softpipe: expand quad pipeline to process >1 quad at a time

This is part one -- we still only pass a single quad down, but
the code can now cope with more.  The quads must all be from the same
tile.

1 files changed, 374 insertions, 356 deletions

diff --git a/src/gallium/drivers/softpipe/sp_quad_blend.c b/src/gallium/drivers/softpipe/sp_quad_blend.c
index 04b5daf3a4..fdf1bb4552 100644
--- a/src/gallium/drivers/softpipe/sp_quad_blend.c
+++ b/src/gallium/drivers/softpipe/sp_quad_blend.c
@@ -117,10 +117,16 @@ do { \
 
 
 static void
-logicop_quad(struct quad_stage *qs, struct quad_header *quad)
+logicop_quad(struct quad_stage *qs, 
+             struct quad_header *quads[],
+             unsigned nr)
 {
    struct softpipe_context *softpipe = qs->softpipe;
    uint cbuf;
+   struct softpipe_cached_tile *
+      tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf],
+                                quads[0]->input.x0, 
+                                quads[0]->input.y0);
 
    /* loop over colorbuffer outputs */
    for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
@@ -129,165 +135,161 @@ logicop_quad(struct quad_stage *qs, struct quad_header *quad)
       uint *src4 = (uint *) src;
       uint *dst4 = (uint *) dst;
       uint *res4 = (uint *) res;
-      struct softpipe_cached_tile *
-         tile = sp_get_cached_tile(softpipe->cbuf_cache[cbuf],
-                                   quad->input.x0, quad->input.y0);
-      float (*quadColor)[4] = quad->output.color[cbuf];
       uint i, j;
 
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
-         for (i = 0; i < 4; i++) {
-            dest[i][j] = tile->data.color[y][x][i];
+      for (i = 0; i < nr; i++) {
+         struct quad_header *quad = quads[i];
+         float (*quadColor)[4] = quad->output.color[cbuf];
+
+         /* get/swizzle dest colors */
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
+            int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
+            for (i = 0; i < 4; i++) {
+               dest[i][j] = tile->data.color[y][x][i];
+            }
          }
-      }
 
-      /* convert to ubyte */
-      for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */
-         dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */
-         dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */
-         dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */
-         dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */
-
-         src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */
-         src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */
-         src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */
-         src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */
-      }
+         /* convert to ubyte */
+         for (j = 0; j < 4; j++) { /* loop over R,G,B,A channels */
+            dst[j][0] = float_to_ubyte(dest[j][0]); /* P0 */
+            dst[j][1] = float_to_ubyte(dest[j][1]); /* P1 */
+            dst[j][2] = float_to_ubyte(dest[j][2]); /* P2 */
+            dst[j][3] = float_to_ubyte(dest[j][3]); /* P3 */
+
+            src[j][0] = float_to_ubyte(quadColor[j][0]); /* P0 */
+            src[j][1] = float_to_ubyte(quadColor[j][1]); /* P1 */
+            src[j][2] = float_to_ubyte(quadColor[j][2]); /* P2 */
+            src[j][3] = float_to_ubyte(quadColor[j][3]); /* P3 */
+         }
 
