Files
Jungfraujoch/receiver/JFJochReceiverLite.cpp
leonarski_fandClaude Opus 5 e7be5447d3
Build Packages / Unit tests (push) Successful in 1h1m55s
Build Packages / build:viewer-tgz:cpu (push) Successful in 8m10s
Build Packages / build:viewer-tgz:cuda (push) Successful in 9m20s
Build Packages / build:rpm (rocky8_nocuda) (push) Successful in 14m6s
Build Packages / build:rpm (rocky9_nocuda) (push) Successful in 14m9s
Build Packages / build:rpm (ubuntu2204_nocuda) (push) Successful in 14m13s
Build Packages / build:rpm (ubuntu2404_nocuda) (push) Successful in 13m43s
Build Packages / build:rpm (rocky8_sls9) (push) Successful in 14m22s
Build Packages / build:rpm (rocky9_sls9) (push) Successful in 13m21s
Build Packages / build:rpm (rocky8) (push) Successful in 12m0s
Build Packages / build:rpm (rocky9) (push) Successful in 13m23s
Build Packages / build:rpm (ubuntu2204) (push) Successful in 13m18s
Build Packages / build:rpm (ubuntu2404) (push) Successful in 13m6s
Build Packages / DIALS test (push) Successful in 13m59s
Build Packages / XDS test (durin plugin) (push) Successful in 8m4s
Build Packages / XDS test (JFJoch plugin) (push) Successful in 8m40s
Build Packages / XDS test (neggia plugin) (push) Successful in 8m1s
Build Packages / Generate python client (push) Successful in 32s
Build Packages / Build documentation (push) Successful in 1m9s
Build Packages / Create release (push) Skipped
Build Packages / build:windows:nocuda (push) Failing after 13m23s
Build Packages / build:windows:cuda (push) Failing after 12m24s
receiver: stop copying every frame back from the device on the Lite path
The Lite workflow built its analysis with the fused GPU engine disabled, which is
also what decides whether the preprocessed image is copied device-to-host after
every frame. So on a machine with a GPU the online path was moving the whole image
back - 72 MB on a large detector, every frame, per worker - for a host reader that
does not exist on that path.

It was left off deliberately when the fused engine was added, to keep the online
path unchanged in that commit, and never revisited. Nothing depends on it: the FPGA
workflow uses a different analysis class, and strong-pixel values are read through a
device gather rather than from the host image.

Turning it on changes no result, and cannot: adaptive detection is unreachable
online, because the REST schema exposes no way to enable it, so the classic GPU
finder runs either way. Measured anyway, both engines on the same frames across five
datasets including very weak ones: 2400 frames, 638260 spots, not one difference -
identical lists, identical indexing rate, identical merge statistics to every
printed digit.

On a large detector with eight workers the median per-image cost falls from 94 to
59 ms and preprocessing from 21 to 6 ms; throughput rises from about 48 to 55 Hz. No
percentile regresses, which is what matters for a service - the ninetieth improves
from 128 to 74 ms and the tail with it. Spot finding gets faster too, because the
large copy no longer contends with the device gather.

Correct two statements while here. The flag's comment and the data-analysis
document both said the online receiver uses the CPU adaptive finder; online never
runs an adaptive finder at all, and the copy the flag really controls was not
mentioned. That copy would be better expressed as what it is - whether a host engine
will read the image, which the constructor already knows - rather than inferred from
which spot finder is wanted.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-08-02 15:16:18 +02:00

432 lines
18 KiB
C++

// SPDX-FileCopyrightText: 2024 Filip Leonarski, Paul Scherrer Institute <filip.leonarski@psi.ch>
// SPDX-License-Identifier: GPL-3.0-only
#include "JFJochReceiverLite.h"
#include "../image_analysis/indexing/IndexerFactory.h"
#include "../common/CUDAWrapper.h"
using namespace std::chrono_literals;
namespace {
// Serialize a processed image into its buffer slot. The slot holds the compressed
// image plus the per-image CBOR metadata (spot list, reflection list, ...); if an
// unusually rich frame makes that metadata too large to fit alongside the image,
// drop just this frame (log + recycle the slot) instead of letting the serializer
// throw, which would abort the whole data collection. Returns true when the image
// was serialized and the slot is ready for the caller to commit.
bool SerializeImageOrDrop(CBORStream2Serializer &serializer, DataMessage &msg,
ZeroCopyReturnValue &loc, size_t slot_size, Logger &logger) {
try {
serializer.SerializeImage(msg);
return true;
} catch (const JFJochException &) {
// Confirm this is a slot-overflow (not some other CBOR error) by re-serializing
// the metadata alone with an empty image placeholder: that always fits when the
// image is what overflowed, and gives the exact metadata size for the log. If it
// throws too, it is not a simple overflow and the exception propagates.
const auto image = msg.image;
msg.image = CompressedImage(nullptr, 0, image.GetWidth(), image.GetHeight(),
image.GetMode(), image.GetCompressionAlgorithm(),
image.GetChannel());
CBORStream2Serializer meas(static_cast<uint8_t *>(loc.GetImage()), slot_size);
meas.SerializeImage(msg);
const size_t metadata_size = meas.GetImageAppendOffset();
msg.image = image;
logger.Error("Dropping image {} - CBOR metadata too large to store image: "
"metadata {} B + image {} B do not fit in buffer slot {} B",
msg.number, metadata_size, image.GetCompressedSize(), slot_size);
loc.release();
return false;
}
}
}
int64_t JFJochReceiverLite::NumberOfDataAnalysisThreads(int64_t requested_thread_number,
const DiffractionExperiment& in_experiment) {
int64_t number_of_images = in_experiment.GetImageNum();
auto image_time = in_experiment.GetImageTime();
if (requested_thread_number <= 0)
return 1;
// For very small datasets no need to go multithreaded
if (number_of_images < 4)
return 1;
if (number_of_images < 16)
return std::min<int64_t>(2, requested_thread_number);
// For 100 Hz, there is no reason to go for a very large number of threads
if (number_of_images < 256 || (image_time >= 10ms))
return std::min<int64_t>(16, requested_thread_number);
return requested_thread_number;
}
JFJochReceiverLite::JFJochReceiverLite(const DiffractionExperiment &in_experiment,
const PixelMask &in_pixel_mask,
ImagePuller &in_image_puller,
ImagePusher &in_image_pusher,
Logger &in_logger,
int64_t forward_and_sum_nthreads,
const SpotFindingSettings &in_spot_finding_settings,
JFJochReceiverCurrentStatus &in_current_status,
JFJochReceiverPlots &in_plots,
ImageBuffer &in_image_buffer,
ZMQPreviewSocket *in_zmq_preview_socket,
ZMQMetadataSocket *in_zmq_metadata_socket,
IndexerThreadPool *indexing_thread_pool)
: JFJochReceiver(in_experiment,
in_image_buffer,
in_image_pusher,
in_current_status,
in_plots,
in_spot_finding_settings,
in_logger,
in_pixel_mask,
in_zmq_preview_socket,
in_zmq_metadata_socket,
indexing_thread_pool),
image_puller(in_image_puller),
data_analysis_nthreads(NumberOfDataAnalysisThreads(forward_and_sum_nthreads, in_experiment)),
data_analysis_started(data_analysis_nthreads),
measurement_started(1),
dark_mask_analysis(in_experiment.GetDarkMaskSettings(), in_experiment.GetPixelsNum()) {
logger.Info("Starting {} data analysis threads", data_analysis_nthreads);
if (experiment.GetDetectorMode() == DetectorMode::DarkMask) {
for (int i = 0; i < data_analysis_nthreads; i++)
data_analysis_futures.emplace_back(
std::async(std::launch::async, &JFJochReceiverLite::MaskThread, this, i)
);
} else {
// Start frame transformation threads
for (int i = 0; i < data_analysis_nthreads; i++)
data_analysis_futures.emplace_back(
std::async(std::launch::async, &JFJochReceiverLite::DataAnalysisThread, this, i)
);
}
measurement = std::async(std::launch::async, &JFJochReceiverLite::MeasurementThread, this);
data_analysis_started.wait();
logger.Info("Data analysis threads ready");
}
JFJochReceiverLite::~JFJochReceiverLite() {
cancelled = true;
try {
if (measurement.valid())
measurement.get();
} catch (const std::exception& e) {
logger.Error("ReceiverLite teardown failed: {}", e.what());
} catch (...) {
logger.Error("ReceiverLite teardown failed with unknown exception");
}
}
void JFJochReceiverLite::MeasurementThread() {
try {
// Wait for start message to arrive
auto msg = image_puller.PollImage();
while (!cancelled && (!msg.has_value() || !msg->cbor || !msg->cbor->start_message))
msg = image_puller.PollImage();
if (cancelled) {
current_status.SetProgress({});
current_status.SetStatus(GetStatus());
measurement_started.count_down();
return; // just quit the function - no measurement, no results
}
Configure(msg->cbor->start_message.value());
image_buffer.StartMeasurement(experiment); // Only at this point we know bit-depth of the images
start_time = std::chrono::system_clock::now();
// Send new start message out
SendStartMessage();
measurement_started.count_down();
} catch (const JFJochException &e) {
logger.ErrorException(e);
Cancel(e);
measurement_started.count_down();
throw;
}
logger.Info("Receiving started");
// Analysis is running
// ...
// Till it is done.
// Combine frame transformation threads
for (auto &f: data_analysis_futures)
f.get();
logger.Info("Data analysis threads finished");
current_status.SetProgress(1.0);
current_status.SetStatus(GetStatus());
// Send end message out
SendEndMessage();
if (!image_buffer.CheckIfBufferReturned(std::chrono::seconds(10))) {
logger.Error("Send commands not finalized in 10 seconds");
throw JFJochException(JFJochExceptionCategory::ZeroMQ, "Send commands not finalized in 10 seconds");
}
logger.Info("All images sent through ZeroMQ");
if (push_images_to_writer)
writer_error = image_pusher.Finalize();
logger.Info("Writing process finalized");
end_time = std::chrono::system_clock::now();
current_status.SetProgress({});
current_status.SetStatus(GetStatus());
}
void JFJochReceiverLite::Configure(const StartMessage &msg) {
if ((experiment.GetXPixelsNum() != msg.image_size_x)
|| (experiment.GetYPixelsNum() != msg.image_size_y))
throw JFJochException(JFJochExceptionCategory::InputParameterInvalid,
"Mismatch in detector size");
if (fabs(experiment.GetPixelSize_mm() - msg.pixel_size_x * 1e3) > 1e-7)
throw JFJochException(JFJochExceptionCategory::InputParameterInvalid,
"Mismatch in pixel size");
experiment.Detector().Description(msg.detector_description);
experiment.Detector().SerialNumber(msg.detector_serial_number);
experiment.Detector().SensorMaterial(msg.sensor_material);
experiment.Detector().SensorThickness_um(msg.sensor_thickness * 1e6);
experiment.Detector().SaturationLimit(msg.saturation_value);
experiment.Detector().BitDepthImage(msg.bit_depth_image);
if (msg.bit_depth_readout)
experiment.Detector().BitDepthReadout(msg.bit_depth_readout.value());
}
void JFJochReceiverLite::MaskThread(uint32_t id) {
std::vector<uint8_t> buffer;
data_analysis_started.count_down();
measurement_started.wait();
while (!cancelled && !end_message_received) {
try {
auto msg = image_puller.PollImage(std::chrono::milliseconds(5));
if (!msg.has_value() || !msg->cbor) {
// Message not received or not parsed
continue;
} else if (msg->cbor->end_message) {
end_message_received = true;
logger.Debug("Thread {} end message received in JFJochReceiverLite", id);
} else if (msg->cbor->data_message) {
++images_collected;
DataMessage data_msg = msg->cbor->data_message.value();
dark_mask_analysis.AnalyzeImage(data_msg, buffer);
compressed_size += data_msg.image.GetCompressedSize();
uncompressed_size += data_msg.image.GetUncompressedSize();
UpdateMaxImageReceived(data_msg.number);
auto loc = image_buffer.GetImageSlot();
if (loc == nullptr)
writer_queue_full = true;
else {
auto writer_buffer = (uint8_t *) loc->GetImage();
CBORStream2Serializer serializer(writer_buffer, experiment.GetImageBufferLocationSize());
if (SerializeImageOrDrop(serializer, data_msg, *loc,
experiment.GetImageBufferLocationSize(), logger)) {
loc->SetImageNumber(data_msg.number);
loc->SetImageSize(serializer.GetBufferSize());
loc->SetIndexed(false);
loc->release();
}
}
}
current_status.SetProgress(GetProgress());
current_status.SetStatus(GetStatus());
} catch (const JFJochException &e) {
logger.ErrorException(e);
Cancel(e);
}
}
}
void JFJochReceiverLite::DataAnalysisThread(uint32_t id) {
std::unique_ptr<MXAnalysisWithoutFPGA> analysis;
logger.Debug("Thread {} started", id);
try {
pin_gpu();
} catch (const JFJochException &e) {
logger.Warning("Error pinning GPU {}", e.what());
}
data_analysis_started.count_down();
measurement_started.wait();
try {
analysis = std::make_unique<MXAnalysisWithoutFPGA>(experiment, *az_int_mapping, pixel_mask, indexer,
/*enable_fused_adaptive_gpu=*/true);
} catch (const JFJochException &e) {
Cancel(e);
return;
}
while (!cancelled && !end_message_received) {
try {
auto msg = image_puller.PollImage(std::chrono::milliseconds(5));
if (!msg.has_value() || !msg->cbor) {
// Message not received or not parsed
continue;
} else if (msg->cbor->end_message) {
// Message is end message
end_message_received = true;
logger.Debug("Thread {} End message received in JFJochReceiverLite", id);
} else if (msg->cbor->calibration) {
// Calibration messages are just forwarded
if (push_images_to_writer)
image_pusher.SendCalibration(msg->cbor->calibration.value());
} else if (msg->cbor->data_message) {
auto start = std::chrono::high_resolution_clock::now();
DataMessage data_msg = msg->cbor->data_message.value();
++images_collected;
auto compressed_size_img = data_msg.image.GetCompressedSize();
auto uncompressed_size_img = data_msg.image.GetUncompressedSize();
compressed_size += compressed_size_img;
uncompressed_size += uncompressed_size_img;
UpdateMaxImageReceived(data_msg.number);
auto image_start_time = std::chrono::high_resolution_clock::now();
AzimuthalIntegrationProfile profile(*az_int_mapping);
analysis->Analyze(data_msg, profile, GetSpotFindingSettings());
auto image_end_time = std::chrono::high_resolution_clock::now();
std::chrono::duration<float> image_duration = image_end_time - image_start_time;
data_msg.processing_time_s = image_duration.count();
data_msg.original_number = data_msg.number;
data_msg.user_data = experiment.GetImageAppendix();
data_msg.run_number = experiment.GetRunNumber();
data_msg.run_name = experiment.GetRunName();
data_msg.receiver_buf_available = image_buffer.GetAvailSlots();
data_msg.receiver_aq_dev_delay = image_puller.GetCurrentFifoUtilization();
// Compression ratio is uncompressed/compressed (e.g. 7x), matching the FPGA path
// (JFJochReceiverFPGA) and the aggregate (JFJochReceiver::GetFinalStatistics) - not the
// reciprocal.
