Merge branch 'gallium-dynamicstencilref'

63 files changed, 563 insertions, 191 deletions

diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index c638239e80b..b5241fa64c6 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -93,6 +93,7 @@ struct cso_context {
    struct pipe_framebuffer_state fb, fb_saved;
    struct pipe_viewport_state vp, vp_saved;
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref, stencil_ref_saved;
 };
 
 
@@ -1057,8 +1058,6 @@ void cso_restore_viewport(struct cso_context *ctx)
 }
 
 
-
-
 enum pipe_error cso_set_blend_color(struct cso_context *ctx,
                                     const struct pipe_blend_color *bc)
 {
@@ -1069,6 +1068,30 @@ enum pipe_error cso_set_blend_color(struct cso_context *ctx,
    return PIPE_OK;
 }
 
+enum pipe_error cso_set_stencil_ref(struct cso_context *ctx,
+                                    const struct pipe_stencil_ref *sr)
+{
+   if (memcmp(&ctx->stencil_ref, sr, sizeof(ctx->stencil_ref))) {
+      ctx->stencil_ref = *sr;
+      ctx->pipe->set_stencil_ref(ctx->pipe, sr);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_stencil_ref(struct cso_context *ctx)
+{
+   ctx->stencil_ref_saved = ctx->stencil_ref;
+}
+
+
+void cso_restore_stencil_ref(struct cso_context *ctx)
+{
+   if (memcmp(&ctx->stencil_ref, &ctx->stencil_ref_saved, sizeof(ctx->stencil_ref))) {
+      ctx->stencil_ref = ctx->stencil_ref_saved;
+      ctx->pipe->set_stencil_ref(ctx->pipe, &ctx->stencil_ref);
+   }
+}
+
 enum pipe_error cso_set_geometry_shader_handle(struct cso_context *ctx,
                                                void *handle)
 {
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
index d2089b1c883..707b3c2cee2 100644
--- a/src/gallium/auxiliary/cso_cache/cso_context.h
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -174,6 +174,12 @@ enum pipe_error cso_set_blend_color(struct cso_context *cso,
                                     const struct pipe_blend_color *bc);
 
 
+enum pipe_error cso_set_stencil_ref(struct cso_context *cso,
+                                    const struct pipe_stencil_ref *sr);
+void cso_save_stencil_ref(struct cso_context *cso);
+void cso_restore_stencil_ref(struct cso_context *cso);
+
+
 #ifdef	__cplusplus
 }
 #endif
diff --git a/src/gallium/auxiliary/util/u_blitter.c b/src/gallium/auxiliary/util/u_blitter.c
index f3b4491d175..1f7e25b84f2 100644
--- a/src/gallium/auxiliary/util/u_blitter.c
+++ b/src/gallium/auxiliary/util/u_blitter.c
@@ -60,13 +60,12 @@ struct blitter_context_priv
    float vertices[4][2][4];   /**< {pos, color} or {pos, texcoord} */
 
    /* Templates for various state objects. */
-   struct pipe_depth_stencil_alpha_state template_dsa;
    struct pipe_sampler_state template_sampler_state;
 
    /* Constant state objects. */
    /* Vertex shaders. */
    void *vs_col; /**< Vertex shader which passes {pos, color} to the output */
-   void *vs_tex; /**<Vertex shader which passes {pos, texcoord} to the output.*/
+   void *vs_tex; /**< Vertex shader which passes {pos, texcoord} to the output.*/
 
    /* Fragment shaders. */
    /* FS which outputs a color to multiple color buffers. */
@@ -85,7 +84,7 @@ struct blitter_context_priv
    void *blend_keep_color;    /**< blend state with writemask of 0 */
 
    /* Depth stencil alpha state. */
-   void *dsa_write_depth_stencil[0xff]; /**< indices are stencil clear values */
+   void *dsa_write_depth_stencil;
    void *dsa_write_depth_keep_stencil;
    void *dsa_keep_depth_stencil;
 
@@ -100,7 +99,7 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
 {
    struct blitter_context_priv *ctx;
    struct pipe_blend_state blend;
-   struct pipe_depth_stencil_alpha_state *dsa;
+   struct pipe_depth_stencil_alpha_state dsa;
    struct pipe_rasterizer_state rs_state;
    struct pipe_sampler_state *sampler_state;
    unsigned i;
@@ -129,23 +128,24 @@ struct blitter_context *util_blitter_create(struct pipe_context *pipe)
    ctx->blend_write_color = pipe->create_blend_state(pipe, &blend);
 
    /* depth stencil alpha state objects */
-   dsa = &ctx->template_dsa;
    ctx->dsa_keep_depth_stencil =
-      pipe->create_depth_stencil_alpha_state(pipe, dsa);
+      pipe->create_depth_stencil_alpha_state(pipe, &dsa);
 
-   dsa->depth.enabled = 1;
-   dsa->depth.writemask = 1;
-   dsa->depth.func = PIPE_FUNC_ALWAYS;
+   dsa.depth.enabled = 1;
+   dsa.depth.writemask = 1;
+   dsa.depth.func = PIPE_FUNC_ALWAYS;
    ctx->dsa_write_depth_keep_stencil =
-      pipe->create_depth_stencil_alpha_state(pipe, dsa);
-
-   dsa->stencil[0].enabled = 1;
-   dsa->stencil[0].func = PIPE_FUNC_ALWAYS;
-   dsa->stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
-   dsa->stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
-   dsa->stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
-   dsa->stencil[0].valuemask = 0xff;
-   dsa->stencil[0].writemask = 0xff;
+      pipe->create_depth_stencil_alpha_state(pipe, &dsa);
+
+   dsa.stencil[0].enabled = 1;
+   dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
+   dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
+   dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
+   dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
+   dsa.stencil[0].valuemask = 0xff;
+   dsa.stencil[0].writemask = 0xff;
+   ctx->dsa_write_depth_stencil =
+      pipe->create_depth_stencil_alpha_state(pipe, &dsa);
    /* The DSA state objects which write depth and stencil are created
     * on-demand. */
 
@@ -210,12 +210,8 @@ void util_blitter_destroy(struct blitter_context *blitter)
    pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
    pipe->delete_depth_stencil_alpha_state(pipe,
                                           ctx->dsa_write_depth_keep_stencil);
-
-   for (i = 0; i < 0xff; i++)
-      if (ctx->dsa_write_depth_stencil[i])
-         pipe->delete_depth_stencil_alpha_state(pipe,
-            ctx->dsa_write_depth_stencil[i]);
-
+   pipe->delete_depth_stencil_alpha_state(pipe, ctx->dsa_write_depth_stencil);
+ 
    pipe->delete_rasterizer_state(pipe, ctx->rs_state);
    pipe->delete_vs_state(pipe, ctx->vs_col);
    pipe->delete_vs_state(pipe, ctx->vs_tex);
@@ -266,6 +262,8 @@ static void blitter_restore_CSOs(struct blitter_context_priv *ctx)
    ctx->blitter.saved_fs = INVALID_PTR;
    ctx->blitter.saved_vs = INVALID_PTR;
 
+   pipe->set_stencil_ref(pipe, &ctx->blitter.saved_stencil_ref);
+
    /* restore the state objects which are required to be saved before copy/fill
     */
    if (ctx->blitter.saved_fb_state.nr_cbufs != ~0) {
@@ -413,26 +411,6 @@ static void blitter_draw_quad(struct blitter_context_priv *ctx)
 }
 
 static INLINE
-void *blitter_get_state_write_depth_stencil(
-               struct blitter_context_priv *ctx,
-               unsigned stencil)
-{
-   struct pipe_context *pipe = ctx->pipe;
-
-   stencil &= 0xff;
-
-   /* Create the DSA state on-demand. */
-   if (!ctx->dsa_write_depth_stencil[stencil]) {
-      ctx->template_dsa.stencil[0].ref_value = stencil;
-
-      ctx->dsa_write_depth_stencil[stencil] =
-         pipe->create_depth_stencil_alpha_state(pipe, &ctx->template_dsa);
-   }
-
-   return ctx->dsa_write_depth_stencil[stencil];
-}
-
-static INLINE
 void **blitter_get_sampler_state(struct blitter_context_priv *ctx,
                                  int miplevel)
 {
@@ -559,9 +537,13 @@ void util_blitter_clear(struct blitter_context *blitter,
    else
       pipe->bind_blend_state(pipe, ctx->blend_keep_color);
 
-   if (clear_buffers & PIPE_CLEAR_DEPTHSTENCIL)
-      pipe->bind_depth_stencil_alpha_state(pipe,
-         blitter_get_state_write_depth_stencil(ctx, stencil));
+   if (clear_buffers & PIPE_CLEAR_DEPTHSTENCIL) {
+      struct pipe_stencil_ref sr;
+      memset (&sr, 0, sizeof(sr));
+      sr.ref_value[0] = stencil & 0xff;
+      pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_write_depth_stencil);
+      pipe->set_stencil_ref(pipe, &sr);
+   }
    else
       pipe->bind_depth_stencil_alpha_state(pipe, ctx->dsa_keep_depth_stencil);
 
@@ -573,6 +555,7 @@ void util_blitter_clear(struct blitter_context *blitter,
    blitter_set_rectangle(ctx, 0, 0, width, height, depth);
    blitter_draw_quad(ctx);
    blitter_restore_CSOs(ctx);
+   /* XXX driver's responsibility to restore stencil refs? */
 }
 
 static boolean
diff --git a/src/gallium/auxiliary/util/u_blitter.h b/src/gallium/auxiliary/util/u_blitter.h
index 3da5a6ca525..a2f17073ac6 100644
--- a/src/gallium/auxiliary/util/u_blitter.h
+++ b/src/gallium/auxiliary/util/u_blitter.h
@@ -47,6 +47,7 @@ struct blitter_context
    void *saved_fs, *saved_vs; /**< fragment shader, vertex shader */
 
    struct pipe_framebuffer_state saved_fb_state;  /**< framebuffer state */
+   struct pipe_stencil_ref saved_stencil_ref;     /**< stencil ref */
 
    int saved_num_sampler_states;
    void *saved_sampler_states[32];
@@ -170,6 +171,13 @@ void util_blitter_save_depth_stencil_alpha(struct blitter_context *blitter,
 }
 
 static INLINE
+void util_blitter_save_stencil_ref(struct blitter_context *blitter,
+                                   const struct pipe_stencil_ref *state)
+{
+   blitter->saved_stencil_ref = *state;
+}
+
+static INLINE
 void util_blitter_save_rasterizer(struct blitter_context *blitter,
                                   void *state)
 {
diff --git a/src/gallium/docs/source/context.rst b/src/gallium/docs/source/context.rst
index a7669575b95..9080addba44 100644
--- a/src/gallium/docs/source/context.rst
+++ b/src/gallium/docs/source/context.rst
@@ -49,8 +49,9 @@ Non-CSO State
 
 These pieces of state are too small, variable, and/or trivial to have CSO
 objects. They all follow simple, one-method binding calls, e.g.
-``set_edgeflags``.
-
+``set_blend_color``.
+* ``set_stencil_ref`` sets the stencil front and back reference values
+  which are used as comparison values in stencil test.
 * ``set_blend_color``
 * ``set_clip_state``
 * ``set_polygon_stipple``
diff --git a/src/gallium/docs/source/cso/dsa.rst b/src/gallium/docs/source/cso/dsa.rst
index 12abaa9d6fe..1bbe381f9e9 100644
--- a/src/gallium/docs/source/cso/dsa.rst
+++ b/src/gallium/docs/source/cso/dsa.rst
@@ -11,9 +11,9 @@ they are all stored in one structure.
 
