double run_benchmark(size_t matrix_size)
{
//
// One alternative: Put the matrices into a contiguous block of memory (allows to use viennacl::fast_copy(), avoiding temporary memory)
//
std::vector<ScalarType> stl_B(matrix_size * matrix_size);
std::vector<ScalarType> stl_C(matrix_size * matrix_size);
//
// Fill the matrix
//
for (unsigned int i = 0; i < matrix_size; ++i)
for (unsigned int j = 0; j < matrix_size; ++j)
stl_B[i*matrix_size + j] = random<ScalarType>();
for (unsigned int i = 0; i < matrix_size; ++i)
for (unsigned int j = 0; j < matrix_size; ++j)
stl_C[i + j*matrix_size] = random<ScalarType>();
//viennacl::ocl::current_context().build_options("-cl-mad-enable -cl-fast-relaxed-math"); //uncomment for additional optimizations
//viennacl::ocl::current_context().build_options("-cl-opt-disable"); //uncomment to get poor performance
viennacl::matrix<ScalarType> vcl_A(matrix_size, matrix_size);
viennacl::matrix<ScalarType,FB> vcl_B(matrix_size, matrix_size);
viennacl::matrix<ScalarType,FC> vcl_C(matrix_size, matrix_size);
typedef viennacl::generator::matrix< viennacl::matrix<ScalarType> > dma_t;
typedef viennacl::generator::matrix< viennacl::matrix<ScalarType,FB> > dmb_t;
typedef viennacl::generator::matrix< viennacl::matrix<ScalarType,FC> > dmc_t;
viennacl::fast_copy(&(stl_B[0]),
&(stl_B[0]) + stl_B.size(),
vcl_B);
viennacl::fast_copy(&(stl_C[0]),
&(stl_C[0]) + stl_C.size(),
vcl_C);
viennacl::generator::custom_operation op;
op.add(dma_t(vcl_A) = viennacl::generator::prod(dmb_t(vcl_B), dmc_t(vcl_C)));
op.program();
op.execute();
viennacl::backend::finish();
double res = 0;
Timer timer;
timer.start();
for(unsigned int r = 0 ; r < N_RUNS ; ++r){
op.execute();
}
viennacl::backend::finish();
res = timer.get();
return res/N_RUNS;
}
int main (int, const char **)
{
typedef float ScalarType; //feel free to change this to 'double' if supported by your hardware
typedef boost::numeric::ublas::matrix<ScalarType> MatrixType;
typedef viennacl::matrix<ScalarType, viennacl::row_major> VCLMatrixType;
std::size_t dim_large = 5;
std::size_t dim_small = 3;
//
// Setup ublas objects and fill with data:
//
MatrixType ublas_A(dim_large, dim_large);
MatrixType ublas_B(dim_small, dim_small);
MatrixType ublas_C(dim_large, dim_small);
MatrixType ublas_D(dim_small, dim_large);
for (std::size_t i=0; i<ublas_A.size1(); ++i)
for (std::size_t j=0; j<ublas_A.size2(); ++j)
ublas_A(i,j) = static_cast<ScalarType>((i+1) + (j+1)*(i+1));
for (std::size_t i=0; i<ublas_B.size1(); ++i)
for (std::size_t j=0; j<ublas_B.size2(); ++j)
ublas_B(i,j) = static_cast<ScalarType>((i+1) + (j+1)*(i+1));
for (std::size_t i=0; i<ublas_C.size1(); ++i)
for (std::size_t j=0; j<ublas_C.size2(); ++j)
ublas_C(i,j) = static_cast<ScalarType>((j+2) + (j+1)*(i+1));
for (std::size_t i=0; i<ublas_D.size1(); ++i)
for (std::size_t j=0; j<ublas_D.size2(); ++j)
ublas_D(i,j) = static_cast<ScalarType>((j+2) + (j+1)*(i+1));
//
// Extract submatrices using the ranges in ublas
//
boost::numeric::ublas::range ublas_r1(0, dim_small); //the first 'dim_small' entries
boost::numeric::ublas::range ublas_r2(dim_large - dim_small, dim_large); //the last 'dim_small' entries
boost::numeric::ublas::matrix_range<MatrixType> ublas_A_sub1(ublas_A, ublas_r1, ublas_r1); //upper left part of A
boost::numeric::ublas::matrix_range<MatrixType> ublas_A_sub2(ublas_A, ublas_r2, ublas_r2); //lower right part of A
boost::numeric::ublas::matrix_range<MatrixType> ublas_C_sub(ublas_C, ublas_r1, ublas_r1); //upper left part of C
boost::numeric::ublas::matrix_range<MatrixType> ublas_D_sub(ublas_D, ublas_r1, ublas_r1); //upper left part of D
//
// Setup ViennaCL objects
//
VCLMatrixType vcl_A(dim_large, dim_large);
VCLMatrixType vcl_B(dim_small, dim_small);
VCLMatrixType vcl_C(dim_large, dim_small);
VCLMatrixType vcl_D(dim_small, dim_large);