if (compressed_size_img > 0)
data_msg.compression_ratio = static_cast<float>(uncompressed_size_img)
/ static_cast<float>(compressed_size_img);
saturated_pixels.Add(data_msg.saturated_pixel_count);
error_pixels.Add(data_msg.error_pixel_count);
if (data_msg.roi.contains("beam")) {
roi_beam_npixel.Add(data_msg.roi["beam"].pixels);
roi_beam_sum.Add(data_msg.roi["beam"].sum);
}
plots.Add(data_msg, profile);
scan_result.Add(data_msg);
if (!serialmx_filter.ApplyFilter(data_msg))
++images_skipped;
else {
auto loc = image_buffer.GetImageSlot();
if (loc == nullptr)
writer_queue_full = true;
else {
auto writer_buffer = (uint8_t *) loc->GetImage();
CBORStream2Serializer serializer(writer_buffer, experiment.GetImageBufferLocationSize());
if (SerializeImageOrDrop(serializer, data_msg, *loc,
experiment.GetImageBufferLocationSize(), logger)) {
loc->SetImageNumber(data_msg.number);
loc->SetImageSize(serializer.GetBufferSize());
loc->SetIndexed(data_msg.indexing_result.value_or(false));
loc->ReadyToSend();
if (zmq_preview_socket != nullptr)
zmq_preview_socket->SendImage(writer_buffer, serializer.GetBufferSize());
if (zmq_metadata_socket != nullptr)
zmq_metadata_socket->AddDataMessage(data_msg);
if (push_images_to_writer) {
// Only count images the pusher accepted; TCP can drop on a
// broken/absent connection or an expired enqueue deadline.
if (image_pusher.SendImage(*loc))
++images_sent;
} else
loc->release();
UpdateMaxImageSent(data_msg.number);
}
}
}
auto end = std::chrono::high_resolution_clock::now();
std::chrono::duration<float> duration_total = end - start;
logger.Debug("Thread {} Image {:6d} processing {:8.04f} s analysis {:8.04f} s",
id, data_msg.number, duration_total.count(), image_duration.count());
}
UpdateMaxDelay(image_puller.GetCurrentFifoUtilization());
current_status.SetProgress(GetProgress());
current_status.SetStatus(GetStatus());
} catch (const JFJochException &e) {
logger.ErrorException(e);
Cancel(e);
}
}
logger.Debug("Thread {} finished", id);
}
void JFJochReceiverLite::StopReceiver() {
if (measurement.valid()) {
measurement.get();
logger.Info("Receiver stopped");
}
}
float JFJochReceiverLite::GetEfficiency() const {
if (experiment.GetFrameNum() == 0)
return 0;
return static_cast<float>(images_collected) / static_cast<float>(experiment.GetFrameNum());
}
float JFJochReceiverLite::GetProgress() const {
if (experiment.GetFrameNum() == 0)
return 0.0;
return static_cast<float>(max_image_number_received) / static_cast<float>(experiment.GetFrameNum());
}
void JFJochReceiverLite::SetSpotFindingSettings(const SpotFindingSettings &in_spot_finding_settings) {
std::unique_lock ul(spot_finding_settings_mutex);
DiffractionExperiment::CheckDataProcessingSettings(in_spot_finding_settings);
spot_finding_settings = in_spot_finding_settings;
}
JFJochReceiverOutput JFJochReceiverLite::GetFinalStatistics() const {
JFJochReceiverOutput ret = JFJochReceiver::GetFinalStatistics();
if (experiment.GetDetectorMode() == DetectorMode::DarkMask)
ret.dark_mask_result = dark_mask_analysis.GetMask();
return ret;
}