llvmpipe: Use the generated SoA blending code.

7 files changed, 225 insertions, 780 deletions

diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 7b5da6ee91..72387301a6 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -46,13 +46,14 @@ struct llvmpipe_tile_cache;
 struct llvmpipe_tex_tile_cache;
 struct lp_fragment_shader;
 struct lp_vertex_shader;
+struct lp_blend_state;
 
 
 struct llvmpipe_context {
    struct pipe_context pipe;  /**< base class */
 
    /** Constant state objects */
-   const struct pipe_blend_state *blend;
+   struct lp_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;
@@ -60,7 +61,7 @@ struct llvmpipe_context {
    const struct lp_vertex_shader *vs;
 
    /** Other rendering state */
-   struct pipe_blend_color blend_color;
+   float ALIGN16_ATTRIB blend_color[4][QUAD_SIZE];
    struct pipe_clip_state clip;
    struct pipe_constant_buffer constants[PIPE_SHADER_TYPES];
    struct pipe_framebuffer_state framebuffer;
diff --git a/src/gallium/drivers/llvmpipe/lp_quad.h b/src/gallium/drivers/llvmpipe/lp_quad.h
index b25c35d327..96cd27de81 100644
--- a/src/gallium/drivers/llvmpipe/lp_quad.h
+++ b/src/gallium/drivers/llvmpipe/lp_quad.h
@@ -83,7 +83,7 @@ struct quad_header_inout
 struct quad_header_output
 {
    /** colors in SOA format (rrrr, gggg, bbbb, aaaa) */
-   float color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE];
+   float ALIGN16_ATTRIB color[PIPE_MAX_COLOR_BUFS][NUM_CHANNELS][QUAD_SIZE];
    float depth[QUAD_SIZE];
 };
 
diff --git a/src/gallium/drivers/llvmpipe/lp_quad_blend.c b/src/gallium/drivers/llvmpipe/lp_quad_blend.c
index fe6d30d7c8..fa2e902b6e 100644
--- a/src/gallium/drivers/llvmpipe/lp_quad_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_quad_blend.c
@@ -26,96 +26,24 @@
  **************************************************************************/
 
 /**
- * quad blending
- * \author Brian Paul
+ * Quad blending.
+ *
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ * @author Brian Paul
  */
 
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_debug_dump.h"
 #include "lp_context.h"
+#include "lp_state.h"
 #include "lp_quad.h"
 #include "lp_surface.h"
 #include "lp_tile_cache.h"
 #include "lp_quad_pipe.h"
 
 
-#define VEC4_COPY(DST, SRC) \
-do { \
-    DST[0] = SRC[0]; \
-    DST[1] = SRC[1]; \
-    DST[2] = SRC[2]; \
-    DST[3] = SRC[3]; \
-} while(0)
-
-#define VEC4_SCALAR(DST, SRC) \
-do { \
-    DST[0] = SRC; \
-    DST[1] = SRC; \
-    DST[2] = SRC; \
-    DST[3] = SRC; \
-} while(0)
-
-#define VEC4_ADD(R, A, B) \
-do { \
-   R[0] = A[0] + B[0]; \
-   R[1] = A[1] + B[1]; \
-   R[2] = A[2] + B[2]; \
-   R[3] = A[3] + B[3]; \
-} while (0)
-
-#define VEC4_SUB(R, A, B) \
-do { \
-   R[0] = A[0] - B[0]; \
-   R[1] = A[1] - B[1]; \
-   R[2] = A[2] - B[2]; \
-   R[3] = A[3] - B[3]; \
-} while (0)
-
-/** Add and limit result to ceiling of 1.0 */
-#define VEC4_ADD_SAT(R, A, B) \
-do { \
-   R[0] = A[0] + B[0];  if (R[0] > 1.0f) R[0] = 1.0f; \
-   R[1] = A[1] + B[1];  if (R[1] > 1.0f) R[1] = 1.0f; \
-   R[2] = A[2] + B[2];  if (R[2] > 1.0f) R[2] = 1.0f; \
-   R[3] = A[3] + B[3];  if (R[3] > 1.0f) R[3] = 1.0f; \
-} while (0)
-
-/** Subtract and limit result to floor of 0.0 */
-#define VEC4_SUB_SAT(R, A, B) \
-do { \
-   R[0] = A[0] - B[0];  if (R[0] < 0.0f) R[0] = 0.0f; \
-   R[1] = A[1] - B[1];  if (R[1] < 0.0f) R[1] = 0.0f; \
-   R[2] = A[2] - B[2];  if (R[2] < 0.0f) R[2] = 0.0f; \
-   R[3] = A[3] - B[3];  if (R[3] < 0.0f) R[3] = 0.0f; \
-} while (0)
-
-#define VEC4_MUL(R, A, B) \
-do { \
-   R[0] = A[0] * B[0]; \
-   R[1] = A[1] * B[1]; \
-   R[2] = A[2] * B[2]; \
-   R[3] = A[3] * B[3]; \
-} while (0)
-
-#define VEC4_MIN(R, A, B) \
-do { \
-   R[0] = (A[0] < B[0]) ? A[0] : B[0]; \
-   R[1] = (A[1] < B[1]) ? A[1] : B[1]; \
-   R[2] = (A[2] < B[2]) ? A[2] : B[2]; \
-   R[3] = (A[3] < B[3]) ? A[3] : B[3]; \
-} while (0)
-
-#define VEC4_MAX(R, A, B) \
-do { \
-   R[0] = (A[0] > B[0]) ? A[0] : B[0]; \
-   R[1] = (A[1] > B[1]) ? A[1] : B[1]; \
-   R[2] = (A[2] > B[2]) ? A[2] : B[2]; \
-   R[3] = (A[3] > B[3]) ? A[3] : B[3]; \
-} while (0)
-
-
-
 static void
 logicop_quad(struct quad_stage *qs, 
              float (*quadColor)[4],
@@ -142,7 +70,7 @@ logicop_quad(struct quad_stage *qs,
       src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */
    }
 
