softpipe: rework texture sampling code

Split into component pieces, stitch together at runtime using function
pointers.  Make it possible to utilize the existing fastpaths as image-level
filters for generic mip-filtering routines.

Remove special case for rectangle filtering, as it can now be handled by
the 2d path.

As most of the mesa demo texturing was already covered by fast paths, its
harder to find examples of speedups, but tunnel gets a boost as mip-nearest
filtering is now able to access the img_2d_linear_wrap_POT functions
for sampling within a mipmap level.

11 files changed, 1350 insertions, 1100 deletions

diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index a0196955c87..ef8faab3bd8 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -228,21 +228,6 @@ softpipe_create( struct pipe_screen *screen )
       softpipe->quad.depth_test = sp_quad_depth_test_stage(softpipe);
       softpipe->quad.blend = sp_quad_blend_stage(softpipe);
 
-   /* vertex shader samplers */
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      softpipe->tgsi.vert_samplers[i].base.get_samples = sp_get_samples;
-      softpipe->tgsi.vert_samplers[i].processor = TGSI_PROCESSOR_VERTEX;
-      softpipe->tgsi.vert_samplers[i].cache = softpipe->tex_cache[i];
-      softpipe->tgsi.vert_samplers_list[i] = &softpipe->tgsi.vert_samplers[i];
-   }
-
-   /* fragment shader samplers */
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples;
-      softpipe->tgsi.frag_samplers[i].processor = TGSI_PROCESSOR_FRAGMENT;
-      softpipe->tgsi.frag_samplers[i].cache = softpipe->tex_cache[i];
-      softpipe->tgsi.frag_samplers_list[i] = &softpipe->tgsi.frag_samplers[i];
-   }
 
    /*
     * Create drawing context and plug our rendering stage into it.
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index fa3306c020a..068a892f250 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -36,7 +36,6 @@
 #include "draw/draw_vertex.h"
 
 #include "sp_quad_pipe.h"
-#include "sp_tex_sample.h"
 
 
 struct softpipe_vbuf_render;
@@ -51,12 +50,12 @@ struct softpipe_context {
    struct pipe_context pipe;  /**< base class */
 
    /** Constant state objects */
-   const struct pipe_blend_state *blend;
-   const struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
-   const struct pipe_depth_stencil_alpha_state *depth_stencil;
-   const struct pipe_rasterizer_state *rasterizer;
-   const struct sp_fragment_shader *fs;
-   const struct sp_vertex_shader *vs;
+   struct pipe_blend_state *blend;
+   struct pipe_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   struct pipe_depth_stencil_alpha_state *depth_stencil;
+   struct pipe_rasterizer_state *rasterizer;
+   struct sp_fragment_shader *fs;
+   struct sp_vertex_shader *vs;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -123,10 +122,8 @@ struct softpipe_context {
 
    /** TGSI exec things */
    struct {
-      struct sp_shader_sampler vert_samplers[PIPE_MAX_SAMPLERS];
-      struct sp_shader_sampler *vert_samplers_list[PIPE_MAX_SAMPLERS];
-      struct sp_shader_sampler frag_samplers[PIPE_MAX_SAMPLERS];
-      struct sp_shader_sampler *frag_samplers_list[PIPE_MAX_SAMPLERS];
+      struct sp_sampler_varient *vert_samplers_list[PIPE_MAX_SAMPLERS];
+      struct sp_sampler_varient *frag_samplers_list[PIPE_MAX_SAMPLERS];
    } tgsi;
 
    /** The primitive drawing context */
@@ -155,5 +152,9 @@ softpipe_context( struct pipe_context *pipe )
    return (struct softpipe_context *)pipe;
 }
 
+void
+softpipe_reset_sampler_varients(struct softpipe_context *softpipe);
+
+
 #endif /* SP_CONTEXT_H */
 
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index 9776e978e3e..77ee3c1136b 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -87,6 +87,7 @@ struct sp_fragment_shader {
 struct sp_vertex_shader {
    struct pipe_shader_state shader;
    struct draw_vertex_shader *draw_data;
+   int max_sampler;             /* -1 if no samplers */
 };
 
 
diff --git a/src/gallium/drivers/softpipe/sp_state_blend.c b/src/gallium/drivers/softpipe/sp_state_blend.c
index 384fe559afd..efed082f823 100644
--- a/src/gallium/drivers/softpipe/sp_state_blend.c
+++ b/src/gallium/drivers/softpipe/sp_state_blend.c
@@ -45,7 +45,7 @@ void softpipe_bind_blend_state( struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   softpipe->blend = (const struct pipe_blend_state *)blend;
+   softpipe->blend = (struct pipe_blend_state *)blend;
 
    softpipe->dirty |= SP_NEW_BLEND;
 }
@@ -86,7 +86,7 @@ softpipe_bind_depth_stencil_state(struct pipe_context *pipe,
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
 
-   softpipe->depth_stencil = (const struct pipe_depth_stencil_alpha_state *)depth_stencil;
+   softpipe->depth_stencil = (struct pipe_depth_stencil_alpha_state *)depth_stencil;
 
    softpipe->dirty |= SP_NEW_DEPTH_STENCIL_ALPHA;
 }
diff --git a/src/gallium/drivers/softpipe/sp_state_derived.c b/src/gallium/drivers/softpipe/sp_state_derived.c
index 1f6e2ccb836..53109283320 100644
--- a/src/gallium/drivers/softpipe/sp_state_derived.c
+++ b/src/gallium/drivers/softpipe/sp_state_derived.c
@@ -198,19 +198,7 @@ update_tgsi_samplers( struct softpipe_context *softpipe )
 {
    unsigned i;
 
-   /* vertex shader samplers */
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      softpipe->tgsi.vert_samplers[i].sampler = softpipe->sampler[i];
-      softpipe->tgsi.vert_samplers[i].texture = softpipe->texture[i];
-      softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples;
-   }
-
-   /* fragment shader samplers */
-   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
-      softpipe->tgsi.frag_samplers[i].sampler = softpipe->sampler[i];
-      softpipe->tgsi.frag_samplers[i].texture = softpipe->texture[i];
-      softpipe->tgsi.frag_samplers[i].base.get_samples = sp_get_samples;
-   }
+   softpipe_reset_sampler_varients( softpipe );
 
    for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
       sp_tile_cache_validate_texture( softpipe->tex_cache[i] );
@@ -232,7 +220,9 @@ void softpipe_update_derived( struct softpipe_context *softpipe )
    }
       
    if (softpipe->dirty & (SP_NEW_SAMPLER |
-                          SP_NEW_TEXTURE))
+                          SP_NEW_TEXTURE |
+                          SP_NEW_FS | 
+                          SP_NEW_VS))
       update_tgsi_samplers( softpipe );
 
    if (softpipe->dirty & (SP_NEW_RASTERIZER |
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index 108ac8b9bb1..3a453219235 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -34,6 +34,7 @@
 #include "pipe/internal/p_winsys_screen.h"
 #include "pipe/p_shader_tokens.h"
 #include "draw/draw_context.h"
+#include "draw/draw_vs.h"
 #include "tgsi/tgsi_dump.h"
 #include "tgsi/tgsi_scan.h"
 #include "tgsi/tgsi_parse.h"
@@ -108,6 +109,8 @@ softpipe_create_vs_state(struct pipe_context *pipe,
    if (state->draw_data == NULL) 
       goto fail;
 
+   state->max_sampler = state->draw_data->info.file_max[TGSI_FILE_SAMPLER];
+
    return state;
 
 fail:
diff --git a/src/gallium/drivers/softpipe/sp_state_sampler.c b/src/gallium/drivers/softpipe/sp_state_sampler.c
index aa2f3f2ccd6..714e6380487 100644
--- a/src/gallium/drivers/softpipe/sp_state_sampler.c
+++ b/src/gallium/drivers/softpipe/sp_state_sampler.c
@@ -38,15 +38,32 @@
 #include "sp_state.h"
 #include "sp_texture.h"
 #include "sp_tile_cache.h"
+#include "sp_tex_sample.h"
 #include "draw/draw_context.h"
 
 
+struct sp_sampler {
+   struct pipe_sampler_state base;
+   struct sp_sampler_varient *varients;
+   struct sp_sampler_varient *current;
+};
+
+static struct sp_sampler *sp_sampler( struct pipe_sampler_state *sampler )
+{
+   return (struct sp_sampler *)sampler;
+}
+
 
 void *
 softpipe_create_sampler_state(struct pipe_context *pipe,
                               const struct pipe_sampler_state *sampler)
 {
-   return mem_dup(sampler, sizeof(*sampler));
+   struct sp_sampler *sp_sampler = CALLOC_STRUCT(sp_sampler);
+
+   sp_sampler->base = *sampler;
+   sp_sampler->varients = NULL;
+
+   return (void *)sp_sampler;
 }
 
 
@@ -106,10 +123,95 @@ softpipe_set_sampler_textures(struct pipe_context *pipe,
 }
 
 
+
+static struct sp_sampler_varient *
+get_sampler_varient( struct sp_sampler *sampler,
+                     struct pipe_texture *texture,
+                     unsigned processor )
+{
+   struct softpipe_texture *sp_texture = softpipe_texture(texture);
+   struct sp_sampler_varient *v = NULL;
+   union sp_sampler_key key;
+
+   key.bits.target = sp_texture->base.target;
+   key.bits.is_pot = sp_texture->pot;
+   key.bits.processor = processor;
+   key.bits.pad = 0;
+
+   if (sampler->current && 
+       key.value == sampler->current->key.value) {
+      v = sampler->current;
+   }
+
+   if (v == NULL) {
+      for (v = sampler->varients; v; v = v->next)
+         if (v->key.value == key.value)
+            break;
+
+      if (v == NULL) {
+         v = sp_create_sampler_varient( &sampler->base, key );
+         v->next = sampler->varients;
+         sampler->varients = v;
+      }
+   }
+   
+   sampler->current = v;
+   return v;
+}
+
+
+
+
+void
+softpipe_reset_sampler_varients(struct softpipe_context *softpipe)
+{
+   int i;
+
+   /* It's a bit hard to build these samplers ahead of time -- don't
+    * really know which samplers are going to be used for vertex and
+    * fragment programs.
+    */
+   for (i = 0; i <= softpipe->vs->max_sampler; i++) {
+      if (softpipe->sampler[i]) {
+         softpipe->tgsi.vert_samplers_list[i] = 
+            get_sampler_varient( sp_sampler(softpipe->sampler[i]),
+                                 softpipe->texture[i],
+                                 TGSI_PROCESSOR_VERTEX );
+
+         sp_sampler_varient_bind_texture( softpipe->tgsi.vert_samplers_list[i], 
+                                          softpipe->tex_cache[i],
+                                          softpipe->texture[i] );
+      }
+   }
+
+   for (i = 0; i <= softpipe->fs->info.file_max[TGSI_FILE_SAMPLER]; i++) {
+      if (softpipe->sampler[i]) {
+         softpipe->tgsi.frag_samplers_list[i] =
+            get_sampler_varient( sp_sampler(softpipe->sampler[i]),
+                                 softpipe->texture[i],
+                                 TGSI_PROCESSOR_FRAGMENT );
+
+         sp_sampler_varient_bind_texture( softpipe->tgsi.frag_samplers_list[i], 
+                                          softpipe->tex_cache[i],
+                                          softpipe->texture[i] );
+      }
+   }
+}
+
+
+
 void
 softpipe_delete_sampler_state(struct pipe_context *pipe,
                               void *sampler)
 {
+   struct sp_sampler *sp_sampler = (struct sp_sampler *)sampler;
+   struct sp_sampler_varient *v, *tmp;
+
+   for (v = sp_sampler->varients; v; v = tmp) {
+      tmp = v->next;
+      sp_sampler_varient_destroy(v);
+   }
+
    FREE( sampler );
 }
 
