Add hadd builtin

(x + y) >> 1 gets changed to:
(x>>1) + (y>>1) + (x&y&1)

Saves us having to do any llvm assembly and overflow checking in the addition.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
llvm-svn: 188476

libclc/generic/include/clc/clc.h

@@ -65,6 +65,7 @@
 #include <clc/integer/abs_diff.h>
 #include <clc/integer/add_sat.h>
 #include <clc/integer/clz.h>
+#include <clc/integer/hadd.h>
 #include <clc/integer/mad24.h>
 #include <clc/integer/mul24.h>
 #include <clc/integer/rotate.h>

libclc/generic/include/clc/integer/hadd.h

+#define __CLC_BODY <clc/integer/hadd.inc>
+#include <clc/integer/gentype.inc>

libclc/generic/include/clc/integer/hadd.inc

+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE hadd(__CLC_GENTYPE x, __CLC_GENTYPE y);

libclc/generic/lib/SOURCES

@@ -11,6 +11,7 @@ integer/add_sat_impl.ll
 integer/clz.cl
 integer/clz_if.ll
 integer/clz_impl.ll
+integer/hadd.cl
 integer/mad24.cl
 integer/mul24.cl
 integer/rotate.cl

libclc/generic/lib/integer/hadd.cl

+#include <clc/clc.h>
+
+#define __CLC_BODY <hadd.inc>
+#include <clc/integer/gentype.inc>

libclc/generic/lib/integer/hadd.inc

+//hadd = (x+y)>>1
+//This can be simplified to x>>1 + y>>1 + (1 if both x and y have the 1s bit set)
+//This saves us having to do any checks for overflow in the addition sum
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE hadd(__CLC_GENTYPE x, __CLC_GENTYPE y) {
+    return (x>>(__CLC_GENTYPE)1)+(y>>(__CLC_GENTYPE)1)+(x&y&(__CLC_GENTYPE)1);
+}