refactor: Restructured python perf benchmark tool

examples/benchmark/BUILD → tools/cpp_benchmark/BUILD

@@ -1,7 +1,7 @@
 package(default_visibility = ["//visibility:public"])
 
 cc_binary(
-    name = "benchmark",
+    name = "cpp_benchmark",
     srcs = [
         "main.cpp",
         "timer.h",

examples/benchmark/README.md → tools/cpp_benchmark/README.md

@@ -9,13 +9,13 @@ Run with bazel:
 > Note: Make sure libtorch and TensorRT are in your LD_LIBRARY_PATH before running, if you need a location you can `export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:[WORKSPACE ROOT]/bazel-Torch-TensorRT/external/libtorch/lib:[WORKSPACE ROOT]/bazel-Torch-TensorRT/external/tensorrt/lib`
 
 ``` sh
-bazel run //cpp/benchmark --cxxopt="-DNDEBUG" --cxxopt="-DJIT" --cxxopt="-DTRT" -- [PATH TO JIT MODULE FILE] [INPUT SIZE (explicit batch)]
+bazel run //tools/cpp_benchmark/int8/benchmark --cxxopt="-DNDEBUG" --cxxopt="-DJIT" --cxxopt="-DTRT" -- [PATH TO JIT MODULE FILE] [INPUT SIZE (explicit batch)]
 ```
 
 For example:
 
 ``` shell
-bazel run //cpp/benchmark  --cxxopt="-DNDEBUG" --cxxopt="-DJIT" --cxxopt="-DTRT" -- $(realpath /tests/models/resnet50.jit.pt) "(32 3 224 224)"
+bazel run //tools/cpp_benchmark/int8/benchmark  --cxxopt="-DNDEBUG" --cxxopt="-DJIT" --cxxopt="-DTRT" -- $(realpath /tests/models/resnet50.jit.pt) "(32 3 224 224)"
 ```
 
 ### Options

examples/int8/datasets/cifar10.cpp → tools/cpp_benchmark/int8/datasets/cifar10.cpp

@@ -1,4 +1,4 @@
-#include "examples/int8/datasets/cifar10.h"
+#include "tools/cpp_benchmark/int8/datasets/cifar10.h"
 #include "torch/data/example.h"
 #include "torch/torch.h"
 #include "torch/types.h"

examples/int8/ptq/BUILD → tools/cpp_benchmark/int8/ptq/BUILD

@@ -13,8 +13,8 @@ cc_binary(
     ],
     deps = [
         "//cpp:torch_tensorrt",
-        "//examples/int8/benchmark",
-        "//examples/int8/datasets:cifar10",
+        "//tools/cpp_benchmark/int8/benchmark",
+        "//tools/cpp_benchmark/int8/datasets:cifar10",
         "@libtorch",
         "@libtorch//:caffe2",
         "@tensorrt//:nvinfer",

examples/int8/ptq/README.md → tools/cpp_benchmark/int8/ptq/README.md

@@ -13,7 +13,7 @@ LibTorch provides a `Dataloader` and `Dataset` API which steamlines preprocessin
 Here is an example interface of a `torch::Dataset` class for CIFAR10:
 
 ```C++
-//examples/int8/ptq/datasets/cifar10.h
+//tools/cpp_benchmark/int8/ptq/datasets/cifar10.h
 #pragma once
 
 #include "torch/data/datasets/base.h"
@@ -120,19 +120,19 @@ This is a short example application that shows how to use Torch-TensorRT to perf
 ## Prerequisites
 
 1. Download CIFAR10 Dataset Binary version ([https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz](https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz))
-2. Train a network on CIFAR10 (see `examples/int8/training/vgg16/README.md` for a VGG16 recipe)
+2. Train a network on CIFAR10 (see `examples/training/vgg16/README.md` for a VGG16 recipe)
 3. Export model to torchscript
 
 ## Compilation using bazel
 
 ``` shell
-bazel run //examples/int8/ptq --compilation_mode=opt <path-to-module> <path-to-cifar10>
+bazel run //tools/cpp_benchmark/int8/ptq --compilation_mode=opt <path-to-module> <path-to-cifar10>
 ```
 
 If you want insight into what is going under the hood or need debug symbols
 
 ``` shell
-bazel run //examples/int8/ptq --compilation_mode=dbg <path-to-module> <path-to-cifar10>
+bazel run //tools/cpp_benchmark/int8/ptq --compilation_mode=dbg <path-to-module> <path-to-cifar10>
 ```
 
 This will build a binary named `ptq` in `bazel-out/k8-<opt|dbg>/bin/cpp/int8/ptq/` directory. Optionally you can add this to `$PATH` environment variable to run `ptq` from anywhere on your system.
@@ -166,7 +166,7 @@ We import header files `cifar10.h` and `benchmark.h` from `ROOT_DIR`. `ROOT_DIR`
 By default it is set to `../../../`. If your Torch-TensorRT directory structure is different, please set `ROOT_DIR` accordingly.
 