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index 6c75158d598..7bc689a298f 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -31,6 +31,7 @@
  *
  * Authors:
  *   Brian Paul
+ *   Keith Whitwell
  */
 
 #include "sp_context.h"
@@ -116,133 +117,157 @@ lerp_3d(float a, float b, float c,
  * \param icoord  returns the integer texcoords
  * \return  integer texture index
  */
-static INLINE void
-nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
+static void
+wrap_nearest_repeat(const float s[4], unsigned size,
+                        int icoord[4])
+{
+   uint ch;
+
+   /* s limited to [0,1) */
+   /* i limited to [0,size-1] */
+   for (ch = 0; ch < 4; ch++) {
+      int i = util_ifloor(s[ch] * size);
+      icoord[ch] = REMAINDER(i, size);
+   }
+}
+
+
+static void
+wrap_nearest_clamp(const float s[4], unsigned size,
                    int icoord[4])
 {
    uint ch;
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_REPEAT:
-      /* s limited to [0,1) */
-      /* i limited to [0,size-1] */
-      for (ch = 0; ch < 4; ch++) {
-         int i = util_ifloor(s[ch] * size);
-         icoord[ch] = REMAINDER(i, size);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP:
+   /* s limited to [0,1] */
+   /* i limited to [0,size-1] */
+   for (ch = 0; ch < 4; ch++) {
+      if (s[ch] <= 0.0F)
+         icoord[ch] = 0;
+      else if (s[ch] >= 1.0F)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(s[ch] * size);
+   }
+}
+
+
+static void
+wrap_nearest_clamp_to_edge(const float s[4], unsigned size,
+                           int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [0, size-1] */
+   const float min = 1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      if (s[ch] < min)
+         icoord[ch] = 0;
+      else if (s[ch] > max)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(s[ch] * size);
+   }
+}
+
+
+static void
+wrap_nearest_clamp_to_border(const float s[4], unsigned size,
+                             int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [-1, size] */
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      if (s[ch] <= min)
+         icoord[ch] = -1;
+      else if (s[ch] >= max)
+         icoord[ch] = size;
+      else
+         icoord[ch] = util_ifloor(s[ch] * size);
+   }
+}
+
+static void
+wrap_nearest_mirror_repeat(const float s[4], unsigned size,
+                           int icoord[4])
+{
+   uint ch;
+   const float min = 1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      const int flr = util_ifloor(s[ch]);
+      float u;
+      if (flr & 1)
+         u = 1.0F - (s[ch] - (float) flr);
+      else
+         u = s[ch] - (float) flr;
+      if (u < min)
+         icoord[ch] = 0;
+      else if (u > max)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(u * size);
+   }
+}
+
+static void
+wrap_nearest_mirror_clamp(const float s[4], unsigned size,
+                          int icoord[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
       /* s limited to [0,1] */
       /* i limited to [0,size-1] */
-      for (ch = 0; ch < 4; ch++) {
-         if (s[ch] <= 0.0F)
-            icoord[ch] = 0;
-         else if (s[ch] >= 1.0F)
-            icoord[ch] = size - 1;
-         else
-            icoord[ch] = util_ifloor(s[ch] * size);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [0, size-1] */
-         const float min = 1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            if (s[ch] < min)
-               icoord[ch] = 0;
-            else if (s[ch] > max)
-               icoord[ch] = size - 1;
-            else
-               icoord[ch] = util_ifloor(s[ch] * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [-1, size] */
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            if (s[ch] <= min)
-               icoord[ch] = -1;
-            else if (s[ch] >= max)
-               icoord[ch] = size;
-            else
-               icoord[ch] = util_ifloor(s[ch] * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_REPEAT:
-      {
-         const float min = 1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            const int flr = util_ifloor(s[ch]);
-            float u;
-            if (flr & 1)
-               u = 1.0F - (s[ch] - (float) flr);
-            else
-               u = s[ch] - (float) flr;
-            if (u < min)
-               icoord[ch] = 0;
-            else if (u > max)
-               icoord[ch] = size - 1;
-            else
-               icoord[ch] = util_ifloor(u * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         /* s limited to [0,1] */
-         /* i limited to [0,size-1] */
-         const float u = fabsf(s[ch]);
-         if (u <= 0.0F)
-            icoord[ch] = 0;
-         else if (u >= 1.0F)
-            icoord[ch] = size - 1;
-         else
-            icoord[ch] = util_ifloor(u * size);
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [0, size-1] */
-         const float min = 1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            const float u = fabsf(s[ch]);
-            if (u < min)
-               icoord[ch] = 0;
-            else if (u > max)
-               icoord[ch] = size - 1;
-            else
-               icoord[ch] = util_ifloor(u * size);
-         }
-      }
-      return;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
-      {
-         /* s limited to [min,max] */
-         /* i limited to [0, size-1] */
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            const float u = fabsf(s[ch]);
-            if (u < min)
-               icoord[ch] = -1;
-            else if (u > max)
-               icoord[ch] = size;
-            else
-               icoord[ch] = util_ifloor(u * size);
-         }
-      }
-      return;
-   default:
-      assert(0);
+      const float u = fabsf(s[ch]);
+      if (u <= 0.0F)
+         icoord[ch] = 0;
+      else if (u >= 1.0F)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(u * size);
+   }
+}
+
+static void
+wrap_nearest_mirror_clamp_to_edge(const float s[4], unsigned size,
+                           int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [0, size-1] */
+   const float min = 1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      const float u = fabsf(s[ch]);
+      if (u < min)
+         icoord[ch] = 0;
+      else if (u > max)
+         icoord[ch] = size - 1;
+      else
+         icoord[ch] = util_ifloor(u * size);
+   }
+}
+
+
+static void
+wrap_nearest_mirror_clamp_to_border(const float s[4], unsigned size,
+                                    int icoord[4])
+{
+   uint ch;
+   /* s limited to [min,max] */
+   /* i limited to [0, size-1] */
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   for (ch = 0; ch < 4; ch++) {
+      const float u = fabsf(s[ch]);
+      if (u < min)
+         icoord[ch] = -1;
+      else if (u > max)
+         icoord[ch] = size;
+      else
+         icoord[ch] = util_ifloor(u * size);
    }
 }
 
@@ -257,125 +282,151 @@ nearest_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
  * \param w  returns blend factor/weight between texture indexes
  * \param icoord  returns the computed integer texture coords
  */
-static INLINE void
-linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
+static void
+wrap_linear_repeat(const float s[4], unsigned size,
+                   int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+
+   for (ch = 0; ch < 4; ch++) {
+      float u = s[ch] * size - 0.5F;
+      icoord0[ch] = REMAINDER(util_ifloor(u), size);
+      icoord1[ch] = REMAINDER(icoord0[ch] + 1, size);
+      w[ch] = FRAC(u);
+   }
+}
+
+static void
+wrap_linear_clamp(const float s[4], unsigned size,
                   int icoord0[4], int icoord1[4], float w[4])
 {
    uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], 0.0F, 1.0F);
+      u = u * size - 0.5f;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = FRAC(u);
+   }
+}
 
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_REPEAT:
-      for (ch = 0; ch < 4; ch++) {
-         float u = s[ch] * size - 0.5F;
-         icoord0[ch] = REMAINDER(util_ifloor(u), size);
-         icoord1[ch] = REMAINDER(icoord0[ch] + 1, size);
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         float u = CLAMP(s[ch], 0.0F, 1.0F);
-         u = u * size - 0.5f;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      for (ch = 0; ch < 4; ch++) {
-         float u = CLAMP(s[ch], 0.0F, 1.0F);
-         u = u * size - 0.5f;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord0[ch] < 0)
-            icoord0[ch] = 0;
-         if (icoord1[ch] >= (int) size)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      {
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            float u = CLAMP(s[ch], min, max);
-            u = u * size - 0.5f;
-            icoord0[ch] = util_ifloor(u);
-            icoord1[ch] = icoord0[ch] + 1;
-            w[ch] = FRAC(u);
-         }
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_REPEAT:
-      for (ch = 0; ch < 4; ch++) {
-         const int flr = util_ifloor(s[ch]);
-         float u;
-         if (flr & 1)
-            u = 1.0F - (s[ch] - (float) flr);
-         else
-            u = s[ch] - (float) flr;
-         u = u * size - 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord0[ch] < 0)
-            icoord0[ch] = 0;
-         if (icoord1[ch] >= (int) size)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         float u = fabsf(s[ch]);
-         if (u >= 1.0F)
-            u = (float) size;
-         else
-            u *= size;
-         u -= 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
-      for (ch = 0; ch < 4; ch++) {
-         float u = fabsf(s[ch]);
-         if (u >= 1.0F)
-            u = (float) size;
-         else
-            u *= size;
-         u -= 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord0[ch] < 0)
-            icoord0[ch] = 0;
-         if (icoord1[ch] >= (int) size)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;;
-   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
-      {
-         const float min = -1.0F / (2.0F * size);
-         const float max = 1.0F - min;
-         for (ch = 0; ch < 4; ch++) {
-            float u = fabsf(s[ch]);
-            if (u <= min)
-               u = min * size;
-            else if (u >= max)
-               u = max * size;
-            else
-               u *= size;
-            u -= 0.5F;
-            icoord0[ch] = util_ifloor(u);
-            icoord1[ch] = icoord0[ch] + 1;
-            w[ch] = FRAC(u);
-         }
-      }
-      break;;
-   default:
-      assert(0);
+static void
+wrap_linear_clamp_to_edge(const float s[4], unsigned size,
+                          int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], 0.0F, 1.0F);
+      u = u * size - 0.5f;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord0[ch] < 0)
+         icoord0[ch] = 0;
+      if (icoord1[ch] >= (int) size)
+         icoord1[ch] = size - 1;
+      w[ch] = FRAC(u);
+   }
+}
+
+static void
+wrap_linear_clamp_to_border(const float s[4], unsigned size,
+                            int icoord0[4], int icoord1[4], float w[4])
+{
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], min, max);
+      u = u * size - 0.5f;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = FRAC(u);
+   }
+}
+
+
+static void
+wrap_linear_mirror_repeat(const float s[4], unsigned size,
+                          int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      const int flr = util_ifloor(s[ch]);
+      float u;
+      if (flr & 1)
+         u = 1.0F - (s[ch] - (float) flr);
+      else
+         u = s[ch] - (float) flr;
+      u = u * size - 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord0[ch] < 0)
+         icoord0[ch] = 0;
+      if (icoord1[ch] >= (int) size)
+         icoord1[ch] = size - 1;
+      w[ch] = FRAC(u);
+   }
+}
+
+static void
+wrap_linear_mirror_clamp(const float s[4], unsigned size,
+                         int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = fabsf(s[ch]);
+      if (u >= 1.0F)
+         u = (float) size;
+      else
+         u *= size;
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = FRAC(u);
+   }
+}
+
+static void
+wrap_linear_mirror_clamp_to_edge(const float s[4], unsigned size,
+                                 int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = fabsf(s[ch]);
+      if (u >= 1.0F)
+         u = (float) size;
+      else
+         u *= size;
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord0[ch] < 0)
+         icoord0[ch] = 0;
+      if (icoord1[ch] >= (int) size)
+         icoord1[ch] = size - 1;
+      w[ch] = FRAC(u);
+   }
+}
+
+static void
+wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size,
+                                   int icoord0[4], int icoord1[4], float w[4])
+{
+   const float min = -1.0F / (2.0F * size);
+   const float max = 1.0F - min;
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = fabsf(s[ch]);
+      if (u <= min)
+         u = min * size;
+      else if (u >= max)
+         u = max * size;
+      else
+         u *= size;
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = FRAC(u);
    }
 }
 
