llvmpipe: Added support for color masks in AoS blending.

Signed-off-by: José Fonseca <jfonseca@vmware.com>

5 files changed, 71 insertions, 33 deletions

diff --git a/src/gallium/auxiliary/util/u_format.h b/src/gallium/auxiliary/util/u_format.h
index bd4e51d279f..1718fb5e240 100644
--- a/src/gallium/auxiliary/util/u_format.h
+++ b/src/gallium/auxiliary/util/u_format.h
@@ -549,6 +549,19 @@ util_format_colormask(const struct util_format_description *desc)
 }
 
 
+/**
+ * Checks if color mask covers every channel for the specified format
+ *
+ * @param desc       a format description to check colormask with
+ * @param colormask  a bit mask for channels, matches format of PIPE_MASK_RGBA
+ */
+static INLINE boolean
+util_format_colormask_full(const struct util_format_description *desc, unsigned colormask)
+{
+   return (~colormask & util_format_colormask(desc)) == 0;
+}
+
+
 boolean
 util_format_is_float(enum pipe_format format);
 
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend.h b/src/gallium/drivers/llvmpipe/lp_bld_blend.h
index f82ae30bb7d..00838a66342 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend.h
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend.h
@@ -64,12 +64,13 @@ lp_build_blend_func(struct lp_build_context *bld,
 LLVMValueRef
 lp_build_blend_aos(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
+                   const enum pipe_format *cbuf_format,
                    struct lp_type type,
                    unsigned rt,
                    LLVMValueRef src,
                    LLVMValueRef dst,
                    LLVMValueRef const_,
-                   unsigned alpha_swizzle);
+                   const unsigned char swizzle[4]);
 
 
 void
diff --git a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
index c342346a36e..678dac81f28 100644
--- a/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
+++ b/src/gallium/drivers/llvmpipe/lp_bld_blend_aos.c
@@ -45,6 +45,7 @@
 
 #include "pipe/p_state.h"
 #include "util/u_debug.h"
+#include "util/u_format.h"
 
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_const.h"
@@ -303,22 +304,20 @@ lp_build_blend_func(struct lp_build_context *bld,
 LLVMValueRef
 lp_build_blend_aos(struct gallivm_state *gallivm,
                    const struct pipe_blend_state *blend,
+                   const enum pipe_format *cbuf_format,
                    struct lp_type type,
                    unsigned rt,
                    LLVMValueRef src,
                    LLVMValueRef dst,
                    LLVMValueRef const_,
-                   unsigned alpha_swizzle)
+                   const unsigned char swizzle[4])
 {
    struct lp_build_blend_aos_context bld;
    LLVMValueRef src_term;
    LLVMValueRef dst_term;
-
-   /* FIXME: color masking not implemented yet */
-   assert(blend->rt[rt].colormask == 0xf);
-
-   if(!blend->rt[rt].blend_enable)
-      return src;
+   LLVMValueRef result;
+   unsigned alpha_swizzle = swizzle[3];
+   boolean fullcolormask;
 
    /* Setup build context */
    memset(&bld, 0, sizeof bld);
@@ -327,30 +326,56 @@ lp_build_blend_aos(struct gallivm_state *gallivm,
    bld.dst = dst;
    bld.const_ = const_;
 
-   /* TODO: There are still a few optimization opportunities here. For certain
-    * combinations it is possible to reorder the operations and therefore saving
-    * some instructions. */
+   if (!blend->rt[rt].blend_enable) {
+      result = src;
+   } else {
+
+      /* TODO: There are still a few optimization opportunities here. For certain
+       * combinations it is possible to reorder the operations and therefore saving
+       * some instructions. */
+
+      src_term = lp_build_blend_factor(&bld, src, blend->rt[rt].rgb_src_factor,
+                                       blend->rt[rt].alpha_src_factor, alpha_swizzle);
+      dst_term = lp_build_blend_factor(&bld, dst, blend->rt[rt].rgb_dst_factor,
+                                       blend->rt[rt].alpha_dst_factor, alpha_swizzle);
+
+      lp_build_name(src_term, "src_term");
+      lp_build_name(dst_term, "dst_term");
+
+      if(blend->rt[rt].rgb_func == blend->rt[rt].alpha_func) {
+         result = lp_build_blend_func(&bld.base, blend->rt[rt].rgb_func, src_term, dst_term);
+      }
+      else {
+         /* Seperate RGB / A functions */
 
-   src_term = lp_build_blend_factor(&bld, src, blend->rt[rt].rgb_src_factor,
-                                    blend->rt[rt].alpha_src_factor, alpha_swizzle);
-   dst_term = lp_build_blend_factor(&bld, dst, blend->rt[rt].rgb_dst_factor,
-                                    blend->rt[rt].alpha_dst_factor, alpha_swizzle);
+         LLVMValueRef rgb;
+         LLVMValueRef alpha;
 
