Implement new doublet cut based on triplet radii

[stephenswat]

This commit adds a new cut to the doublet finding which is able to determine based on only the doublets if it is possible to find a third spacepoint which will form a seed within the required helix radius bounds.

Note that this cut should, mathematically, not remove any triplets, but it should reduce the time taken to find triplets.

This is the cut discussed on the traccc Mattermost recently.

Pinging @krasznaa and @pbutti as interested parties.

[stephenswat]

This commit was inspired by the G-200 performance breakdown shown today, where the seed finding takes up an outsided amount of time:

Thus, optimising the seeding is once again the flavour of the month.

[stephenswat]

Performance summary

Here is a summary of the performance effects of this PR:

Graphical

Tabular

Kernel	32950a5	ef228f5	Delta
fit_forward	109.46 ms	109.15 ms	-0.3%
propagate_to_next_surface	69.93 ms	69.96 ms	0.0%
fit_backward	54.04 ms	53.82 ms	-0.4%
count_triplets	14.24 ms	8.19 ms	-42.5%
find_triplets	6.01 ms	4.37 ms	-27.2%
find_tracks	4.28 ms	4.28 ms	0.0%
count_doublets	618.27 μs	1.57 ms	154.0%
find_doublets	890.84 μs	1.33 ms	49.7%
fit_prelude	1.12 ms	1.12 ms	0.2%
ccl_kernel	872.06 μs	872.65 μs	0.1%
select_seeds	360.96 μs	348.10 μs	-3.6%
Thrust::sort	271.94 μs	271.38 μs	-0.2%
build_tracks	194.24 μs	192.64 μs	-0.8%
update_triplet_weights	100.28 μs	99.14 μs	-1.1%
DeviceRadixSortHistogramKernel	97.85 μs	97.69 μs	-0.2%
apply_interaction	63.69 μs	63.84 μs	0.2%
DeviceRadixSortOnesweepKernel	54.83 μs	54.61 μs	-0.4%
fill_sort_keys	44.78 μs	44.58 μs	-0.4%
estimate_track_params	34.46 μs	34.37 μs	-0.3%
populate_grid	30.32 μs	30.36 μs	0.1%
count_grid_capacities	29.20 μs	29.09 μs	-0.4%
unknown	21.50 μs	20.52 μs	-4.6%
form_spacepoints	12.59 μs	12.46 μs	-1.0%
reduce_triplet_counts	6.62 μs	5.36 μs	-19.1%
make_barcode_sequence	1.00 μs	1.02 μs	1.2%
DeviceRadixSortExclusiveSumKernel	510.28 ns	510.45 ns	0.0%
fill_prefix_sum	165.49 ns	165.37 ns	-0.1%
Total	262.78 ms	255.97 ms	-2.6%

Important

All metrics in this report are given as reciprocal throughput, not as wallclock runtime.

Warning

At least one kernel incurred a significant performance regression.

Note

This is an automated message produced on the explicit request of a human being.

[stephenswat]

Quick note, this looks unspectacular but if I put in the numbers from the ITk execution and keep the percentages the same it looks like this:

Kernel	32950a5	ef228f5	Delta
count_triplets	295.23 ms	169.75 ms	-42.5%
find_triplets	67.29 ms	48.98 ms	-27.2%
count_doublets	19.64 ms	49.88 ms	154.0%
find_doublets	31.46 ms	47.09 ms	49.7%
Total	413.62 ms	315.70 ms	-23.7%

[stephenswat]

This is now ready to go!

[krasznaa]

I like it. As long as the comparison test with Acts doesn't break, I'm happy to go forward with this.

[sonarqubecloud]

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud