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

Clang-format tests, examples, libraries, benchmarks, etc.

Browse Source
This commit is contained in:
Antonio Sánchez
2023-12-05 21:22:55 +00:00
committed by Rasmus Munk Larsen
parent 3252ecc7a4
commit 46e9cdb7fe
876 changed files with 33453 additions and 37795 deletions
Download Patch File Download Diff File Expand all files Collapse all files

106
unsupported/test/cxx11_tensor_concatenation_sycl.cpp
View File

@@ -22,9 +22,8 @@
using Eigen::Tensor;
template<typename DataType, int DataLayout, typename IndexType>
static void test_simple_concatenation(const Eigen::SyclDevice& sycl_device)
{
template <typename DataType, int DataLayout, typename IndexType>
static void test_simple_concatenation(const Eigen::SyclDevice& sycl_device) {
IndexType leftDim1 = 2;
IndexType leftDim2 = 3;
IndexType leftDim3 = 1;
@@ -34,31 +33,36 @@ static void test_simple_concatenation(const Eigen::SyclDevice& sycl_device)
IndexType rightDim3 = 1;
Eigen::array<IndexType, 3> rightRange = {{rightDim1, rightDim2, rightDim3}};
//IndexType concatDim1 = 3;
// IndexType concatDim2 = 3;
// IndexType concatDim3 = 1;
//Eigen::array<IndexType, 3> concatRange = {{concatDim1, concatDim2, concatDim3}};
// IndexType concatDim1 = 3;
// IndexType concatDim2 = 3;
// IndexType concatDim3 = 1;
// Eigen::array<IndexType, 3> concatRange = {{concatDim1, concatDim2, concatDim3}};
Tensor<DataType, 3, DataLayout, IndexType> left(leftRange);
Tensor<DataType, 3, DataLayout, IndexType> right(rightRange);
left.setRandom();
right.setRandom();
DataType * gpu_in1_data = static_cast<DataType*>(sycl_device.allocate(left.dimensions().TotalSize()*sizeof(DataType)));
DataType * gpu_in2_data = static_cast<DataType*>(sycl_device.allocate(right.dimensions().TotalSize()*sizeof(DataType)));
DataType* gpu_in1_data =
static_cast<DataType*>(sycl_device.allocate(left.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_in2_data =
static_cast<DataType*>(sycl_device.allocate(right.dimensions().TotalSize() * sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_in1(gpu_in1_data, leftRange);
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_in2(gpu_in2_data, rightRange);
sycl_device.memcpyHostToDevice(gpu_in1_data, left.data(),(left.dimensions().TotalSize())*sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in2_data, right.data(),(right.dimensions().TotalSize())*sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in1_data, left.data(), (left.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in2_data, right.data(), (right.dimensions().TotalSize()) * sizeof(DataType));
///
Tensor<DataType, 3, DataLayout, IndexType> concatenation1(leftDim1+rightDim1, leftDim2, leftDim3);
DataType * gpu_out_data1 = static_cast<DataType*>(sycl_device.allocate(concatenation1.dimensions().TotalSize()*sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_out1(gpu_out_data1, concatenation1.dimensions());
Tensor<DataType, 3, DataLayout, IndexType> concatenation1(leftDim1 + rightDim1, leftDim2, leftDim3);
DataType* gpu_out_data1 =
static_cast<DataType*>(sycl_device.allocate(concatenation1.dimensions().TotalSize() * sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_out1(gpu_out_data1,
concatenation1.dimensions());
//concatenation = left.concatenate(right, 0);
gpu_out1.device(sycl_device) =gpu_in1.concatenate(gpu_in2, 0);
sycl_device.memcpyDeviceToHost(concatenation1.data(), gpu_out_data1,(concatenation1.dimensions().TotalSize())*sizeof(DataType));
// concatenation = left.concatenate(right, 0);
gpu_out1.device(sycl_device) = gpu_in1.concatenate(gpu_in2, 0);
