diff -r 9923b81a1d34 perf.py
--- a/perf.py	Wed Feb 03 14:19:18 2016 -0600
+++ b/perf.py	Mon Sep 12 11:04:21 2016 -0700
@@ -818,6 +818,8 @@
     try:
         changed_data = benchmark_function(changed_python, options,
                                           *args, **kwargs)
+        if options.testonly:
+            return
         if options.raw:
             return FormatRawData(changed_data, options)
         base_data = benchmark_function(base_python, options,
@@ -1215,9 +1217,9 @@
                    extra_args=[], iteration_scaling=1):
     """Abstract measurement function for Unladen's bm_* scripts.
 
-    Based on the values of options.fast/rigorous, will pass -n {5,50,100} to
-    the benchmark script. MeasureGeneric takes care of parsing out the running
-    times from the memory usage data.
+    Based on the values of options.testonly/fast/rigorous, will pass -n
+    {1,5,50,100} to the benchmark script. MeasureGeneric takes care of
+    parsing out the running times from the memory usage data.
 
     Args:
         python: start of the argv list for running Python.
@@ -1237,7 +1239,10 @@
         bm_env = {}
 
     trials = 50
-    if options.rigorous:
+    if options.testonly:
+        trials = 1
+        iteration_scaling = min(1, iteration_scaling)
+    elif options.rigorous:
         trials = 100
     elif options.fast:
         trials = 5
@@ -1323,6 +1328,16 @@
     PYBENCH_PATH = Relative("performance/pybench/pybench.py")
     PYBENCH_ENV = BuildEnv(inherit_env=options.inherit_env)
 
+    if options.testonly:
+        with open(os.devnull, "wb") as dev_null:
+            subprocess.check_call(LogCall(changed_python + [
+                                          PYBENCH_PATH,
+                                          "-n", "1",
+                                          "-C", "0",
+                                  ]), stdout=dev_null,
+                                  env=PYBENCH_ENV)
+        return
+
     try:
         with contextlib.nested(open(os.devnull, "wb"),
                                TemporaryFilename(prefix="baseline."),
@@ -1435,7 +1450,7 @@
                    inherit_env=options.inherit_env)
 
     # This can be compressed, but it's harder to understand.
-    if options.fast:
+    if options.testonly or options.fast:
         trials = 1
         target = fast_target
     elif options.rigorous:
@@ -1461,7 +1476,9 @@
     hg_env = BuildEnv({"PYTHONPATH": hg_path}, options.inherit_env)
 
     trials = 500
-    if options.rigorous:
+    if options.testonly:
+        trials = 1
+    elif options.rigorous:
         trials = 1000
     elif options.fast:
         trials = 100
@@ -1480,7 +1497,9 @@
     bzr_env = BuildEnv({"PYTHONPATH": bzr_path}, options.inherit_env)
 
     trials = 100
-    if options.rigorous:
+    if options.testonly:
+        trials = 1
+    elif options.rigorous:
         trials = 200
     elif options.fast:
         trials = 10
@@ -1694,6 +1713,8 @@
     try:
         changed_data = MeasureSpitfire(changed_python, options,
                                        spitfire_env, extra_args)
+        if options.testonly:
+            return
         if options.raw:
             return FormatRawData(changed_data, options)
         base_data = MeasureSpitfire(base_python, options,
@@ -2090,7 +2111,9 @@
 
 @VersionRange()
 def BM_normal_startup(base_python, changed_python, options):
-    if options.rigorous:
+    if options.testonly:
+        num_loops = 1
+    elif options.rigorous:
         num_loops = 100
     elif options.fast:
         num_loops = 5
@@ -2100,6 +2123,8 @@
     opts = []
     changed_data = MeasureStartup(changed_python, opts, num_loops,
                                   options.track_memory, options.inherit_env)
+    if options.testonly:
+        return
     if options.raw:
         return FormatRawData(changed_data, options)
     base_data = MeasureStartup(base_python, opts, num_loops,
@@ -2109,7 +2134,9 @@
 
