AI/rocm-systems
2
0
Fork 0
Κώδικας Ζητήματα Pull Requests Δράσεις Πακέτα Έργα Κυκλοφορίες Wiki Δραστηριότητα

keep converter in csv_converter.py

Περιήγηση Στο Πηγαίο 
Signed-off-by: josantos <josantos@amd.com>


[ROCm/rocprofiler-compute commit: 2948f73ae8]
Αυτή η υποβολή περιλαμβάνεται σε:
josantos
2023-08-09 10:46:57 -05:00
γονέας 67971bca46
υποβολή 9d01f81aac
2 αρχεία άλλαξαν με 62 προσθήκες και 132 διαγραφές
Λήψη Αρχείου Patch Λήψη Αρχείου Diff Ανάπτυξη όλων των αρχείων Σύμπτυξη όλων των αρχείων
projects/rocprofiler-compute/src/omniperf
+2 -72
Προβολή Αρχείου
@@ -262,76 +262,6 @@ def mongo_import(args, profileAndImport):
csv_converter.convert_folder(connectionInfo, Extractionlvl) csv_converter.convert_folder(connectionInfo, Extractionlvl)
print("-- Complete! --") print("-- Complete! --")
def kernel_name_shortener(df, cache, level):
if level >= 5:
return df
columnName = ""
if "KernelName" in df:
columnName = "KernelName"
if "Name" in df:
columnName = "Name"
if columnName == "KernelName" or columnName == "Name":
# loop through all indices
for index in df.index:
original_name = df.loc[index, columnName]
if original_name in cache:
continue
# cache miss, add the shortened name to the dictionary
new_name = ""
matches = ""
names_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]+)(::)?")
# works for name Kokkos::namespace::init_lock_array_kernel_threadid(int) [clone .kd]
if names_and_args.search(original_name):
matches = names_and_args.findall(original_name)
else:
# Works for first case '__amd_rocclr_fillBuffer.kd'
# remove .kd and then parse through original regex
first_case = re.compile(r"([^\s]+)(.kd)")
Mod_name_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]*)")
interim_name = first_case.search(original_name).group(1)
matches = Mod_name_and_args.findall(interim_name)
current_level = 0
for name in matches:
##can cause errors if a function name or argument is equal to 'clone'
if name[0] == "clone":
continue
if len(name) == 3:
if name[2] == "::":
continue
if current_level < level:
new_name += name[0]
# closing '>' is to be taken account by the while loop
if name[1].count(">") == 0:
if current_level < level:
if not (current_level == level - 1 and name[1].count("<") > 0):
new_name += name[1]
current_level += name[1].count("<")
curr_index = 0
# cases include '>' '> >, ' have to go in depth here to not lose account of commas and current level
while name[1].count(">") > 0 and curr_index < len(name[1]):
if current_level < level:
new_name += name[1][curr_index:]
current_level -= name[1][curr_index:].count(">")
curr_index = len(name[1])
elif name[1][curr_index] == (">"):
current_level -= 1
curr_index += 1
cache[original_name] = new_name
if new_name == None or new_name == "":
cache[original_name] = original_name
df[columnName] = df[columnName].map(cache)
return df
################################################ ################################################
# Roofline Helpers # Roofline Helpers
################################################ ################################################
@@ -860,7 +790,7 @@ def main():
engine="python", engine="python",
) )
t2 = kernel_name_shortener(t1, cache, level=Extractionlvl) t2 = csv_converter.kernel_name_shortener(t1, cache, level=Extractionlvl)
except pd.errors.EmptyDataError: except pd.errors.EmptyDataError:
print("Skipping empty csv " + file) print("Skipping empty csv " + file)
@@ -881,7 +811,7 @@ def main():
engine="python", engine="python",
) )
t2 = kernel_name_shortener(t1, cache, level=Extractionlvl) t2 = csv_converter.kernel_name_shortener(t1, cache, level=Extractionlvl)
except pd.errors.EmptyDataError: except pd.errors.EmptyDataError:
print("Skipping empty csv " + file) print("Skipping empty csv " + file)
projects/rocprofiler-compute/src/utils/csv_converter.py
+60 -60
Προβολή Αρχείου
@@ -33,81 +33,81 @@ from pymongo import MongoClient
from tqdm import tqdm from tqdm import tqdm
import shutil import shutil
# cache = dict() cache = dict()
supported_arch = {"gfx906": "mi50", "gfx908": "mi100", "gfx90a": "mi200"} supported_arch = {"gfx906": "mi50", "gfx908": "mi100", "gfx90a": "mi200"}
MAX_SERVER_SEL_DELAY = 5000 # 5 sec connection timeout MAX_SERVER_SEL_DELAY = 5000 # 5 sec connection timeout