sycl_device.memcpyDeviceToHost(concatenation1.data(), gpu_out_data1,
(concatenation1.dimensions().TotalSize()) * sizeof(DataType));
VERIFY_IS_EQUAL(concatenation1.dimension(0), 4);
VERIFY_IS_EQUAL(concatenation1.dimension(1), 3);
@@ -73,13 +77,16 @@ static void test_simple_concatenation(const Eigen::SyclDevice& sycl_device)
}
sycl_device.deallocate(gpu_out_data1);
Tensor<DataType, 3, DataLayout, IndexType> concatenation2(leftDim1, leftDim2 +rightDim2, leftDim3);
DataType * gpu_out_data2 = static_cast<DataType*>(sycl_device.allocate(concatenation2.dimensions().TotalSize()*sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_out2(gpu_out_data2, concatenation2.dimensions());
gpu_out2.device(sycl_device) =gpu_in1.concatenate(gpu_in2, 1);
sycl_device.memcpyDeviceToHost(concatenation2.data(), gpu_out_data2,(concatenation2.dimensions().TotalSize())*sizeof(DataType));
Tensor<DataType, 3, DataLayout, IndexType> concatenation2(leftDim1, leftDim2 + rightDim2, leftDim3);
DataType* gpu_out_data2 =
static_cast<DataType*>(sycl_device.allocate(concatenation2.dimensions().TotalSize() * sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_out2(gpu_out_data2,
concatenation2.dimensions());
gpu_out2.device(sycl_device) = gpu_in1.concatenate(gpu_in2, 1);
sycl_device.memcpyDeviceToHost(concatenation2.data(), gpu_out_data2,
(concatenation2.dimensions().TotalSize()) * sizeof(DataType));
//concatenation = left.concatenate(right, 1);
// concatenation = left.concatenate(right, 1);
VERIFY_IS_EQUAL(concatenation2.dimension(0), 2);
VERIFY_IS_EQUAL(concatenation2.dimension(1), 6);
VERIFY_IS_EQUAL(concatenation2.dimension(2), 1);
@@ -92,13 +99,16 @@ static void test_simple_concatenation(const Eigen::SyclDevice& sycl_device)
}
}
sycl_device.deallocate(gpu_out_data2);
Tensor<DataType, 3, DataLayout, IndexType> concatenation3(leftDim1, leftDim2, leftDim3+rightDim3);
DataType * gpu_out_data3 = static_cast<DataType*>(sycl_device.allocate(concatenation3.dimensions().TotalSize()*sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_out3(gpu_out_data3, concatenation3.dimensions());
gpu_out3.device(sycl_device) =gpu_in1.concatenate(gpu_in2, 2);
sycl_device.memcpyDeviceToHost(concatenation3.data(), gpu_out_data3,(concatenation3.dimensions().TotalSize())*sizeof(DataType));
Tensor<DataType, 3, DataLayout, IndexType> concatenation3(leftDim1, leftDim2, leftDim3 + rightDim3);
DataType* gpu_out_data3 =
static_cast<DataType*>(sycl_device.allocate(concatenation3.dimensions().TotalSize() * sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 3, DataLayout, IndexType>> gpu_out3(gpu_out_data3,
concatenation3.dimensions());
gpu_out3.device(sycl_device) = gpu_in1.concatenate(gpu_in2, 2);
sycl_device.memcpyDeviceToHost(concatenation3.data(), gpu_out_data3,
(concatenation3.dimensions().TotalSize()) * sizeof(DataType));
//concatenation = left.concatenate(right, 2);
// concatenation = left.concatenate(right, 2);
VERIFY_IS_EQUAL(concatenation3.dimension(0), 2);
VERIFY_IS_EQUAL(concatenation3.dimension(1), 3);
VERIFY_IS_EQUAL(concatenation3.dimension(2), 2);
@@ -112,10 +122,8 @@ static void test_simple_concatenation(const Eigen::SyclDevice& sycl_device)
sycl_device.deallocate(gpu_in1_data);
sycl_device.deallocate(gpu_in2_data);
}
template<typename DataType, int DataLayout, typename IndexType>
static void test_concatenation_as_lvalue(const Eigen::SyclDevice& sycl_device)
{
template <typename DataType, int DataLayout, typename IndexType>
static void test_concatenation_as_lvalue(const Eigen::SyclDevice& sycl_device) {
IndexType leftDim1 = 2;
IndexType leftDim2 = 3;
Eigen::array<IndexType, 2> leftRange = {{leftDim1, leftDim2}};