@@ -384,27 +435,26 @@ linear_texcoord_4(unsigned wrapMode, const float s[4], unsigned size,
  * For RECT textures / unnormalized texcoords
  * Only a subset of wrap modes supported.
  */
-static INLINE void
-nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
+static void
+wrap_nearest_unorm_clamp(const float s[4], unsigned size,
                           int icoord[4])
 {
    uint ch;
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         int i = util_ifloor(s[ch]);
-         icoord[ch]= CLAMP(i, 0, (int) size-1);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      /* fall-through */
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      for (ch = 0; ch < 4; ch++) {
-         icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
-      }
-      return;
-   default:
-      assert(0);
+   for (ch = 0; ch < 4; ch++) {
+      int i = util_ifloor(s[ch]);
+      icoord[ch]= CLAMP(i, 0, (int) size-1);
+   }
+}
+
+/* Handles clamp_to_edge and clamp_to_border:
+ */
+static void
+wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
+                               int icoord[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
    }
 }
 
@@ -413,157 +463,82 @@ nearest_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
  * For RECT textures / unnormalized texcoords.
  * Only a subset of wrap modes supported.
  */
-static INLINE void
-linear_texcoord_unnorm_4(unsigned wrapMode, const float s[4], unsigned size,
+static void
+wrap_linear_unorm_clamp(const float s[4], unsigned size,
                          int icoord0[4], int icoord1[4], float w[4])
 {
    uint ch;
-   switch (wrapMode) {
-   case PIPE_TEX_WRAP_CLAMP:
-      for (ch = 0; ch < 4; ch++) {
-         /* Not exactly what the spec says, but it matches NVIDIA output */
-         float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         w[ch] = FRAC(u);
-      }
-      return;
-   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
-      /* fall-through */
-   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
-      for (ch = 0; ch < 4; ch++) {
-         float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
-         u -= 0.5F;
-         icoord0[ch] = util_ifloor(u);
-         icoord1[ch] = icoord0[ch] + 1;
-         if (icoord1[ch] > (int) size - 1)
-            icoord1[ch] = size - 1;
-         w[ch] = FRAC(u);
-      }
-      break;
-   default:
-      assert(0);
+   for (ch = 0; ch < 4; ch++) {
+      /* Not exactly what the spec says, but it matches NVIDIA output */
+      float u = CLAMP(s[ch] - 0.5F, 0.0f, (float) size - 1.0f);
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      w[ch] = FRAC(u);
    }
 }
 
-
-static unsigned
-choose_cube_face(float rx, float ry, float rz, float *newS, float *newT)
+static void
+wrap_linear_unorm_clamp_to_border( const float s[4], unsigned size,
+                                   int icoord0[4], int icoord1[4], float w[4])
 {
-   /*
-      major axis
-      direction     target                             sc     tc    ma
-      ----------    -------------------------------    ---    ---   ---
-       +rx          TEXTURE_CUBE_MAP_POSITIVE_X_EXT    -rz    -ry   rx
-       -rx          TEXTURE_CUBE_MAP_NEGATIVE_X_EXT    +rz    -ry   rx
-       +ry          TEXTURE_CUBE_MAP_POSITIVE_Y_EXT    +rx    +rz   ry
-       -ry          TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT    +rx    -rz   ry
-       +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
-       -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
-   */
-   const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
-   unsigned face;
-   float sc, tc, ma;
-
-   if (arx > ary && arx > arz) {
-      if (rx >= 0.0F) {
-         face = PIPE_TEX_FACE_POS_X;
-         sc = -rz;
-         tc = -ry;
-         ma = arx;
-      }
-      else {
-         face = PIPE_TEX_FACE_NEG_X;
-         sc = rz;
-         tc = -ry;
-         ma = arx;
-      }
-   }
-   else if (ary > arx && ary > arz) {
-      if (ry >= 0.0F) {
-         face = PIPE_TEX_FACE_POS_Y;
-         sc = rx;
-         tc = rz;
-         ma = ary;
-      }
-      else {
-         face = PIPE_TEX_FACE_NEG_Y;
-         sc = rx;
-         tc = -rz;
-         ma = ary;
-      }
-   }
-   else {
-      if (rz > 0.0F) {
-         face = PIPE_TEX_FACE_POS_Z;
-         sc = rx;
-         tc = -ry;
-         ma = arz;
-      }
-      else {
-         face = PIPE_TEX_FACE_NEG_Z;
-         sc = -rx;
-         tc = -ry;
-         ma = arz;
-      }
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord1[ch] > (int) size - 1)
+         icoord1[ch] = size - 1;
+      w[ch] = FRAC(u);
    }
+}
+
 
-   *newS = ( sc / ma + 1.0F ) * 0.5F;
-   *newT = ( tc / ma + 1.0F ) * 0.5F;
 
-   return face;
-}
 
 
 /**
  * Examine the quad's texture coordinates to compute the partial
  * derivatives w.r.t X and Y, then compute lambda (level of detail).
- *
- * This is only done for fragment shaders, not vertex shaders.
  */
 static float
-compute_lambda(struct tgsi_sampler *tgsi_sampler,
-               const float s[QUAD_SIZE],
-               const float t[QUAD_SIZE],
-               const float p[QUAD_SIZE],
-               float lodbias)
+compute_lambda_1d(const struct sp_sampler_varient *samp,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias)
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
    const struct pipe_sampler_state *sampler = samp->sampler;
-   float rho, lambda;
+   float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+   float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
+   float rho = MAX2(dsdx, dsdy) * texture->width[0];
+   float lambda;
 
-   if (samp->processor == TGSI_PROCESSOR_VERTEX)
-      return lodbias;
+   lambda = util_fast_log2(rho);
+   lambda += lodbias + sampler->lod_bias;
+   lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
 
-   assert(sampler->normalized_coords);
+   return lambda;
+}
 
-   assert(s);
-   {
-      float dsdx = s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT];
-      float dsdy = s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT];
-      dsdx = fabsf(dsdx);
-      dsdy = fabsf(dsdy);
-      rho = MAX2(dsdx, dsdy) * texture->width[0];
-   }
-   if (t) {
-      float dtdx = t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT];
-      float dtdy = t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT];
-      float max;
-      dtdx = fabsf(dtdx);
-      dtdy = fabsf(dtdy);
-      max = MAX2(dtdx, dtdy) * texture->height[0];
-      rho = MAX2(rho, max);
-   }
-   if (p) {
-      float dpdx = p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT];
-      float dpdy = p[QUAD_TOP_LEFT]     - p[QUAD_BOTTOM_LEFT];
-      float max;
-      dpdx = fabsf(dpdx);
-      dpdy = fabsf(dpdy);
-      max = MAX2(dpdx, dpdy) * texture->depth[0];
-      rho = MAX2(rho, max);
-   }
+static float
+compute_lambda_2d(const struct sp_sampler_varient *samp,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias)
+{
+   const struct pipe_texture *texture = samp->texture;
+   const struct pipe_sampler_state *sampler = samp->sampler;
+   float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+   float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
+   float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
+   float dtdy = fabsf(t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT]);
+   float maxx = MAX2(dsdx, dsdy) * texture->width[0];
+   float maxy = MAX2(dtdx, dtdy) * texture->height[0];
+   float rho  = MAX2(maxx, maxy);
+   float lambda;
 
    lambda = util_fast_log2(rho);
    lambda += lodbias + sampler->lod_bias;
@@ -573,82 +548,50 @@ compute_lambda(struct tgsi_sampler *tgsi_sampler,
 }
 
 
-/**
- * Do several things here:
- * 1. Compute lambda from the texcoords, if needed
- * 2. Determine if we're minifying or magnifying
- * 3. If minifying, choose mipmap levels
- * 4. Return image filter to use within mipmap images
- * \param level0  Returns first mipmap level to sample from
- * \param level1  Returns second mipmap level to sample from
- * \param levelBlend  Returns blend factor between levels, in [0,1]
- * \param imgFilter  Returns either the min or mag filter, depending on lambda
- */
-static void
-choose_mipmap_levels(struct tgsi_sampler *tgsi_sampler,
-                     const float s[QUAD_SIZE],
-                     const float t[QUAD_SIZE],
-                     const float p[QUAD_SIZE],
-                     float lodbias,
-                     unsigned *level0, unsigned *level1, float *levelBlend,
-                     unsigned *imgFilter)
+static float
+compute_lambda_3d(const struct sp_sampler_varient *samp,
+                  const float s[QUAD_SIZE],
+                  const float t[QUAD_SIZE],
+                  const float p[QUAD_SIZE],
+                  float lodbias)
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
    const struct pipe_sampler_state *sampler = samp->sampler;
+   float dsdx = fabsf(s[QUAD_BOTTOM_RIGHT] - s[QUAD_BOTTOM_LEFT]);
+   float dsdy = fabsf(s[QUAD_TOP_LEFT]     - s[QUAD_BOTTOM_LEFT]);
+   float dtdx = fabsf(t[QUAD_BOTTOM_RIGHT] - t[QUAD_BOTTOM_LEFT]);
+   float dtdy = fabsf(t[QUAD_TOP_LEFT]     - t[QUAD_BOTTOM_LEFT]);
+   float dpdx = fabsf(p[QUAD_BOTTOM_RIGHT] - p[QUAD_BOTTOM_LEFT]);
+   float dpdy = fabsf(p[QUAD_TOP_LEFT]     - p[QUAD_BOTTOM_LEFT]);
+   float maxx = MAX2(dsdx, dsdy) * texture->width[0];
+   float maxy = MAX2(dtdx, dtdy) * texture->height[0];
+   float maxz = MAX2(dpdx, dpdy) * texture->depth[0];
+   float rho, lambda;
 
-   if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
-      /* no mipmap selection needed */
-      *level0 = *level1 = CLAMP((int) sampler->min_lod,
-                                0, (int) texture->last_level);
-
-      if (sampler->min_img_filter != sampler->mag_img_filter) {
-         /* non-mipmapped texture, but still need to determine if doing
-          * minification or magnification.
-          */
-         float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias);
-         if (lambda <= 0.0) {
-            *imgFilter = sampler->mag_img_filter;
-         }
-         else {
-            *imgFilter = sampler->min_img_filter;
-         }
-      }
-      else {
-         *imgFilter = sampler->mag_img_filter;
-      }
-   }
-   else {
-      float lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias);
+   rho = MAX2(maxx, maxy);
+   rho = MAX2(rho, maxz);
 
