Merge branch 'lp-offset-twoside'

6 files changed, 306 insertions, 69 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index db09c95b27..a35e09e8b4 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -104,9 +104,18 @@ struct llvmpipe_context {
    /** Vertex format */
    struct vertex_info vertex_info;
    
+   /** Which vertex shader output slot contains color */
+   int color_slot[2];
+
+   /** Which vertex shader output slot contains bcolor */
+   int bcolor_slot[2];
+
    /** Which vertex shader output slot contains point size */
    int psize_slot;
 
+   /**< minimum resolvable depth value, for polygon offset */   
+   double mrd;
+   
    /** The tiling engine */
    struct lp_setup_context *setup;
    struct lp_setup_variant setup_variant;
diff --git a/src/gallium/drivers/llvmpipe/lp_state_derived.c b/src/gallium/drivers/llvmpipe/lp_state_derived.c
index 0f5f7369e0..8725ea39fe 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_derived.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_derived.c
@@ -53,6 +53,11 @@ compute_vertex_info(struct llvmpipe_context *llvmpipe)
    unsigned vs_index;
    uint i;
 
+   llvmpipe->color_slot[0] = ~0;
+   llvmpipe->color_slot[1] = ~0;
+   llvmpipe->bcolor_slot[0] = ~0;
+   llvmpipe->bcolor_slot[1] = ~0;
+
    /*
     * Match FS inputs against VS outputs, emitting the necessary
     * attributes.  Could cache these structs and look them up with a
@@ -76,12 +81,31 @@ compute_vertex_info(struct llvmpipe_context *llvmpipe)
                                          lpfs->info.base.input_semantic_name[i],
                                          lpfs->info.base.input_semantic_index[i]);
 
+      if (lpfs->info.base.input_semantic_name[i] == TGSI_SEMANTIC_COLOR &&
+          lpfs->info.base.input_semantic_index[i] < 2) {
+         int idx = lpfs->info.base.input_semantic_index[i];
+         llvmpipe->color_slot[idx] = vinfo->num_attribs;
+      }
+
       /*
        * Emit the requested fs attribute for all but position.
        */
       draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, vs_index);
    }
 
+   /* Figure out if we need bcolor as well.
+    */
+   for (i = 0; i < 2; i++) {
+      vs_index = draw_find_shader_output(llvmpipe->draw,
+                                         TGSI_SEMANTIC_BCOLOR, i);
+
+      if (vs_index > 0) {
+         llvmpipe->bcolor_slot[i] = vinfo->num_attribs;
+         draw_emit_vertex_attr(vinfo, EMIT_4F, INTERP_PERSPECTIVE, vs_index);
+      }
+   }
+
+
    /* Figure out if we need pointsize as well.
     */
    vs_index = draw_find_shader_output(llvmpipe->draw,
diff --git a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
index dbd73812e4..574f9e940e 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_rasterizer.c
@@ -32,16 +32,64 @@
 #include "lp_setup.h"
 #include "draw/draw_context.h"
 
+struct lp_rast_state {
+   struct pipe_rasterizer_state lp_state;
+   struct pipe_rasterizer_state draw_state;
+};
+
+/* State which might be handled in either the draw module or locally.
+ * This function is used to turn that state off in one of the two
+ * places.
+ */
+static void
+clear_flags(struct pipe_rasterizer_state *rast)
+{
+   rast->light_twoside = 0;
+   rast->offset_tri = 0;
+}
+
 