 During actual execution, the order of operations done on fragments is always:
 
+* Alpha
 * Stencil
 * Depth
-* Alpha
 
 Depth Members
 -------------
@@ -28,15 +28,18 @@ func
 Stencil Members
 ---------------
 
-XXX document valuemask, writemask
-
 enabled
     Whether the stencil test is enabled. For the second stencil, whether the
-    two-sided stencil is enabled.
+    two-sided stencil is enabled. If two-sided stencil is disabled, the other
+    fields for the second array member are not valid.
 func
     The stencil test function. One of PIPE_FUNC.
-ref_value
-    Stencil test reference value; used for certain functions.
+valuemask
+    Stencil test value mask; this is ANDed with the value in the stencil
+    buffer and the reference value before doing the stencil comparison test.
+writemask
+    Stencil test writemask; this controls which bits of the stencil buffer
+    are written.
 fail_op
     The operation to carry out if the stencil test fails. One of
     PIPE_STENCIL_OP.
diff --git a/src/gallium/docs/source/cso/sampler.rst b/src/gallium/docs/source/cso/sampler.rst
index 044ffffcb4f..77979fc44d1 100644
--- a/src/gallium/docs/source/cso/sampler.rst
+++ b/src/gallium/docs/source/cso/sampler.rst
@@ -45,4 +45,6 @@ border_color
     RGBA color used for out-of-bounds coordinates.
 max_anisotropy
     Maximum filtering to apply anisotropically to textures. Setting this to
-    1.0 effectively disables anisotropic filtering.
+    0 disables anisotropic filtering. Any other setting enables anisotropic
+    filtering, however it's not unexpected some drivers only will change their
+    filtering with a setting of 2 and higher.
diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index 7f2b33c2dcc..bbb112fd336 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -230,6 +230,7 @@ struct cell_command_rasterizer
 {
    opcode_t opcode;    /**< CELL_CMD_STATE_RASTERIZER */
    struct pipe_rasterizer_state rasterizer;
+   uint32_t pad[1];
 };
 
 
@@ -326,7 +327,7 @@ struct cell_command_sampler
    opcode_t opcode;         /**< CELL_CMD_STATE_SAMPLER */
    uint unit;
    struct pipe_sampler_state state;
-   uint32_t pad_[2];
+   uint32_t pad_[3];
 };
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index 905cd5db390..a77cc5b9067 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -114,6 +114,7 @@ struct cell_context
    struct spe_function logic_op;
 
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
    struct pipe_buffer *constants[2];
    struct pipe_framebuffer_state framebuffer;
diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
index 0dab34075da..70683bb3676 100644
--- a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
@@ -1175,7 +1175,8 @@ gen_colormask(struct spe_function *f,
  */
 static void
 gen_stencil_test(struct spe_function *f,
-                 const struct pipe_stencil_state *state, 
+                 const struct pipe_stencil_state *state,
+                 const unsigned ref_value,
                  uint stencil_max_value,
                  int fragment_mask_reg,
                  int fbS_reg, 
@@ -1189,7 +1190,7 @@ gen_stencil_test(struct spe_function *f,
    case PIPE_FUNC_EQUAL:
       if (state->valuemask == stencil_max_value) {
          /* stencil_pass = fragment_mask & (s == reference) */
-         spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+         spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, ref_value);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
       }
       else {
@@ -1197,7 +1198,7 @@ gen_stencil_test(struct spe_function *f,
          uint tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
          spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil,
-                                state->valuemask & state->ref_value);
+                                state->valuemask & ref_value);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1206,7 +1207,7 @@ gen_stencil_test(struct spe_function *f,
    case PIPE_FUNC_NOTEQUAL:
       if (state->valuemask == stencil_max_value) {
          /* stencil_pass = fragment_mask & ~(s == reference) */
-         spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+         spe_compare_equal_uint(f, stencil_pass_reg, fbS_reg, ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
       }
       else {
@@ -1214,7 +1215,7 @@ gen_stencil_test(struct spe_function *f,
          int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
          spe_compare_equal_uint(f, stencil_pass_reg, tmp_masked_stencil,
-                                state->valuemask & state->ref_value);
+                                state->valuemask & ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1223,7 +1224,7 @@ gen_stencil_test(struct spe_function *f,
    case PIPE_FUNC_LESS:
       if (state->valuemask == stencil_max_value) {
          /* stencil_pass = fragment_mask & (reference < s)  */
-         spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, state->ref_value);
+         spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg, ref_value);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
       }
       else {
@@ -1231,7 +1232,7 @@ gen_stencil_test(struct spe_function *f,
          int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
          spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil,
-                                  state->valuemask & state->ref_value);
+                                  state->valuemask & ref_value);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1246,7 +1247,7 @@ gen_stencil_test(struct spe_function *f,
           * treats its operands as unsigned - no sign extension.
           */
          int tmp_reg = spe_allocate_available_register(f);
-         spe_load_uint(f, tmp_reg, state->ref_value);
+         spe_load_uint(f, tmp_reg, ref_value);
          spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_reg);
@@ -1255,7 +1256,7 @@ gen_stencil_test(struct spe_function *f,
          /* stencil_pass = fragment_mask & ((reference&mask) > (s&mask)) */
          int tmp_reg = spe_allocate_available_register(f);
          int tmp_masked_stencil = spe_allocate_available_register(f);
-         spe_load_uint(f, tmp_reg, state->valuemask & state->ref_value);
+         spe_load_uint(f, tmp_reg, state->valuemask & ref_value);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
          spe_clgt(f, stencil_pass_reg, tmp_reg, tmp_masked_stencil);
          spe_and(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
@@ -1269,7 +1270,7 @@ gen_stencil_test(struct spe_function *f,
          /* stencil_pass = fragment_mask & (reference >= s) 
           *              = fragment_mask & ~(s > reference) */
          spe_compare_greater_uint(f, stencil_pass_reg, fbS_reg,
-                                  state->ref_value);
+                                  ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
       }
       else {
@@ -1277,7 +1278,7 @@ gen_stencil_test(struct spe_function *f,
          int tmp_masked_stencil = spe_allocate_available_register(f);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
          spe_compare_greater_uint(f, stencil_pass_reg, tmp_masked_stencil,
-                                  state->valuemask & state->ref_value);
+                                  state->valuemask & ref_value);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_masked_stencil);
       }
@@ -1289,7 +1290,7 @@ gen_stencil_test(struct spe_function *f,
           *               = fragment_mask & ~(reference > s) */
          /* As above, we have to do this by loading a register */
          int tmp_reg = spe_allocate_available_register(f);
-         spe_load_uint(f, tmp_reg, state->ref_value);
+         spe_load_uint(f, tmp_reg, ref_value);
          spe_clgt(f, stencil_pass_reg, tmp_reg, fbS_reg);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
          spe_release_register(f, tmp_reg);
@@ -1298,7 +1299,7 @@ gen_stencil_test(struct spe_function *f,
          /* stencil_pass = fragment_mask & ~((reference&mask) > (s&mask)) */
          int tmp_reg = spe_allocate_available_register(f);
          int tmp_masked_stencil = spe_allocate_available_register(f);
-         spe_load_uint(f, tmp_reg, state->ref_value & state->valuemask);
+         spe_load_uint(f, tmp_reg, ref_value & state->valuemask);
          spe_and_uint(f, tmp_masked_stencil, fbS_reg, state->valuemask);
          spe_clgt(f, stencil_pass_reg, tmp_reg, tmp_masked_stencil);
          spe_andc(f, stencil_pass_reg, fragment_mask_reg, stencil_pass_reg);
@@ -1453,6 +1454,7 @@ gen_stencil_values(struct spe_function *f,
 static void
 gen_get_stencil_values(struct spe_function *f,
                        const struct pipe_stencil_state *stencil,
+                       const unsigned ref_value,
                        const uint depth_enabled,
                        int fbS_reg, 
                        int *fail_reg,
@@ -1488,7 +1490,7 @@ gen_get_stencil_values(struct spe_function *f,
    }
    else {
       *fail_reg = spe_allocate_available_register(f);
-      gen_stencil_values(f, stencil->fail_op, stencil->ref_value, 
+      gen_stencil_values(f, stencil->fail_op, ref_value, 
          0xff, fbS_reg, *fail_reg);
    }
 
@@ -1501,7 +1503,7 @@ gen_get_stencil_values(struct spe_function *f,
    }
    else {
       *zfail_reg = spe_allocate_available_register(f);
-      gen_stencil_values(f, stencil->zfail_op, stencil->ref_value, 
+      gen_stencil_values(f, stencil->zfail_op, ref_value, 
          0xff, fbS_reg, *zfail_reg);
    }
 
@@ -1516,7 +1518,7 @@ gen_get_stencil_values(struct spe_function *f,
    }
    else {
       *zpass_reg = spe_allocate_available_register(f);
-      gen_stencil_values(f, stencil->zpass_op, stencil->ref_value, 
+      gen_stencil_values(f, stencil->zpass_op, ref_value, 
          0xff, fbS_reg, *zpass_reg);
    }
 }
@@ -1528,7 +1530,8 @@ gen_get_stencil_values(struct spe_function *f,
  */
 static boolean
 gen_stencil_depth_test(struct spe_function *f, 
-                       const struct pipe_depth_stencil_alpha_state *dsa, 
+                       const struct pipe_depth_stencil_alpha_state *dsa,
+                       const struct pipe_stencil_ref *stencil_ref,
                        const uint facing,
                        const int mask_reg, const int fragZ_reg, 
                        const int fbZ_reg, const int fbS_reg)
@@ -1551,6 +1554,7 @@ gen_stencil_depth_test(struct spe_function *f,
    int stencil_writemask_reg;
    int zmask_reg;
    int newS_reg;
+   unsigned ref_value;
 
    /* Stenciling is quite complex: up to six different configurable stencil 
     * operations/calculations can be required (three each for front-facing
@@ -1579,9 +1583,11 @@ gen_stencil_depth_test(struct spe_function *f,
     */
    if (facing == CELL_FACING_BACK && dsa->stencil[1].enabled) {
       stencil = &dsa->stencil[1];
+      ref_value = stencil_ref->ref_value[1];
    }
    else {
       stencil = &dsa->stencil[0];
+      ref_value = stencil_ref->ref_value[0];
    }
 
    /* Calculate the writemask.  If the writemask is trivial (either
@@ -1641,7 +1647,7 @@ gen_stencil_depth_test(struct spe_function *f,
     */
    spe_comment(f, 0, "Running basic stencil test");
    stencil_pass_reg = spe_allocate_available_register(f);
-   gen_stencil_test(f, stencil, 0xff, mask_reg, fbS_reg, stencil_pass_reg);
+   gen_stencil_test(f, stencil, ref_value, 0xff, mask_reg, fbS_reg, stencil_pass_reg);
 
    /* Generate code that, given the mask of valid fragments and the
     * mask of valid fragments that passed the stencil test, computes
@@ -1678,7 +1684,7 @@ gen_stencil_depth_test(struct spe_function *f,
       spe_comment(f, 0, facing == CELL_FACING_FRONT
                   ? "Computing front-facing stencil values"
                   : "Computing back-facing stencil values");
-      gen_get_stencil_values(f, stencil, dsa->depth.enabled, fbS_reg, 
+      gen_get_stencil_values(f, stencil, ref_value, dsa->depth.enabled, fbS_reg, 
          &stencil_fail_values, &stencil_pass_depth_fail_values, 
          &stencil_pass_depth_pass_values);
    }  
@@ -1818,6 +1824,7 @@ gen_stencil_depth_test(struct spe_function *f,
 static void
 gen_depth_stencil(struct cell_context *cell,
                   const struct pipe_depth_stencil_alpha_state *dsa,
+                  const struct pipe_stencil_ref *stencil_ref,
                   struct spe_function *f,
                   uint facing,
                   int mask_reg,
@@ -1940,7 +1947,7 @@ gen_depth_stencil(struct cell_context *cell,
        * gen_stencil_depth_test() function must ignore the
        * fbZ_reg register if depth is not enabled.
        */
-      write_depth_stencil = gen_stencil_depth_test(f, dsa, facing,
+      write_depth_stencil = gen_stencil_depth_test(f, dsa, stencil_ref, facing,
                                                    mask_reg, fragZ_reg,
                                                    fbZ_reg, fbS_reg);
    }
@@ -2029,6 +2036,7 @@ cell_gen_fragment_function(struct cell_context *cell,
                            struct spe_function *f)
 {
    const struct pipe_depth_stencil_alpha_state *dsa = cell->depth_stencil;
+   const struct pipe_stencil_ref *stencil_ref = &cell->stencil_ref;
    const struct pipe_blend_state *blend = cell->blend;
    const struct pipe_blend_color *blend_color = &cell->blend_color;
    const enum pipe_format color_format = cell->framebuffer.cbufs[0]->format;
@@ -2101,7 +2109,7 @@ cell_gen_fragment_function(struct cell_context *cell,
 
    /* generate depth and/or stencil test code */
    if (dsa->depth.enabled || dsa->stencil[0].enabled) {
-      gen_depth_stencil(cell, dsa, f,
+      gen_depth_stencil(cell, dsa, stencil_ref, f,
                         facing,
                         mask_reg,
                         depth_tile_reg,
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index 3259c58687c..3d8b4409c75 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -113,6 +113,19 @@ cell_delete_depth_stencil_alpha_state(struct pipe_context *pipe, void *dsa)
 
 
 static void
+cell_set_stencil_ref(struct pipe_context *pipe,
+                     const struct pipe_stencil_ref *stencil_ref)
+{
+   struct cell_context *cell = cell_context(pipe);
+
+   draw_flush(cell->draw);
+
+   cell->stencil_ref = *stencil_ref;
+
+   cell->dirty |= CELL_NEW_DEPTH_STENCIL;
+}
+
+static void
 cell_set_clip_state(struct pipe_context *pipe,
                     const struct pipe_clip_state *clip)
 {
@@ -397,6 +410,7 @@ cell_init_state_functions(struct cell_context *cell)
    cell->pipe.delete_rasterizer_state = cell_delete_rasterizer_state;
 
    cell->pipe.set_blend_color = cell_set_blend_color;
+   cell->pipe.set_stencil_ref = cell_set_stencil_ref;
    cell->pipe.set_clip_state = cell_set_clip_state;
 
    cell->pipe.set_framebuffer_state = cell_set_framebuffer_state;
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index 7681e3411e8..9214ac162be 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -33,6 +33,7 @@
 #include "pipe/p_screen.h"
 