-      if (lambda <= 0.0) { /* XXX threshold depends on the filter */
-         /* magnifying */
-         *imgFilter = sampler->mag_img_filter;
-         *level0 = *level1 = 0;
-      }
-      else {
-         /* minifying */
-         *imgFilter = sampler->min_img_filter;
-
-         /* choose mipmap level(s) and compute the blend factor between them */
-         if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
-            /* Nearest mipmap level */
-            const int lvl = (int) (lambda + 0.5);
-            *level0 =
-            *level1 = CLAMP(lvl, 0, (int) texture->last_level);
-         }
-         else {
-            /* Linear interpolation between mipmap levels */
-            const int lvl = (int) lambda;
-            *level0 = CLAMP(lvl,     0, (int) texture->last_level);
-            *level1 = CLAMP(lvl + 1, 0, (int) texture->last_level);
-            *levelBlend = FRAC(lambda);  /* blending weight between levels */
-         }
-      }
-   }
+   lambda = util_fast_log2(rho);
+   lambda += lodbias + sampler->lod_bias;
+   lambda = CLAMP(lambda, sampler->min_lod, sampler->max_lod);
+
+   return lambda;
 }
 
 
+
+static float
+compute_lambda_vert(const struct sp_sampler_varient *samp,
+                    const float s[QUAD_SIZE],
+                    const float t[QUAD_SIZE],
+                    const float p[QUAD_SIZE],
+                    float lodbias)
+{
+   return lodbias;
+}
+
+
+
 /**
  * Get a texel from a texture, using the texture tile cache.
  *
@@ -663,12 +606,12 @@ choose_mipmap_levels(struct tgsi_sampler *tgsi_sampler,
  * XXX maybe move this into sp_tile_cache.c and merge with the
  * sp_get_cached_tile_tex() function.  Also, get 4 texels instead of 1...
  */
-static void
+static INLINE void
 get_texel_quad_2d(const struct tgsi_sampler *tgsi_sampler,
                   unsigned face, unsigned level, int x, int y, 
                   const float *out[4])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
 
    const struct softpipe_cached_tile *tile
       = sp_get_cached_tile_tex(samp->cache,
@@ -687,7 +630,7 @@ static INLINE const float *
 get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler,
                  unsigned face, unsigned level, int x, int y)
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
 
    const struct softpipe_cached_tile *tile
       = sp_get_cached_tile_tex(samp->cache,
@@ -700,7 +643,7 @@ get_texel_2d_ptr(const struct tgsi_sampler *tgsi_sampler,
 }
 
 
-static void
+static INLINE void
 get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler,
                      unsigned face, unsigned level, 
                      int x0, int y0, 
@@ -717,12 +660,12 @@ get_texel_quad_2d_mt(const struct tgsi_sampler *tgsi_sampler,
    }
 }
 
-static void
+static INLINE void
 get_texel(const struct tgsi_sampler *tgsi_sampler,
                  unsigned face, unsigned level, int x, int y, int z,
                  float rgba[NUM_CHANNELS][QUAD_SIZE], unsigned j)
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
    const struct pipe_sampler_state *sampler = samp->sampler;
 
@@ -756,140 +699,18 @@ get_texel(const struct tgsi_sampler *tgsi_sampler,
 }
 
 
-/**
- * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
- * When we sampled the depth texture, the depth value was put into all
- * RGBA channels.  We look at the red channel here.
- * \param rgba  quad of (depth) texel values
- * \param p  texture 'P' components for four pixels in quad
- * \param j  which pixel in the quad to test [0..3]
- */
-static INLINE void
-shadow_compare(const struct pipe_sampler_state *sampler,
-               float rgba[NUM_CHANNELS][QUAD_SIZE],
-               const float p[QUAD_SIZE],
-               uint j)
-{
-   int k;
-   switch (sampler->compare_func) {
-   case PIPE_FUNC_LESS:
-      k = p[j] < rgba[0][j];
-      break;
-   case PIPE_FUNC_LEQUAL:
-      k = p[j] <= rgba[0][j];
-      break;
-   case PIPE_FUNC_GREATER:
-      k = p[j] > rgba[0][j];
-      break;
-   case PIPE_FUNC_GEQUAL:
-      k = p[j] >= rgba[0][j];
-      break;
-   case PIPE_FUNC_EQUAL:
-      k = p[j] == rgba[0][j];
-      break;
-   case PIPE_FUNC_NOTEQUAL:
-      k = p[j] != rgba[0][j];
-      break;
-   case PIPE_FUNC_ALWAYS:
-      k = 1;
-      break;
-   case PIPE_FUNC_NEVER:
-      k = 0;
-      break;
-   default:
-      k = 0;
-      assert(0);
-      break;
-   }
 
-   /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
-   rgba[0][j] = rgba[1][j] = rgba[2][j] = (float) k;
-   rgba[3][j] = 1.0F;
-}
 
 
-/**
- * As above, but do four z/texture comparisons.
- */
 static INLINE void
-shadow_compare4(const struct pipe_sampler_state *sampler,
-                float rgba[NUM_CHANNELS][QUAD_SIZE],
-                const float p[QUAD_SIZE])
-{
-   int j, k0, k1, k2, k3;
-   float val;
-
-   /* compare four texcoords vs. four texture samples */
-   switch (sampler->compare_func) {
-   case PIPE_FUNC_LESS:
-      k0 = p[0] < rgba[0][0];
-      k1 = p[1] < rgba[0][1];
-      k2 = p[2] < rgba[0][2];
-      k3 = p[3] < rgba[0][3];
-      break;
-   case PIPE_FUNC_LEQUAL:
-      k0 = p[0] <= rgba[0][0];
-      k1 = p[1] <= rgba[0][1];
-      k2 = p[2] <= rgba[0][2];
-      k3 = p[3] <= rgba[0][3];
-      break;
-   case PIPE_FUNC_GREATER:
-      k0 = p[0] > rgba[0][0];
-      k1 = p[1] > rgba[0][1];
-      k2 = p[2] > rgba[0][2];
-      k3 = p[3] > rgba[0][3];
-      break;
-   case PIPE_FUNC_GEQUAL:
-      k0 = p[0] >= rgba[0][0];
-      k1 = p[1] >= rgba[0][1];
-      k2 = p[2] >= rgba[0][2];
-      k3 = p[3] >= rgba[0][3];
-      break;
-   case PIPE_FUNC_EQUAL:
-      k0 = p[0] == rgba[0][0];
-      k1 = p[1] == rgba[0][1];
-      k2 = p[2] == rgba[0][2];
-      k3 = p[3] == rgba[0][3];
-      break;
-   case PIPE_FUNC_NOTEQUAL:
-      k0 = p[0] != rgba[0][0];
-      k1 = p[1] != rgba[0][1];
-      k2 = p[2] != rgba[0][2];
-      k3 = p[3] != rgba[0][3];
-      break;
-   case PIPE_FUNC_ALWAYS:
-      k0 = k1 = k2 = k3 = 1;
-      break;
-   case PIPE_FUNC_NEVER:
-      k0 = k1 = k2 = k3 = 0;
-      break;
-   default:
-      k0 = k1 = k2 = k3 = 0;
-      assert(0);
-      break;
-   }
-
-   /* convert four pass/fail values to an intensity in [0,1] */
-   val = 0.25F * (k0 + k1 + k2 + k3);
-
-   /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
-   for (j = 0; j < 4; j++) {
-      rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
-      rgba[3][j] = 1.0F;
-   }
-}
-
-
-
-static void
-sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
-                                    const float s[QUAD_SIZE],
-                                    const float t[QUAD_SIZE],
-                                    const float p[QUAD_SIZE],
-                                    float lodbias,
-                                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+img_filter_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
+                                  const float s[QUAD_SIZE],
+                                  const float t[QUAD_SIZE],
+                                  const float p[QUAD_SIZE],
+                                  float lodbias,
+                                  float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    unsigned  j;
    unsigned level = samp->level;
    unsigned xpot = 1 << (samp->xpot - level);
@@ -940,15 +761,15 @@ sp_get_samples_2d_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
 }
 
 
-static void
-sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
-                                     const float s[QUAD_SIZE],
-                                     const float t[QUAD_SIZE],
-                                     const float p[QUAD_SIZE],
-                                     float lodbias,
-                                     float rgba[NUM_CHANNELS][QUAD_SIZE])
+static INLINE void
+img_filter_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
+                                 const float s[QUAD_SIZE],
+                                 const float t[QUAD_SIZE],
+                                 const float p[QUAD_SIZE],
+                                 float lodbias,
+                                 float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    unsigned  j;
    unsigned level = samp->level;
    unsigned xpot = 1 << (samp->xpot - level);
@@ -975,15 +796,15 @@ sp_get_samples_2d_nearest_repeat_POT(struct tgsi_sampler *tgsi_sampler,
 }
 
 
-static void
-sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
-                                     const float s[QUAD_SIZE],
-                                     const float t[QUAD_SIZE],
-                                     const float p[QUAD_SIZE],
-                                     float lodbias,
-                                     float rgba[NUM_CHANNELS][QUAD_SIZE])
+static INLINE void
+img_filter_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
+                                const float s[QUAD_SIZE],
+                                const float t[QUAD_SIZE],
+                                const float p[QUAD_SIZE],
+                                float lodbias,
+                                float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    unsigned  j;
    unsigned level = samp->level;
    unsigned xpot = 1 << (samp->xpot - level);
@@ -1018,238 +839,193 @@ sp_get_samples_2d_nearest_clamp_POT(struct tgsi_sampler *tgsi_sampler,
    }
 }
 
-
 static void
-sp_get_samples_2d_linear_mip_linear_repeat_POT(struct tgsi_sampler *tgsi_sampler,
-                                               const float s[QUAD_SIZE],
-                                               const float t[QUAD_SIZE],
-                                               const float p[QUAD_SIZE],
-                                               float lodbias,
-                                               float rgba[NUM_CHANNELS][QUAD_SIZE])
+img_filter_1d_nearest(struct tgsi_sampler *tgsi_sampler,
+                        const float s[QUAD_SIZE],
+                        const float t[QUAD_SIZE],
+                        const float p[QUAD_SIZE],
+                        float lodbias,
+                        float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
-   int level0;
-   float lambda;
-
-   lambda = compute_lambda(tgsi_sampler, s, t, p, lodbias);
-   level0 = (int)lambda;
+   unsigned level0, j;
+   int width;
+   int x[4];
 
-   if (lambda < 0.0) { 
-      samp->level = 0;
-      sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
-                                           s, t, p, 0, rgba );
-   }
-   else if (level0 >= texture->last_level) {
-      samp->level = texture->last_level;
-      sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
-                                           s, t, p, 0, rgba );
-   }
-   else {
-      float levelBlend = lambda - level0;
-      float rgba0[4][4];
-      float rgba1[4][4];
-      int c,j;
+   level0 = samp->level;
+   width = texture->width[level0];
 
-      samp->level = level0;
-      sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
-                                           s, t, p, 0, rgba0 );
+   assert(width > 0);
 
-      samp->level = level0+1;
-      sp_get_samples_2d_linear_repeat_POT( tgsi_sampler,
-                                           s, t, p, 0, rgba1 );
+   samp->nearest_texcoord_s(s, width, x);
 
-      for (j = 0; j < QUAD_SIZE; j++) {
-         for (c = 0; c < 4; c++) {
-            rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
-         }
-      }
+   for (j = 0; j < QUAD_SIZE; j++) {
+      get_texel(tgsi_sampler, 0, level0, x[j], 0, 0, rgba, j);
    }
 }
 
