HDF5: Use explicit mapping for VDS
Build Packages / build:rpm (rocky8_nocuda) (push) Successful in 10m30s
Build Packages / build:rpm (ubuntu2404_nocuda) (push) Successful in 14m25s
Build Packages / build:rpm (ubuntu2204_nocuda) (push) Successful in 16m8s
Build Packages / build:rpm (rocky9_nocuda) (push) Successful in 16m21s
Build Packages / build:rpm (rocky8) (push) Successful in 18m13s
Build Packages / build:rpm (rocky8_sls9) (push) Successful in 18m25s
Build Packages / build:rpm (rocky9) (push) Successful in 19m12s
Build Packages / build:rpm (rocky9_sls9) (push) Successful in 19m21s
Build Packages / Generate python client (push) Successful in 51s
Build Packages / Build documentation (push) Successful in 1m10s
Build Packages / Create release (push) Has been skipped
Build Packages / build:rpm (ubuntu2204) (push) Successful in 10m38s
Build Packages / build:rpm (ubuntu2404) (push) Successful in 8m29s
Build Packages / XDS test (neggia plugin) (push) Successful in 9m50s
Build Packages / XDS test (durin plugin) (push) Successful in 13m7s
Build Packages / XDS test (JFJoch plugin) (push) Successful in 11m16s
Build Packages / DIALS test (push) Successful in 15m26s
Build Packages / Unit tests (push) Successful in 49m40s

This commit is contained in:
2026-05-05 11:58:08 +02:00
parent 7d34e8a049
commit d905cc230c
5 changed files with 407 additions and 13 deletions
+65 -7
View File
@@ -62,6 +62,37 @@ std::vector<hsize_t> GetDimension(HDF5Object& object, const std::string& path) {
return dim;
}
std::vector<HDF5VirtualDatasetMapping> ReadVDSImageMappings(HDF5Object &file,
const std::string &dataset_name) {
HDF5DataSet dataset(file, dataset_name);
HDF5Dcpl dcpl(dataset);
auto mappings = dcpl.GetVirtualMappings();
if (mappings.empty())
throw JFJochException(JFJochExceptionCategory::HDF5,
dataset_name + " is not a virtual dataset");
for (const auto &mapping: mappings) {
if (mapping.dataset.empty())
throw JFJochException(JFJochExceptionCategory::HDF5,
"VDS mapping has empty source dataset name");
if (mapping.virtual_start.size() != 3)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Only 3D image VDS mappings are supported");
}
return mappings;
}
std::string ResolveRelativeToMaster(const std::string &directory,
const std::string &filename) {
std::filesystem::path path(filename);
if (path.is_absolute() || directory.empty())
return filename;
return (std::filesystem::path(directory) / path).string();
}
template <class T>
void JFJochHDF5Reader::ReadVector(std::vector<T> &v,
HDF5Object &file,
@@ -104,6 +135,10 @@ void JFJochHDF5Reader::ReadFile(const std::string& filename) {
master_file = std::make_shared<HDF5ReadOnlyFile>(filename);
dataset->experiment = default_experiment;
std::filesystem::path master_path(filename);
master_file_directory = master_path.parent_path().string();
vds_data_mappings.clear();
dataset->arm_date = master_file->GetString("/entry/start_time");
dataset->experiment.FilePrefix(dataset_name(filename));
@@ -115,19 +150,19 @@ void JFJochHDF5Reader::ReadFile(const std::string& filename) {
size_t image_size_x = 0;
size_t image_size_y = 0;
if (master_file->Exists("/entry/data/data"))
format = FileWriterFormat::NXmxVDS;
else if (master_file->Exists("/entry/data/data_000001")) {
format = FileWriterFormat::NXmxLegacy;
}
if (master_file->Exists("/entry/data/data")) {
HDF5DataSet data_dataset(*master_file, "/entry/data/data");
HDF5Dcpl dcpl(data_dataset);
data_layout = dcpl.GetLayout();
if (format == FileWriterFormat::NXmxVDS ) {
auto dim = GetDimension(*master_file, "/entry/data/data");
number_of_images = dim[0];
image_size_y = dim[1];