 #include "cell/common.h"
+#include "cell_context.h"
 #include "cell_screen.h"
 #include "cell_texture.h"
 #include "cell_winsys.h"
diff --git a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
index 21af7ed1c3f..07be5e92ea7 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
@@ -282,6 +282,7 @@ emit_stencil_op(struct spe_function *f,
  */
 static int
 emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
+                  struct pipe_stencil_ref *sr,
                   unsigned face,
                   struct spe_function *f,
                   int mask,
@@ -296,7 +297,7 @@ emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
    int stencil_pass = spe_allocate_available_register(f);
    int face_stencil = spe_allocate_available_register(f);
    int stencil_src = stencil;
-   const unsigned ref = (dsa->stencil[face].ref_value
+   const unsigned ref = (sr->ref_value[face]
                          & dsa->stencil[face].valuemask);
    boolean complement = FALSE;
    int stored;
@@ -406,7 +407,7 @@ emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
 
          emit_stencil_op(f, face_stencil, stencil_src, stencil_fail,
                          dsa->stencil[face].fail_op,
-                         dsa->stencil[face].ref_value);
+                         sr->ref_value[face]);
 
          stencil_src = face_stencil;
       }
@@ -421,7 +422,7 @@ emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
 
          emit_stencil_op(f, face_stencil, stencil_src, depth_fail,
                          dsa->stencil[face].zfail_op,
-                         dsa->stencil[face].ref_value);
+                         sr->ref_value[face]);
          stencil_src = face_stencil;
       }
 
@@ -429,7 +430,7 @@ emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
           && (dsa->stencil[face].zpass_op != PIPE_STENCIL_OP_KEEP)) {
          emit_stencil_op(f, face_stencil, stencil_src, depth_pass,
                          dsa->stencil[face].zpass_op,
-                         dsa->stencil[face].ref_value);
+                         sr->ref_value[face]);
          stencil_src = face_stencil;
       }
    }
@@ -463,7 +464,8 @@ emit_stencil_test(struct pipe_depth_stencil_alpha_state *dsa,
 
 
 void
-cell_generate_depth_stencil_test(struct cell_depth_stencil_alpha_state *cdsa)
+cell_generate_depth_stencil_test(struct cell_depth_stencil_alpha_state *cdsa,
+                                 struct pipe_stencil_ref *sr)
 {
    struct pipe_depth_stencil_alpha_state *const dsa = &cdsa->base;
    struct spe_function *const f = &cdsa->code;
@@ -499,13 +501,13 @@ cell_generate_depth_stencil_test(struct cell_depth_stencil_alpha_state *cdsa)
 
    if (dsa->stencil[0].enabled) {
       const int front_depth_pass = spe_allocate_available_register(f);
-      int front_stencil = emit_stencil_test(dsa, 0, f, mask,
+      int front_stencil = emit_stencil_test(dsa, sr, 0, f, mask,
                                             depth_mask, depth_complement,
                                             stencil, front_depth_pass);
 
       if (dsa->stencil[1].enabled) {
          const int back_depth_pass = spe_allocate_available_register(f);
-         int back_stencil = emit_stencil_test(dsa, 1, f, mask,
+         int back_stencil = emit_stencil_test(dsa, sr, 1, f, mask,
                                               depth_mask,  depth_complement,
                                               stencil, back_depth_pass);
 
@@ -579,7 +581,7 @@ cell_generate_depth_stencil_test(struct cell_depth_stencil_alpha_state *cdsa)
                 dsa->stencil[i].zfail_op,
                 dsa->stencil[i].zpass_op);
          printf("#    ref value / value mask / write mask: %02x %02x %02x\n",
-                dsa->stencil[i].ref_value,
+                sr->ref_value[i],
                 dsa->stencil[i].valuemask,
                 dsa->stencil[i].writemask);
       }
diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h
index 191a44c3dfc..bb1a168ea7a 100644
--- a/src/gallium/drivers/failover/fo_context.h
+++ b/src/gallium/drivers/failover/fo_context.h
@@ -51,9 +51,10 @@
 #define FO_NEW_VERTEX          0x2000
 #define FO_NEW_VERTEX_SHADER   0x4000
 #define FO_NEW_BLEND_COLOR     0x8000
-#define FO_NEW_CLEAR_COLOR     0x10000
-#define FO_NEW_VERTEX_BUFFER   0x20000
-#define FO_NEW_VERTEX_ELEMENT  0x40000
+#define FO_NEW_STENCIL_REF     0x10000
+#define FO_NEW_CLEAR_COLOR     0x20000
+#define FO_NEW_VERTEX_BUFFER   0x40000
+#define FO_NEW_VERTEX_ELEMENT  0x80000
 
 
 
@@ -79,6 +80,7 @@ struct failover_context {
    const struct fo_state     *vertex_shader;
 
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
    struct pipe_framebuffer_state framebuffer;
    struct pipe_poly_stipple poly_stipple;
diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c
index c189d1d82cd..970606a3f50 100644
--- a/src/gallium/drivers/failover/fo_state.c
+++ b/src/gallium/drivers/failover/fo_state.c
@@ -90,7 +90,7 @@ failover_delete_blend_state( struct pipe_context *pipe,
 
 static void
 failover_set_blend_color( struct pipe_context *pipe,
-			  const struct pipe_blend_color *blend_color )
+                          const struct pipe_blend_color *blend_color )
 {
    struct failover_context *failover = failover_context(pipe);
 
@@ -100,9 +100,21 @@ failover_set_blend_color( struct pipe_context *pipe,
    failover->hw->set_blend_color( failover->hw, blend_color );
 }
 
+static void
+failover_set_stencil_ref( struct pipe_context *pipe,
+                          const struct pipe_stencil_ref *stencil_ref )
+{
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->stencil_ref = *stencil_ref;
+   failover->dirty |= FO_NEW_STENCIL_REF;
+   failover->sw->set_stencil_ref( failover->sw, stencil_ref );
+   failover->hw->set_stencil_ref( failover->hw, stencil_ref );
+}
+
 static void 
 failover_set_clip_state( struct pipe_context *pipe,
-			 const struct pipe_clip_state *clip )
+                         const struct pipe_clip_state *clip )
 {
    struct failover_context *failover = failover_context(pipe);
 
@@ -533,6 +545,7 @@ failover_init_state_functions( struct failover_context *failover )
    failover->pipe.delete_vs_state = failover_delete_vs_state;
 
    failover->pipe.set_blend_color = failover_set_blend_color;
+   failover->pipe.set_stencil_ref = failover_set_stencil_ref;
    failover->pipe.set_clip_state = failover_set_clip_state;
    failover->pipe.set_framebuffer_state = failover_set_framebuffer_state;
    failover->pipe.set_polygon_stipple = failover_set_polygon_stipple;
diff --git a/src/gallium/drivers/failover/fo_state_emit.c b/src/gallium/drivers/failover/fo_state_emit.c
index a3341e33f80..5c000808425 100644
--- a/src/gallium/drivers/failover/fo_state_emit.c
+++ b/src/gallium/drivers/failover/fo_state_emit.c
@@ -65,7 +65,10 @@ failover_state_emit( struct failover_context *failover )
 
    if (failover->dirty & FO_NEW_DEPTH_STENCIL)
       failover->sw->bind_depth_stencil_alpha_state( failover->sw,
-						    failover->depth_stencil->sw_state );
+                                                    failover->depth_stencil->sw_state );
+
+   if (failover->dirty & FO_NEW_STENCIL_REF)
+      failover->sw->set_stencil_ref( failover->sw, &failover->stencil_ref );
 
    if (failover->dirty & FO_NEW_FRAMEBUFFER)
       failover->sw->set_framebuffer_state( failover->sw, &failover->framebuffer );
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index 1479d2201a9..da769e7b290 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -232,6 +232,7 @@ struct i915_context
    struct i915_fragment_shader *fs;
 
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
    /* XXX unneded */
    struct pipe_buffer *constants[PIPE_SHADER_TYPES];
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index beb26e996a4..62169918e2b 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -190,7 +190,7 @@ static void i915_delete_blend_state(struct pipe_context *pipe, void *blend)
 }
 
 static void i915_set_blend_color( struct pipe_context *pipe,
-			     const struct pipe_blend_color *blend_color )
+                                  const struct pipe_blend_color *blend_color )
 {
    struct i915_context *i915 = i915_context(pipe);
    draw_flush(i915->draw);
@@ -200,6 +200,17 @@ static void i915_set_blend_color( struct pipe_context *pipe,
    i915->dirty |= I915_NEW_BLEND;
 }
 
+static void i915_set_stencil_ref( struct pipe_context *pipe,
+                                  const struct pipe_stencil_ref *stencil_ref )
+{
+   struct i915_context *i915 = i915_context(pipe);
+   draw_flush(i915->draw);
+
+   i915->stencil_ref = *stencil_ref;
+
+   i915->dirty |= I915_NEW_DEPTH_STENCIL;
+}
+
 static void *
 i915_create_sampler_state(struct pipe_context *pipe,
                           const struct pipe_sampler_state *sampler)
@@ -217,10 +228,10 @@ i915_create_sampler_state(struct pipe_context *pipe,
    minFilt = translate_img_filter( sampler->min_img_filter );
    magFilt = translate_img_filter( sampler->mag_img_filter );
    
-   if (sampler->max_anisotropy > 1.0)
+   if (sampler->max_anisotropy > 1)
       minFilt = magFilt = FILTER_ANISOTROPIC;
 
-   if (sampler->max_anisotropy > 2.0) {
+   if (sampler->max_anisotropy > 2) {
       cso->state[0] |= SS2_MAX_ANISO_4;
    }
 
@@ -334,11 +345,9 @@ i915_create_depth_stencil_state(struct pipe_context *pipe,
       int fop  = i915_translate_stencil_op(depth_stencil->stencil[0].fail_op);
       int dfop = i915_translate_stencil_op(depth_stencil->stencil[0].zfail_op);
       int dpop = i915_translate_stencil_op(depth_stencil->stencil[0].zpass_op);
-      int ref  = depth_stencil->stencil[0].ref_value & 0xff;
 
       cso->stencil_LIS5 |= (S5_STENCIL_TEST_ENABLE |
                             S5_STENCIL_WRITE_ENABLE |
-                            (ref  << S5_STENCIL_REF_SHIFT) |
                             (test << S5_STENCIL_TEST_FUNC_SHIFT) |
                             (fop  << S5_STENCIL_FAIL_SHIFT) |
                             (dfop << S5_STENCIL_PASS_Z_FAIL_SHIFT) |
@@ -350,7 +359,6 @@ i915_create_depth_stencil_state(struct pipe_context *pipe,
       int fop   = i915_translate_stencil_op(depth_stencil->stencil[1].fail_op);
       int dfop  = i915_translate_stencil_op(depth_stencil->stencil[1].zfail_op);
       int dpop  = i915_translate_stencil_op(depth_stencil->stencil[1].zpass_op);
-      int ref   = depth_stencil->stencil[1].ref_value & 0xff;
       int tmask = depth_stencil->stencil[1].valuemask & 0xff;
       int wmask = depth_stencil->stencil[1].writemask & 0xff;
 
@@ -359,7 +367,6 @@ i915_create_depth_stencil_state(struct pipe_context *pipe,
                      BFO_ENABLE_STENCIL_TWO_SIDE |
                      BFO_ENABLE_STENCIL_REF |
                      BFO_STENCIL_TWO_SIDE |
-                     (ref  << BFO_STENCIL_REF_SHIFT) |
                      (test << BFO_STENCIL_TEST_SHIFT) |
                      (fop  << BFO_STENCIL_FAIL_SHIFT) |
                      (dfop << BFO_STENCIL_PASS_Z_FAIL_SHIFT) |
@@ -777,6 +784,7 @@ i915_init_state_functions( struct i915_context *i915 )
    i915->base.delete_vs_state = i915_delete_vs_state;
 
    i915->base.set_blend_color = i915_set_blend_color;
+   i915->base.set_stencil_ref = i915_set_stencil_ref;
    i915->base.set_clip_state = i915_set_clip_state;
    i915->base.set_constant_buffer = i915_set_constant_buffer;
    i915->base.set_framebuffer_state = i915_set_framebuffer_state;
diff --git a/src/gallium/drivers/i915/i915_state_dynamic.c b/src/gallium/drivers/i915/i915_state_dynamic.c
index 86126a5a152..9c6723b3916 100644
--- a/src/gallium/drivers/i915/i915_state_dynamic.c
+++ b/src/gallium/drivers/i915/i915_state_dynamic.c
@@ -94,9 +94,16 @@ const struct i915_tracked_state i915_upload_MODES4 = {
 
 static void upload_BFO( struct i915_context *i915 )
 {
+   unsigned bfo[2];
+   bfo[0] = i915->depth_stencil->bfo[0];
+   bfo[1] = i915->depth_stencil->bfo[1];
+   /* I don't get it only allowed to set a ref mask when the enable bit is set? */
+   if (bfo[0] & BFO_ENABLE_STENCIL_REF) {
+      bfo[0] |= i915->stencil_ref.ref_value[1] << BFO_STENCIL_REF_SHIFT;
+   }
    set_dynamic_indirect( i915,
 			 I915_DYNAMIC_BFO_0,
-			 &(i915->depth_stencil->bfo[0]),
+			 &(bfo[0]),
 			 2 );
 }
 
diff --git a/src/gallium/drivers/i915/i915_state_immediate.c b/src/gallium/drivers/i915/i915_state_immediate.c
index 8c16bb4e271..d2c6f151434 100644
--- a/src/gallium/drivers/i915/i915_state_immediate.c
+++ b/src/gallium/drivers/i915/i915_state_immediate.c
@@ -129,6 +129,8 @@ static void upload_S5( struct i915_context *i915 )
    unsigned LIS5 = 0;
 