-/**
- * Common code for sampling 1D/2D/cube textures.
- * Could probably extend for 3D...
- */
+
 static void
-sp_get_samples_2d_common(struct tgsi_sampler *tgsi_sampler,
-                         const float s[QUAD_SIZE],
-                         const float t[QUAD_SIZE],
-                         const float p[QUAD_SIZE],
-                         float lodbias,
-                         float rgba[NUM_CHANNELS][QUAD_SIZE],
-                         const unsigned faces[4])
+img_filter_2d_nearest(struct tgsi_sampler *tgsi_sampler,
+                      const float s[QUAD_SIZE],
+                      const float t[QUAD_SIZE],
+                      const float p[QUAD_SIZE],
+                      float lodbias,
+                      float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
-   const struct pipe_sampler_state *sampler = samp->sampler;
-   unsigned level0, level1, j, imgFilter;
+   const unsigned *faces = samp->faces; /* zero when not cube-mapping */
+   unsigned level0, j;
    int width, height;
-   float levelBlend;
+   int x[4], y[4];
+
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+
+   assert(width > 0);
+
+   samp->nearest_texcoord_s(s, width, x);
+   samp->nearest_texcoord_t(t, height, y);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j);
+   }
+}
 
-   choose_mipmap_levels(tgsi_sampler, s, t, p, 
-                        lodbias,
-                        &level0, &level1, &levelBlend, &imgFilter);
 
-   assert(sampler->normalized_coords);
+static void
+img_filter_3d_nearest(struct tgsi_sampler *tgsi_sampler,
+                      const float s[QUAD_SIZE],
+                      const float t[QUAD_SIZE],
+                      const float p[QUAD_SIZE],
+                      float lodbias,
+                      float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width, height, depth;
+   int x[4], y[4], z[4];
 
+   level0 = samp->level;
    width = texture->width[level0];
    height = texture->height[level0];
+   depth = texture->depth[level0];
 
    assert(width > 0);
+   assert(height > 0);
+   assert(depth > 0);
 
-   switch (imgFilter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      {
-         int x[4], y[4];
-         nearest_texcoord_4(sampler->wrap_s, s, width, x);
-         nearest_texcoord_4(sampler->wrap_t, t, height, y);
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            get_texel(tgsi_sampler, faces[j], level0, x[j], y[j], 0, rgba, j);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare(sampler, rgba, p, j);
-            }
+   samp->nearest_texcoord_s(s, width,  x);
+   samp->nearest_texcoord_t(t, height, y);
+   samp->nearest_texcoord_p(p, depth,  z);
 
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               unsigned c;
-               x[j] /= 2;
-               y[j] /= 2;
-               get_texel(tgsi_sampler, faces[j], level1, x[j], y[j], 0,
-                         rgba2, j);
-               if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
-                  shadow_compare(sampler, rgba2, p, j);
-               }
-
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
-               }
-            }
-         }
-      }
-      break;
-   case PIPE_TEX_FILTER_LINEAR:
-   case PIPE_TEX_FILTER_ANISO:
-      {
-         int x0[4], y0[4], x1[4], y1[4];
-         float xw[4], yw[4]; /* weights */
-
-         linear_texcoord_4(sampler->wrap_s, s, width, x0, x1, xw);
-         linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            float tx[4][4]; /* texels */
-            int c;
-            get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0);
-            get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1);
-            get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2);
-            get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare4(sampler, tx, p);
-            }
+   for (j = 0; j < QUAD_SIZE; j++) {
+      get_texel(tgsi_sampler, 0, level0, x[j], y[j], z[j], rgba, j);
+   }
+}
 
-            /* interpolate R, G, B, A */
-            for (c = 0; c < 4; c++) {
-               rgba[c][j] = lerp_2d(xw[j], yw[j],
-                                    tx[c][0], tx[c][1],
-                                    tx[c][2], tx[c][3]);
-            }
 
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-
-               /* XXX: This is incorrect -- will often end up with (x0
-                *  == x1 && y0 == y1), meaning that we fetch the same
-                *  texel four times and linearly interpolate between
-                *  identical values.  The correct approach would be to
-                *  call linear_texcoord again for the second level.
-                */
-               x0[j] /= 2;
-               y0[j] /= 2;
-               x1[j] /= 2;
-               y1[j] /= 2;
-               get_texel(tgsi_sampler, faces[j], level1, x0[j], y0[j], 0, tx, 0);
-               get_texel(tgsi_sampler, faces[j], level1, x1[j], y0[j], 0, tx, 1);
-               get_texel(tgsi_sampler, faces[j], level1, x0[j], y1[j], 0, tx, 2);
-               get_texel(tgsi_sampler, faces[j], level1, x1[j], y1[j], 0, tx, 3);
-               if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE){
-                  shadow_compare4(sampler, tx, p);
-               }
-
-               /* interpolate R, G, B, A */
-               for (c = 0; c < 4; c++) {
-                  rgba2[c][j] = lerp_2d(xw[j], yw[j],
-                                        tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
-               }
-
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
-               }
-            }
-         }
+static void
+img_filter_1d_linear(struct tgsi_sampler *tgsi_sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   unsigned level0, j;
+   int width;
+   int x0[4], x1[4];
+   float xw[4]; /* weights */
+
+
+   level0 = samp->level;
+   width = texture->width[level0];
+
+   assert(width > 0);
+
+   samp->linear_texcoord_s(s, width, x0, x1, xw);
+
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      float tx[4][4]; /* texels */
+      int c;
+      get_texel(tgsi_sampler, 0, level0, x0[j], 0, 0, tx, 0);
+      get_texel(tgsi_sampler, 0, level0, x1[j], 0, 0, tx, 1);
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp(xw[j], tx[c][0], tx[c][1]);
       }
-      break;
-   default:
-      assert(0);
    }
 }
 
-
-static INLINE void
-sp_get_samples_1d(struct tgsi_sampler *sampler,
-                  const float s[QUAD_SIZE],
-                  const float t[QUAD_SIZE],
-                  const float p[QUAD_SIZE],
-                  float lodbias,
-                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+static void
+img_filter_2d_linear(struct tgsi_sampler *tgsi_sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   static const unsigned faces[4] = {0, 0, 0, 0};
-   static const float tzero[4] = {0, 0, 0, 0};
-   sp_get_samples_2d_common(sampler, s, tzero, NULL,
-                            lodbias, rgba, faces);
-}
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   const unsigned *faces = samp->faces; /* zero when not cube-mapping */
+   unsigned level0, j;
+   int width, height;
+   int x0[4], y0[4], x1[4], y1[4];
+   float xw[4], yw[4]; /* weights */
 
 
-static INLINE void
-sp_get_samples_2d(struct tgsi_sampler *sampler,
-                  const float s[QUAD_SIZE],
-                  const float t[QUAD_SIZE],
-                  const float p[QUAD_SIZE],
-                  float lodbias,
-                  float rgba[NUM_CHANNELS][QUAD_SIZE])
-{
-   static const unsigned faces[4] = {0, 0, 0, 0};
-   sp_get_samples_2d_common(sampler, s, t, p,
-                            lodbias, rgba, faces);
+   level0 = samp->level;
+   width = texture->width[level0];
+   height = texture->height[level0];
+
+   assert(width > 0);
+
+   samp->linear_texcoord_s(s, width,  x0, x1, xw);
+   samp->linear_texcoord_s(t, height, y0, y1, yw);
+
+   for (j = 0; j < QUAD_SIZE; j++) {
+      float tx[4][4]; /* texels */
+      int c;
+      get_texel(tgsi_sampler, faces[j], level0, x0[j], y0[j], 0, tx, 0);
+      get_texel(tgsi_sampler, faces[j], level0, x1[j], y0[j], 0, tx, 1);
+      get_texel(tgsi_sampler, faces[j], level0, x0[j], y1[j], 0, tx, 2);
+      get_texel(tgsi_sampler, faces[j], level0, x1[j], y1[j], 0, tx, 3);
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp_2d(xw[j], yw[j],
+                              tx[c][0], tx[c][1],
+                              tx[c][2], tx[c][3]);
+      }
+   }
 }
 
 
-static INLINE void
-sp_get_samples_3d(struct tgsi_sampler *tgsi_sampler,
-                  const float s[QUAD_SIZE],
-                  const float t[QUAD_SIZE],
-                  const float p[QUAD_SIZE],
-                  float lodbias,
-                  float rgba[NUM_CHANNELS][QUAD_SIZE])
+static void
+img_filter_3d_linear(struct tgsi_sampler *tgsi_sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   const struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
-   const struct pipe_sampler_state *sampler = samp->sampler;
-   /* get/map pipe_surfaces corresponding to 3D tex slices */
-   unsigned level0, level1, j, imgFilter;
+   unsigned level0, j;
    int width, height, depth;
-   float levelBlend;
-   const uint face = 0;
-
-   choose_mipmap_levels(tgsi_sampler, s, t, p, 
-                        lodbias,
-                        &level0, &level1, &levelBlend, &imgFilter);
-
-   assert(sampler->normalized_coords);
+   int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
+   float xw[4], yw[4], zw[4]; /* interpolation weights */
 
+   level0 = samp->level;
    width = texture->width[level0];
    height = texture->height[level0];
    depth = texture->depth[level0];
@@ -1258,323 +1034,631 @@ sp_get_samples_3d(struct tgsi_sampler *tgsi_sampler,
    assert(height > 0);
    assert(depth > 0);
 
-   switch (imgFilter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      {
-         int x[4], y[4], z[4];
-         nearest_texcoord_4(sampler->wrap_s, s, width, x);
-         nearest_texcoord_4(sampler->wrap_t, t, height, y);
-         nearest_texcoord_4(sampler->wrap_r, p, depth, z);
-         for (j = 0; j < QUAD_SIZE; j++) {
-            get_texel(tgsi_sampler, face, level0, x[j], y[j], z[j], rgba, j);
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               unsigned c;
-               x[j] /= 2;
-               y[j] /= 2;
-               z[j] /= 2;
-               get_texel(tgsi_sampler, face, level1, x[j], y[j], z[j], rgba2, j);
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba2[c][j], rgba[c][j]);
-               }
-            }
-         }
-      }
-      break;
-   case PIPE_TEX_FILTER_LINEAR:
-   case PIPE_TEX_FILTER_ANISO:
-      {
-         int x0[4], x1[4], y0[4], y1[4], z0[4], z1[4];
-         float xw[4], yw[4], zw[4]; /* interpolation weights */
-         linear_texcoord_4(sampler->wrap_s, s, width,  x0, x1, xw);
-         linear_texcoord_4(sampler->wrap_t, t, height, y0, y1, yw);
-         linear_texcoord_4(sampler->wrap_r, p, depth,  z0, z1, zw);
-
-         for (j = 0; j < QUAD_SIZE; j++) {
-            int c;
-            float tx0[4][4], tx1[4][4];
-            get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z0[j], tx0, 0);
-            get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z0[j], tx0, 1);
-            get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z0[j], tx0, 2);
-            get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z0[j], tx0, 3);
-            get_texel(tgsi_sampler, face, level0, x0[j], y0[j], z1[j], tx1, 0);
-            get_texel(tgsi_sampler, face, level0, x1[j], y0[j], z1[j], tx1, 1);
-            get_texel(tgsi_sampler, face, level0, x0[j], y1[j], z1[j], tx1, 2);
-            get_texel(tgsi_sampler, face, level0, x1[j], y1[j], z1[j], tx1, 3);
-
-            /* interpolate R, G, B, A */
-            for (c = 0; c < 4; c++) {
-               rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
-                                    tx0[c][0], tx0[c][1],
-                                    tx0[c][2], tx0[c][3],
-                                    tx1[c][0], tx1[c][1],
-                                    tx1[c][2], tx1[c][3]);
-            }
+   samp->linear_texcoord_s(s, width,  x0, x1, xw);
+   samp->linear_texcoord_s(t, height, y0, y1, yw);
+   samp->linear_texcoord_s(p, depth,  z0, z1, zw);
 