-   switch (llvmpipe->blend->logicop_func) {
+   switch (llvmpipe->blend->base.logicop_func) {
    case PIPE_LOGICOP_CLEAR:
       for (j = 0; j < 4; j++)
          res4[j] = 0;
@@ -220,488 +148,6 @@ logicop_quad(struct quad_stage *qs,
 }
 
 
-
-static void
-blend_quad(struct quad_stage *qs, 
-           float (*quadColor)[4],
-           float (*dest)[4])
-{
-   static const float zero[4] = { 0, 0, 0, 0 };
-   static const float one[4] = { 1, 1, 1, 1 };
-   struct llvmpipe_context *llvmpipe = qs->llvmpipe;
-   float source[4][QUAD_SIZE];
-
-   /*
-    * Compute src/first term RGB
-    */
-   switch (llvmpipe->blend->rgb_src_factor) {
-   case PIPE_BLENDFACTOR_ONE:
-      VEC4_COPY(source[0], quadColor[0]); /* R */
-      VEC4_COPY(source[1], quadColor[1]); /* G */
-      VEC4_COPY(source[2], quadColor[2]); /* B */
-      break;
-   case PIPE_BLENDFACTOR_SRC_COLOR:
-      VEC4_MUL(source[0], quadColor[0], quadColor[0]); /* R */
-      VEC4_MUL(source[1], quadColor[1], quadColor[1]); /* G */
-      VEC4_MUL(source[2], quadColor[2], quadColor[2]); /* B */
-      break;
-   case PIPE_BLENDFACTOR_SRC_ALPHA:
-   {
-      const float *alpha = quadColor[3];
-      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_DST_COLOR:
-      VEC4_MUL(source[0], quadColor[0], dest[0]); /* R */
-      VEC4_MUL(source[1], quadColor[1], dest[1]); /* G */
-      VEC4_MUL(source[2], quadColor[2], dest[2]); /* B */
-      break;
-   case PIPE_BLENDFACTOR_DST_ALPHA:
-   {
-      const float *alpha = dest[3];
-      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-   {
-      const float *alpha = quadColor[3];
-      float diff[4], temp[4];
-      VEC4_SUB(diff, one, dest[3]);
-      VEC4_MIN(temp, alpha, diff);
-      VEC4_MUL(source[0], quadColor[0], temp); /* R */
-      VEC4_MUL(source[1], quadColor[1], temp); /* G */
-      VEC4_MUL(source[2], quadColor[2], temp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_CONST_COLOR:
-   {
-      float comp[4];
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[0]); /* R */
-      VEC4_MUL(source[0], quadColor[0], comp); /* R */
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[1]); /* G */
-      VEC4_MUL(source[1], quadColor[1], comp); /* G */
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[2]); /* B */
-      VEC4_MUL(source[2], quadColor[2], comp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_CONST_ALPHA:
-   {
-      float alpha[4];
-      VEC4_SCALAR(alpha, llvmpipe->blend_color.color[3]);
-      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_SRC1_COLOR:
-      assert(0); /* to do */
-      break;
-   case PIPE_BLENDFACTOR_SRC1_ALPHA:
-      assert(0); /* to do */
-      break;
-   case PIPE_BLENDFACTOR_ZERO:
-      VEC4_COPY(source[0], zero); /* R */
-      VEC4_COPY(source[1], zero); /* G */
-      VEC4_COPY(source[2], zero); /* B */
-      break;
-   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-   {
-      float inv_comp[4];
-      VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
-      VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
-      VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
-      VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
-      VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
-      VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-   {
-      float inv_alpha[4];
-      VEC4_SUB(inv_alpha, one, quadColor[3]);
-      VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-   {
-      float inv_alpha[4];
-      VEC4_SUB(inv_alpha, one, dest[3]);
-      VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_DST_COLOR:
-   {
-      float inv_comp[4];
-      VEC4_SUB(inv_comp, one, dest[0]); /* R */
-      VEC4_MUL(source[0], quadColor[0], inv_comp); /* R */
-      VEC4_SUB(inv_comp, one, dest[1]); /* G */
-      VEC4_MUL(source[1], quadColor[1], inv_comp); /* G */
-      VEC4_SUB(inv_comp, one, dest[2]); /* B */
-      VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-   {
-      float inv_comp[4];
-      /* R */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[0]);
-      VEC4_MUL(source[0], quadColor[0], inv_comp);
-      /* G */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[1]);
-      VEC4_MUL(source[1], quadColor[1], inv_comp);
-      /* B */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[2]);
-      VEC4_MUL(source[2], quadColor[2], inv_comp);
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-   {
-      float inv_alpha[4];
-      VEC4_SCALAR(inv_alpha, 1.0f - llvmpipe->blend_color.color[3]);
-      VEC4_MUL(source[0], quadColor[0], inv_alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], inv_alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
-      assert(0); /* to do */
-      break;
-   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
-      assert(0); /* to do */
-      break;
-   default:
-      assert(0);
-   }
-
-   /*
-    * Compute src/first term A
-    */
-   switch (llvmpipe->blend->alpha_src_factor) {
-   case PIPE_BLENDFACTOR_ONE:
