Files
Jungfraujoch/image_analysis/spot_finding/SpotExtractorGPU.h
T
leonarski_fandClaude Opus 5 0af109b838
Build Packages / build:rugnux:aarch64 (cross) (push) Successful in 8m39s
Build Packages / build:windows:nocuda (push) Successful in 17m3s
Build Packages / build:windows:cuda (push) Successful in 19m14s
Build Packages / build:rugnux-tgz (x86_64) (push) Successful in 19m38s
Build Packages / build:viewer-tgz:cpu (push) Successful in 20m32s
Build Packages / build:rpm (rocky9_nocuda) (push) Successful in 23m30s
Build Packages / build:viewer-tgz:cuda (push) Successful in 23m43s
Build Packages / build:rpm (ubuntu2204_nocuda) (push) Successful in 27m8s
Build Packages / build:rpm (rocky8_nocuda) (push) Successful in 27m12s
Build Packages / build:rugnux:windows (push) Successful in 10m52s
Build Packages / build:rpm (ubuntu2404_nocuda) (push) Successful in 19m27s
Build Packages / build:rpm (rocky9_sls9) (push) Successful in 21m13s
Build Packages / build:rpm (rocky8_sls9) (push) Successful in 26m49s
Build Packages / build:rpm (rocky9) (push) Successful in 23m36s
Build Packages / build:rpm (ubuntu2404) (push) Successful in 24m9s
Build Packages / Generate python client (push) Successful in 32s
Build Packages / build:rpm (rocky8) (push) Successful in 29m17s
Build Packages / Create release (push) Skipped
Build Packages / XDS test (durin plugin) (push) Successful in 11m18s
Build Packages / Build documentation (push) Successful in 1m10s
Build Packages / DIALS test (push) Successful in 25m11s
Build Packages / build:rpm (ubuntu2204) (push) Successful in 28m6s
Build Packages / XDS test (neggia plugin) (push) Successful in 8m56s
Build Packages / XDS test (JFJoch plugin) (push) Successful in 9m51s
Build Packages / Unit tests (push) Successful in 1h26m0s
Spot finding: bound how large a spot may be, not how bright
A component of more than 50 connected pixels was discarded. Under the
self-calibrating threshold that is an INTENSITY CEILING, not a size bound: the
threshold is an absolute per-ring contour, so a component's area above it grows
as sigma^2*ln(A/T), without bound in the peak amplitude. Measured on the strong
rotation set, footprints run 3 px at 30-100 counts to 50 px above 10000 - a
slope of 4.4 px per ln(peak) against the fixed local-box test's 0.8, which
saturates at 13 px and never reaches the bound at all. So the brighter a
reflection, the more certainly it was thrown away: 59% of the box finder's
d>3 A spots were missing from the adaptive finder's list, including every one
of its ten strongest, at an intensity ratio of 1.085 for those that did match.
Indexed spots per image collapsed from 220 to 7.

Raise the bound to 200 - CrystFEL peakfinder8's --max-pix-count, the only
directly comparable number in the field; XDS has no such parameter and guards
on shape instead - and ask a component above 50 pixels to be COMPACT: it must
fill a fifth of the square its bounding box fits inside. A Bragg reflection is
round and fills about half of that square however bright it is; an ice arc, a
cosmic-ray track or a lit detector row fills a fifth or less, and those are what
an upper bound was ever protecting against. Below 50 nothing is asked of the
shape, so every component accepted before still is. Integer arithmetic on both
sides, and on the GPU the bounding side fits in what was padding, so the device
struct does not grow.

The shape test is what makes the raise safe. With a flat 200 alone, two battery
crystals moved: one lost a little I/sigma, and the other's de-novo lattice was
NOT MONOTONE in the bound - correct at 50, 100 and 200, wrong at 150 and at 250
and above - so 200 was partly luck. With the shape test both are unchanged to
three decimals.

De novo at defaults on the strong set: indexing rate 0.574 -> 0.804,
completeness 97.3 -> 99.6%, <I/sigma> 3.42 -> 6.07, R_meas 0.299 -> 0.249,
CC1/2 0.947 -> 0.961, ISa 3.29 -> 4.13. The full 37-crystal battery, scored per
shell, is still owed.

