Merge remote-tracking branch 'origin/released'

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

9 files changed, 573 insertions, 192 deletions

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);