Implement mem_fence on ptx

PTX does not differentiate between read and write fences. Hence, these a
lowered to a mem_fence call. The mem_fence function compiles to the
“member.cta” instruction, which commits all outstanding reads and writes
of a thread such that these become visible to all other threads in the same
CTA (i.e., work-group). The instruction does not differentiate between
global and local memory. Hence, the flags parameter is ignored, except
for deciding whether a “member.cta” instruction should be issued at all.

Reviewed-by: Jan Vesely <jan.vesely@rutgers.edu>
llvm-svn: 315235

libclc/ptx-nvidiacl/lib/SOURCES

@@ -1,3 +1,4 @@
+mem_fence/fence.cl
 synchronization/barrier.cl
 workitem/get_global_id.cl
 workitem/get_group_id.cl

libclc/ptx-nvidiacl/lib/mem_fence/fence.cl

@@ -0,0 +1,15 @@
+#include <clc/clc.h>
+
+_CLC_DEF void mem_fence(cl_mem_fence_flags flags) {
+   if (flags & (CLK_GLOBAL_MEM_FENCE | CLK_LOCAL_MEM_FENCE))
+     __nvvm_membar_cta();
+}
+
+// We do not have separate mechanism for read and write fences.
+_CLC_DEF void read_mem_fence(cl_mem_fence_flags flags) {
+  mem_fence(flags);
+}
+
+_CLC_DEF void write_mem_fence(cl_mem_fence_flags flags) {
+  mem_fence(flags);
+}