image_size_x = dim[2];
images_per_file = number_of_images;
if (data_layout == HDF5DataSetLayout::VIRTUAL)
vds_data_mappings = ReadVDSImageMappings(*master_file, "/entry/data/data");
if (master_file->Exists("/entry/instrument/detector/detectorSpecific/data_collection_efficiency_image"))
dataset->efficiency = master_file->ReadVector<float>(
@@ -166,7 +201,12 @@ void JFJochHDF5Reader::ReadFile(const std::string& filename) {
}
if (master_file->Exists("/entry/image"))
dataset->max_value = master_file->ReadOptVector<int64_t>("/entry/image/max_value");
} else if (format == FileWriterFormat::NXmxLegacy) {
format = FileWriterFormat::NXmxVDS;
} else if (master_file->Exists("/entry/data/data_000001")) {
format = FileWriterFormat::NXmxLegacy;
data_layout = HDF5DataSetLayout::CONTIGUOUS;
legacy_format_files.clear();
image_size_x = master_file->GetInt("/entry/instrument/detector/detectorSpecific/x_pixels_in_detector");
@@ -509,6 +549,22 @@ std::pair<std::shared_ptr<HDF5ReadOnlyFile>, uint32_t> JFJochHDF5Reader::GetImag
uint32_t file_id = image_number / images_per_file;
image_id = image_number % images_per_file;
data_file = std::make_shared<HDF5ReadOnlyFile>(legacy_format_files.at(file_id));
} else if (format == FileWriterFormat::NXmxVDS && data_layout == HDF5DataSetLayout::VIRTUAL) {
const auto image = static_cast<hsize_t>(image_number);
for (const auto &mapping: vds_data_mappings) {
if (!mapping.ContainsVirtualImage(image))
continue;
image_id = static_cast<uint32_t>(mapping.SourceImage(image));
data_file = std::make_shared<HDF5ReadOnlyFile>(
ResolveRelativeToMaster(master_file_directory, mapping.filename)
);
return {std::move(data_file), image_id};
}
throw JFJochException(JFJochExceptionCategory::HDF5,
"Image not covered by /entry/data/data VDS mappings");
} else {
image_id = image_number;
data_file = master_file;
@@ -774,6 +830,8 @@ void JFJochHDF5Reader::Close() {
master_file = {};
number_of_images = 0;
legacy_format_files.clear();
vds_data_mappings.clear();
master_file_directory.clear();
SetStartMessage({});
}
+3
View File
@@ -9,10 +9,13 @@
class JFJochHDF5Reader : public JFJochReader {
FileWriterFormat format = FileWriterFormat::NoFile;
HDF5DataSetLayout data_layout = HDF5DataSetLayout::CONTIGUOUS;
std::shared_ptr<HDF5ReadOnlyFile> master_file;
std::vector<std::string> legacy_format_files;
std::vector<HDF5VirtualDatasetMapping> vds_data_mappings;
std::string master_file_directory;
size_t images_per_file = 1;
size_t number_of_images = 0;
+97
View File
@@ -1344,3 +1344,100 @@ TEST_CASE("FileWriter_TIFF", "[HDF5][Full]") {
REQUIRE(!file_set.GetZMQAddr());
}
}
TEST_CASE("HDF5Objects_VDS_reverse_contiguous", "[HDF5][Unit]") {
{
RegisterHDF5Filter();
HDF5File file("scratch_vds_reverse_contiguous.h5", true);
HDF5Dcpl dcpl;
HDF5DataType data_type((int16_t) 0);
HDF5DataSpace full_space({5, 4, 3});
{
HDF5DataSpace source_space({2, 4, 3});
HDF5DataSpace virtual_space({5, 4, 3});
virtual_space.SelectHyperslab({0, 0, 0}, {2, 4, 3});
dcpl.SetVirtual("file_000001.h5", "/entry/data/data", source_space, virtual_space);
}
{
HDF5DataSpace source_space({3, 4, 3});
HDF5DataSpace virtual_space({5, 4, 3});
virtual_space.SelectHyperslab({2, 0, 0}, {3, 4, 3});
dcpl.SetVirtual("file_000002.h5", "/entry/data/data", source_space, virtual_space);
}
HDF5DataSet dataset(file, "/data", data_type, full_space, dcpl);
HDF5Dcpl read_dcpl(dataset);
auto mappings = read_dcpl.GetVirtualMappings();
REQUIRE(mappings.size() == 2);
REQUIRE(mappings[0].ContainsVirtualImage(0));
REQUIRE(mappings[0].ContainsVirtualImage(1));
REQUIRE(!mappings[0].ContainsVirtualImage(2));
