cell: propogate blend color to SPUs for the fallback fragment ops code

6 files changed, 74 insertions, 9 deletions

diff --git a/src/gallium/drivers/cell/common.h b/src/gallium/drivers/cell/common.h
index b0169b8e329..3b5a25e1656 100644
--- a/src/gallium/drivers/cell/common.h
+++ b/src/gallium/drivers/cell/common.h
@@ -118,12 +118,16 @@
 
 /**
  * Command to specify per-fragment operations state and generated code.
+ * Note that the dsa, blend, blend_color fields are really only needed
+ * for the fallback/C per-pixel code.  They're not used when we generate
+ * dynamic SPU fragment code (which is the normal case).
  */
 struct cell_command_fragment_ops
 {
    uint64_t opcode;      /**< CELL_CMD_STATE_FRAGMENT_OPS */
    struct pipe_depth_stencil_alpha_state dsa;
    struct pipe_blend_state blend;
+   struct pipe_blend_color blend_color;
    unsigned code[SPU_MAX_FRAGMENT_OPS_INSTS];
 };
 
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index 80a9b3d7e13..1fcf03c2b82 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -74,6 +74,7 @@ struct cell_fragment_shader_state
 struct cell_fragment_ops_key
 {
    struct pipe_blend_state blend;
+   struct pipe_blend_color blend_color;
    struct pipe_depth_stencil_alpha_state dsa;
    enum pipe_format color_format;
    enum pipe_format zs_format;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index bb694aa1073..d2427584ba4 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -52,6 +52,7 @@ lookup_fragment_ops(struct cell_context *cell)
     */
    memset(&key, 0, sizeof(key));
    key.blend = *cell->blend;
+   key.blend_color = cell->blend_color;
    key.dsa = *cell->depth_stencil;
 
    if (cell->framebuffer.cbufs[0])
diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c
index a07b312111e..b521c3aecfd 100644
--- a/src/gallium/drivers/cell/spu/spu_command.c
+++ b/src/gallium/drivers/cell/spu/spu_command.c
@@ -195,6 +195,7 @@ cmd_state_fragment_ops(const struct cell_command_fragment_ops *fops)
    /* Copy state info (for fallback case only) */
    memcpy(&spu.depth_stencil_alpha, &fops->dsa, sizeof(fops->dsa));
    memcpy(&spu.blend, &fops->blend, sizeof(fops->blend));
+   memcpy(&spu.blend_color, &fops->blend_color, sizeof(fops->blend_color));
 
    /* Parity twist!  For now, always use the fallback code by default,
     * only switching to codegen when specifically requested.  This
diff --git a/src/gallium/drivers/cell/spu/spu_main.h b/src/gallium/drivers/cell/spu/spu_main.h
index eff43b870ca..ca72baea8b3 100644
--- a/src/gallium/drivers/cell/spu/spu_main.h
+++ b/src/gallium/drivers/cell/spu/spu_main.h
@@ -145,6 +145,7 @@ struct spu_global
    struct spu_framebuffer fb;
    struct pipe_depth_stencil_alpha_state depth_stencil_alpha;
    struct pipe_blend_state blend;
+   struct pipe_blend_color blend_color;
    struct pipe_sampler_state sampler[PIPE_MAX_SAMPLERS];
    struct spu_texture texture[PIPE_MAX_SAMPLERS];
    struct vertex_info vertex_info;
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
index 9404704abf8..f8ffc704926 100644
--- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
@@ -260,7 +260,7 @@ spu_fallback_fragment_ops(uint x, uint y,
       }
 
       /*
-       * Compute Src RGB terms
+       * Compute Src RGB terms (fragment color * factor)
        */
       switch (spu.blend.rgb_src_factor) {
       case PIPE_BLENDFACTOR_ONE:
@@ -283,13 +283,33 @@ spu_fallback_fragment_ops(uint x, uint y,
          term1g = spu_mul(fragG, fragA);
          term1b = spu_mul(fragB, fragA);
          break;
+      case PIPE_BLENDFACTOR_DST_COLOR:
+         term1r = spu_mul(fragR, fbRGBA[0]);
+         term1g = spu_mul(fragG, fbRGBA[1]);
+         term1b = spu_mul(fragB, fbRGBA[1]);
+         break;
+      case PIPE_BLENDFACTOR_DST_ALPHA:
+         term1r = spu_mul(fragR, fbRGBA[3]);
+         term1g = spu_mul(fragG, fbRGBA[3]);
+         term1b = spu_mul(fragB, fbRGBA[3]);
+         break;
+      case PIPE_BLENDFACTOR_CONST_COLOR:
+         term1r = spu_mul(fragR, spu_splats(spu.blend_color.color[0]));
+         term1g = spu_mul(fragG, spu_splats(spu.blend_color.color[1]));
+         term1b = spu_mul(fragB, spu_splats(spu.blend_color.color[2]));
+         break;
+      case PIPE_BLENDFACTOR_CONST_ALPHA:
+         term1r = spu_mul(fragR, spu_splats(spu.blend_color.color[3]));
+         term1g = spu_mul(fragG, spu_splats(spu.blend_color.color[3]));
+         term1b = spu_mul(fragB, spu_splats(spu.blend_color.color[3]));
+         break;
       /* XXX more cases */
       default:
          ASSERT(0);
       }
 
