format code

Author: VitorRamos

.github/workflows/main.yml

@@ -33,4 +33,3 @@ jobs:
 #    - name: tests
 #      run: |
 #       ./run_tests.sh
-

.pre-commit-config.yaml

@@ -0,0 +1,11 @@
+repos:
+-   repo: https://github.com/pre-commit/pre-commit-hooks
+    rev: v2.3.0
+    hooks:
+    -   id: check-yaml
+    -   id: end-of-file-fixer
+    -   id: trailing-whitespace
+-   repo: https://github.com/psf/black
+    rev: 22.10.0
+    hooks:
+    -   id: black

README.md

@@ -67,4 +67,4 @@ Performance counters are special hardware registers available on most modern CPU
   + **rdmsr**: Reads the contents of a 64-bit model specific register (MSR) specified in the ECX register into registers EDX:EAX. This instruction must be executed at privilege level 0 or in real-address mode
 
   + **rdpmc**: Is slightly faster that the equivelent rdmsr instruction. rdpmc can also be configured to allow access to the counters from userspace, without being priviledged.
-+ **From Userspace** (Linux) : The Linux Performance Counter subsystem provides an abstraction of these hardware capabilities. It provides per task and per CPU counters, counter groups, and it provides event capabilities on top of those. It provides "virtual" 64-bit counters, regardless of the width of the underlying hardware counters. Performance counters are accessed via special file descriptors. There's one file descriptor per virtual counter used. The special file descriptor is opened via the perf_event_open() system call. These system call do not use rdpmc but rdpmc is not necessarily faster than other methods for reading event values.
++ **From Userspace** (Linux) : The Linux Performance Counter subsystem provides an abstraction of these hardware capabilities. It provides per task and per CPU counters, counter groups, and it provides event capabilities on top of those. It provides "virtual" 64-bit counters, regardless of the width of the underlying hardware counters. Performance counters are accessed via special file descriptors. There's one file descriptor per virtual counter used. The special file descriptor is opened via the perf_event_open() system call. These system call do not use rdpmc but rdpmc is not necessarily faster than other methods for reading event values.

install.sh

@@ -5,4 +5,3 @@ sudo apt install g++ gcc swig libpfm4-dev python3-dev python3-pip
 cd profiler/
 python3 setup.py build
 python3 setup.py install
-

main.py

@@ -3,11 +3,11 @@
 import numpy as np
 
 try:
-    evs_monitor= [["PERF_COUNT_HW_INSTRUCTIONS"],["SYSTEMWIDE:RAPL_ENERGY_PKG"]]
-    program= Profiler(program_args=['./tests/simple_bench'], events_groups=evs_monitor)
-    data= program.run(sample_period=0.01,reset_on_sample=False)
-    df= pd.DataFrame(data, columns= ["inst","energy"] )
-    df["energy"]*=2.3283064365386962890625e-10
+    evs_monitor = [["PERF_COUNT_HW_INSTRUCTIONS"], ["SYSTEMWIDE:RAPL_ENERGY_PKG"]]
+    program = Profiler(program_args=["./tests/simple_bench"], events_groups=evs_monitor)
+    data = program.run(sample_period=0.01, reset_on_sample=False)
+    df = pd.DataFrame(data, columns=["inst", "energy"])
+    df["energy"] *= 2.3283064365386962890625e-10
     print(df)
 except RuntimeError as e:
-    print(e.args[0])
+    print(e.args[0])

profiler/MANIFEST.in

@@ -1,4 +1,4 @@
 recursive-include  perfmon/include/ *
 include  profiler/workload.cpp
 include  profiler/workload.h
-include  ../README.md
+include  ../README.md

profiler/perfmon/pmu.py

@@ -25,78 +25,83 @@
 import os
 from perfmon import *
 
+
 def public_members(self):
     s = "{ "
     for k, v in self.__dict__.items():
-      if not k[0] == '_':
-        s += "%s : %s, " % (k, v)
+        if not k[0] == "_":
+            s += "%s : %s, " % (k, v)
     s += " }"
     return s
 
+
 class System:
-  # Use the os that gives us everything
-  os = PFM_OS_PERF_EVENT_EXT
-
-  def __init__(self):
-    self.ncpus = os.sysconf('SC_NPROCESSORS_ONLN')
-    self.pmus = []
-    for i in range(0, PFM_PMU_MAX):
-      try:
-        pmu = PMU(i)
-      except:
-        pass
-      else:
-        self.pmus.append(pmu)
-
-  def __repr__(self):
-    return public_members(self)
+    # Use the os that gives us everything
+    os = PFM_OS_PERF_EVENT_EXT
+
+    def __init__(self):
+        self.ncpus = os.sysconf("SC_NPROCESSORS_ONLN")
+        self.pmus = []
+        for i in range(0, PFM_PMU_MAX):
+            try:
+                pmu = PMU(i)
+            except:
+                pass
+            else:
+                self.pmus.append(pmu)
+
+    def __repr__(self):
+        return public_members(self)
+
 
 class Event:
-  def __init__(self, info):
-    self.info = info
-    self.__attrs = []
+    def __init__(self, info):
+        self.info = info
+        self.__attrs = []
+
+    def __repr__(self):
+        return "\n" + public_members(self)
 
-  def __repr__(self):
-    return '\n' + public_members(self)
+    def __parse_attrs(self):
+        info = self.info
+        for index in range(0, info.nattrs):
+            self.__attrs.append(pfm_get_event_attr_info(info.idx, index, System.os)[1])
 
