Added more ranges. changed everything to MB/s. Account for the loop overhead.

4 files changed, 63 insertions, 16 deletions

diff --git a/scripts/bench.py b/scripts/bench.py
index aa0e6fc..e99ab4a 100644
--- a/scripts/bench.py
+++ b/scripts/bench.py
@@ -16,15 +16,19 @@ def run(cache, variant, function, bytes, loops):
         xbuild = build
         cmd = '%(xbuild)s%(variant)s -t %(function)s -c %(bytes)s -l %(loops)s' % locals()
 
-        got = subprocess.check_output(cmd.split()).strip()
-        cache[key] = got
-        print got
+        try:
+            got = subprocess.check_output(cmd.split()).strip()
+            cache[key] = got
+            print got
+        except OSError, ex:
+            assert False, 'Error %s while running %s' % (ex, cmd)
 
     sys.stdout.flush()
 
 
 def run_bytes(cache, variant, function, bytes):
     for b in bytes:
+        # Figure out the number of loops to give a roughly consistent run
         loops = int(500000000/5 / math.sqrt(b))
         run(cache, variant, function, b, loops)
 
@@ -35,8 +39,8 @@ def run_functions(cache, variant, functions, bytes):
 
 
 HAS = {
-    'this': 'strcmp strcpy memchr strchr strlen memcpy memset',
-    'bionc': 'strlen memset memcpy',
+    'this': 'bounce strcmp strcpy memchr strchr strlen memcpy memset',
+    'bionic': 'strlen memset memcpy',
     'glibc': 'memset strlen memcpy  strcmp strcpy memchr strchr',
     'newlib': 'strcmp strlen strcpy',
     'plain': 'memset memcpy strcmp strcpy',
@@ -56,7 +60,11 @@ def run_variants(cache, variants, bytes):
 
 def run_top(cache):
     variants = HAS.keys()
-    bytes = [2**x for x in range(1, 12)]
+    power2 = [2**x for x in range(1, 12)]
+    minus1 = [x-1 for x in power2]
+    various = [int(1.7*x) for x in range(1, 12)]
+
+    bytes = sorted(set(power2 + minus1 + various))
     run_variants(cache, variants, bytes)
 
 
@@ -81,4 +89,5 @@ def main():
             for line in cache.values():
                 print >> f, line
 
-main()
+if __name__ == '__main__':
+    main()
diff --git a/scripts/libplot.py b/scripts/libplot.py
index f83d142..129d79f 100644
--- a/scripts/libplot.py
+++ b/scripts/libplot.py
@@ -1,3 +1,5 @@
+"""Shared routines for the plotters."""
+
 import fileinput
 import collections
 
@@ -5,6 +7,7 @@ Record = collections.namedtuple('Record', 'variant function bytes loops elapsed
 
 
 def parse_value(v):
+    """Turn text into a primitive"""
     try:
         return float(v)
     except ValueError:
@@ -12,9 +15,38 @@ def parse_value(v):
 
 
 def unique(records, name):
+    """Return the unique values of a column in the records"""
     values = set(getattr(x, name) for x in records)
     return sorted(values)
 
 
 def parse():
-    return [Record(*[parse_value(y) for y in x.split(':')]) for x in fileinput.input()]
+    """Parse a record file into named tuples, correcting for loop
+    overhead along the way.
+    """
+    records = [Record(*[parse_value(y) for y in x.split(':')]) for x in fileinput.input()]
+
+    # Pull out any bounce values
+    costs = {}
+
+    for record in [x for x in records if x.function=='bounce']:
+        costs[(record.bytes, record.loops)] = record.elapsed
+
+    # Fix up all of the records for cost
+    out = []
+
+    for record in records:
+        if record.function == 'bounce':
+            continue
+
+        cost = costs.get((record.bytes, record.loops), None)
+
+        if not cost:
+            out.append(record)
+        else:
+            # Unfortunately you can't update a namedtuple...
+            values = list(record)
+            values[-2] -= cost
+            out.append(Record(*values))
+
+    return out
diff --git a/scripts/plot-sizes.py b/scripts/plot-sizes.py
index d1fbdf9..0e05f4e 100644
--- a/scripts/plot-sizes.py
+++ b/scripts/plot-sizes.py
@@ -16,6 +16,7 @@ def plot(records, function):
     bytes = libplot.unique(records, 'bytes')
 
     colours = iter('bgrcmyk')
+    all_x = []
 
     pylab.clf()
 
@@ -26,6 +27,7 @@ def plot(records, function):
         X = [x.bytes for x in matches]
         Y = [x.bytes*x.loops/x.elapsed/(1024*1024) for x in matches]
 
+        all_x.extend(X)
         colour = colours.next()
 
         if X:
@@ -38,7 +40,7 @@ def plot(records, function):
     pylab.xlabel('Size (B)')
     pylab.ylabel('Rate (MB/s)')
     pylab.semilogx()
-    pylab.xlim(0, max(X))
+    pylab.xlim(0, max(all_x))
     pylab.ylim(0, pylab.ylim()[1])
 
 def main():
diff --git a/scripts/plot-top.py b/scripts/plot-top.py
index 02dc571..4cf1965 100644
--- a/scripts/plot-top.py
+++ b/scripts/plot-top.py
@@ -1,7 +1,7 @@
 #!/usr/bin/env python
 
-"""Plot the results of benchmarking different variants of the string
-routines for one size.
+"""Plot the performance of different variants of the string routines
+for one size.
 """
 
 import libplot
@@ -9,10 +9,7 @@ import libplot
 import pylab
 
 
-def main():
-    bytes = 64
-
-    records = libplot.parse()
+def plot(records, bytes):
     records = [x for x in records if x.bytes==bytes]
 
     variants = libplot.unique(records, 'variant')
@@ -31,7 +28,7 @@ def main():
 
             if matches:
                 match = matches[0]
-                heights.append(match.bytes*match.loops/match.elapsed)
+                heights.append(match.bytes*match.loops/match.elapsed/(1024*1024))
             else:
                 heights.append(0)
 
@@ -41,8 +38,15 @@ def main():
     axes.set_xticklabels(functions)
     axes.set_xticks(X + 0.5)
 
+    pylab.title('Performance of different variants for %d byte blocks' % bytes)
+    pylab.ylabel('Rate (MB/s)')
     pylab.legend()
     pylab.grid()
+    pylab.savefig('top-%d.png' % bytes)
+
+def main():
+    records = libplot.parse()
+    plot(records, 64)
     pylab.show()
 
 if __name__ == '__main__':