    LIS5 |= i915->depth_stencil->stencil_LIS5;
+   /* hope it's safe to set stencil ref value even if stencil test is disabled? */
+   LIS5 |= i915->stencil_ref.ref_value[0] << S5_STENCIL_REF_SHIFT;
 
    LIS5 |= i915->blend->LIS5;
 
diff --git a/src/gallium/drivers/i965/brw_cc.c b/src/gallium/drivers/i965/brw_cc.c
index 4a543276f54..cc8e380c68b 100644
--- a/src/gallium/drivers/i965/brw_cc.c
+++ b/src/gallium/drivers/i965/brw_cc.c
@@ -65,14 +65,33 @@ combine_cc3( struct brw_cc3 a, struct brw_cc3 b )
    return ca.cc3;
 }
 
+static INLINE struct brw_cc1
+combine_cc1( struct brw_cc1 a, struct brw_cc1 b )
+{
+   union { struct brw_cc1 cc1; unsigned i; } ca, cb;
+   ca.cc1 = a;
+   cb.cc1 = b;
+   ca.i |= cb.i;
+   return ca.cc1;
+}
+
+static INLINE struct brw_cc2
+combine_cc2( struct brw_cc2 a, struct brw_cc2 b )
+{
+   union { struct brw_cc2 cc2; unsigned i; } ca, cb;
+   ca.cc2 = a;
+   cb.cc2 = b;
+   ca.i |= cb.i;
+   return ca.cc2;
+}
 
 static int prepare_cc_unit( struct brw_context *brw )
 {
    brw->cc.cc.cc0 = brw->curr.zstencil->cc0;
-   brw->cc.cc.cc1 = brw->curr.zstencil->cc1;
-   brw->cc.cc.cc2 = brw->curr.zstencil->cc2;
+   brw->cc.cc.cc1 = combine_cc1( brw->curr.zstencil->cc1, brw->curr.cc1_stencil_ref );
+   brw->cc.cc.cc2 = combine_cc2( brw->curr.zstencil->cc2, brw->curr.blend->cc2 );
    brw->cc.cc.cc3 = combine_cc3( brw->curr.zstencil->cc3, brw->curr.blend->cc3 );
-   
+
    brw->cc.cc.cc5 = brw->curr.blend->cc5;
    brw->cc.cc.cc6 = brw->curr.blend->cc6;
    brw->cc.cc.cc7 = brw->curr.zstencil->cc7;
diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h
index 19fda423de0..12cfa7b049c 100644
--- a/src/gallium/drivers/i965/brw_context.h
+++ b/src/gallium/drivers/i965/brw_context.h
@@ -153,7 +153,6 @@ struct brw_blend_state {
    struct brw_surf_ss0 ss0;
 };
 
-
 struct brw_rasterizer_state;
 
 struct brw_immediate_data {
@@ -560,12 +559,14 @@ struct brw_context
 
       struct pipe_scissor_state scissor;
       struct pipe_viewport_state viewport;
+      struct pipe_stencil_ref stencil_ref;
       struct pipe_framebuffer_state fb;
       struct pipe_clip_state ucp;
       struct pipe_buffer *vertex_constants;
       struct pipe_buffer *fragment_constants;
 
       struct brw_blend_constant_color bcc;
+      struct brw_cc1 cc1_stencil_ref;
       struct brw_polygon_stipple bps;
       struct brw_cc_viewport ccv;
 
diff --git a/src/gallium/drivers/i965/brw_pipe_depth.c b/src/gallium/drivers/i965/brw_pipe_depth.c
index e010d76e0d3..b7000d5e334 100644
--- a/src/gallium/drivers/i965/brw_pipe_depth.c
+++ b/src/gallium/drivers/i965/brw_pipe_depth.c
@@ -72,7 +72,6 @@ static void create_bcc_state( struct brw_depth_stencil_state *zstencil,
 	 translate_stencil_op(templ->stencil[0].zfail_op);
       zstencil->cc0.stencil_pass_depth_pass_op =
 	 translate_stencil_op(templ->stencil[0].zpass_op);
-      zstencil->cc1.stencil_ref = templ->stencil[0].ref_value;
       zstencil->cc1.stencil_write_mask = templ->stencil[0].writemask;
       zstencil->cc1.stencil_test_mask = templ->stencil[0].valuemask;
 
@@ -86,7 +85,6 @@ static void create_bcc_state( struct brw_depth_stencil_state *zstencil,
 	    translate_stencil_op(templ->stencil[1].zfail_op);
 	 zstencil->cc0.bf_stencil_pass_depth_pass_op =
 	    translate_stencil_op(templ->stencil[1].zpass_op);
-	 zstencil->cc1.bf_stencil_ref = templ->stencil[1].ref_value;
 	 zstencil->cc2.bf_stencil_write_mask = templ->stencil[1].writemask;
 	 zstencil->cc2.bf_stencil_test_mask = templ->stencil[1].valuemask;
       }
@@ -159,9 +157,19 @@ static void brw_delete_depth_stencil_state(struct pipe_context *pipe,
    FREE(cso);
 }
 
+static void brw_set_stencil_ref(struct pipe_context *pipe,
+                                const struct pipe_stencil_ref *stencil_ref)
+{
+   struct brw_context *brw = brw_context(pipe);
+   brw->curr.cc1_stencil_ref.stencil_ref = stencil_ref->ref_value[0];
+   brw->curr.cc1_stencil_ref.bf_stencil_ref = stencil_ref->ref_value[1];
+
+   brw->state.dirty.mesa |= PIPE_NEW_DEPTH_STENCIL_ALPHA;
+}
 
 void brw_pipe_depth_stencil_init( struct brw_context *brw )
 {
+   brw->base.set_stencil_ref = brw_set_stencil_ref;
    brw->base.create_depth_stencil_alpha_state = brw_create_depth_stencil_state;
    brw->base.bind_depth_stencil_alpha_state = brw_bind_depth_stencil_state;
    brw->base.delete_depth_stencil_alpha_state = brw_delete_depth_stencil_state;
diff --git a/src/gallium/drivers/i965/brw_pipe_sampler.c b/src/gallium/drivers/i965/brw_pipe_sampler.c
index 6aab5610043..c7c0e2ae95e 100644
--- a/src/gallium/drivers/i965/brw_pipe_sampler.c
+++ b/src/gallium/drivers/i965/brw_pipe_sampler.c
@@ -114,14 +114,12 @@ brw_create_sampler_state( struct pipe_context *pipe,
 
    /* XXX: anisotropy logic slightly changed: 
     */
-   if (template->max_anisotropy > 1.0) {
+   if (template->max_anisotropy > 1) {
       sampler->ss0.min_filter = BRW_MAPFILTER_ANISOTROPIC; 
       sampler->ss0.mag_filter = BRW_MAPFILTER_ANISOTROPIC;
 
-      if (template->max_anisotropy > 2.0) {
-	 sampler->ss3.max_aniso = MIN2((template->max_anisotropy - 2) / 2,
-				       BRW_ANISORATIO_16);
-      }
+      sampler->ss3.max_aniso = MIN2((template->max_anisotropy - 2) / 2,
+                                    BRW_ANISORATIO_16);
    }
 
    sampler->ss1.r_wrap_mode = translate_wrap_mode(template->wrap_r);
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index 9955380e1fe..8248b2a4132 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -389,6 +389,17 @@ identity_set_blend_color(struct pipe_context *_pipe,
 }
 
 static void
+identity_set_stencil_ref(struct pipe_context *_pipe,
+                         const struct pipe_stencil_ref *stencil_ref)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   pipe->set_stencil_ref(pipe,
+                         stencil_ref);
+}
+
+static void
 identity_set_clip_state(struct pipe_context *_pipe,
                         const struct pipe_clip_state *clip)
 {
@@ -723,6 +734,7 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.bind_vs_state = identity_bind_vs_state;
    id_pipe->base.delete_vs_state = identity_delete_vs_state;
    id_pipe->base.set_blend_color = identity_set_blend_color;
+   id_pipe->base.set_stencil_ref = identity_set_stencil_ref;
    id_pipe->base.set_clip_state = identity_set_clip_state;
    id_pipe->base.set_constant_buffer = identity_set_constant_buffer;
    id_pipe->base.set_framebuffer_state = identity_set_framebuffer_state;
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index eded68bbc39..9120226de0c 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -146,6 +146,7 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.delete_vs_state = llvmpipe_delete_vs_state;
 
    llvmpipe->pipe.set_blend_color = llvmpipe_set_blend_color;
+   llvmpipe->pipe.set_stencil_ref = llvmpipe_set_stencil_ref;
    llvmpipe->pipe.set_clip_state = llvmpipe_set_clip_state;
    llvmpipe->pipe.set_constant_buffer = llvmpipe_set_constant_buffer;
    llvmpipe->pipe.set_framebuffer_state = llvmpipe_set_framebuffer_state;
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 3bde485ac0c..955c7eb8e0e 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -61,6 +61,7 @@ struct llvmpipe_context {
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
    struct pipe_buffer *constants[PIPE_SHADER_TYPES];
    struct pipe_framebuffer_state framebuffer;
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index 8f68f12bed7..9beba32271f 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -152,13 +152,16 @@ void llvmpipe_bind_rasterizer_state(struct pipe_context *, void *);
 void llvmpipe_delete_rasterizer_state(struct pipe_context *, void *);
 
 void llvmpipe_set_framebuffer_state( struct pipe_context *,
-			     const struct pipe_framebuffer_state * );
+                                     const struct pipe_framebuffer_state * );
 
 void llvmpipe_set_blend_color( struct pipe_context *pipe,
                                const struct pipe_blend_color *blend_color );
 
+void llvmpipe_set_stencil_ref( struct pipe_context *pipe,
+                               const struct pipe_stencil_ref *stencil_ref );
+
 void llvmpipe_set_clip_state( struct pipe_context *,
-			     const struct pipe_clip_state * );
+                              const struct pipe_clip_state * );
 
 void llvmpipe_set_constant_buffer(struct pipe_context *,
                                   uint shader, uint index,
diff --git a/src/gallium/drivers/llvmpipe/lp_state_blend.c b/src/gallium/drivers/llvmpipe/lp_state_blend.c
index 9b950e82d89..8bb89f96c27 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_blend.c
@@ -121,3 +121,24 @@ llvmpipe_delete_depth_stencil_state(struct pipe_context *pipe, void *depth)
 {
    FREE( depth );
 }
+
+void llvmpipe_set_stencil_ref( struct pipe_context *pipe,
+                               const struct pipe_stencil_ref *stencil_ref )
+{
+   struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+
+   if(!stencil_ref)
+      return;
+
+   if(memcmp(&llvmpipe->stencil_ref, stencil_ref, sizeof *stencil_ref) == 0)
+      return;
+
+   draw_flush(llvmpipe->draw);
+
+   memcpy(&llvmpipe->stencil_ref, stencil_ref, sizeof *stencil_ref);
+
+   /* not sure. want new flag? */
+   llvmpipe->dirty |= LP_NEW_DEPTH_STENCIL_ALPHA;
+}
+
+
diff --git a/src/gallium/drivers/nv30/nv30_context.h b/src/gallium/drivers/nv30/nv30_context.h
index b3b26f7f94a..ca3d6aca7fa 100644
--- a/src/gallium/drivers/nv30/nv30_context.h
+++ b/src/gallium/drivers/nv30/nv30_context.h
@@ -61,7 +61,8 @@ enum nv30_state_index {
 	NV30_STATE_VTXBUF = 31,
 	NV30_STATE_VTXFMT = 32,
 	NV30_STATE_VTXATTR = 33,
-	NV30_STATE_MAX = 34
+	NV30_STATE_SR = 34,
+	NV30_STATE_MAX = 35
 };
 
 #include "nv30_screen.h"
@@ -79,6 +80,7 @@ enum nv30_state_index {
 #define NV30_NEW_FRAGPROG	(1 << 10)
 #define NV30_NEW_ARRAYS		(1 << 11)
 #define NV30_NEW_UCP		(1 << 12)
+#define NV30_NEW_SR		(1 << 13)
 
 struct nv30_rasterizer_state {
 	struct pipe_rasterizer_state pipe;
@@ -129,6 +131,7 @@ struct nv30_context {
 	struct nv30_zsa_state *zsa;
 	struct nv30_blend_state *blend;
 	struct pipe_blend_color blend_colour;
+	struct pipe_stencil_ref stencil_ref;
 	struct pipe_viewport_state viewport;
 	struct pipe_framebuffer_state framebuffer;
 	struct pipe_buffer *idxbuf;
@@ -194,6 +197,7 @@ extern struct nv30_state_entry nv30_state_viewport;
 extern struct nv30_state_entry nv30_state_framebuffer;
 extern struct nv30_state_entry nv30_state_fragtex;
 extern struct nv30_state_entry nv30_state_vbo;
+extern struct nv30_state_entry nv30_state_sr;
 
 /* nv30_vbo.c */
 extern void nv30_draw_arrays(struct pipe_context *, unsigned mode,
diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c
index f775938ba79..d911c807074 100644
--- a/src/gallium/drivers/nv30/nv30_state.c
+++ b/src/gallium/drivers/nv30/nv30_state.c
@@ -139,13 +139,13 @@ nv30_sampler_state_create(struct pipe_context *pipe,
 
 	ps->en = 0;
 
