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

95
unsupported/test/cxx11_tensor_scan_sycl.cpp
View File

@@ -23,31 +23,25 @@ using Eigen::Tensor;
typedef Tensor<float, 1>::DimensionPair DimPair;
template <typename DataType, int DataLayout, typename IndexType>
void test_sycl_cumsum(const Eigen::SyclDevice& sycl_device, IndexType m_size,
IndexType k_size, IndexType n_size, int consume_dim,
bool exclusive) {
void test_sycl_cumsum(const Eigen::SyclDevice& sycl_device, IndexType m_size, IndexType k_size, IndexType n_size,
int consume_dim, bool exclusive) {
static const DataType error_threshold = 1e-4f;
std::cout << "Testing for (" << m_size << "," << k_size << "," << n_size
<< " consume_dim : " << consume_dim << ")" << std::endl;
std::cout << "Testing for (" << m_size << "," << k_size << "," << n_size << " consume_dim : " << consume_dim << ")"
<< std::endl;
Tensor<DataType, 3, DataLayout, IndexType> t_input(m_size, k_size, n_size);
Tensor<DataType, 3, DataLayout, IndexType> t_result(m_size, k_size, n_size);
Tensor<DataType, 3, DataLayout, IndexType> t_result_gpu(m_size, k_size,
n_size);
Tensor<DataType, 3, DataLayout, IndexType> t_result_gpu(m_size, k_size, n_size);
t_input.setRandom();
std::size_t t_input_bytes = t_input.size() * sizeof(DataType);
std::size_t t_result_bytes = t_result.size() * sizeof(DataType);
DataType* gpu_data_in =
static_cast<DataType*>(sycl_device.allocate(t_input_bytes));
DataType* gpu_data_out =
static_cast<DataType*>(sycl_device.allocate(t_result_bytes));
DataType* gpu_data_in = static_cast<DataType*>(sycl_device.allocate(t_input_bytes));
DataType* gpu_data_out = static_cast<DataType*>(sycl_device.allocate(t_result_bytes));
array<IndexType, 3> tensorRange = {{m_size, k_size, n_size}};
TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_t_input(
gpu_data_in, tensorRange);
TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_t_result(
gpu_data_out, tensorRange);
TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_t_input(gpu_data_in, tensorRange);
TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_t_result(gpu_data_out, tensorRange);
sycl_device.memcpyHostToDevice(gpu_data_in, t_input.data(), t_input_bytes);
sycl_device.memcpyHostToDevice(gpu_data_out, t_input.data(), t_input_bytes);
@@ -55,21 +49,18 @@ void test_sycl_cumsum(const Eigen::SyclDevice& sycl_device, IndexType m_size,
t_result = t_input.cumsum(consume_dim, exclusive);
sycl_device.memcpyDeviceToHost(t_result_gpu.data(), gpu_data_out,
t_result_bytes);
sycl_device.memcpyDeviceToHost(t_result_gpu.data(), gpu_data_out, t_result_bytes);
sycl_device.synchronize();
for (IndexType i = 0; i < t_result.size(); i++) {
if (static_cast<DataType>(std::fabs(static_cast<DataType>(
t_result(i) - t_result_gpu(i)))) < error_threshold) {
if (static_cast<DataType>(std::fabs(static_cast<DataType>(t_result(i) - t_result_gpu(i)))) < error_threshold) {
continue;
}
if (Eigen::internal::isApprox(t_result(i), t_result_gpu(i),
error_threshold)) {
if (Eigen::internal::isApprox(t_result(i), t_result_gpu(i), error_threshold)) {
continue;
}
std::cout << "mismatch detected at index " << i << " CPU : " << t_result(i)
<< " vs SYCL : " << t_result_gpu(i) << std::endl;
std::cout << "mismatch detected at index " << i << " CPU : " << t_result(i) << " vs SYCL : " << t_result_gpu(i)
<< std::endl;
assert(false);
}
sycl_device.deallocate(gpu_data_in);
@@ -78,64 +69,44 @@ void test_sycl_cumsum(const Eigen::SyclDevice& sycl_device, IndexType m_size,
template <typename DataType, typename Dev>
void sycl_scan_test_exclusive_dim0_per_device(const Dev& sycl_device) {
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 2049, 1023, 127, 0,
