Debugging CUDA kernels with VSCode - c++

Question

I am trying to debug a CUDA application on VSCode.
Configuration:
Ubuntu 20.04
VSCode 1.56.2
CUDA 11.3
gcc/g++ 9.3
For this, I have the following (up to date) extensions:
C/C++
Nsight Visual Studio Code Edition
I have no problem compiling/running my program. However when I'm debugging, I am able to use the debugger correctly in CPU side but not on GPU side. Indeed, when I'm trying to add a breakpoint anywhere in the kernel, running the program always moves my breakpoint to the closing bracket and I can't see the variables.
Here are the files:
add.cu
#include <iostream>
#include <math.h>
// Kernel function to add the elements of two arrays
__global__
void add(float *x, float *y)
{
y[blockIdx.x] = x[blockIdx.x] + y[blockIdx.x];
}
int main(void)
{
const int N = 1<<20;
float *x, *y;
// Allocate Unified Memory – accessible from CPU or GPU
cudaMallocManaged(&x, N*sizeof(float));
cudaMallocManaged(&y, N*sizeof(float));
// initialize x and y arrays on the host
for (int i = 0; i < N; i++) {
x[i] = 1.0f;
y[i] = 2.0f;
}
// Run kernel on 1M elements on the GPU
add<<<N, 1>>>(x, y);
// Wait for GPU to finish before accessing on host
cudaDeviceSynchronize();
// Check for errors (all values should be 3.0f)
float maxError = 0.0f;
for (int i = 0; i < N; i++)
maxError = fmax(maxError, fabs(y[i]-3.0f));
std::cout << "Max error: " << maxError << std::endl;
// Free memory
cudaFree(x);
cudaFree(y);
return 0;
}
CMakeLists.txt
cmake_minimum_required(VERSION 3.10)
# set the project name
project(add CUDA)
# add the executable
add_executable(add add.cu)
.vscode/launch.json
{
// Use IntelliSense to learn about possible attributes.
// Hover to view descriptions of existing attributes.
// For more information, visit: https://go.microsoft.com/fwlink/?linkid=830387
"version": "0.2.0",
"configurations": [
{
"name": "CUDA C++: Launch",
"type": "cuda-gdb",
"request": "launch",
"program": "${workspaceFolder}/build/add"
}
]
}
I'm compiling this way:
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Debug ..
cmake --build .
Every program I am trying do the exact same thing on any kernel breakpoint.
What do I miss here?

Answer 1

I finally solved it by forcing -G flag when in debug mode in cmake, adding the following lines after add_executable:
if(CMAKE_BUILD_TYPE STREQUAL "Debug")
target_compile_options(add PRIVATE $<$<COMPILE_LANGUAGE:CUDA>:-G>)
endif()
With this, debugging on device works.
Thanks #RobertCrovella for the tip.