regent: Emit a faster atomic instruction for doubles when sm >= 60

AlexisPerry · Dec 6, 2018 · 9dde639 · 9dde639
1 parent ce40a0f
commit 9dde639
Showing 1 changed file with 3 additions and 0 deletions.
diff --git a/language/src/regent/cudahelper.t b/language/src/regent/cudahelper.t
@@ -312,6 +312,9 @@ local function generate_atomic(op, typ)
   if op == "+" and typ == float then
     return terralib.intrinsic("llvm.nvvm.atomic.load.add.f32.p0f32",
                               {&float,float} -> {float})
+  elseif op == "+" and typ == double and get_cuda_version() >= 60 then
+    return terralib.intrinsic("llvm.nvvm.atomic.load.add.f64.p0f64",
+                              {&double,double} -> {double})
   end
 
   local cas_type