-      VEC4_COPY(source[3], quadColor[3]); /* A */
-      break;
-   case PIPE_BLENDFACTOR_SRC_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_SRC_ALPHA:
-   {
-      const float *alpha = quadColor[3];
-      VEC4_MUL(source[3], quadColor[3], alpha); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_DST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_DST_ALPHA:
-      VEC4_MUL(source[3], quadColor[3], dest[3]); /* A */
-      break;
-   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-      /* multiply alpha by 1.0 */
-      VEC4_COPY(source[3], quadColor[3]); /* A */
-      break;
-   case PIPE_BLENDFACTOR_CONST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_CONST_ALPHA:
-   {
-      float comp[4];
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[3]); /* A */
-      VEC4_MUL(source[3], quadColor[3], comp); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_ZERO:
-      VEC4_COPY(source[3], zero); /* A */
-      break;
-   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-   {
-      float inv_alpha[4];
-      VEC4_SUB(inv_alpha, one, quadColor[3]);
-      VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_DST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-   {
-      float inv_alpha[4];
-      VEC4_SUB(inv_alpha, one, dest[3]);
-      VEC4_MUL(source[3], quadColor[3], inv_alpha); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-   {
-      float inv_comp[4];
-      /* A */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[3]);
-      VEC4_MUL(source[3], quadColor[3], inv_comp);
-   }
-   break;
-   default:
-      assert(0);
-   }
-
-
-   /*
-    * Compute dest/second term RGB
-    */
-   switch (llvmpipe->blend->rgb_dst_factor) {
-   case PIPE_BLENDFACTOR_ONE:
-      /* dest = dest * 1   NO-OP, leave dest as-is */
-      break;
-   case PIPE_BLENDFACTOR_SRC_COLOR:
-      VEC4_MUL(dest[0], dest[0], quadColor[0]); /* R */
-      VEC4_MUL(dest[1], dest[1], quadColor[1]); /* G */
-      VEC4_MUL(dest[2], dest[2], quadColor[2]); /* B */
-      break;
-   case PIPE_BLENDFACTOR_SRC_ALPHA:
-      VEC4_MUL(dest[0], dest[0], quadColor[3]); /* R * A */
-      VEC4_MUL(dest[1], dest[1], quadColor[3]); /* G * A */
-      VEC4_MUL(dest[2], dest[2], quadColor[3]); /* B * A */
-      break;
-   case PIPE_BLENDFACTOR_DST_ALPHA:
-      VEC4_MUL(dest[0], dest[0], dest[3]); /* R * A */
-      VEC4_MUL(dest[1], dest[1], dest[3]); /* G * A */
-      VEC4_MUL(dest[2], dest[2], dest[3]); /* B * A */
-      break;
-   case PIPE_BLENDFACTOR_DST_COLOR:
-      VEC4_MUL(dest[0], dest[0], dest[0]); /* R */
-      VEC4_MUL(dest[1], dest[1], dest[1]); /* G */
-      VEC4_MUL(dest[2], dest[2], dest[2]); /* B */
-      break;
-   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-      assert(0); /* illegal */
-      break;
-   case PIPE_BLENDFACTOR_CONST_COLOR:
-   {
-      float comp[4];
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[0]); /* R */
-      VEC4_MUL(dest[0], dest[0], comp); /* R */
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[1]); /* G */
-      VEC4_MUL(dest[1], dest[1], comp); /* G */
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[2]); /* B */
-      VEC4_MUL(dest[2], dest[2], comp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_CONST_ALPHA:
-   {
-      float comp[4];
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[3]); /* A */
-      VEC4_MUL(dest[0], dest[0], comp); /* R */
-      VEC4_MUL(dest[1], dest[1], comp); /* G */
-      VEC4_MUL(dest[2], dest[2], comp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_ZERO:
-      VEC4_COPY(dest[0], zero); /* R */
-      VEC4_COPY(dest[1], zero); /* G */
-      VEC4_COPY(dest[2], zero); /* B */
-      break;
-   case PIPE_BLENDFACTOR_SRC1_COLOR:
-   case PIPE_BLENDFACTOR_SRC1_ALPHA:
-      /* XXX what are these? */
-      assert(0);
-      break;
-   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-   {
-      float inv_comp[4];
-      VEC4_SUB(inv_comp, one, quadColor[0]); /* R */
-      VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
-      VEC4_SUB(inv_comp, one, quadColor[1]); /* G */
-      VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
-      VEC4_SUB(inv_comp, one, quadColor[2]); /* B */
-      VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-   {
-      float one_minus_alpha[QUAD_SIZE];
-      VEC4_SUB(one_minus_alpha, one, quadColor[3]);
-      VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */
-      VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */
-      VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-   {
-      float inv_comp[4];
-      VEC4_SUB(inv_comp, one, dest[3]); /* A */
-      VEC4_MUL(dest[0], inv_comp, dest[0]); /* R */
-      VEC4_MUL(dest[1], inv_comp, dest[1]); /* G */
-      VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_DST_COLOR:
-   {
-      float inv_comp[4];
-      VEC4_SUB(inv_comp, one, dest[0]); /* R */
-      VEC4_MUL(dest[0], dest[0], inv_comp); /* R */
-      VEC4_SUB(inv_comp, one, dest[1]); /* G */