-   lp_build_name(src_term, "src_term");
-   lp_build_name(dst_term, "dst_term");
+         rgb   = lp_build_blend_func(&bld.base, blend->rt[rt].rgb_func,   src_term, dst_term);
+         alpha = lp_build_blend_func(&bld.base, blend->rt[rt].alpha_func, src_term, dst_term);
 
-   if(blend->rt[rt].rgb_func == blend->rt[rt].alpha_func) {
-      return lp_build_blend_func(&bld.base, blend->rt[rt].rgb_func, src_term, dst_term);
+         result = lp_build_blend_swizzle(&bld, rgb, alpha, LP_BUILD_BLEND_SWIZZLE_RGBA, alpha_swizzle);
+      }
    }
-   else {
-      /* Seperate RGB / A functions */
 
-      LLVMValueRef rgb;
-      LLVMValueRef alpha;
+   /* Apply color masking if necessary */
+   fullcolormask = util_format_colormask_full(util_format_description(cbuf_format[rt]), blend->rt[rt].colormask);
 
-      rgb   = lp_build_blend_func(&bld.base, blend->rt[rt].rgb_func,   src_term, dst_term);
-      alpha = lp_build_blend_func(&bld.base, blend->rt[rt].alpha_func, src_term, dst_term);
+   if (!fullcolormask) {
+      LLVMValueRef mask;
+      unsigned mask_swizzle;
 
-      return lp_build_blend_swizzle(&bld, rgb, alpha, LP_BUILD_BLEND_SWIZZLE_RGBA, alpha_swizzle);
+      /* Swizzle the color mask to ensure it matches target format */
+      mask_swizzle =
+               ((blend->rt[rt].colormask & (1 << swizzle[0])) >> swizzle[0])
+            | (((blend->rt[rt].colormask & (1 << swizzle[1])) >> swizzle[1]) << 1)
+            | (((blend->rt[rt].colormask & (1 << swizzle[2])) >> swizzle[2]) << 2)
+            | (((blend->rt[rt].colormask & (1 << swizzle[3])) >> swizzle[3]) << 3);
+
+      mask = lp_build_const_mask_aos(gallivm, bld.base.type, mask_swizzle);
+
+      result = lp_build_select(&bld.base, mask, result, dst);
    }
+
+   return result;
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 482f4e38632..582c6db15d5 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -924,6 +924,7 @@ generate_variant(struct llvmpipe_context *lp,
                  const struct lp_fragment_shader_variant_key *key)
 {
    struct lp_fragment_shader_variant *variant;
+   const struct util_format_description *cbuf0_format_desc;
    boolean fullcolormask;
 
    variant = CALLOC_STRUCT(lp_fragment_shader_variant);
@@ -942,12 +943,8 @@ generate_variant(struct llvmpipe_context *lp,
     */
    fullcolormask = FALSE;
    if (key->nr_cbufs == 1) {
-      const struct util_format_description *format_desc;
-      format_desc = util_format_description(key->cbuf_format[0]);
-      if ((~key->blend.rt[0].colormask &
-           util_format_colormask(format_desc)) == 0) {
-         fullcolormask = TRUE;
-      }
+      cbuf0_format_desc = util_format_description(key->cbuf_format[0]);
+      fullcolormask = util_format_colormask_full(cbuf0_format_desc, key->blend.rt[0].colormask);
    }
 
    variant->opaque =
diff --git a/src/gallium/drivers/llvmpipe/lp_test_blend.c b/src/gallium/drivers/llvmpipe/lp_test_blend.c
index b3ca134131d..af66ac388f7 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_blend.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_blend.c
@@ -179,7 +179,9 @@ add_blend_test(struct gallivm_state *gallivm,
    LLVMValueRef res_ptr;
    LLVMBasicBlockRef block;
    LLVMBuilderRef builder;
+   const enum pipe_format format = PIPE_FORMAT_R8G8B8A8_UNORM;
    const unsigned rt = 0;
+   const unsigned char swizzle[4] = { 0, 1, 2, 3 };
 
    vec_type = lp_build_vec_type(gallivm, type);
 
@@ -205,7 +207,7 @@ add_blend_test(struct gallivm_state *gallivm,
       dst = LLVMBuildLoad(builder, dst_ptr, "dst");
       con = LLVMBuildLoad(builder, const_ptr, "const");
 
-      res = lp_build_blend_aos(gallivm, blend, type, rt, src, dst, con, 3);
+      res = lp_build_blend_aos(gallivm, blend, &format, type, rt, src, dst, con, swizzle);
 
       lp_build_name(res, "res");