LINK : fatal error LNK1181: cannot open input file 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\lib\x64.obj'

Question

I have a piece of code:

#include <CL/cl.h>
#include <stdlib.h>

void axpy(float a, float* x, float* y, int n) {
  cl_context context;
  cl_command_queue queue;
  cl_mem x_buffer;
  cl_mem y_buffer;
  cl_program program;
  cl_kernel kernel;

  // Get the list of available OpenCL devices.
  cl_device_id* devices;
  cl_uint num_devices;
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_GPU, 0, NULL, &num_devices);
  devices = (cl_device_id*)malloc(num_devices * sizeof(cl_device_id));
  clGetDeviceIDs(NULL, CL_DEVICE_TYPE_GPU, num_devices, devices, NULL);

  // Create a context and command queue.
  cl_int err;
  context = clCreateContext(NULL, num_devices, devices, NULL, NULL, &err);
  queue = clCreateCommandQueue(context, devices[0], 0, &err);

  // Create buffers for the input and output vectors.
  x_buffer = clCreateBuffer(context, CL_MEM_READ_ONLY, n * sizeof(float), NULL, &err);
  y_buffer = clCreateBuffer(context, CL_MEM_WRITE_ONLY, n * sizeof(float), NULL, &err);

  // Write the input vectors to the buffers.
  clEnqueueWriteBuffer(queue, x_buffer, CL_TRUE, 0, n * sizeof(float), x, 0, NULL, NULL);

  // Create the kernel object.
  const char* kernel_source =
    "__kernel void axpy(__global float* x, __global float* y, float a, int n) {\
      int i = get_global_id(0);\
      if (i < n) {\
        y[i] = a * x[i] + y[i];\
      }\
    }";
  program = clCreateProgramWithSource(context, 1, &kernel_source, NULL, &err);
  err = clBuildProgram(program, 1, devices, NULL, NULL, NULL);

  // Create the kernel object.
  kernel = clCreateKernel(program, "axpy", &err);

  // Set the kernel arguments.
  clSetKernelArg(kernel, 0, sizeof(cl_mem), &x_buffer);
  clSetKernelArg(kernel, 1, sizeof(cl_mem), &y_buffer);
  clSetKernelArg(kernel, 2, sizeof(float), &a);
  clSetKernelArg(kernel, 3, sizeof(int), &n);

  // Execute the kernel.
  size_t global_work_size = n;
  clEnqueueNDRangeKernel(queue, kernel, 1, NULL, &global_work_size, NULL, 0, NULL, NULL);

  // Wait for the kernel to finish executing.
  clFinish(queue);

  // Read the output vector from the buffer.
  clEnqueueReadBuffer(queue, y_buffer, CL_TRUE, 0, n * sizeof(float), y, 0, NULL, NULL);

  // Release the resources.
  clReleaseMemObject(x_buffer);
  clReleaseMemObject(y_buffer);
  clReleaseProgram(program);
  clReleaseKernel(kernel);
  clReleaseCommandQueue(queue);
  clReleaseContext(context);

  // Free the list of devices.
  free(devices);
}

int main(int argc, char** argv) {
  // Your code here
  return 0;
}

that I am trying to compile with

cl /EHsc axpy.cpp /I"%CUDA_PATH%\include" /link "C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v12.2/lib/x64"

however, I get the error message:

LINK : fatal error LNK1181: cannot open input file 'C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\lib\x64.obj'

I would appreciate it if you could help me know how to resolve the problem.

Answer 1

I think I have figured my problem out. I should have specified the libraries that I need to link to specifically, instead of just pointing to the directory. Here is a correct way to specify the link options for cl.exe:

cl /EHsc axpy.cpp /I"%CUDA_PATH%\include" /link /LIBPATH:"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.2\lib\x64" OpenCL.lib

In case we need to link with multiple libraries, we can specify them one after another:

cl /EHsc axpy.cpp /I"%CUDA_PATH%\include" /link /LIBPATH:"path\to\libraries" lib1.lib lib2.lib lib3.lib

and to get rid of the

cl_version.h: CL_TARGET_OPENCL_VERSION is not defined. Defaulting to 300 (OpenCL 3.0)

warning, I simply added

#define CL_TARGET_OPENCL_VERSION 120

at the top of the code.