 @VersionRange()
 def BM_startup_nosite(base_python, changed_python, options):
-    if options.rigorous:
+    if options.testonly:
+        num_loops = 1
+    elif options.rigorous:
         num_loops = 200
     elif options.fast:
         num_loops = 10
@@ -2119,6 +2146,8 @@
     opts = ["-S"]
     changed_data = MeasureStartup(changed_python, opts, num_loops,
                                   options.track_memory, options.inherit_env)
+    if options.testonly:
+        return
     if options.raw:
         return FormatRawData(changed_data, options)
     base_data = MeasureStartup(base_python, opts, num_loops,
@@ -2310,7 +2339,7 @@
                       options.inherit_env)
 
     trials = 5
-    if options.fast:
+    if options.testonly or options.fast:
         trials = 1
     elif options.rigorous:
         trials = 10
@@ -2570,6 +2599,9 @@
                             " accurate results"))
     parser.add_option("-f", "--fast", action="store_true",
                       help="Get rough answers quickly")
+    parser.add_option("--testonly", action="store_true",
+                      help="Run each benchmark only once. Provide only "
+                           "baseline_python, not changed_python")
     parser.add_option("-v", "--verbose", action="store_true",
                       help="Print more output")
     parser.add_option("-m", "--track_memory", action="store_true",
@@ -2636,7 +2668,7 @@
 
 
     options, args = parser.parse_args(argv)
-    expected = 1 if options.raw else 2
+    expected = 1 if options.raw or options.testonly else 2
     if len(args) != expected:
         parser.error("incorrect number of arguments")
     if expected == 1:
@@ -2652,16 +2684,19 @@
         options.experiment_label = options.changed_binary
 
     base_args, changed_args = ParsePythonArgsOption(options.args)
-    if options.raw:
+    if options.raw or options.testonly:
         if base_args != changed_args:
-            parser.error('provide args for only one interpreter in raw mode')
+            parser.error('provide args for only one interpreter '
+                         'in raw or testonly mode')
         if options.track_memory:
             # XXX this might be worth fixing someday?
-            parser.error('raw mode is not compatible with memory tracking')
+            parser.error('raw and testonly modes are not compatible with '
+                         'memory tracking')
         if options.diff_instrumentation:
-            parser.error('raw mode is not compatible with instrumentation')
+            parser.error('raw and testonly modes are not compatible with '
+                         'instrumentation')
         if options.csv:
-            parser.error('raw mode does not support csv output')
+            parser.error('raw and testonly modes do not support csv output')
     base_cmd_prefix = [base] + base_args
     changed_cmd_prefix = [changed] + changed_args
 
@@ -2714,6 +2749,9 @@
     print("Report on %s" % " ".join(platform.uname()))
     if multiprocessing:
         print("Total CPU cores:", multiprocessing.cpu_count())
+    if options.testonly:
+        print("Successfully ran benchmarks: %s" % " ".join(to_run))
+        return
     if options.raw:
         for name, result in results:
             print()
diff -r 9923b81a1d34 performance/pybench/pybench.py
--- a/performance/pybench/pybench.py	Wed Feb 03 14:19:18 2016 -0600
+++ b/performance/pybench/pybench.py	Mon Sep 12 11:04:21 2016 -0700
@@ -220,10 +220,11 @@
 
         # Set parameters
         if warp is not None:
-            self.rounds = int(self.rounds / warp)
-            if self.rounds == 0:
-                raise ValueError('warp factor set too high')
+            rounds = int(self.rounds / warp)
+            if rounds == 0:
+                raise ValueError('warp factor set too high (%s vs %s)' % (warp, self.rounds))
             self.warp = warp
+            self.rounds = rounds
         if calibration_runs is not None:
             if (not ALLOW_SKIPPING_CALIBRATION and
                 calibration_runs < 1):