File indexing completed on 2025-01-18 09:43:01
0001
0002
0003
0004
0005
0006
0007
0008
0009
0010 #ifndef boost_numeric_ublas_opencl_transpose_hpp_
0011 #define boost_numeric_ublas_opencl_transpose_hpp_
0012
0013 #include <boost/numeric/ublas/opencl/library.hpp>
0014 #include <boost/numeric/ublas/opencl/vector.hpp>
0015 #include <boost/numeric/ublas/opencl/matrix.hpp>
0016
0017
0018 #define OPENCL_TRANSPOSITION_KERNEL(DATA_TYPE) \
0019 "__kernel void transpose(__global " #DATA_TYPE "* in, __global " #DATA_TYPE "* result, unsigned int width, unsigned int height) \n" \
0020 "{ \n" \
0021 " unsigned int column_index = get_global_id(0); \n" \
0022 " unsigned int row_index = get_global_id(1); \n" \
0023 " if (column_index < width && row_index < height) \n" \
0024 " { \n" \
0025 " unsigned int index_in = column_index + width * row_index; \n" \
0026 " unsigned int index_result = row_index + height * column_index; \n" \
0027 " result[index_result] = in[index_in]; \n" \
0028 " } \n" \
0029 "} \n"
0030
0031
0032 namespace boost { namespace numeric { namespace ublas { namespace opencl {
0033
0034 template<class T, class L1, class L2>
0035 typename std::enable_if<is_numeric<T>::value>::type
0036 change_layout(ublas::matrix<T, L1, opencl::storage> const &m,
0037 ublas::matrix<T, L2, opencl::storage> &result,
0038 compute::command_queue& queue)
0039 {
0040 assert(m.size1() == result.size1() && m.size2() == result.size2());
0041 assert(m.device() == result.device() && m.device() == queue.get_device());
0042 assert(!(std::is_same<L1, L2>::value));
0043 char const *kernel;
0044 if (std::is_same<T, float>::value)
0045 kernel = OPENCL_TRANSPOSITION_KERNEL(float);
0046 else if (std::is_same<T, double>::value)
0047 kernel = OPENCL_TRANSPOSITION_KERNEL(double);
0048 else if (std::is_same<T, std::complex<float>>::value)
0049 kernel = OPENCL_TRANSPOSITION_KERNEL(float2);
0050 else if (std::is_same<T, std::complex<double>>::value)
0051 kernel = OPENCL_TRANSPOSITION_KERNEL(double2);
0052 size_t len = strlen(kernel);
0053 cl_int err;
0054 cl_context c_context = queue.get_context().get();
0055 cl_program program = clCreateProgramWithSource(c_context, 1, &kernel, &len, &err);
0056 clBuildProgram(program, 1, &queue.get_device().get(), NULL, NULL, NULL);
0057 cl_kernel c_kernel = clCreateKernel(program, "transpose", &err);
0058 size_t width = std::is_same < L1, ublas::basic_row_major<>>::value ? m.size2() : m.size1();
0059 size_t height = std::is_same < L1, ublas::basic_row_major<>>::value ? m.size1() : m.size2();
0060 size_t global_size[2] = { width , height };
0061 clSetKernelArg(c_kernel, 0, sizeof(T*), &m.begin().get_buffer().get());
0062 clSetKernelArg(c_kernel, 1, sizeof(T*), &result.begin().get_buffer().get());
0063 clSetKernelArg(c_kernel, 2, sizeof(unsigned int), &width);
0064 clSetKernelArg(c_kernel, 3, sizeof(unsigned int), &height);
0065 cl_command_queue c_queue = queue.get();
0066 cl_event event = NULL;
0067 clEnqueueNDRangeKernel(c_queue, c_kernel, 2, NULL, global_size, NULL, 0, NULL, &event);
0068 clWaitForEvents(1, &event);
0069 }
0070
0071 template<class T, class L1, class L2, class A>
