Add performance section to documentation

.gitignore

@@ -1,6 +1,9 @@
 # build directory
 build/
 
+# python compiled files
+*.pyc
+
 # vim temp files
 .*.sw*
 

doc/compute.qbk

@@ -22,4 +22,5 @@
 [include porting_guide.qbk]
 [include platforms_and_compilers.qbk]
 [include reference.qbk]
+[include performance.qbk]
 [include faq.qbk]

doc/performance.qbk

@@ -0,0 +1,46 @@
+[section:performance Performance]
+
+The following tests were run with an AMD HD 7970 "Tahiti" GPU on a system
+with an Intel i7-4771 3.50GHz CPU.
+
+Source code for the benchmarks can be found under the
+[@https://github.com/kylelutz/compute/tree/master/perf perf] directory. All
+benchmarks were compiled with optimizations enabled (i.e. "gcc -O3").
+
+[h3 Accumulate]
+[$images/perf/accumulate_time_plot.png [width 850px] [align center]]
+
+[h3 Count]
+[$images/perf/count_time_plot.png [width 850px] [align center]]
+
+[h3 Inner Product]
+[$images/perf/inner_product_time_plot.png [width 850px] [align center]]
+
+[h3 Merge]
+[$images/perf/merge_time_plot.png [width 850px] [align center]]
+
+[h3 Partial Sum]
+[$images/perf/partial_sum_time_plot.png [width 850px] [align center]]
+
+[h3 Partition]
+[$images/perf/partition_time_plot.png [width 850px] [align center]]
+
+[h3 Reverse]
+[$images/perf/reverse_time_plot.png [width 850px] [align center]]
+
+[h3 Rotate]
+[$images/perf/rotate_time_plot.png [width 850px] [align center]]
+
+[h3 Set Difference]
+[$images/perf/set_difference_time_plot.png [width 850px] [align center]]
+
+[h3 Sort]
+[$images/perf/sort_time_plot.png [width 850px] [align center]]
+
+[h3 Transform]
+[$images/perf/saxpy_time_plot.png [width 850px] [align center]]
+
+[h3 Unique]
+[$images/perf/unique_time_plot.png [width 850px] [align center]]
+
+[endsect] [/ performance]

perf/perf.py

@@ -5,7 +5,6 @@
 
 import os
 import sys
-import random
 import subprocess
 
 try:
@@ -132,14 +131,15 @@ def run_benchmark(name, sizes, vs=[]):
                 "is_permutation",
                 "merge",
                 "next_permutation",
-                "nth_element"
+                "nth_element",
                 "partial_sum",
                 "partition",
                 "partition_point",
                 "prev_permutation",
                 "reverse",
                 "rotate",
                 "rotate_copy",
+                "saxpy",
                 "search",
                 "search_n",
                 "set_difference",

perf/perfdoc.py

@@ -0,0 +1,62 @@
+#!/usr/bin/python
+
+import os
+import sys
+import pylab
+
+from perf import run_benchmark
+
+fignum = 0
+
+def plot_to_file(report, filename):
+    global fignum
+    fignum += 1
+    pylab.figure(fignum)
+
+    run_to_label = {
+        "stl" : "C++ STL",
+        "compute" : "Boost.Compute"
+    }
+
+    for run in report.samples.keys():
+        x = []
+        y = []
+
+        for sample in report.samples[run]:
+            x.append(sample[0])
+            y.append(sample[1])
+
+        pylab.plot(x, y, marker='o', label=run_to_label[run])
+
+    pylab.xlabel("Size")
+    pylab.ylabel("Time (ms)")
+    pylab.legend(loc='upper left')
+    pylab.savefig(filename)
+
+if __name__ == '__main__':
+    sizes = [pow(2, x) for x in range(10, 26)]
+    algorithms = [
+        "accumulate",
+        "count",
+        "inner_product",
+        "merge",
+        "partial_sum",
+        "partition",
+        "reverse",
+        "rotate",
+        "saxpy",
+        "set_difference",
+        "sort",
+        "unique",
+    ]
+
+    try:
+        os.mkdir("perf_plots")
+    except OSError:
+        pass
+
+    for algorithm in algorithms:
+        print "running '%s'" % (algorithm)
+        report = run_benchmark(algorithm, sizes, ["stl"])
+        plot_to_file(report, "perf_plots/%s_time_plot.png" % algorithm)
+