| 1 |
#!/usr/bin/env python3 |
| 2 |
# SPDX-License-Identifier: MIT OR MPL-2.0 OR LGPL-2.1-or-later OR GPL-2.0-or-later |
| 3 |
# Copyright 2015, SIL International, All rights reserved. |
| 4 |
''' |
| 5 |
Tool to read callgrind output and return the net cost of the first call |
| 6 |
to a given function |
| 7 |
''' |
| 8 |
|
| 9 |
import argparse |
| 10 |
import json |
| 11 |
import os |
| 12 |
import re |
| 13 |
import tempfile |
| 14 |
from contextlib import suppress |
| 15 |
from multiprocessing import Pool |
| 16 |
from pathlib import Path |
| 17 |
from subprocess import call |
| 18 |
|
| 19 |
p = argparse.ArgumentParser() |
| 20 |
p.add_argument('file', nargs='?', type=argparse.FileType('r'), |
| 21 |
help="control file") |
| 22 |
p.add_argument('-g', '--graph', action='store_true', |
| 23 |
help='Display a graph instead of csv results') |
| 24 |
p.add_argument('-s', '--single', action='store_true', |
| 25 |
help='Do single processing') |
| 26 |
p.add_argument('-f', '--font', |
| 27 |
help='font file to use for a single test') |
| 28 |
p.add_argument('-t', '--text', |
| 29 |
help='text file to use for a single test') |
| 30 |
p.add_argument('-l', '--line', type=int, default=0, |
| 31 |
help='line in text file for single test') |
| 32 |
p.add_argument('-n', '--normalize', action="store_true", |
| 33 |
help="Normalize results") |
| 34 |
p.add_argument('-k', '--keep', |
| 35 |
help='Store intermediate callgrind results in ' |
| 36 |
'given directory') |
| 37 |
p.add_argument('-u', '--usp', type=Path, |
| 38 |
help="Include usp results") |
| 39 |
p.add_argument('-v', '--verbose', action='store_true', |
| 40 |
help='Print the commands executed') |
| 41 |
|
| 42 |
args = p.parse_args() |
| 43 |
|
| 44 |
|
| 45 |
def parsecg(fname): |
| 46 |
idcache = {} |
| 47 |
|
| 48 |
def parseline(line): |
| 49 |
(_, s) = line.split('=', 1) |
| 50 |
s = s.lstrip() |
| 51 |
m = re.match(r'(?:\((\d+)\))?\s*(.*)$', s) |
| 52 |
ident = m.group(1) |
| 53 |
name = m.group(2) |
| 54 |
if ident and name: idcache[ident] = name |
| 55 |
if ident: return (ident, idcache.get(ident, name)) |
| 56 |
else: return (None, name) |
| 57 |
|
| 58 |
fncosts = {} |
| 59 |
with fname.open() as fh: |
| 60 |
incfn = False |
| 61 |
currfn = '' |
| 62 |
for line in fh: |
| 63 |
line = line.strip() |
| 64 |
if incfn: |
| 65 |
if line.startswith('calls'): continue |
| 66 |
cost = line.split(' ')[1] |
| 67 |
fncosts[currfn] = cost |
| 68 |
incfn = False |
| 69 |
elif line.startswith('cfn'): |
| 70 |
(ident, name) = parseline(line) |
| 71 |
incfn = True |
| 72 |
currfn = name |
| 73 |
elif line.startswith('fn'): |
| 74 |
(id, name) = parseline(line) |
| 75 |
return fncosts |
| 76 |
|
| 77 |
|
| 78 |
modes = { |
| 79 |
'ot': ( |
| 80 |
'ot', |
| 81 |
# {'ot': 'hb_shape_plan_execute', |
| 82 |
{'ot': '_hb_ot_shape', |
| 83 |
'ot_GSUB': 'hb_ot_map_t::substitute(hb_ot_shape_plan_t const*, hb_font_t*, hb_buffer_t*) const', |
| 84 |
'ot_GPOS': 'hb_ot_map_t::position(hb_ot_shape_plan_t const*, hb_font_t*, hb_buffer_t*) const'}), |
| 85 |
'gr': ( |
| 86 |
'graphite2', |
| 87 |
# {'hbgr': 'hb_shape_plan_execute', |
| 88 |
{'hbgr': '_hb_graphite2_shape', |