-	if (cso->max_anisotropy >= 8.0) {
+	if (cso->max_anisotropy >= 8) {
 		ps->en |= NV34TCL_TX_ENABLE_ANISO_8X;
 	} else
-	if (cso->max_anisotropy >= 4.0) {
+	if (cso->max_anisotropy >= 4) {
 		ps->en |= NV34TCL_TX_ENABLE_ANISO_4X;
 	} else
-	if (cso->max_anisotropy >= 2.0) {
+	if (cso->max_anisotropy >= 2) {
 		ps->en |= NV34TCL_TX_ENABLE_ANISO_2X;
 	}
 
@@ -435,7 +435,7 @@ nv30_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 {
 	struct nv30_context *nv30 = nv30_context(pipe);
 	struct nv30_zsa_state *zsaso = CALLOC(1, sizeof(*zsaso));
-	struct nouveau_stateobj *so = so_new(5, 21, 0);
+	struct nouveau_stateobj *so = so_new(6, 20, 0);
 	struct nouveau_grobj *rankine = nv30->screen->rankine;
 
 	so_method(so, rankine, NV34TCL_DEPTH_FUNC, 3);
@@ -449,11 +449,11 @@ nv30_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 	so_data  (so, float_to_ubyte(cso->alpha.ref_value));
 
 	if (cso->stencil[0].enabled) {
-		so_method(so, rankine, NV34TCL_STENCIL_FRONT_ENABLE, 8);
+		so_method(so, rankine, NV34TCL_STENCIL_FRONT_ENABLE, 3);
 		so_data  (so, cso->stencil[0].enabled ? 1 : 0);
 		so_data  (so, cso->stencil[0].writemask);
 		so_data  (so, nvgl_comparison_op(cso->stencil[0].func));
-		so_data  (so, cso->stencil[0].ref_value);
+		so_method(so, rankine, NV34TCL_STENCIL_FRONT_FUNC_MASK, 4);
 		so_data  (so, cso->stencil[0].valuemask);
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].zfail_op));
@@ -464,11 +464,11 @@ nv30_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 	}
 
 	if (cso->stencil[1].enabled) {
-		so_method(so, rankine, NV34TCL_STENCIL_BACK_ENABLE, 8);
+		so_method(so, rankine, NV34TCL_STENCIL_BACK_ENABLE, 3);
 		so_data  (so, cso->stencil[1].enabled ? 1 : 0);
 		so_data  (so, cso->stencil[1].writemask);
 		so_data  (so, nvgl_comparison_op(cso->stencil[1].func));
-		so_data  (so, cso->stencil[1].ref_value);
+		so_method(so, rankine, NV34TCL_STENCIL_BACK_FUNC_MASK, 4);
 		so_data  (so, cso->stencil[1].valuemask);
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].zfail_op));
@@ -583,6 +583,16 @@ nv30_set_blend_color(struct pipe_context *pipe,
 }
 
 static void
+nv30_set_stencil_ref(struct pipe_context *pipe,
+		     const struct pipe_stencil_ref *sr)
+{
+	struct nv30_context *nv30 = nv30_context(pipe);
+
+	nv30->stencil_ref = *sr;
+	nv30->dirty |= NV30_NEW_SR;
+}
+
+static void
 nv30_set_clip_state(struct pipe_context *pipe,
 		    const struct pipe_clip_state *clip)
 {
@@ -704,6 +714,7 @@ nv30_init_state_functions(struct nv30_context *nv30)
 	nv30->pipe.delete_fs_state = nv30_fp_state_delete;
 
 	nv30->pipe.set_blend_color = nv30_set_blend_color;
+        nv30->pipe.set_stencil_ref = nv30_set_stencil_ref;
 	nv30->pipe.set_clip_state = nv30_set_clip_state;
 	nv30->pipe.set_constant_buffer = nv30_set_constant_buffer;
 	nv30->pipe.set_framebuffer_state = nv30_set_framebuffer_state;
diff --git a/src/gallium/drivers/nv30/nv30_state_emit.c b/src/gallium/drivers/nv30/nv30_state_emit.c
index d9650f63eb2..deefe7fd8db 100644
--- a/src/gallium/drivers/nv30/nv30_state_emit.c
+++ b/src/gallium/drivers/nv30/nv30_state_emit.c
@@ -12,6 +12,7 @@ static struct nv30_state_entry *render_states[] = {
 	&nv30_state_blend,
 	&nv30_state_blend_colour,
 	&nv30_state_zsa,
+	&nv30_state_sr,
 	&nv30_state_viewport,
 	&nv30_state_vbo,
 	NULL
diff --git a/src/gallium/drivers/nv30/nv30_state_zsa.c b/src/gallium/drivers/nv30/nv30_state_zsa.c
index 0940b7269b2..88cd74f180a 100644
--- a/src/gallium/drivers/nv30/nv30_state_zsa.c
+++ b/src/gallium/drivers/nv30/nv30_state_zsa.c
@@ -15,3 +15,27 @@ struct nv30_state_entry nv30_state_zsa = {
 		.hw = NV30_STATE_ZSA
 	}
 };
+
+static boolean
+nv30_state_sr_validate(struct nv30_context *nv30)
+{
+	struct nouveau_stateobj *so = so_new(2, 2, 0);
+	struct pipe_stencil_ref *sr = &nv30->stencil_ref;
+
+	so_method(so, nv30->screen->rankine, NV34TCL_STENCIL_FRONT_FUNC_REF, 1);
+	so_data  (so, sr->ref_value[0]);
+	so_method(so, nv30->screen->rankine, NV34TCL_STENCIL_BACK_FUNC_REF, 1);
+	so_data  (so, sr->ref_value[1]);
+
+	so_ref(so, &nv30->state.hw[NV30_STATE_SR]);
+	so_ref(NULL, &so);
+	return TRUE;
+}
+
+struct nv30_state_entry nv30_state_sr = {
+	.validate = nv30_state_sr_validate,
+	.dirty = {
+		.pipe = NV30_NEW_SR,
+		.hw = NV30_STATE_SR
+	}
+};
diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h
index 958a48f2a4a..4861924dac7 100644
--- a/src/gallium/drivers/nv40/nv40_context.h
+++ b/src/gallium/drivers/nv40/nv40_context.h
@@ -61,7 +61,8 @@ enum nv40_state_index {
 	NV40_STATE_VTXBUF = 31,
 	NV40_STATE_VTXFMT = 32,
 	NV40_STATE_VTXATTR = 33,
-	NV40_STATE_MAX = 34
+	NV40_STATE_SR = 34,
+	NV40_STATE_MAX = 35
 };
 
 #include "nv40_screen.h"
@@ -79,6 +80,7 @@ enum nv40_state_index {
 #define NV40_NEW_FRAGPROG	(1 << 10)
 #define NV40_NEW_ARRAYS		(1 << 11)
 #define NV40_NEW_UCP		(1 << 12)
+#define NV40_NEW_SR		(1 << 13)
 
 struct nv40_rasterizer_state {
 	struct pipe_rasterizer_state pipe;
@@ -144,6 +146,7 @@ struct nv40_context {
 	struct nv40_zsa_state *zsa;
 	struct nv40_blend_state *blend;
 	struct pipe_blend_color blend_colour;
+	struct pipe_stencil_ref stencil_ref;
 	struct pipe_viewport_state viewport;
 	struct pipe_framebuffer_state framebuffer;
 	struct pipe_buffer *idxbuf;
@@ -215,6 +218,7 @@ extern struct nv40_state_entry nv40_state_framebuffer;
 extern struct nv40_state_entry nv40_state_fragtex;
 extern struct nv40_state_entry nv40_state_vbo;
 extern struct nv40_state_entry nv40_state_vtxfmt;
+extern struct nv40_state_entry nv40_state_sr;
 
 /* nv40_vbo.c */
 extern void nv40_draw_arrays(struct pipe_context *, unsigned mode,
diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c
index 51b40e51e49..2073bf07353 100644
--- a/src/gallium/drivers/nv40/nv40_state.c
+++ b/src/gallium/drivers/nv40/nv40_state.c
@@ -132,26 +132,26 @@ nv40_sampler_state_create(struct pipe_context *pipe,
 		    (wrap_mode(cso->wrap_r) << NV40TCL_TEX_WRAP_R_SHIFT));
 
 	ps->en = 0;
-	if (cso->max_anisotropy >= 2.0) {
+	if (cso->max_anisotropy >= 2) {
 		/* no idea, binary driver sets it, works without it.. meh.. */
 		ps->wrap |= (1 << 5);
 
-		if (cso->max_anisotropy >= 16.0) {
+		if (cso->max_anisotropy >= 16) {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_16X;
 		} else
-		if (cso->max_anisotropy >= 12.0) {
+		if (cso->max_anisotropy >= 12) {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_12X;
 		} else
-		if (cso->max_anisotropy >= 10.0) {
+		if (cso->max_anisotropy >= 10) {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_10X;
 		} else
-		if (cso->max_anisotropy >= 8.0) {
+		if (cso->max_anisotropy >= 8) {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_8X;
 		} else
-		if (cso->max_anisotropy >= 6.0) {
+		if (cso->max_anisotropy >= 6) {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_6X;
 		} else
-		if (cso->max_anisotropy >= 4.0) {
+		if (cso->max_anisotropy >= 4) {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_4X;
 		} else {
 			ps->en |= NV40TCL_TEX_ENABLE_ANISO_2X;
@@ -445,7 +445,7 @@ nv40_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 {
 	struct nv40_context *nv40 = nv40_context(pipe);
 	struct nv40_zsa_state *zsaso = CALLOC(1, sizeof(*zsaso));
-	struct nouveau_stateobj *so = so_new(4, 21, 0);
+	struct nouveau_stateobj *so = so_new(6, 20, 0);
 	struct nouveau_grobj *curie = nv40->screen->curie;
 
 	so_method(so, curie, NV40TCL_DEPTH_FUNC, 3);
@@ -459,11 +459,11 @@ nv40_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 	so_data  (so, float_to_ubyte(cso->alpha.ref_value));
 
 	if (cso->stencil[0].enabled) {
-		so_method(so, curie, NV40TCL_STENCIL_FRONT_ENABLE, 8);
+		so_method(so, curie, NV40TCL_STENCIL_FRONT_ENABLE, 3);
 		so_data  (so, cso->stencil[0].enabled ? 1 : 0);
 		so_data  (so, cso->stencil[0].writemask);
 		so_data  (so, nvgl_comparison_op(cso->stencil[0].func));
-		so_data  (so, cso->stencil[0].ref_value);
+		so_method(so, curie, NV40TCL_STENCIL_FRONT_FUNC_MASK, 4);
 		so_data  (so, cso->stencil[0].valuemask);
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].zfail_op));
@@ -474,11 +474,11 @@ nv40_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 	}
 
 	if (cso->stencil[1].enabled) {
-		so_method(so, curie, NV40TCL_STENCIL_BACK_ENABLE, 8);
+		so_method(so, curie, NV40TCL_STENCIL_BACK_ENABLE, 3);
 		so_data  (so, cso->stencil[1].enabled ? 1 : 0);
 		so_data  (so, cso->stencil[1].writemask);
 		so_data  (so, nvgl_comparison_op(cso->stencil[1].func));
-		so_data  (so, cso->stencil[1].ref_value);
+		so_method(so, curie, NV40TCL_STENCIL_BACK_FUNC_MASK, 4);
 		so_data  (so, cso->stencil[1].valuemask);
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].zfail_op));
@@ -592,6 +592,16 @@ nv40_set_blend_color(struct pipe_context *pipe,
 	nv40->dirty |= NV40_NEW_BCOL;
 }
 
