1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
|
"""Print a summary of specialization stats for all files in the
default stats folders.
"""
import collections
import os.path
import opcode
from datetime import date
import itertools
import argparse
if os.name == "nt":
DEFAULT_DIR = "c:\\temp\\py_stats\\"
else:
DEFAULT_DIR = "/tmp/py_stats/"
#Create list of all instruction names
specialized = iter(opcode._specialized_instructions)
opname = ["<0>"]
for name in opcode.opname[1:]:
if name.startswith("<"):
try:
name = next(specialized)
except StopIteration:
pass
opname.append(name)
TOTAL = "specialization.deferred", "specialization.hit", "specialization.miss", "execution_count"
def print_specialization_stats(name, family_stats, defines):
if "specializable" not in family_stats:
return
total = sum(family_stats.get(kind, 0) for kind in TOTAL)
if total == 0:
return
with Section(name, 3, f"specialization stats for {name} family"):
rows = []
for key in sorted(family_stats):
if key.startswith("specialization.failure_kinds"):
continue
if key.startswith("specialization."):
label = key[len("specialization."):]
elif key == "execution_count":
label = "unquickened"
else:
label = key
if key not in ("specialization.success", "specialization.failure", "specializable"):
rows.append((f"{label:>12}", f"{family_stats[key]:>12}", f"{100*family_stats[key]/total:0.1f}%"))
emit_table(("Kind", "Count", "Ratio"), rows)
print_title("Specialization attempts", 4)
total_attempts = 0
for key in ("specialization.success", "specialization.failure"):
total_attempts += family_stats.get(key, 0)
rows = []
for key in ("specialization.success", "specialization.failure"):
label = key[len("specialization."):]
label = label[0].upper() + label[1:]
val = family_stats.get(key, 0)
rows.append((label, val, f"{100*val/total_attempts:0.1f}%"))
emit_table(("", "Count:", "Ratio:"), rows)
total_failures = family_stats.get("specialization.failure", 0)
failure_kinds = [ 0 ] * 30
for key in family_stats:
if not key.startswith("specialization.failure_kind"):
continue
_, index = key[:-1].split("[")
index = int(index)
failure_kinds[index] = family_stats[key]
failures = [(value, index) for (index, value) in enumerate(failure_kinds)]
failures.sort(reverse=True)
rows = []
for value, index in failures:
if not value:
continue
rows.append((kind_to_text(index, defines, name), value, f"{100*value/total_failures:0.1f}%"))
emit_table(("Failure kind", "Count:", "Ratio:"), rows)
def gather_stats():
stats = collections.Counter()
for filename in os.listdir(DEFAULT_DIR):
with open(os.path.join(DEFAULT_DIR, filename)) as fd:
for line in fd:
key, value = line.split(":")
key = key.strip()
value = int(value)
stats[key] += value
return stats
def extract_opcode_stats(stats):
opcode_stats = [ {} for _ in range(256) ]
for key, value in stats.items():
if not key.startswith("opcode"):
continue
n, _, rest = key[7:].partition("]")
opcode_stats[int(n)][rest.strip(".")] = value
return opcode_stats
def parse_kinds(spec_src):
defines = collections.defaultdict(list)
for line in spec_src:
line = line.strip()
if not line.startswith("#define SPEC_FAIL_"):
continue
line = line[len("#define SPEC_FAIL_"):]
name, val = line.split()
defines[int(val.strip())].append(name.strip())
return defines
def pretty(defname):
return defname.replace("_", " ").lower()
def kind_to_text(kind, defines, opname):
if kind < 7:
return pretty(defines[kind][0])
if opname.endswith("ATTR"):
opname = "ATTR"
if opname.endswith("SUBSCR"):
opname = "SUBSCR"
if opname.startswith("PRECALL"):
opname = "CALL"
for name in defines[kind]:
if name.startswith(opname):
return pretty(name[len(opname)+1:])
return "kind " + str(kind)
def categorized_counts(opcode_stats):
basic = 0
specialized = 0
not_specialized = 0
specialized_instructions = {
op for op in opcode._specialized_instructions
if "__" not in op and "ADAPTIVE" not in op}
adaptive_instructions = {
op for op in opcode._specialized_instructions
if "ADAPTIVE" in op}
for i, opcode_stat in enumerate(opcode_stats):
if "execution_count" not in opcode_stat:
continue
count = opcode_stat['execution_count']
name = opname[i]
if "specializable" in opcode_stat:
not_specialized += count
elif name in adaptive_instructions:
not_specialized += count
elif name in specialized_instructions:
miss = opcode_stat.get("specialization.miss", 0)
not_specialized += miss
specialized += count - miss
else:
basic += count
return basic, not_specialized, specialized
def print_title(name, level=2):
print("#"*level, name)
print()
class Section:
