Series.sum optimization (#585)

* Series.sum optimization

* Add bool sum

* combination of two implementations

* Small fix

Author: 1e-to

sdc/datatypes/hpat_pandas_series_functions.py

@@ -2838,8 +2838,8 @@ def hpat_pandas_series_sum_impl(
             _skipna = skipna
 
         if _skipna:
-            return numpy.nansum(self._data)
-        return numpy.sum(self._data)
+            return numpy_like.nansum(self._data)
+        return numpy_like.sum(self._data)
 
     return hpat_pandas_series_sum_impl
 

sdc/functions/numpy_like.py

@@ -201,6 +201,19 @@ def sdc_isnan_float_impl(self):
     ty_checker.raise_exc(dtype, 'int or float', 'self.dtype')
 
 
+def gen_sum_bool_impl():
+    """Generate sum bool implementation."""
+    def _sum_bool_impl(self):
+        length = len(self)
+        result = 0
+        for i in prange(length):
+            result += self[i]
+
+        return result
+
+    return _sum_bool_impl
+
+
 @sdc_overload(sum)
 def sdc_sum_overload(self):
     """
@@ -230,9 +243,12 @@ def sdc_sum_number_impl(self):
 
         return sdc_sum_number_impl
 
+    if isinstance(dtype, (types.Boolean, bool)):
+        return gen_sum_bool_impl()
+
 
 @sdc_overload(nansum)
-def sdc_sum_overload(self):
+def sdc_nansum_overload(self):
     """
     Intel Scalable Dataframe Compiler Developer Guide
     *************************************************
@@ -258,6 +274,9 @@ def sdc_nansum_number_impl(self):
 
         return sdc_nansum_number_impl
 
+    if isinstance(dtype, (types.Boolean, bool)):
+        return gen_sum_bool_impl()
+
 
 def nanmin(a):
     pass

sdc/tests/test_series.py

@@ -2319,6 +2319,14 @@ def test_impl(S):
         S = pd.Series([1., 2., 3.])
         self.assertEqual(hpat_func(S), test_impl(S))
 
+    def test_series_sum_bool(self):
+        def test_impl(S):
+            return S.sum()
+        hpat_func = self.jit(test_impl)
+
+        S = pd.Series([True, True, False])
+        self.assertEqual(hpat_func(S), test_impl(S))
+
     def test_series_sum_nan(self):
         def test_impl(S):
             return S.sum()