| 89 |
'gr': 'gr_make_seg', |
| 90 |
'grcoll': 'graphite2::Pass::collisionShift(graphite2::Segment*, int, graphite2::json*) const', |
| 91 |
'grkern': 'graphite2::Pass::collisionKern(graphite2::Segment*, int, graphite2::json*) const'}), |
| 92 |
'usp': ( |
| 93 |
'uniscribe', |
| 94 |
{'usp': '0x61829dc6', |
| 95 |
'abc': 'GetCharABCWidthsI', |
| 96 |
'usp_GSUB': '0x6f90e350', |
| 97 |
'usp_GPOS': '0x6f90e660'}, |
| 98 |
{'valgrind_parms': ['--trace-children=yes', '--combine-dumps=yes'], |
| 99 |
'dir': args.usp, |
| 100 |
'exe': ['wine', |
| 101 |
args.usp and args.usp.joinpath('util', 'hb-shape.exe')], |
| 102 |
'env': { |
| 103 |
'WINEPATH': args.usp, |
| 104 |
'WINEDLLPATH': args.usp, |
| 105 |
'WINEDLLOVERRIDES': '"usp10=n"', |
| 106 |
'WINEDEBUG': 'trace+dll'}}, |
| 107 |
{'uspnorm': lambda x: int(x['usp']) - int(x['abc'])} |
| 108 |
) |
| 109 |
} |
| 110 |
|
| 111 |
|
| 112 |
def getval(dat, key, defkey=None): |
| 113 |
if not hasattr(dat, 'keys'): |
| 114 |
return dat |
| 115 |
if defkey is not None: |
| 116 |
return dat.get(key, dat[defkey]) |
| 117 |
else: |
| 118 |
return dat.get(key, None) |
| 119 |
|
| 120 |
|
| 121 |
def dorow(x): |
| 122 |
(c, r) = x |
| 123 |
results = r.copy() |
| 124 |
for k, job in modes.items(): |
| 125 |
if k == 'usp' and not args.usp: |
| 126 |
for l in ('usp', 'abc', 'uspnorm', 'usp_GSUB', 'usp_GPOS'): |
| 127 |
results[l] = 0 |
| 128 |
continue |
| 129 |
resf = tmpd.joinpath('callgrind_'+k+'_'+str(c)+'.out').resolve() |
| 130 |
with open(r['textfile'], encoding='utf_8') as ft: |
| 131 |
t = ft.readlines()[int(r['textline'])].strip() |
| 132 |
tfile = tmpd.joinpath('t'+k+str(c)+'.out').resolve() |
| 133 |
with tfile.open('w') as ft: |
| 134 |
ft.write(t+"\n") |
| 135 |
results['char'] = len(t) |
| 136 |
font = getval(r['font'], k, defkey='ot') |
| 137 |
cmd = ['valgrind', '-q', |
| 138 |
'--tool=callgrind', |
| 139 |
'--callgrind-out-file='+str(resf), |
| 140 |
'hb-shape', |
| 141 |
'--font-file='+font, |
| 142 |
'--shapers='+job[0], |
| 143 |
'--text-file='+str(tfile)] |
| 144 |
cwd = None |
| 145 |
env = os.environ.copy() |
| 146 |
if len(job) > 2: |
| 147 |
cmd = cmd[:4] + job[2]['valgrind_parms'] + job[2]['exe'] + cmd[5:] |
| 148 |
cwd = job[2]['dir'] |
| 149 |
env.update(job[2]['env']) |
| 150 |
# copt = 3 # This is never used |
| 151 |
if 'rtl' in r and r['rtl']: |
| 152 |
cmd.append('--direction=rtl') |
| 153 |
if 'cmdext' in r: |
| 154 |
f = getval(r['cmdext'], k) |
| 155 |
if f is not None: |
| 156 |
cmd.append(f) |
| 157 |
if args.verbose: |
| 158 |
print(" ".join(cmd)) |
| 159 |
|
| 160 |
with suppress(Exception): |
| 161 |
call(cmd, stdout=dumpfh, cwd=cwd) |
| 162 |
|
| 163 |
cfns = parsecg(resf) |
| 164 |
for rk, rf in job[1].items(): |
| 165 |
if 'ignore' in r and k in r['ignore']: |
| 166 |
results[rk] = 0 |
| 167 |
else: |
| 168 |
results[rk] = int(cfns.get(rf, 0)) |
| 169 |
if args.normalize: |
| 170 |
results[rk] /= results['char'] |
| 171 |
if len(job) > 3: |
| 172 |
for rk, rl in job[3].items(): |
| 173 |
results[rk] = rl(results) |
| 174 |
if not args.keep: |
| 175 |
resf.unlink() |
| 176 |
tfile.unlink() |
| 177 |
return results |
