blit_state *ctx)
+set_depthstencil_fragment_shader(struct blit_state *ctx,
+                                 enum pipe_texture_target pipe_tex)
 {
-   if (!ctx->fs_depth)
-      ctx->fs_depth =
-         util_make_fragment_tex_shader_writedepth(ctx->pipe, TGSI_TEXTURE_2D,
+   if (!ctx->fs_depthstencil[pipe_tex]) {
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(pipe_tex);
+
+      ctx->fs_depthstencil[pipe_tex] =
+         util_make_fragment_tex_shader_writedepthstencil(ctx->pipe, tgsi_tex,
                                                   TGSI_INTERPOLATE_LINEAR);
+   }
 
-   cso_set_fragment_shader_handle(ctx->cso, ctx->fs_depth);
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs_depthstencil[pipe_tex]);
+}
+
+
+/**
+ * Helper function to set the shader which writes depth.
+ */
+static INLINE void
+set_depth_fragment_shader(struct blit_state *ctx,
+                          enum pipe_texture_target pipe_tex)
+{
+   if (!ctx->fs_depth[pipe_tex]) {
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(pipe_tex);
+
+      ctx->fs_depth[pipe_tex] =
+         util_make_fragment_tex_shader_writedepth(ctx->pipe, tgsi_tex,
+                                                  TGSI_INTERPOLATE_LINEAR);
+   }
+
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs_depth[pipe_tex]);
+}
+
+
+/**
+ * Helper function to set the shader which writes stencil.
+ */
+static INLINE void
+set_stencil_fragment_shader(struct blit_state *ctx,
+                            enum pipe_texture_target pipe_tex)
+{
+   if (!ctx->fs_stencil[pipe_tex]) {
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(pipe_tex);
+
+      ctx->fs_stencil[pipe_tex] =
+         util_make_fragment_tex_shader_writestencil(ctx->pipe, tgsi_tex,
+                                                    TGSI_INTERPOLATE_LINEAR);
+   }
+
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs_stencil[pipe_tex]);
 }
 
 
@@ -350,20 +420,19 @@ formats_compatible(enum pipe_format src_format,
  * \param writemask  controls which channels in the dest surface are sourced
  *                   from the src surface.  Disabled channels are sourced
  *                   from (0,0,0,1).
- * XXX need some control over blitting stencil.
  */
 void
-util_blit_pixels_writemask(struct blit_state *ctx,
-                           struct pipe_resource *src_tex,
-                           unsigned src_level,
-                           int srcX0, int srcY0,
-                           int srcX1, int srcY1,
-                           int srcZ0,
-                           struct pipe_surface *dst,
-                           int dstX0, int dstY0,
-                           int dstX1, int dstY1,
-                           float z, uint filter,
-                           uint writemask)
+util_blit_pixels(struct blit_state *ctx,
+                 struct pipe_resource *src_tex,
+                 unsigned src_level,
+                 int srcX0, int srcY0,
+                 int srcX1, int srcY1,
+                 int srcZ0,
+                 struct pipe_surface *dst,
+                 int dstX0, int dstY0,
+                 int dstX1, int dstY1,
+                 float z, uint filter,
+                 uint writemask, uint zs_writemask)
 {
    struct pipe_context *pipe = ctx->pipe;
    struct pipe_screen *screen = pipe->screen;
@@ -375,9 +444,12 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    const int srcW = abs(srcX1 - srcX0);
    const int srcH = abs(srcY1 - srcY0);
    unsigned offset;
-   boolean overlap, dst_is_depth;
+   boolean overlap;
    float s0, t0, s1, t1;
    boolean normalized;
+   boolean is_stencil, is_depth, blit_depth, blit_stencil;
+   const struct util_format_description *src_desc =
+         util_format_description(src_tex->format);
 
    assert(filter == PIPE_TEX_MIPFILTER_NEAREST ||
           filter == PIPE_TEX_MIPFILTER_LINEAR);
@@ -394,12 +466,24 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    src_format = util_format_linear(src_tex->format);
    dst_format = util_format_linear(dst->format);
 
+   /* See whether we will blit depth or stencil. */
+   is_depth = util_format_has_depth(src_desc);
+   is_stencil = util_format_has_stencil(src_desc);
+
+   blit_depth = is_depth && (zs_writemask & BLIT_WRITEMASK_Z);
+   blit_stencil = is_stencil && (zs_writemask & BLIT_WRITEMASK_STENCIL);
+
+   assert((writemask && !zs_writemask && !is_depth && !is_stencil) ||
+          (!writemask && (blit_depth || blit_stencil)));
+
    /*
     * Check for simple case:  no format conversion, no flipping, no stretching,
     * no overlapping.
     * Filter mode should not matter since there's no stretching.
     */
    if (formats_compatible(src_format, dst_format) &&
+       is_stencil == blit_stencil &&
+       is_depth == blit_depth &&
        srcX0 < srcX1 &&
        dstX0 < dstX1 &&
        srcY0 < srcY1 &&
@@ -422,6 +506,17 @@ util_blit_pixels_writemask(struct blit_state *ctx,
       return;
    }
 
+   /* It's a mistake to call this function with a stencil format and
+    * without shader stencil export. We don't do software fallbacks here.
+    * Ignore stencil and only copy depth.
+    */
+   if (blit_stencil && !ctx->has_stencil_export) {
+      blit_stencil = FALSE;
+
+      if (!blit_depth)
+         return;
+   }
+
    if (dst_format == dst->format) {
       dst_surface = dst;
    } else {
@@ -430,20 +525,11 @@ util_blit_pixels_writemask(struct blit_state *ctx,
       dst_surface = pipe->create_surface(pipe, dst->texture, &templ);
    }
 
-   /* Create a temporary texture when src and dest alias or when src
-    * is anything other than a 2d texture.
-    * XXX should just use appropriate shader to access 1d / 3d slice / cube face,
-    * much like the u_blitter code does (should be pretty trivial).
-    * 
-    * This can still be improved upon.
+   /* Create a temporary texture when src and dest alias.
     */
-   if ((src_tex == dst_surface->texture &&
+   if (src_tex == dst_surface->texture &&
        dst_surface->u.tex.level == src_level &&
-       dst_surface->u.tex.first_layer == srcZ0) ||
-       (src_tex->target != PIPE_TEXTURE_2D &&
-       src_tex->target != PIPE_TEXTURE_2D &&
-       src_tex->target != PIPE_TEXTURE_RECT))
-   {
+       dst_surface->u.tex.first_layer == srcZ0) {
       /* Make a temporary texture which contains a copy of the source pixels.
        * Then we'll sample from the temporary texture.
        */
@@ -509,6 +595,11 @@ util_blit_pixels_writemask(struct blit_state *ctx,
       }
 
       u_sampler_view_default_template(&sv_templ, tex, tex->format);
+      if (!blit_depth && blit_stencil) {
+         /* set a stencil-only format, e.g. Z24S8 --> X24S8 */
+         sv_templ.format = util_format_stencil_only(tex->format);
+         assert(sv_templ.format != PIPE_FORMAT_NONE);
+      }
       sampler_view = pipe->create_sampler_view(pipe, tex, &sv_templ);
 
       if (!sampler_view) {
@@ -520,6 +611,11 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    else {
       /* Directly sample from the source resource/texture */
       u_sampler_view_default_template(&sv_templ, src_tex, src_format);
+      if (!blit_depth && blit_stencil) {
+         /* set a stencil-only format, e.g. Z24S8 --> X24S8 */
+         sv_templ.format = util_format_stencil_only(src_format);
+         assert(sv_templ.format != PIPE_FORMAT_NONE);
+      }
       sampler_view = pipe->create_sampler_view(pipe, src_tex, &sv_templ);
 
       if (!sampler_view) {
@@ -540,15 +636,14 @@ util_blit_pixels_writemask(struct blit_state *ctx,
       }
    }
 
-   dst_is_depth = util_format_is_depth_or_stencil(dst_format);
-
-   assert(screen->is_format_supported(screen, sampler_view->format, ctx->internal_target,
-                                      sampler_view->texture->nr_samples,
-                                      PIPE_BIND_SAMPLER_VIEW));
+   assert(screen->is_format_supported(screen, sampler_view->format,
+                     ctx->internal_target, sampler_view->texture->nr_samples,
+                     PIPE_BIND_SAMPLER_VIEW));
    assert(screen->is_format_supported(screen, dst_format, ctx->internal_target,
-                                      dst_surface->texture->nr_samples,
-                                      dst_is_depth ? PIPE_BIND_DEPTH_STENCIL :
-                                                     PIPE_BIND_RENDER_TARGET));
+                     dst_surface->texture->nr_samples,
+                     is_depth || is_stencil ? PIPE_BIND_DEPTH_STENCIL :
+                                              PIPE_BIND_RENDER_TARGET));
+
    /* save state (restored below) */
    cso_save_blend(ctx->cso);
    cso_save_depth_stencil_alpha(ctx->cso);
@@ -565,23 +660,76 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    cso_save_vertex_buffers(ctx->cso);
 
    /* set misc state we care about */
-   cso_set_blend(ctx->cso, &ctx->blend);
-   cso_set_depth_stencil_alpha(ctx->cso,
-                               dst_is_depth ? &ctx->depthstencil_write :
-                                              &ctx->depthstencil_keep);
+   if (writemask)
+      cso_set_blend(ctx->cso, &ctx->blend_write_color);
+   else
+      cso_set_blend(ctx->cso, &ctx->blend_keep_color);
+
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
    cso_set_stream_outputs(ctx->cso, 0, NULL, 0);
 
-   /* sampler */
+   /* default sampler state */
    ctx->sampler.normalized_coords = normalized;
    ctx->sampler.min_img_filter = filter;
    ctx->sampler.mag_img_filter = filter;
    ctx->sampler.min_lod = src_level;
    ctx->sampler.max_lod = src_level;
-   cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+
+   /* Depth stencil state, fragment shader and sampler setup depending on what
+    * we blit.
+    */
+   if (blit_depth && blit_stencil) {
+      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+      /* don't filter stencil */
+      ctx->sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
+      ctx->sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
+      cso_single_sampler(ctx->cso, 1, &ctx->sampler);
+
+      cso_set_depth_stencil_alpha(ctx->cso, &ctx->dsa_write_depthstencil);
+      set_depthstencil_fragment_shader(ctx, sampler_view->texture->target);
+   }
+   else if (blit_depth) {
+      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+      cso_set_depth_stencil_alpha(ctx->cso, &ctx->dsa_write_depth);
+      set_depth_fragment_shader(ctx, sampler_view->texture->target);
+   }
+   else if (blit_stencil) {
+      /* don't filter stencil */
+      ctx->sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
+      ctx->sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
+      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+
+      cso_set_depth_stencil_alpha(ctx->cso, &ctx->dsa_write_stencil);
+      set_stencil_fragment_shader(ctx, sampler_view->texture->target);
+   }
+   else { /* color */
+      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+      cso_set_depth_stencil_alpha(ctx->cso, &ctx->dsa_keep_depthstencil);
+      set_fragment_shader(ctx, writemask, sampler_view->texture->target);
+   }
    cso_single_sampler_done(ctx->cso);
 
+   /* textures */
+   if (blit_depth && blit_stencil) {
+      /* Setup two samplers, one for depth and the other one for stencil. */
+      struct pipe_sampler_view templ;
+      struct pipe_sampler_view *views[2];
+
+      templ = *sampler_view;
+      templ.format = util_format_stencil_only(templ.format);
+      assert(templ.format != PIPE_FORMAT_NONE);
+
+      views[0] = sampler_view;
+      views[1] = pipe->create_sampler_view(pipe, views[0]->texture, &templ);
+      cso_set_fragment_sampler_views(ctx->cso, 2, views);
+
+      pipe_sampler_view_reference(&views[1], NULL);
+   }
+   else {
+      cso_set_fragment_sampler_views(ctx->cso, 1, &sampler_view);
+   }
+
    /* viewport */
    ctx->viewport.scale[0] = 0.5f * dst_surface->width;
    ctx->viewport.scale[1] = 0.5f * dst_surface->height;
@@ -593,15 +741,6 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    ctx->viewport.translate[3] = 0.0f;
    cso_set_viewport(ctx->cso, &ctx->viewport);
 
-   /* texture */
-   cso_set_fragment_sampler_views(ctx->cso, 1, &sampler_view);
-
-   /* shaders */
-   if (dst_is_depth) {
-      set_depth_fragment_shader(ctx);
-   } else {
-      set_fragment_shader(ctx, writemask);
-   }
    set_vertex_shader(ctx);
    cso_set_geometry_shader_handle(ctx->cso, NULL);
 
@@ -609,7 +748,7 @@ util_blit_pixels_writemask(struct blit_state *ctx,
    memset(&fb, 0, sizeof(fb));
    fb.width = dst_surface->width;
    fb.height = dst_surface->height;
-   if (dst_is_depth) {
+   if (blit_depth || blit_stencil) {
       fb.zsbuf = dst_surface;
    } else {
       fb.nr_cbufs = 1;
@@ -655,31 +794,6 @@ util_blit_pixels_writemask(struct blit_state *ctx,
 }
 
 
-void
-util_blit_pixels(struct blit_state *ctx,
-                 struct pipe_resource *src_tex,
-                 unsigned src_level,
-                 int srcX0, int srcY0,
-                 int srcX1, int srcY1,
-                 int srcZ,
-                 struct pipe_surface *dst,
-                 int dstX0, int dstY0,
-                 int dstX1, int dstY1,
-                 float z, uint filter )
-{
-   util_blit_pixels_writemask( ctx, src_tex,
-                               src_level,
-                               srcX0, srcY0,
-                               srcX1, srcY1,
-                               srcZ,
-                               dst,
-                               dstX0, dstY0,
-                               dstX1, dstY1,
-                               z, filter,
-                               TGSI_WRITEMASK_XYZW );
-}
-
-
 /**
  * Copy pixel block from src texture to dst surface.
  * The sampler view's first_level field indicates the source
@@ -747,8 +861,8 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_save_vertex_buffers(ctx->cso);
 