CHECK(mappings[0].SourceImage(0) == 0);
CHECK(mappings[0].SourceImage(1) == 1);
REQUIRE(mappings[1].ContainsVirtualImage(2));
REQUIRE(mappings[1].ContainsVirtualImage(4));
CHECK(mappings[1].SourceImage(2) == 0);
CHECK(mappings[1].SourceImage(3) == 1);
CHECK(mappings[1].SourceImage(4) == 2);
}
remove("scratch_vds_reverse_contiguous.h5");
REQUIRE(H5Fget_obj_count(H5F_OBJ_ALL, H5F_OBJ_ALL) == 0);
}
TEST_CASE("HDF5Objects_VDS_reverse_strided", "[HDF5][Unit]") {
{
RegisterHDF5Filter();
HDF5File file("scratch_vds_reverse_strided.h5", true);
HDF5Dcpl dcpl;
HDF5DataType data_type((int16_t) 0);
HDF5DataSpace full_space({6, 4, 3});
{
HDF5DataSpace source_space({3, 4, 3});
HDF5DataSpace virtual_space({6, 4, 3});
virtual_space.SelectHyperslabWithStride({0, 0, 0}, {3, 4, 3}, {2, 1, 1});
dcpl.SetVirtual("file_even.h5", "/entry/data/data", source_space, virtual_space);
}
{
HDF5DataSpace source_space({3, 4, 3});
HDF5DataSpace virtual_space({6, 4, 3});
virtual_space.SelectHyperslabWithStride({1, 0, 0}, {3, 4, 3}, {2, 1, 1});
dcpl.SetVirtual("file_odd.h5", "/entry/data/data", source_space, virtual_space);
}
HDF5DataSet dataset(file, "/data", data_type, full_space, dcpl);
HDF5Dcpl read_dcpl(dataset);
auto mappings = read_dcpl.GetVirtualMappings();
REQUIRE(mappings.size() == 2);
REQUIRE(mappings[0].ContainsVirtualImage(0));
REQUIRE(mappings[0].ContainsVirtualImage(2));
REQUIRE(mappings[0].ContainsVirtualImage(4));
REQUIRE(!mappings[0].ContainsVirtualImage(1));
CHECK(mappings[0].SourceImage(0) == 0);
CHECK(mappings[0].SourceImage(2) == 1);
CHECK(mappings[0].SourceImage(4) == 2);
REQUIRE(mappings[1].ContainsVirtualImage(1));
REQUIRE(mappings[1].ContainsVirtualImage(3));
REQUIRE(mappings[1].ContainsVirtualImage(5));
REQUIRE(!mappings[1].ContainsVirtualImage(0));
CHECK(mappings[1].SourceImage(1) == 0);
CHECK(mappings[1].SourceImage(3) == 1);
CHECK(mappings[1].SourceImage(5) == 2);
}
remove("scratch_vds_reverse_strided.h5");
REQUIRE(H5Fget_obj_count(H5F_OBJ_ALL, H5F_OBJ_ALL) == 0);
}
+219 -5
View File
@@ -231,23 +231,26 @@ bool HDF5DataType::IsFloat() const {
HDF5Dcpl::HDF5Dcpl() : HDF5Id() {
id = H5Pcreate(H5P_DATASET_CREATE);
ndim = 0;
layout = HDF5DataSetLayout::CONTIGUOUS;
}
HDF5Dcpl::HDF5Dcpl(const HDF5DataSet &data_set) : HDF5Id() {
id = H5Dget_create_plist(data_set.GetID());
// Check if chunking is enabled
H5D_layout_t layout = H5Pget_layout(id);
if (layout != H5D_CHUNKED) {
ndim = 0;
} else {
H5D_layout_t h5_layout = H5Pget_layout(id);
if (h5_layout == H5D_VIRTUAL)
layout = HDF5DataSetLayout::VIRTUAL;
else if (h5_layout == H5D_CHUNKED) {
layout = HDF5DataSetLayout::CHUNKED;
ndim = H5Pget_chunk(id, 0, nullptr);
if (ndim <= 0) {
H5Pclose(id);
throw JFJochException(JFJochExceptionCategory::HDF5,
"Error getting number of chunk dimensions");
}
}
} else
layout = HDF5DataSetLayout::CONTIGUOUS;
}
HDF5Dcpl::~HDF5Dcpl() {
@@ -255,6 +258,7 @@ HDF5Dcpl::~HDF5Dcpl() {
}
void HDF5Dcpl::SetChunking(const std::vector<hsize_t> &dims) {
layout = HDF5DataSetLayout::CHUNKED;
if ((dims.empty()) || dims[0] == 0)
throw JFJochException(JFJochExceptionCategory::HDF5, "Value dimension cannot be 0");
ndim = dims.size();
@@ -298,11 +302,16 @@ void HDF5Dcpl::SetFillValueU16(uint16_t val) {
}
void HDF5Dcpl::SetVirtual(const std::string &path, const std::string &dataset, const HDF5DataSpace &src_dataspace, const HDF5DataSpace &dest_dataspace) {
layout = HDF5DataSetLayout::VIRTUAL;
std::string filename = ExtractFilename(path);
if (H5Pset_virtual(id, dest_dataspace.GetID(), filename.c_str(), dataset.c_str(), src_dataspace.GetID()) < 0)
