From 01c1726f7dd8365465ff5ca1c4e430827989702d Mon Sep 17 00:00:00 2001
From: Martin Bauer <martin.bauer@fau.de>
Date: Mon, 18 Jun 2018 13:59:40 +0200
Subject: [PATCH] Bugfixes & Improvements for MPI parallel runs
- fix for field collection and reduction operations
- support for aligned fields in parallel data handling
---
lbstep.py | 19 ++++++++++---------
1 file changed, 10 insertions(+), 9 deletions(-)
diff --git a/lbstep.py b/lbstep.py
index a71bec99..8ce5ed3d 100644
--- a/lbstep.py
+++ b/lbstep.py
@@ -184,17 +184,16 @@ class LatticeBoltzmannStep:
slice_obj = make_slice[:, :] if self.dim == 2 else make_slice[:, :, 0.5]
result = self._data_handling.gather_array(data_name, slice_obj)
- if result is None:
- return
if masked:
mask = self.boundary_handling.get_mask(slice_obj[:self.dim], 'domain', True)
- if len(mask.shape) < len(result.shape):
- assert len(mask.shape) + 1 == len(result.shape)
- mask = np.repeat(mask[..., np.newaxis], result.shape[-1], axis=2)
+ if result is not None:
+ if len(mask.shape) < len(result.shape):
+ assert len(mask.shape) + 1 == len(result.shape)
+ mask = np.repeat(mask[..., np.newaxis], result.shape[-1], axis=2)
- result = np.ma.masked_array(result, mask)
- return result.squeeze()
+ result = np.ma.masked_array(result, mask).squeeze()
+ return result
def velocity_slice(self, slice_obj=None, masked=True):
return self._get_slice(self.velocity_data_name, slice_obj, masked)
@@ -281,10 +280,12 @@ class LatticeBoltzmannStep:
self.time_steps_run += time_steps
- def benchmark_run(self, time_steps):
+ def benchmark_run(self, time_steps, number_of_cells=None):
+ if number_of_cells is None:
+ number_of_cells = self.number_of_cells
time_loop = self.get_time_loop()
duration_of_time_step = time_loop.benchmark_run(time_steps)
- mlups = self.number_of_cells / duration_of_time_step * 1e-6
+ mlups = number_of_cells / duration_of_time_step * 1e-6
self.time_steps_run += time_loop.time_steps_run
return mlups
--
GitLab