-      VEC4_MUL(dest[1], dest[1], inv_comp); /* G */
-      VEC4_SUB(inv_comp, one, dest[2]); /* B */
-      VEC4_MUL(dest[2], dest[2], inv_comp); /* B */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-   {
-      float inv_comp[4];
-      /* R */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[0]);
-      VEC4_MUL(dest[0], dest[0], inv_comp);
-      /* G */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[1]);
-      VEC4_MUL(dest[1], dest[1], inv_comp);
-      /* B */
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[2]);
-      VEC4_MUL(dest[2], dest[2], inv_comp);
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-   {
-      float inv_comp[4];
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[3]);
-      VEC4_MUL(dest[0], dest[0], inv_comp);
-      VEC4_MUL(dest[1], dest[1], inv_comp);
-      VEC4_MUL(dest[2], dest[2], inv_comp);
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
-   case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
-      /* XXX what are these? */
-      assert(0);
-      break;
-   default:
-      assert(0);
-   }
-
-   /*
-    * Compute dest/second term A
-    */
-   switch (llvmpipe->blend->alpha_dst_factor) {
-   case PIPE_BLENDFACTOR_ONE:
-      /* dest = dest * 1   NO-OP, leave dest as-is */
-      break;
-   case PIPE_BLENDFACTOR_SRC_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_SRC_ALPHA:
-      VEC4_MUL(dest[3], dest[3], quadColor[3]); /* A * A */
-      break;
-   case PIPE_BLENDFACTOR_DST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_DST_ALPHA:
-      VEC4_MUL(dest[3], dest[3], dest[3]); /* A */
-      break;
-   case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
-      assert(0); /* illegal */
-      break;
-   case PIPE_BLENDFACTOR_CONST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_CONST_ALPHA:
-   {
-      float comp[4];
-      VEC4_SCALAR(comp, llvmpipe->blend_color.color[3]); /* A */
-      VEC4_MUL(dest[3], dest[3], comp); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_ZERO:
-      VEC4_COPY(dest[3], zero); /* A */
-      break;
-   case PIPE_BLENDFACTOR_INV_SRC_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
-   {
-      float one_minus_alpha[QUAD_SIZE];
-      VEC4_SUB(one_minus_alpha, one, quadColor[3]);
-      VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_DST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_INV_DST_ALPHA:
-   {
-      float inv_comp[4];
-      VEC4_SUB(inv_comp, one, dest[3]); /* A */
-      VEC4_MUL(dest[3], inv_comp, dest[3]); /* A */
-   }
-   break;
-   case PIPE_BLENDFACTOR_INV_CONST_COLOR:
-      /* fall-through */
-   case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
-   {
-      float inv_comp[4];
-      VEC4_SCALAR(inv_comp, 1.0f - llvmpipe->blend_color.color[3]);
-      VEC4_MUL(dest[3], dest[3], inv_comp);
-   }
-   break;
-   default:
-      assert(0);
-   }
-
-   /*
-    * Combine RGB terms
-    */
-   switch (llvmpipe->blend->rgb_func) {
-   case PIPE_BLEND_ADD:
-      VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */
-      VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */
-      VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */
-      break;
-   case PIPE_BLEND_SUBTRACT:
-      VEC4_SUB_SAT(quadColor[0], source[0], dest[0]); /* R */
-      VEC4_SUB_SAT(quadColor[1], source[1], dest[1]); /* G */
-      VEC4_SUB_SAT(quadColor[2], source[2], dest[2]); /* B */
-      break;
-   case PIPE_BLEND_REVERSE_SUBTRACT:
-      VEC4_SUB_SAT(quadColor[0], dest[0], source[0]); /* R */
-      VEC4_SUB_SAT(quadColor[1], dest[1], source[1]); /* G */
-      VEC4_SUB_SAT(quadColor[2], dest[2], source[2]); /* B */
-      break;
-   case PIPE_BLEND_MIN:
-      VEC4_MIN(quadColor[0], source[0], dest[0]); /* R */
-      VEC4_MIN(quadColor[1], source[1], dest[1]); /* G */
-      VEC4_MIN(quadColor[2], source[2], dest[2]); /* B */
-      break;
-   case PIPE_BLEND_MAX:
-      VEC4_MAX(quadColor[0], source[0], dest[0]); /* R */
-      VEC4_MAX(quadColor[1], source[1], dest[1]); /* G */
-      VEC4_MAX(quadColor[2], source[2], dest[2]); /* B */
-      break;
-   default:
-      assert(0);
-   }
-
-   /*
-    * Combine A terms
-    */
-   switch (llvmpipe->blend->alpha_func) {
-   case PIPE_BLEND_ADD:
-      VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */
-      break;
-   case PIPE_BLEND_SUBTRACT:
-      VEC4_SUB_SAT(quadColor[3], source[3], dest[3]); /* A */
-      break;
-   case PIPE_BLEND_REVERSE_SUBTRACT:
-      VEC4_SUB_SAT(quadColor[3], dest[3], source[3]); /* A */
-      break;
-   case PIPE_BLEND_MIN:
-      VEC4_MIN(quadColor[3], source[3], dest[3]); /* A */
-      break;
-   case PIPE_BLEND_MAX:
-      VEC4_MAX(quadColor[3], source[3], dest[3]); /* A */
-      break;
-   default:
-      assert(0);
-   }
-}
-
 static void
 colormask_quad(struct quad_stage *qs,
                float (*quadColor)[4],
@@ -710,40 +156,45 @@ colormask_quad(struct quad_stage *qs,
    struct llvmpipe_context *llvmpipe = qs->llvmpipe;
 
