Adding cusolverSp

Accelerated Computing CUDA CUDA NVCC Compiler
1

Good morning, I’m trying to solve a least square problem on GPU but I’m fighting against some problem with external libraries.
I’m using CUDA 12 on a NVIDIA GeForce RTX 3060 Laptop GPU.

#include <cuda.h>
#include <cuda_runtime.h>
#include <cusolverSp.h>

int main(int argc, char** argv) {
  // Set up the input data
  int m = 1000;
  int n = 100;
  float* A = new float[m * n];
  float* b = new float[m];
  // Initialize A and b with some values

  // Allocate device memory for the input data
  float* d_A;
  cudaMalloc((void**)&d_A, m * n * sizeof(float));
  float* d_b;
  cudaMalloc((void**)&d_b, m * sizeof(float));

  // Copy the input data to the device
  cudaMemcpy(d_A, A, m * n * sizeof(float), cudaMemcpyHostToDevice);
  cudaMemcpy(d_b, b, m * sizeof(float), cudaMemcpyHostToDevice);

  // Create the CUSOLVER context
  cusolverSpHandle_t solver;
  cusolverSpCreate(&solver);

  // Allocate device memory for the solution
  float* d_x;
  cudaMalloc((void**)&d_x, n * sizeof(float));

  // Set up the CUSOLVER LSQR parameters
  cusolverSpLsqrParams params;
  params.tol = 1e-6;
  params.max_iter = 1000;
  params.normA = 0;  // Automatically compute the norm of A
  params.normB = 0;  // Automatically compute the norm of b

  // Solve the least squares problem
  int status = cusolverSpLsqr(solver, m, n, &params, d_A, d_b, d_x);
  if (status != CUSOLVER_STATUS_SUCCESS) {
    // Handle error
  }

  // Copy the solution from the device to the host
  float* x = new float[n];
  cudaMemcpy(x, d_x, n * sizeof(float), cudaMemcpyDeviceToHost);

  // Clean up
  delete[] A;
  delete[] b;
  delete[] x;
  cudaFree(d_A);
  cudaFree(d_b);
  cudaFree(d_x);
  cusolverSpDestroy(solver);
}

During the compilation I get these errors:

../src/QR.cu(32): error: identifier "cusolverSpLsqrParams" is undefined

../src/QR.cu(39): error: identifier "cusolverSpLsqr" is undefined

The compile command generated by ECLIPSE is the following:

Invoking: NVCC Compiler
/usr/local/cuda/bin/nvcc -I../cublas --device-debug --debug -gencode arch=compute_52,code=sm_52 -gencode arch=compute_52,code=compute_52 -ccbin g++ -c -o "src/QR.o" "../src/QR.cu"

I’m pretty sure that there is some error in the libraries linking, but I have found no solution on the web.
Can someone indicate my how to fix this problem?

EDIT:
I understood that the support for cublas has been removed after cuda 10.
But this version with cusolver should work aslo for cuda 12.
Any Hint?

Thank you very much.