-            if (level0 != level1) {
-               /* get texels from second mipmap level and blend */
-               float rgba2[4][4];
-               x0[j] /= 2;
-               y0[j] /= 2;
-               z0[j] /= 2;
-               x1[j] /= 2;
-               y1[j] /= 2;
-               z1[j] /= 2;
-               get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z0[j], tx0, 0);
-               get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z0[j], tx0, 1);
-               get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z0[j], tx0, 2);
-               get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z0[j], tx0, 3);
-               get_texel(tgsi_sampler, face, level1, x0[j], y0[j], z1[j], tx1, 0);
-               get_texel(tgsi_sampler, face, level1, x1[j], y0[j], z1[j], tx1, 1);
-               get_texel(tgsi_sampler, face, level1, x0[j], y1[j], z1[j], tx1, 2);
-               get_texel(tgsi_sampler, face, level1, x1[j], y1[j], z1[j], tx1, 3);
-
-               /* interpolate R, G, B, A */
-               for (c = 0; c < 4; c++) {
-                  rgba2[c][j] = lerp_3d(xw[j], yw[j], zw[j],
-                                        tx0[c][0], tx0[c][1],
-                                        tx0[c][2], tx0[c][3],
-                                        tx1[c][0], tx1[c][1],
-                                        tx1[c][2], tx1[c][3]);
-               }
-
-               /* blend mipmap levels */
-               for (c = 0; c < NUM_CHANNELS; c++) {
-                  rgba[c][j] = lerp(levelBlend, rgba[c][j], rgba2[c][j]);
-               }
-            }
-         }
+   for (j = 0; j < QUAD_SIZE; j++) {
+      float tx0[4][4], tx1[4][4];
+      int c;
+      
+      get_texel(tgsi_sampler, 0, level0, x0[j], y0[j], z0[j], tx0, 0);
+      get_texel(tgsi_sampler, 0, level0, x1[j], y0[j], z0[j], tx0, 1);
+      get_texel(tgsi_sampler, 0, level0, x0[j], y1[j], z0[j], tx0, 2);
+      get_texel(tgsi_sampler, 0, level0, x1[j], y1[j], z0[j], tx0, 3);
+      get_texel(tgsi_sampler, 0, level0, x0[j], y0[j], z1[j], tx1, 0);
+      get_texel(tgsi_sampler, 0, level0, x1[j], y0[j], z1[j], tx1, 1);
+      get_texel(tgsi_sampler, 0, level0, x0[j], y1[j], z1[j], tx1, 2);
+      get_texel(tgsi_sampler, 0, level0, x1[j], y1[j], z1[j], tx1, 3);
+
+      /* interpolate R, G, B, A */
+      for (c = 0; c < 4; c++) {
+         rgba[c][j] = lerp_3d(xw[j], yw[j], zw[j],
+                              tx0[c][0], tx0[c][1],
+                              tx0[c][2], tx0[c][3],
+                              tx1[c][0], tx1[c][1],
+                              tx1[c][2], tx1[c][3]);
       }
-      break;
-   default:
-      assert(0);
    }
 }
 
 
+
+
+
+
+
 static void
-sp_get_samples_cube(struct tgsi_sampler *sampler,
-                    const float s[QUAD_SIZE],
-                    const float t[QUAD_SIZE],
-                    const float p[QUAD_SIZE],
-                    float lodbias,
-                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+mip_filter_linear(struct tgsi_sampler *tgsi_sampler,
+                     const float s[QUAD_SIZE],
+                     const float t[QUAD_SIZE],
+                     const float p[QUAD_SIZE],
+                     float lodbias,
+                     float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   unsigned faces[QUAD_SIZE], j;
-   float ssss[4], tttt[4];
-   for (j = 0; j < QUAD_SIZE; j++) {
-      faces[j] = choose_cube_face(s[j], t[j], p[j], ssss + j, tttt + j);
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   int level0;
+   float lambda;
+
+   lambda = samp->compute_lambda(samp, s, t, p, lodbias);
+   level0 = (int)lambda;
+
+   if (lambda < 0.0) { 
+      samp->level = 0;
+      samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else if (level0 >= texture->last_level) {
+      samp->level = texture->last_level;
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else {
+      float levelBlend = lambda - level0;
+      float rgba0[4][4];
+      float rgba1[4][4];
+      int c,j;
+
+      samp->level = level0;
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba0 );
+
+      samp->level = level0+1;
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba1 );
+
+      for (j = 0; j < QUAD_SIZE; j++) {
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
+         }
+      }
    }
-   sp_get_samples_2d_common(sampler, ssss, tttt, NULL,
-                            lodbias, rgba, faces);
 }
 
 
+
 static void
-sp_get_samples_rect(struct tgsi_sampler *tgsi_sampler,
-                    const float s[QUAD_SIZE],
-                    const float t[QUAD_SIZE],
-                    const float p[QUAD_SIZE],
-                    float lodbias,
-                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+mip_filter_nearest(struct tgsi_sampler *tgsi_sampler,
+                   const float s[QUAD_SIZE],
+                   const float t[QUAD_SIZE],
+                   const float p[QUAD_SIZE],
+                   float lodbias,
+                   float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   const struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    const struct pipe_texture *texture = samp->texture;
-   const struct pipe_sampler_state *sampler = samp->sampler;
-   const uint face = 0;
-   unsigned level0, level1, j, imgFilter;
-   int width, height;
-   float levelBlend;
+   float lambda;
 
-   choose_mipmap_levels(tgsi_sampler, s, t, p, 
-                        lodbias,
-                        &level0, &level1, &levelBlend, &imgFilter);
+   lambda = samp->compute_lambda(samp, s, t, p, lodbias);
 
-   /* texture RECTS cannot be mipmapped */
-   assert(level0 == level1);
+   if (lambda < 0.0) { 
+      samp->level = 0;
+      samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else {
+      samp->level = (int)(lambda + 0.5) ;
+      samp->level = MIN2(samp->level, (int)texture->last_level);
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+}
 
-   width = texture->width[level0];
-   height = texture->height[level0];
 
-   assert(width > 0);
+static void
+mip_filter_none(struct tgsi_sampler *tgsi_sampler,
+                const float s[QUAD_SIZE],
+                const float t[QUAD_SIZE],
+                const float p[QUAD_SIZE],
+                float lodbias,
+                float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   float lambda = samp->compute_lambda(samp, s, t, p, lodbias);
 
-   switch (imgFilter) {
-   case PIPE_TEX_FILTER_NEAREST:
-      {
-         int x[4], y[4];
-         nearest_texcoord_unnorm_4(sampler->wrap_s, s, width, x);
-         nearest_texcoord_unnorm_4(sampler->wrap_t, t, height, y);
-         for (j = 0; j < QUAD_SIZE; j++) {
-            get_texel(tgsi_sampler, face, level0, x[j], y[j], 0, rgba, j);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare(sampler, rgba, p, j);
-            }
-         }
-      }
-      break;
-   case PIPE_TEX_FILTER_LINEAR:
-   case PIPE_TEX_FILTER_ANISO:
-      {
-         int x0[4], y0[4], x1[4], y1[4];
-         float xw[4], yw[4]; /* weights */
-         linear_texcoord_unnorm_4(sampler->wrap_s, s, width,  x0, x1, xw);
-         linear_texcoord_unnorm_4(sampler->wrap_t, t, height, y0, y1, yw);
-         for (j = 0; j < QUAD_SIZE; j++) {
-            float tx[4][4]; /* texels */
-            int c;
-            get_texel(tgsi_sampler, face, level0, x0[j], y0[j], 0, tx, 0);
-            get_texel(tgsi_sampler, face, level0, x1[j], y0[j], 0, tx, 1);
-            get_texel(tgsi_sampler, face, level0, x0[j], y1[j], 0, tx, 2);
-            get_texel(tgsi_sampler, face, level0, x1[j], y1[j], 0, tx, 3);
-            if (sampler->compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
-               shadow_compare4(sampler, tx, p);
-            }
-            for (c = 0; c < 4; c++) {
-               rgba[c][j] = lerp_2d(xw[j], yw[j],
-                                    tx[c][0], tx[c][1], tx[c][2], tx[c][3]);
-            }
-         }
-      }
-      break;
-   default:
-      assert(0);
+   if (lambda < 0.0) { 
+      samp->mag_img_filter( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else {
+      samp->min_img_filter( tgsi_sampler, s, t, p, 0, rgba );
    }
 }
 
 
-/**
- * Error condition handler
+
+/* Specialized version of mip_filter_linear with hard-wired calls to
+ * 2d lambda calculation and 2d_linear_repeat_POT img filters.
  */
-static INLINE void
-sp_get_samples_null(struct tgsi_sampler *tgsi_sampler,
-                    const float s[QUAD_SIZE],
-                    const float t[QUAD_SIZE],
-                    const float p[QUAD_SIZE],
-                    float lodbias,
-                    float rgba[NUM_CHANNELS][QUAD_SIZE])
+static void
+mip_filter_linear_2d_linear_repeat_POT(
+   struct tgsi_sampler *tgsi_sampler,
+   const float s[QUAD_SIZE],
+   const float t[QUAD_SIZE],
+   const float p[QUAD_SIZE],
+   float lodbias,
+   float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   int i,j;
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   const struct pipe_texture *texture = samp->texture;
+   int level0;
+   float lambda;
+
+   lambda = compute_lambda_2d(samp, s, t, p, lodbias);
+   level0 = (int)lambda;
+
+   /* Catches both negative and large values of level0:
+    */
+   if ((unsigned)level0 >= texture->last_level) { 
+      if (level0 < 0)
+         samp->level = 0;
+      else
+         samp->level = texture->last_level;
 
-   for (i = 0; i < 4; i++)
-      for (j = 0; j < 4; j++)
-         rgba[i][j] = 1.0;
+      img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba );
+   }
+   else {
+      float levelBlend = lambda - level0;
+      float rgba0[4][4];
+      float rgba1[4][4];
+      int c,j;
+
+      samp->level = level0;
+      img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba0 );
+
+      samp->level = level0+1;
+      img_filter_2d_linear_repeat_POT( tgsi_sampler, s, t, p, 0, rgba1 );
+
+      for (j = 0; j < QUAD_SIZE; j++) {
+         for (c = 0; c < 4; c++) {
+            rgba[c][j] = lerp(levelBlend, rgba0[c][j], rgba1[c][j]);
+         }
+      }
+   }
 }
 
-/**
- * Called via tgsi_sampler::get_samples() when using a sampler for the
- * first time.  Determine the actual sampler function, link it in and
- * call it.
+
+
+/* Compare stage in the little sampling pipeline.
  */
-void
-sp_get_samples(struct tgsi_sampler *tgsi_sampler,
+static void
+sample_compare(struct tgsi_sampler *tgsi_sampler,
                const float s[QUAD_SIZE],
                const float t[QUAD_SIZE],
                const float p[QUAD_SIZE],
                float lodbias,
                float rgba[NUM_CHANNELS][QUAD_SIZE])
 {
-   struct sp_shader_sampler *samp = sp_shader_sampler(tgsi_sampler);
-   const struct pipe_texture *texture = samp->texture;
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
    const struct pipe_sampler_state *sampler = samp->sampler;
+   int j, k0, k1, k2, k3;
+   float val;
 