throw JFJochException(JFJochExceptionCategory::HDF5, "Cannot set virtual mapping");
}
HDF5DataSetLayout HDF5Dcpl::GetLayout() const {
return layout;
}
HDF5Fapl::HDF5Fapl() : HDF5Id() {
id = H5Pcreate(H5P_FILE_ACCESS);
}
@@ -995,3 +1004,208 @@ std::string HDF5Object::GetLinkedFileName(const std::string& name) const {
return s;
}
namespace {
std::vector<hsize_t> GetDataspaceDimensions(hid_t dataspace_id) {
const int ndims = H5Sget_simple_extent_ndims(dataspace_id);
if (ndims < 0)
throw JFJochException(JFJochExceptionCategory::HDF5, "Cannot read dataspace dimensions");
std::vector<hsize_t> dims(ndims);
if (ndims > 0 && H5Sget_simple_extent_dims(dataspace_id, dims.data(), nullptr) < 0)
throw JFJochException(JFJochExceptionCategory::HDF5, "Cannot read dataspace dimensions");
return dims;
}
void GetRegularSelection(hid_t dataspace_id,
std::vector<hsize_t> &start,
std::vector<hsize_t> &stride,
std::vector<hsize_t> &count,
std::vector<hsize_t> &block) {
const auto dims = GetDataspaceDimensions(dataspace_id);
const H5S_sel_type selection_type = H5Sget_select_type(dataspace_id);
if (selection_type == H5S_SEL_ALL) {
start.clear();
stride.clear();
count.clear();
block.clear();
return;
}
start.assign(dims.size(), 0);
stride.assign(dims.size(), 1);
count.assign(dims.size(), 1);
block = dims;
if (selection_type != H5S_SEL_HYPERSLABS)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Only regular hyperslab VDS selections are supported");
if (H5Sis_regular_hyperslab(dataspace_id) <= 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Only regular hyperslab VDS selections are supported");
if (H5Sget_regular_hyperslab(dataspace_id,
start.data(),
stride.data(),
count.data(),
block.data()) < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot decode VDS hyperslab selection");
}
bool ContainsInRegularHyperslab(hsize_t value,
hsize_t start,
hsize_t stride,
hsize_t count,
hsize_t block) {
for (hsize_t i = 0; i < count; i++) {
const hsize_t block_start = start + i * stride;
if ((value >= block_start) && (value < block_start + block))
return true;
}
return false;
}
hsize_t IndexInRegularHyperslab(hsize_t value,
hsize_t start,
hsize_t stride,
hsize_t count,
hsize_t block) {
for (hsize_t i = 0; i < count; i++) {
const hsize_t block_start = start + i * stride;
if ((value >= block_start) && (value < block_start + block))
return i * block + (value - block_start);
}
throw JFJochException(JFJochExceptionCategory::HDF5,
"Image is not part of VDS hyperslab");
}
hsize_t ValueFromRegularHyperslabIndex(hsize_t index,
hsize_t start,
hsize_t stride,
hsize_t count,
hsize_t block) {
if (index >= count * block)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Source image is outside of VDS source hyperslab");
const hsize_t block_number = index / block;
const hsize_t in_block = index % block;
return start + block_number * stride + in_block;
}
}
bool HDF5VirtualDatasetMapping::ContainsVirtualImage(hsize_t image_number) const {
if (virtual_start.empty() || virtual_stride.empty() || virtual_count.empty() || virtual_block.empty())
return false;
return ContainsInRegularHyperslab(image_number,
virtual_start[0],
virtual_stride[0],
virtual_count[0],
virtual_block[0]);
}
hsize_t HDF5VirtualDatasetMapping::SourceImage(hsize_t image_number) const {
if (!ContainsVirtualImage(image_number))
throw JFJochException(JFJochExceptionCategory::HDF5,
"Image is outside of VDS mapping");
const hsize_t source_index = IndexInRegularHyperslab(image_number,
virtual_start[0],
virtual_stride[0],
virtual_count[0],
virtual_block[0]);
if (source_start.empty() || source_stride.empty() || source_count.empty() || source_block.empty())