    /* set misc state we care about */
-   cso_set_blend(ctx->cso, &ctx->blend);
-   cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil_keep);
+   cso_set_blend(ctx->cso, &ctx->blend_write_color);
+   cso_set_depth_stencil_alpha(ctx->cso, &ctx->dsa_keep_depthstencil);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
    cso_set_stream_outputs(ctx->cso, 0, NULL, 0);
@@ -775,7 +889,8 @@ util_blit_pixels_tex(struct blit_state *ctx,
    cso_set_fragment_sampler_views(ctx->cso, 1, &src_sampler_view);
 
    /* shaders */
-   set_fragment_shader(ctx, TGSI_WRITEMASK_XYZW);
+   set_fragment_shader(ctx, TGSI_WRITEMASK_XYZW,
+                       src_sampler_view->texture->target);
    set_vertex_shader(ctx);
    cso_set_geometry_shader_handle(ctx->cso, NULL);
 
diff --git a/mesalib/src/gallium/auxiliary/util/u_blit.h b/mesalib/src/gallium/auxiliary/util/u_blit.h
index 810d01b04..56ab03083 100644
--- a/mesalib/src/gallium/auxiliary/util/u_blit.h
+++ b/mesalib/src/gallium/auxiliary/util/u_blit.h
@@ -31,6 +31,8 @@
 
 
 #include "pipe/p_compiler.h"
+/* for TGSI_WRITEMASK_* specification in util_blit_pixels */
+#include "pipe/p_shader_tokens.h"
 
 
 #ifdef __cplusplus
@@ -44,6 +46,8 @@ struct pipe_resource;
 struct pipe_sampler_view;
 struct pipe_surface;
 
+#define BLIT_WRITEMASK_Z         1
+#define BLIT_WRITEMASK_STENCIL   2
 
 extern struct blit_state *
 util_create_blit(struct pipe_context *pipe, struct cso_context *cso);
@@ -61,20 +65,8 @@ util_blit_pixels(struct blit_state *ctx,
                  struct pipe_surface *dst,
                  int dstX0, int dstY0,
                  int dstX1, int dstY1,
-                 float z, uint filter);
-
-void
-util_blit_pixels_writemask(struct blit_state *ctx,
-                           struct pipe_resource *src_tex,
-                           unsigned src_level,
-                           int srcX0, int srcY0,
-                           int srcX1, int srcY1,
-                           int srcZ0,
-                           struct pipe_surface *dst,
-                           int dstX0, int dstY0,
-                           int dstX1, int dstY1,
-                           float z, uint filter,
-                           uint writemask);
+                 float z, uint filter,
+                 uint writemask, uint zs_writemask);
 
 extern void
 util_blit_pixels_tex(struct blit_state *ctx,
diff --git a/mesalib/src/gallium/auxiliary/util/u_blitter.c b/mesalib/src/gallium/auxiliary/util/u_blitter.c
index 47d0227a9..b31ac2d5c 100644
--- a/mesalib/src/gallium/auxiliary/util/u_blitter.c
+++ b/mesalib/src/gallium/auxiliary/util/u_blitter.c
@@ -78,6 +78,8 @@ struct blitter_context_priv
    /* FS which outputs a depth from a texture,
       where the index is PIPE_TEXTURE_* to be sampled. */
    void *fs_texfetch_depth[PIPE_MAX_TEXTURE_TYPES];
+   void *fs_texfetch_depthstencil[PIPE_MAX_TEXTURE_TYPES];
+   void *fs_texfetch_stencil[PIPE_MAX_TEXTURE_TYPES];
 
    /* Blend state. */
    void *blend_write_color;   /**< blend state with writemask of RGBA */
@@ -112,6 +114,7 @@ struct blitter_context_priv
    boolean has_geometry_shader;
    boolean vertex_has_integers;
    boolean has_stream_out;
+   boolean has_stencil_export;
 };
 
 static void blitter_draw_rectangle(struct blitter_context *blitter,
@@ -163,6 +166,10 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
       pipe->screen->get_param(pipe->screen,
                               PIPE_CAP_MAX_STREAM_OUTPUT_BUFFERS) != 0;
 
+   ctx->has_stencil_export =
+         pipe->screen->get_param(pipe->screen,
+                                 PIPE_CAP_SHADER_STENCIL_EXPORT);
+
    /* blend state objects */
    memset(&blend, 0, sizeof(blend));
    ctx->blend_keep_color = pipe->create_blend_state(pipe, &blend);
@@ -314,6 +321,10 @@ void util_blitter_destroy(struct blitter_context *blitter)
          pipe->delete_fs_state(pipe, ctx->fs_texfetch_col[i]);
       if (ctx->fs_texfetch_depth[i])
          pipe->delete_fs_state(pipe, ctx->fs_texfetch_depth[i]);
+      if (ctx->fs_texfetch_depthstencil[i])
+         pipe->delete_fs_state(pipe, ctx->fs_texfetch_depthstencil[i]);
+      if (ctx->fs_texfetch_stencil[i])
+         pipe->delete_fs_state(pipe, ctx->fs_texfetch_stencil[i]);
    }
 
    for (i = 0; i <= PIPE_MAX_COLOR_BUFS; i++) {
@@ -653,32 +664,6 @@ void *blitter_get_fs_col(struct blitter_context_priv *ctx, unsigned num_cbufs,
    }
 }
 
-/** Convert PIPE_TEXTURE_x to TGSI_TEXTURE_x */
-static unsigned
-pipe_tex_to_tgsi_tex(enum pipe_texture_target pipe_tex_target)
-{
-   switch (pipe_tex_target) {
-   case PIPE_TEXTURE_1D:
-      return TGSI_TEXTURE_1D;
-   case PIPE_TEXTURE_2D:
-      return TGSI_TEXTURE_2D;
-   case PIPE_TEXTURE_RECT:
-      return TGSI_TEXTURE_RECT;
-   case PIPE_TEXTURE_3D:
-      return TGSI_TEXTURE_3D;
-   case PIPE_TEXTURE_CUBE:
-      return TGSI_TEXTURE_CUBE;
-   case PIPE_TEXTURE_1D_ARRAY:
-      return TGSI_TEXTURE_1D_ARRAY;
-   case PIPE_TEXTURE_2D_ARRAY:
-      return TGSI_TEXTURE_2D_ARRAY;
-   default:
-      assert(0 && "unexpected texture target");
-      return TGSI_TEXTURE_UNKNOWN;
-   }
-}
-
-
 static INLINE
 void *blitter_get_fs_texfetch_col(struct blitter_context_priv *ctx,
                                   unsigned tex_target)
@@ -689,7 +674,7 @@ void *blitter_get_fs_texfetch_col(struct blitter_context_priv *ctx,
 
    /* Create the fragment shader on-demand. */
    if (!ctx->fs_texfetch_col[tex_target]) {
-      unsigned tgsi_tex = pipe_tex_to_tgsi_tex(tex_target);
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(tex_target);
 
       ctx->fs_texfetch_col[tex_target] =
         util_make_fragment_tex_shader(pipe, tgsi_tex, TGSI_INTERPOLATE_LINEAR);
@@ -708,7 +693,7 @@ void *blitter_get_fs_texfetch_depth(struct blitter_context_priv *ctx,
 
    /* Create the fragment shader on-demand. */
    if (!ctx->fs_texfetch_depth[tex_target]) {
-      unsigned tgsi_tex = pipe_tex_to_tgsi_tex(tex_target);
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(tex_target);
 
       ctx->fs_texfetch_depth[tex_target] =
          util_make_fragment_tex_shader_writedepth(pipe, tgsi_tex,
@@ -718,6 +703,46 @@ void *blitter_get_fs_texfetch_depth(struct blitter_context_priv *ctx,
    return ctx->fs_texfetch_depth[tex_target];
 }
 
+static INLINE
+void *blitter_get_fs_texfetch_depthstencil(struct blitter_context_priv *ctx,
+                                           unsigned tex_target)
+{
+   struct pipe_context *pipe = ctx->base.pipe;
+
+   assert(tex_target < PIPE_MAX_TEXTURE_TYPES);
+
+   /* Create the fragment shader on-demand. */
+   if (!ctx->fs_texfetch_depthstencil[tex_target]) {
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(tex_target);
+
+      ctx->fs_texfetch_depthstencil[tex_target] =
+         util_make_fragment_tex_shader_writedepthstencil(pipe, tgsi_tex,
+                                                  TGSI_INTERPOLATE_LINEAR);
+   }
+
+   return ctx->fs_texfetch_depthstencil[tex_target];
+}
+
+static INLINE
+void *blitter_get_fs_texfetch_stencil(struct blitter_context_priv *ctx,
+                                      unsigned tex_target)
+{
+   struct pipe_context *pipe = ctx->base.pipe;
+
+   assert(tex_target < PIPE_MAX_TEXTURE_TYPES);
+
+   /* Create the fragment shader on-demand. */
+   if (!ctx->fs_texfetch_stencil[tex_target]) {
+      unsigned tgsi_tex = util_pipe_tex_to_tgsi_tex(tex_target);
+
+      ctx->fs_texfetch_stencil[tex_target] =
+         util_make_fragment_tex_shader_writestencil(pipe, tgsi_tex,
+                                                    TGSI_INTERPOLATE_LINEAR);
+   }
+
+   return ctx->fs_texfetch_stencil[tex_target];
+}
+
 static void blitter_set_common_draw_rect_state(struct blitter_context_priv *ctx)
 {
    struct pipe_context *pipe = ctx->base.pipe;
@@ -892,7 +917,7 @@ void util_blitter_default_src_texture(struct pipe_sampler_view *src_templ,
     src_templ->u.tex.last_level = srclevel;
     src_templ->u.tex.first_layer = 0;
     src_templ->u.tex.last_layer =
-        src->target == PIPE_TEXTURE_3D ? src->depth0 - 1
+        src->target == PIPE_TEXTURE_3D ? u_minify(src->depth0, srclevel) - 1
                                        : src->array_size - 1;
     src_templ->swizzle_r = PIPE_SWIZZLE_RED;
     src_templ->swizzle_g = PIPE_SWIZZLE_GREEN;
@@ -916,6 +941,8 @@ void util_blitter_copy_texture(struct blitter_context *blitter,
    struct pipe_sampler_view src_templ, *src_view;
    unsigned bind;
    boolean is_stencil, is_depth;
+   const struct util_format_description *src_desc =
+         util_format_description(src->format);
 
    /* Give up if textures are not set. */
    assert(dst && src);
@@ -925,8 +952,8 @@ void util_blitter_copy_texture(struct blitter_context *blitter,
    assert(src->target < PIPE_MAX_TEXTURE_TYPES);
 
    /* Is this a ZS format? */
-   is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
-   is_stencil = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 1) != 0;
+   is_depth = util_format_has_depth(src_desc);
+   is_stencil = util_format_has_stencil(src_desc);
 
    if (is_depth || is_stencil)
       bind = PIPE_BIND_DEPTH_STENCIL;
@@ -935,7 +962,7 @@ void util_blitter_copy_texture(struct blitter_context *blitter,
 
    /* Check if we can sample from and render to the surfaces. */
    /* (assuming copying a stencil buffer is not possible) */
-   if ((!ignore_stencil && is_stencil) ||
+   if ((!ignore_stencil && is_stencil && !ctx->has_stencil_export) ||
        !screen->is_format_supported(screen, dst->format, dst->target,
                                     dst->nr_samples, bind) ||
        !screen->is_format_supported(screen, src->format, src->target,
@@ -976,6 +1003,21 @@ void util_blitter_copy_texture_view(struct blitter_context *blitter,
    enum pipe_texture_target src_target = src->texture->target;
    unsigned width = srcbox->width;
    unsigned height = srcbox->height;
+   boolean is_stencil, is_depth;
+   const struct util_format_description *src_desc =
+         util_format_description(src->format);
+
+   is_depth = util_format_has_depth(src_desc);
+   is_stencil = util_format_has_stencil(src_desc);
+
+   /* If you want a fallback for stencil copies,
+    * use util_blitter_copy_texture. */
+   if (is_stencil && !ctx->has_stencil_export) {
+      is_stencil = FALSE;
+
+      if (!is_depth)
+         return;
+   }
 
    /* Sanity checks. */
    if (dst->texture == src->texture &&
@@ -997,12 +1039,25 @@ void util_blitter_copy_texture_view(struct blitter_context *blitter,
    fb_state.width = dst->width;
    fb_state.height = dst->height;
 
-   if (util_format_is_depth_or_stencil(dst->format)) {
+   if (is_depth || is_stencil) {
       pipe->bind_blend_state(pipe, ctx->blend_keep_color);
-      pipe->bind_depth_stencil_alpha_state(pipe,
-                                           ctx->dsa_write_depth_keep_stencil);
-      pipe->bind_fs_state(pipe,
-            blitter_get_fs_texfetch_depth(ctx, src_target));
+
+      if (is_depth && is_stencil) {
+         pipe->bind_depth_stencil_alpha_state(pipe,
+                                              ctx->dsa_write_depth_stencil);
+         pipe->bind_fs_state(pipe,
+               blitter_get_fs_texfetch_depthstencil(ctx, src_target));
+      } else if (is_depth) {
+         pipe->bind_depth_stencil_alpha_state(pipe,
+                                              ctx->dsa_write_depth_keep_stencil);
+         pipe->bind_fs_state(pipe,
+               blitter_get_fs_texfetch_depth(ctx, src_target));
+      } else { /* is_stencil */
+         pipe->bind_depth_stencil_alpha_state(pipe,
+                                              ctx->dsa_keep_depth_write_stencil);
+         pipe->bind_fs_state(pipe,
+               blitter_get_fs_texfetch_stencil(ctx, src_target));
+      }
 
       fb_state.nr_cbufs = 0;
       fb_state.zsbuf = dst;
@@ -1017,9 +1072,29 @@ void util_blitter_copy_texture_view(struct blitter_context *blitter,
       fb_state.zsbuf = 0;
    }
 
