4 Replies Latest reply on Jun 1, 2010 9:46 AM by omkaranathan

    Cannot find 64bit length function at link time on the CPU


      Hello again,

      I've got a kernel that compiles fine but when it comes to actually use the kernel I get an error

      symbol lookup error: : /tmp/OCLHdvABy.so

      undefined symbol: __length_4f64

      I'm running this on the 2.1 stream SDK, and on the CPU, with both of the pragma's defined

      #pragma OPENCL EXTENSION cl_khr_fp64 : enable

      #pragma OPENCL EXTENSION cl_amd_fp64 : enable

      Is the function not supported yet? I thought from knowledge base article KB88 it is supported on the CPU.


      The following set of double-precision floating point functionality is supported  in OpenCL™ C kernels for x86 CPUs

      The following geometric functions: minmaxclampdegreesradiansstepsmoothstepsigndot, length.

      Edited: Added a minimal example code


      #include <iostream> //The OpenCL C++ bindings, with exceptions #define __CL_ENABLE_EXCEPTIONS #include <CL/cl.hpp> std::string KernelSrc = "#pragma OPENCL EXTENSION cl_khr_fp64 : enable\n" "#pragma OPENCL EXTENSION cl_amd_fp64 : enable\n" "__kernel void BadKernel(__global double4* Input, __global double* Output)\n" "{ Output[get_global_id(0)] = length(Input[get_global_id(0)]); }"; int main(int argc, char * argv[]) { try { std::vector<cl::Platform> platforms; cl::Platform::get(&platforms); std::vector<cl::Device> devices; platforms[0].getDevices(CL_DEVICE_TYPE_CPU, &devices); devices.resize(1); std::cout << "Using device " << devices[0].getInfo<CL_DEVICE_NAME>() << std::endl; cl::Context context(devices); cl::Program::Sources source(1, std::pair<const char *, size_t>(KernelSrc.c_str(), KernelSrc.size())); cl::Program program(context, source); try { program.build(devices); } catch (cl::Error& err) { std::cerr << "Building failed, " << err.what() << "(" << err.err() << ")\n" << program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[0]) << "\n"; return -1; } cl::CommandQueue CmdQ(context, devices[0]); const size_t WrkGrpSize = 64; const size_t N = WrkGrpSize * 10; cl::Buffer InputBuffer(context, CL_MEM_ALLOC_HOST_PTR, sizeof(cl_double4) * N); cl_double4* Input = (cl_double4*)CmdQ.enqueueMapBuffer(InputBuffer, true, CL_MAP_WRITE, 0, N * sizeof(cl_double4)); for (size_t index(0); index < N; ++index) for (size_t dim(0); dim < 4; ++dim) Input[index].s[dim] = dim; CmdQ.enqueueUnmapMemObject(InputBuffer, (void*)Input); cl::Buffer OutputBuffer(context, CL_MEM_ALLOC_HOST_PTR, sizeof(cl_double) * N); cl::Kernel BadKernel = cl::Kernel(program, "BadKernel"); cl::KernelFunctor BadKernelFunc(BadKernel.bind(CmdQ, cl::NDRange(N), cl::NDRange(WrkGrpSize))); /*************** Launch the kernel ****************/ //This causes a linker error BadKernelFunc(InputBuffer, OutputBuffer).wait(); std::cout << "Finished without error!\n"; } catch (cl::Error& err) { std::cerr << "An OpenCL error occured, " << err.what() << "\nError num of " << err.err() << "\n"; return -1; } }