def __init__(self, title, level=2, summary=None):
self.title = title
self.level = level
if summary is None:
self.summary = title.lower()
else:
self.summary = summary
def __enter__(self):
print_title(self.title, self.level)
print("<details>")
print("<summary>", self.summary, "</summary>")
print()
return self
def __exit__(*args):
print()
print("</details>")
print()
def emit_table(header, rows):
width = len(header)
header_line = "|"
under_line = "|"
for item in header:
under = "---"
if item.endswith(":"):
item = item[:-1]
under += ":"
header_line += item + " | "
under_line += under + "|"
print(header_line)
print(under_line)
for row in rows:
if width is not None and len(row) != width:
raise ValueError("Wrong number of elements in row '" + str(rows) + "'")
print("|", " | ".join(str(i) for i in row), "|")
print()
def emit_execution_counts(opcode_stats, total):
with Section("Execution counts", summary="execution counts for all instructions"):
counts = []
for i, opcode_stat in enumerate(opcode_stats):
if "execution_count" in opcode_stat:
count = opcode_stat['execution_count']
miss = 0
if "specializable" not in opcode_stat:
miss = opcode_stat.get("specialization.miss")
counts.append((count, opname[i], miss))
counts.sort(reverse=True)
cumulative = 0
rows = []
for (count, name, miss) in counts:
cumulative += count
if miss:
miss = f"{100*miss/count:0.1f}%"
else:
miss = ""
rows.append((name, count, f"{100*count/total:0.1f}%",
f"{100*cumulative/total:0.1f}%", miss))
emit_table(
("Name", "Count:", "Self:", "Cumulative:", "Miss ratio:"),
rows
)
def emit_specialization_stats(opcode_stats):
spec_path = os.path.join(os.path.dirname(__file__), "../../Python/specialize.c")
with open(spec_path) as spec_src:
defines = parse_kinds(spec_src)
with Section("Specialization stats", summary="specialization stats by family"):
for i, opcode_stat in enumerate(opcode_stats):
name = opname[i]
print_specialization_stats(name, opcode_stat, defines)
def emit_specialization_overview(opcode_stats, total):
basic, not_specialized, specialized = categorized_counts(opcode_stats)
with Section("Specialization effectiveness"):
emit_table(("Instructions", "Count:", "Ratio:"), (
("Basic", basic, f"{basic*100/total:0.1f}%"),
("Not specialized", not_specialized, f"{not_specialized*100/total:0.1f}%"),
("Specialized", specialized, f"{specialized*100/total:0.1f}%"),
))
def emit_call_stats(stats):
with Section("Call stats", summary="Inlined calls and frame stats"):
total = 0
for key, value in stats.items():
if "Calls to" in key:
total += value
rows = []
for key, value in stats.items():
if "Calls to" in key:
rows.append((key, value, f"{100*value/total:0.1f}%"))
for key, value in stats.items():
if key.startswith("Frame"):
rows.append((key, value, f"{100*value/total:0.1f}%"))
emit_table(("", "Count:", "Ratio:"), rows)
def emit_object_stats(stats):
with Section("Object stats", summary="allocations, frees and dict materializatons"):
total = stats.get("Object new values")
rows = []
for key, value in stats.items():
if key.startswith("Object"):
if "materialize" in key:
materialize = f"{100*value/total:0.1f}%"
else:
materialize = ""
label = key[6:].strip()
label = label[0].upper() + label[1:]
rows.append((label, value, materialize))
emit_table(("", "Count:", "Ratio:"), rows)
def get_total(opcode_stats):
total = 0
for opcode_stat in opcode_stats:
if "execution_count" in opcode_stat:
total += opcode_stat['execution_count']
return total
def emit_pair_counts(opcode_stats, total):
with Section("Pair counts", summary="Pair counts for top 100 pairs"):
pair_counts = []
for i, opcode_stat in enumerate(opcode_stats):
if i == 0:
continue
for key, value in opcode_stat.items():
if key.startswith("pair_count"):
x, _, _ = key[11:].partition("]")
if value:
pair_counts.append((value, (i, int(x))))
pair_counts.sort(reverse=True)
cumulative = 0
rows = []
for (count, pair) in itertools.islice(pair_counts, 100):
i, j = pair
cumulative += count
rows.append((opname[i] + " " + opname[j], count, f"{100*count/total:0.1f}%",
f"{100*cumulative/total:0.1f}%"))
emit_table(("Pair", "Count:", "Self:", "Cumulative:"),
rows
)
def main():
stats = gather_stats()
opcode_stats = extract_opcode_stats(stats)
total = get_total(opcode_stats)
emit_execution_counts(opcode_stats, total)
emit_pair_counts(opcode_stats, total)
emit_specialization_stats(opcode_stats)
emit_specialization_overview(opcode_stats, total)
emit_call_stats(stats)
emit_object_stats(stats)
print("---")
print("Stats gathered on:", date.today())
if __name__ == "__main__":
main()
|