Also documents what StrongPixelSet::AddStrongPixel has required since the
component search became linear - pixels in raster order - and puts the existing
test's insertion order into it. Both callers scan a bitmap in ascending flat
index and always satisfied it; the test did not, and was the only thing that
did not.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FBumeJVx4oeXxiBRpkrE5H
2026-08-28 11:49:49 +02:00

101 lines
5.0 KiB
C++

// SPDX-FileCopyrightText: 2026 Filip Leonarski, Paul Scherrer Institute <filip.leonarski@psi.ch>
// SPDX-License-Identifier: GPL-3.0-only
#pragma once
// Device-side connected-component extraction for the GPU spot finders.
//
// The GPU finders flag strong pixels into a packed bit buffer ON THE DEVICE. Reading spots out of it
// used to mean copying that whole buffer back (2.26 MB per frame at 18 MP) and scanning it bit by bit
// on the host. This does the whole extraction where the data already is, so nothing about the image
// comes back - only the finished spot list, a few hundred entries.
//
// The algorithm is the sparse formulation the ACTS/traccc project settled on for the same problem
// (sparse silicon-detector hits): the strong pixels are compacted into a list that is sorted by flat
// index, each pixel finds its at most FOUR backward 8-neighbours by binary search in that list, and
// the resulting graph is labelled with a lock-free union-find. A dense image-wide labelling
// (Playne-equivalence, BUF/BKE, nppiLabelMarkers, cv::cuda::connectedComponents) would label 18
// million pixels to find five hundred.
//
// It reproduces the host StrongPixelSet::sparseccl EXACTLY, not just equivalently:
// * both make a component's root its lowest list index, so both find the same roots;
// * labels are handed out by a prefix sum over the roots in ascending order, which is the order the
// host's second scan hands them out in, so the SPOT ORDER is identical;
// * the centroid sums are accumulated per component in ascending list order, in integers, term for
// term as DiffractionSpot::AddPixel does them, so there is no rounding for the two compilers to
// disagree about.
// tests/SpotExtractorGPUParityTest.cpp holds the two to each other on realistic, occupancy-swept and
// pathological frames, and checks that repeating a frame gives byte-identical output.
#include <cstdint>
#include <memory>
#include <vector>
#include "../../common/DiffractionSpot.h"
#include "../indexing/CUDAMemHelpers.h"
#include "SpotFindingSettings.h"
// Per-component sums, in exactly the form DiffractionSpot holds them: x and y are sum(col*photons)
// and sum(line*photons), not a centroid.
struct SpotExtractorGPUSpot {
int64_t x;
int64_t y;
int64_t photons;
int64_t max_photons;
int32_t pixel_count;
// Longer side of the component's bounding box, for SpotShapeAccepted. It fits in what used to be
// padding, so carrying it costs nothing.
int32_t bbox_side;
};
class SpotExtractorGPU {
std::shared_ptr<CudaStream> stream;
const int32_t width;
const size_t nwords;
// Strong pixels this engine's buffers hold, and above which the extraction gives up on the frame -
// StrongPixelLimit, so it follows the detector rather than standing at a constant. 104 bytes of
// device memory apiece, 29 MB on an 18-megapixel detector.
const uint32_t max_strong;
// Spots copied back together with their count in one transfer. A frame with more than this many
// surviving spots - far past anything indexable - simply takes a second copy.
static constexpr uint32_t SPOT_PREFIX = 4096;
int compact_blocks = 0;
CudaDevicePtr<uint32_t> gpu_res_mask; // packed, bit set = pixel excluded
CudaDevicePtr<uint32_t> gpu_block_count;
CudaDevicePtr<uint32_t> gpu_block_offset;
CudaDevicePtr<uint32_t> gpu_nstrong;
CudaDevicePtr<uint32_t> gpu_index; // strong pixels, sorted by flat index
CudaDevicePtr<int32_t> gpu_value;
CudaDevicePtr<uint32_t> gpu_parent; // union-find parent
CudaDevicePtr<uint32_t> gpu_root;
CudaDevicePtr<uint32_t> gpu_label; // compact label, indexed by root
CudaDevicePtr<int32_t> gpu_count; // pixels per component
CudaDevicePtr<SpotExtractorGPUSpot> gpu_spot;
CudaDevicePtr<SpotExtractorGPUSpot> gpu_spot_out;
CudaDevicePtr<uint32_t> gpu_nspot;
CudaHostPtr<uint32_t> host_nstrong;
CudaHostPtr<uint32_t> host_nspot;
CudaHostPtr<SpotExtractorGPUSpot> host_spot; // SPOT_PREFIX entries, pinned
std::vector<SpotExtractorGPUSpot> overflow_spot; // only for a frame with more spots than that
public:
SpotExtractorGPU(int32_t width, int32_t height, std::shared_ptr<CudaStream> stream);
void SetResolutionMask(const std::vector<uint32_t> &packed_mask);
// gpu_strong is the finder's device bit buffer, gpu_image the preprocessed image it was built
// from. Fills spots with every component of at most max-pix pixels, in the same order the host
// extractor would.
void Extract(const uint32_t *gpu_strong, const int32_t *gpu_image,
const SpotFindingSettings &settings, std::vector<DiffractionSpot> &spots);
// Strong pixels the last Extract() saw, after the resolution mask. Reported whether or not the
// frame was given up on, which is the point of it: a frame at or above max_strong yields no spots
// at all, and this is what says so.
[[nodiscard]] uint32_t StrongPixelCount() const { return *host_nstrong.get(); }
};