-   pipe->bind_fragment_sampler_states(pipe, 1, &ctx->sampler_state);
+   if (is_depth && is_stencil) {
+      /* Setup two samplers, one for depth and the other one for stencil. */
+      struct pipe_sampler_view templ;
+      struct pipe_sampler_view *views[2];
+      void *samplers[2] = {ctx->sampler_state, ctx->sampler_state};
+
+      templ = *src;
+      templ.format = util_format_stencil_only(templ.format);
+      assert(templ.format != PIPE_FORMAT_NONE);
+
+      views[0] = src;
+      views[1] = pipe->create_sampler_view(pipe, src->texture, &templ);
+
+      pipe->set_fragment_sampler_views(pipe, 2, views);
+      pipe->bind_fragment_sampler_states(pipe, 2, samplers);
+
+      pipe_sampler_view_reference(&views[1], NULL);
+   } else {
+      pipe->set_fragment_sampler_views(pipe, 1, &src);
+      pipe->bind_fragment_sampler_states(pipe, 1, &ctx->sampler_state);
+   }
+
    pipe->bind_vertex_elements_state(pipe, ctx->velem_state);
-   pipe->set_fragment_sampler_views(pipe, 1, &src);
    pipe->set_framebuffer_state(pipe, &fb_state);
 
    blitter_set_common_draw_rect_state(ctx);
diff --git a/mesalib/src/gallium/auxiliary/util/u_blitter.h b/mesalib/src/gallium/auxiliary/util/u_blitter.h
index d4d30852b..2db984c8a 100644
--- a/mesalib/src/gallium/auxiliary/util/u_blitter.h
+++ b/mesalib/src/gallium/auxiliary/util/u_blitter.h
@@ -169,8 +169,8 @@ void util_blitter_clear_depth_custom(struct blitter_context *blitter,
  * The same holds for depth-stencil formats with the exception that stencil
  * cannot be copied unless you set ignore_stencil to FALSE. In that case,
  * a software fallback path is taken and both surfaces must be of the same
- * format.
- * XXX implement hw-accel stencil copy using shader stencil export.
+ * format. If the shader stencil export is supported, stencil copy is always
+ * accelerated.
  *
  * Use pipe_screen->is_format_supported to know your options.
  *
diff --git a/mesalib/src/gallium/auxiliary/util/u_format.h b/mesalib/src/gallium/auxiliary/util/u_format.h
index e35e164b4..a718389dd 100644
--- a/mesalib/src/gallium/auxiliary/util/u_format.h
+++ b/mesalib/src/gallium/auxiliary/util/u_format.h
@@ -882,6 +882,35 @@ util_format_linear(enum pipe_format format)
 }
 
 /**
+ * Given a depth-stencil format, return the corresponding stencil-only format.
+ * For stencil-only formats, return the format unchanged.
+ */
+static INLINE enum pipe_format
+util_format_stencil_only(enum pipe_format format)
+{
+   switch (format) {
+   /* mask out the depth component */
+   case PIPE_FORMAT_Z24_UNORM_S8_UINT:
+      return PIPE_FORMAT_X24S8_UINT;
+   case PIPE_FORMAT_S8_UINT_Z24_UNORM:
+      return PIPE_FORMAT_S8X24_UINT;
+   case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
+      return PIPE_FORMAT_X32_S8X24_UINT;
+
+   /* stencil only formats */
+   case PIPE_FORMAT_X24S8_UINT:
+   case PIPE_FORMAT_S8X24_UINT:
+   case PIPE_FORMAT_X32_S8X24_UINT:
+   case PIPE_FORMAT_S8_UINT:
+      return format;
+
+   default:
+      assert(0);
+      return PIPE_FORMAT_NONE;
+   }
+}
+
+/**
  * Return the number of components stored.
  * Formats with block size != 1x1 will always have 1 component (the block).
  */
diff --git a/mesalib/src/gallium/auxiliary/util/u_gen_mipmap.c b/mesalib/src/gallium/auxiliary/util/u_gen_mipmap.c
index 2ff1af70a..a96cf6a30 100644
--- a/mesalib/src/gallium/auxiliary/util/u_gen_mipmap.c
+++ b/mesalib/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -59,14 +59,17 @@ struct gen_mipmap_state
    struct pipe_context *pipe;
    struct cso_context *cso;
 
-   struct pipe_blend_state blend;
-   struct pipe_depth_stencil_alpha_state depthstencil;
+   struct pipe_blend_state blend_keep_color, blend_write_color;
+   struct pipe_depth_stencil_alpha_state dsa_keep_depth, dsa_write_depth;
    struct pipe_rasterizer_state rasterizer;
    struct pipe_sampler_state sampler;
    struct pipe_vertex_element velem[2];
 
    void *vs;
-   void *fs[TGSI_TEXTURE_COUNT]; /**< Not all are used, but simplifies code */
+
+   /** Not all are used, but simplifies code */
+   void *fs_color[TGSI_TEXTURE_COUNT];
+   void *fs_depth[TGSI_TEXTURE_COUNT];
 
    struct pipe_resource *vbuf;  /**< quad vertices */
    unsigned vbuf_slot;
@@ -1272,11 +1275,16 @@ util_create_gen_mipmap(struct pipe_context *pipe,
    ctx->cso = cso;
 
    /* disabled blending/masking */
-   memset(&ctx->blend, 0, sizeof(ctx->blend));
-   ctx->blend.rt[0].colormask = PIPE_MASK_RGBA;
+   memset(&ctx->blend_keep_color, 0, sizeof(ctx->blend_keep_color));
+   memset(&ctx->blend_write_color, 0, sizeof(ctx->blend_write_color));
+   ctx->blend_write_color.rt[0].colormask = PIPE_MASK_RGBA;
 
    /* no-op depth/stencil/alpha */
-   memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
+   memset(&ctx->dsa_keep_depth, 0, sizeof(ctx->dsa_keep_depth));
+   memset(&ctx->dsa_write_depth, 0, sizeof(ctx->dsa_write_depth));
+   ctx->dsa_write_depth.depth.enabled = 1;
+   ctx->dsa_write_depth.depth.func = PIPE_FUNC_ALWAYS;
+   ctx->dsa_write_depth.depth.writemask = 1;
 
    /* rasterizer */
    memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
@@ -1318,14 +1326,25 @@ util_create_gen_mipmap(struct pipe_context *pipe,
  * Helper function to set the fragment shaders.
  */
 static INLINE void
-set_fragment_shader(struct gen_mipmap_state *ctx, uint type)
+set_fragment_shader(struct gen_mipmap_state *ctx, uint type,
+                    boolean output_depth)
 {
-   if (!ctx->fs[type])
-      ctx->fs[type] =
-         util_make_fragment_tex_shader(ctx->pipe, type,
-                                       TGSI_INTERPOLATE_LINEAR);
+   if (output_depth) {
+      if (!ctx->fs_depth[type])
+         ctx->fs_depth[type] =
+            util_make_fragment_tex_shader_writedepth(ctx->pipe, type,
+                                                     TGSI_INTERPOLATE_LINEAR);
 
-   cso_set_fragment_shader_handle(ctx->cso, ctx->fs[type]);
+      cso_set_fragment_shader_handle(ctx->cso, ctx->fs_depth[type]);
+   }
+   else {
+      if (!ctx->fs_color[type])
+         ctx->fs_color[type] =
+            util_make_fragment_tex_shader(ctx->pipe, type,
+                                          TGSI_INTERPOLATE_LINEAR);
+
+      cso_set_fragment_shader_handle(ctx->cso, ctx->fs_color[type]);
+   }
 }
 
 
@@ -1464,9 +1483,13 @@ util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
    struct pipe_context *pipe = ctx->pipe;
    unsigned i;
 
-   for (i = 0; i < Elements(ctx->fs); i++)
-      if (ctx->fs[i])
-         pipe->delete_fs_state(pipe, ctx->fs[i]);
+   for (i = 0; i < Elements(ctx->fs_color); i++)
+      if (ctx->fs_color[i])
+         pipe->delete_fs_state(pipe, ctx->fs_color[i]);
+
+   for (i = 0; i < Elements(ctx->fs_depth); i++)
+      if (ctx->fs_depth[i])
+         pipe->delete_fs_state(pipe, ctx->fs_depth[i]);
 
    if (ctx->vs)
       pipe->delete_vs_state(pipe, ctx->vs);
@@ -1500,6 +1523,7 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    uint dstLevel;
    uint offset;
    uint type;
+   boolean is_depth = util_format_is_depth_or_stencil(psv->format);
 
    /* The texture object should have room for the levels which we're
     * about to generate.
@@ -1538,7 +1562,9 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
 
    /* check if we can render in the texture's format */
    if (!screen->is_format_supported(screen, psv->format, pt->target,
-                                    pt->nr_samples, PIPE_BIND_RENDER_TARGET)) {
+                                    pt->nr_samples,
+                                    is_depth ? PIPE_BIND_DEPTH_STENCIL :
+                                               PIPE_BIND_RENDER_TARGET)) {
       fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
       return;
    }
@@ -1559,28 +1585,25 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
    cso_save_vertex_buffers(ctx->cso);
 
    /* bind our state */
-   cso_set_blend(ctx->cso, &ctx->blend);
-   cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
+   cso_set_blend(ctx->cso, is_depth ? &ctx->blend_keep_color :
+                                      &ctx->blend_write_color);
+   cso_set_depth_stencil_alpha(ctx->cso, is_depth ? &ctx->dsa_write_depth :
+                                                    &ctx->dsa_keep_depth);
    cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
    cso_set_vertex_elements(ctx->cso, 2, ctx->velem);
    cso_set_stream_outputs(ctx->cso, 0, NULL, 0);
 
-   set_fragment_shader(ctx, type);
+   set_fragment_shader(ctx, type, is_depth);
    set_vertex_shader(ctx);
    cso_set_geometry_shader_handle(ctx->cso, NULL);
 
    /* init framebuffer state */
    memset(&fb, 0, sizeof(fb));
-   fb.nr_cbufs = 1;
 
    /* set min/mag to same filter for faster sw speed */
    ctx->sampler.mag_img_filter = filter;
    ctx->sampler.min_img_filter = filter;
 
-   /*
-    * XXX for small mipmap levels, it may be faster to use the software
-    * fallback path...
-    */
    for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
       const uint srcLevel = dstLevel - 1;
       struct pipe_viewport_state vp;
@@ -1609,7 +1632,9 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
             layer = face;
 
          memset(&surf_templ, 0, sizeof(surf_templ));
-         u_surface_default_template(&surf_templ, pt, PIPE_BIND_RENDER_TARGET);
+         u_surface_default_template(&surf_templ, pt,
+                                    is_depth ? PIPE_BIND_DEPTH_STENCIL :
+                                               PIPE_BIND_RENDER_TARGET);
          surf_templ.u.tex.level = dstLevel;
          surf_templ.u.tex.first_layer = layer;
          surf_templ.u.tex.last_layer = layer;
@@ -1618,7 +1643,14 @@ util_gen_mipmap(struct gen_mipmap_state *ctx,
          /*
           * Setup framebuffer / dest surface
           */
-         fb.cbufs[0] = surf;
+         if (is_depth) {
+            fb.nr_cbufs = 0;
+            fb.zsbuf = surf;
+         }
+         else {
+            fb.nr_cbufs = 1;
+            fb.cbufs[0] = surf;
+         }
          fb.width = u_minify(pt->width0, dstLevel);
          fb.height = u_minify(pt->height0, dstLevel);
          cso_set_framebuffer(ctx->cso, &fb);
diff --git a/mesalib/src/gallium/auxiliary/util/u_inlines.h b/mesalib/src/gallium/auxiliary/util/u_inlines.h
index 2ec153c58..2d603e4dd 100644
--- a/mesalib/src/gallium/auxiliary/util/u_inlines.h
+++ b/mesalib/src/gallium/auxiliary/util/u_inlines.h
@@ -30,6 +30,7 @@
 
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
 #include "pipe/p_state.h"
 #include "pipe/p_screen.h"
 #include "util/u_debug.h"
@@ -545,6 +546,31 @@ util_query_clear_result(union pipe_query_result *result, unsigned type)
    }
 }
 
+/** Convert PIPE_TEXTURE_x to TGSI_TEXTURE_x */
+static INLINE unsigned
+util_pipe_tex_to_tgsi_tex(enum pipe_texture_target pipe_tex_target)
+{
+   switch (pipe_tex_target) {
+   case PIPE_TEXTURE_1D:
+      return TGSI_TEXTURE_1D;
+   case PIPE_TEXTURE_2D:
+      return TGSI_TEXTURE_2D;
+   case PIPE_TEXTURE_RECT:
+      return TGSI_TEXTURE_RECT;
+   case PIPE_TEXTURE_3D:
+      return TGSI_TEXTURE_3D;
+   case PIPE_TEXTURE_CUBE:
+      return TGSI_TEXTURE_CUBE;
+   case PIPE_TEXTURE_1D_ARRAY:
+      return TGSI_TEXTURE_1D_ARRAY;
+   case PIPE_TEXTURE_2D_ARRAY:
+      return TGSI_TEXTURE_2D_ARRAY;
+   default:
+      assert(0 && "unexpected texture target");
+      return TGSI_TEXTURE_UNKNOWN;
+   }
+}
+
 #ifdef __cplusplus
 }
 #endif
diff --git a/mesalib/src/gallium/auxiliary/util/u_simple_shaders.c b/mesalib/src/gallium/auxiliary/util/u_simple_shaders.c
index 320c0f7a8..3476b6ce0 100644
--- a/mesalib/src/gallium/auxiliary/util/u_simple_shaders.c
+++ b/mesalib/src/gallium/auxiliary/util/u_simple_shaders.c
@@ -211,6 +211,106 @@ util_make_fragment_tex_shader_writedepth(struct pipe_context *pipe,
 