# def kernel_name_shortener(df, cache, level): def kernel_name_shortener(df, cache, level):
# if level >= 5: if level >= 5:
# return df return df
# columnName = "" columnName = ""
# if "KernelName" in df: if "KernelName" in df:
# columnName = "KernelName" columnName = "KernelName"
# if "Name" in df: if "Name" in df:
# columnName = "Name" columnName = "Name"
# if columnName == "KernelName" or columnName == "Name": if columnName == "KernelName" or columnName == "Name":
# # loop through all indices # loop through all indices
# for index in df.index: for index in df.index:
# original_name = df.loc[index, columnName] original_name = df.loc[index, columnName]
# if original_name in cache: if original_name in cache:
# continue continue
# # cache miss, add the shortened name to the dictionary # cache miss, add the shortened name to the dictionary
# new_name = "" new_name = ""
# matches = "" matches = ""
# names_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]+)(::)?") names_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]+)(::)?")
# # works for name Kokkos::namespace::init_lock_array_kernel_threadid(int) [clone .kd] # works for name Kokkos::namespace::init_lock_array_kernel_threadid(int) [clone .kd]
# if names_and_args.search(original_name): if names_and_args.search(original_name):
# matches = names_and_args.findall(original_name) matches = names_and_args.findall(original_name)
# else: else:
# # Works for first case '__amd_rocclr_fillBuffer.kd' # Works for first case '__amd_rocclr_fillBuffer.kd'
# # remove .kd and then parse through original regex # remove .kd and then parse through original regex
# first_case = re.compile(r"([^\s]+)(.kd)") first_case = re.compile(r"([^\s]+)(.kd)")
# Mod_name_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]*)") Mod_name_and_args = re.compile(r"(?P<name>[( )A-Za-z0-9_]+)([ ,*<>()]*)")
# interim_name = first_case.search(original_name).group(1) interim_name = first_case.search(original_name).group(1)
# matches = Mod_name_and_args.findall(interim_name) matches = Mod_name_and_args.findall(interim_name)
# current_level = 0 current_level = 0
# for name in matches: for name in matches:
# ##can cause errors if a function name or argument is equal to 'clone' ##can cause errors if a function name or argument is equal to 'clone'
# if name[0] == "clone": if name[0] == "clone":
# continue continue
# if len(name) == 3: if len(name) == 3:
# if name[2] == "::": if name[2] == "::":
# continue continue
# if current_level < level: if current_level < level:
# new_name += name[0] new_name += name[0]
# # closing '>' is to be taken account by the while loop # closing '>' is to be taken account by the while loop
# if name[1].count(">") == 0: if name[1].count(">") == 0:
# if current_level < level: if current_level < level:
# if not (current_level == level - 1 and name[1].count("<") > 0): if not (current_level == level - 1 and name[1].count("<") > 0):
# new_name += name[1] new_name += name[1]
# current_level += name[1].count("<") current_level += name[1].count("<")
# curr_index = 0 curr_index = 0
# # cases include '>' '> >, ' have to go in depth here to not lose account of commas and current level # cases include '>' '> >, ' have to go in depth here to not lose account of commas and current level
# while name[1].count(">") > 0 and curr_index < len(name[1]): while name[1].count(">") > 0 and curr_index < len(name[1]):
# if current_level < level: if current_level < level:
# new_name += name[1][curr_index:] new_name += name[1][curr_index:]
# current_level -= name[1][curr_index:].count(">") current_level -= name[1][curr_index:].count(">")
# curr_index = len(name[1]) curr_index = len(name[1])
# elif name[1][curr_index] == (">"): elif name[1][curr_index] == (">"):
# current_level -= 1 current_level -= 1
# curr_index += 1 curr_index += 1
# cache[original_name] = new_name cache[original_name] = new_name
# if new_name == None or new_name == "": if new_name == None or new_name == "":
# cache[original_name] = original_name cache[original_name] = original_name
# df[columnName] = df[columnName].map(cache) df[columnName] = df[columnName].map(cache)
# return df return df
# Verify target directory and setup connection # Verify target directory and setup connection