summaryrefslogtreecommitdiffstats
path: root/Tools
diff options
context:
space:
mode:
authorBenjamin Peterson <benjamin@python.org>2009-02-26 18:55:48 (GMT)
committerBenjamin Peterson <benjamin@python.org>2009-02-26 18:55:48 (GMT)
commitefb06b0d91a6f7f4a34bd2dc46f4afbffdf49e0e (patch)
tree5ab5f2a98bec4f2f164928aace5e47a7cc810736 /Tools
parente3a29806449132490029fafa4cc1349d9d4077a8 (diff)
downloadcpython-efb06b0d91a6f7f4a34bd2dc46f4afbffdf49e0e.zip
cpython-efb06b0d91a6f7f4a34bd2dc46f4afbffdf49e0e.tar.gz
cpython-efb06b0d91a6f7f4a34bd2dc46f4afbffdf49e0e.tar.bz2
Merged revisions 69811,69947 via svnmerge from
svn+ssh://pythondev@svn.python.org/python/trunk ........ r69811 | collin.winter | 2009-02-20 13:30:41 -0600 (Fri, 20 Feb 2009) | 2 lines Issue 5176: special-case string formatting in BINARY_MODULO implementation. This shows a modest (1-3%) speed-up in templating systems, for example. ........ r69947 | jeffrey.yasskin | 2009-02-24 16:48:34 -0600 (Tue, 24 Feb 2009) | 3 lines Tools/scripts/analyze_dxp.py, a module with some helper functions to analyze the output of sys.getdxp(). ........
Diffstat (limited to 'Tools')
-rw-r--r--Tools/scripts/README1
-rw-r--r--Tools/scripts/analyze_dxp.py129
2 files changed, 130 insertions, 0 deletions
diff --git a/Tools/scripts/README b/Tools/scripts/README
index f5c9791..64fc2bd 100644
--- a/Tools/scripts/README
+++ b/Tools/scripts/README
@@ -4,6 +4,7 @@ dutree or lll) are also generally useful UNIX tools.
See also the Demo/scripts directory!
+analyze_dxp.py Analyzes the result of sys.getdxp()
byext.py Print lines/words/chars stats of files by extension
byteyears.py Print product of a file's size and age
checkappend.py Search for multi-argument .append() calls
diff --git a/Tools/scripts/analyze_dxp.py b/Tools/scripts/analyze_dxp.py
new file mode 100644
index 0000000..bde931e
--- /dev/null
+++ b/Tools/scripts/analyze_dxp.py
@@ -0,0 +1,129 @@
+"""
+Some helper functions to analyze the output of sys.getdxp() (which is
+only available if Python was built with -DDYNAMIC_EXECUTION_PROFILE).
+These will tell you which opcodes have been executed most frequently
+in the current process, and, if Python was also built with -DDXPAIRS,
+will tell you which instruction _pairs_ were executed most frequently,
+which may help in choosing new instructions.
+
+If Python was built without -DDYNAMIC_EXECUTION_PROFILE, importing
+this module will raise a RuntimeError.
+
+If you're running a script you want to profile, a simple way to get
+the common pairs is:
+
+$ PYTHONPATH=$PYTHONPATH:<python_srcdir>/Tools/scripts \
+./python -i -O the_script.py --args
+...
+> from analyze_dxp import *
+> s = render_common_pairs()
+> open('/tmp/some_file', 'w').write(s)
+"""
+
+import copy
+import opcode
+import operator
+import sys
+import threading
+
+if not hasattr(sys, "getdxp"):
+ raise RuntimeError("Can't import analyze_dxp: Python built without"
+ " -DDYNAMIC_EXECUTION_PROFILE.")
+
+
+_profile_lock = threading.RLock()
+_cumulative_profile = sys.getdxp()
+
+# If Python was built with -DDXPAIRS, sys.getdxp() returns a list of
+# lists of ints. Otherwise it returns just a list of ints.
+def has_pairs(profile):
+ """Returns True if the Python that produced the argument profile
+ was built with -DDXPAIRS."""
+
+ return len(profile) > 0 and isinstance(profile[0], list)
+
+
+def reset_profile():
+ """Forgets any execution profile that has been gathered so far."""
+ with _profile_lock:
+ sys.getdxp() # Resets the internal profile
+ global _cumulative_profile
+ _cumulative_profile = sys.getdxp() # 0s out our copy.
+
+
+def merge_profile():
+ """Reads sys.getdxp() and merges it into this module's cached copy.
+
+ We need this because sys.getdxp() 0s itself every time it's called."""
+
+ with _profile_lock:
+ new_profile = sys.getdxp()
+ if has_pairs(new_profile):
+ for first_inst in range(len(_cumulative_profile)):
+ for second_inst in range(len(_cumulative_profile[first_inst])):
+ _cumulative_profile[first_inst][second_inst] += (
+ new_profile[first_inst][second_inst])
+ else:
+ for inst in range(len(_cumulative_profile)):
+ _cumulative_profile[inst] += new_profile[inst]
+
+
+def snapshot_profile():
+ """Returns the cumulative execution profile until this call."""
+ with _profile_lock:
+ merge_profile()
+ return copy.deepcopy(_cumulative_profile)
+
+
+def common_instructions(profile):
+ """Returns the most common opcodes in order of descending frequency.
+
+ The result is a list of tuples of the form
+ (opcode, opname, # of occurrences)
+
+ """
+ if has_pairs(profile) and profile:
+ inst_list = profile[-1]
+ else:
+ inst_list = profile
+ result = [(op, opcode.opname[op], count)
+ for op, count in enumerate(inst_list)
+ if count > 0]
+ result.sort(key=operator.itemgetter(2), reverse=True)
+ return result
+
+
+def common_pairs(profile):
+ """Returns the most common opcode pairs in order of descending frequency.
+
+ The result is a list of tuples of the form
+ ((1st opcode, 2nd opcode),
+ (1st opname, 2nd opname),
+ # of occurrences of the pair)
+
+ """
+ if not has_pairs(profile):
+ return []
+ result = [((op1, op2), (opcode.opname[op1], opcode.opname[op2]), count)
+ # Drop the row of single-op profiles with [:-1]
+ for op1, op1profile in enumerate(profile[:-1])
+ for op2, count in enumerate(op1profile)
+ if count > 0]
+ result.sort(key=operator.itemgetter(2), reverse=True)
+ return result
+
+
+def render_common_pairs(profile=None):
+ """Renders the most common opcode pairs to a string in order of
+ descending frequency.
+
+ The result is a series of lines of the form:
+ # of occurrences: ('1st opname', '2nd opname')
+
+ """
+ if profile is None:
+ profile = snapshot_profile()
+ def seq():
+ for _, ops, count in common_pairs(profile):
+ yield "%s: %s\n" % (count, ops)
+ return ''.join(seq())