 
 /**
+ * Make a simple fragment texture shader which reads the texture unit 0 and 1
+ * and writes it as depth and stencil, respectively.
+ */
+void *
+util_make_fragment_tex_shader_writedepthstencil(struct pipe_context *pipe,
+                                                unsigned tex_target,
+                                                unsigned interp_mode)
+{
+   struct ureg_program *ureg;
+   struct ureg_src depth_sampler, stencil_sampler;
+   struct ureg_src tex;
+   struct ureg_dst out, depth, stencil;
+   struct ureg_src imm;
+
+   ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
+   if (ureg == NULL)
+      return NULL;
+
+   depth_sampler = ureg_DECL_sampler( ureg, 0 );
+   stencil_sampler = ureg_DECL_sampler( ureg, 1 );
+
+   tex = ureg_DECL_fs_input( ureg,
+                             TGSI_SEMANTIC_GENERIC, 0,
+                             interp_mode );
+
+   out = ureg_DECL_output( ureg,
+                           TGSI_SEMANTIC_COLOR,
+                           0 );
+
+   depth = ureg_DECL_output( ureg,
+                             TGSI_SEMANTIC_POSITION,
+                             0 );
+
+   stencil = ureg_DECL_output( ureg,
+                             TGSI_SEMANTIC_STENCIL,
+                             0 );
+
+   imm = ureg_imm4f( ureg, 0, 0, 0, 1 );
+
+   ureg_MOV( ureg, out, imm );
+
+   ureg_TEX( ureg,
+             ureg_writemask(depth, TGSI_WRITEMASK_Z),
+             tex_target, tex, depth_sampler );
+   ureg_TEX( ureg,
+             ureg_writemask(stencil, TGSI_WRITEMASK_Y),
+             tex_target, tex, stencil_sampler );
+   ureg_END( ureg );
+
+   return ureg_create_shader_and_destroy( ureg, pipe );
+}
+
+
+/**
+ * Make a simple fragment texture shader which reads a texture and writes it
+ * as stencil.
+ */
+void *
+util_make_fragment_tex_shader_writestencil(struct pipe_context *pipe,
+                                           unsigned tex_target,
+                                           unsigned interp_mode)
+{
+   struct ureg_program *ureg;
+   struct ureg_src stencil_sampler;
+   struct ureg_src tex;
+   struct ureg_dst out, stencil;
+   struct ureg_src imm;
+
+   ureg = ureg_create( TGSI_PROCESSOR_FRAGMENT );
+   if (ureg == NULL)
+      return NULL;
+
+   stencil_sampler = ureg_DECL_sampler( ureg, 0 );
+
+   tex = ureg_DECL_fs_input( ureg,
+                             TGSI_SEMANTIC_GENERIC, 0,
+                             interp_mode );
+
+   out = ureg_DECL_output( ureg,
+                           TGSI_SEMANTIC_COLOR,
+                           0 );
+
+   stencil = ureg_DECL_output( ureg,
+                             TGSI_SEMANTIC_STENCIL,
+                             0 );
+
+   imm = ureg_imm4f( ureg, 0, 0, 0, 1 );
+
+   ureg_MOV( ureg, out, imm );
+
+   ureg_TEX( ureg,
+             ureg_writemask(stencil, TGSI_WRITEMASK_Y),
+             tex_target, tex, stencil_sampler );
+   ureg_END( ureg );
+
+   return ureg_create_shader_and_destroy( ureg, pipe );
+}
+
+
+/**
  * Make simple fragment color pass-through shader.
  */
 void *
diff --git a/mesalib/src/gallium/auxiliary/util/u_simple_shaders.h b/mesalib/src/gallium/auxiliary/util/u_simple_shaders.h
index 5f31b72c4..0764998a3 100644
--- a/mesalib/src/gallium/auxiliary/util/u_simple_shaders.h
+++ b/mesalib/src/gallium/auxiliary/util/u_simple_shaders.h
@@ -75,6 +75,18 @@ util_make_fragment_tex_shader_writedepth(struct pipe_context *pipe,
 
 
 extern void *
+util_make_fragment_tex_shader_writedepthstencil(struct pipe_context *pipe,
+                                                unsigned tex_target,
+                                                unsigned interp_mode);
+
+
+extern void *
+util_make_fragment_tex_shader_writestencil(struct pipe_context *pipe,
+                                           unsigned tex_target,
+                                           unsigned interp_mode);
+
+
+extern void *
 util_make_fragment_passthrough_shader(struct pipe_context *pipe);
 
 
diff --git a/mesalib/src/glsl/ir.h b/mesalib/src/glsl/ir.h
index b54e2f2e0..9bbf3b7ef 100644
--- a/mesalib/src/glsl/ir.h
+++ b/mesalib/src/glsl/ir.h
@@ -1341,7 +1341,8 @@ enum ir_texture_opcode {
 class ir_texture : public ir_rvalue {
 public:
    ir_texture(enum ir_texture_opcode op)
-      : op(op), projector(NULL), shadow_comparitor(NULL), offset(NULL)
+      : op(op), coordinate(NULL), projector(NULL), shadow_comparitor(NULL),
+        offset(NULL)
    {
       this->ir_type = ir_type_texture;
    }
diff --git a/mesalib/src/glsl/ir_builder.cpp b/mesalib/src/glsl/ir_builder.cpp
index 9a16c90e5..d96e25c18 100644
--- a/mesalib/src/glsl/ir_builder.cpp
+++ b/mesalib/src/glsl/ir_builder.cpp
@@ -77,6 +77,21 @@ swizzle(operand a, int swizzle, int components)
 }
 
 ir_swizzle *
+swizzle_for_size(operand a, int components)
+{
+   void *mem_ctx = ralloc_parent(a.val);
+
+   if (a.val->type->vector_elements < components)
+      components = a.val->type->vector_elements;
+
+   unsigned s[4] = { 0, 1, 2, 3 };
+   for (int i = components; i < 4; i++)
+      s[i] = components - 1;
+
+   return new(mem_ctx) ir_swizzle(a.val, s, components);
+}
+
+ir_swizzle *
 swizzle_xxxx(operand a)
 {
    return swizzle(a, SWIZZLE_XXXX, 4);
@@ -143,6 +158,14 @@ swizzle_xyzw(operand a)
 }
 
 ir_expression *
+expr(ir_expression_operation op, operand a)
+{
+   void *mem_ctx = ralloc_parent(a.val);
+
+   return new(mem_ctx) ir_expression(op, a.val);
+}
+
+ir_expression *
 expr(ir_expression_operation op, operand a, operand b)
 {
    void *mem_ctx = ralloc_parent(a.val);
diff --git a/mesalib/src/glsl/ir_builder.h b/mesalib/src/glsl/ir_builder.h
index 0ebcbab95..7a0a196ee 100644
--- a/mesalib/src/glsl/ir_builder.h
+++ b/mesalib/src/glsl/ir_builder.h
@@ -83,6 +83,7 @@ public:
 ir_assignment *assign(deref lhs, operand rhs);
 ir_assignment *assign(deref lhs, operand rhs, int writemask);
 
+ir_expression *expr(ir_expression_operation op, operand a);
 ir_expression *expr(ir_expression_operation op, operand a, operand b);
 ir_expression *add(operand a, operand b);
 ir_expression *sub(operand a, operand b);
@@ -90,6 +91,11 @@ ir_expression *mul(operand a, operand b);
 ir_expression *dot(operand a, operand b);
 ir_expression *saturate(operand a);
 
+/**
+ * Swizzle away later components, but preserve the ordering.
+ */
+ir_swizzle *swizzle_for_size(operand a, int components);
+
 ir_swizzle *swizzle_xxxx(operand a);
 ir_swizzle *swizzle_yyyy(operand a);
 ir_swizzle *swizzle_zzzz(operand a);
diff --git a/mesalib/src/glsl/linker.cpp b/mesalib/src/glsl/linker.cpp
index 310944752..f06298cf6 100644
--- a/mesalib/src/glsl/linker.cpp
+++ b/mesalib/src/glsl/linker.cpp
@@ -1439,8 +1439,7 @@ public:
    bool accumulate_num_outputs(struct gl_shader_program *prog, unsigned *count);
    bool store(struct gl_context *ctx, struct gl_shader_program *prog,
               struct gl_transform_feedback_info *info, unsigned buffer,
-              unsigned varying, const unsigned max_outputs) const;
-
+              const unsigned max_outputs) const;
 
    /**
     * True if assign_location() has been called for this object.
@@ -1450,6 +1449,16 @@ public:
       return this->location != -1;
    }
 
+   bool is_next_buffer_separator() const
+   {
+      return this->next_buffer_separator;
+   }
+
+   bool is_varying() const
+   {
+      return !this->next_buffer_separator && !this->skip_components;
+   }
+
    /**
     * Determine whether this object refers to the variable var.
     */
@@ -1527,6 +1536,17 @@ private:
     * glGetTransformFeedbackVarying().
     */
    unsigned size;
+
+   /**
+    * How many components to skip. If non-zero, this is
+    * gl_SkipComponents{1,2,3,4} from ARB_transform_feedback3.
+    */
+   unsigned skip_components;
+
+   /**
+    * Whether this is gl_NextBuffer from ARB_transform_feedback3.
+    */
+   bool next_buffer_separator;
 };
 
 
@@ -1546,7 +1566,31 @@ tfeedback_decl::init(struct gl_context *ctx, struct gl_shader_program *prog,
    this->location = -1;
    this->orig_name = input;
    this->is_clip_distance_mesa = false;
+   this->skip_components = 0;
+   this->next_buffer_separator = false;
 
+   if (ctx->Extensions.ARB_transform_feedback3) {
+      /* Parse gl_NextBuffer. */
+      if (strcmp(input, "gl_NextBuffer") == 0) {
+         this->next_buffer_separator = true;
+         return true;
+      }
+
+      /* Parse gl_SkipComponents. */
+      if (strcmp(input, "gl_SkipComponents1") == 0)
+         this->skip_components = 1;
+      else if (strcmp(input, "gl_SkipComponents2") == 0)
+         this->skip_components = 2;
+      else if (strcmp(input, "gl_SkipComponents3") == 0)
+         this->skip_components = 3;
+      else if (strcmp(input, "gl_SkipComponents4") == 0)
+         this->skip_components = 4;
+
+      if (this->skip_components)
+         return true;
+   }
+
+   /* Parse a declaration. */
    const char *bracket = strrchr(input, '[');
 
    if (bracket) {
@@ -1581,6 +1625,8 @@ tfeedback_decl::init(struct gl_context *ctx, struct gl_shader_program *prog,
 bool
 tfeedback_decl::is_same(const tfeedback_decl &x, const tfeedback_decl &y)
 {
+   assert(x.is_varying() && y.is_varying());
+
    if (strcmp(x.var_name, y.var_name) != 0)
       return false;
    if (x.is_subscripted != y.is_subscripted)
@@ -1603,6 +1649,8 @@ tfeedback_decl::assign_location(struct gl_context *ctx,
                                 struct gl_shader_program *prog,
                                 ir_variable *output_var)
 {
+   assert(this->is_varying());
+
    if (output_var->type->is_array()) {
       /* Array variable */
       const unsigned matrix_cols =
@@ -1677,6 +1725,10 @@ bool
 tfeedback_decl::accumulate_num_outputs(struct gl_shader_program *prog,
                                        unsigned *count)
 {
+   if (!this->is_varying()) {
+      return true;
+   }
+
    if (!this->is_assigned()) {
       /* From GL_EXT_transform_feedback:
        *   A program will fail to link if:
@@ -1709,9 +1761,16 @@ tfeedback_decl::accumulate_num_outputs(struct gl_shader_program *prog,
 bool
 tfeedback_decl::store(struct gl_context *ctx, struct gl_shader_program *prog,
                       struct gl_transform_feedback_info *info,
-                      unsigned buffer,
-                      unsigned varying, const unsigned max_outputs) const
+                      unsigned buffer, const unsigned max_outputs) const
 {
+   assert(!this->next_buffer_separator);
+
+   /* Handle gl_SkipComponents. */
+   if (this->skip_components) {
+      info->BufferStride[buffer] += this->skip_components;
+      return true;
+   }
+
    /* From GL_EXT_transform_feedback:
     *   A program will fail to link if:
     *
@@ -1757,9 +1816,9 @@ tfeedback_decl::store(struct gl_context *ctx, struct gl_shader_program *prog,
    }
    assert(components_so_far == this->num_components());
 
-   info->Varyings[varying].Name = ralloc_strdup(prog, this->orig_name);
-   info->Varyings[varying].Type = this->type;
-   info->Varyings[varying].Size = this->size;
+   info->Varyings[info->NumVarying].Name = ralloc_strdup(prog, this->orig_name);
+   info->Varyings[info->NumVarying].Type = this->type;
+   info->Varyings[info->NumVarying].Size = this->size;
    info->NumVarying++;
 
    return true;
@@ -1781,6 +1840,10 @@ parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog,
    for (unsigned i = 0; i < num_names; ++i) {
       if (!decls[i].init(ctx, prog, mem_ctx, varying_names[i]))
          return false;
+
+      if (!decls[i].is_varying())
+         continue;
+
       /* From GL_EXT_transform_feedback:
        *   A program will fail to link if:
        *
@@ -1792,6 +1855,9 @@ parse_tfeedback_decls(struct gl_context *ctx, struct gl_shader_program *prog,
        * feedback of arrays would be useless otherwise.
        */
       for (unsigned j = 0; j < i; ++j) {
+         if (!decls[j].is_varying())
+            continue;
+
          if (tfeedback_decl::is_same(decls[i], decls[j])) {
             linker_error(prog, "Transform feedback varying %s specified "
                          "more than once.", varying_names[i]);
@@ -1948,6 +2014,9 @@ assign_varying_locations(struct gl_context *ctx,
       }
 
       for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
+         if (!tfeedback_decls[i].is_varying())
+            continue;
+
          if (!tfeedback_decls[i].is_assigned() &&
              tfeedback_decls[i].matches_var(output_var)) {
             if (output_var->location == -1) {
@@ -2059,9 +2128,6 @@ store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog,
    memset(&prog->LinkedTransformFeedback, 0,
           sizeof(prog->LinkedTransformFeedback));
 
-   prog->LinkedTransformFeedback.NumBuffers =
-      separate_attribs_mode ? num_tfeedback_decls : 1;
-
    prog->LinkedTransformFeedback.Varyings =
       rzalloc_array(prog,
 		    struct gl_transform_feedback_varying_info,
@@ -2077,14 +2143,37 @@ store_tfeedback_info(struct gl_context *ctx, struct gl_shader_program *prog,
                     struct gl_transform_feedback_output,
                     num_outputs);
 
