Author: David Schneider <[email protected]>
Branch: extradoc
Changeset: r4357:d35d75773797
Date: 2012-07-25 12:20 +0200
http://bitbucket.org/pypy/extradoc/changeset/d35d75773797/
Log: add code to build backend data tables about machine code and guard
data sizes and add the latest results
diff --git a/talk/vmil2012/Makefile b/talk/vmil2012/Makefile
--- a/talk/vmil2012/Makefile
+++ b/talk/vmil2012/Makefile
@@ -1,5 +1,5 @@
-jit-guards.pdf: paper.tex paper.bib figures/log.tex figures/example.tex
figures/benchmarks_table.tex
+jit-guards.pdf: paper.tex paper.bib figures/log.tex figures/example.tex
figures/benchmarks_table.tex figures/backend_table.tex
pdflatex paper
bibtex paper
pdflatex paper
@@ -18,15 +18,18 @@
%.tex: %.py
pygmentize -l python -o $@ $<
-figures/benchmarks_table.tex: tool/build_tables.py logs/summary.csv
tool/table_template.tex
+figures/%_table.tex: tool/build_tables.py logs/backend_summary.csv
logs/summary.csv tool/table_template.tex
tool/setup.sh
- paper_env/bin/python tool/build_tables.py logs/summary.csv
tool/table_template.tex figures/benchmarks_table.tex
+ paper_env/bin/python tool/build_tables.py $@
logs/logbench*:;
logs/summary.csv: logs/logbench* tool/difflogs.py
@if ls logs/logbench* &> /dev/null; then python tool/difflogs.py
--diffall logs; fi
+logs/backend_summary.csv: logs/logbench* tool/backenddata.py
+ @if ls logs/logbench* &> /dev/null; then python tool/backenddata.py
logs; fi
+
logs::
tool/run_benchmarks.sh
diff --git a/talk/vmil2012/logs/backend_summary.csv
b/talk/vmil2012/logs/backend_summary.csv
new file mode 100644
--- /dev/null
+++ b/talk/vmil2012/logs/backend_summary.csv
@@ -0,0 +1,12 @@
+exe,bench,asm size,guard map size
+pypy-c,chaos,154,24
+pypy-c,crypto_pyaes,167,24
+pypy-c,django,220,47
+pypy-c,go,4802,874
+pypy-c,pyflate-fast,719,150
+pypy-c,raytrace-simple,486,75
+pypy-c,richards,153,17
+pypy-c,spambayes,2502,337
+pypy-c,sympy_expand,918,211
+pypy-c,telco,506,77
+pypy-c,twisted_names,1604,211
diff --git a/talk/vmil2012/paper.tex b/talk/vmil2012/paper.tex
--- a/talk/vmil2012/paper.tex
+++ b/talk/vmil2012/paper.tex
@@ -456,6 +456,7 @@
\label{sec:evaluation}
\include{figures/benchmarks_table}
+\include{figures/backend_table}
* Evaluation
* Measure guard memory consumption and machine code size
diff --git a/talk/vmil2012/tool/backenddata.py
b/talk/vmil2012/tool/backenddata.py
new file mode 100644
--- /dev/null
+++ b/talk/vmil2012/tool/backenddata.py
@@ -0,0 +1,93 @@
+#!/usr/bin/env python
+"""
+Parse and summarize the traces produced by pypy-c-jit when PYPYLOG is set.
+only works for logs when unrolling is disabled
+"""
+
+import csv
+import optparse
+import os
+import re
+import sys
+from pypy.jit.metainterp.history import ConstInt
+from pypy.jit.tool.oparser import parse
+from pypy.rpython.lltypesystem import llmemory, lltype
+from pypy.tool import logparser
+
+
+def collect_logfiles(path):
+ if not os.path.isdir(path):
+ logs = [os.path.basename(path)]
+ else:
+ logs = os.listdir(path)
+ all = []
+ for log in logs:
+ parts = log.split(".")