-      switch (softpipe->blend->logicop_func) {
-      case PIPE_LOGICOP_CLEAR:
-         for (j = 0; j < 4; j++)
-            res4[j] = 0;
-         break;
-      case PIPE_LOGICOP_NOR:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~(src4[j] | dst4[j]);
-         break;
-      case PIPE_LOGICOP_AND_INVERTED:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~src4[j] & dst4[j];
-         break;
-      case PIPE_LOGICOP_COPY_INVERTED:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~src4[j];
-         break;
-      case PIPE_LOGICOP_AND_REVERSE:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] & ~dst4[j];
-         break;
-      case PIPE_LOGICOP_INVERT:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~dst4[j];
-         break;
-      case PIPE_LOGICOP_XOR:
-         for (j = 0; j < 4; j++)
-            res4[j] = dst4[j] ^ src4[j];
-         break;
-      case PIPE_LOGICOP_NAND:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~(src4[j] & dst4[j]);
-         break;
-      case PIPE_LOGICOP_AND:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] & dst4[j];
-         break;
-      case PIPE_LOGICOP_EQUIV:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~(src4[j] ^ dst4[j]);
-         break;
-      case PIPE_LOGICOP_NOOP:
-         for (j = 0; j < 4; j++)
-            res4[j] = dst4[j];
-         break;
-      case PIPE_LOGICOP_OR_INVERTED:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~src4[j] | dst4[j];
-         break;
-      case PIPE_LOGICOP_COPY:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j];
-         break;
-      case PIPE_LOGICOP_OR_REVERSE:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] | ~dst4[j];
-         break;
-      case PIPE_LOGICOP_OR:
-         for (j = 0; j < 4; j++)
-            res4[j] = src4[j] | dst4[j];
-         break;
-      case PIPE_LOGICOP_SET:
-         for (j = 0; j < 4; j++)
-            res4[j] = ~0;
-         break;
-      default:
-         assert(0);
-      }
+         switch (softpipe->blend->logicop_func) {
+         case PIPE_LOGICOP_CLEAR:
+            for (j = 0; j < 4; j++)
+               res4[j] = 0;
+            break;
+         case PIPE_LOGICOP_NOR:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~(src4[j] | dst4[j]);
+            break;
+         case PIPE_LOGICOP_AND_INVERTED:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~src4[j] & dst4[j];
+            break;
+         case PIPE_LOGICOP_COPY_INVERTED:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~src4[j];
+            break;
+         case PIPE_LOGICOP_AND_REVERSE:
+            for (j = 0; j < 4; j++)
+               res4[j] = src4[j] & ~dst4[j];
+            break;
+         case PIPE_LOGICOP_INVERT:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~dst4[j];
+            break;
+         case PIPE_LOGICOP_XOR:
+            for (j = 0; j < 4; j++)
+               res4[j] = dst4[j] ^ src4[j];
+            break;
+         case PIPE_LOGICOP_NAND:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~(src4[j] & dst4[j]);
+            break;
+         case PIPE_LOGICOP_AND:
+            for (j = 0; j < 4; j++)
+               res4[j] = src4[j] & dst4[j];
+            break;
+         case PIPE_LOGICOP_EQUIV:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~(src4[j] ^ dst4[j]);
+            break;
+         case PIPE_LOGICOP_NOOP:
+            for (j = 0; j < 4; j++)
+               res4[j] = dst4[j];
+            break;
+         case PIPE_LOGICOP_OR_INVERTED:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~src4[j] | dst4[j];
+            break;
+         case PIPE_LOGICOP_COPY:
+            for (j = 0; j < 4; j++)
+               res4[j] = src4[j];
+            break;
+         case PIPE_LOGICOP_OR_REVERSE:
+            for (j = 0; j < 4; j++)
+               res4[j] = src4[j] | ~dst4[j];
+            break;
+         case PIPE_LOGICOP_OR:
+            for (j = 0; j < 4; j++)
+               res4[j] = src4[j] | dst4[j];
+            break;
+         case PIPE_LOGICOP_SET:
+            for (j = 0; j < 4; j++)
+               res4[j] = ~0;
+            break;
+         default:
+            assert(0);
+         }
 
-      for (j = 0; j < 4; j++) {
-         quadColor[j][0] = ubyte_to_float(res[j][0]);
-         quadColor[j][1] = ubyte_to_float(res[j][1]);
-         quadColor[j][2] = ubyte_to_float(res[j][2]);
-         quadColor[j][3] = ubyte_to_float(res[j][3]);
+         for (j = 0; j < 4; j++) {
+            quadColor[j][0] = ubyte_to_float(res[j][0]);
+            quadColor[j][1] = ubyte_to_float(res[j][1]);
+            quadColor[j][2] = ubyte_to_float(res[j][2]);
+            quadColor[j][3] = ubyte_to_float(res[j][3]);
+         }
       }
    }
-
-   /* pass quad to next stage */
-   qs->next->run(qs->next, quad);
 }
 