-   for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
-      unsigned buffer = separate_attribs_mode ? i : 0;
-      if (!tfeedback_decls[i].store(ctx, prog, &prog->LinkedTransformFeedback,
-                                    buffer, i, num_outputs))
-         return false;
+   unsigned num_buffers = 0;
+
+   if (separate_attribs_mode) {
+      /* GL_SEPARATE_ATTRIBS */
+      for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
+         if (!tfeedback_decls[i].store(ctx, prog, &prog->LinkedTransformFeedback,
+                                       num_buffers, num_outputs))
+            return false;
+
+         num_buffers++;
+      }
+   }
+   else {
+      /* GL_INVERLEAVED_ATTRIBS */
+      for (unsigned i = 0; i < num_tfeedback_decls; ++i) {
+         if (tfeedback_decls[i].is_next_buffer_separator()) {
+            num_buffers++;
+            continue;
+         }
+
+         if (!tfeedback_decls[i].store(ctx, prog,
+                                       &prog->LinkedTransformFeedback,
+                                       num_buffers, num_outputs))
+            return false;
+      }
+      num_buffers++;
    }
+
    assert(prog->LinkedTransformFeedback.NumOutputs == num_outputs);
 
+   prog->LinkedTransformFeedback.NumBuffers = num_buffers;
    return true;
 }
 
diff --git a/mesalib/src/mapi/glapi/gen/gl_API.xml b/mesalib/src/mapi/glapi/gen/gl_API.xml
index 01460b49b..8bdc47908 100644
--- a/mesalib/src/mapi/glapi/gen/gl_API.xml
+++ b/mesalib/src/mapi/glapi/gen/gl_API.xml
@@ -8005,7 +8005,36 @@
 
 <xi:include href="ARB_vertex_type_2_10_10_10_rev.xml" xmlns:xi="http://www.w3.org/2001/XInclude"/>
 
-<!-- ARB extensions #86...#94 -->
+<!-- ARB extensions #86...#93 -->
+
+<category name="GL_ARB_transform_feedback3" number="94">
+  <enum name="MAX_TRANSFORM_FEEDBACK_BUFFERS" value="0x8E70"/>
+  <enum name="MAX_VERTEX_STREAMS"             value="0x8E71"/>
+
+  <function name="DrawTransformFeedbackStream" offset="assign">
+    <param name="mode" type="GLenum"/>
+    <param name="id" type="GLuint"/>
+    <param name="stream" type="GLuint"/>
+  </function>
+
+  <function name="BeginQueryIndexed" offset="assign">
+    <param name="target" type="GLenum"/>
+    <param name="index" type="GLuint"/>
+    <param name="id" type="GLuint"/>
+  </function>
+
+  <function name="EndQueryIndexed" offset="assign">
+    <param name="target" type="GLenum"/>
+    <param name="index" type="GLuint"/>
+  </function>
+
+  <function name="GetQueryIndexediv" offset="assign">
+    <param name="target" type="GLenum"/>
+    <param name="index" type="GLuint"/>
+    <param name="pname" type="GLenum"/>
+    <param name="params" type="GLint *"/>
+  </function>
+</category>
 
 <xi:include href="ARB_ES2_compatibility.xml" xmlns:xi="http://www.w3.org/2001/XInclude"/>
 
@@ -8015,7 +8044,24 @@
 
 <xi:include href="ARB_robustness.xml" xmlns:xi="http://www.w3.org/2001/XInclude"/>
 
-<!-- ARB extensions #106...#116 -->
+<!-- ARB extensions #106...#108 -->
+
+<category name="GL_ARB_transform_feedback_instanced" number="109">
+  <function name="DrawTransformFeedbackInstanced" offset="assign">
+    <param name="mode" type="GLenum"/>
+    <param name="id" type="GLuint"/>
+    <param name="primcount" type="GLsizei"/>
+  </function>
+
+  <function name="DrawTransformFeedbackStreamInstanced" offset="assign">
+    <param name="mode" type="GLenum"/>
+    <param name="id" type="GLuint"/>
+    <param name="stream" type="GLuint"/>
+    <param name="primcount" type="GLsizei"/>
+  </function>
+</category>
+
+<!-- ARB extensions #110...#116 -->
 
 <xi:include href="ARB_base_instance.xml" xmlns:xi="http://www.w3.org/2001/XInclude"/>
 
diff --git a/mesalib/src/mesa/main/api_validate.c b/mesalib/src/mesa/main/api_validate.c
index cf6aaf0c6..ece0a2b4a 100644
--- a/mesalib/src/mesa/main/api_validate.c
+++ b/mesalib/src/mesa/main/api_validate.c
@@ -593,26 +593,42 @@ _mesa_validate_DrawElementsInstanced(struct gl_context *ctx,
 GLboolean
 _mesa_validate_DrawTransformFeedback(struct gl_context *ctx,
                                      GLenum mode,
-                                     struct gl_transform_feedback_object *obj)
+                                     struct gl_transform_feedback_object *obj,
+                                     GLuint stream,
+                                     GLsizei numInstances)
 {
    ASSERT_OUTSIDE_BEGIN_END_WITH_RETVAL(ctx, GL_FALSE);
    FLUSH_CURRENT(ctx, 0);
 
-   if (!_mesa_valid_prim_mode(ctx, mode, "glDrawTransformFeedback")) {
+   if (!_mesa_valid_prim_mode(ctx, mode, "glDrawTransformFeedback*(mode)")) {
       return GL_FALSE;
    }
 
    if (!obj) {
-      _mesa_error(ctx, GL_INVALID_VALUE, "glDrawTransformFeedback(name)");
+      _mesa_error(ctx, GL_INVALID_VALUE, "glDrawTransformFeedback*(name)");
       return GL_FALSE;
    }
 
    if (!obj->EndedAnytime) {
-      _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawTransformFeedback");
+      _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawTransformFeedback*");
       return GL_FALSE;
    }
 
-   if (!check_valid_to_render(ctx, "glDrawTransformFeedback")) {
+   if (stream >= ctx->Const.MaxVertexStreams) {
+      _mesa_error(ctx, GL_INVALID_VALUE,
+                  "glDrawTransformFeedbackStream*(index>=MaxVertexStream)");
+      return GL_FALSE;
+   }
+
+   if (numInstances <= 0) {
+      if (numInstances < 0)
+         _mesa_error(ctx, GL_INVALID_VALUE,
+                     "glDrawTransformFeedback*Instanced(numInstances=%d)",
+                     numInstances);
+      return GL_FALSE;
+   }
+
+   if (!check_valid_to_render(ctx, "glDrawTransformFeedback*")) {
       return GL_FALSE;
    }
 
diff --git a/mesalib/src/mesa/main/api_validate.h b/mesalib/src/mesa/main/api_validate.h
index 59f329768..8dc8df4b2 100644
--- a/mesalib/src/mesa/main/api_validate.h
+++ b/mesalib/src/mesa/main/api_validate.h
@@ -83,7 +83,9 @@ _mesa_validate_DrawElementsInstanced(struct gl_context *ctx,
 extern GLboolean
 _mesa_validate_DrawTransformFeedback(struct gl_context *ctx,
                                      GLenum mode,
-                                     struct gl_transform_feedback_object *obj);
+                                     struct gl_transform_feedback_object *obj,
+                                     GLuint stream,
+                                     GLsizei numInstances);
 
 #endif
 
diff --git a/mesalib/src/mesa/main/context.c b/mesalib/src/mesa/main/context.c
index 7de714b81..d5ccce076 100644
--- a/mesalib/src/mesa/main/context.c
+++ b/mesalib/src/mesa/main/context.c
@@ -655,6 +655,7 @@ _mesa_init_constants(struct gl_context *ctx)
    ctx->Const.MaxTransformFeedbackBuffers = MAX_FEEDBACK_BUFFERS;
    ctx->Const.MaxTransformFeedbackSeparateComponents = 4 * MAX_FEEDBACK_ATTRIBS;
    ctx->Const.MaxTransformFeedbackInterleavedComponents = 4 * MAX_FEEDBACK_ATTRIBS;
+   ctx->Const.MaxVertexStreams = 1;
 
    /** GL_ARB_uniform_buffer_object */
    ctx->Const.MaxCombinedUniformBlocks = 36;
diff --git a/mesalib/src/mesa/main/dd.h b/mesalib/src/mesa/main/dd.h
index c8a765f47..e60d019bb 100644
--- a/mesalib/src/mesa/main/dd.h
+++ b/mesalib/src/mesa/main/dd.h
@@ -1052,6 +1052,14 @@ typedef struct {
                                                                   GLsizei primcount, GLint basevertex,
                                                                   GLuint baseinstance);
    void (GLAPIENTRYP DrawTransformFeedback)(GLenum mode, GLuint name);
+   void (GLAPIENTRYP DrawTransformFeedbackStream)(GLenum mode, GLuint name,
+                                                  GLuint stream);
+   void (GLAPIENTRYP DrawTransformFeedbackInstanced)(GLenum mode, GLuint name,
+                                                     GLsizei primcount);
+   void (GLAPIENTRYP DrawTransformFeedbackStreamInstanced)(GLenum mode,
+                                                           GLuint name,
+                                                           GLuint stream,
+                                                           GLsizei primcount);
    /*@}*/
 
    /**
diff --git a/mesalib/src/mesa/main/dlist.c b/mesalib/src/mesa/main/dlist.c
index a827d132d..40961b15c 100644
--- a/mesalib/src/mesa/main/dlist.c
+++ b/mesalib/src/mesa/main/dlist.c
@@ -472,6 +472,15 @@ typedef enum
    /* ARB_timer_query */
    OPCODE_QUERY_COUNTER,
 
+   /* ARB_transform_feedback3 */
+   OPCODE_BEGIN_QUERY_INDEXED,
+   OPCODE_END_QUERY_INDEXED,
+   OPCODE_DRAW_TRANSFORM_FEEDBACK_STREAM,
+
+   /* ARB_transform_feedback_instanced */
+   OPCODE_DRAW_TRANSFORM_FEEDBACK_INSTANCED,
+   OPCODE_DRAW_TRANSFORM_FEEDBACK_STREAM_INSTANCED,
+
    /* The following three are meta instructions */
    OPCODE_ERROR,                /* raise compiled-in error */
    OPCODE_CONTINUE,
@@ -504,6 +513,7 @@ union gl_dlist_node
    GLuint ui;
    GLenum e;
    GLfloat f;
+   GLsizei si;
    GLvoid *data;
    void *next;                  /* If prev node's opcode==OPCODE_CONTINUE */
 };
@@ -5338,7 +5348,6 @@ save_BeginQueryARB(GLenum target, GLuint id)
    }
 }
 
-
 static void GLAPIENTRY
 save_EndQueryARB(GLenum target)
 {
@@ -5354,7 +5363,6 @@ save_EndQueryARB(GLenum target)
    }
 }
 
-
 static void GLAPIENTRY
 save_QueryCounter(GLuint id, GLenum target)
 {
@@ -5371,6 +5379,39 @@ save_QueryCounter(GLuint id, GLenum target)
    }
 }
 
+static void GLAPIENTRY
+save_BeginQueryIndexed(GLenum target, GLuint index, GLuint id)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   Node *n;
+   ASSERT_OUTSIDE_SAVE_BEGIN_END_AND_FLUSH(ctx);
+   n = alloc_instruction(ctx, OPCODE_BEGIN_QUERY_INDEXED, 3);
+   if (n) {
+      n[1].e = target;
+      n[2].ui = index;
+      n[3].ui = id;
+   }
+   if (ctx->ExecuteFlag) {
+      CALL_BeginQueryIndexed(ctx->Exec, (target, index, id));
+   }
+}
+
+static void GLAPIENTRY
+save_EndQueryIndexed(GLenum target, GLuint index)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   Node *n;
+   ASSERT_OUTSIDE_SAVE_BEGIN_END_AND_FLUSH(ctx);
+   n = alloc_instruction(ctx, OPCODE_END_QUERY_INDEXED, 2);
+   if (n) {
+      n[1].e = target;
+      n[2].ui = index;
+   }
+   if (ctx->ExecuteFlag) {
+      CALL_EndQueryIndexed(ctx->Exec, (target, index));
+   }
+}
+
 #endif /* FEATURE_queryobj */
 
 
@@ -6441,6 +6482,60 @@ save_DrawTransformFeedback(GLenum mode, GLuint name)
    }
 }
 
+static void GLAPIENTRY
+save_DrawTransformFeedbackStream(GLenum mode, GLuint name, GLuint stream)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   Node *n;
+   ASSERT_OUTSIDE_SAVE_BEGIN_END_AND_FLUSH(ctx);
+   n = alloc_instruction(ctx, OPCODE_DRAW_TRANSFORM_FEEDBACK_STREAM, 3);
+   if (n) {
+      n[1].e = mode;
+      n[2].ui = name;
+      n[3].ui = stream;
+   }
+   if (ctx->ExecuteFlag) {
+      CALL_DrawTransformFeedbackStream(ctx->Exec, (mode, name, stream));
+   }
+}
+
+static void GLAPIENTRY
+save_DrawTransformFeedbackInstanced(GLenum mode, GLuint name,
+                                    GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   Node *n;
+   ASSERT_OUTSIDE_SAVE_BEGIN_END_AND_FLUSH(ctx);
+   n = alloc_instruction(ctx, OPCODE_DRAW_TRANSFORM_FEEDBACK_INSTANCED, 3);
+   if (n) {
+      n[1].e = mode;
+      n[2].ui = name;
+      n[3].si = primcount;
+   }
+   if (ctx->ExecuteFlag) {
+      CALL_DrawTransformFeedbackInstanced(ctx->Exec, (mode, name, primcount));
+   }
+}
+
+static void GLAPIENTRY
+save_DrawTransformFeedbackStreamInstanced(GLenum mode, GLuint name,
+                                          GLuint stream, GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   Node *n;
+   ASSERT_OUTSIDE_SAVE_BEGIN_END_AND_FLUSH(ctx);
+   n = alloc_instruction(ctx, OPCODE_DRAW_TRANSFORM_FEEDBACK_STREAM_INSTANCED, 4);
+   if (n) {
+      n[1].e = mode;
+      n[2].ui = name;
+      n[3].ui = stream;
+      n[4].si = primcount;
+   }
+   if (ctx->ExecuteFlag) {
+      CALL_DrawTransformFeedbackStreamInstanced(ctx->Exec, (mode, name, stream,
+                                                            primcount));
+   }
+}
 