-   /* Default to the 'undefined' case:
-    */
-   tgsi_sampler->get_samples = sp_get_samples_null;
+   samp->mip_filter( tgsi_sampler, s, t, p, lodbias, rgba );
 
-   if (!texture) {
-      assert(0);                /* is this legal?? */
-      goto out;
-   }
 
-   if (!sampler->normalized_coords) {
-      assert (texture->target == PIPE_TEXTURE_2D);
-      tgsi_sampler->get_samples = sp_get_samples_rect;
-      goto out;
-   }
+   /**
+    * Compare texcoord 'p' (aka R) against texture value 'rgba[0]'
+    * When we sampled the depth texture, the depth value was put into all
+    * RGBA channels.  We look at the red channel here.
+    */
 
-   switch (texture->target) {
-   case PIPE_TEXTURE_1D:
-      tgsi_sampler->get_samples = sp_get_samples_1d;
+   /* compare four texcoords vs. four texture samples */
+   switch (sampler->compare_func) {
+   case PIPE_FUNC_LESS:
+      k0 = p[0] < rgba[0][0];
+      k1 = p[1] < rgba[0][1];
+      k2 = p[2] < rgba[0][2];
+      k3 = p[3] < rgba[0][3];
       break;
-   case PIPE_TEXTURE_2D:
-      tgsi_sampler->get_samples = sp_get_samples_2d;
+   case PIPE_FUNC_LEQUAL:
+      k0 = p[0] <= rgba[0][0];
+      k1 = p[1] <= rgba[0][1];
+      k2 = p[2] <= rgba[0][2];
+      k3 = p[3] <= rgba[0][3];
       break;
-   case PIPE_TEXTURE_3D:
-      tgsi_sampler->get_samples = sp_get_samples_3d;
+   case PIPE_FUNC_GREATER:
+      k0 = p[0] > rgba[0][0];
+      k1 = p[1] > rgba[0][1];
+      k2 = p[2] > rgba[0][2];
+      k3 = p[3] > rgba[0][3];
       break;
-   case PIPE_TEXTURE_CUBE:
-      tgsi_sampler->get_samples = sp_get_samples_cube;
+   case PIPE_FUNC_GEQUAL:
+      k0 = p[0] >= rgba[0][0];
+      k1 = p[1] >= rgba[0][1];
+      k2 = p[2] >= rgba[0][2];
+      k3 = p[3] >= rgba[0][3];
+      break;
+   case PIPE_FUNC_EQUAL:
+      k0 = p[0] == rgba[0][0];
+      k1 = p[1] == rgba[0][1];
+      k2 = p[2] == rgba[0][2];
+      k3 = p[3] == rgba[0][3];
+      break;
+   case PIPE_FUNC_NOTEQUAL:
+      k0 = p[0] != rgba[0][0];
+      k1 = p[1] != rgba[0][1];
+      k2 = p[2] != rgba[0][2];
+      k3 = p[3] != rgba[0][3];
+      break;
+   case PIPE_FUNC_ALWAYS:
+      k0 = k1 = k2 = k3 = 1;
+      break;
+   case PIPE_FUNC_NEVER:
+      k0 = k1 = k2 = k3 = 0;
       break;
    default:
+      k0 = k1 = k2 = k3 = 0;
       assert(0);
       break;
    }
 
-   /* Do this elsewhere: 
-    */
-   samp->xpot = util_unsigned_logbase2( samp->texture->width[0] );
-   samp->ypot = util_unsigned_logbase2( samp->texture->height[0] );
+   /* convert four pass/fail values to an intensity in [0,1] */
+   val = 0.25F * (k0 + k1 + k2 + k3);
 
-   /* Try to hook in a faster sampler.  Ultimately we'll have to
-    * code-generate these.  Luckily most of this looks like it is
-    * orthogonal state within the sampler.
+   /* XXX returning result for default GL_DEPTH_TEXTURE_MODE = GL_LUMINANCE */
+   for (j = 0; j < 4; j++) {
+      rgba[0][j] = rgba[1][j] = rgba[2][j] = val;
+      rgba[3][j] = 1.0F;
+   }
+}
+
+/* Calculate cube faces.
+ */
+static void
+sample_cube(struct tgsi_sampler *tgsi_sampler,
+            const float s[QUAD_SIZE],
+            const float t[QUAD_SIZE],
+            const float p[QUAD_SIZE],
+            float lodbias,
+            float rgba[NUM_CHANNELS][QUAD_SIZE])
+{
+   struct sp_sampler_varient *samp = sp_sampler_varient(tgsi_sampler);
+   unsigned j;
+   float ssss[4], tttt[4];
+
+   /*
+     major axis
+     direction     target                             sc     tc    ma
+     ----------    -------------------------------    ---    ---   ---
+     +rx          TEXTURE_CUBE_MAP_POSITIVE_X_EXT    -rz    -ry   rx
+     -rx          TEXTURE_CUBE_MAP_NEGATIVE_X_EXT    +rz    -ry   rx
+     +ry          TEXTURE_CUBE_MAP_POSITIVE_Y_EXT    +rx    +rz   ry
+     -ry          TEXTURE_CUBE_MAP_NEGATIVE_Y_EXT    +rx    -rz   ry
+     +rz          TEXTURE_CUBE_MAP_POSITIVE_Z_EXT    +rx    -ry   rz
+     -rz          TEXTURE_CUBE_MAP_NEGATIVE_Z_EXT    -rx    -ry   rz
+   */
+   for (j = 0; j < QUAD_SIZE; j++) {
+      float rx = s[j];
+      float ry = t[j];
+      float rz = p[j];
+      const float arx = fabsf(rx), ary = fabsf(ry), arz = fabsf(rz);
+      unsigned face;
+      float sc, tc, ma;
+
+      if (arx > ary && arx > arz) {
+         if (rx >= 0.0F) {
+            face = PIPE_TEX_FACE_POS_X;
+            sc = -rz;
+            tc = -ry;
+            ma = arx;
+         }
+         else {
+            face = PIPE_TEX_FACE_NEG_X;
+            sc = rz;
+            tc = -ry;
+            ma = arx;
+         }
+      }
+      else if (ary > arx && ary > arz) {
+         if (ry >= 0.0F) {
+            face = PIPE_TEX_FACE_POS_Y;
+            sc = rx;
+            tc = rz;
+            ma = ary;
+         }
+         else {
+            face = PIPE_TEX_FACE_NEG_Y;
+            sc = rx;
+            tc = -rz;
+            ma = ary;
+         }
+      }
+      else {
+         if (rz > 0.0F) {
+            face = PIPE_TEX_FACE_POS_Z;
+            sc = rx;
+            tc = -ry;
+            ma = arz;
+         }
+         else {
+            face = PIPE_TEX_FACE_NEG_Z;
+            sc = -rx;
+            tc = -ry;
+            ma = arz;
+         }
+      }
+
+      ssss[j] = ( sc / ma + 1.0F ) * 0.5F;
+      tttt[j] = ( tc / ma + 1.0F ) * 0.5F;
+      samp->faces[j] = face;
+   }
+
+   /* In our little pipeline, the compare stage is next.  If compare
+    * is not active, this will point somewhere deeper into the
+    * pipeline, eg. to mip_filter or even img_filter.
     */
-   if (texture->target == PIPE_TEXTURE_2D &&
-       sampler->min_img_filter == sampler->mag_img_filter &&
-       sampler->wrap_s == sampler->wrap_t &&
-       sampler->compare_mode == FALSE &&
-       sampler->normalized_coords) 
-   {
-      if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_NONE) {
-         samp->level = CLAMP((int) sampler->min_lod,
-                             0, (int) texture->last_level);
-
-         if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) {
-            switch (sampler->min_img_filter) {
+   samp->compare(tgsi_sampler, ssss, tttt, NULL, lodbias, rgba);
+}
+
+
+
+
+static wrap_nearest_func get_nearest_unorm_wrap( unsigned mode )
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_nearest_unorm_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_nearest_unorm_clamp_to_border;
+   default:
+      assert(0);
+      return wrap_nearest_unorm_clamp;
+   }
+}
+
+
+static wrap_nearest_func get_nearest_wrap( unsigned mode )
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      return wrap_nearest_repeat;
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_nearest_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return wrap_nearest_clamp_to_edge;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_nearest_clamp_to_border;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      return wrap_nearest_mirror_repeat;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      return wrap_nearest_mirror_clamp;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      return wrap_nearest_mirror_clamp_to_edge;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      return wrap_nearest_mirror_clamp_to_border;
+   default:
+      assert(0);
+      return wrap_nearest_repeat;
+   }
+}
+
+static wrap_linear_func get_linear_unorm_wrap( unsigned mode )
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_linear_unorm_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_linear_unorm_clamp_to_border;
+   default:
+      assert(0);
+      return wrap_linear_unorm_clamp;
+   }
+}
+
+static wrap_linear_func get_linear_wrap( unsigned mode )
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+      return wrap_linear_repeat;
+   case PIPE_TEX_WRAP_CLAMP:
+      return wrap_linear_clamp;
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return wrap_linear_clamp_to_edge;
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+      return wrap_linear_clamp_to_border;
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+      return wrap_linear_mirror_repeat;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      return wrap_linear_mirror_clamp;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      return wrap_linear_mirror_clamp_to_edge;
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      return wrap_linear_mirror_clamp_to_border;
+   default:
+      assert(0);
+      return wrap_linear_repeat;
+   }
+}
+
+static compute_lambda_func get_lambda_func( const union sp_sampler_key key )
+{
+   if (key.bits.processor == TGSI_PROCESSOR_VERTEX)
+      return compute_lambda_vert;
+   
+   switch (key.bits.target) {
+   case PIPE_TEXTURE_1D:
+      return compute_lambda_1d;
+   case PIPE_TEXTURE_2D:
+      return compute_lambda_2d;
+   case PIPE_TEXTURE_3D:
+      return compute_lambda_3d;
+   default:
+      assert(0);
+      return compute_lambda_1d;
+   }
+}
+
+static filter_func get_img_filter( const union sp_sampler_key key,
+                                   unsigned filter,
+                                   const struct pipe_sampler_state *sampler )
+{
+   switch (key.bits.target) {
+   case PIPE_TEXTURE_1D:
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_1d_nearest;
+      else
+         return img_filter_1d_linear;
+      break;
+   case PIPE_TEXTURE_2D:
+      /* Try for fast path:
+       */
+      if (key.bits.is_pot &&
+          sampler->wrap_s == sampler->wrap_t &&
+          sampler->normalized_coords) 
+      {
+         switch (sampler->wrap_s) {
+         case PIPE_TEX_WRAP_REPEAT:
+            switch (filter) {
             case PIPE_TEX_FILTER_NEAREST:
-               tgsi_sampler->get_samples = sp_get_samples_2d_nearest_repeat_POT;
-               break;
+               return img_filter_2d_nearest_repeat_POT;
             case PIPE_TEX_FILTER_LINEAR:
-               tgsi_sampler->get_samples = sp_get_samples_2d_linear_repeat_POT;
-               break;
+               return img_filter_2d_linear_repeat_POT;
             default:
                break;
             }
-         } 
-         else if (sampler->wrap_s == PIPE_TEX_WRAP_CLAMP) {
-            switch (sampler->min_img_filter) {
+            break;
+         case PIPE_TEX_WRAP_CLAMP:
+            switch (filter) {
             case PIPE_TEX_FILTER_NEAREST:
-               tgsi_sampler->get_samples = sp_get_samples_2d_nearest_clamp_POT;
-               break;
+               return img_filter_2d_nearest_clamp_POT;
             default:
                break;
             }
          }
       }
-      else if (sampler->min_mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
-         if (sampler->wrap_s == PIPE_TEX_WRAP_REPEAT) {
-            switch (sampler->min_img_filter) {
-            case PIPE_TEX_FILTER_LINEAR:
-               tgsi_sampler->get_samples = sp_get_samples_2d_linear_mip_linear_repeat_POT;
-               break;
-            default:
-               break;
-            }
-         } 
+      /* Fallthrough to default versions:
+       */
+   case PIPE_TEXTURE_CUBE:
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_2d_nearest;
+      else
+         return img_filter_2d_linear;
+      break;
+   case PIPE_TEXTURE_3D:
+      if (filter == PIPE_TEX_FILTER_NEAREST) 
+         return img_filter_3d_nearest;
+      else
+         return img_filter_3d_linear;
+      break;
+   default:
+      assert(0);
+      return img_filter_1d_nearest;
+   }
+}
+
+
+void
+sp_sampler_varient_bind_texture( struct sp_sampler_varient *samp,
+                                 struct softpipe_tile_cache *tex_cache,
+                                 const struct pipe_texture *texture )
+{
+   const struct pipe_sampler_state *sampler = samp->sampler;
+
+   samp->texture = texture;
+   samp->cache = tex_cache;
+   samp->xpot = util_unsigned_logbase2( texture->width[0] );
+   samp->ypot = util_unsigned_logbase2( texture->height[0] );
+   samp->level = CLAMP((int) sampler->min_lod, 0, (int) texture->last_level);
+}
+
+/* Create a sampler varient for a given set of non-orthogonal state.  Currently the 
+ */
+struct sp_sampler_varient *
+sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
+                           const union sp_sampler_key key )
+{
+   struct sp_sampler_varient *samp = CALLOC_STRUCT(sp_sampler_varient);
+   if (!samp)
+      return NULL;
+
+   samp->sampler = sampler;
+   samp->key = key;
+
+   /* Note that (for instance) linear_texcoord_s and
+    * nearest_texcoord_s may be active at the same time, if the
+    * sampler min_img_filter differs from its mag_img_filter.
+    */
+   if (sampler->normalized_coords) {
+      samp->linear_texcoord_s = get_linear_wrap( sampler->wrap_s );
+      samp->linear_texcoord_t = get_linear_wrap( sampler->wrap_t );
+      samp->linear_texcoord_p = get_linear_wrap( sampler->wrap_r );
+      
+      samp->nearest_texcoord_s = get_nearest_wrap( sampler->wrap_s );
+      samp->nearest_texcoord_t = get_nearest_wrap( sampler->wrap_t );
+      samp->nearest_texcoord_p = get_nearest_wrap( sampler->wrap_r );
+   }
+   else {
+      samp->linear_texcoord_s = get_linear_unorm_wrap( sampler->wrap_s );
+      samp->linear_texcoord_t = get_linear_unorm_wrap( sampler->wrap_t );
+      samp->linear_texcoord_p = get_linear_unorm_wrap( sampler->wrap_r );
+      
+      samp->nearest_texcoord_s = get_nearest_unorm_wrap( sampler->wrap_s );
+      samp->nearest_texcoord_t = get_nearest_unorm_wrap( sampler->wrap_t );
+      samp->nearest_texcoord_p = get_nearest_unorm_wrap( sampler->wrap_r );
+   }
+   
+   samp->compute_lambda = get_lambda_func( key );
+
+   samp->min_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
+   samp->mag_img_filter = get_img_filter(key, sampler->min_img_filter, sampler);
+
+   switch (sampler->min_mip_filter) {
+   case PIPE_TEX_MIPFILTER_NONE:
+      if (sampler->min_img_filter == sampler->mag_img_filter) 
+         samp->mip_filter = samp->min_img_filter;         
+      else
+         samp->mip_filter = mip_filter_none;
+      break;
+
+   case PIPE_TEX_MIPFILTER_NEAREST:
+      samp->mip_filter = mip_filter_nearest;
+      break;
+
+   case PIPE_TEX_MIPFILTER_LINEAR:
+      if (key.bits.is_pot &&
+          sampler->min_img_filter == sampler->mag_img_filter &&
+          sampler->wrap_s == sampler->wrap_t &&
+          sampler->normalized_coords &&
+          sampler->wrap_s == sampler->wrap_t &&
+          sampler->wrap_s == PIPE_TEX_WRAP_REPEAT &&
+          sampler->min_img_filter == PIPE_TEX_FILTER_LINEAR)
+      {
+         samp->mip_filter = mip_filter_linear_2d_linear_repeat_POT;
+      }
+      else 
+      {
+         samp->mip_filter = mip_filter_linear;
       }
+      break;
+   }
+
+   if (sampler->compare_mode != FALSE) {
+      samp->compare = sample_compare;
+   }
+   else {
+      /* Skip compare operation by promoting the mip_filter function
+       * pointer:
+       */
+      samp->compare = samp->mip_filter;
    }
-   else if (0) {
-      _debug_printf("target %d/%d min_mip %d/%d min_img %d/%d wrap %d/%d compare %d/%d norm %d/%d\n",
-                    texture->target, PIPE_TEXTURE_2D,
-                    sampler->min_mip_filter, PIPE_TEX_MIPFILTER_NONE,
-                    sampler->min_img_filter, sampler->mag_img_filter,
-                    sampler->wrap_s, sampler->wrap_t,
-                    sampler->compare_mode, FALSE,
-                    sampler->normalized_coords, TRUE);
+   
+   if (key.bits.target == PIPE_TEXTURE_CUBE) {
+      samp->base.get_samples = sample_cube;
    }
+   else {
+      samp->faces[0] = 0;
+      samp->faces[1] = 0;
+      samp->faces[2] = 0;
+      samp->faces[3] = 0;
 