+ static void
+nv40_set_stencil_ref(struct pipe_context *pipe,
+		     const struct pipe_stencil_ref *sr)
+{
+	struct nv40_context *nv40 = nv40_context(pipe);
+
+	nv40->stencil_ref = *sr;
+	nv40->dirty |= NV40_NEW_SR;
+}
+
 static void
 nv40_set_clip_state(struct pipe_context *pipe,
 		    const struct pipe_clip_state *clip)
@@ -719,6 +729,7 @@ nv40_init_state_functions(struct nv40_context *nv40)
 	nv40->pipe.delete_fs_state = nv40_fp_state_delete;
 
 	nv40->pipe.set_blend_color = nv40_set_blend_color;
+        nv40->pipe.set_stencil_ref = nv40_set_stencil_ref;
 	nv40->pipe.set_clip_state = nv40_set_clip_state;
 	nv40->pipe.set_constant_buffer = nv40_set_constant_buffer;
 	nv40->pipe.set_framebuffer_state = nv40_set_framebuffer_state;
diff --git a/src/gallium/drivers/nv40/nv40_state_emit.c b/src/gallium/drivers/nv40/nv40_state_emit.c
index 1c4007a129e..8990f303ce4 100644
--- a/src/gallium/drivers/nv40/nv40_state_emit.c
+++ b/src/gallium/drivers/nv40/nv40_state_emit.c
@@ -13,6 +13,7 @@ static struct nv40_state_entry *render_states[] = {
 	&nv40_state_blend,
 	&nv40_state_blend_colour,
 	&nv40_state_zsa,
+	&nv40_state_sr,
 	&nv40_state_viewport,
 	&nv40_state_vbo,
 	NULL
@@ -29,6 +30,7 @@ static struct nv40_state_entry *swtnl_states[] = {
 	&nv40_state_blend,
 	&nv40_state_blend_colour,
 	&nv40_state_zsa,
+	&nv40_state_sr,
 	&nv40_state_viewport,
 	&nv40_state_vtxfmt,
 	NULL
diff --git a/src/gallium/drivers/nv40/nv40_state_zsa.c b/src/gallium/drivers/nv40/nv40_state_zsa.c
index fb760677c88..9cbe7da6db1 100644
--- a/src/gallium/drivers/nv40/nv40_state_zsa.c
+++ b/src/gallium/drivers/nv40/nv40_state_zsa.c
@@ -15,3 +15,27 @@ struct nv40_state_entry nv40_state_zsa = {
 		.hw = NV40_STATE_ZSA
 	}
 };
+
+static boolean
+nv40_state_sr_validate(struct nv40_context *nv40)
+{
+	struct nouveau_stateobj *so = so_new(2, 2, 0);
+	struct pipe_stencil_ref *sr = &nv40->stencil_ref;
+
+	so_method(so, nv40->screen->curie, NV40TCL_STENCIL_FRONT_FUNC_REF, 1);
+	so_data  (so, sr->ref_value[0]);
+	so_method(so, nv40->screen->curie, NV40TCL_STENCIL_BACK_FUNC_REF, 1);
+	so_data  (so, sr->ref_value[1]);
+
+	so_ref(so, &nv40->state.hw[NV40_STATE_SR]);
+	so_ref(NULL, &so);
+	return TRUE;
+}
+
+struct nv40_state_entry nv40_state_sr = {
+	.validate = nv40_state_sr_validate,
+	.dirty = {
+		.pipe = NV40_NEW_SR,
+		.hw = NV40_STATE_SR
+	}
+};
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index 14cef4c0bf8..b4de3e2ba57 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -50,6 +50,7 @@
 #define NV50_NEW_ARRAYS		(1 << 14)
 #define NV50_NEW_SAMPLER	(1 << 15)
 #define NV50_NEW_TEXTURE	(1 << 16)
+#define NV50_NEW_STENCIL_REF	(1 << 17)
 
 struct nv50_blend_stateobj {
 	struct pipe_blend_state pipe;
@@ -120,6 +121,7 @@ struct nv50_state {
 	struct nouveau_stateobj *blend;
 	struct nouveau_stateobj *blend_colour;
 	struct nouveau_stateobj *zsa;
+	struct nouveau_stateobj *stencil_ref;
 	struct nouveau_stateobj *rast;
 	struct nouveau_stateobj *stipple;
 	struct nouveau_stateobj *scissor;
@@ -155,6 +157,7 @@ struct nv50_context {
 	struct nv50_zsa_stateobj *zsa;
 	struct nv50_rasterizer_stateobj *rasterizer;
 	struct pipe_blend_color blend_colour;
+	struct pipe_stencil_ref stencil_ref;
 	struct pipe_poly_stipple stipple;
 	struct pipe_scissor_state scissor;
 	struct pipe_viewport_state viewport;
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index 7c531b50a50..7d304907b65 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -202,18 +202,18 @@ nv50_sampler_state_create(struct pipe_context *pipe,
 		break;
 	}
 
-	if (cso->max_anisotropy >= 16.0)
+	if (cso->max_anisotropy >= 16)
 		tsc[0] |= (7 << 20);
 	else
-	if (cso->max_anisotropy >= 12.0)
+	if (cso->max_anisotropy >= 12)
 		tsc[0] |= (6 << 20);
 	else {
-		tsc[0] |= (int)(cso->max_anisotropy * 0.5f) << 20;
+		tsc[0] |= (cso->max_anisotropy >> 1) << 20;
 
-		if (cso->max_anisotropy >= 4.0)
+		if (cso->max_anisotropy >= 4)
 			tsc[1] |= NV50TSC_1_1_UNKN_ANISO_35;
 		else
-		if (cso->max_anisotropy >= 2.0)
+		if (cso->max_anisotropy >= 2)
 			tsc[1] |= NV50TSC_1_1_UNKN_ANISO_15;
 	}
 
@@ -447,7 +447,7 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 {
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_zsa_stateobj *zsa = CALLOC_STRUCT(nv50_zsa_stateobj);
-	struct nouveau_stateobj *so = so_new(8, 22, 0);
+	struct nouveau_stateobj *so = so_new(9, 21, 0);
 
 	so_method(so, tesla, NV50TCL_DEPTH_WRITE_ENABLE, 1);
 	so_data  (so, cso->depth.writemask ? 1 : 0);
@@ -462,13 +462,13 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 	}
 
 	if (cso->stencil[0].enabled) {
-		so_method(so, tesla, NV50TCL_STENCIL_FRONT_ENABLE, 8);
+		so_method(so, tesla, NV50TCL_STENCIL_FRONT_ENABLE, 5);
 		so_data  (so, 1);
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].fail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].zfail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[0].zpass_op));
 		so_data  (so, nvgl_comparison_op(cso->stencil[0].func));
-		so_data  (so, cso->stencil[0].ref_value);
+		so_method(so, tesla, NV50TCL_STENCIL_FRONT_MASK, 2);
 		so_data  (so, cso->stencil[0].writemask);
 		so_data  (so, cso->stencil[0].valuemask);
 	} else {
@@ -483,8 +483,7 @@ nv50_depth_stencil_alpha_state_create(struct pipe_context *pipe,
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].zfail_op));
 		so_data  (so, nvgl_stencil_op(cso->stencil[1].zpass_op));
 		so_data  (so, nvgl_comparison_op(cso->stencil[1].func));
-		so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 3);
-		so_data  (so, cso->stencil[1].ref_value);
+		so_method(so, tesla, NV50TCL_STENCIL_BACK_MASK, 2);
 		so_data  (so, cso->stencil[1].writemask);
 		so_data  (so, cso->stencil[1].valuemask);
 	} else {
@@ -633,6 +632,16 @@ nv50_set_blend_color(struct pipe_context *pipe,
 	nv50->dirty |= NV50_NEW_BLEND_COLOUR;
 }
 
+ static void
+nv50_set_stencil_ref(struct pipe_context *pipe,
+		     const struct pipe_stencil_ref *sr)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
+
+	nv50->stencil_ref = *sr;
+	nv50->dirty |= NV50_NEW_STENCIL_REF;
+}
+
 static void
 nv50_set_clip_state(struct pipe_context *pipe,
 		    const struct pipe_clip_state *clip)
@@ -761,6 +770,7 @@ nv50_init_state_functions(struct nv50_context *nv50)
 	nv50->pipe.delete_gs_state = nv50_gp_state_delete;
 
 	nv50->pipe.set_blend_color = nv50_set_blend_color;
+        nv50->pipe.set_stencil_ref = nv50_set_stencil_ref;
 	nv50->pipe.set_clip_state = nv50_set_clip_state;
 	nv50->pipe.set_constant_buffer = nv50_set_constant_buffer;
 	nv50->pipe.set_framebuffer_state = nv50_set_framebuffer_state;
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index ee28fa63c14..0be6abbb171 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -205,6 +205,8 @@ nv50_state_emit(struct nv50_context *nv50)
 			nv50->state.dirty |= NV50_NEW_RASTERIZER;
 		if (nv50->state.blend_colour)
 			nv50->state.dirty |= NV50_NEW_BLEND_COLOUR;
+		if (nv50->state.stencil_ref)
+			nv50->state.dirty |= NV50_NEW_STENCIL_REF;
 		if (nv50->state.stipple)
 			nv50->state.dirty |= NV50_NEW_STIPPLE;
 		if (nv50->state.scissor)
@@ -242,6 +244,8 @@ nv50_state_emit(struct nv50_context *nv50)
 		so_emit(chan, nv50->state.rast);
 	if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR)
 		so_emit(chan, nv50->state.blend_colour);
+	if (nv50->state.dirty & NV50_NEW_STENCIL_REF)
+		so_emit(chan, nv50->state.stencil_ref);
 	if (nv50->state.dirty & NV50_NEW_STIPPLE)
 		so_emit(chan, nv50->state.stipple);
 	if (nv50->state.dirty & NV50_NEW_SCISSOR)
@@ -325,6 +329,16 @@ nv50_state_validate(struct nv50_context *nv50)
 		so_ref(NULL, &so);
 	}
 
+	if (nv50->dirty & NV50_NEW_STENCIL_REF) {
+		so = so_new(2, 2, 0);
+		so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1);
+		so_data  (so, nv50->stencil_ref.ref_value[0]);
+		so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1);
+		so_data  (so, nv50->stencil_ref.ref_value[1]);
+		so_ref(so, &nv50->state.stencil_ref);
+		so_ref(NULL, &so);
+	}
+
 	if (nv50->dirty & NV50_NEW_STIPPLE) {
 		so = so_new(1, 32, 0);
 		so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index cdedb302209..faaf9523cb2 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -27,6 +27,7 @@ static void r300_blitter_save_states(struct r300_context* r300)
 {
     util_blitter_save_blend(r300->blitter, r300->blend_state.state);
     util_blitter_save_depth_stencil_alpha(r300->blitter, r300->dsa_state.state);
+    util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
     util_blitter_save_fragment_shader(r300->blitter, r300->fs);
     util_blitter_save_vertex_shader(r300->blitter, r300->vs);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 84617578128..ba64b3eb725 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -318,6 +318,8 @@ struct r300_context {
     struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
     int vertex_element_count;
 
+    struct pipe_stencil_ref stencil_ref;
+
     /* Bitmask of dirty state objects. */
     uint32_t dirty_state;
     /* Flag indicating whether or not the HW is dirty. */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index de6ba651d14..d31336ed76f 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -114,6 +114,7 @@ void r300_emit_dsa_state(struct r300_context* r300, void* state)
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     struct pipe_framebuffer_state* fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
+    struct pipe_stencil_ref stencil_ref = r300->stencil_ref;
     CS_LOCALS(r300);
 
     BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6);
@@ -128,10 +129,10 @@ void r300_emit_dsa_state(struct r300_context* r300, void* state)
         OUT_CS(0);
     }
 
-    OUT_CS(dsa->stencil_ref_mask);
+    OUT_CS(dsa->stencil_ref_mask | stencil_ref.ref_value[0]);
 
     if (r300screen->caps->is_r500) {
-        OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf);
+        OUT_CS_REG(R500_ZB_STENCILREFMASK_BF, dsa->stencil_ref_bf | stencil_ref.ref_value[1]);
     }
     END_CS;
 }
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 99ecae9f5f8..4d158cff7c7 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -425,7 +425,7 @@ static void*
             (r300_translate_stencil_op(state->stencil[0].zfail_op) <<
                 R300_S_FRONT_ZFAIL_OP_SHIFT);
 
-        dsa->stencil_ref_mask = (state->stencil[0].ref_value) |
+        dsa->stencil_ref_mask =
                 (state->stencil[0].valuemask << R300_STENCILMASK_SHIFT) |
                 (state->stencil[0].writemask << R300_STENCILWRITEMASK_SHIFT);
 
@@ -444,7 +444,7 @@ static void*
             if (caps->is_r500)
             {
                 dsa->z_buffer_control |= R500_STENCIL_REFMASK_FRONT_BACK;
-                dsa->stencil_ref_bf = (state->stencil[1].ref_value) |
+                dsa->stencil_ref_bf =
                     (state->stencil[1].valuemask <<
                     R300_STENCILMASK_SHIFT) |
                     (state->stencil[1].writemask <<
@@ -488,6 +488,14 @@ static void r300_delete_dsa_state(struct pipe_context* pipe,
     FREE(state);
 }
 
+static void r300_set_stencil_ref(struct pipe_context* pipe,
+                                 const struct pipe_stencil_ref* sr)
+{
+    struct r300_context* r300 = r300_context(pipe);
+    r300->stencil_ref = *sr;
+    r300->dsa_state.dirty = TRUE;
+}
+
 static void
     r300_set_framebuffer_state(struct pipe_context* pipe,
                                const struct pipe_framebuffer_state* state)
@@ -794,7 +802,7 @@ static void*
     sampler->filter0 |= r300_translate_tex_filters(state->min_img_filter,
                                                    state->mag_img_filter,
                                                    state->min_mip_filter,
-                                                   state->max_anisotropy > 1.0);
+                                                   state->max_anisotropy > 0);
 
     /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
     /* We must pass these to the emit function to clamp them properly. */
@@ -1112,6 +1120,8 @@ void r300_init_state_functions(struct r300_context* r300)
     r300->context.bind_depth_stencil_alpha_state = r300_bind_dsa_state;
     r300->context.delete_depth_stencil_alpha_state = r300_delete_dsa_state;
 
+    r300->context.set_stencil_ref = r300_set_stencil_ref;
+
     r300->context.set_framebuffer_state = r300_set_framebuffer_state;
 
     r300->context.create_fs_state = r300_create_fs_state;
diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h
index 5df6815221f..779ba2c7be4 100644
--- a/src/gallium/drivers/r300/r300_state_inlines.h
+++ b/src/gallium/drivers/r300/r300_state_inlines.h
@@ -312,15 +312,15 @@ static INLINE uint32_t r300_translate_tex_filters(int min, int mag, int mip,
     return retval;
 }
 