 
-
-
 static void
-blend_quad(struct quad_stage *qs, struct quad_header *quad)
+blend_quads(struct quad_stage *qs, 
+            struct quad_header *quads[],
+            unsigned nr)
 {
    static const float zero[4] = { 0, 0, 0, 0 };
    static const float one[4] = { 1, 1, 1, 1 };
-
    struct softpipe_context *softpipe = qs->softpipe;
    uint cbuf;
 
-   if (softpipe->blend->logicop_enable) {
-      logicop_quad(qs, quad);
-      return;
-   }
-
    /* loop over colorbuffer outputs */
    for (cbuf = 0; cbuf < softpipe->framebuffer.nr_cbufs; cbuf++) {
       float source[4][QUAD_SIZE], dest[4][QUAD_SIZE];
       struct softpipe_cached_tile *tile
          = sp_get_cached_tile(softpipe->cbuf_cache[cbuf],
-                              quad->input.x0, quad->input.y0);
-      float (*quadColor)[4] = quad->output.color[cbuf];
-      uint i, j;
-
-      /* get/swizzle dest colors */
-      for (j = 0; j < QUAD_SIZE; j++) {
-         int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
-         int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
-         for (i = 0; i < 4; i++) {
-            dest[i][j] = tile->data.color[y][x][i];
+                              quads[0]->input.x0, 
+                              quads[0]->input.y0);
+      uint q, i, j;
+
+      for (q = 0; q < nr; q++) {
+         struct quad_header *quad = quads[q];
+         float (*quadColor)[4] = quad->output.color[cbuf];
+
+         /* get/swizzle dest colors */
+         for (j = 0; j < QUAD_SIZE; j++) {
+            int x = (quad->input.x0 & (TILE_SIZE-1)) + (j & 1);
+            int y = (quad->input.y0 & (TILE_SIZE-1)) + (j >> 1);
+            for (i = 0; i < 4; i++) {
+               dest[i][j] = tile->data.color[y][x][i];
+            }
          }
-      }
 
-      /*
-       * Compute src/first term RGB
-       */
-      switch (softpipe->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:
+         /*
+          * Compute src/first term RGB
+          */
+         switch (softpipe->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 */
@@ -295,12 +297,12 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             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:
+         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 */
@@ -308,7 +310,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], alpha); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
+         case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
          {
             const float *alpha = quadColor[3];
             float diff[4], temp[4];
@@ -319,7 +321,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], temp); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_CONST_COLOR:
+         case PIPE_BLENDFACTOR_CONST_COLOR:
          {
             float comp[4];
             VEC4_SCALAR(comp, softpipe->blend_color.color[0]); /* R */
@@ -330,7 +332,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], comp); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_CONST_ALPHA:
+         case PIPE_BLENDFACTOR_CONST_ALPHA:
          {
             float alpha[4];
             VEC4_SCALAR(alpha, softpipe->blend_color.color[3]);
@@ -339,18 +341,18 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             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:
+         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 */
@@ -361,7 +363,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+         case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
          {
             float inv_alpha[4];
             VEC4_SUB(inv_alpha, one, quadColor[3]);
@@ -370,7 +372,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+         case PIPE_BLENDFACTOR_INV_DST_ALPHA:
          {
             float inv_alpha[4];
             VEC4_SUB(inv_alpha, one, dest[3]);
@@ -379,7 +381,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], inv_alpha); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_DST_COLOR:
+         case PIPE_BLENDFACTOR_INV_DST_COLOR:
          {
             float inv_comp[4];
             VEC4_SUB(inv_comp, one, dest[0]); /* R */
@@ -390,7 +392,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], inv_comp); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+         case PIPE_BLENDFACTOR_INV_CONST_COLOR:
          {
             float inv_comp[4];
             /* R */
@@ -404,7 +406,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[2], quadColor[2], inv_comp);
          }
          break;
-      case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+         case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
          {
             float inv_alpha[4];
             VEC4_SCALAR(inv_alpha, 1.0f - softpipe->blend_color.color[3]);
@@ -413,73 +415,73 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             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);
-      }
+         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 (softpipe->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:
+         /*
+          * Compute src/first term A
+          */
+         switch (softpipe->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:
+         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, softpipe->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:
+         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:
+         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:
+         case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+            /* fall-through */
+         case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
          {
             float inv_comp[4];
             /* A */
@@ -487,42 +489,42 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(source[3], quadColor[3], inv_comp);
          }
          break;
-      default:
-         assert(0);
-      }
+         default:
+            assert(0);
+         }
 