viennacl::copy(ublas_A, vcl_A);
viennacl::copy(ublas_B, vcl_B);
viennacl::copy(ublas_C, vcl_C);
viennacl::copy(ublas_D, vcl_D);
//
// Extract submatrices using the ranges in ViennaCL
//
viennacl::range vcl_r1(0, dim_small); //the first 'dim_small' entries
viennacl::range vcl_r2(dim_large - dim_small, dim_large); //the last 'dim_small' entries
viennacl::matrix_range<VCLMatrixType> vcl_A_sub1(vcl_A, vcl_r1, vcl_r1); //upper left part of A
viennacl::matrix_range<VCLMatrixType> vcl_A_sub2(vcl_A, vcl_r2, vcl_r2); //lower right part of A
viennacl::matrix_range<VCLMatrixType> vcl_C_sub(vcl_C, vcl_r1, vcl_r1); //upper left part of C
viennacl::matrix_range<VCLMatrixType> vcl_D_sub(vcl_D, vcl_r1, vcl_r1); //upper left part of D
//
// Copy from ublas to submatrices and back:
//
ublas_A_sub1 = ublas_B;
viennacl::copy(ublas_B, vcl_A_sub1);
viennacl::copy(vcl_A_sub1, ublas_B);
//
// Addition:
//
// range to range:
ublas_A_sub2 += ublas_A_sub2;
vcl_A_sub2 += vcl_A_sub2;
// range to matrix:
ublas_B += ublas_A_sub2;
vcl_B += vcl_A_sub2;
//
// use matrix range with matrix-matrix product:
//
ublas_A_sub1 += prod(ublas_C_sub, ublas_D_sub);
vcl_A_sub1 += viennacl::linalg::prod(vcl_C_sub, vcl_D_sub);
//
// Print result matrices:
//
std::cout << "Result ublas: " << ublas_A << std::endl;
std::cout << "Result ViennaCL: " << vcl_A << std::endl;
//
// That's it.
//
std::cout << "!!!! TUTORIAL COMPLETED SUCCESSFULLY !!!!" << std::endl;
return EXIT_SUCCESS;
}
int run_benchmark()
{
Timer timer;
double exec_time;
//
// One alternative: Put the matrices into a contiguous block of memory (allows to use viennacl::fast_copy(), avoiding temporary memory)
//
std::vector<ScalarType> stl_A(BLAS3_MATRIX_SIZE * BLAS3_MATRIX_SIZE);
std::vector<ScalarType> stl_B(BLAS3_MATRIX_SIZE * BLAS3_MATRIX_SIZE);
std::vector<ScalarType> stl_C(BLAS3_MATRIX_SIZE * BLAS3_MATRIX_SIZE);
//
// Fill the matrix
//
for (unsigned int i = 0; i < BLAS3_MATRIX_SIZE; ++i)
for (unsigned int j = 0; j < BLAS3_MATRIX_SIZE; ++j)
stl_A[i*BLAS3_MATRIX_SIZE + j] = random<ScalarType>();
for (unsigned int i = 0; i < BLAS3_MATRIX_SIZE; ++i)
for (unsigned int j = 0; j < BLAS3_MATRIX_SIZE; ++j)
stl_B[i + j*BLAS3_MATRIX_SIZE] = random<ScalarType>();
//
// Set up some ViennaCL objects
//
#ifdef VIENNACL_WITH_OPENCL
viennacl::ocl::set_context_device_type(0, viennacl::ocl::gpu_tag());
#endif
//viennacl::ocl::current_context().build_options("-cl-mad-enable -cl-fast-relaxed-math"); //uncomment for additional optimizations
//viennacl::ocl::current_context().build_options("-cl-opt-disable"); //uncomment to get poor performance
viennacl::matrix<ScalarType> vcl_A(BLAS3_MATRIX_SIZE, BLAS3_MATRIX_SIZE);
viennacl::matrix<ScalarType> vcl_B(BLAS3_MATRIX_SIZE, BLAS3_MATRIX_SIZE);
viennacl::matrix<ScalarType> vcl_C(BLAS3_MATRIX_SIZE, BLAS3_MATRIX_SIZE);
/////////////////////////////////////////////////
//////////// Matrix-matrix products /////////////
/////////////////////////////////////////////////
//
// Now iterate over all OpenCL devices in the context and compute the matrix-matrix product
//
std::cout << " ------ Benchmark 1: Matrix-Matrix product ------ " << std::endl;
#ifdef VIENNACL_WITH_OPENCL
std::vector<viennacl::ocl::device> devices = viennacl::ocl::current_context().devices();
#else
std::vector<long> devices(1);
#endif
for (std::size_t i=0; i<devices.size(); ++i)
{
#ifdef VIENNACL_WITH_OPENCL
viennacl::ocl::current_context().switch_device(devices[i]);
std::cout << " - Device Name: " << viennacl::ocl::current_device().name() << std::endl;
#endif
viennacl::fast_copy(&(stl_A[0]),
&(stl_A[0]) + stl_A.size(),
vcl_A);
viennacl::fast_copy(&(stl_B[0]),
&(stl_B[0]) + stl_B.size(),
vcl_B);
vcl_C = viennacl::linalg::prod(vcl_A, vcl_B);
viennacl::backend::finish();
timer.start();
