devtools/eigen
1
0
Fork 0
mirror of https://gitlab.com/libeigen/eigen.git synced 2026-04-10 11:34:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

Merged eigen/eigen into default

Browse Source
This commit is contained in:
Benoit Steiner
2016-10-12 22:42:33 -07:00
parent 5c68051cd7 e74612b9a0
commit 7e4a6754b2
15 changed files with 131 additions and 73 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
unsupported/test/cxx11_tensor_argmax_cuda.cu
View File

@@ -116,10 +116,10 @@ void test_cuda_argmax_dim()
assert(cudaMemcpyAsync(tensor_arg.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess);
assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
VERIFY_IS_EQUAL(tensor_arg.dimensions().TotalSize(),
VERIFY_IS_EQUAL(tensor_arg.size(),
size_t(2*3*5*7 / tensor.dimension(dim)));
for (size_t n = 0; n < tensor_arg.dimensions().TotalSize(); ++n) {
for (DenseIndex n = 0; n < tensor_arg.size(); ++n) {
// Expect max to be in the first index of the reduced dimension
VERIFY_IS_EQUAL(tensor_arg.data()[n], 0);
}
@@ -144,7 +144,7 @@ void test_cuda_argmax_dim()
assert(cudaMemcpyAsync(tensor_arg.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess);
assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
for (size_t n = 0; n < tensor_arg.dimensions().TotalSize(); ++n) {
for (DenseIndex n = 0; n < tensor_arg.size(); ++n) {
// Expect max to be in the last index of the reduced dimension
VERIFY_IS_EQUAL(tensor_arg.data()[n], tensor.dimension(dim) - 1);
}
@@ -205,10 +205,10 @@ void test_cuda_argmin_dim()
assert(cudaMemcpyAsync(tensor_arg.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess);
assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
VERIFY_IS_EQUAL(tensor_arg.dimensions().TotalSize(),
size_t(2*3*5*7 / tensor.dimension(dim)));
VERIFY_IS_EQUAL(tensor_arg.size(),
2*3*5*7 / tensor.dimension(dim));
for (size_t n = 0; n < tensor_arg.dimensions().TotalSize(); ++n) {
for (DenseIndex n = 0; n < tensor_arg.size(); ++n) {
// Expect min to be in the first index of the reduced dimension
VERIFY_IS_EQUAL(tensor_arg.data()[n], 0);
}
@@ -233,7 +233,7 @@ void test_cuda_argmin_dim()
assert(cudaMemcpyAsync(tensor_arg.data(), d_out, out_bytes, cudaMemcpyDeviceToHost, gpu_device.stream()) == cudaSuccess);
assert(cudaStreamSynchronize(gpu_device.stream()) == cudaSuccess);
for (size_t n = 0; n < tensor_arg.dimensions().TotalSize(); ++n) {
for (DenseIndex n = 0; n < tensor_arg.size(); ++n) {
// Expect max to be in the last index of the reduced dimension
VERIFY_IS_EQUAL(tensor_arg.data()[n], tensor.dimension(dim) - 1);
}

38
unsupported/test/cxx11_tensor_complex_cuda.cu
View File

@@ -108,8 +108,46 @@ static void test_cuda_sum_reductions() {
}
static void test_cuda_product_reductions() {
Eigen::CudaStreamDevice stream;
Eigen::GpuDevice gpu_device(&stream);
const int num_rows = internal::random<int>(1024, 5*1024);
const int num_cols = internal::random<int>(1024, 5*1024);
Tensor<std::complex<float>, 2> in(num_rows, num_cols);
in.setRandom();
Tensor<std::complex<float>, 0> full_redux;
full_redux = in.prod();
std::size_t in_bytes = in.size() * sizeof(std::complex<float>);
std::size_t out_bytes = full_redux.size() * sizeof(std::complex<float>);
std::complex<float>* gpu_in_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(in_bytes));
std::complex<float>* gpu_out_ptr = static_cast<std::complex<float>*>(gpu_device.allocate(out_bytes));
gpu_device.memcpyHostToDevice(gpu_in_ptr, in.data(), in_bytes);
TensorMap<Tensor<std::complex<float>, 2> > in_gpu(gpu_in_ptr, num_rows, num_cols);
TensorMap<Tensor<std::complex<float>, 0> > out_gpu(gpu_out_ptr);
out_gpu.device(gpu_device) = in_gpu.prod();
Tensor<std::complex<float>, 0> full_redux_gpu;
gpu_device.memcpyDeviceToHost(full_redux_gpu.data(), gpu_out_ptr, out_bytes);
gpu_device.synchronize();
// Check that the CPU and GPU reductions return the same result.
VERIFY_IS_APPROX(full_redux(), full_redux_gpu());
gpu_device.deallocate(gpu_in_ptr);
gpu_device.deallocate(gpu_out_ptr);
}
void test_cxx11_tensor_complex()
{
CALL_SUBTEST(test_cuda_nullary());
CALL_SUBTEST(test_cuda_sum_reductions());
CALL_SUBTEST(test_cuda_product_reductions());
}