 
 static void *
 llvmpipe_create_rasterizer_state(struct pipe_context *pipe,
                                  const struct pipe_rasterizer_state *rast)
 {
-   /* We do nothing special with rasterizer state.
-    * The CSO handle is just a pointer to a pipe_rasterizer_state object.
+   boolean need_pipeline;
+
+   /* Partition rasterizer state into what we want the draw module to
+    * handle, and what we'll look after ourselves.
+    */
+   struct lp_rast_state *state = MALLOC_STRUCT(lp_rast_state);
+   if (state == NULL)
+      return NULL;
+
+   memcpy(&state->draw_state, rast, sizeof *rast);
+   memcpy(&state->lp_state, rast, sizeof *rast);
+
+   /* We rely on draw module to do unfilled polyons, AA lines and
+    * points and stipple.
+    * 
+    * Over time, reduce this list of conditions, and expand the list
+    * of flags which get cleared in clear_flags().
     */
-   return mem_dup(rast, sizeof(*rast));
+   need_pipeline = (rast->fill_front != PIPE_POLYGON_MODE_FILL ||
+		    rast->fill_back != PIPE_POLYGON_MODE_FILL ||
+		    rast->point_smooth ||
+		    rast->line_smooth ||
+		    rast->line_stipple_enable ||
+		    rast->poly_stipple_enable);
+
+   /* If not using the pipeline, clear out the flags which we can
+    * handle ourselves.  If we *are* using the pipeline, do everything
+    * on the pipeline and clear those flags on our internal copy of
+    * the state.
+    */
+   if (need_pipeline)
+      clear_flags(&state->lp_state);
+   else
+      clear_flags(&state->draw_state);
+
+   return state;
 }
 
 
@@ -50,36 +98,33 @@ static void
 llvmpipe_bind_rasterizer_state(struct pipe_context *pipe, void *handle)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
-   const struct pipe_rasterizer_state *rasterizer =
-      (const struct pipe_rasterizer_state *) handle;
-
-   if (llvmpipe->rasterizer == rasterizer)
-      return;
+   const struct lp_rast_state *state =
+      (const struct lp_rast_state *) handle;
 
-   /* pass-through to draw module */
-   draw_set_rasterizer_state(llvmpipe->draw, rasterizer, handle);
+   if (state) {
+      llvmpipe->rasterizer = &state->lp_state;
+      draw_set_rasterizer_state(llvmpipe->draw, &state->draw_state, handle);
 
-   llvmpipe->rasterizer = rasterizer;
-
-   /* Note: we can immediately set the triangle state here and
-    * not worry about binning because we handle culling during
-    * triangle setup, not when rasterizing the bins.
-    */
-   if (llvmpipe->rasterizer) {
+      /* XXX: just pass lp_state directly to setup.
+       */
       lp_setup_set_triangle_state( llvmpipe->setup,
-                   llvmpipe->rasterizer->cull_face,
-                   llvmpipe->rasterizer->front_ccw,
-                   llvmpipe->rasterizer->scissor,
-                   llvmpipe->rasterizer->gl_rasterization_rules);
+				   state->lp_state.cull_face,
+				   state->lp_state.front_ccw,
+				   state->lp_state.scissor,
+				   state->lp_state.gl_rasterization_rules);
       lp_setup_set_flatshade_first( llvmpipe->setup,
-                   llvmpipe->rasterizer->flatshade_first);
+				    state->lp_state.flatshade_first);
       lp_setup_set_line_state( llvmpipe->setup,
-                   llvmpipe->rasterizer->line_width);
+			       state->lp_state.line_width);
       lp_setup_set_point_state( llvmpipe->setup,
-                   llvmpipe->rasterizer->point_size,
-                   llvmpipe->rasterizer->point_size_per_vertex,
-                   llvmpipe->rasterizer->sprite_coord_enable,
-                   llvmpipe->rasterizer->sprite_coord_mode);
+				state->lp_state.point_size,
+				state->lp_state.point_size_per_vertex,
+				state->lp_state.sprite_coord_enable,
+				state->lp_state.sprite_coord_mode);
+   }
+   else {
+      llvmpipe->rasterizer = NULL;
+      draw_set_rasterizer_state(llvmpipe->draw, NULL, handle);      
    }
 