vcl_C = viennacl::linalg::prod(vcl_A, vcl_B);
viennacl::backend::finish();
exec_time = timer.get();
std::cout << " - Execution time on device (no setup time included): " << exec_time << std::endl;
std::cout << " - GFLOPs (counting multiply&add as separate operations): " << 2.0 * (vcl_A.size1() / 1000.0) * (vcl_A.size2() / 1000.0) * (vcl_B.size2() / 1000.0) / exec_time << std::endl;
std::cout << std::endl;
}
std::cout << " ------ Benchmark 2: Matrix-Matrix product using ranges ------ " << std::endl;
viennacl::range r(BLAS3_MATRIX_SIZE/4, 3 * BLAS3_MATRIX_SIZE/4);
for (std::size_t i=0; i<devices.size(); ++i)
{
#ifdef VIENNACL_WITH_OPENCL
viennacl::ocl::current_context().switch_device(devices[i]);
std::cout << " - Device Name: " << viennacl::ocl::current_device().name() << std::endl;
#endif
viennacl::fast_copy(&(stl_A[0]),
&(stl_A[0]) + stl_A.size(),
vcl_A);
viennacl::fast_copy(&(stl_B[0]),
&(stl_B[0]) + stl_B.size(),
vcl_B);
viennacl::project(vcl_C, r, r) = viennacl::linalg::prod(viennacl::project(vcl_A, r, r), viennacl::project(vcl_B, r, r));
viennacl::backend::finish();
timer.start();
viennacl::project(vcl_C, r, r) = viennacl::linalg::prod(viennacl::project(vcl_A, r, r), viennacl::project(vcl_B, r, r));
viennacl::backend::finish();
exec_time = timer.get();
std::cout << " - Execution time on device (no setup time included): " << exec_time << std::endl;
std::cout << " - GFLOPs (counting multiply&add as separate operations): " << 2.0 * (vcl_A.size1() / 2000.0) * (vcl_A.size2() / 2000.0) * (vcl_B.size2() / 2000.0) / exec_time << std::endl;
std::cout << std::endl;
}
std::cout << " ------ Benchmark 3: Matrix-Matrix product using slices ------ " << std::endl;
viennacl::slice s(0, 2, BLAS3_MATRIX_SIZE/2);
for (std::size_t i=0; i<devices.size(); ++i)
{
#ifdef VIENNACL_WITH_OPENCL
viennacl::ocl::current_context().switch_device(devices[i]);
std::cout << " - Device Name: " << viennacl::ocl::current_device().name() << std::endl;
#endif
viennacl::fast_copy(&(stl_A[0]),
&(stl_A[0]) + stl_A.size(),
vcl_A);
viennacl::fast_copy(&(stl_B[0]),
&(stl_B[0]) + stl_B.size(),
vcl_B);
viennacl::project(vcl_C, s, s) = viennacl::linalg::prod(viennacl::project(vcl_A, s, s), viennacl::project(vcl_B, s, s));
viennacl::backend::finish();
timer.start();
viennacl::project(vcl_C, s, s) = viennacl::linalg::prod(viennacl::project(vcl_A, s, s), viennacl::project(vcl_B, s, s));
viennacl::backend::finish();
exec_time = timer.get();
std::cout << " - Execution time on device (no setup time included): " << exec_time << std::endl;
std::cout << " - GFLOPs (counting multiply&add as separate operations): " << 2.0 * (vcl_A.size1() / 2000.0) * (vcl_A.size2() / 2000.0) * (vcl_B.size2() / 2000.0) / exec_time << std::endl;
std::cout << std::endl;
}
std::cout << " ------ Benchmark 4: LU factorization ------ " << std::endl;
for (std::size_t i=0; i<devices.size(); ++i)
{
#ifdef VIENNACL_WITH_OPENCL
viennacl::ocl::current_context().switch_device(devices[i]);
std::cout << " - Device Name: " << viennacl::ocl::current_device().name() << std::endl;
#endif
viennacl::fast_copy(&(stl_A[0]),
&(stl_A[0]) + stl_A.size(),
vcl_A);
viennacl::linalg::lu_factorize(vcl_A);
viennacl::backend::finish();
timer.start();
viennacl::linalg::lu_factorize(vcl_A);
viennacl::backend::finish();
exec_time = timer.get();
std::cout << " - Execution time on device (no setup time included): " << exec_time << std::endl;
std::cout << " - GFLOPs (counting multiply&add as separate operations): " << 2.0 * (vcl_A.size1() / 1000.0) * (vcl_A.size2() / 1000.0) * (vcl_A.size2() / 1000.0) / exec_time << std::endl;
std::cout << std::endl;
}
return EXIT_SUCCESS;
}
int test_prod(Epsilon const& epsilon)
{
int ret;
viennacl::tools::uniform_random_numbers<NumericT> randomNumber;
std::size_t matrix_size1 = 29; //some odd number, not too large
