Small call optimizations

Summary: carlosatorres caught this while looking at some ARM perf. For `JITRT_BindKeywordArgs` our memory is already zero-initialized by `std::unique_ptr`. For `JITRT_BindKeywordArgsSimple` we don't need to zero the memory for the values that exist.

Reviewed By: mpage

Differential Revision: D109753053

fbshipit-source-id: c7637385642fce54de647bda07f9e22cddecb106

Author: DinoV

cinderx/Jit/jit_rt.cpp

@@ -156,7 +156,7 @@ static int JITRT_BindKeywordArgs(
   Py_ssize_t argcount = PyVectorcall_NARGS(nargsf);
 
   for (int i = 0; i < arg_space.size(); i++) {
-    arg_space[i] = nullptr;
+    JIT_DCHECK(arg_space[i] == nullptr, "should be initialized");
   }
 
   // Copy all positional arguments into local variables
@@ -245,16 +245,17 @@ static int JITRT_BindKeywordArgsSimple(
     return 0;
   }
 
-  for (int i = 0; i < arg_space.size(); i++) {
-    arg_space[i] = nullptr;
-  }
-
   // Copy all positional arguments into local variables
   Py_ssize_t n = std::min<Py_ssize_t>(argcount, co->co_argcount);
+
   for (Py_ssize_t j = 0; j < n; j++) {
     arg_space[j] = args[j];
   }
 
+  for (int i = n; i < arg_space.size(); i++) {
+    arg_space[i] = nullptr;
+  }
+
   // Check the number of positional arguments
   return JITRT_BindKeywords(args, kwnames, arg_space, argcount, co, nullptr) &&
       JITRT_BindDefaults(argcount, arg_space, co, func);