 /* aka UseProgram() */
 static void GLAPIENTRY
@@ -8368,6 +8463,12 @@ execute_list(struct gl_context *ctx, GLuint list)
          case OPCODE_QUERY_COUNTER:
             CALL_QueryCounter(ctx->Exec, (n[1].ui, n[2].e));
             break;
+         case OPCODE_BEGIN_QUERY_INDEXED:
+            CALL_BeginQueryIndexed(ctx->Exec, (n[1].e, n[2].ui, n[3].ui));
+            break;
+         case OPCODE_END_QUERY_INDEXED:
+            CALL_EndQueryIndexed(ctx->Exec, (n[1].e, n[2].ui));
+            break;
 #endif
          case OPCODE_DRAW_BUFFERS_ARB:
             {
@@ -8686,6 +8787,18 @@ execute_list(struct gl_context *ctx, GLuint list)
          case OPCODE_DRAW_TRANSFORM_FEEDBACK:
             CALL_DrawTransformFeedback(ctx->Exec, (n[1].e, n[2].ui));
             break;
+         case OPCODE_DRAW_TRANSFORM_FEEDBACK_STREAM:
+            CALL_DrawTransformFeedbackStream(ctx->Exec,
+                                             (n[1].e, n[2].ui, n[3].ui));
+            break;
+         case OPCODE_DRAW_TRANSFORM_FEEDBACK_INSTANCED:
+            CALL_DrawTransformFeedbackInstanced(ctx->Exec,
+                                                (n[1].e, n[2].ui, n[3].si));
+            break;
+         case OPCODE_DRAW_TRANSFORM_FEEDBACK_STREAM_INSTANCED:
+            CALL_DrawTransformFeedbackStreamInstanced(ctx->Exec,
+                                       (n[1].e, n[2].ui, n[3].ui, n[4].si));
+            break;
 
 
          case OPCODE_BIND_SAMPLER:
@@ -10461,6 +10574,15 @@ _mesa_create_save_table(void)
    SET_PauseTransformFeedback(table, save_PauseTransformFeedback);
    SET_ResumeTransformFeedback(table, save_ResumeTransformFeedback);
    SET_DrawTransformFeedback(table, save_DrawTransformFeedback);
+   SET_DrawTransformFeedbackStream(table, save_DrawTransformFeedbackStream);
+   SET_DrawTransformFeedbackInstanced(table,
+                                      save_DrawTransformFeedbackInstanced);
+   SET_DrawTransformFeedbackStreamInstanced(table,
+                                save_DrawTransformFeedbackStreamInstanced);
+#if FEATURE_queryobj
+   SET_BeginQueryIndexed(table, save_BeginQueryIndexed);
+   SET_EndQueryIndexed(table, save_EndQueryIndexed);
+#endif
 #endif
 
    /* GL_ARB_instanced_arrays */
diff --git a/mesalib/src/mesa/main/extensions.c b/mesalib/src/mesa/main/extensions.c
index dbc2813f2..0675ce75d 100644
--- a/mesalib/src/mesa/main/extensions.c
+++ b/mesalib/src/mesa/main/extensions.c
@@ -141,6 +141,8 @@ static const struct extension extension_table[] = {
    { "GL_ARB_texture_swizzle",                     o(EXT_texture_swizzle),                     GL,             2008 },
    { "GL_ARB_timer_query",                         o(ARB_timer_query),                         GL,             2010 },
    { "GL_ARB_transform_feedback2",                 o(ARB_transform_feedback2),                 GL,             2010 },
+   { "GL_ARB_transform_feedback3",                 o(ARB_transform_feedback3),                 GL,             2010 },
+   { "GL_ARB_transform_feedback_instanced",        o(ARB_transform_feedback_instanced),        GL,             2011 },
    { "GL_ARB_transpose_matrix",                    o(ARB_transpose_matrix),                    GL,             1999 },
    { "GL_ARB_uniform_buffer_object",               o(ARB_uniform_buffer_object),               GL,             2009 },
    { "GL_ARB_vertex_array_bgra",                   o(EXT_vertex_array_bgra),                   GL,             2008 },
diff --git a/mesalib/src/mesa/main/fbobject.c b/mesalib/src/mesa/main/fbobject.c
index cfaea62bb..4370c7218 100644
--- a/mesalib/src/mesa/main/fbobject.c
+++ b/mesalib/src/mesa/main/fbobject.c
@@ -1296,6 +1296,9 @@ _mesa_base_fbo_format(struct gl_context *ctx, GLenum internalFormat)
 
    case GL_RGB10_A2UI:
       return ctx->Extensions.ARB_texture_rgb10_a2ui ? GL_RGBA : 0;
+
+   case GL_RGB565:
+      return ctx->Extensions.ARB_ES2_compatibility ? GL_RGB : 0;
    default:
       return 0;
    }
diff --git a/mesalib/src/mesa/main/formats.c b/mesalib/src/mesa/main/formats.c
index ccc0b1707..c65c9c2e1 100644
--- a/mesalib/src/mesa/main/formats.c
+++ b/mesalib/src/mesa/main/formats.c
@@ -1860,8 +1860,7 @@ _mesa_format_image_size(gl_format format, GLsizei width,
       const GLuint wblocks = (width + bw - 1) / bw;
       const GLuint hblocks = (height + bh - 1) / bh;
       const GLuint sz = wblocks * hblocks * info->BytesPerBlock;
-      assert(depth == 1);
-      return sz;
+      return sz * depth;
    }
    else {
       /* non-compressed */
@@ -1887,8 +1886,7 @@ _mesa_format_image_size64(gl_format format, GLsizei width,
       const uint64_t wblocks = (width + bw - 1) / bw;
       const uint64_t hblocks = (height + bh - 1) / bh;
       const uint64_t sz = wblocks * hblocks * info->BytesPerBlock;
-      assert(depth == 1);
-      return sz;
+      return sz * depth;
    }
    else {
       /* non-compressed */
diff --git a/mesalib/src/mesa/main/get.c b/mesalib/src/mesa/main/get.c
index cff4cf394..15de321e4 100644
--- a/mesalib/src/mesa/main/get.c
+++ b/mesalib/src/mesa/main/get.c
@@ -328,6 +328,7 @@ EXTRA_EXT(ARB_sync);
 EXTRA_EXT(ARB_vertex_shader);
 EXTRA_EXT(EXT_transform_feedback);
 EXTRA_EXT(ARB_transform_feedback2);
+EXTRA_EXT(ARB_transform_feedback3);
 EXTRA_EXT(EXT_pixel_buffer_object);
 EXTRA_EXT(ARB_vertex_program);
 EXTRA_EXT2(NV_point_sprite, ARB_point_sprite);
@@ -1247,6 +1248,14 @@ static const struct value_desc values[] = {
    { GL_TRANSFORM_FEEDBACK_BINDING, LOC_CUSTOM, TYPE_INT, 0,
      extra_ARB_transform_feedback2 },
 
+   /* GL_ARB_transform_feedback3 */
+   { GL_MAX_TRANSFORM_FEEDBACK_BUFFERS,
+     CONTEXT_INT(Const.MaxTransformFeedbackBuffers),
+     extra_ARB_transform_feedback3 },
+   { GL_MAX_VERTEX_STREAMS,
+     CONTEXT_INT(Const.MaxVertexStreams),
+     extra_ARB_transform_feedback3 },
+
    /* GL_ARB_geometry_shader4 */
    { GL_MAX_GEOMETRY_TEXTURE_IMAGE_UNITS_ARB,
      CONTEXT_INT(Const.MaxGeometryTextureImageUnits),
diff --git a/mesalib/src/mesa/main/image.c b/mesalib/src/mesa/main/image.c
index b6c2645e9..678dfeb2b 100644
--- a/mesalib/src/mesa/main/image.c
+++ b/mesalib/src/mesa/main/image.c
@@ -770,6 +770,7 @@ _mesa_is_color_format(GLenum format)
       case GL_R3_G3_B2:
       case GL_RGB4:
       case GL_RGB5:
+      case GL_RGB565:
       case GL_RGB8:
       case GL_RGB10:
       case GL_RGB12:
diff --git a/mesalib/src/mesa/main/mtypes.h b/mesalib/src/mesa/main/mtypes.h
index bdbb5137e..1f74e6a83 100644
--- a/mesalib/src/mesa/main/mtypes.h
+++ b/mesalib/src/mesa/main/mtypes.h
@@ -2837,6 +2837,7 @@ struct gl_constants
    GLuint MaxTransformFeedbackBuffers;
    GLuint MaxTransformFeedbackSeparateComponents;
    GLuint MaxTransformFeedbackInterleavedComponents;
+   GLuint MaxVertexStreams;
 
    /** GL_EXT_gpu_shader4 */
    GLint MinProgramTexelOffset, MaxProgramTexelOffset;
@@ -2937,6 +2938,8 @@ struct gl_extensions
    GLboolean ARB_texture_storage;
    GLboolean ARB_timer_query;
    GLboolean ARB_transform_feedback2;
+   GLboolean ARB_transform_feedback3;
+   GLboolean ARB_transform_feedback_instanced;
    GLboolean ARB_transpose_matrix;
    GLboolean ARB_uniform_buffer_object;
    GLboolean ARB_vertex_array_object;
diff --git a/mesalib/src/mesa/main/queryobj.c b/mesalib/src/mesa/main/queryobj.c
index 805c031ae..a578aa594 100644
--- a/mesalib/src/mesa/main/queryobj.c
+++ b/mesalib/src/mesa/main/queryobj.c
@@ -263,28 +263,51 @@ _mesa_IsQueryARB(GLuint id)
       return GL_FALSE;
 }
 
+static GLboolean
+query_error_check_index(struct gl_context *ctx, GLenum target, GLuint index)
+{
+   switch (target) {
+   case GL_TRANSFORM_FEEDBACK_PRIMITIVES_WRITTEN:
+   case GL_PRIMITIVES_GENERATED:
+      if (index >= ctx->Const.MaxVertexStreams) {
+         _mesa_error(ctx, GL_INVALID_VALUE,
+                     "glBeginQueryIndexed(index>=MaxVertexStreams)");
+         return GL_FALSE;
+      }
+      break;
+   default:
+      if (index > 0) {
+         _mesa_error(ctx, GL_INVALID_VALUE, "glBeginQueryIndexed(index>0)");
+         return GL_FALSE;
+      }
+   }
+   return GL_TRUE;
+}
 
 static void GLAPIENTRY
-_mesa_BeginQueryARB(GLenum target, GLuint id)
+_mesa_BeginQueryIndexed(GLenum target, GLuint index, GLuint id)
 {
    struct gl_query_object *q, **bindpt;
    GET_CURRENT_CONTEXT(ctx);
    ASSERT_OUTSIDE_BEGIN_END(ctx);
 
    if (MESA_VERBOSE & VERBOSE_API)
-      _mesa_debug(ctx, "glBeginQuery(%s, %u)\n",
-                  _mesa_lookup_enum_by_nr(target), id);
+      _mesa_debug(ctx, "glBeginQueryIndexed(%s, %u, %u)\n",
+                  _mesa_lookup_enum_by_nr(target), index, id);
+
+   if (!query_error_check_index(ctx, target, index))
+      return;
 
    FLUSH_VERTICES(ctx, _NEW_DEPTH);
 
    bindpt = get_query_binding_point(ctx, target);
    if (!bindpt) {
-      _mesa_error(ctx, GL_INVALID_ENUM, "glBeginQueryARB(target)");
+      _mesa_error(ctx, GL_INVALID_ENUM, "glBeginQuery{Indexed}(target)");
       return;
    }
 
    if (id == 0) {
-      _mesa_error(ctx, GL_INVALID_OPERATION, "glBeginQueryARB(id==0)");
+      _mesa_error(ctx, GL_INVALID_OPERATION, "glBeginQuery{Indexed}(id==0)");
       return;
    }
 
@@ -293,7 +316,7 @@ _mesa_BeginQueryARB(GLenum target, GLuint id)
       /* create new object */
       q = ctx->Driver.NewQueryObject(ctx, id);
       if (!q) {
-         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glBeginQueryARB");
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glBeginQuery{Indexed}");
          return;
       }
       _mesa_HashInsert(ctx->Query.QueryObjects, id, q);
@@ -302,7 +325,7 @@ _mesa_BeginQueryARB(GLenum target, GLuint id)
       /* pre-existing object */
       if (q->Active) {
          _mesa_error(ctx, GL_INVALID_OPERATION,
-                     "glBeginQueryARB(query already active)");
+                     "glBeginQuery{Indexed}(query already active)");
          return;
       }
    }
@@ -320,20 +343,24 @@ _mesa_BeginQueryARB(GLenum target, GLuint id)
 
 
 static void GLAPIENTRY
-_mesa_EndQueryARB(GLenum target)
+_mesa_EndQueryIndexed(GLenum target, GLuint index)
 {
    struct gl_query_object *q, **bindpt;
    GET_CURRENT_CONTEXT(ctx);
    ASSERT_OUTSIDE_BEGIN_END(ctx);
 
    if (MESA_VERBOSE & VERBOSE_API)
-      _mesa_debug(ctx, "glEndQuery(%s)\n", _mesa_lookup_enum_by_nr(target));
+      _mesa_debug(ctx, "glEndQueryIndexed(%s, %u)\n",
+                  _mesa_lookup_enum_by_nr(target), index);
+
+   if (!query_error_check_index(ctx, target, index))
+      return;
 
    FLUSH_VERTICES(ctx, _NEW_DEPTH);
 
    bindpt = get_query_binding_point(ctx, target);
    if (!bindpt) {
-      _mesa_error(ctx, GL_INVALID_ENUM, "glEndQueryARB(target)");
+      _mesa_error(ctx, GL_INVALID_ENUM, "glEndQuery{Indexed}(target)");
       return;
    }
 
@@ -343,7 +370,7 @@ _mesa_EndQueryARB(GLenum target)
 
    if (!q || !q->Active) {
       _mesa_error(ctx, GL_INVALID_OPERATION,
-                  "glEndQueryARB(no matching glBeginQueryARB)");
+                  "glEndQuery{Indexed}(no matching glBeginQuery{Indexed})");
       return;
    }
 