-out:
-   tgsi_sampler->get_samples( tgsi_sampler, s, t, p, lodbias, rgba );
+      /* Skip cube face determination by promoting the compare
+       * function pointer:
+       */
+      samp->base.get_samples = samp->compare;
+   }
+
+   return samp;
 }
 
+
+
+
+
+
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.h b/src/gallium/drivers/softpipe/sp_tex_sample.h
index c73ae441314..26f80eb88a8 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.h
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.h
@@ -31,14 +31,61 @@
 
 #include "tgsi/tgsi_exec.h"
 
+struct sp_sampler_varient;
+
+typedef void (*wrap_nearest_func)(const float s[4],
+                                  unsigned size,
+                                  int icoord[4]);
+
+typedef void (*wrap_linear_func)(const float s[4], 
+                                 unsigned size,
+                                 int icoord0[4],
+                                 int icoord1[4],
+                                 float w[4]);
+
+typedef float (*compute_lambda_func)(const struct sp_sampler_varient *sampler,
+                                     const float s[QUAD_SIZE],
+                                     const float t[QUAD_SIZE],
+                                     const float p[QUAD_SIZE],
+                                     float lodbias);
+
+typedef void (*filter_func)(struct tgsi_sampler *tgsi_sampler,
+                            const float s[QUAD_SIZE],
+                            const float t[QUAD_SIZE],
+                            const float p[QUAD_SIZE],
+                            float lodbias,
+                            float rgba[NUM_CHANNELS][QUAD_SIZE]);
+
+
+union sp_sampler_key {
+   struct {
+      unsigned target:3;
+      unsigned is_pot:1;
+      unsigned processor:2;
+      unsigned pad:26;
+   } bits;
+   unsigned value;
+};
 
 /**
  * Subclass of tgsi_sampler
  */
-struct sp_shader_sampler
+struct sp_sampler_varient
 {
    struct tgsi_sampler base;  /**< base class */
 
+   union sp_sampler_key key;
+
+   /* The owner of this struct:
+    */
+   const struct pipe_sampler_state *sampler;
+
+
+   /* Currently bound texture:
+    */
+   const struct pipe_texture *texture;
+   struct softpipe_tile_cache *cache;
+
    unsigned processor;
 
    /* For sp_get_samples_2d_linear_POT:
@@ -47,22 +94,51 @@ struct sp_shader_sampler
    unsigned ypot;
    unsigned level;
 
-   const struct pipe_texture *texture;
-   const struct pipe_sampler_state *sampler;
+   unsigned faces[4];
+   
+   wrap_nearest_func nearest_texcoord_s;
+   wrap_nearest_func nearest_texcoord_t;
+   wrap_nearest_func nearest_texcoord_p;
 
-   struct softpipe_tile_cache *cache;
+   wrap_linear_func linear_texcoord_s;
+   wrap_linear_func linear_texcoord_t;
+   wrap_linear_func linear_texcoord_p;
+
+   filter_func min_img_filter;
+   filter_func mag_img_filter;
+
+   compute_lambda_func compute_lambda;
+
+   filter_func mip_filter;
+   filter_func compare;
+   
+   /* Linked list:
+    */
+   struct sp_sampler_varient *next;
 };
 
+struct sp_sampler;
+
+/* Create a sampler varient for a given set of non-orthogonal state.  Currently the 
+ */
+struct sp_sampler_varient *
+sp_create_sampler_varient( const struct pipe_sampler_state *sampler,
+                           const union sp_sampler_key key );
 
+void sp_sampler_varient_bind_texture( struct sp_sampler_varient *varient,
+                                      struct softpipe_tile_cache *tex_cache,
+                                      const struct pipe_texture *tex );
 
-static INLINE struct sp_shader_sampler *
-sp_shader_sampler(const struct tgsi_sampler *sampler)
-{
-   return (struct sp_shader_sampler *) sampler;
-}
+void sp_sampler_varient_destroy( struct sp_sampler_varient * );
 
 
 
+static INLINE struct sp_sampler_varient *
+sp_sampler_varient(const struct tgsi_sampler *sampler)
+{
+   return (struct sp_sampler_varient *) sampler;
+}
+
 extern void
 sp_get_samples(struct tgsi_sampler *tgsi_sampler,
                const float s[QUAD_SIZE],
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 0c84375bf10..a3a54dada4a 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -120,16 +120,20 @@ softpipe_displaytarget_layout(struct pipe_screen *screen,
 
 static struct pipe_texture *
 softpipe_texture_create(struct pipe_screen *screen,
-                        const struct pipe_texture *templat)
+                        const struct pipe_texture *template)
 {
    struct softpipe_texture *spt = CALLOC_STRUCT(softpipe_texture);
    if (!spt)
       return NULL;
 
-   spt->base = *templat;
+   spt->base = *template;
    pipe_reference_init(&spt->base.reference, 1);
    spt->base.screen = screen;
 
+   spt->pot = (util_is_power_of_two(template->width[0]) &&
+               util_is_power_of_two(template->height[0]) &&
+               util_is_power_of_two(template->depth[0]));
+
    if (spt->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) {
       if (!softpipe_displaytarget_layout(screen, spt))
          goto fail;
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index 42df722a2d1..4dd0c1239e2 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -48,6 +48,10 @@ struct softpipe_texture
     */
    struct pipe_buffer *buffer;
 
+   /* True if texture images are power-of-two in all dimensions:
+    */
+   boolean pot;
+
    unsigned timestamp;
 };