 
-      /*
-       * Compute dest/second term RGB
-       */
-      switch (softpipe->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:
+         /*
+          * Compute dest/second term RGB
+          */
+         switch (softpipe->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, softpipe->blend_color.color[0]); /* R */
@@ -533,7 +535,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(dest[2], dest[2], comp); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_CONST_ALPHA:
+         case PIPE_BLENDFACTOR_CONST_ALPHA:
          {
             float comp[4];
             VEC4_SCALAR(comp, softpipe->blend_color.color[3]); /* A */
@@ -542,17 +544,17 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             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:
+         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 */
@@ -563,7 +565,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
+         case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
          {
             float one_minus_alpha[QUAD_SIZE];
             VEC4_SUB(one_minus_alpha, one, quadColor[3]);
@@ -572,7 +574,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(dest[2], dest[2], one_minus_alpha); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_DST_ALPHA:
+         case PIPE_BLENDFACTOR_INV_DST_ALPHA:
          {
             float inv_comp[4];
             VEC4_SUB(inv_comp, one, dest[3]); /* A */
@@ -581,7 +583,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(dest[2], inv_comp, dest[2]); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_DST_COLOR:
+         case PIPE_BLENDFACTOR_INV_DST_COLOR:
          {
             float inv_comp[4];
             VEC4_SUB(inv_comp, one, dest[0]); /* R */
@@ -592,7 +594,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(dest[2], dest[2], inv_comp); /* B */
          }
          break;
-      case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+         case PIPE_BLENDFACTOR_INV_CONST_COLOR:
          {
             float inv_comp[4];
             /* R */
@@ -606,7 +608,7 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             VEC4_MUL(dest[2], dest[2], inv_comp);
          }
          break;
-      case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
+         case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
          {
             float inv_comp[4];
             VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
@@ -615,138 +617,154 @@ blend_quad(struct quad_stage *qs, struct quad_header *quad)
             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);
-      }
+         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 (softpipe->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:
+         /*
+          * Compute dest/second term A
+          */
+         switch (softpipe->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, softpipe->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:
+         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:
+         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:
+         case PIPE_BLENDFACTOR_INV_CONST_COLOR:
+            /* fall-through */
+         case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
          {
             float inv_comp[4];
             VEC4_SCALAR(inv_comp, 1.0f - softpipe->blend_color.color[3]);
             VEC4_MUL(dest[3], dest[3], inv_comp);
          }
          break;
-      default:
-         assert(0);
-      }
+         default:
+            assert(0);
+         }
 
-      /*
-       * Combine RGB terms
-       */
-      switch (softpipe->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 RGB terms
+          */
+         switch (softpipe->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 (softpipe->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);
+         /*
+          * Combine A terms
+          */
+         switch (softpipe->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);
+         }
       }
-
    } /* cbuf loop */
+}
+
+
+static void
+blend_quad(struct quad_stage *qs, 
+           struct quad_header *quads[],
+           unsigned nr)
+{
+   struct softpipe_context *softpipe = qs->softpipe;
+
+   if (softpipe->blend->logicop_enable) {
+      logicop_quad(qs, quads, nr);
+   }
+   else if (softpipe->blend->blend_enable) {
+      blend_quads(qs, quads, nr );
+   }
 
    /* pass blended quad to next stage */
-   qs->next->run(qs->next, quad);
+   qs->next->run(qs->next, quads, nr);
 }