0072 typename std::enable_if<is_numeric<T>::value>::type
0073 change_layout(ublas::matrix<T, L1, A> const &m,
0074 ublas::matrix<T, L2, A> &result,
0075 compute::command_queue& queue)
0076 {
0077 ublas::matrix<T, L1, opencl::storage> mdev(m, queue);
0078 ublas::matrix<T, L2, opencl::storage> rdev(result.size1(), result.size2(), queue.get_context());
0079 change_layout(mdev, rdev, queue);
0080 rdev.to_host(result, queue);
0081 }
0082
0083 template<class T, class L>
0084 typename std::enable_if<is_numeric<T>::value>::type
0085 trans(ublas::matrix<T, L, opencl::storage> const &m,
0086 ublas::matrix<T, L, opencl::storage> &result,
0087 compute::command_queue& queue)
0088 {
0089 assert(m.size1() == result.size2() && m.size2() == result.size1());
0090 assert(m.device() == result.device() && m.device() == queue.get_device());
0091 char const *kernel;
0092 if (std::is_same<T, float>::value)
0093 kernel = OPENCL_TRANSPOSITION_KERNEL(float);
0094 else if (std::is_same<T, double>::value)
0095 kernel = OPENCL_TRANSPOSITION_KERNEL(double);
0096 else if (std::is_same<T, std::complex<float>>::value)
0097 kernel = OPENCL_TRANSPOSITION_KERNEL(float2);
0098 else if (std::is_same<T, std::complex<double>>::value)
0099 kernel = OPENCL_TRANSPOSITION_KERNEL(double2);
0100 size_t len = strlen(kernel);
0101 cl_int err;
0102 cl_context c_context = queue.get_context().get();
0103 cl_program program = clCreateProgramWithSource(c_context, 1, &kernel, &len, &err);
0104 clBuildProgram(program, 1, &queue.get_device().get(), NULL, NULL, NULL);
0105 cl_kernel c_kernel = clCreateKernel(program, "transpose", &err);
0106 size_t width = std::is_same <L, ublas::basic_row_major<>>::value ? m.size2() : m.size1();
0107 size_t height = std::is_same <L, ublas::basic_row_major<>>::value ? m.size1() : m.size2();
0108 size_t global_size[2] = { width , height };
0109 clSetKernelArg(c_kernel, 0, sizeof(T*), &m.begin().get_buffer().get());
0110 clSetKernelArg(c_kernel, 1, sizeof(T*), &result.begin().get_buffer().get());
0111 clSetKernelArg(c_kernel, 2, sizeof(unsigned int), &width);
0112 clSetKernelArg(c_kernel, 3, sizeof(unsigned int), &height);
0113 cl_command_queue c_queue = queue.get();
0114 cl_event event = NULL;
0115 clEnqueueNDRangeKernel(c_queue, c_kernel, 2, NULL, global_size, NULL, 0, NULL, &event);
0116 clWaitForEvents(1, &event);
0117 }
0118
0119 template<class T, class L, class A>
0120 typename std::enable_if<is_numeric<T>::value>::type
0121 trans(ublas::matrix<T, L, A> const &m,
0122 ublas::matrix<T, L, A> &result,
0123 compute::command_queue& queue)
0124 {
0125 ublas::matrix<T, L, opencl::storage> mdev(m, queue);
0126 ublas::matrix<T, L, opencl::storage> rdev(result.size1(), result.size2(), queue.get_context());
0127 trans(mdev, rdev, queue);
0128 rdev.to_host(result, queue);
0129 }
0130
0131 template<class T, class L, class A>
0132 typename std::enable_if<is_numeric<T>::value, ublas::matrix<T, L, A>>::type
0133 trans(ublas::matrix<T, L, A>& m, compute::command_queue& queue)
0134 {
0135 ublas::matrix<T, L, A> result(m.size2(), m.size1());
0136 trans(m, result, queue);
0137 return result;
0138 }
0139
0140 }}}}
0141
0142 #endif