-  def __parse_attrs(self):
-    info = self.info
-    for index in range(0, info.nattrs):
-      self.__attrs.append(pfm_get_event_attr_info(info.idx, index,
-                                                  System.os)[1])
+    def attrs(self):
+        if not self.__attrs:
+            self.__parse_attrs()
+        return self.__attrs
 
-  def attrs(self):
-    if not self.__attrs:
-      self.__parse_attrs()
-    return self.__attrs
 
 class PMU:
-  def __init__(self, i):
-    self.info = pfm_get_pmu_info(i)[1]
-    self.__events = []
-
-  def __parse_events(self):
-    index = self.info.first_event
-    while index != -1:
-      self.__events.append(Event(pfm_get_event_info(index, System.os)[1]))
-      index = pfm_get_event_next(index)
-
-  def events(self):
-    if not self.__events:
-      self.__parse_events()
-    return self.__events
-
-  def __repr__(self):
-    return public_members(self)
-
-if __name__ == '__main__':
-  from perfmon import *
-  s = System()
-  for pmu in s.pmus:
-    info = pmu.info
-    if info.flags.is_present:
-      print(info.name, info.size, info.nevents)
-      for e in pmu.events():
-        print(e.info.name, e.info.code)
-        for a in e.attrs():
-          print('\t\t', a.name, a.code)
+    def __init__(self, i):
+        self.info = pfm_get_pmu_info(i)[1]
+        self.__events = []
+
+    def __parse_events(self):
+        index = self.info.first_event
+        while index != -1:
+            self.__events.append(Event(pfm_get_event_info(index, System.os)[1]))
+            index = pfm_get_event_next(index)
+
+    def events(self):
+        if not self.__events:
+            self.__parse_events()
+        return self.__events
+
+    def __repr__(self):
+        return public_members(self)
+
+
+if __name__ == "__main__":
+    from perfmon import *
+
+    s = System()
+    for pmu in s.pmus:
+        info = pmu.info
+        if info.flags.is_present:
+            print(info.name, info.size, info.nevents)
+            for e in pmu.events():
+                print(e.info.name, e.info.code)
+                for a in e.attrs():
+                    print("\t\t", a.name, a.code)

profiler/perfmon/session.py

@@ -27,54 +27,57 @@
 
 # Common base class
 class Session:
-  def __init__(self, events):
-    self.system = System()
-    self.event_names = events
-    self.events = []
-    self.fds = []
-    for e in events:
-      err, encoding = pfm_get_perf_event_encoding(e, PFM_PLM0 | PFM_PLM3,
-                                                  None, None)
-      self.events.append(encoding)
+    def __init__(self, events):
+        self.system = System()
+        self.event_names = events
+        self.events = []
+        self.fds = []
+        for e in events:
+            err, encoding = pfm_get_perf_event_encoding(
+                e, PFM_PLM0 | PFM_PLM3, None, None
+            )
+            self.events.append(encoding)
 
-  def __del__(self):
-    pass
+    def __del__(self):
+        pass
+
+    def read(self, fd):
+        # TODO: determine counter width
+        return os.read(fd, 8)
 
-  def read(self, fd):
-    # TODO: determine counter width
-    return os.read(fd, 8)
 
 class SystemWideSession(Session):
-  def __init__(self, cpus, events):
-    self.cpus = cpus
-    Session.__init__(self, events)
+    def __init__(self, cpus, events):
+        self.cpus = cpus
+        Session.__init__(self, events)
+
+    def __del__(self):
+        Session.__del__(self)
 
-  def __del__(self):
-    Session.__del__(self)
+    def start(self):
+        self.cpu_fds = []
+        for c in self.cpus:
+            self.cpu_fds.append([])
+            cur_cpu_fds = self.cpu_fds[-1]
+            for e in self.events:
+                cur_cpu_fds.append(perf_event_open(e, -1, c, -1, 0))
 
-  def start(self):
-    self.cpu_fds = []
-    for c in self.cpus:
-      self.cpu_fds.append([])
-      cur_cpu_fds = self.cpu_fds[-1]
-      for e in self.events:
-        cur_cpu_fds.append(perf_event_open(e, -1, c, -1, 0))
+    def read(self, c, i):
+        index = self.cpus.index(c)
+        return Session.read(self, self.cpu_fds[index][i])
 
-  def read(self, c, i):
-    index = self.cpus.index(c)
-    return Session.read(self, self.cpu_fds[index][i])
 
 class PerThreadSession(Session):
-  def __init__(self, pid, events):
-    self.pid = pid
-    Session.__init__(self, events)
+    def __init__(self, pid, events):
+        self.pid = pid
+        Session.__init__(self, events)
 
-  def __del__(self):
-    Session.__del__(self)
+    def __del__(self):
+        Session.__del__(self)
 
-  def start(self):
-    for e in self.events:
-      self.fds.append(perf_event_open(e, self.pid, -1, -1, 0))
+    def start(self):
+        for e in self.events:
+            self.fds.append(perf_event_open(e, self.pid, -1, -1, 0))
 
-  def read(self, i):
-    return Session.read(self, self.fds[i])
+    def read(self, i):
+        return Session.read(self, self.fds[i])

profiler/profiler/__init__.py

@@ -1,3 +1,3 @@
-from . profiler import *
-from . events import *
-from . compare import *
+from .profiler import *
+from .events import *
+from .compare import *