[PATCH 1/3] Implement barrier() builtin

I accidentally sent this series twice, sorry for the spam.

-Tom

From: Tom Stellard <thomas.stellard@amd.com>

---
generic/include/clc/clc.h | 2 ++
generic/include/clc/integer/mad24.h | 3 +++
generic/include/clc/integer/mad24.inc | 1 +
generic/include/clc/integer/mul24.h | 3 +++
generic/include/clc/integer/mul24.inc | 1 +
generic/lib/SOURCES | 2 ++
generic/lib/integer/mad24.cl | 4 ++++
generic/lib/integer/mad24.inc | 3 +++
generic/lib/integer/mul24.cl | 4 ++++
generic/lib/integer/mul24.inc | 11 +++++++++++
10 files changed, 34 insertions(+)
create mode 100644 generic/include/clc/integer/mad24.h
create mode 100644 generic/include/clc/integer/mad24.inc
create mode 100644 generic/include/clc/integer/mul24.h
create mode 100644 generic/include/clc/integer/mul24.inc
create mode 100644 generic/lib/integer/mad24.cl
create mode 100644 generic/lib/integer/mad24.inc
create mode 100644 generic/lib/integer/mul24.cl
create mode 100644 generic/lib/integer/mul24.inc

diff --git a/generic/include/clc/clc.h b/generic/include/clc/clc.h
index 10d30e0..ffa2a24 100644
--- a/generic/include/clc/clc.h
+++ b/generic/include/clc/clc.h
@@ -64,6 +64,8 @@
#include <clc/integer/abs_diff.h>
#include <clc/integer/add_sat.h>
#include <clc/integer/clz.h>
+#include <clc/integer/mad24.h>
+#include <clc/integer/mul24.h>
#include <clc/integer/rotate.h>
#include <clc/integer/sub_sat.h>

diff --git a/generic/include/clc/integer/mad24.h b/generic/include/clc/integer/mad24.h
new file mode 100644
index 0000000..0c120fa
--- /dev/null
+++ b/generic/include/clc/integer/mad24.h
@@ -0,0 +1,3 @@
+#define __CLC_BODY <clc/integer/mad24.inc>
+#include <clc/integer/integer-gentype.inc>
+#undef __CLC_BODY
diff --git a/generic/include/clc/integer/mad24.inc b/generic/include/clc/integer/mad24.inc
new file mode 100644
index 0000000..81fe0c2
--- /dev/null
+++ b/generic/include/clc/integer/mad24.inc
@@ -0,0 +1 @@
+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE mad24(__CLC_GENTYPE x, __CLC_GENTYPE y, __CLC_GENTYPE z);
diff --git a/generic/include/clc/integer/mul24.h b/generic/include/clc/integer/mul24.h
new file mode 100644
index 0000000..4f97098
--- /dev/null
+++ b/generic/include/clc/integer/mul24.h
@@ -0,0 +1,3 @@
+#define __CLC_BODY <clc/integer/mul24.inc>
+#include <clc/integer/integer-gentype.inc>
+#undef __CLC_BODY
diff --git a/generic/include/clc/integer/mul24.inc b/generic/include/clc/integer/mul24.inc
new file mode 100644
index 0000000..8cbf7c1
--- /dev/null
+++ b/generic/include/clc/integer/mul24.inc
@@ -0,0 +1 @@
+_CLC_OVERLOAD _CLC_DECL __CLC_GENTYPE mul24(__CLC_GENTYPE x, __CLC_GENTYPE y);
diff --git a/generic/lib/SOURCES b/generic/lib/SOURCES
index 8cda14a..c2da3d7 100644
--- a/generic/lib/SOURCES
+++ b/generic/lib/SOURCES
@@ -11,6 +11,8 @@ integer/add_sat_impl.ll
integer/clz.cl
integer/clz_if.ll
integer/clz_impl.ll
+integer/mad24.cl
+integer/mul24.cl
integer/rotate.cl
integer/sub_sat.cl
integer/sub_sat_if.ll
diff --git a/generic/lib/integer/mad24.cl b/generic/lib/integer/mad24.cl
new file mode 100644
index 0000000..e29e99f
--- /dev/null
+++ b/generic/lib/integer/mad24.cl
@@ -0,0 +1,4 @@
+#include <clc/clc.h>
+
+#define __CLC_BODY <mad24.inc>
+#include <clc/integer/integer-gentype.inc>
diff --git a/generic/lib/integer/mad24.inc b/generic/lib/integer/mad24.inc
new file mode 100644
index 0000000..902b0aa
--- /dev/null
+++ b/generic/lib/integer/mad24.inc
@@ -0,0 +1,3 @@
+_CLC_OVERLOAD _CLC_DEF __CLC_GENTYPE mad24(__CLC_GENTYPE x, __CLC_GENTYPE y, __CLC_GENTYPE z){
+ return mul24(x, y) + z;
+}
diff --git a/generic/lib/integer/mul24.cl b/generic/lib/integer/mul24.cl
new file mode 100644
index 0000000..8aedca6
--- /dev/null
+++ b/generic/lib/integer/mul24.cl
@@ -0,0 +1,4 @@
+#include <clc/clc.h>
+
+#define __CLC_BODY <mul24.inc>
+#include <clc/integer/integer-gentype.inc>
diff --git a/generic/lib/integer/mul24.inc b/generic/lib/integer/mul24.inc
new file mode 100644
index 0000000..c669029
--- /dev/null
+++ b/generic/lib/integer/mul24.inc
@@ -0,0 +1,11 @@
+
+// We need to use shifts here in order to mantain the sign bit for signed
+// integers. The compiler should optimize this to (x & 0x00FFFFFF) for
+// unsgined integers.

s/unsgined/unsigned/

With that fixed:
Reviewed-by: Aaron Watry <awatry@gmail.com>

Reviewed and Tested-by: Aaron Watry <awatry@gmail.com>

Tested on r600g w/ a CEDAR and svn/git head LLVM.

... Eventually I'll learn that reply-all is my friend.