+ if len(parts) != 3:
+ continue
+ l, exe, bench = parts
+ if l != "logbench":
+ continue
+ all.append((exe, bench, log))
+ all.sort()
+ return all
+
+
+def collect_guard_data(log):
+ """Calculate the total size in bytes of the locations maps for all guards
+ in a logfile"""
+ guards = logparser.extract_category(log, 'jit-backend-guard-size')
+ return sum(int(x[6:]) for x in guards if x.startswith('chars'))
+
+
+def collect_asm_size(log, guard_size=0):
+ """Calculate the size of the machine code pieces of a logfile. If
+ guard_size is passed it is substracted from result under the assumption
+ that the guard location maps are encoded in the instruction stream"""
+ asm = logparser.extract_category(log, 'jit-backend-dump')
+ asmlen = 0
+ for block in asm:
+ expr = re.compile("CODE_DUMP @\w+ \+\d+\s+(.*$)")
+ match = expr.search(block)
+ assert match is not None # no match found
+ code = match.group(1)
+ asmlen += len(code)
+ return asmlen - guard_size
+
+
+def collect_data(dirname, logs):
+ for exe, name, log in logs:
+ path = os.path.join(dirname, log)
+ logfile = logparser.parse_log_file(path)
+ guard_size = collect_guard_data(logfile)
+ asm_size = collect_asm_size(logfile, guard_size)
+ yield (exe, name, log, asm_size, guard_size)
+
+
+def main(path):
+ logs = collect_logfiles(path)
+ if os.path.isdir(path):
+ dirname = path
+ else:
+ dirname = os.path.dirname(path)
+ results = collect_data(dirname, logs)
+
+ with file("logs/backend_summary.csv", "w") as f:
+ csv_writer = csv.writer(f)
+ row = ["exe", "bench", "asm size", "guard map size"]
+ csv_writer.writerow(row)
+ print row
+ for exe, bench, log, asm_size, guard_size in results:
+ row = [exe, bench, asm_size / 1024, guard_size / 1024]
+ csv_writer.writerow(row)
+ print row
+
+if __name__ == '__main__':
+ parser = optparse.OptionParser(usage="%prog logdir_or_file")
+
+ options, args = parser.parse_args()
+ if len(args) != 1:
+ parser.print_help()
+ sys.exit(2)
+ else:
+ main(args[0])
diff --git a/talk/vmil2012/tool/build_tables.py
b/talk/vmil2012/tool/build_tables.py
--- a/talk/vmil2012/tool/build_tables.py
+++ b/talk/vmil2012/tool/build_tables.py
@@ -2,17 +2,21 @@
import csv
import django
from django.template import Template, Context
-import optparse
-from os import path
+import os
import sys
-#
+# This line is required for Django configuration
+django.conf.settings.configure()
-def main(csvfile, template, texfile):
+def getlines(csvfile):
with open(csvfile, 'rb') as f:
reader = csv.DictReader(f, delimiter=',')
- lines = [l for l in reader]
+ return [l for l in reader]
+
+
+def build_ops_count_table(csvfile, texfile, template):
+ lines = getlines(csvfile)
head = ['Benchmark',
'ops b/o',
@@ -20,7 +24,7 @@
'ops a/o',
'\\% guards a/o',
'opt. rate',
- 'guard opt. rate',]
+ 'guard opt. rate']
table = []
# collect data
@@ -33,22 +37,43 @@
res = [
bench['bench'].replace('_', '\\_'),
ops_bo,
- "%.2f (%s)" % (guards_bo / ops_bo * 100, bench['guard
before']),
+ "%.2f (%s)" % (guards_bo / ops_bo * 100,
+ bench['guard before']),
ops_ao,
- "%.2f (%s)" % (guards_ao / ops_ao * 100, bench['guard after']),
- "%.2f" % ((1 - ops_ao/ops_bo) * 100,),
- "%.2f" % ((1 - guards_ao/guards_bo) * 100,),
+ "%.2f (%s)" % (guards_ao / ops_ao * 100,
+ bench['guard after']),
+ "%.2f" % ((1 - ops_ao / ops_bo) * 100,),
+ "%.2f" % ((1 - guards_ao / guards_bo) * 100,),
]
table.append(res)
output = render_table(template, head, sorted(table))
+ write_table(output, texfile)
+
+
+def build_backend_count_table(csvfile, texfile, template):
+ lines = getlines(csvfile)
+
+ head = ['Benchmark',
+ 'Machine code size (kB)',
+ 'll resume data (kB)']
+
+ table = []
+ # collect data
+ for bench in lines:
+ bench['bench'] = bench['bench'].replace('_', '\\_')
+ keys = ['bench', 'asm size', 'guard map size']
+ table.append([bench[k] for k in keys])
+ output = render_table(template, head, sorted(table))
+ write_table(output, texfile)
+
+
+def write_table(output, texfile):
# Write the output to a file
with open(texfile, 'w') as out_f:
out_f.write(output)
def render_table(ttempl, head, table):
- # This line is required for Django configuration
- django.conf.settings.configure()
# open and read template
with open(ttempl) as f:
t = Template(f.read())
@@ -56,12 +81,25 @@
return t.render(c)
+tables = {
+ 'benchmarks_table.tex':
+ ('summary.csv', build_ops_count_table),
+ 'backend_table.tex':
+ ('backend_summary.csv', build_backend_count_table)
+ }
+
+
+def main(table):
+ tablename = os.path.basename(table)
+ if tablename not in tables:
+ raise AssertionError('unsupported table')
+ data, builder = tables[tablename]
+ csvfile = os.path.join('logs', data)
+ texfile = os.path.join('figures', tablename)
+ template = os.path.join('tool', 'table_template.tex')
+ builder(csvfile, texfile, template)
+
+
if __name__ == '__main__':
- parser = optparse.OptionParser(usage="%prog csvfile template.tex
output.tex")
- options, args = parser.parse_args()
- if len(args) < 3:
- parser.print_help()
- sys.exit(2)
- else:
- main(args[0], args[1], args[2])
-
+ assert len(sys.argv) > 1
+ main(sys.argv[1])
_______________________________________________
pypy-commit mailing list
[email protected]
http://mail.python.org/mailman/listinfo/pypy-commit