| 178 |
|
| 179 |
|
| 180 |
outtop = 'name, length, ot, ot_GSUB, ot_GPOS, hbgr, gr,' \ |
| 181 |
' grcoll, grkern, usp, abc, uspnorm, usp_GSUB, usp_GPOS' |
| 182 |
outformat = '{label}, {char}, {ot}, {ot_GSUB}, {ot_GPOS}, {hbgr}, {gr},' \ |
| 183 |
' {grcoll}, {grkern}, {usp}, {abc}, {uspnorm}, {usp_GSUB},' \ |
| 184 |
' {usp_GPOS}' |
| 185 |
|
| 186 |
tmpd = Path(args.keep if args.keep else tempfile.mkdtemp(prefix='hbspeeds-')) |
| 187 |
dumpf = tmpd.joinpath('dump.out') |
| 188 |
with dumpf.open('w') as dumpfh: |
| 189 |
if args.file: |
| 190 |
c = json.load(args.file) |
| 191 |
args.file.close() |
| 192 |
else: |
| 193 |
c = [{'font': args.font, |
| 194 |
'textfile': args.text, |
| 195 |
'textline': args.line, |
| 196 |
'label': 'Commandline'}] |
| 197 |
|
| 198 |
if args.single or args.file is None: |
| 199 |
res = list(map(dorow, enumerate(c))) |
| 200 |
else: |
| 201 |
pool = Pool() |
| 202 |
res = pool.map(dorow, enumerate(c)) |
| 203 |
pool.close() |
| 204 |
pool.join() |
| 205 |
dumpf.unlink() |
| 206 |
|
| 207 |
if not args.keep: |
| 208 |
tmpd.rmdir() |
| 209 |
|
| 210 |
if not args.graph: |
| 211 |
print(outtop) |
| 212 |
for r in res: |
| 213 |
if r is not None: |
| 214 |
print(outformat.format(**r)) |
| 215 |
else: |
| 216 |
import matplotlib.pyplot as plt |
| 217 |
|
| 218 |
def stackplot(xs, ydat, width, **kwargs): |
| 219 |
res = [] |
| 220 |
h = len(ydat[0]) |
| 221 |
# yacc = [0] * len(ydat) # This is never used |
| 222 |
for i in range(h): |
| 223 |
res.append(plt.bar(xs, |
| 224 |
[y[i] - (y[i-1] if i > 0 else 0) for y in ydat], |
| 225 |
width, |
| 226 |
bottom=[y[i-1] if i > 0 else 0 for y in ydat], |
| 227 |
**kwargs)) |
| 228 |
return res |
| 229 |
|
| 230 |
width = 0.25 |
| 231 |
fig, ax = plt.subplots(figsize=(10, 5)) |
| 232 |
dat = [ |
| 233 |
[(x['uspnorm'], ) if args.normalize |
| 234 |
else (x['uspnorm'] / x['char'], ) for x in res], |
| 235 |
[(x['ot'], ) if args.normalize |
| 236 |
else (x['ot'] / x['char'], ) for x in res], |
| 237 |
[(x['gr'], ) if args.normalize |
| 238 |
else (x['gr'] / x['char'], ) for x in res]] |
| 239 |
# [(x['usp_GSUB'], x['usp_GSUB'] + x['usp_GPOS'] - x['abc'], |
| 240 |
# x['uspnorm']) for x in res], |
| 241 |
# [(x['ot_GSUB'], x['ot_GSUB'] + x['ot_GPOS'], x['ot']) for x in res], |
| 242 |
# [(x['gr'], x['hbgr']) for x in res]] |
| 243 |
pos = list(range(len(dat[1]))) |
| 244 |
plots = [stackplot(pos, dat[0], width, |
| 245 |
alpha=0.6, color='blue', label='usp'), |
| 246 |
stackplot([p + width for p in pos], dat[1], width, |
| 247 |
alpha=0.6, color='red', label='ot'), |
| 248 |
stackplot([p + width * 2 for p in pos], dat[2], width, |
| 249 |
alpha=0.6, color='green', label='gr')] |
| 250 |
ax.set_ylabel('Cycles / character') |
| 251 |
ax.set_title('Font Speed Test') |
| 252 |
ax.set_xticks([p + 1.5 * width for p in pos]) |
| 253 |
ax.set_xticklabels([x['label'] for x in res]) |
| 254 |
plt.xlim(min(pos) - width, max(pos) + width * 4) |
| 255 |
# ymax = max([max(x) for x in dat]) |
| 256 |
# plt.ylim(0, ymax) |
| 257 |
plt.legend([x[0] for x in plots], ['usp', 'ot', 'gr'], loc='upper left') |
| 258 |
plt.grid() |
| 259 |
plt.show() |