    /* R */
-   if (!(llvmpipe->blend->colormask & PIPE_MASK_R))
+   if (!(llvmpipe->blend->base.colormask & PIPE_MASK_R))
       COPY_4V(quadColor[0], dest[0]);
 
    /* G */
-   if (!(llvmpipe->blend->colormask & PIPE_MASK_G))
+   if (!(llvmpipe->blend->base.colormask & PIPE_MASK_G))
       COPY_4V(quadColor[1], dest[1]);
 
    /* B */
-   if (!(llvmpipe->blend->colormask & PIPE_MASK_B))
+   if (!(llvmpipe->blend->base.colormask & PIPE_MASK_B))
       COPY_4V(quadColor[2], dest[2]);
 
    /* A */
-   if (!(llvmpipe->blend->colormask & PIPE_MASK_A))
+   if (!(llvmpipe->blend->base.colormask & PIPE_MASK_A))
       COPY_4V(quadColor[3], dest[3]);
 }
 
 
+static void blend_begin(struct quad_stage *qs)
+{
+}
+
+
 static void
-blend_fallback(struct quad_stage *qs, 
-               struct quad_header *quads[],
-               unsigned nr)
+blend_run(struct quad_stage *qs,
+          struct quad_header *quads[],
+          unsigned nr)
 {
    struct llvmpipe_context *llvmpipe = qs->llvmpipe;
-   const struct pipe_blend_state *blend = llvmpipe->blend;
+   struct lp_blend_state *blend = llvmpipe->blend;
    unsigned cbuf;
+   uint q, i, j;
 
    for (cbuf = 0; cbuf < llvmpipe->framebuffer.nr_cbufs; cbuf++) 
    {
-      float dest[4][QUAD_SIZE];
+      float ALIGN16_ATTRIB dest[4][QUAD_SIZE];
       struct llvmpipe_cached_tile *tile
          = lp_get_cached_tile(llvmpipe->cbuf_cache[cbuf],
                               quads[0]->input.x0, 
                               quads[0]->input.y0);
-      uint q, i, j;
 
       for (q = 0; q < nr; q++) {
          struct quad_header *quad = quads[q];
@@ -762,14 +213,19 @@ blend_fallback(struct quad_stage *qs,
          }
 
 
-         if (blend->logicop_enable) {
+         if (blend->base.logicop_enable) {
             logicop_quad( qs, quadColor, dest );
          }
-         else if (blend->blend_enable) {
-            blend_quad( qs, quadColor, dest );
+         else {
+            assert(blend->jit_function);
+            assert((((uintptr_t)quadColor) & 0xf) == 0);
+            assert((((uintptr_t)dest) & 0xf) == 0);
+            assert((((uintptr_t)llvmpipe->blend_color) & 0xf) == 0);
+            if(blend->jit_function)
+               blend->jit_function( quadColor, dest, llvmpipe->blend_color, quadColor );
          }
 
-         if (blend->colormask != 0xf)
+         if (blend->base.colormask != 0xf)
             colormask_quad( qs, quadColor, dest );
    
          /* Output color values
@@ -788,197 +244,6 @@ blend_fallback(struct quad_stage *qs,
 }
 
