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

2
unsupported/Eigen/AlignedVector3
View File

@@ -61,7 +61,7 @@ template<typename _Scalar> class AlignedVector3
Scalar* data() { return m_coeffs.data(); }
const Scalar* data() const { return m_coeffs.data(); }
Index innerStride() const { return 1; }
Index outerStride() const { return m_coeffs.outerStride(); }
Index outerStride() const { return 3; }
inline const Scalar& coeff(Index row, Index col) const
{ return m_coeffs.coeff(row, col); }

2
unsupported/Eigen/CXX11/Tensor
View File

@@ -34,6 +34,8 @@
#include <cstring>
#ifdef _WIN32
typedef __int16 int16_t;
typedef unsigned __int16 uint16_t;
typedef __int32 int32_t;
typedef unsigned __int32 uint32_t;
typedef __int64 int64_t;

12
unsupported/Eigen/CXX11/src/Tensor/TensorFunctors.h
View File

@@ -124,7 +124,8 @@ template <typename T> struct SumReducer
}
template <typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizeBoth(const T saccum, const Packet& vaccum) const {
return saccum + predux(vaccum);
internal::scalar_sum_op<T> sum_op;
return sum_op(saccum, predux(vaccum));
}
};
@@ -173,7 +174,8 @@ template <typename T> struct MeanReducer
}
template <typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizeBoth(const T saccum, const Packet& vaccum) const {
return (saccum + predux(vaccum)) / (scalarCount_ + packetCount_ * unpacket_traits<Packet>::size);
internal::scalar_sum_op<T> sum_op;
return sum_op(saccum, predux(vaccum)) / (scalarCount_ + packetCount_ * unpacket_traits<Packet>::size);
}
protected:
@@ -304,7 +306,8 @@ template <typename T> struct ProdReducer
static const bool IsStateful = false;
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reduce(const T t, T* accum) const {
(*accum) *= t;
internal::scalar_product_op<T> prod_op;
(*accum) = prod_op(*accum, t);
}
template <typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void reducePacket(const Packet& p, Packet* accum) const {
@@ -328,7 +331,8 @@ template <typename T> struct ProdReducer
}
template <typename Packet>
EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T finalizeBoth(const T saccum, const Packet& vaccum) const {
return saccum * predux_mul(vaccum);
internal::scalar_product_op<T> prod_op;
return prod_op(saccum, predux_mul(vaccum));
}
};

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());
}