PERF: fixed performance degration becasue doesn't slice values array in to_native_types of DateTimeBlock

asv_bench/benchmarks/io/csv.py

@@ -50,6 +50,24 @@ def time_frame_date_formatting(self):
         self.data.to_csv(self.fname, date_format='%Y%m%d')
 
 
+class ToCSVDatetimeBig(BaseIO):
+
+    fname = '__test__.csv'
+    timeout = 1500
+    params = [1000, 10000, 100000]
+    param_names = ['obs']
+
+    def setup(self, obs):
+        d = '2018-11-29'
+        dt = '2018-11-26 11:18:27.0'
+        self.data = DataFrame({'dt': [np.datetime64(dt)] * obs,
+                               'd': [np.datetime64(d)] * obs,
+                               'r': [np.random.uniform()] * obs})
+
+    def time_frame(self, obs):
+        self.data.to_csv(self.fname)
+
+
 class StringIORewind:
 
     def data(self, stringio_object):

doc/source/whatsnew/v0.25.0.rst

@@ -249,6 +249,7 @@ Performance Improvements
 - Improved performance of :meth:`pandas.core.groupby.GroupBy.quantile` (:issue:`20405`)
 - Improved performance of :meth:`read_csv` by faster tokenizing and faster parsing of small float numbers (:issue:`25784`)
 - Improved performance of :meth:`read_csv` by faster parsing of N/A and boolean values (:issue:`25804`)
+- Improved performance of :meth:DataFrame.`to_csv` when write datetime dtype data (:issue:`25708`)
 
 .. _whatsnew_0250.bug_fixes:
 

pandas/core/internals/blocks.py

@@ -2178,6 +2178,7 @@ def to_native_types(self, slicer=None, na_rep=None, date_format=None,
         i8values = self.values.view('i8')
 
         if slicer is not None:
+            values = values[..., slicer]
             i8values = i8values[..., slicer]
 
         from pandas.io.formats.format import _get_format_datetime64_from_values