 
-static void
-blend_single_add_src_alpha_inv_src_alpha(struct quad_stage *qs, 
-                                         struct quad_header *quads[],
-                                         unsigned nr)
-{
-   static const float one[4] = { 1, 1, 1, 1 };
-   float one_minus_alpha[QUAD_SIZE];
-   float dest[4][QUAD_SIZE];
-   float source[4][QUAD_SIZE];
-   uint i, j, q;
-
-   struct llvmpipe_cached_tile *tile
-      = lp_get_cached_tile(qs->llvmpipe->cbuf_cache[0],
-                           quads[0]->input.x0, 
-                           quads[0]->input.y0);
-
-   for (q = 0; q < nr; q++) {
-      struct quad_header *quad = quads[q];
-      float (*quadColor)[4] = quad->output.color[0];
-      const float *alpha = quadColor[3];
-      const int itx = (quad->input.x0 & (TILE_SIZE-1));
-      const int ity = (quad->input.y0 & (TILE_SIZE-1));
-      
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = itx + (j & 1);
-         int y = ity + (j >> 1);
-         for (i = 0; i < 4; i++) {
-            dest[i][j] = tile->data.color[y][x][i];
-         }
-      }
-
-      VEC4_MUL(source[0], quadColor[0], alpha); /* R */
-      VEC4_MUL(source[1], quadColor[1], alpha); /* G */
-      VEC4_MUL(source[2], quadColor[2], alpha); /* B */
-      VEC4_MUL(source[3], quadColor[3], alpha); /* A */
-
-      VEC4_SUB(one_minus_alpha, one, alpha);
-      VEC4_MUL(dest[0], dest[0], one_minus_alpha); /* R */
-      VEC4_MUL(dest[1], dest[1], one_minus_alpha); /* G */
-      VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
-      VEC4_MUL(dest[3], dest[3], one_minus_alpha); /* B */
-
-      VEC4_ADD_SAT(quadColor[0], source[0], dest[0]); /* R */
-      VEC4_ADD_SAT(quadColor[1], source[1], dest[1]); /* G */
-      VEC4_ADD_SAT(quadColor[2], source[2], dest[2]); /* B */
-      VEC4_ADD_SAT(quadColor[3], source[3], dest[3]); /* A */
-
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (quad->inout.mask & (1 << j)) {
-            int x = itx + (j & 1);
-            int y = ity + (j >> 1);
-            for (i = 0; i < 4; i++) { /* loop over color chans */
-               tile->data.color[y][x][i] = quadColor[i][j];
-            }
-         }
-      }
-   }
-}
-
-static void
-blend_single_add_one_one(struct quad_stage *qs, 
-                         struct quad_header *quads[],
-                         unsigned nr)
-{
-   float dest[4][QUAD_SIZE];
-   uint i, j, q;
-
-   struct llvmpipe_cached_tile *tile
-      = lp_get_cached_tile(qs->llvmpipe->cbuf_cache[0],
-                           quads[0]->input.x0, 
-                           quads[0]->input.y0);
-
-   for (q = 0; q < nr; q++) {
-      struct quad_header *quad = quads[q];
-      float (*quadColor)[4] = quad->output.color[0];
-      const int itx = (quad->input.x0 & (TILE_SIZE-1));
-      const int ity = (quad->input.y0 & (TILE_SIZE-1));
-      
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = itx + (j & 1);
-         int y = ity + (j >> 1);
-         for (i = 0; i < 4; i++) {
-            dest[i][j] = tile->data.color[y][x][i];
-         }
-      }
-     
-      VEC4_ADD_SAT(quadColor[0], quadColor[0], dest[0]); /* R */
-      VEC4_ADD_SAT(quadColor[1], quadColor[1], dest[1]); /* G */
-      VEC4_ADD_SAT(quadColor[2], quadColor[2], dest[2]); /* B */
-      VEC4_ADD_SAT(quadColor[3], quadColor[3], dest[3]); /* A */
-
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (quad->inout.mask & (1 << j)) {
-            int x = itx + (j & 1);
-            int y = ity + (j >> 1);
-            for (i = 0; i < 4; i++) { /* loop over color chans */
-               tile->data.color[y][x][i] = quadColor[i][j];
-            }
-         }
-      }
-   }
-}
-
-
-static void
-single_output_color(struct quad_stage *qs, 
-                    struct quad_header *quads[],
-                    unsigned nr)
-{
-   uint i, j, q;
-
-   struct llvmpipe_cached_tile *tile
-      = lp_get_cached_tile(qs->llvmpipe->cbuf_cache[0],
-                           quads[0]->input.x0, 
-                           quads[0]->input.y0);
-
-   for (q = 0; q < nr; q++) {
-      struct quad_header *quad = quads[q];
-      float (*quadColor)[4] = quad->output.color[0];
-      const int itx = (quad->input.x0 & (TILE_SIZE-1));
-      const int ity = (quad->input.y0 & (TILE_SIZE-1));
-      
-      for (j = 0; j < QUAD_SIZE; j++) {
-         if (quad->inout.mask & (1 << j)) {
-            int x = itx + (j & 1);
-            int y = ity + (j >> 1);
-            for (i = 0; i < 4; i++) { /* loop over color chans */
-               tile->data.color[y][x][i] = quadColor[i][j];
-            }
-         }
-      }
-   }
-}
-
-static void
-blend_noop(struct quad_stage *qs, 
-           struct quad_header *quads[],
-           unsigned nr)
-{
-}
-
-
-static void
-choose_blend_quad(struct quad_stage *qs, 
-                  struct quad_header *quads[],
-                  unsigned nr)
-{
-   struct llvmpipe_context *llvmpipe = qs->llvmpipe;
-   const struct pipe_blend_state *blend = llvmpipe->blend;
-
-   qs->run = blend_fallback;
-   
-   if (llvmpipe->framebuffer.nr_cbufs == 0) {
-      qs->run = blend_noop;
-   }
-   else if (!llvmpipe->blend->logicop_enable &&
-            llvmpipe->blend->colormask == 0xf) 
-   {
-      if (!blend->blend_enable) {
-         qs->run = single_output_color;
-      }
-      else if (blend->rgb_src_factor == blend->alpha_src_factor &&
-               blend->rgb_dst_factor == blend->alpha_dst_factor &&
-               blend->rgb_func == blend->alpha_func &&
-               llvmpipe->framebuffer.nr_cbufs == 1)
-      {
-         if (blend->alpha_func == PIPE_BLEND_ADD) {
-            if (blend->rgb_src_factor == PIPE_BLENDFACTOR_ONE &&
-                blend->rgb_dst_factor == PIPE_BLENDFACTOR_ONE) {
-               qs->run = blend_single_add_one_one;
-            }
-            else if (blend->rgb_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA &&
-                blend->rgb_dst_factor == PIPE_BLENDFACTOR_INV_SRC_ALPHA)
-               qs->run = blend_single_add_src_alpha_inv_src_alpha;
-
-         }
-      }
-   }
-
-   qs->run(qs, quads, nr);
-}
-
-
-static void blend_begin(struct quad_stage *qs)
-{
-   qs->run = choose_blend_quad;
-}
-
-
 static void blend_destroy(struct quad_stage *qs)
 {
    FREE( qs );
@@ -991,7 +256,7 @@ struct quad_stage *lp_quad_blend_stage( struct llvmpipe_context *llvmpipe )
 
    stage->llvmpipe = llvmpipe;
    stage->begin = blend_begin;
-   stage->run = choose_blend_quad;
+   stage->run = blend_run;
    stage->destroy = blend_destroy;
 
    return stage;
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index b71f15678e..2236711e9b 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -26,6 +26,8 @@
  **************************************************************************/
 
