PyTorch with C++ 1
Published:
PyTorch is one of the well established libraries for modeling deep neural networks. The exposed Python API is the most commonly used one. However, the library also exposes bindings for C++. In this series of notebooks, I will try to demonstrate how to use the latter. I will be following to a large extent the documentation for the C++ frontend.
PyTorch with C++ 1
I will start by performing some basic manipulations with the Tensor class. Instances of this class are used around the library to perform numerics. Whilst at this I will also show how to link against the C++ bindings.
Install and link against libtorch
To start with, download the binaries from here: https://pytorch.org/get-started/locally/. I use the pre-build binaries. Also make sure that you download the C++11 ABI.
Let’s create now the first C++ PyTorch program. The following program is taken from the official documentation.
#include <torch/torch.h>
#include <iostream>
int main() {
torch::Tensor tensor = torch::eye(3);
std::cout << tensor << std::endl;
return 0;
}
I use the following CMakeLists.txt file to generate the needed Makefile
cmake_minimum_required(VERSION 3.0 FATAL_ERROR)
PROJECT(example_1 VERSION 1.0.0 LANGUAGES CXX)
SET(SOURCE example_1.cpp)
SET(EXECUTABLE example_1)
# default optionsSET(BUILD_SHARED_LIBS ON)
SET(CMAKE_BUILD_TYPE "Debug")
SET(CMAKE_CXX_COMPILER g++)
SET(CMAKE_CXX_STANDARD 20)
SET(CMAKE_CXX_STANDARD_REQUIRED True)
SET(CMAKE_C_COMPILER gcc)
SET(CMAKE_LINKER_FLAGS "-pthread")
LIST(APPEND CMAKE_PREFIX_PATH /home/alex/MySoftware/libtorch)
FIND_PACKAGE(Torch REQUIRED CONFIG)
MESSAGE(STATUS "TORCH Include directory ${TORCH_INCLUDE_DIRS}")
MESSAGE(STATUS "Build type: ${CMAKE_BUILD_TYPE}")
MESSAGE(STATUS "C++ Compiler: ${CMAKE_CXX_COMPILER}")
MESSAGE(STATUS "C Compiler: ${CMAKE_C_COMPILER}")
INCLUDE_DIRECTORIES(${TORCH_INCLUDE_DIRS})
ADD_EXECUTABLE(${EXECUTABLE} ${SOURCE})
TARGET_LINK_LIBRARIES(${EXECUTABLE} ${TORCH_LIBRARIES})
The program above produces the following output when built and executed:
1 0 0
0 1 0
0 0 1
[ CPUFloatType{3,3} ]
I will now extend the example above to check on the provided API. Still, this is very elementary. Here is the updated example
#include <torch/torch.h>
#include <iostream>
#include <vector>
int main() {
if(torch::cuda::is_available()){
std::cout<<"CUDA is available on this machine"<<std::endl;
}
else{
std::cout<<"CUDA is not available on this machine"<<std::endl;
}
torch::Tensor tensor = torch::eye(3);
std::cout << tensor << std::endl;
std::vector<double> data(3, 2.0);
auto tensor_from_data_1 = torch::tensor(data);
std::cout << tensor_from_data_1 << std::endl;
data[0] = data[1] = data[2] = 1.0;
auto tensor_from_data_2 = torch::tensor(data);
std::cout << tensor_from_data_2 << std::endl;
auto sum = tensor_from_data_2 + tensor_from_data_1;
std::cout << sum << std::endl;
if(torch::cuda::is_available()){
// create a tensor and send it to the GPU
auto cuda_tensor = torch::tensor({1.0, 2.0, 3.0}).to("cuda");
}
// compute element-wise product
auto tensor1 = torch::tensor({1.0, 2.0, 3.0});
auto product = tensor1 * tensor1;
std::cout << product << std::endl;
return 0;
}
The output of the program above is shown below
CUDA is not available on this machine
1 0 0
0 1 0
0 0 1
[ CPUFloatType{3,3} ]
2
2
2
[ CPUFloatType{3} ]
1
1
1
[ CPUFloatType{3} ]
3
3
3
[ CPUFloatType{3} ]
1
4
9
[ CPUFloatType{3} ]
The driver code above can be found in this github repository.