       /*
-       * Compute Src Alpha term
+       * Compute Src Alpha term (fragment alpha * factor)
        */
       switch (spu.blend.alpha_src_factor) {
       case PIPE_BLENDFACTOR_ONE:
@@ -301,13 +321,23 @@ spu_fallback_fragment_ops(uint x, uint y,
       case PIPE_BLENDFACTOR_SRC_ALPHA:
          term1a = spu_mul(fragA, fragA);
          break;
+      case PIPE_BLENDFACTOR_DST_COLOR:
+         /* fall-through */
+      case PIPE_BLENDFACTOR_DST_ALPHA:
+         term1a = spu_mul(fragA, fbRGBA[3]);
+         break;
+      case PIPE_BLENDFACTOR_CONST_COLOR:
+         /* fall-through */
+      case PIPE_BLENDFACTOR_CONST_ALPHA:
+         term1a = spu_mul(fragR, spu_splats(spu.blend_color.color[3]));
+         break;
       /* XXX more cases */
       default:
          ASSERT(0);
       }
 
       /*
-       * Compute Dest RGB terms
+       * Compute Dest RGB terms (framebuffer color * factor)
        */
       switch (spu.blend.rgb_dst_factor) {
       case PIPE_BLENDFACTOR_ONE:
@@ -337,17 +367,37 @@ spu_fallback_fragment_ops(uint x, uint y,
          term2g = spu_mul(fbRGBA[1], tmp);
          term2b = spu_mul(fbRGBA[2], tmp);
          break;
-      /* XXX more cases */
+      case PIPE_BLENDFACTOR_DST_COLOR:
+         term2r = spu_mul(fbRGBA[0], fbRGBA[0]);
+         term2g = spu_mul(fbRGBA[1], fbRGBA[1]);
+         term2b = spu_mul(fbRGBA[2], fbRGBA[2]);
+         break;
+      case PIPE_BLENDFACTOR_DST_ALPHA:
+         term2r = spu_mul(fbRGBA[0], fbRGBA[3]);
+         term2g = spu_mul(fbRGBA[1], fbRGBA[3]);
+         term2b = spu_mul(fbRGBA[2], fbRGBA[3]);
+         break;
+      case PIPE_BLENDFACTOR_CONST_COLOR:
+         term2r = spu_mul(fbRGBA[0], spu_splats(spu.blend_color.color[0]));
+         term2g = spu_mul(fbRGBA[1], spu_splats(spu.blend_color.color[1]));
+         term2b = spu_mul(fbRGBA[2], spu_splats(spu.blend_color.color[2]));
+         break;
+      case PIPE_BLENDFACTOR_CONST_ALPHA:
+         term2r = spu_mul(fbRGBA[0], spu_splats(spu.blend_color.color[3]));
+         term2g = spu_mul(fbRGBA[1], spu_splats(spu.blend_color.color[3]));
+         term2b = spu_mul(fbRGBA[2], spu_splats(spu.blend_color.color[3]));
+         break;
+       /* XXX more cases */
       default:
          ASSERT(0);
       }
 
       /*
-       * Compute Dest Alpha term
+       * Compute Dest Alpha term (framebuffer alpha * factor)
        */
       switch (spu.blend.alpha_dst_factor) {
       case PIPE_BLENDFACTOR_ONE:
-         term2a = fragA;
+         term2a = fbRGBA[3];
          break;
       case PIPE_BLENDFACTOR_SRC_COLOR:
          term2a = spu_splats(0.0f);
@@ -360,6 +410,16 @@ spu_fallback_fragment_ops(uint x, uint y,
          tmp = spu_sub(one, fragA);
          term2a = spu_mul(fbRGBA[3], tmp);
          break;
+      case PIPE_BLENDFACTOR_DST_COLOR:
+         /* fall-through */
+      case PIPE_BLENDFACTOR_DST_ALPHA:
+         term2a = spu_mul(fbRGBA[3], fbRGBA[3]);
+         break;
+      case PIPE_BLENDFACTOR_CONST_COLOR:
+         /* fall-through */
+      case PIPE_BLENDFACTOR_CONST_ALPHA:
+         term2a = spu_mul(fbRGBA[3], spu_splats(spu.blend_color.color[3]));
+         break;
       /* XXX more cases */
       default:
          ASSERT(0);
@@ -394,9 +454,7 @@ spu_fallback_fragment_ops(uint x, uint y,
          fragG = spu_max(term1g, term2g);
          fragB = spu_max(term1b, term2b);
          break;
-      /* XXX more cases */
       default:
-         printf("unsup 0x%x\n", spu.blend.rgb_func);
          ASSERT(0);
       }
 
@@ -419,7 +477,6 @@ spu_fallback_fragment_ops(uint x, uint y,
       case PIPE_BLEND_MAX:
          fragA = spu_max(term1a, term2a);
          break;
-      /* XXX more cases */
       default:
          ASSERT(0);
       }