[cig-commits] r19905 - short/3D/PyLith/benchmarks/trunk/static/performance/solvertest

[brad at geodynamics.org]

Author: brad
Date: 2012-03-29 15:23:48 -0700 (Thu, 29 Mar 2012)
New Revision: 19905

Added:
   short/3D/PyLith/benchmarks/trunk/static/performance/solvertest/plot_scaling.py
Log:
Added initial version of script for plotting scaling. Uses bogus event data from logs, so it will need to be updated to use the staging info.

Added: short/3D/PyLith/benchmarks/trunk/static/performance/solvertest/plot_scaling.py
===================================================================
--- short/3D/PyLith/benchmarks/trunk/static/performance/solvertest/plot_scaling.py	                        (rev 0)
+++ short/3D/PyLith/benchmarks/trunk/static/performance/solvertest/plot_scaling.py	2012-03-29 22:23:48 UTC (rev 19905)
@@ -0,0 +1,102 @@
+#!/usr/bin/env python
+# ----------------------------------------------------------------------
+#
+# Brad T. Aagaard, U.S. Geological Survey
+#
+# ----------------------------------------------------------------------
+#
+# Plot performance for weak scaling.
+#
+# PREREQUISITES: matplotlib, numpy
+#
+# Create subdirectory 'logs' with python log summaries.
+
+import matplotlib.pyplot as pyplot
+import numpy
+import sys
+
+sys.path.append("../../../figures")
+import matplotlibext
+
+nprocs = [1,2,4,8,16,32]
+events = ["Solve",
+          "Reform Jacobian",
+          "Reform Residual",
+          ]
+cells = ["Hex8",
+         "Tet4",
+         ]
+
+symdict = {'Hex8': 's',
+           'Tet4': '^',
+           }
+styledict = {'Setup': ('orange', (None,None)),
+             'Prestep': ('green', (None, None)),
+             'Reform Jacobian': ('blue', (3,1.5)),
+             'Reform Residual': ('purple', (6,1.5)),
+             'Solve': ('red', (None, None)),
+             'Poststep': ('ltblue', (None, None)),
+             }
+             
+
+# Allocate storage for stats
+data = {}
+for c in cells:
+    data[c] = {}
+    for e in events:
+        data[c][e] = numpy.zeros(len(nprocs), dtype=numpy.float32)
+
+# Get stats
+for c in cells:
+    for ip in xrange(len(nprocs)):
+        sys.path.append("logs")
+        log = __import__("%s_np%03d" % (c.lower(), nprocs[ip]))
+        for e in events:
+            total = 0.0
+            eattr = e.replace(' ','_')
+            for sube in log.__getattribute__(eattr).event.values():
+                total += numpy.mean(sube.Time)
+            data[c][e][ip] = total
+
+
+figure = matplotlibext.Figure()
+figure.open(3.0, 3.25, margins=[[0.6, 0, 0.1], [0.5, 0, 0.05]], dpi=150)
+ax = figure.axes(1.35, 1, 1.35, 1)
+
+for c in cells:
+    for e in events:
+        ax.loglog(nprocs, data[c][e], 
+                  marker=symdict[c], 
+                  color=styledict[e][0], 
+                  linewidth=1,
+                  dashes=styledict[e][1])
+        ax.hold(True)
+
+ax.set_xlim((1, 128))
+ax.set_xlabel("# Processors", fontsize=10)
+
+ax.set_ylim((0.01, 500))
+ax.set_ylabel("Time (s)", fontsize=10)
+
+import matplotlib.lines as lines
+proxies = []
+for c in cells:
+    proxies.append(lines.Line2D((0,0),(1,1), 
+                                marker=symdict[c], 
+                                color=styledict['Solve'][0],
+                                linewidth=0))
+l1 = ax.legend(proxies, cells, 
+          loc='lower left', bbox_to_anchor=(0,1.05), borderaxespad=0)
+
+proxies = []
+for e in events:
+    proxies.append(lines.Line2D((0,0),(1,1), 
+                                marker=None, 
+                                color=styledict[e][0],
+                                dashes=styledict[e][1]))
+ax.legend(proxies, events, 
+          loc='lower right', bbox_to_anchor=(1,1.05), borderaxespad=0)
+ax.add_artist(l1)
+
+pyplot.show()
+pyplot.savefig('scaling.pdf')


Property changes on: short/3D/PyLith/benchmarks/trunk/static/performance/solvertest/plot_scaling.py
___________________________________________________________________
Name: svn:executable
   + *