-static INLINE uint32_t r300_anisotropy(float max_aniso)
+static INLINE uint32_t r300_anisotropy(unsigned max_aniso)
 {
-    if (max_aniso >= 16.0f) {
+    if (max_aniso >= 16) {
         return R300_TX_MAX_ANISO_16_TO_1;
-    } else if (max_aniso >= 8.0f) {
+    } else if (max_aniso >= 8) {
         return R300_TX_MAX_ANISO_8_TO_1;
-    } else if (max_aniso >= 4.0f) {
+    } else if (max_aniso >= 4) {
         return R300_TX_MAX_ANISO_4_TO_1;
-    } else if (max_aniso >= 2.0f) {
+    } else if (max_aniso >= 2) {
         return R300_TX_MAX_ANISO_2_TO_1;
     } else {
         return R300_TX_MAX_ANISO_1_TO_1;
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 2b22ce256e6..ddc35bcd629 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -246,6 +246,7 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->pipe.delete_gs_state = softpipe_delete_gs_state;
 
    softpipe->pipe.set_blend_color = softpipe_set_blend_color;
+   softpipe->pipe.set_stencil_ref = softpipe_set_stencil_ref;
    softpipe->pipe.set_clip_state = softpipe_set_clip_state;
    softpipe->pipe.set_constant_buffer = softpipe_set_constant_buffer;
    softpipe->pipe.set_framebuffer_state = softpipe_set_framebuffer_state;
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 62f9e7aad3d..95def72c541 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -62,6 +62,7 @@ struct softpipe_context {
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
    struct pipe_buffer *constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
    struct pipe_framebuffer_state framebuffer;
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index a981775cbd3..499eebd671b 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -519,7 +519,7 @@ depth_stencil_test_quad(struct quad_stage *qs,
    failOp  = softpipe->depth_stencil->stencil[face].fail_op;
    zFailOp = softpipe->depth_stencil->stencil[face].zfail_op;
    zPassOp = softpipe->depth_stencil->stencil[face].zpass_op;
-   ref     = softpipe->depth_stencil->stencil[face].ref_value;
+   ref     = softpipe->stencil_ref.ref_value[face];
    wrtMask = softpipe->depth_stencil->stencil[face].writemask;
    valMask = softpipe->depth_stencil->stencil[face].valuemask;
 
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index a83cae73617..4370bbeaee2 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -132,13 +132,16 @@ void softpipe_bind_rasterizer_state(struct pipe_context *, void *);
 void softpipe_delete_rasterizer_state(struct pipe_context *, void *);
 
 void softpipe_set_framebuffer_state( struct pipe_context *,
-			     const struct pipe_framebuffer_state * );
+                                     const struct pipe_framebuffer_state * );
 
 void softpipe_set_blend_color( struct pipe_context *pipe,
                                const struct pipe_blend_color *blend_color );
 
+void softpipe_set_stencil_ref( struct pipe_context *pipe,
+                               const struct pipe_stencil_ref *stencil_ref );
+
 void softpipe_set_clip_state( struct pipe_context *,
-			     const struct pipe_clip_state * );
+                              const struct pipe_clip_state * );
 
 void softpipe_set_constant_buffer(struct pipe_context *,
                                   uint shader, uint index,
diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c
index 95ab3234337..c63a49e90b0 100644
--- a/src/gallium/drivers/softpipe/sp_state_blend.c
+++ b/src/gallium/drivers/softpipe/sp_state_blend.c
@@ -61,7 +61,7 @@ void softpipe_delete_blend_state(struct pipe_context *pipe,
 
 
 void softpipe_set_blend_color( struct pipe_context *pipe,
-			     const struct pipe_blend_color *blend_color )
+                               const struct pipe_blend_color *blend_color )
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
@@ -80,7 +80,7 @@ void softpipe_set_blend_color( struct pipe_context *pipe,
 
 void *
 softpipe_create_depth_stencil_state(struct pipe_context *pipe,
-				    const struct pipe_depth_stencil_alpha_state *depth_stencil)
+                                    const struct pipe_depth_stencil_alpha_state *depth_stencil)
 {
    return mem_dup(depth_stencil, sizeof(*depth_stencil));
 }
@@ -101,3 +101,13 @@ softpipe_delete_depth_stencil_state(struct pipe_context *pipe, void *depth)
 {
    FREE( depth );
 }
+
+void softpipe_set_stencil_ref( struct pipe_context *pipe,
+                               const struct pipe_stencil_ref *stencil_ref )
+{
+   struct softpipe_context *softpipe = softpipe_context(pipe);
+
+   softpipe->stencil_ref = *stencil_ref;
+
+   softpipe->dirty |= SP_NEW_DEPTH_STENCIL_ALPHA;
+}
diff --git a/src/gallium/drivers/softpipe/sp_video_context.c b/src/gallium/drivers/softpipe/sp_video_context.c
index d8b5b31e954..7a3a6361670 100644
--- a/src/gallium/drivers/softpipe/sp_video_context.c
+++ b/src/gallium/drivers/softpipe/sp_video_context.c
@@ -209,7 +209,6 @@ init_pipe_state(struct sp_mpeg12_context *ctx)
       dsa.stencil[i].fail_op = PIPE_STENCIL_OP_KEEP;
       dsa.stencil[i].zpass_op = PIPE_STENCIL_OP_KEEP;
       dsa.stencil[i].zfail_op = PIPE_STENCIL_OP_KEEP;
-      dsa.stencil[i].ref_value = 0;
       dsa.stencil[i].valuemask = 0;
       dsa.stencil[i].writemask = 0;
    }
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index f9a641c6df6..03302e2a6ec 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -116,7 +116,6 @@ struct svga_depth_stencil_state {
    /* SVGA3D has one ref/mask/writemask triple shared between front &
     * back face stencil.  We really need two:
     */
-   unsigned stencil_ref:8;
    unsigned stencil_mask:8;
    unsigned stencil_writemask:8;
 
@@ -199,6 +198,7 @@ struct svga_state
    struct pipe_poly_stipple poly_stipple;
    struct pipe_scissor_state scissor;
    struct pipe_blend_color blend_color;
+   struct pipe_stencil_ref stencil_ref;
    struct pipe_clip_state clip;
    struct pipe_viewport_state viewport;
 
@@ -376,6 +376,7 @@ struct svga_context
 #define SVGA_NEW_VS_RESULT           0x1000000
 #define SVGA_NEW_ZERO_STRIDE         0x2000000
 #define SVGA_NEW_TEXTURE_FLAGS       0x4000000
+#define SVGA_NEW_STENCIL_REF         0x8000000
 
 
 
diff --git a/src/gallium/drivers/svga/svga_pipe_depthstencil.c b/src/gallium/drivers/svga/svga_pipe_depthstencil.c
index 12bbd233a58..c317bec6d57 100644
--- a/src/gallium/drivers/svga/svga_pipe_depthstencil.c
+++ b/src/gallium/drivers/svga/svga_pipe_depthstencil.c
@@ -89,7 +89,6 @@ svga_create_depth_stencil_state(struct pipe_context *pipe,
       /* SVGA3D has one ref/mask/writemask triple shared between front &
        * back face stencil.  We really need two:
        */
-      ds->stencil_ref       = templ->stencil[0].ref_value & 0xff;
       ds->stencil_mask      = templ->stencil[0].valuemask & 0xff;
       ds->stencil_writemask = templ->stencil[0].writemask & 0xff;
    }
@@ -102,7 +101,6 @@ svga_create_depth_stencil_state(struct pipe_context *pipe,
       ds->stencil[1].zfail  = svga_translate_stencil_op(templ->stencil[1].zfail_op);
       ds->stencil[1].pass   = svga_translate_stencil_op(templ->stencil[1].zpass_op);
 
-      ds->stencil_ref       = templ->stencil[1].ref_value & 0xff;
       ds->stencil_mask      = templ->stencil[1].valuemask & 0xff;
       ds->stencil_writemask = templ->stencil[1].writemask & 0xff;
    }
@@ -139,12 +137,24 @@ static void svga_delete_depth_stencil_state(struct pipe_context *pipe,
 }
 
 
+static void svga_set_stencil_ref( struct pipe_context *pipe,
+                                  const struct pipe_stencil_ref *stencil_ref )
+{
+   struct svga_context *svga = svga_context(pipe);
+
+   svga->curr.stencil_ref = *stencil_ref;
+
+   svga->dirty |= SVGA_NEW_STENCIL_REF;
+}
+
 
 void svga_init_depth_stencil_functions( struct svga_context *svga )
 {
    svga->pipe.create_depth_stencil_alpha_state = svga_create_depth_stencil_state;
    svga->pipe.bind_depth_stencil_alpha_state = svga_bind_depth_stencil_state;
    svga->pipe.delete_depth_stencil_alpha_state = svga_delete_depth_stencil_state;
+
+   svga->pipe.set_stencil_ref = svga_set_stencil_ref;
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_pipe_sampler.c b/src/gallium/drivers/svga/svga_pipe_sampler.c
index b70081343d1..2a9adfbb067 100644
--- a/src/gallium/drivers/svga/svga_pipe_sampler.c
+++ b/src/gallium/drivers/svga/svga_pipe_sampler.c
@@ -102,8 +102,8 @@ svga_create_sampler_state(struct pipe_context *pipe,
    cso->mipfilter = translate_mip_filter(sampler->min_mip_filter);
    cso->magfilter = translate_img_filter( sampler->mag_img_filter );
    cso->minfilter = translate_img_filter( sampler->min_img_filter );
-   cso->aniso_level = MAX2( (unsigned) sampler->max_anisotropy, 1 );
-   if(cso->aniso_level != 1)
+   cso->aniso_level = MAX2( sampler->max_anisotropy, 1 );
+   if(sampler->max_anisotropy)
       cso->magfilter = cso->minfilter = SVGA3D_TEX_FILTER_ANISOTROPIC;
    cso->lod_bias = sampler->lod_bias;
    cso->addressu = translate_wrap_mode(sampler->wrap_s);
diff --git a/src/gallium/drivers/svga/svga_state_rss.c b/src/gallium/drivers/svga/svga_state_rss.c
index 5ce9b4ef4f9..68828c252f4 100644
--- a/src/gallium/drivers/svga/svga_state_rss.c
+++ b/src/gallium/drivers/svga/svga_state_rss.c
@@ -26,6 +26,9 @@
 #include "util/u_inlines.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
+#if 0
+#include "util/u_pack_color.h"
+#endif
 
 #include "svga_context.h"
 #include "svga_state.h"
@@ -100,6 +103,23 @@ static int emit_rss( struct svga_context *svga,
       }
    }
 
+#if 0
+   /* FIXME: shouldn't we emit blend color here */
+   if (dirty & SVGA_NEW_BLEND_COLOR) {
+      union util_color uc;
+      ubyte r = float_to_ubyte(svga->curr.blend_color.color[0]);
+      ubyte g = float_to_ubyte(svga->curr.blend_color.color[1]);
+      ubyte b = float_to_ubyte(svga->curr.blend_color.color[2]);
+      ubyte a = float_to_ubyte(svga->curr.blend_color.color[3]);
+
+      util_pack_color_ub( r, g, b, a,
+                          PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
+
+      EMIT_RS( svga, uc.ui, BLENDCOLOR, fail );
+   }
+#endif
+
+
 
    if (dirty & (SVGA_NEW_DEPTH_STENCIL | SVGA_NEW_RAST)) {
       const struct svga_depth_stencil_state *curr = svga->curr.depth; 
@@ -123,8 +143,7 @@ static int emit_rss( struct svga_context *svga,
          EMIT_RS( svga, curr->stencil[0].fail,  STENCILFAIL, fail );
          EMIT_RS( svga, curr->stencil[0].zfail, STENCILZFAIL, fail );
          EMIT_RS( svga, curr->stencil[0].pass,  STENCILPASS, fail );
-         
-         EMIT_RS( svga, curr->stencil_ref, STENCILREF, fail );
+
          EMIT_RS( svga, curr->stencil_mask, STENCILMASK, fail );
          EMIT_RS( svga, curr->stencil_writemask, STENCILWRITEMASK, fail );
       }
@@ -160,7 +179,6 @@ static int emit_rss( struct svga_context *svga,
          EMIT_RS( svga, curr->stencil[ccw].zfail, CCWSTENCILZFAIL, fail );
          EMIT_RS( svga, curr->stencil[ccw].pass,  CCWSTENCILPASS, fail );
 
-         EMIT_RS( svga, curr->stencil_ref, STENCILREF, fail );
          EMIT_RS( svga, curr->stencil_mask, STENCILMASK, fail );
          EMIT_RS( svga, curr->stencil_writemask, STENCILWRITEMASK, fail );
       }
@@ -178,6 +196,9 @@ static int emit_rss( struct svga_context *svga,
       }
    }
 
+   if (dirty & SVGA_NEW_STENCIL_REF) {
+      EMIT_RS( svga, svga->curr.stencil_ref.ref_value[0], STENCILREF, fail );
+   }
 
    if (dirty & SVGA_NEW_RAST)
    {
@@ -257,7 +278,11 @@ struct svga_tracked_state svga_hw_rss =
    "hw rss state",
 