std::size_t matrix_size2 = 47; //some odd number, not too large
std::size_t matrix_size3 = 33; //some odd number, not too large
//std::size_t matrix_size1 = 128; //some odd number, not too large
//std::size_t matrix_size2 = 64; //some odd number, not too large
//std::size_t matrix_size3 = 128; //some odd number, not too large
//std::size_t matrix_size1 = 256; // for testing AMD kernels
//std::size_t matrix_size2 = 256; // for testing AMD kernels
//std::size_t matrix_size3 = 256; // for testing AMD kernels
// --------------------------------------------------------------------------
// ublas reference:
std::vector<std::vector<NumericT> > A(matrix_size1, std::vector<NumericT>(matrix_size2));
std::vector<std::vector<NumericT> > big_A(4*matrix_size1, std::vector<NumericT>(4*matrix_size2, NumericT(3.1415)));
std::vector<std::vector<NumericT> > B(matrix_size2, std::vector<NumericT>(matrix_size3));
std::vector<std::vector<NumericT> > big_B(4*matrix_size2, std::vector<NumericT>(4*matrix_size3, NumericT(42.0)));
std::vector<std::vector<NumericT> > C(matrix_size1, std::vector<NumericT>(matrix_size3));
//fill A and B:
for (std::size_t i = 0; i < A.size(); ++i)
for (std::size_t j = 0; j < A[0].size(); ++j)
A[i][j] = static_cast<NumericT>(0.1) * randomNumber();
for (std::size_t i = 0; i < B.size(); ++i)
for (std::size_t j = 0; j < B[0].size(); ++j)
B[i][j] = static_cast<NumericT>(0.1) * randomNumber();
std::vector<std::vector<NumericT> > A_trans(A[0].size(), std::vector<NumericT>(A.size()));
for (std::size_t i = 0; i < A.size(); ++i)
for (std::size_t j = 0; j < A[0].size(); ++j)
A_trans[j][i] = A[i][j];
std::vector<std::vector<NumericT> > big_A_trans(big_A[0].size(), std::vector<NumericT>(big_A.size()));
for (std::size_t i = 0; i < big_A.size(); ++i)
for (std::size_t j = 0; j < big_A[0].size(); ++j)
big_A_trans[j][i] = big_A[i][j];
std::vector<std::vector<NumericT> > B_trans(B[0].size(), std::vector<NumericT>(B.size()));
for (std::size_t i = 0; i < B.size(); ++i)
for (std::size_t j = 0; j < B[0].size(); ++j)
B_trans[j][i] = B[i][j];
std::vector<std::vector<NumericT> > big_B_trans(big_B[0].size(), std::vector<NumericT>(big_B.size()));
for (std::size_t i = 0; i < big_B.size(); ++i)
for (std::size_t j = 0; j < big_B[0].size(); ++j)
big_B_trans[j][i] = big_B[i][j];
//
// ViennaCL objects
//
// A
viennacl::range range1_A(matrix_size1, 2*matrix_size1);
viennacl::range range2_A(matrix_size2, 2*matrix_size2);
viennacl::slice slice1_A(matrix_size1, 2, matrix_size1);
viennacl::slice slice2_A(matrix_size2, 3, matrix_size2);
viennacl::matrix<NumericT, F_A> vcl_A(matrix_size1, matrix_size2);
viennacl::copy(A, vcl_A);
viennacl::matrix<NumericT, F_A> vcl_big_range_A(4*matrix_size1, 4*matrix_size2);
viennacl::matrix_range<viennacl::matrix<NumericT, F_A> > vcl_range_A(vcl_big_range_A, range1_A, range2_A);
viennacl::copy(A, vcl_range_A);
viennacl::matrix<NumericT, F_A> vcl_big_slice_A(4*matrix_size1, 4*matrix_size2);
viennacl::matrix_slice<viennacl::matrix<NumericT, F_A> > vcl_slice_A(vcl_big_slice_A, slice1_A, slice2_A);
viennacl::copy(A, vcl_slice_A);
// A^T
viennacl::matrix<NumericT, F_A> vcl_A_trans(matrix_size2, matrix_size1);
viennacl::copy(A_trans, vcl_A_trans);
viennacl::matrix<NumericT, F_A> vcl_big_range_A_trans(4*matrix_size2, 4*matrix_size1);
viennacl::matrix_range<viennacl::matrix<NumericT, F_A> > vcl_range_A_trans(vcl_big_range_A_trans, range2_A, range1_A);
viennacl::copy(A_trans, vcl_range_A_trans);
viennacl::matrix<NumericT, F_A> vcl_big_slice_A_trans(4*matrix_size2, 4*matrix_size1);
viennacl::matrix_slice<viennacl::matrix<NumericT, F_A> > vcl_slice_A_trans(vcl_big_slice_A_trans, slice2_A, slice1_A);
viennacl::copy(A_trans, vcl_slice_A_trans);
// B
viennacl::range range1_B(2*matrix_size2, 3*matrix_size2);
viennacl::range range2_B(2*matrix_size3, 3*matrix_size3);