    llvmpipe->dirty |= LP_NEW_RASTERIZER;
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.c b/src/gallium/drivers/llvmpipe/lp_state_setup.c
index 2c8b8b9a92..194b014436 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.c
@@ -30,9 +30,12 @@
 #include "util/u_memory.h"
 #include "util/u_simple_list.h"
 #include "os/os_time.h"
+#include "gallivm/lp_bld_arit.h"
 #include "gallivm/lp_bld_debug.h"
 #include "gallivm/lp_bld_init.h"
 #include "gallivm/lp_bld_intr.h"
+#include "gallivm/lp_bld_flow.h"
+#include "gallivm/lp_bld_type.h"
 #include <llvm-c/Analysis.h>	/* for LLVMVerifyFunction */
 
 #include "lp_perf.h"
@@ -74,6 +77,12 @@ struct lp_setup_args
    LLVMValueRef dy01_ooa;
    LLVMValueRef dx20_ooa;
    LLVMValueRef dx01_ooa;
+
+   /* Temporary, per-attribute:
+    */
+   LLVMValueRef v0a;
+   LLVMValueRef v1a;
+   LLVMValueRef v2a;
 };
 
 static LLVMTypeRef type4f(void)
@@ -149,16 +158,11 @@ static void
 emit_constant_coef4( LLVMBuilderRef builder,
 		     struct lp_setup_args *args,
 		     unsigned slot,
-		     LLVMValueRef vert,
-		     unsigned attr)
+		     LLVMValueRef vert)
 {
    LLVMValueRef zero      = LLVMConstReal(LLVMFloatType(), 0.0);
    LLVMValueRef zerovec   = vec4f_from_scalar(builder, zero, "zero");
-   LLVMValueRef idx       = LLVMConstInt(LLVMInt32Type(), attr, 0);
-   LLVMValueRef attr_ptr  = LLVMBuildGEP(builder, vert, &idx, 1, "attr_ptr");
-   LLVMValueRef vert_attr = LLVMBuildLoad(builder, attr_ptr, "vert_attr");
-
-   store_coef(builder, args, slot, vert_attr, zerovec, zerovec);
+   store_coef(builder, args, slot, vert, zerovec, zerovec);
 }
 
 
@@ -195,7 +199,150 @@ vert_attrib(LLVMBuilderRef b,
    return LLVMBuildLoad(b, LLVMBuildGEP(b, vert, idx, 2, ""), name);
 }
 
+static LLVMValueRef
+vert_clamp(LLVMBuilderRef b,
+           LLVMValueRef x,
+           LLVMValueRef min,
+           LLVMValueRef max)
+{
+   LLVMValueRef min_result = LLVMBuildFCmp(b, LLVMRealUGT, min, x, "");
+   LLVMValueRef max_result = LLVMBuildFCmp(b, LLVMRealUGT, x, max, "");
+   LLVMValueRef clamp_value;
+
+   clamp_value = LLVMBuildSelect(b, min_result, min, x, "");
+   clamp_value = LLVMBuildSelect(b, max_result, max, x, "");
+
+   return clamp_value;
+}
+
+static void
+lp_twoside(LLVMBuilderRef b, 
+           struct lp_setup_args *args,
+           const struct lp_setup_variant_key *key,
+           int bcolor_slot)
+{
+   LLVMValueRef a0_back, a1_back, a2_back;
+   LLVMValueRef idx2 = LLVMConstInt(LLVMInt32Type(), bcolor_slot, 0);
+
+   LLVMValueRef facing = args->facing;
+   LLVMValueRef front_facing = LLVMBuildICmp(b, LLVMIntEQ, facing, LLVMConstInt(LLVMInt32Type(), 0, 0), ""); /** need i1 for if condition */
+   
+   a0_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx2, 1, ""), "v0a_back");
+   a1_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx2, 1, ""), "v1a_back");
+   a2_back = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx2, 1, ""), "v2a_back");
+
+   /* Possibly swap the front and back attrib values,
+    *
+    * Prefer select to if so we don't have to worry about phis or
+    * allocas.
+    */
+   args->v0a = LLVMBuildSelect(b, front_facing, a0_back, args->v0a, "");
+   args->v1a = LLVMBuildSelect(b, front_facing, a1_back, args->v1a, "");
+   args->v2a = LLVMBuildSelect(b, front_facing, a2_back, args->v2a, "");
+
+}
 