    (SVGA_NEW_BLEND |
+#if 0
+    SVGA_NEW_BLEND_COLOR |
+#endif
     SVGA_NEW_DEPTH_STENCIL |
+    SVGA_NEW_STENCIL_REF |
     SVGA_NEW_RAST |
     SVGA_NEW_FRAME_BUFFER |
     SVGA_NEW_NEED_PIPELINE),
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index 34ceaa41c15..866e228cecf 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -792,6 +792,24 @@ trace_context_set_blend_color(struct pipe_context *_pipe,
 
 
 static INLINE void
+trace_context_set_stencil_ref(struct pipe_context *_pipe,
+                              const struct pipe_stencil_ref *state)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+
+   trace_dump_call_begin("pipe_context", "set_stencil_ref");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(stencil_ref, state);
+
+   pipe->set_stencil_ref(pipe, state);
+
+   trace_dump_call_end();
+}
+
+
+static INLINE void
 trace_context_set_clip_state(struct pipe_context *_pipe,
                              const struct pipe_clip_state *state)
 {
@@ -1291,6 +1309,7 @@ trace_context_create(struct trace_screen *tr_scr,
    tr_ctx->base.bind_vs_state = trace_context_bind_vs_state;
    tr_ctx->base.delete_vs_state = trace_context_delete_vs_state;
    tr_ctx->base.set_blend_color = trace_context_set_blend_color;
+   tr_ctx->base.set_stencil_ref = trace_context_set_stencil_ref;
    tr_ctx->base.set_clip_state = trace_context_set_clip_state;
    tr_ctx->base.set_constant_buffer = trace_context_set_constant_buffer;
    tr_ctx->base.set_framebuffer_state = trace_context_set_framebuffer_state;
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 6648539a0fa..a4c72551269 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -301,7 +301,6 @@ void trace_dump_depth_stencil_alpha_state(const struct pipe_depth_stencil_alpha_
       trace_dump_member(uint, &state->stencil[i], fail_op);
       trace_dump_member(uint, &state->stencil[i], zpass_op);
       trace_dump_member(uint, &state->stencil[i], zfail_op);
-      trace_dump_member(uint, &state->stencil[i], ref_value);
       trace_dump_member(uint, &state->stencil[i], valuemask);
       trace_dump_member(uint, &state->stencil[i], writemask);
       trace_dump_struct_end();
@@ -381,6 +380,22 @@ void trace_dump_blend_color(const struct pipe_blend_color *state)
    trace_dump_struct_end();
 }
 
+void trace_dump_stencil_ref(const struct pipe_stencil_ref *state)
+{
+   if (!trace_dumping_enabled_locked())
+      return;
+
+   if(!state) {
+      trace_dump_null();
+      return;
+   }
+
+   trace_dump_struct_begin("pipe_stencil_ref");
+
+   trace_dump_member_array(uint, state, ref_value);
+
+   trace_dump_struct_end();
+}
 
 void trace_dump_framebuffer_state(const struct pipe_framebuffer_state *state)
 {
@@ -420,11 +435,11 @@ void trace_dump_sampler_state(const struct pipe_sampler_state *state)
    trace_dump_member(uint, state, compare_mode);
    trace_dump_member(uint, state, compare_func);
    trace_dump_member(bool, state, normalized_coords);
+   trace_dump_member(uint, state, max_anisotropy);
    trace_dump_member(float, state, lod_bias);
    trace_dump_member(float, state, min_lod);
    trace_dump_member(float, state, max_lod);
    trace_dump_member_array(float, state, border_color);
-   trace_dump_member(float, state, max_anisotropy);
 
    trace_dump_struct_end();
 }
diff --git a/src/gallium/drivers/trace/tr_dump_state.h b/src/gallium/drivers/trace/tr_dump_state.h
index c7860fd6e18..825519d0444 100644
--- a/src/gallium/drivers/trace/tr_dump_state.h
+++ b/src/gallium/drivers/trace/tr_dump_state.h
@@ -59,6 +59,8 @@ void trace_dump_blend_state(const struct pipe_blend_state *state);
 
 void trace_dump_blend_color(const struct pipe_blend_color *state);
 
+void trace_dump_stencil_ref(const struct pipe_stencil_ref *state);
+
 void trace_dump_framebuffer_state(const struct pipe_framebuffer_state *state);
 
 void trace_dump_sampler_state(const struct pipe_sampler_state *state);
diff --git a/src/gallium/include/pipe/p_context.h b/src/gallium/include/pipe/p_context.h
index f1e6a60e041..f82b77903e9 100644
--- a/src/gallium/include/pipe/p_context.h
+++ b/src/gallium/include/pipe/p_context.h
@@ -186,8 +186,11 @@ struct pipe_context {
    void (*set_blend_color)( struct pipe_context *,
                             const struct pipe_blend_color * );
 
+   void (*set_stencil_ref)( struct pipe_context *,
+                            const struct pipe_stencil_ref * );
+
    void (*set_clip_state)( struct pipe_context *,
-			   const struct pipe_clip_state * );
+                            const struct pipe_clip_state * );
 
    void (*set_constant_buffer)( struct pipe_context *,
                                 uint shader, uint index,
diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 68369570b95..5ac5c878135 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -199,9 +199,8 @@ struct pipe_stencil_state
    unsigned fail_op:3;  /**< PIPE_STENCIL_OP_x */
    unsigned zpass_op:3; /**< PIPE_STENCIL_OP_x */
    unsigned zfail_op:3; /**< PIPE_STENCIL_OP_x */
-   ubyte ref_value;
-   ubyte valuemask;
-   ubyte writemask;
+   unsigned valuemask:8;
+   unsigned writemask:8;
 };
 
 
@@ -251,6 +250,10 @@ struct pipe_blend_color
    float color[4];
 };
 
+struct pipe_stencil_ref
+{
+   ubyte ref_value[2];
+};
 
 struct pipe_framebuffer_state
 {
@@ -278,10 +281,10 @@ struct pipe_sampler_state
    unsigned compare_mode:1;      /**< PIPE_TEX_COMPARE_x */
    unsigned compare_func:3;      /**< PIPE_FUNC_x */
    unsigned normalized_coords:1; /**< Are coords normalized to [0,1]? */
+   unsigned max_anisotropy:6;
    float lod_bias;               /**< LOD/lambda bias */
    float min_lod, max_lod;       /**< LOD clamp range, after bias */
    float border_color[4];
-   float max_anisotropy;
 };
 
 
diff --git a/src/gallium/state_trackers/python/p_context.i b/src/gallium/state_trackers/python/p_context.i
index ce893dad453..3f36ccb6217 100644
--- a/src/gallium/state_trackers/python/p_context.i
+++ b/src/gallium/state_trackers/python/p_context.i
@@ -130,11 +130,15 @@ struct st_context {
    /*
     * Parameter-like state (or properties)
     */
-   
+
    void set_blend_color(const struct pipe_blend_color *state ) {
       cso_set_blend_color($self->cso, state);
    }
 
+   void set_stencil_ref(const struct pipe_stencil_ref *state ) {
+      cso_set_stencil_ref($self->cso, state);
+   }
+
    void set_clip(const struct pipe_clip_state *state ) {
       $self->pipe->set_clip_state($self->pipe, state);
    }
diff --git a/src/gallium/state_trackers/python/retrace/interpreter.py b/src/gallium/state_trackers/python/retrace/interpreter.py
index 190db43b082..7277701279c 100755
--- a/src/gallium/state_trackers/python/retrace/interpreter.py
+++ b/src/gallium/state_trackers/python/retrace/interpreter.py
@@ -430,12 +430,15 @@ class Context(Object):
 
     def delete_fs_state(self, state):
         pass
-    
+
     delete_vs_state = delete_fs_state
-    
+
     def set_blend_color(self, state):
         self.real.set_blend_color(state)
 
+    def set_stencil_ref(self, state):
+        self.real.set_stencil_ref(state)
+
     def set_clip_state(self, state):
         _state = gallium.Clip()
         _state.nr = state.nr
diff --git a/src/gallium/state_trackers/vega/polygon.c b/src/gallium/state_trackers/vega/polygon.c
index f56ea0c8b44..c06dbf52069 100644
--- a/src/gallium/state_trackers/vega/polygon.c
+++ b/src/gallium/state_trackers/vega/polygon.c
@@ -307,6 +307,7 @@ static void draw_polygon(struct vg_context *ctx,
 void polygon_fill(struct polygon *poly, struct vg_context *ctx)
 {
    struct pipe_depth_stencil_alpha_state dsa;
+   struct pipe_stencil_ref sr;
    struct pipe_blend_state blend;
    VGfloat bounds[4];
    VGfloat min_x, min_y, max_x, max_y;
@@ -325,6 +326,9 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
    set_blend_for_fill(&blend);
 
    memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
+   memset(&sr, 0, sizeof(struct pipe_stencil_ref));
+   /* only need a fixed 0. Rely on default or move it out at least? */
+   cso_set_stencil_ref(ctx->cso_context, &sr);
 
    cso_save_blend(ctx->cso_context);
    cso_save_depth_stencil_alpha(ctx->cso_context);
@@ -336,7 +340,6 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
       dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
       dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT;
       dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-      dsa.stencil[0].ref_value = 0;
       dsa.stencil[0].valuemask = ~0;
 
       cso_set_blend(ctx->cso_context, &blend);
@@ -352,7 +355,6 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
          dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
          dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
          dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].ref_value = 0;
          dsa.stencil[0].valuemask = ~0;
 
          /* back */
@@ -362,7 +364,6 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
          dsa.stencil[1].zfail_op = PIPE_STENCIL_OP_KEEP;
          dsa.stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
          dsa.stencil[1].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[1].ref_value = 0;
          dsa.stencil[1].valuemask = ~0;
 
          cso_set_blend(ctx->cso_context, &blend);
@@ -375,7 +376,6 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
 
          cso_save_rasterizer(ctx->cso_context);
          dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].ref_value = 0;
          dsa.stencil[0].valuemask = ~0;
 
          raster.cull_mode = raster.front_winding ^ PIPE_WINDING_BOTH;
@@ -407,7 +407,6 @@ void polygon_fill(struct polygon *poly, struct vg_context *ctx)
    dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
    dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
    dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].ref_value = 0;
    dsa.stencil[0].valuemask = dsa.stencil[0].writemask;
    dsa.stencil[1].enabled = 0;
    memcpy(&dsa.depth, &ctx->state.g3d.dsa.depth,
@@ -425,6 +424,7 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
 {
    struct array *polys = polyarray->array;
    struct pipe_depth_stencil_alpha_state dsa;
+   struct pipe_stencil_ref sr;
    struct pipe_blend_state blend;
    VGfloat min_x = polyarray->min_x;
    VGfloat min_y = polyarray->min_y;
@@ -442,6 +442,9 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
    set_blend_for_fill(&blend);
 
    memset(&dsa, 0, sizeof(struct pipe_depth_stencil_alpha_state));
+   memset(&sr, 0, sizeof(struct pipe_stencil_ref));
+   /* only need a fixed 0. Rely on default or move it out at least? */
+   cso_set_stencil_ref(ctx->cso_context, &sr);
 
    cso_save_blend(ctx->cso_context);
    cso_save_depth_stencil_alpha(ctx->cso_context);
@@ -453,7 +456,6 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
       dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
       dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INVERT;
       dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-      dsa.stencil[0].ref_value = 0;
       dsa.stencil[0].valuemask = ~0;
 
       cso_set_blend(ctx->cso_context, &blend);
@@ -472,7 +474,6 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
          dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_KEEP;
          dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_INCR_WRAP;
          dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].ref_value = 0;
          dsa.stencil[0].valuemask = ~0;
 
          /* back */
@@ -482,7 +483,6 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
          dsa.stencil[1].zfail_op = PIPE_STENCIL_OP_KEEP;
          dsa.stencil[1].zpass_op = PIPE_STENCIL_OP_DECR_WRAP;
          dsa.stencil[1].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[1].ref_value = 0;
          dsa.stencil[1].valuemask = ~0;
 
          cso_set_blend(ctx->cso_context, &blend);
@@ -498,7 +498,6 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
 
          cso_save_rasterizer(ctx->cso_context);
          dsa.stencil[0].func = PIPE_FUNC_ALWAYS;
-         dsa.stencil[0].ref_value = 0;
          dsa.stencil[0].valuemask = ~0;
 
          raster.cull_mode = raster.front_winding ^ PIPE_WINDING_BOTH;
@@ -536,7 +535,6 @@ void polygon_array_fill(struct polygon_array *polyarray, struct vg_context *ctx)
    dsa.stencil[0].fail_op = PIPE_STENCIL_OP_REPLACE;
    dsa.stencil[0].zfail_op = PIPE_STENCIL_OP_REPLACE;
    dsa.stencil[0].zpass_op = PIPE_STENCIL_OP_REPLACE;
-   dsa.stencil[0].ref_value = 0;
    dsa.stencil[0].valuemask = dsa.stencil[0].writemask;
    dsa.stencil[1].enabled = 0;
    memcpy(&dsa.depth, &ctx->state.g3d.dsa.depth,