viennacl::slice slice1_B(matrix_size2, 3, matrix_size2);
viennacl::slice slice2_B(matrix_size3, 2, matrix_size3);
viennacl::matrix<NumericT, F_B> vcl_B(matrix_size2, matrix_size3);
viennacl::copy(B, vcl_B);
viennacl::matrix<NumericT, F_B> vcl_big_range_B(4*matrix_size2, 4*matrix_size3);
viennacl::matrix_range<viennacl::matrix<NumericT, F_B> > vcl_range_B(vcl_big_range_B, range1_B, range2_B);
viennacl::copy(B, vcl_range_B);
viennacl::matrix<NumericT, F_B> vcl_big_slice_B(4*matrix_size2, 4*matrix_size3);
viennacl::matrix_slice<viennacl::matrix<NumericT, F_B> > vcl_slice_B(vcl_big_slice_B, slice1_B, slice2_B);
viennacl::copy(B, vcl_slice_B);
// B^T
viennacl::matrix<NumericT, F_B> vcl_B_trans(matrix_size3, matrix_size2);
viennacl::copy(B_trans, vcl_B_trans);
viennacl::matrix<NumericT, F_B> vcl_big_range_B_trans(4*matrix_size3, 4*matrix_size2);
viennacl::matrix_range<viennacl::matrix<NumericT, F_B> > vcl_range_B_trans(vcl_big_range_B_trans, range2_B, range1_B);
viennacl::copy(B_trans, vcl_range_B_trans);
viennacl::matrix<NumericT, F_B> vcl_big_slice_B_trans(4*matrix_size3, 4*matrix_size2);
viennacl::matrix_slice<viennacl::matrix<NumericT, F_B> > vcl_slice_B_trans(vcl_big_slice_B_trans, slice2_B, slice1_B);
viennacl::copy(B_trans, vcl_slice_B_trans);
// C
viennacl::range range1_C(matrix_size1-1, 2*matrix_size1-1);
viennacl::range range2_C(matrix_size3-1, 2*matrix_size3-1);
viennacl::slice slice1_C(matrix_size1-1, 3, matrix_size1);
viennacl::slice slice2_C(matrix_size3-1, 3, matrix_size3);
viennacl::matrix<NumericT, F_C> vcl_C(matrix_size1, matrix_size3);
viennacl::matrix<NumericT, F_C> vcl_big_range_C(4*matrix_size1, 4*matrix_size3);
viennacl::matrix_range<viennacl::matrix<NumericT, F_C> > vcl_range_C(vcl_big_range_C, range1_C, range2_C);
viennacl::matrix<NumericT, F_C> vcl_big_slice_C(4*matrix_size1, 4*matrix_size3);
viennacl::matrix_slice<viennacl::matrix<NumericT, F_C> > vcl_slice_C(vcl_big_slice_C, slice1_C, slice2_C);
std::cout << "--- Part 1: Testing matrix-matrix products ---" << std::endl;
//////
////// A: matrix
//////
//
//
std::cout << "Now using A=matrix, B=matrix, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_B, vcl_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=matrix, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_B, vcl_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=matrix, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_B, vcl_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=range, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=range, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=range, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=slice, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=slice, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=matrix, B=slice, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_A, vcl_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//////
////// A: range
//////
//
//
std::cout << "Now using A=range, B=matrix, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_B, vcl_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=matrix, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_B, vcl_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=matrix, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_B, vcl_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=range, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=range, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=range, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=slice, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=slice, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=range, B=slice, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_range_A, vcl_range_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//////