@@ -136,28 +144,30 @@ static void test_concatenation_as_lvalue(const Eigen::SyclDevice& sycl_device)
right.setRandom();
result.setRandom();
DataType * gpu_in1_data = static_cast<DataType*>(sycl_device.allocate(left.dimensions().TotalSize()*sizeof(DataType)));
DataType * gpu_in2_data = static_cast<DataType*>(sycl_device.allocate(right.dimensions().TotalSize()*sizeof(DataType)));
DataType * gpu_out_data = static_cast<DataType*>(sycl_device.allocate(result.dimensions().TotalSize()*sizeof(DataType)));
DataType* gpu_in1_data =
static_cast<DataType*>(sycl_device.allocate(left.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_in2_data =
static_cast<DataType*>(sycl_device.allocate(right.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data =
static_cast<DataType*>(sycl_device.allocate(result.dimensions().TotalSize() * sizeof(DataType)));
Eigen::TensorMap<Eigen::Tensor<DataType, 2, DataLayout, IndexType>> gpu_in1(gpu_in1_data, leftRange);
Eigen::TensorMap<Eigen::Tensor<DataType, 2, DataLayout, IndexType>> gpu_in2(gpu_in2_data, rightRange);
Eigen::TensorMap<Eigen::Tensor<DataType, 2, DataLayout, IndexType>> gpu_out(gpu_out_data, resRange);
sycl_device.memcpyHostToDevice(gpu_in1_data, left.data(),(left.dimensions().TotalSize())*sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in2_data, right.data(),(right.dimensions().TotalSize())*sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_out_data, result.data(),(result.dimensions().TotalSize())*sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in1_data, left.data(), (left.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in2_data, right.data(), (right.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_out_data, result.data(), (result.dimensions().TotalSize()) * sizeof(DataType));
// t1.concatenate(t2, 0) = result;
gpu_in1.concatenate(gpu_in2, 0).device(sycl_device) =gpu_out;
sycl_device.memcpyDeviceToHost(left.data(), gpu_in1_data,(left.dimensions().TotalSize())*sizeof(DataType));
sycl_device.memcpyDeviceToHost(right.data(), gpu_in2_data,(right.dimensions().TotalSize())*sizeof(DataType));
// t1.concatenate(t2, 0) = result;
gpu_in1.concatenate(gpu_in2, 0).device(sycl_device) = gpu_out;
sycl_device.memcpyDeviceToHost(left.data(), gpu_in1_data, (left.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyDeviceToHost(right.data(), gpu_in2_data, (right.dimensions().TotalSize()) * sizeof(DataType));
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
VERIFY_IS_EQUAL(left(i, j), result(i, j));
VERIFY_IS_EQUAL(right(i, j), result(i+2, j));
VERIFY_IS_EQUAL(right(i, j), result(i + 2, j));
}
}
sycl_device.deallocate(gpu_in1_data);
@@ -165,8 +175,8 @@ static void test_concatenation_as_lvalue(const Eigen::SyclDevice& sycl_device)
sycl_device.deallocate(gpu_out_data);
}
template <typename DataType, typename Dev_selector> void tensorConcat_perDevice(Dev_selector s){
template <typename DataType, typename Dev_selector>
void tensorConcat_perDevice(Dev_selector s) {
QueueInterface queueInterface(s);
auto sycl_device = Eigen::SyclDevice(&queueInterface);
test_simple_concatenation<DataType, RowMajor, int64_t>(sycl_device);
@@ -174,7 +184,7 @@ template <typename DataType, typename Dev_selector> void tensorConcat_perDevice(
test_concatenation_as_lvalue<DataType, ColMajor, int64_t>(sycl_device);
}
EIGEN_DECLARE_TEST(cxx11_tensor_concatenation_sycl) {
for (const auto& device :Eigen::get_sycl_supported_devices()) {
for (const auto& device : Eigen::get_sycl_supported_devices()) {
CALL_SUBTEST(tensorConcat_perDevice<half>(device));
CALL_SUBTEST(tensorConcat_perDevice<float>(device));
}