+static void
+lp_do_offset_tri(LLVMBuilderRef b, 
+                 struct lp_setup_args *args,
+                 const struct lp_setup_variant_key *key)
+{
+   struct lp_build_context bld;
+   LLVMValueRef zoffset, mult;
+   LLVMValueRef z0_new, z1_new, z2_new;
+   LLVMValueRef dzdx0, dzdx, dzdy0, dzdy;
+   LLVMValueRef max, max_value;
+   
+   LLVMValueRef one  = LLVMConstReal(LLVMFloatType(), 1.0);
+   LLVMValueRef zero  = LLVMConstReal(LLVMFloatType(), 0.0);
+
+   /* edge vectors: e = v0 - v2, f = v1 - v2 */
+   LLVMValueRef v0_x = vert_attrib(b, args->v0, 0, 0, "v0_x");
+   LLVMValueRef v1_x = vert_attrib(b, args->v1, 0, 0, "v1_x");
+   LLVMValueRef v2_x = vert_attrib(b, args->v2, 0, 0, "v2_x");
+   LLVMValueRef v0_y = vert_attrib(b, args->v0, 0, 1, "v0_y");
+   LLVMValueRef v1_y = vert_attrib(b, args->v1, 0, 1, "v1_y");
+   LLVMValueRef v2_y = vert_attrib(b, args->v2, 0, 1, "v2_y");
+   LLVMValueRef v0_z = vert_attrib(b, args->v0, 0, 2, "v0_z");
+   LLVMValueRef v1_z = vert_attrib(b, args->v1, 0, 2, "v1_z");
+   LLVMValueRef v2_z = vert_attrib(b, args->v2, 0, 2, "v2_z");
+ 
+   /* edge vectors: e = v0 - v2, f = v1 - v2 */
+   LLVMValueRef dx02 = LLVMBuildFSub(b, v0_x, v2_x, "dx02");
+   LLVMValueRef dy02 = LLVMBuildFSub(b, v0_y, v2_y, "dy02");
+   LLVMValueRef dz02 = LLVMBuildFSub(b, v0_z, v2_z, "dz02");
+   LLVMValueRef dx12 = LLVMBuildFSub(b, v1_x, v2_x, "dx12"); 
+   LLVMValueRef dy12 = LLVMBuildFSub(b, v1_y, v2_y, "dy12");
+   LLVMValueRef dz12 = LLVMBuildFSub(b, v1_z, v2_z, "dz12");
+ 
+   /* det = cross(e,f).z */
+   LLVMValueRef dx02_dy12  = LLVMBuildFMul(b, dx02, dy12, "dx02_dy12");
+   LLVMValueRef dy02_dx12  = LLVMBuildFMul(b, dy02, dx12, "dy02_dx12");
+   LLVMValueRef det  = LLVMBuildFSub(b, dx02_dy12, dy02_dx12, "det");
+   LLVMValueRef inv_det = LLVMBuildFDiv(b, one, det, "inv_det"); 
+   
+   /* (res1,res2) = cross(e,f).xy */
+   LLVMValueRef dy02_dz12    = LLVMBuildFMul(b, dy02, dz12, "dy02_dz12");
+   LLVMValueRef dz02_dy12    = LLVMBuildFMul(b, dz02, dy12, "dz02_dy12");
+   LLVMValueRef dz02_dx12    = LLVMBuildFMul(b, dz02, dx12, "dz02_dx12");
+   LLVMValueRef dx02_dz12    = LLVMBuildFMul(b, dx02, dz12, "dx02_dz12");
+   LLVMValueRef res1  = LLVMBuildFSub(b, dy02_dz12, dz02_dy12, "res1");
+   LLVMValueRef res2  = LLVMBuildFSub(b, dz02_dx12, dx02_dz12, "res2");