////// A: slice
//////
//
//
std::cout << "Now using A=slice, B=matrix, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_B, vcl_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=matrix, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_B, vcl_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=matrix, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_B, vcl_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=range, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=range, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=range, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_range_B, vcl_range_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=slice, C=matrix" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=slice, C=range" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_range_C);
if (ret != EXIT_SUCCESS)
return ret;
//
//
std::cout << "Now using A=slice, B=slice, C=slice" << std::endl;
ret = test_prod<NumericT>(epsilon,
A, A_trans, B, B_trans, C,
vcl_slice_A, vcl_slice_A_trans,
vcl_slice_B, vcl_slice_B_trans,
vcl_slice_C);
if (ret != EXIT_SUCCESS)
return ret;
return ret;
}
int run_test()
{
//typedef float ScalarType;
typedef boost::numeric::ublas::matrix<ScalarType> MatrixType;
typedef boost::numeric::ublas::vector<ScalarType> VectorType;
typedef viennacl::matrix<ScalarType, T> VCLMatrixType;
typedef viennacl::vector<ScalarType> VCLVectorType;
viennacl::scalar<ScalarType> gpu_pi = ScalarType(3.1415);
std::size_t dim_large = 151;
std::size_t dim_small = 37;
//std::size_t dim_large = 35;
//std::size_t dim_small = 11;
//setup ublas objects:
MatrixType ublas_A(dim_large, dim_large);
for (std::size_t i=0; i<ublas_A.size1(); ++i)
for (std::size_t j=0; j<ublas_A.size2(); ++j)
ublas_A(i,j) = ScalarType((i+1) + (j+1)*(i+1));
MatrixType ublas_B(dim_small, dim_small);
for (std::size_t i=0; i<ublas_B.size1(); ++i)
for (std::size_t j=0; j<ublas_B.size2(); ++j)
ublas_B(i,j) = ScalarType((i+1) + (j+1)*(i+1));
MatrixType ublas_C(dim_large, dim_small);
for (std::size_t i=0; i<ublas_C.size1(); ++i)
for (std::size_t j=0; j<ublas_C.size2(); ++j)
ublas_C(i,j) = ScalarType((j+2) + (j+1)*(i+1));
MatrixType ublas_D(dim_small, dim_large);
for (std::size_t i=0; i<ublas_D.size1(); ++i)
for (std::size_t j=0; j<ublas_D.size2(); ++j)
ublas_D(i,j) = ScalarType((j+2) + (j+1)*(i+1));
boost::numeric::ublas::range ublas_r1(0, dim_small);
boost::numeric::ublas::range ublas_r2(dim_large - dim_small, dim_large);
boost::numeric::ublas::matrix_range<MatrixType> ublas_A_sub1(ublas_A, ublas_r1, ublas_r1);
boost::numeric::ublas::matrix_range<MatrixType> ublas_A_sub2(ublas_A, ublas_r2, ublas_r2);
boost::numeric::ublas::matrix_range<MatrixType> ublas_C_sub(ublas_C, ublas_r1, ublas_r1);
boost::numeric::ublas::matrix_range<MatrixType> ublas_D_sub(ublas_D, ublas_r1, ublas_r1);
//Setup ViennaCL objects
VCLMatrixType vcl_A(dim_large, dim_large);
viennacl::copy(ublas_A, vcl_A);
VCLMatrixType vcl_B(dim_small, dim_small);
viennacl::copy(ublas_B, vcl_B);
VCLMatrixType vcl_C(dim_large, dim_small);
viennacl::copy(ublas_C, vcl_C);
VCLMatrixType vcl_D(dim_small, dim_large);
viennacl::copy(ublas_D, vcl_D);
viennacl::range vcl_r1(0, dim_small);
viennacl::range vcl_r2(dim_large - dim_small, dim_large);
viennacl::matrix_range<VCLMatrixType> vcl_A_sub1(vcl_A, vcl_r1, vcl_r1);
viennacl::matrix_range<VCLMatrixType> vcl_A_sub2(vcl_A, vcl_r2, vcl_r2);
viennacl::matrix_range<VCLMatrixType> vcl_C_sub(vcl_C, vcl_r1, vcl_r1);
viennacl::matrix_range<VCLMatrixType> vcl_D_sub(vcl_D, vcl_r1, vcl_r1);
std::cout << std::endl;
std::cout << "//" << std::endl;
std::cout << "////////// Test: Copy CTOR //////////" << std::endl;
std::cout << "//" << std::endl;
{
std::cout << "Testing matrix created from range... ";
ublas_B = ublas_A_sub1;
