Draft: Develop

This MR adds two features to pystencils. First, the base pointer specification is revealed to the user which allows producing kernels with less register usage. Second, the summands insider the summation printer are printer recursively now which allows for more parallelism inside a single core.

pystencils/backends/cbackend.py

@@ -627,7 +627,23 @@ class VectorizedCustomSympyPrinter(CustomSympyPrinter):
             result = self.instruction_set['|'].format(result, item)
         return result
 
-    def _print_Add(self, expr, order=None):
+    def _print_Add(self, expr):
+        # recursivley add the summands to increase the level of parallelism on the core level
+        def visit(summands):
+            if len(summands) == 2:
+                sign = summands[0].sign * summands[1].sign
+                func = self.instruction_set['-' + suffix] if sign == -1 else self.instruction_set['+' + suffix]
+                return func.format(summands[0].term, summands[1].term)
+            else:
+                elements = len(summands) // 2
+                if len(summands[:elements]) < 2:
+                    func = self.instruction_set['-' + suffix] \
+                        if summands[0].sign == -1 else self.instruction_set['+' + suffix]
+                    return func.format(summands[0].term, visit(summands[elements:]))
+                else:
+                    func = self.instruction_set['+' + suffix]
+                    return func.format(visit(summands[:elements]), visit(summands[elements:]))
+
         try:
             result = self._scalarFallback('_print_Add', expr)
         except Exception:
@@ -662,10 +678,8 @@ class VectorizedCustomSympyPrinter(CustomSympyPrinter):
             summands.insert(0, self.SummandInfo(1, "0"))
 
         assert len(summands) >= 2
-        processed = summands[0].term
-        for summand in summands[1:]:
-            func = self.instruction_set['-' + suffix] if summand.sign == -1 else self.instruction_set['+' + suffix]
-            processed = func.format(processed, summand.term)
+
+        processed = visit(summands)
         return processed
 
     def _print_Pow(self, expr):