+
+   /* dzdx = fabsf(res1 * inv_det), dydx = fabsf(res2 * inv_det)*/
+   lp_build_context_init(&bld, b, lp_type_float(32));
+   dzdx0 = LLVMBuildFMul(b, res1, inv_det, "dzdx");
+   dzdx  = lp_build_abs(&bld, dzdx0);
+   dzdy0 = LLVMBuildFMul(b, res2, inv_det, "dzdy");
+   dzdy  = lp_build_abs(&bld, dzdy0);
+
+   /* zoffset = offset->units + MAX2(dzdx, dzdy) * offset->scale */
+   max = LLVMBuildFCmp(b, LLVMRealUGT, dzdx, dzdy, "");
+   max_value = LLVMBuildSelect(b, max, dzdx, dzdy, "max"); 
+
+   mult = LLVMBuildFMul(b, max_value, LLVMConstReal(LLVMFloatType(), key->scale), "");
+   zoffset = LLVMBuildFAdd(b, LLVMConstReal(LLVMFloatType(), key->units), mult, "zoffset");
+
+   /* clamp and do offset */
+   z0_new = vert_clamp(b, LLVMBuildFAdd(b, v0_z, zoffset, ""), zero, one);
+   z1_new = vert_clamp(b, LLVMBuildFAdd(b, v1_z, zoffset, ""), zero, one);
+   z2_new = vert_clamp(b, LLVMBuildFAdd(b, v2_z, zoffset, ""), zero, one);
+
+   /* insert into args->a0.z, a1.z, a2.z:
+    */   
+   args->v0a = LLVMBuildInsertElement(b, args->v0a, z0_new, LLVMConstInt(LLVMInt32Type(), 2, 0), "");
+   args->v1a = LLVMBuildInsertElement(b, args->v1a, z1_new, LLVMConstInt(LLVMInt32Type(), 2, 0), "");
+   args->v2a = LLVMBuildInsertElement(b, args->v2a, z2_new, LLVMConstInt(LLVMInt32Type(), 2, 0), "");
+}
+
+static void
+load_attribute(LLVMBuilderRef b, 
+               struct lp_setup_args *args,
+               const struct lp_setup_variant_key *key,
+               unsigned vert_attr)
+{
+   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
+
+   /* Load the vertex data
+    */
+   args->v0a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
+   args->v1a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
+   args->v2a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
+
+
+   /* Potentially modify it according to twoside, offset, etc:
+    */
+   if (vert_attr == 0 && (key->scale != 0.0f || key->units != 0.0f)) {
+      lp_do_offset_tri(b, args, key);
+   }
+
+   if (key->twoside) {
+      if (vert_attr == key->color_slot && key->bcolor_slot != ~0)
+         lp_twoside(b, args, key, key->bcolor_slot);
+      else if (vert_attr == key->spec_slot && key->bspec_slot != ~0)
+         lp_twoside(b, args, key, key->bspec_slot);
+   }
+}
 
