gallium: Force blend color to 16-byte alignment

This aligns the 4-element color float array to 16 byte boundaries.  This
should allow compiler vectorizers to generate better optimizations.
Also fixes broken vectorization generated by Intel compiler.

v2: Fixed indentation and added a lengthy comment explaining the
    reason for the alignment.

Cc: <[email protected]>
Reported-by: Tim Rowley <[email protected]>
Tested-by: Tim Rowley <[email protected]>
Signed-off-by: Chuck Atkins <[email protected]>
Acked-by: Roland Scheidegger <[email protected]>

1 files changed, 11 insertions, 1 deletions

diff --git a/src/gallium/include/pipe/p_state.h b/src/gallium/include/pipe/p_state.h
index 1543e90972e..5526c392aa7 100644
--- a/src/gallium/include/pipe/p_state.h
+++ b/src/gallium/include/pipe/p_state.h
@@ -326,7 +326,17 @@ struct pipe_blend_state
 
 struct pipe_blend_color
 {
-   float color[4];
+   /**
+    * Making the color array explicitly 16-byte aligned provides a hint to
+    * compilers to make more efficient auto-vectorization optimizations.
+    * The actual performance gains from vectorizing the blend color array are
+    * fairly minimal, if any, but the alignment is necessary to work around
+    * buggy vectorization in some compilers which fail to generate the correct
+    * unaligned accessors resulting in a segfault.  Specifically several
+    * versions of the Intel compiler are known to be affected but it's likely
+    * others are as well.
+    */
+   PIPE_ALIGN_VAR(16) float color[4];
 };