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

100
unsupported/test/cxx11_tensor_reverse_sycl.cpp
View File

@@ -38,30 +38,24 @@ static void test_simple_reverse(const Eigen::SyclDevice& sycl_device) {
dim_rev[2] = true;
dim_rev[3] = false;
DataType* gpu_in_data = static_cast<DataType*>(
sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data = static_cast<DataType*>(sycl_device.allocate(
reversed_tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_in_data =
static_cast<DataType*>(sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data =
static_cast<DataType*>(sycl_device.allocate(reversed_tensor.dimensions().TotalSize() * sizeof(DataType)));
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > in_gpu(gpu_in_data,
tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu(gpu_out_data,
tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu(gpu_out_data, tensorRange);
sycl_device.memcpyHostToDevice(
gpu_in_data, tensor.data(),
(tensor.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in_data, tensor.data(), (tensor.dimensions().TotalSize()) * sizeof(DataType));
out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
sycl_device.memcpyDeviceToHost(
reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
sycl_device.memcpyDeviceToHost(reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
// Check that the CPU and GPU reductions return the same result.
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
for (IndexType k = 0; k < 5; ++k) {
for (IndexType l = 0; l < 7; ++l) {
VERIFY_IS_EQUAL(tensor(i, j, k, l),
reversed_tensor(i, 2 - j, 4 - k, l));
VERIFY_IS_EQUAL(tensor(i, j, k, l), reversed_tensor(i, 2 - j, 4 - k, l));
}
}
}
@@ -72,9 +66,8 @@ static void test_simple_reverse(const Eigen::SyclDevice& sycl_device) {
dim_rev[3] = false;
out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
sycl_device.memcpyDeviceToHost(
reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
sycl_device.memcpyDeviceToHost(reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
@@ -91,16 +84,14 @@ static void test_simple_reverse(const Eigen::SyclDevice& sycl_device) {
dim_rev[2] = false;
dim_rev[3] = true;
out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
sycl_device.memcpyDeviceToHost(
reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
sycl_device.memcpyDeviceToHost(reversed_tensor.data(), gpu_out_data,
reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
for (IndexType i = 0; i < 2; ++i) {
for (IndexType j = 0; j < 3; ++j) {
for (IndexType k = 0; k < 5; ++k) {
for (IndexType l = 0; l < 7; ++l) {
VERIFY_IS_EQUAL(tensor(i, j, k, l),
reversed_tensor(1 - i, j, k, 6 - l));
VERIFY_IS_EQUAL(tensor(i, j, k, l), reversed_tensor(1 - i, j, k, 6 - l));
}
}
}
@@ -111,8 +102,7 @@ static void test_simple_reverse(const Eigen::SyclDevice& sycl_device) {
}
template <typename DataType, int DataLayout, typename IndexType>
static void test_expr_reverse(const Eigen::SyclDevice& sycl_device,
bool LValue) {
static void test_expr_reverse(const Eigen::SyclDevice& sycl_device, bool LValue) {
IndexType dim1 = 2;
IndexType dim2 = 3;
IndexType dim3 = 5;
@@ -130,32 +120,26 @@ static void test_expr_reverse(const Eigen::SyclDevice& sycl_device,
dim_rev[2] = false;
dim_rev[3] = true;
DataType* gpu_in_data = static_cast<DataType*>(
sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data_expected = static_cast<DataType*>(sycl_device.allocate(
expected.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data_result = static_cast<DataType*>(
sycl_device.allocate(result.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_in_data =
static_cast<DataType*>(sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data_expected =
static_cast<DataType*>(sycl_device.allocate(expected.dimensions().TotalSize() * sizeof(DataType)));
DataType* gpu_out_data_result =
static_cast<DataType*>(sycl_device.allocate(result.dimensions().TotalSize() * sizeof(DataType)));
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > in_gpu(gpu_in_data,
tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_expected(
gpu_out_data_expected, tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_result(
gpu_out_data_result, tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > in_gpu(gpu_in_data, tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_expected(gpu_out_data_expected, tensorRange);
TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_result(gpu_out_data_result, tensorRange);
sycl_device.memcpyHostToDevice(
gpu_in_data, tensor.data(),
(tensor.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_in_data, tensor.data(), (tensor.dimensions().TotalSize()) * sizeof(DataType));
if (LValue) {
out_gpu_expected.reverse(dim_rev).device(sycl_device) = in_gpu;
} else {
out_gpu_expected.device(sycl_device) = in_gpu.reverse(dim_rev);
}
sycl_device.memcpyDeviceToHost(
expected.data(), gpu_out_data_expected,
expected.dimensions().TotalSize() * sizeof(DataType));
sycl_device.memcpyDeviceToHost(expected.data(), gpu_out_data_expected,
expected.dimensions().TotalSize() * sizeof(DataType));
array<IndexType, 4> src_slice_dim;
src_slice_dim[0] = 2;
@@ -172,9 +156,8 @@ static void test_expr_reverse(const Eigen::SyclDevice& sycl_device,
for (IndexType i = 0; i < 5; ++i) {
if (LValue) {
out_gpu_result.slice(dst_slice_start, dst_slice_dim)
.reverse(dim_rev)
.device(sycl_device) = in_gpu.slice(src_slice_start, src_slice_dim);
out_gpu_result.slice(dst_slice_start, dst_slice_dim).reverse(dim_rev).device(sycl_device) =
in_gpu.slice(src_slice_start, src_slice_dim);
} else {
out_gpu_result.slice(dst_slice_start, dst_slice_dim).device(sycl_device) =
in_gpu.slice(src_slice_start, src_slice_dim).reverse(dim_rev);
@@ -182,9 +165,8 @@ static void test_expr_reverse(const Eigen::SyclDevice& sycl_device,
src_slice_start[2] += 1;
dst_slice_start[2] += 1;
}
sycl_device.memcpyDeviceToHost(
result.data(), gpu_out_data_result,
result.dimensions().TotalSize() * sizeof(DataType));
sycl_device.memcpyDeviceToHost(result.data(), gpu_out_data_result,
result.dimensions().TotalSize() * sizeof(DataType));
for (IndexType i = 0; i < expected.dimension(0); ++i) {
for (IndexType j = 0; j < expected.dimension(1); ++j) {
@@ -198,23 +180,20 @@ static void test_expr_reverse(const Eigen::SyclDevice& sycl_device,
dst_slice_start[2] = 0;
result.setRandom();
sycl_device.memcpyHostToDevice(
gpu_out_data_result, result.data(),
(result.dimensions().TotalSize()) * sizeof(DataType));
sycl_device.memcpyHostToDevice(gpu_out_data_result, result.data(),
(result.dimensions().TotalSize()) * sizeof(DataType));
for (IndexType i = 0; i < 5; ++i) {
if (LValue) {
out_gpu_result.slice(dst_slice_start, dst_slice_dim)
.reverse(dim_rev)
.device(sycl_device) = in_gpu.slice(dst_slice_start, dst_slice_dim);
out_gpu_result.slice(dst_slice_start, dst_slice_dim).reverse(dim_rev).device(sycl_device) =
in_gpu.slice(dst_slice_start, dst_slice_dim);
} else {
out_gpu_result.slice(dst_slice_start, dst_slice_dim).device(sycl_device) =
in_gpu.reverse(dim_rev).slice(dst_slice_start, dst_slice_dim);
}
dst_slice_start[2] += 1;
}
sycl_device.memcpyDeviceToHost(
result.data(), gpu_out_data_result,
result.dimensions().TotalSize() * sizeof(DataType));
sycl_device.memcpyDeviceToHost(result.data(), gpu_out_data_result,
result.dimensions().TotalSize() * sizeof(DataType));
for (IndexType i = 0; i < expected.dimension(0); ++i) {
for (IndexType j = 0; j < expected.dimension(1); ++j) {
@@ -240,8 +219,7 @@ void sycl_reverse_test_per_device(const cl::sycl::device& d) {
}
EIGEN_DECLARE_TEST(cxx11_tensor_reverse_sycl) {
for (const auto& device : Eigen::get_sycl_supported_devices()) {
std::cout << "Running on "
<< device.get_info<cl::sycl::info::device::name>() << std::endl;
std::cout << "Running on " << device.get_info<cl::sycl::info::device::name>() << std::endl;
CALL_SUBTEST_1(sycl_reverse_test_per_device<short>(device));
CALL_SUBTEST_2(sycl_reverse_test_per_device<int>(device));
CALL_SUBTEST_3(sycl_reverse_test_per_device<unsigned int>(device));