 static void 
 emit_coef4( LLVMBuilderRef b,
@@ -243,20 +390,17 @@ emit_coef4( LLVMBuilderRef b,
 static void 
 emit_linear_coef( LLVMBuilderRef b,
 		  struct lp_setup_args *args,
-		  unsigned slot,
-		  unsigned vert_attr)
+		  unsigned slot)
 {
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
-
-   LLVMValueRef a0 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
-   LLVMValueRef a1 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
-   LLVMValueRef a2 = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
-
-   emit_coef4(b, args, slot, a0, a1, a2);
+   /* nothing to do anymore */
+   emit_coef4(b,
+              args, slot, 
+              args->v0a,
+              args->v1a,
+              args->v2a);
 }
 
 
-
 /**
  * Compute a0, dadx and dady for a perspective-corrected interpolant,
  * for a triangle.
@@ -268,24 +412,17 @@ emit_linear_coef( LLVMBuilderRef b,
 static void 
 emit_perspective_coef( LLVMBuilderRef b,
 		       struct lp_setup_args *args,
-		       unsigned slot,
-		       unsigned vert_attr)
+		       unsigned slot)
 {
    /* premultiply by 1/w  (v[0][3] is always 1/w):
     */
-   LLVMValueRef idx = LLVMConstInt(LLVMInt32Type(), vert_attr, 0);
-
-   LLVMValueRef v0a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v0, &idx, 1, ""), "v0a");
-   LLVMValueRef v1a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v1, &idx, 1, ""), "v1a");
-   LLVMValueRef v2a = LLVMBuildLoad(b, LLVMBuildGEP(b, args->v2, &idx, 1, ""), "v2a");
-
    LLVMValueRef v0_oow = vec4f_from_scalar(b, vert_attrib(b, args->v0, 0, 3, ""), "v0_oow");
    LLVMValueRef v1_oow = vec4f_from_scalar(b, vert_attrib(b, args->v1, 0, 3, ""), "v1_oow");
    LLVMValueRef v2_oow = vec4f_from_scalar(b, vert_attrib(b, args->v2, 0, 3, ""), "v2_oow");
 
-   LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, v0a, v0_oow, "v0_oow_v0a");
-   LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, v1a, v1_oow, "v1_oow_v1a");
-   LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, v2a, v2_oow, "v2_oow_v2a");
+   LLVMValueRef v0_oow_v0a = LLVMBuildFMul(b, args->v0a, v0_oow, "v0_oow_v0a");
+   LLVMValueRef v1_oow_v1a = LLVMBuildFMul(b, args->v1a, v1_oow, "v1_oow_v1a");
+   LLVMValueRef v2_oow_v2a = LLVMBuildFMul(b, args->v2a, v2_oow, "v2_oow_v2a");
 
    emit_coef4(b, args, slot, v0_oow_v0a, v1_oow_v1a, v2_oow_v2a);
 }
@@ -294,9 +431,9 @@ emit_perspective_coef( LLVMBuilderRef b,
 static void
 emit_position_coef( LLVMBuilderRef builder,
 		    struct lp_setup_args *args,
-		    int slot, int attrib )
+		    int slot )
 {
-   emit_linear_coef(builder, args, slot, attrib);
+   emit_linear_coef(builder, args, slot);
 }
 
 
@@ -314,29 +451,34 @@ emit_tri_coef( LLVMBuilderRef builder,
 
    /* The internal position input is in slot zero:
     */
-   emit_position_coef(builder, args, 0, 0);
+   load_attribute(builder, args, key, 0);
+   emit_position_coef(builder, args, 0);
 
    /* setup interpolation for all the remaining attributes:
     */
    for (slot = 0; slot < key->num_inputs; slot++) {
-      unsigned vert_attr = key->inputs[slot].src_index;
+
+      if (key->inputs[slot].interp == LP_INTERP_CONSTANT ||
+          key->inputs[slot].interp == LP_INTERP_LINEAR ||
+          key->inputs[slot].interp == LP_INTERP_PERSPECTIVE)
+         load_attribute(builder, args, key, key->inputs[slot].src_index);
 
       switch (key->inputs[slot].interp) {
       case LP_INTERP_CONSTANT:
 	 if (key->flatshade_first) {
-	    emit_constant_coef4(builder, args, slot+1, args->v0, vert_attr);
+	    emit_constant_coef4(builder, args, slot+1, args->v0a);
 	 }
 	 else {
-	    emit_constant_coef4(builder, args, slot+1, args->v2, vert_attr);
+	    emit_constant_coef4(builder, args, slot+1, args->v2a);
 	 }
 	 break;
 