return source_index;
return ValueFromRegularHyperslabIndex(source_index,
source_start[0],
source_stride[0],
source_count[0],
source_block[0]);
}
std::vector<HDF5VirtualDatasetMapping> HDF5Dcpl::GetVirtualMappings() const {
size_t mapping_count = 0;
if (H5Pget_virtual_count(id, &mapping_count) < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get number of VDS mappings");
if (mapping_count < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get number of VDS mappings");
std::vector<HDF5VirtualDatasetMapping> ret;
ret.reserve(static_cast<size_t>(mapping_count));
for (size_t i = 0; i < static_cast<size_t>(mapping_count); i++) {
HDF5VirtualDatasetMapping mapping;
const ssize_t filename_size = H5Pget_virtual_filename(id, i, nullptr, 0);
if (filename_size < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get VDS source filename size");
std::vector<char> filename(filename_size + 1, '\0');
if (H5Pget_virtual_filename(id, i, filename.data(), filename.size()) < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get VDS source filename");
mapping.filename = filename.data();
const ssize_t dataset_size = H5Pget_virtual_dsetname(id, i, nullptr, 0);
if (dataset_size < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get VDS source dataset size");
std::vector<char> dataset(dataset_size + 1, '\0');
if (H5Pget_virtual_dsetname(id, i, dataset.data(), dataset.size()) < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get VDS source dataset");
mapping.dataset = dataset.data();
const hid_t virtual_space = H5Pget_virtual_vspace(id, i);
if (virtual_space < 0)
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get VDS virtual dataspace");
const hid_t source_space = H5Pget_virtual_srcspace(id, i);
if (source_space < 0) {
H5Sclose(virtual_space);
throw JFJochException(JFJochExceptionCategory::HDF5,
"Cannot get VDS source dataspace");
}
try {
GetRegularSelection(virtual_space,
mapping.virtual_start,
mapping.virtual_stride,
mapping.virtual_count,
mapping.virtual_block);
GetRegularSelection(source_space,
mapping.source_start,
mapping.source_stride,
mapping.source_count,
mapping.source_block);
} catch (...) {
H5Sclose(source_space);
H5Sclose(virtual_space);
throw;
}
H5Sclose(source_space);
H5Sclose(virtual_space);
ret.emplace_back(std::move(mapping));
}
return ret;
}
+23 -1
View File
@@ -19,6 +19,24 @@ extern std::mutex hdf5_mutex;
class HDF5DataSet;
struct HDF5VirtualDatasetMapping {
std::string filename;
std::string dataset;
std::vector<hsize_t> virtual_start;
std::vector<hsize_t> virtual_stride;
std::vector<hsize_t> virtual_count;
std::vector<hsize_t> virtual_block;
std::vector<hsize_t> source_start;
std::vector<hsize_t> source_stride;
std::vector<hsize_t> source_count;
std::vector<hsize_t> source_block;
bool ContainsVirtualImage(hsize_t image_number) const;
hsize_t SourceImage(hsize_t image_number) const;
};
class HDF5Id {
protected:
hid_t id;
@@ -70,8 +88,11 @@ public:
~HDF5DataType();
};
enum class HDF5DataSetLayout {CONTIGUOUS, CHUNKED, VIRTUAL};
class HDF5Dcpl : public HDF5Id {
uint8_t ndim = 0;
HDF5DataSetLayout layout = HDF5DataSetLayout::CONTIGUOUS;
public:
HDF5Dcpl();
HDF5Dcpl(const HDF5DataSet& data_set);
@@ -87,7 +108,8 @@ public:
void SetCompression(CompressionAlgorithm algorithm, size_t block_size);
void SetVirtual(const std::string& filename, const std::string& dataset, const HDF5DataSpace& src_dataspace,
const HDF5DataSpace& virtual_dataspace);
HDF5DataSetLayout GetLayout() const;
std::vector<HDF5VirtualDatasetMapping> GetVirtualMappings() const;
uint8_t GetNumOfDimensions() const;
};