VCLMatrixType vcl_temp = vcl_A_sub1;
if (check_for_equality(ublas_B, vcl_temp))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Testing range created from range... ";
//ublas_A_sub1 = ublas_A_sub1;
VCLMatrixType vcl_ctor_sub1 = vcl_A_sub1; //Note: This is mostly a compilation test only
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
}
std::cout << std::endl;
std::cout << "//" << std::endl;
std::cout << "////////// Test: Assignments //////////" << std::endl;
std::cout << "//" << std::endl;
std::cout << "Testing matrix assigned to range... ";
ublas_A_sub1 = ublas_B;
vcl_A_sub1 = vcl_B;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Testing range assigned to matrix... ";
ublas_B = ublas_A_sub2;
vcl_B = vcl_A_sub2;
if (check_for_equality(ublas_B, vcl_B))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Testing range assigned to range... ";
ublas_A_sub1 = ublas_C_sub;
vcl_A_sub1 = vcl_C_sub;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << std::endl;
std::cout << "//" << std::endl;
std::cout << "////////// Test 1: Copy to GPU //////////" << std::endl;
std::cout << "//" << std::endl;
ublas_A_sub1 = ublas_B;
viennacl::copy(ublas_B, vcl_A_sub1);
std::cout << "Testing upper left copy to A... ";
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
ublas_A_sub2 = ublas_B;
viennacl::copy(ublas_B, vcl_A_sub2);
std::cout << "Testing lower right copy to A... ";
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
ublas_C_sub = ublas_B;
viennacl::copy(ublas_B, vcl_C_sub);
std::cout << "Testing upper copy to C... ";
if (check_for_equality(ublas_C, vcl_C))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
ublas_D_sub = ublas_B;
viennacl::copy(ublas_B, vcl_D_sub);
std::cout << "Testing left copy to D... ";
if (check_for_equality(ublas_D, vcl_D))
std::cout << "PASSED!" << std::endl;
else
std::cout << std::endl << "TEST failed!";
std::cout << std::endl;
std::cout << "//" << std::endl;
std::cout << "////////// Test 2: Copy from GPU //////////" << std::endl;
std::cout << "//" << std::endl;
std::cout << "Testing upper left copy to A... ";
if (check_for_equality(ublas_A_sub1, vcl_A_sub1))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Testing lower right copy to A... ";
if (check_for_equality(ublas_A_sub2, vcl_A_sub2))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Testing upper copy to C... ";
if (check_for_equality(ublas_C_sub, vcl_C_sub))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Testing left copy to D... ";
if (check_for_equality(ublas_D_sub, vcl_D_sub))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "//" << std::endl;
std::cout << "////////// Test 3: Addition //////////" << std::endl;
std::cout << "//" << std::endl;
viennacl::copy(ublas_A_sub2, vcl_A_sub2);
std::cout << "Inplace add to submatrix: ";
ublas_A_sub2 += ublas_A_sub2;
vcl_A_sub2 += vcl_A_sub2;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Inplace add to matrix: ";
ublas_B += ublas_A_sub2;
vcl_B += vcl_A_sub2;
if (check_for_equality(ublas_B, vcl_B))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Inplace add of matrix: ";
ublas_A_sub2 += ublas_B;
vcl_A_sub2 += vcl_B;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Add to submatrix: ";
ublas_A_sub2 = ublas_A_sub2 + ublas_A_sub2;
vcl_A_sub2 = vcl_A_sub2 + vcl_A_sub2;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Add to matrix: ";
ublas_B = ublas_A_sub2 + ublas_A_sub2;
vcl_B = vcl_A_sub2 + vcl_A_sub2;
if (check_for_equality(ublas_B, vcl_B))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "//" << std::endl;
std::cout << "////////// Test 4: Subtraction //////////" << std::endl;
std::cout << "//" << std::endl;
viennacl::copy(ublas_A_sub2, vcl_A_sub2);
std::cout << "Inplace sub to submatrix: ";
ublas_A_sub2 -= ublas_A_sub2;