@@ -351,6 +378,17 @@ _mesa_EndQueryARB(GLenum target)
    ctx->Driver.EndQuery(ctx, q);
 }
 
+static void GLAPIENTRY
+_mesa_BeginQueryARB(GLenum target, GLuint id)
+{
+   _mesa_BeginQueryIndexed(target, 0, id);
+}
+
+static void GLAPIENTRY
+_mesa_EndQueryARB(GLenum target)
+{
+   _mesa_EndQueryIndexed(target, 0);
+}
 
 static void GLAPIENTRY
 _mesa_QueryCounter(GLuint id, GLenum target)
@@ -410,17 +448,22 @@ _mesa_QueryCounter(GLuint id, GLenum target)
 
 
 static void GLAPIENTRY
-_mesa_GetQueryivARB(GLenum target, GLenum pname, GLint *params)
+_mesa_GetQueryIndexediv(GLenum target, GLuint index, GLenum pname,
+                        GLint *params)
 {
    struct gl_query_object *q = NULL, **bindpt = NULL;
    GET_CURRENT_CONTEXT(ctx);
    ASSERT_OUTSIDE_BEGIN_END(ctx);
 
    if (MESA_VERBOSE & VERBOSE_API)
-      _mesa_debug(ctx, "glGetQueryiv(%s, %s)\n",
+      _mesa_debug(ctx, "glGetQueryIndexediv(%s, %u, %s)\n",
                   _mesa_lookup_enum_by_nr(target),
+                  index,
                   _mesa_lookup_enum_by_nr(pname));
 
+   if (!query_error_check_index(ctx, target, index))
+      return;
+
    if (target == GL_TIMESTAMP) {
       if (!ctx->Extensions.ARB_timer_query) {
          _mesa_error(ctx, GL_INVALID_ENUM, "glGetQueryARB(target)");
@@ -430,7 +473,7 @@ _mesa_GetQueryivARB(GLenum target, GLenum pname, GLint *params)
    else {
       bindpt = get_query_binding_point(ctx, target);
       if (!bindpt) {
-         _mesa_error(ctx, GL_INVALID_ENUM, "glGetQueryARB(target)");
+         _mesa_error(ctx, GL_INVALID_ENUM, "glGetQuery{Indexed}iv(target)");
          return;
       }
 
@@ -445,11 +488,16 @@ _mesa_GetQueryivARB(GLenum target, GLenum pname, GLint *params)
          *params = q ? q->Id : 0;
          break;
       default:
-         _mesa_error(ctx, GL_INVALID_ENUM, "glGetQueryivARB(pname)");
+         _mesa_error(ctx, GL_INVALID_ENUM, "glGetQuery{Indexed}iv(pname)");
          return;
    }
 }
 
+static void GLAPIENTRY
+_mesa_GetQueryivARB(GLenum target, GLenum pname, GLint *params)
+{
+   _mesa_GetQueryIndexediv(target, 0, pname, params);
+}
 
 static void GLAPIENTRY
 _mesa_GetQueryObjectivARB(GLuint id, GLenum pname, GLint *params)
@@ -650,6 +698,10 @@ _mesa_init_queryobj_dispatch(struct _glapi_table *disp)
 
    SET_GetQueryObjecti64vEXT(disp, _mesa_GetQueryObjecti64vEXT);
    SET_GetQueryObjectui64vEXT(disp, _mesa_GetQueryObjectui64vEXT);
+
+   SET_BeginQueryIndexed(disp, _mesa_BeginQueryIndexed);
+   SET_EndQueryIndexed(disp, _mesa_EndQueryIndexed);
+   SET_GetQueryIndexediv(disp, _mesa_GetQueryIndexediv);
 }
 
 
diff --git a/mesalib/src/mesa/main/texformat.c b/mesalib/src/mesa/main/texformat.c
index 5fdc2ab1f..26bcbc10a 100644
--- a/mesalib/src/mesa/main/texformat.c
+++ b/mesalib/src/mesa/main/texformat.c
@@ -258,6 +258,16 @@ _mesa_choose_tex_format( struct gl_context *ctx, GLint internalFormat,
          ; /* fallthrough */
    }
 
+   if (ctx->Extensions.ARB_ES2_compatibility) {
+      switch (internalFormat) {
+         case GL_RGB565:
+            RETURN_IF_SUPPORTED(MESA_FORMAT_RGB565);
+            break;
+         default:
+         ; /* fallthrough */
+      }
+   }
+
    if (ctx->Extensions.MESA_ycbcr_texture) {
       if (internalFormat == GL_YCBCR_MESA) {
          if (type == GL_UNSIGNED_SHORT_8_8_MESA)
diff --git a/mesalib/src/mesa/main/teximage.c b/mesalib/src/mesa/main/teximage.c
index b16baaf91..126386ebe 100644
--- a/mesalib/src/mesa/main/teximage.c
+++ b/mesalib/src/mesa/main/teximage.c
@@ -138,6 +138,15 @@ _mesa_base_tex_format( struct gl_context *ctx, GLint internalFormat )
       }
    }
 
+   if (ctx->Extensions.ARB_ES2_compatibility) {
+      switch (internalFormat) {
+         case GL_RGB565:
+            return GL_RGB;
+         default:
+            ; /* fallthrough */
+      }
+   }
+
    if (ctx->Extensions.ARB_depth_texture) {
       switch (internalFormat) {
          case GL_DEPTH_COMPONENT:
diff --git a/mesalib/src/mesa/main/transformfeedback.c b/mesalib/src/mesa/main/transformfeedback.c
index 6f8221201..7679b4b08 100644
--- a/mesalib/src/mesa/main/transformfeedback.c
+++ b/mesalib/src/mesa/main/transformfeedback.c
@@ -603,6 +603,38 @@ _mesa_TransformFeedbackVaryings(GLuint program, GLsizei count,
       return;
    }
 
+   if (ctx->Extensions.ARB_transform_feedback3) {
+      if (bufferMode == GL_INTERLEAVED_ATTRIBS) {
+         unsigned buffers = 1;
+
+         for (i = 0; i < count; i++) {
+            if (strcmp(varyings[i], "gl_NextBuffer") == 0)
+               buffers++;
+         }
+
+         if (buffers > ctx->Const.MaxTransformFeedbackBuffers) {
+            _mesa_error(ctx, GL_INVALID_OPERATION,
+                        "glTransformFeedbackVaryings(too many gl_NextBuffer "
+                        "occurences)");
+            return;
+         }
+      } else {
+         for (i = 0; i < count; i++) {
+            if (strcmp(varyings[i], "gl_NextBuffer") == 0 ||
+                strcmp(varyings[i], "gl_SkipComponents1") == 0 ||
+                strcmp(varyings[i], "gl_SkipComponents2") == 0 ||
+                strcmp(varyings[i], "gl_SkipComponents3") == 0 ||
+                strcmp(varyings[i], "gl_SkipComponents4") == 0) {
+               _mesa_error(ctx, GL_INVALID_OPERATION,
+                           "glTransformFeedbackVaryings(SEPARATE_ATTRIBS,"
+                           "varying=%s)",
+                           varyings[i]);
+               return;
+            }
+         }
+      }
+   }
+
    /* free existing varyings, if any */
    for (i = 0; i < shProg->TransformFeedback.NumVarying; i++) {
       free(shProg->TransformFeedback.VaryingNames[i]);
diff --git a/mesalib/src/mesa/main/vtxfmt.c b/mesalib/src/mesa/main/vtxfmt.c
index a27596a98..bf7a54c0a 100644
--- a/mesalib/src/mesa/main/vtxfmt.c
+++ b/mesalib/src/mesa/main/vtxfmt.c
@@ -111,6 +111,11 @@ install_vtxfmt( struct _glapi_table *tab, const GLvertexformat *vfmt )
    SET_DrawElementsInstancedBaseVertex(tab, vfmt->DrawElementsInstancedBaseVertex);
    SET_DrawElementsInstancedBaseVertexBaseInstance(tab, vfmt->DrawElementsInstancedBaseVertexBaseInstance);
    SET_DrawTransformFeedback(tab, vfmt->DrawTransformFeedback);
+   SET_DrawTransformFeedbackStream(tab, vfmt->DrawTransformFeedbackStream);
+   SET_DrawTransformFeedbackInstanced(tab,
+                                      vfmt->DrawTransformFeedbackInstanced);
+   SET_DrawTransformFeedbackStreamInstanced(tab,
+                                vfmt->DrawTransformFeedbackStreamInstanced);
 
    /* GL_NV_vertex_program */
    SET_VertexAttrib1fNV(tab, vfmt->VertexAttrib1fNV);
diff --git a/mesalib/src/mesa/state_tracker/st_cb_blit.c b/mesalib/src/mesa/state_tracker/st_cb_blit.c
index 27da2c633..1486779fd 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_blit.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_blit.c
@@ -242,7 +242,7 @@ st_BlitFramebuffer(struct gl_context *ctx,
                           srcX0, srcY0, srcX1, srcY1,
                           srcAtt->Zoffset + srcAtt->CubeMapFace,
                           dstSurf, dstX0, dstY0, dstX1, dstY1,
-                          0.0, pFilter);
+                          0.0, pFilter, TGSI_WRITEMASK_XYZW, 0);
       }
       else {
          struct st_renderbuffer *srcRb =
@@ -257,7 +257,7 @@ st_BlitFramebuffer(struct gl_context *ctx,
                           srcX0, srcY0, srcX1, srcY1,
                           srcSurf->u.tex.first_layer,
                           dstSurf, dstX0, dstY0, dstX1, dstY1,
-                          0.0, pFilter);
+                          0.0, pFilter, TGSI_WRITEMASK_XYZW, 0);
       }
    }
 
@@ -281,6 +281,13 @@ st_BlitFramebuffer(struct gl_context *ctx,
       struct pipe_surface *dstDepthSurf =
          dstDepthRb ? dstDepthRb->surface : NULL;
 
+      struct st_renderbuffer *srcStencilRb =
+         st_renderbuffer(readFB->Attachment[BUFFER_STENCIL].Renderbuffer);
+      struct st_renderbuffer *dstStencilRb =
+         st_renderbuffer(drawFB->Attachment[BUFFER_STENCIL].Renderbuffer);
+      struct pipe_surface *dstStencilSurf =
+         dstStencilRb ? dstStencilRb->surface : NULL;
+
       if ((mask & depthStencil) == depthStencil &&
           st_is_depth_stencil_combined(srcDepth, srcStencil) &&
           st_is_depth_stencil_combined(dstDepth, dstStencil)) {
@@ -294,7 +301,15 @@ st_BlitFramebuffer(struct gl_context *ctx,
                           srcX0, srcY0, srcX1, srcY1,
                           srcDepthRb->surface->u.tex.first_layer,
                           dstDepthSurf, dstX0, dstY0, dstX1, dstY1,
-                          0.0, pFilter);
+                          0.0, pFilter, 0,
+                          BLIT_WRITEMASK_Z |
+                          (st->has_stencil_export ? BLIT_WRITEMASK_STENCIL
+                                                  : 0));
+
+         if (!st->has_stencil_export) {
+            _mesa_problem(ctx, "st_BlitFramebuffer(STENCIL) "
+                               "software fallback not implemented");
+         }
       }
       else {
          /* blitting depth and stencil separately */
@@ -305,12 +320,22 @@ st_BlitFramebuffer(struct gl_context *ctx,
                              srcX0, srcY0, srcX1, srcY1,
                              srcDepthRb->surface->u.tex.first_layer,
                              dstDepthSurf, dstX0, dstY0, dstX1, dstY1,
-                             0.0, pFilter);
+                             0.0, pFilter, 0, BLIT_WRITEMASK_Z);
          }
 
          if (mask & GL_STENCIL_BUFFER_BIT) {
-            /* blit stencil only */
-            _mesa_problem(ctx, "st_BlitFramebuffer(STENCIL) not completed");
+            if (st->has_stencil_export) {
+               util_blit_pixels(st->blit, srcStencilRb->texture,
+                                srcStencilRb->surface->u.tex.level,
+                                srcX0, srcY0, srcX1, srcY1,
+                                srcStencilRb->surface->u.tex.first_layer,
+                                dstStencilSurf, dstX0, dstY0, dstX1, dstY1,
+                                0.0, pFilter, 0, BLIT_WRITEMASK_STENCIL);
+            }
+            else {
+               _mesa_problem(ctx, "st_BlitFramebuffer(STENCIL) "
+                                  "software fallback not implemented");
+            }
          }
       }
    }
diff --git a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
index 2bcbada4f..c5f36316b 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_drawpixels.c
@@ -403,6 +403,8 @@ internal_format(struct gl_context *ctx, GLenum format, GLenum type)
 
          case GL_UNSIGNED_SHORT_5_6_5:
          case GL_UNSIGNED_SHORT_5_6_5_REV:
+            return GL_RGB565;
+
          case GL_UNSIGNED_SHORT_5_5_5_1:
          case GL_UNSIGNED_SHORT_1_5_5_5_REV:
             return GL_RGB5_A1;
@@ -1165,27 +1167,8 @@ st_DrawPixels(struct gl_context *ctx, GLint x, GLint y,
              * The stencil is written using the shader stencil export
              * functionality. */
             if (write_stencil) {
-               enum pipe_format stencil_format = PIPE_FORMAT_NONE;
-
-               switch (pt->format) {
-               case PIPE_FORMAT_Z24_UNORM_S8_UINT:
-               case PIPE_FORMAT_X24S8_UINT:
-                  stencil_format = PIPE_FORMAT_X24S8_UINT;
-                  break;
-               case PIPE_FORMAT_S8_UINT_Z24_UNORM:
-               case PIPE_FORMAT_S8X24_UINT:
-                  stencil_format = PIPE_FORMAT_S8X24_UINT;
-                  break;
-               case PIPE_FORMAT_Z32_FLOAT_S8X24_UINT:
-               case PIPE_FORMAT_X32_S8X24_UINT:
-                  stencil_format = PIPE_FORMAT_X32_S8X24_UINT;
-                  break;
-               case PIPE_FORMAT_S8_UINT:
-                  stencil_format = PIPE_FORMAT_S8_UINT;
-                  break;
-               default:
-                  assert(0);
-               }
+               enum pipe_format stencil_format =
+                     util_format_stencil_only(pt->format);
 
                sv[1] = st_create_texture_sampler_view_format(st->pipe, pt,
                                                              stencil_format);
diff --git a/mesalib/src/mesa/state_tracker/st_cb_texture.c b/mesalib/src/mesa/state_tracker/st_cb_texture.c
index 67c3f9590..a7f57b96f 100644
--- a/mesalib/src/mesa/state_tracker/st_cb_texture.c
+++ b/mesalib/src/mesa/state_tracker/st_cb_texture.c
@@ -944,7 +944,7 @@ st_CopyTexSubImage(struct gl_context *ctx, GLuint dims,
    struct pipe_screen *screen = pipe->screen;
    enum pipe_format dest_format, src_format;
    GLboolean matching_base_formats;
-   GLuint format_writemask, sample_count;
+   GLuint color_writemask, zs_writemask, sample_count;
    struct pipe_surface *dest_surface = NULL;
    GLboolean do_flip = (st_fb_orientation(ctx->ReadBuffer) == Y_0_TOP);
    struct pipe_surface surf_tmpl;
@@ -1025,15 +1025,17 @@ st_CopyTexSubImage(struct gl_context *ctx, GLuint dims,
    }
 
    if (texBaseFormat == GL_DEPTH_COMPONENT) {
-      format_writemask = TGSI_WRITEMASK_XYZW;
+      color_writemask = 0;
+      zs_writemask = BLIT_WRITEMASK_Z;
       dst_usage = PIPE_BIND_DEPTH_STENCIL;
    }
    else {
-      format_writemask = compatible_src_dst_formats(ctx, &strb->Base, texImage);
+      color_writemask = compatible_src_dst_formats(ctx, &strb->Base, texImage);
+      zs_writemask = 0;
       dst_usage = PIPE_BIND_RENDER_TARGET;
    }
 