 
+#include <llvm-c/Transforms/Scalar.h>
+
 #include "util/u_memory.h"
 #include "util/u_simple_screen.h"
 #include "pipe/internal/p_winsys_screen.h"
@@ -149,9 +151,17 @@ llvmpipe_is_format_supported( struct pipe_screen *screen,
 
 
 static void
-llvmpipe_destroy_screen( struct pipe_screen *screen )
+llvmpipe_destroy_screen( struct pipe_screen *_screen )
 {
-   struct pipe_winsys *winsys = screen->winsys;
+   struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
+
+   struct pipe_winsys *winsys = _screen->winsys;
+
+   if(screen->engine)
+      LLVMDisposeExecutionEngine(screen->engine);
+
+   if(screen->pass)
+      LLVMDisposePassManager(screen->pass);
 
    if(winsys->destroy)
       winsys->destroy(winsys);
@@ -169,6 +179,7 @@ struct pipe_screen *
 llvmpipe_create_screen(struct pipe_winsys *winsys)
 {
    struct llvmpipe_screen *screen = CALLOC_STRUCT(llvmpipe_screen);
+   char *error = NULL;
 
    if (!screen)
       return NULL;
@@ -186,5 +197,25 @@ llvmpipe_create_screen(struct pipe_winsys *winsys)
    llvmpipe_init_screen_texture_funcs(&screen->base);
    u_simple_screen_init(&screen->base);
 
+   screen->module = LLVMModuleCreateWithName("llvmpipe");
+
+   screen->provider = LLVMCreateModuleProviderForExistingModule(screen->module);
+
+   if (LLVMCreateJITCompiler(&screen->engine, screen->provider, 1, &error)) {
+      fprintf(stderr, "%s\n", error);
+      LLVMDisposeMessage(error);
+      abort();
+   }
+
+   screen->pass = LLVMCreateFunctionPassManager(screen->provider);
+   LLVMAddTargetData(LLVMGetExecutionEngineTargetData(screen->engine), screen->pass);
+   /* These are the passes currently listed in llvm-c/Transforms/Scalar.h,
+    * but there are more on SVN. */
+   LLVMAddConstantPropagationPass(screen->pass);
+   LLVMAddInstructionCombiningPass(screen->pass);
+   LLVMAddPromoteMemoryToRegisterPass(screen->pass);
+   LLVMAddGVNPass(screen->pass);
+   LLVMAddCFGSimplificationPass(screen->pass);
+
    return &screen->base;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.h b/src/gallium/drivers/llvmpipe/lp_screen.h
index 7386ebefe6..c3ff1531d2 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.h
+++ b/src/gallium/drivers/llvmpipe/lp_screen.h
@@ -31,14 +31,23 @@
 #ifndef LP_SCREEN_H
 #define LP_SCREEN_H
 
+#include <llvm-c/Core.h>
+#include <llvm-c/Analysis.h>
+#include <llvm-c/ExecutionEngine.h>
+
 #include "pipe/p_screen.h"
 #include "pipe/p_defines.h"
 
 
-
-struct llvmpipe_screen {
+struct llvmpipe_screen
+{
    struct pipe_screen base;
 
+   LLVMModuleRef module;
+   LLVMExecutionEngineRef engine;
+   LLVMModuleProviderRef provider;
+   LLVMPassManagerRef pass;
+
    /* Increments whenever textures are modified.  Contexts can track
     * this.
     */
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index 125567b677..78c4778580 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -31,6 +31,8 @@
 #ifndef LP_STATE_H
 #define LP_STATE_H
 
+#include <llvm-c/Core.h>
+
 #include "pipe/p_state.h"
 #include "tgsi/tgsi_scan.h"
 
@@ -90,6 +92,23 @@ struct lp_vertex_shader {
 };
 
 
+typedef void
+(*lp_blend_func)(float src[][4],
+                 float dst[][4],
+                 float con[][4],
+                 float res[][4]);
+
+
+struct lp_blend_state
+{
+   struct pipe_blend_state base;
+
+   LLVMValueRef function;
+
+   lp_blend_func jit_function;
+};
+
+
 
 void *
 llvmpipe_create_blend_state(struct pipe_context *,
diff --git a/src/gallium/drivers/llvmpipe/lp_state_blend.c b/src/gallium/drivers/llvmpipe/lp_state_blend.c
index 041a54f13e..ed4ea3b330 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_blend.c
@@ -1,5 +1,6 @@
 /**************************************************************************
  * 
+ * Copyright 2009 VMware, Inc.
  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  *
@@ -25,19 +26,126 @@
  * 
  **************************************************************************/
 
-/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+/**
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ * @author Keith Whitwell <keith@tungstengraphics.com>
  */
 
 #include "util/u_memory.h"
+#include "util/u_debug_dump.h"
+#include "lp_screen.h"
 #include "lp_context.h"
 #include "lp_state.h"
 