vcl_A_sub2 -= vcl_A_sub2;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Inplace sub to matrix: ";
ublas_B -= ublas_A_sub2;
vcl_B -= vcl_A_sub2;
if (check_for_equality(ublas_B, vcl_B))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Inplace sub of matrix: ";
ublas_A_sub2 -= ublas_B;
vcl_A_sub2 -= vcl_B;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Sub from submatrix: ";
ublas_A_sub2 = ublas_A_sub2 - ublas_A_sub2;
vcl_A_sub2 = vcl_A_sub2 - vcl_A_sub2;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Sub from matrix: ";
ublas_B = ublas_A_sub2 - ublas_A_sub2;
vcl_B = vcl_A_sub2 - vcl_A_sub2;
if (check_for_equality(ublas_B, vcl_B))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "//" << std::endl;
std::cout << "////////// Test 5: Scaling //////////" << std::endl;
std::cout << "//" << std::endl;
viennacl::copy(ublas_A, vcl_A);
std::cout << "Multiplication with CPU scalar: ";
ublas_A_sub2 *= ScalarType(3.1415);
vcl_A_sub2 *= ScalarType(3.1415);
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Multiplication with GPU scalar: ";
ublas_A_sub2 *= gpu_pi;
vcl_A_sub2 *= gpu_pi;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Division with CPU scalar: ";
ublas_A_sub2 /= ScalarType(3.1415);
vcl_A_sub2 /= ScalarType(3.1415);
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Division with GPU scalar: ";
ublas_A_sub2 /= gpu_pi;
vcl_A_sub2 /= gpu_pi;
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "//" << std::endl;
std::cout << "////////// Test 6: Matrix-Matrix Products //////////" << std::endl;
std::cout << "//" << std::endl;
std::cout << "Assigned C = A * B: ";
ublas_A_sub1 = prod(ublas_C_sub, ublas_D_sub);
vcl_A_sub1 = viennacl::linalg::prod(vcl_C_sub, vcl_D_sub);
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Assigned C = A^T * B: ";
ublas_A_sub1 = prod(trans(ublas_C_sub), ublas_D_sub);
vcl_A_sub1 = viennacl::linalg::prod(trans(vcl_C_sub), vcl_D_sub);
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Assigned C = A * B^T: ";
ublas_A_sub1 = prod(ublas_C_sub, trans(ublas_D_sub));
vcl_A_sub1 = viennacl::linalg::prod(vcl_C_sub, trans(vcl_D_sub));
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Assigned C = A^T * B^T: ";
ublas_A_sub1 = prod(trans(ublas_C_sub), trans(ublas_D_sub));
vcl_A_sub1 = viennacl::linalg::prod(trans(vcl_C_sub), trans(vcl_D_sub));
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "Inplace add of prod(): ";
ublas_A_sub1 += prod(ublas_C_sub, ublas_D_sub);
vcl_A_sub1 += viennacl::linalg::prod(vcl_C_sub, vcl_D_sub);
if (check_for_equality(ublas_A, vcl_A))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << "//" << std::endl;
std::cout << "////////// Test 7: Matrix-Vector Products //////////" << std::endl;
std::cout << "//" << std::endl;
VectorType ublas_v1(dim_large);
for (std::size_t i=0; i<ublas_v1.size(); ++i)
ublas_v1(i) = i;
boost::numeric::ublas::vector_range<VectorType> ublas_v1_sub(ublas_v1, ublas_r1);
VectorType ublas_v2(dim_large);
for (std::size_t i=0; i<ublas_v2.size(); ++i)
ublas_v2(i) = i - 5;
boost::numeric::ublas::vector_range<VectorType> ublas_v2_sub(ublas_v2, ublas_r1);
VCLVectorType vcl_v1(ublas_v1.size());
viennacl::vector_range<VCLVectorType> vcl_v1_sub(vcl_v1, vcl_r1);
VCLVectorType vcl_v2(ublas_v2.size());
viennacl::vector_range<VCLVectorType> vcl_v2_sub(vcl_v2, vcl_r1);
viennacl::copy(ublas_v1, vcl_v1);
viennacl::copy(ublas_v2, vcl_v2);
viennacl::copy(ublas_A_sub1, vcl_A_sub1);
ublas_v2_sub = prod(ublas_A_sub1, ublas_v1_sub);
vcl_v2_sub = viennacl::linalg::prod(vcl_A_sub1, vcl_v1_sub);
if (check_for_equality_vector(ublas_v2, vcl_v2))
std::cout << "PASSED!" << std::endl;
else
{
std::cout << std::endl << "TEST failed!";
return EXIT_FAILURE;
}
std::cout << std::endl;
std::cout << "----------------------------------------------" << std::endl;
std::cout << std::endl;
return EXIT_SUCCESS;
}