Implement async_work_group_copy builtin v3

This is a simple implementation which just copies data synchronously.

v2:
  - Use size_t.

v3:
  - Fix possible race condition by splitting the copy among multiple
    work items.

llvm-svn: 219008

2 files changed, 26 insertions, 0 deletions

diff --git a/libclc/generic/lib/async/async_work_group_copy.cl b/libclc/generic/lib/async/async_work_group_copy.cl
new file mode 100644
index 00000000000..fe20ecfd9bb
--- /dev/null
+++ b/libclc/generic/lib/async/async_work_group_copy.cl
@@ -0,0 +1,9 @@
+#include <clc/clc.h>
+
+#ifdef cl_khr_fp64
+#pragma OPENCL EXTENSION cl_khr_fp64 : enable
+#endif
+
+#define __CLC_BODY <async_work_group_copy.inc>
+#include <clc/async/gentype.inc>
+#undef __CLC_BODY
diff --git a/libclc/generic/lib/async/async_work_group_copy.inc b/libclc/generic/lib/async/async_work_group_copy.inc
new file mode 100644
index 00000000000..a143ddfb9f6
--- /dev/null
+++ b/libclc/generic/lib/async/async_work_group_copy.inc
@@ -0,0 +1,17 @@
+_CLC_OVERLOAD _CLC_DEF event_t async_work_group_copy(
+    local __CLC_GENTYPE *dst,
+    const global __CLC_GENTYPE *src,
+    size_t num_gentypes,
+    event_t event) {
+
+  return async_work_group_strided_copy(dst, src, num_gentypes, 1, event);
+}
+
+_CLC_OVERLOAD _CLC_DEF event_t async_work_group_copy(
+    global __CLC_GENTYPE *dst,
+    const local __CLC_GENTYPE *src,
+    size_t num_gentypes,
+    event_t event) {
+
+  return async_work_group_strided_copy(dst, src, num_gentypes, 1, event);
+}