Merge pull request #12 from hariharan-devarajan/feature/thread_id

Features to Improve documentation

README.md

@@ -7,40 +7,33 @@ A low-level profiler for capture I/O calls from deep learning applications.
 
 Requirements
 1. Python > 3.8
-2. spack
 
-Using spack dlio-profiler
-```
-spack env create -d ./spack-env
-spack env activate -p ./spack-env
-spack add [email protected]
-spack install
-```
 
-create a virtual env for your python package where u will use dlio_profiler.
-```
-python3 -m venv ./venv
-source venv/bin/activate
-pip install .
-```
-install in local user
-```
-export DLIO_LOGGER_USER=1
-pip install .
-```
-install directly from github
-```
-pip install git+https://github.com/hariharan-devarajan/dlio-profiler.git
-```
-Build dlio profiler through cmake
+## Build DLIO Profiler with pip
+
+Users can easily install DLIO profiler using pip. This is the way most python packages are installed.
+This method would work for both native python environments and conda environments.
+
+### From source
+
+```bash
+    git clone [email protected]:hariharan-devarajan/dlio-profiler.git
+    cd dlio-profiler
+    # You can skip this for installing the dev branch.
+    # for latest stable version use master branch.
+    git checkout tags/<Release> -b <Release>
+    pip install .
 ```
-cd dlio-profiler
-mkdir build
-cd build
-cmake -DCMAKE_INSTALL_PREFIX=../venv ../
-make install -j
+
+### From Github
+
+```bash
+DLP_VERSION=dev
+pip install git+https://github.com/hariharan-devarajan/dlio-profiler.git@${DLP_VERSION}
 ```
 
+For more build instructions check [here](https://dlio-profiler.readthedocs.io/en/latest/build.html)
+
 Usage
 
 ```

docs/build.rst

@@ -6,13 +6,13 @@ This section describes how to build DLIO Profiler.
 
 There are three build options:
 
-* build DLIO Profiler with pip (recommended),
-* build DLIO Profiler with Spack, and
-* build DLIO Profiler with cmake
+- build DLIO Profiler with pip (recommended),
+- build DLIO Profiler with Spack, and
+- build DLIO Profiler with cmake
 
 ----------
 
------------------------------------------
+----------------------------
 Build DLIO Profiler with pip
 ----------------------------
 

docs/examples.rst

@@ -635,3 +635,27 @@ Job submition script
      LD_PRELOAD=./dlio_ml_workloads/PolarisAT/conda-envs/ml_workload_latest_conda/lib/libdlio_profiler_preload.so aprun -n 4 -N 4 python resnet_hvd_dlio.py --batch-size 64 --epochs 1 > dlio_log 2>&1
    
      cat *.pfw > combined_logs.pfw # To combine to a single pfw file. 
+
+
+***********************
+Integrated Applications
+***********************
+
+Here is the list applications that currently use DLIO Profiler.
+
+1. `DLIO Benchmark <https://github.com/argonne-lcf/dlio_benchmark>`_
+2. MuMMI
+3. Resnet50 with pytorch and torchvision
+
+****************************
+Example Chrome Tracing Plots
+****************************
+
+Example of Unet3D application with DLIO Benchmark. This trace shows the first few steps of the benchmark.
+Here, we can see that we can get application level calls (e.g., ``train`` and ``TorchDataset``) as well as low-level I/O calls (dark green color).
+
+.. image:: images/tracing/trace.png
+  :width: 400
+  :alt: Unet3D applications
+
+

docs/limitations.rst

@@ -4,3 +4,9 @@ Limitations
 
 For certain system the spawning of processes create new processes but do not carry env variable.
 In those cases the LD_PRELOAD or python module load would not load DLIO Profiler as they are removed by system and result in missing profiling info.
+
+----------------
+OS Compatibility
+----------------
+
+The profiler internally uses system calls such as ``getpid()`` and ``gettid()`` which are only implemented in Linux OS.

src/dlio_profiler/core/dlio_profiler_main.cpp

@@ -100,15 +100,15 @@ dlio_profiler::DLIOProfilerCore::initlialize(bool is_init, bool _bind, const cha
         gotcha_priority = atoi(dlio_profiler_priority_str); // GCOV_EXCL_LINE
       }
       if (_process_id == nullptr || *_process_id == -1) {
-        this->process_id = getpid();
+        this->process_id = dlp_getpid();
       } else {
         this->process_id = *_process_id;
       }
       DLIO_PROFILER_LOGINFO("Setting process_id to %d", this->process_id);
       if (_log_file == nullptr) {
         char *dlio_profiler_log = getenv(DLIO_PROFILER_LOG_FILE);
         char proc_name[PATH_MAX], cmd[128];
-        sprintf(cmd, "/proc/%d/cmdline", getpid());
+        sprintf(cmd, "/proc/%d/cmdline", dlp_getpid());
         int fd = dlp_open(cmd, O_RDONLY);
         const char *exec_file_name = nullptr;
         if (fd != -1) {

src/dlio_profiler/dlio_logger.h

@@ -51,7 +51,7 @@ class DLIOLogger {
             fd = fileno(stdout);
             log_file = "STDOUT";
           } else {
-            int pid = getpid();
+            int pid = dlp_getpid();
             log_file = std::string(dlio_profiler_log_dir) + "/" + "trace_ll_" + std::to_string(pid) + ".pfw";
           } // GCOV_EXCL_STOP
         }

src/dlio_profiler/utils/posix_internal.cpp

@@ -46,3 +46,11 @@ int dlp_fsync(int fd) { // GCOV_EXCL_START
 ssize_t dlp_readlink(const char *path, char *buf, size_t bufsize) {
   return syscall(SYS_readlink, path, buf, bufsize);
 }
+
+pid_t dlp_gettid(){
+  return syscall(SYS_gettid);
+}
+
+pid_t dlp_getpid(){
+  return syscall(SYS_getpid);
+}

src/dlio_profiler/utils/posix_internal.h

@@ -25,4 +25,8 @@ int dlp_fsync(int fd);
 
 ssize_t dlp_readlink(const char *path, char *buf, size_t bufsize);
 
+pid_t dlp_gettid();
+
+pid_t dlp_getpid();
+
 #endif // DLIO_PROFILER_POSIX_INTERNAL_H

src/dlio_profiler/writer/chrome_writer.cpp

@@ -78,11 +78,11 @@ dlio_profiler::ChromeWriter::convert_json(std::string &event_name, std::string &
   std::stringstream all_stream;
   int tid, pid;
   if (process_id == -1) {
-    tid = std::hash<std::thread::id>{}(std::this_thread::get_id()) % 100000;
-    pid = getpid();
+    tid = dlp_gettid();
+    pid = dlp_getpid();
   } else {
     pid = process_id;
-    tid = getpid() + std::hash<std::thread::id>{}(std::this_thread::get_id()) % 100000;
+    tid = dlp_getpid() + dlp_gettid();
   }
   auto start_sec = std::chrono::duration<TimeResolution, std::ratio<1>>(start_time);
   auto duration_sec = std::chrono::duration<TimeResolution, std::ratio<1>>(duration);

src/dlio_profiler/writer/chrome_writer.h

@@ -60,7 +60,7 @@ namespace dlio_profiler {
           if (enable_core_affinity_str != nullptr && strcmp(enable_core_affinity_str, "1") == 0) {
             enable_core_affinity = true;
           }
-          process_id = getpid();
+          process_id = dlp_getpid();
           this->fd = fd;
           if (enable_core_affinity) {
             hwloc_topology_init(&topology);  // initialization