true);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 2049, 1023, 127, 0,
true);
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 2049, 1023, 127, 0, true);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 2049, 1023, 127, 0, true);
}
template <typename DataType, typename Dev>
void sycl_scan_test_exclusive_dim1_per_device(const Dev& sycl_device) {
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 2049, 127, 1,
true);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 2049, 127, 1,
true);
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 2049, 127, 1, true);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 2049, 127, 1, true);
}
template <typename DataType, typename Dev>
void sycl_scan_test_exclusive_dim2_per_device(const Dev& sycl_device) {
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 127, 2049, 2,
true);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 127, 2049, 2,
true);
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 127, 2049, 2, true);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 127, 2049, 2, true);
}
template <typename DataType, typename Dev>
void sycl_scan_test_inclusive_dim0_per_device(const Dev& sycl_device) {
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 2049, 1023, 127, 0,
false);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 2049, 1023, 127, 0,
false);
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 2049, 1023, 127, 0, false);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 2049, 1023, 127, 0, false);
}
template <typename DataType, typename Dev>
void sycl_scan_test_inclusive_dim1_per_device(const Dev& sycl_device) {
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 2049, 127, 1,
false);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 2049, 127, 1,
false);
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 2049, 127, 1, false);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 2049, 127, 1, false);
}
template <typename DataType, typename Dev>
void sycl_scan_test_inclusive_dim2_per_device(const Dev& sycl_device) {
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 127, 2049, 2,
false);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 127, 2049, 2,
false);
test_sycl_cumsum<DataType, ColMajor, int64_t>(sycl_device, 1023, 127, 2049, 2, false);
test_sycl_cumsum<DataType, RowMajor, int64_t>(sycl_device, 1023, 127, 2049, 2, false);
}
EIGEN_DECLARE_TEST(cxx11_tensor_scan_sycl) {
for (const auto& device : Eigen::get_sycl_supported_devices()) {
std::cout << "Running on "
<< device.template get_info<cl::sycl::info::device::name>()
<< std::endl;
std::cout << "Running on " << device.template get_info<cl::sycl::info::device::name>() << std::endl;
QueueInterface queueInterface(device);
auto sycl_device = Eigen::SyclDevice(&queueInterface);
CALL_SUBTEST_1(
sycl_scan_test_exclusive_dim0_per_device<float>(sycl_device));
CALL_SUBTEST_2(
sycl_scan_test_exclusive_dim1_per_device<float>(sycl_device));
CALL_SUBTEST_3(
sycl_scan_test_exclusive_dim2_per_device<float>(sycl_device));
CALL_SUBTEST_4(
sycl_scan_test_inclusive_dim0_per_device<float>(sycl_device));
CALL_SUBTEST_5(
sycl_scan_test_inclusive_dim1_per_device<float>(sycl_device));
CALL_SUBTEST_6(
sycl_scan_test_inclusive_dim2_per_device<float>(sycl_device));
CALL_SUBTEST_1(sycl_scan_test_exclusive_dim0_per_device<float>(sycl_device));
CALL_SUBTEST_2(sycl_scan_test_exclusive_dim1_per_device<float>(sycl_device));
CALL_SUBTEST_3(sycl_scan_test_exclusive_dim2_per_device<float>(sycl_device));
CALL_SUBTEST_4(sycl_scan_test_inclusive_dim0_per_device<float>(sycl_device));
CALL_SUBTEST_5(sycl_scan_test_inclusive_dim1_per_device<float>(sycl_device));
CALL_SUBTEST_6(sycl_scan_test_inclusive_dim2_per_device<float>(sycl_device));
}
}