wangzelin.wzl 93a1074b0c patch 4.0
2022-10-24 17:57:12 +08:00

173 lines
4.4 KiB
Python

#!/bin/env python
__author__ = 'dongyun.zdy'
import sys
import os
import numpy as np
import getopt
file_name = "scan_model.res.formal"
if len(sys.argv) >= 2:
file_name = sys.argv[1]
out_file_name = file_name + ".prep"
time_per_case = 5
use_delete_min_max = False
filters = []
out_columns = [c for c in xrange(100)]
cols_supplied = False
wrong_arg = False
target_column_id = 0
#sys.argv.extend("-i sort_result -o sort.8.test -t 7 -C 2 -f 1,e,8".split(" "))
sys.argv.extend("-i nestloop_result -o nl_result".split(" "))
opts,args = getopt.getopt(sys.argv[1:],"i:o:t:f:a:dc:C:")
for op, value in opts:
if "-i" == op:
file_name = value
elif "-o" == op:
out_file_name = value
elif "-t" == op:
time_per_case = int(value)
elif "-f" == op:
filter_str = value
filter_elements = filter_str.split(",")
if not filter_elements[1] in ["g","l","ge","le","e","ne"]:
print "invalid filter type"
sys.exit(1)
filters.append(filter_str.split(","))
elif "-a" == op:
time_per_case = int(value)
elif "-d" == op:
use_delete_min_max = True
elif "-C" == op:
target_column_id = int(value)
elif "-c" == op:
if not cols_supplied:
cols_supplied = True
out_columns = []
out_columns.extend([int(c) for c in value.split(",")])
else:
wrong_arg = True
if wrong_arg:
print "wrong arg"
sys.exit(1)
if time_per_case < 5:
use_delete_min_max = False
if os.path.exists(out_file_name):
os.remove(out_file_name)
origin_file = open(file_name, "r")
out_file = open(out_file_name,"w")
i = 0
column_nums = []
avgs = []
avg_strs = []
def delete(li, index):
li = li[:index] + li[index+1:]
return li
def find_max_index(l):
max = -9999999999999999999999
max_i = -1
for i in xrange(len(l)):
if l[i] > max:
max = l[i]
max_i = i
return max_i
def find_min_index(l):
min = 999999999999999999999999
min_i = -1
for i in xrange(len(l)):
if l[i] < min:
min = l[i]
min_i = i
return min_i
def delete_max_min_case(column_nums, column_id):
# min_i = find_min_index(column_nums[len(column_nums) - 1])
# for j in xrange(len(column_nums)):
# column_nums[j] = delete(column_nums[j], min_i)
max_i = find_max_index(column_nums[column_id])
for j in xrange(len(column_nums)):
column_nums[j] = delete(column_nums[j], max_i)
max_i = find_max_index(column_nums[column_id])
for j in xrange(len(column_nums)):
column_nums[j] = delete(column_nums[j], max_i)
def do_filter(column_strs):
filtered = False
for f in filters:
if f[1] == "g" and float(column_strs[int(f[0])]) <= int(f[2]) :
filtered = True
break
elif f[1] == "l" and float(column_strs[int(f[0])]) >= int(f[2]) :
filtered = True
break
elif f[1] == "ge" and float(column_strs[int(f[0])]) < int(f[2]) :
filtered = True
break
elif f[1] == "le" and float(column_strs[int(f[0])]) > int(f[2]) :
filtered = True
break
elif f[1] == "e" and float(column_strs[int(f[0])]) != int(f[2]) :
filtered = True
break
elif f[1] == "ne" and float(column_strs[int(f[0])]) == int(f[2]) :
filtered = True
break
return filtered
state = 0 #comment line
elements = []
for line in origin_file:
line = line.strip()
if state == 0:
out_file.write(line + "\n")
elif state == 1:
elements = line.split(",row_count : ")
elif state == 2:
pass
elif state == 3:
pass
elif state == 4:
elements.append(line.split("join_time except conds : ")[1])
elif state == 5:
elements.append(line.split("equal_eval : ")[1])
elif state == 6:
pass
elif state == 7:
elements.append(line.split("other_eval : ")[1])
elif state == 8:
pass
elif state == 9:
elements.append(line.split("right_cache_put : ")[1])
elif state == 10:
elements.append(line.split("right_cache_acc : ")[1])
elif state == 11:
elements.append(line.split("match_group_count : ")[1])
out_file.write(",".join(elements) + "\n")
else:
print "wrong state"
state = (state + 1) % 12
origin_file.close()
out_file.close()