       case LP_INTERP_LINEAR:
-	 emit_linear_coef(builder, args, slot+1, vert_attr);
+	 emit_linear_coef(builder, args, slot+1);
          break;
 
       case LP_INTERP_PERSPECTIVE:
-	 emit_perspective_coef(builder, args, slot+1, vert_attr);
+	 emit_perspective_coef(builder, args, slot+1);
          break;
 
       case LP_INTERP_POSITION:
@@ -478,7 +620,8 @@ init_args(LLVMBuilderRef b,
  */
 static struct lp_setup_variant *
 generate_setup_variant(struct llvmpipe_screen *screen,
-		       struct lp_setup_variant_key *key)
+		       struct lp_setup_variant_key *key,
+                       struct llvmpipe_context *lp)
 {
    struct lp_setup_variant *variant = NULL;
    struct lp_setup_args args;
@@ -605,14 +748,20 @@ lp_make_setup_variant_key(struct llvmpipe_context *lp,
    key->num_inputs = fs->info.base.num_inputs;
    key->flatshade_first = lp->rasterizer->flatshade_first;
    key->pixel_center_half = lp->rasterizer->gl_rasterization_rules;
+   key->twoside = lp->rasterizer->light_twoside;
    key->size = Offset(struct lp_setup_variant_key,
 		      inputs[key->num_inputs]);
+   key->color_slot = lp->color_slot[0];
+   key->bcolor_slot = lp->bcolor_slot[0];
+   key->spec_slot = lp->color_slot[1];
+   key->bspec_slot = lp->bcolor_slot[1];
+   key->units = (float) (lp->rasterizer->offset_units * lp->mrd);
+   key->scale = lp->rasterizer->offset_scale;
    key->pad = 0;
-
    memcpy(key->inputs, fs->inputs, key->num_inputs * sizeof key->inputs[0]);
    for (i = 0; i < key->num_inputs; i++) {
       if (key->inputs[i].interp == LP_INTERP_COLOR) {
-	 if (lp->rasterizer->flatshade)
+         if (lp->rasterizer->flatshade)
 	    key->inputs[i].interp = LP_INTERP_CONSTANT;
 	 else
 	    key->inputs[i].interp = LP_INTERP_LINEAR;
@@ -702,7 +851,7 @@ llvmpipe_update_setup(struct llvmpipe_context *lp)
 	 cull_setup_variants(lp);
       }
 
-      variant = generate_setup_variant(screen, key);
+      variant = generate_setup_variant(screen, key, lp);
       insert_at_head(&lp->setup_variants_list, &variant->list_item_global);
       lp->nr_setup_variants++;
    }
diff --git a/src/gallium/drivers/llvmpipe/lp_state_setup.h b/src/gallium/drivers/llvmpipe/lp_state_setup.h
index b0c81baa75..90c55ca4ce 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_state_setup.h
@@ -15,11 +15,20 @@ struct lp_setup_variant_list_item
 
 
 struct lp_setup_variant_key {   
+   unsigned size:16;
    unsigned num_inputs:8;
+   unsigned color_slot:8;
+
+   unsigned bcolor_slot:8;
+   unsigned spec_slot:8;
+   unsigned bspec_slot:8;
    unsigned flatshade_first:1;
    unsigned pixel_center_half:1;
-   unsigned pad:7;
-   unsigned size:16;
+   unsigned twoside:1;
+   unsigned pad:5;
+
+   float units;
+   float scale;      
    struct lp_shader_input inputs[PIPE_MAX_SHADER_INPUTS];
 };
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state_surface.c b/src/gallium/drivers/llvmpipe/lp_state_surface.c
index cd1a5b1980..375ceb2b94 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_surface.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_surface.c
@@ -77,6 +77,7 @@ llvmpipe_set_framebuffer_state(struct pipe_context *pipe,
          else {
             mrd = 0.00002;
          }
+         lp->mrd = mrd;
          draw_set_mrd(lp->draw, mrd);
       }