Implement barrier() builtin

Reviewed and Tested-by: Aaron Watry <[email protected]> llvm-svn: 185837
ckissane · Jul 8, 2013 · 3a81b5d · 3a81b5d
1 parent a4cadba
commit 3a81b5d
Show file tree

Hide file tree

Showing 3 changed files with 29 additions and 0 deletions.
diff --git a/libclc/r600/lib/SOURCES b/libclc/r600/lib/SOURCES
@@ -2,3 +2,5 @@ workitem/get_group_id.ll
 workitem/get_local_size.ll
 workitem/get_local_id.ll
 workitem/get_global_size.ll
+synchronization/barrier.cl
+synchronization/barrier_impl.ll
diff --git a/libclc/r600/lib/synchronization/barrier.cl b/libclc/r600/lib/synchronization/barrier.cl
@@ -0,0 +1,15 @@
+
+#include <clc/clc.h>
+
+void barrier_local(void);
+void barrier_global(void);
+
+void barrier(cl_mem_fence_flags flags) {
+  if (flags & CLK_LOCAL_MEM_FENCE) {
+    barrier_local();
+  }
+
+  if (flags & CLK_GLOBAL_MEM_FENCE) {
+    barrier_global();
+  }
+}
diff --git a/libclc/r600/lib/synchronization/barrier_impl.ll b/libclc/r600/lib/synchronization/barrier_impl.ll
@@ -0,0 +1,12 @@
+declare void @llvm.AMDGPU.barrier.local() nounwind
+declare void @llvm.AMDGPU.barrier.global() nounwind
+
+define void @barrier_local() nounwind alwaysinline {
+  call void @llvm.AMDGPU.barrier.local()
+  ret void
+}
+
+define void @barrier_global() nounwind alwaysinline {
+  call void @llvm.AMDGPU.barrier.global()
+  ret void
+}