 ```sh
-cd examples/int8/ptq
+cd tools/cpp_benchmark/int8/ptq
 # This will generate a ptq binary
 make ROOT_DIR=<PATH> CUDA_VERSION=11.1
 ./ptq <path-to-module> <path-to-cifar10>

examples/int8/ptq/main.cpp → tools/cpp_benchmark/int8/ptq/main.cpp

@@ -3,8 +3,8 @@
 #include "torch_tensorrt/ptq.h"
 #include "torch_tensorrt/torch_tensorrt.h"
 
-#include "examples/int8/benchmark/benchmark.h"
-#include "examples/int8/datasets/cifar10.h"
+#include "tools/cpp_benchmark/int8/benchmark/benchmark.h"
+#include "tools/cpp_benchmark//int8/datasets/cifar10.h"
 
 #include <sys/stat.h>
 #include <iostream>

examples/int8/qat/BUILD → tools/cpp_benchmark/int8/qat/BUILD

@@ -13,8 +13,8 @@ cc_binary(
     ],
     deps = [
         "//cpp:torch_tensorrt",
-        "//examples/int8/benchmark",
-        "//examples/int8/datasets:cifar10",
+        "//tools/cpp_benchmark/int8/benchmark",
+        "//tools/cpp_benchmark/int8/datasets:cifar10",
         "@libtorch",
         "@libtorch//:caffe2",
         "@tensorrt//:nvinfer",

examples/int8/qat/README.md → tools/cpp_benchmark/int8/qat/README.md

@@ -12,21 +12,21 @@ This is a short example application that shows how to use Torch-TensorRT to perf
 ## Prerequisites
 
 1. Download CIFAR10 Dataset Binary version ([https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz](https://www.cs.toronto.edu/~kriz/cifar-10-binary.tar.gz))
-2. Train a network on CIFAR10 and perform quantization aware training on it. Refer to `examples/int8/training/vgg16/README.md` for detailed instructions.
+2. Train a network on CIFAR10 and perform quantization aware training on it. Refer to `examples/training/vgg16/README.md` for detailed instructions.
    Export the QAT model to Torchscript.
 3. Install NVIDIA's <a href="https://github.com/NVIDIA/TensorRT/tree/master/tools/pytorch-quantization">pytorch quantization toolkit</a>
 4. TensorRT 8.0.1.6 or above
 
 ## Compilation using bazel
 
 ``` shell
-bazel run //examples/int8/qat --compilation_mode=opt <path-to-module> <path-to-cifar10>
+bazel run //tools/cpp_benchmark/int8/qat --compilation_mode=opt <path-to-module> <path-to-cifar10>
 ```
 
 If you want insight into what is going under the hood or need debug symbols
 
 ``` shell
-bazel run //examples/int8/qat --compilation_mode=dbg <path-to-module> <path-to-cifar10>
+bazel run //tools/cpp_benchmark/int8/qat --compilation_mode=dbg <path-to-module> <path-to-cifar10>
 ```
 
 This will build a binary named `qat` in `bazel-out/k8-<opt|dbg>/bin/cpp/int8/qat/` directory. Optionally you can add this to `$PATH` environment variable to run `qat` from anywhere on your system.
@@ -60,7 +60,7 @@ We import header files `cifar10.h` and `benchmark.h` from `ROOT_DIR`. `ROOT_DIR`
 By default it is set to `../../../`. If your Torch-TensorRT directory structure is different, please set `ROOT_DIR` accordingly.
 
 ```sh
-cd examples/int8/qat
+cd tools/cpp_benchmark/int8/qat
 # This will generate a ptq binary
 make ROOT_DIR=<PATH> CUDA_VERSION=11.1
 ./qat <path-to-module> <path-to-cifar10>

examples/int8/qat/main.cpp → tools/cpp_benchmark/int8/qat/main.cpp

@@ -2,8 +2,8 @@
 #include "torch/torch.h"
 #include "torch_tensorrt/torch_tensorrt.h"
 
-#include "examples/int8/benchmark/benchmark.h"
-#include "examples/int8/datasets/cifar10.h"
+#include "tools/cpp_benchmark/int8/benchmark/benchmark.h"
+#include "tools/cpp_benchmark/int8/datasets/cifar10.h"
 
 #include <sys/stat.h>
 #include <iostream>