+#include "lp_bld_type.h"
+#include "lp_bld_arit.h"
+#include "lp_bld_blend.h"
+
+
+static void
+blend_generate(struct llvmpipe_screen *screen,
+               struct lp_blend_state *blend)
+{
+   union lp_type type;
+   LLVMTypeRef vec_type;
+   LLVMTypeRef args[4];
+   LLVMValueRef src_ptr;
+   LLVMValueRef dst_ptr;
+   LLVMValueRef const_ptr;
+   LLVMValueRef res_ptr;
+   LLVMBasicBlockRef block;
+   LLVMBuilderRef builder;
+   LLVMValueRef src[4];
+   LLVMValueRef dst[4];
+   LLVMValueRef con[4];
+   LLVMValueRef res[4];
+   char src_name[5] = "src?";
+   char dst_name[5] = "dst?";
+   char con_name[5] = "con?";
+   char res_name[5] = "res?";
+   unsigned i;
+
+   type.value = 0;
+   type.floating = TRUE;
+   type.sign = FALSE;
+   type.norm = TRUE;
+   type.width = 32;
+   type.length = 4;
+
+   vec_type = lp_build_vec_type(type);
+
+   args[3] = args[2] = args[1] = args[0] = LLVMPointerType(vec_type, 0);
+   blend->function = LLVMAddFunction(screen->module, "blend", LLVMFunctionType(LLVMVoidType(), args, 4, 0));
+   LLVMSetFunctionCallConv(blend->function, LLVMCCallConv);
+   src_ptr = LLVMGetParam(blend->function, 0);
+   dst_ptr = LLVMGetParam(blend->function, 1);
+   const_ptr = LLVMGetParam(blend->function, 2);
+   res_ptr = LLVMGetParam(blend->function, 3);
+
+   block = LLVMAppendBasicBlock(blend->function, "entry");
+   builder = LLVMCreateBuilder();
+   LLVMPositionBuilderAtEnd(builder, block);
+
+   for(i = 0; i < 4; ++i) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      con_name[3] = dst_name[3] = src_name[3] = "rgba"[i];
+      src[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, src_ptr, &index, 1, ""), src_name);
+      dst[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, dst_ptr, &index, 1, ""), dst_name);
+      con[i] = LLVMBuildLoad(builder, LLVMBuildGEP(builder, const_ptr, &index, 1, ""), con_name);
+   }
+
+   lp_build_blend_soa(builder, &blend->base, type, src, dst, con, res);
+
+   for(i = 0; i < 4; ++i) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      res_name[3] = "rgba"[i];
+      LLVMSetValueName(res[i], res_name);
+      LLVMBuildStore(builder, res[i], LLVMBuildGEP(builder, res_ptr, &index, 1, ""));
+   }
+
+   LLVMBuildRetVoid(builder);;
+
+   LLVMDisposeBuilder(builder);
+}
+
 
 void *
 llvmpipe_create_blend_state(struct pipe_context *pipe,
-                            const struct pipe_blend_state *blend)
+                            const struct pipe_blend_state *base)
 {
-   return mem_dup(blend, sizeof(*blend));
+   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
+   struct lp_blend_state *blend;
+
+   blend = CALLOC_STRUCT(lp_blend_state);
+   if(!blend)
+      return NULL;
+
+   blend->base = *base;
+
+   blend_generate(screen, blend);
+
+   LLVMRunFunctionPassManager(screen->pass, blend->function);
+
+#if 1
+   debug_printf("%s=%s %s=%s %s=%s %s=%s %s=%s %s=%s\n",
+                "rgb_func",         debug_dump_blend_func  (blend->base.rgb_func, TRUE),
+                "rgb_src_factor",   debug_dump_blend_factor(blend->base.rgb_src_factor, TRUE),
+                "rgb_dst_factor",   debug_dump_blend_factor(blend->base.rgb_dst_factor, TRUE),
+                "alpha_func",       debug_dump_blend_func  (blend->base.alpha_func, TRUE),
+                "alpha_src_factor", debug_dump_blend_factor(blend->base.alpha_src_factor, TRUE),
+                "alpha_dst_factor", debug_dump_blend_factor(blend->base.alpha_dst_factor, TRUE));
+   LLVMDumpModule(screen->module);
+   debug_printf("\n");
+#endif
+
+   if(LLVMVerifyFunction(blend->function, LLVMPrintMessageAction)) {
+      LLVMDumpModule(screen->module);
+      abort();
+   }
+
+   blend->jit_function = (lp_blend_func)LLVMGetPointerToGlobal(screen->engine, blend->function);
+
+   return blend;
 }
 
 void llvmpipe_bind_blend_state( struct pipe_context *pipe,
@@ -45,14 +153,23 @@ void llvmpipe_bind_blend_state( struct pipe_context *pipe,
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
 
-   llvmpipe->blend = (const struct pipe_blend_state *)blend;
+   llvmpipe->blend = (struct lp_blend_state *)blend;
 
    llvmpipe->dirty |= LP_NEW_BLEND;
 }
 
 void llvmpipe_delete_blend_state(struct pipe_context *pipe,
-                                 void *blend)
+                                 void *_blend)
 {
+   struct llvmpipe_screen *screen = llvmpipe_screen(pipe->screen);
+   struct lp_blend_state *blend = (struct lp_blend_state *)_blend;
+
+   if(blend->function) {
+      if(blend->jit_function)
+         LLVMFreeMachineCodeForFunction(screen->engine, blend->function);
+      LLVMDeleteFunction(blend->function);
+   }
+
    FREE( blend );
 }
 
@@ -61,8 +178,11 @@ void llvmpipe_set_blend_color( struct pipe_context *pipe,
 			     const struct pipe_blend_color *blend_color )
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+   unsigned i, j;
 
-   llvmpipe->blend_color = *blend_color;
+   for (i = 0; i < 4; ++i)
+      for (j = 0; j < QUAD_SIZE; ++j)
+         llvmpipe->blend_color[i][j] = blend_color->color[i];
 
    llvmpipe->dirty |= LP_NEW_BLEND;
 }