-   if (!format_writemask ||
+   if ((!color_writemask && !zs_writemask) ||
        !screen->is_format_supported(screen, src_format,
                                     PIPE_TEXTURE_2D, sample_count,
                                     PIPE_BIND_SAMPLER_VIEW) ||
@@ -1066,17 +1068,17 @@ st_CopyTexSubImage(struct gl_context *ctx, GLuint dims,
 
    dest_surface = pipe->create_surface(pipe, stImage->pt,
                                        &surf_tmpl);
-   util_blit_pixels_writemask(st->blit,
-                              strb->texture,
-                              strb->surface->u.tex.level,
-                              srcX, srcY0,
-                              srcX + width, srcY1,
-                              strb->surface->u.tex.first_layer,
-                              dest_surface,
-                              destX, destY,
-                              destX + width, destY + height,
-                              0.0, PIPE_TEX_MIPFILTER_NEAREST,
-                              format_writemask);
+   util_blit_pixels(st->blit,
+                    strb->texture,
+                    strb->surface->u.tex.level,
+                    srcX, srcY0,
+                    srcX + width, srcY1,
+                    strb->surface->u.tex.first_layer,
+                    dest_surface,
+                    destX, destY,
+                    destX + width, destY + height,
+                    0.0, PIPE_TEX_MIPFILTER_NEAREST,
+                    color_writemask, zs_writemask);
    pipe_surface_reference(&dest_surface, NULL);
 
    /* Restore conditional rendering state. */
diff --git a/mesalib/src/mesa/state_tracker/st_context.c b/mesalib/src/mesa/state_tracker/st_context.c
index 132dcc02f..117ea90f8 100644
--- a/mesalib/src/mesa/state_tracker/st_context.c
+++ b/mesalib/src/mesa/state_tracker/st_context.c
@@ -187,6 +187,8 @@ st_create_context_priv( struct gl_context *ctx, struct pipe_context *pipe )
    st->pixel_xfer.cache = _mesa_new_program_cache();
 
    st->force_msaa = st_get_msaa();
+   st->has_stencil_export =
+      screen->get_param(screen, PIPE_CAP_SHADER_STENCIL_EXPORT);
 
    /* GL limits and extensions */
    st_init_limits(st);
diff --git a/mesalib/src/mesa/state_tracker/st_context.h b/mesalib/src/mesa/state_tracker/st_context.h
index 734b4d9c1..a3f44b3ab 100644
--- a/mesalib/src/mesa/state_tracker/st_context.h
+++ b/mesalib/src/mesa/state_tracker/st_context.h
@@ -81,7 +81,7 @@ struct st_context
    struct draw_stage *rastpos_stage;  /**< For glRasterPos */
    GLboolean clamp_frag_color_in_shader;
    GLboolean clamp_vert_color_in_shader;
-
+   boolean has_stencil_export; /**< can do shader stencil export? */
 
    /* On old libGL's for linux we need to invalidate the drawables
     * on glViewpport calls, this is set via a option.
diff --git a/mesalib/src/mesa/state_tracker/st_extensions.c b/mesalib/src/mesa/state_tracker/st_extensions.c
index 5b333ad2c..5099acdb5 100644
--- a/mesalib/src/mesa/state_tracker/st_extensions.c
+++ b/mesalib/src/mesa/state_tracker/st_extensions.c
@@ -355,6 +355,7 @@ void st_init_extensions(struct st_context *st)
       { o(ARB_shadow),                       PIPE_CAP_TEXTURE_SHADOW_MAP               },
       { o(ARB_texture_non_power_of_two),     PIPE_CAP_NPOT_TEXTURES                    },
       { o(ARB_transform_feedback2),          PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME       },
+      { o(ARB_transform_feedback3),          PIPE_CAP_STREAM_OUTPUT_PAUSE_RESUME       },
 
       { o(EXT_blend_equation_separate),      PIPE_CAP_BLEND_EQUATION_SEPARATE          },
       { o(EXT_draw_buffers2),                PIPE_CAP_INDEP_BLEND_ENABLE               },
@@ -641,4 +642,9 @@ void st_init_extensions(struct st_context *st)
        screen->get_param(screen, PIPE_CAP_QUERY_TIMESTAMP)) {
       ctx->Extensions.ARB_timer_query = GL_TRUE;
    }
+
+   if (ctx->Extensions.ARB_transform_feedback2 &&
+       ctx->Extensions.ARB_draw_instanced) {
+      ctx->Extensions.ARB_transform_feedback_instanced = GL_TRUE;
+   }
 }
diff --git a/mesalib/src/mesa/state_tracker/st_format.c b/mesalib/src/mesa/state_tracker/st_format.c
index 4265d14b4..57d34411f 100644
--- a/mesalib/src/mesa/state_tracker/st_format.c
+++ b/mesalib/src/mesa/state_tracker/st_format.c
@@ -793,6 +793,10 @@ static const struct format_mapping format_map[] = {
       { PIPE_FORMAT_B5G6R5_UNORM, PIPE_FORMAT_B5G5R5A1_UNORM,
         DEFAULT_RGBA_FORMATS }
    },
+   {
+      { GL_RGB565 },
+      { PIPE_FORMAT_B5G6R5_UNORM, DEFAULT_RGBA_FORMATS }
+   },
 
    /* basic Alpha formats */
    {
diff --git a/mesalib/src/mesa/vbo/vbo_exec_array.c b/mesalib/src/mesa/vbo/vbo_exec_array.c
index d2854dd6c..a923aa1fa 100644
--- a/mesalib/src/mesa/vbo/vbo_exec_array.c
+++ b/mesalib/src/mesa/vbo/vbo_exec_array.c
@@ -1286,12 +1286,17 @@ vbo_exec_MultiDrawElementsBaseVertex(GLenum mode,
 static void
 vbo_draw_transform_feedback(struct gl_context *ctx, GLenum mode,
                             struct gl_transform_feedback_object *obj,
-                            GLuint numInstances)
+                            GLuint stream, GLuint numInstances)
 {
    struct vbo_context *vbo = vbo_context(ctx);
    struct vbo_exec_context *exec = &vbo->exec;
    struct _mesa_prim prim[2];
 
+   if (!_mesa_validate_DrawTransformFeedback(ctx, mode, obj, stream,
+                                             numInstances)) {
+      return;
+   }
+
    vbo_bind_arrays(ctx);
 
    /* init most fields to zero */
@@ -1334,11 +1339,52 @@ vbo_exec_DrawTransformFeedback(GLenum mode, GLuint name)
       _mesa_debug(ctx, "glDrawTransformFeedback(%s, %d)\n",
                   _mesa_lookup_enum_by_nr(mode), name);
 
-   if (!_mesa_validate_DrawTransformFeedback(ctx, mode, obj)) {
-      return;
-   }
+   vbo_draw_transform_feedback(ctx, mode, obj, 0, 1);
+}
+
+static void GLAPIENTRY
+vbo_exec_DrawTransformFeedbackStream(GLenum mode, GLuint name, GLuint stream)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct gl_transform_feedback_object *obj =
+      _mesa_lookup_transform_feedback_object(ctx, name);
+
+   if (MESA_VERBOSE & VERBOSE_DRAW)
+      _mesa_debug(ctx, "glDrawTransformFeedbackStream(%s, %u, %u)\n",
+                  _mesa_lookup_enum_by_nr(mode), name, stream);
+
+   vbo_draw_transform_feedback(ctx, mode, obj, stream, 1);
+}
+
+static void GLAPIENTRY
+vbo_exec_DrawTransformFeedbackInstanced(GLenum mode, GLuint name,
+                                        GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct gl_transform_feedback_object *obj =
+      _mesa_lookup_transform_feedback_object(ctx, name);
+
+   if (MESA_VERBOSE & VERBOSE_DRAW)
+      _mesa_debug(ctx, "glDrawTransformFeedbackInstanced(%s, %d)\n",
+                  _mesa_lookup_enum_by_nr(mode), name);
+
+   vbo_draw_transform_feedback(ctx, mode, obj, 0, primcount);
+}
+
+static void GLAPIENTRY
+vbo_exec_DrawTransformFeedbackStreamInstanced(GLenum mode, GLuint name,
+                                              GLuint stream, GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct gl_transform_feedback_object *obj =
+      _mesa_lookup_transform_feedback_object(ctx, name);
+
+   if (MESA_VERBOSE & VERBOSE_DRAW)
+      _mesa_debug(ctx, "glDrawTransformFeedbackStreamInstanced"
+                  "(%s, %u, %u, %i)\n",
+                  _mesa_lookup_enum_by_nr(mode), name, stream, primcount);
 
-   vbo_draw_transform_feedback(ctx, mode, obj, 1);
+   vbo_draw_transform_feedback(ctx, mode, obj, stream, primcount);
 }
 
 #endif
@@ -1365,6 +1411,12 @@ vbo_exec_array_init( struct vbo_exec_context *exec )
    exec->vtxfmt.DrawElementsInstancedBaseVertexBaseInstance = vbo_exec_DrawElementsInstancedBaseVertexBaseInstance;
 #if FEATURE_EXT_transform_feedback
    exec->vtxfmt.DrawTransformFeedback = vbo_exec_DrawTransformFeedback;
+   exec->vtxfmt.DrawTransformFeedbackStream =
+         vbo_exec_DrawTransformFeedbackStream;
+   exec->vtxfmt.DrawTransformFeedbackInstanced =
+         vbo_exec_DrawTransformFeedbackInstanced;
+   exec->vtxfmt.DrawTransformFeedbackStreamInstanced =
+         vbo_exec_DrawTransformFeedbackStreamInstanced;
 #endif
 }
 
diff --git a/mesalib/src/mesa/vbo/vbo_save_api.c b/mesalib/src/mesa/vbo/vbo_save_api.c
index d27525812..a02a13db5 100644
--- a/mesalib/src/mesa/vbo/vbo_save_api.c
+++ b/mesalib/src/mesa/vbo/vbo_save_api.c
@@ -1068,6 +1068,45 @@ _save_DrawTransformFeedback(GLenum mode, GLuint name)
 
 
 static void GLAPIENTRY
+_save_DrawTransformFeedbackStream(GLenum mode, GLuint name, GLuint stream)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode;
+   (void) name;
+   (void) stream;
+   _mesa_compile_error(ctx, GL_INVALID_OPERATION,
+                       "glDrawTransformFeedbackStream");
+}
+
+
+static void GLAPIENTRY
+_save_DrawTransformFeedbackInstanced(GLenum mode, GLuint name,
+                                     GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode;
+   (void) name;
+   (void) primcount;
+   _mesa_compile_error(ctx, GL_INVALID_OPERATION,
+                       "glDrawTransformFeedbackInstanced");
+}
+
+
+static void GLAPIENTRY
+_save_DrawTransformFeedbackStreamInstanced(GLenum mode, GLuint name,
+                                           GLuint stream, GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode;
+   (void) name;
+   (void) stream;
+   (void) primcount;
+   _mesa_compile_error(ctx, GL_INVALID_OPERATION,
+                       "glDrawTransformFeedbackStreamInstanced");
+}
+
+
+static void GLAPIENTRY
 _save_Rectf(GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2)
 {
    GET_CURRENT_CONTEXT(ctx);
@@ -1416,9 +1455,13 @@ _save_vtxfmt_init(struct gl_context *ctx)
    vfmt->DrawRangeElements = _save_DrawRangeElements;
    vfmt->DrawElementsBaseVertex = _save_DrawElementsBaseVertex;
    vfmt->DrawRangeElementsBaseVertex = _save_DrawRangeElementsBaseVertex;
-   vfmt->DrawTransformFeedback = _save_DrawTransformFeedback;
    vfmt->MultiDrawElementsEXT = _save_MultiDrawElements;
    vfmt->MultiDrawElementsBaseVertex = _save_MultiDrawElementsBaseVertex;
+   vfmt->DrawTransformFeedback = _save_DrawTransformFeedback;
+   vfmt->DrawTransformFeedbackStream = _save_DrawTransformFeedbackStream;
+   vfmt->DrawTransformFeedbackInstanced = _save_DrawTransformFeedbackInstanced;
+   vfmt->DrawTransformFeedbackStreamInstanced =
+         _save_DrawTransformFeedbackStreamInstanced;
 }