-- Instead of storing a list of list of alignment starts, which is expensive to manipulate, we instead store a linear list of alignment starts. Not grouped as previously. This enables us to simplify iteration and update operations, making them much faster
-- Critically, the downsampler still requires this list of list. We convert back and forth between these two representations as required, which is very rarely for normal data sets (WGS NA12878 on chr20 is 0.2%, 4x WGS is even less).
-- No longer update the total counts in each per-sample state manager, but instead return delta counts that are updated by the overall ReadStateManager
-- One step on the way to improving the underlying representation of the data in PerSampleReadStateManager
-- Make LocusIteratorByState final
-- Use a linked hash map instead of a hash map since we want to iterate through the map fairly often
-- Ensure that we call doneSubmittingReads before getting reads for samples. This function call fell out before and since it wasn't enforced I only noticed the problem while writing comments
-- Don't make unnecessary calls to contains for map. Just use get() and check that the result is null
-- Use a LinkedList in PassThroughDownsampler, since this is faster for add() than the existing ArrayList, and we were's using random access to any resulting
-- Made LIBSPerformance a full featured CommandLineProgram, and it can be used to assess the LIBS performance by reading a provided BAM
-- ReadStateManager now provides a clean interface to iterate in sample order the per-sample read states, allowing us to avoid many map.get calls
-- Moved updateReadStates to ReadStateManager
-- Removed the unnecessary wrapping of an iterator in ReadStateManager
-- readStatesBySample is now a LinkedHashMap so that iteration occurs in LIBS sample order, allowing us to avoid many unnecessary calls to map.get iterating over samples. Now those are just map native iterations
-- Restructured collectPendingReads for simplicity, removing redundant and consolidating common range checks. The new piece is code is much clearer and avoids several unnecessary function calls
-- Only ReadBackedPileupImpl (concrete class) and ReadBackedPileup (interface) live, moved all functionality of AbstractReadBackedPileup into the impl
-- ReadBackedPileupImpl was literally a shell class after we removed extended events. A few bits of code cleanup and we reduced a bunch of class complexity in the gatk
-- ReadBackedPileups no longer accept pre-cached values (size, nMapQ reads, etc) but now lazy load these values as needed
-- Created optimized calculation routines to iterator over all of the reads in the pileup in whatever order is most efficient as well.
-- New LIBS no longer calculates size, n mapq, and n deletion reads while making pileups.
-- Added commons-collections for IteratorChain
-- function to create pileup elements in AlignmentStateMachine and LIBS
-- Cleanup pileup element constructors, directing users to LIBS.createPileupFromRead() that really does the right thing
-- Optimizations to AlignmentStateMachine
-- Properly count deletions. Added unit test for counting routines
-- AlignmentStateMachine.java is no longer recursive
-- Traversals now use new LIBS, not the old one
-- AlignmentStateMachine does what SAMRecordAlignmentState should really do. It's correct in that it's more accurate than the LIB_position tests themselves. This is a non-broken, correct implementation. Needs cleanup, contracts, etc.
-- This version is like 6x slower than the original implementation (according to the google caliper benchmark here). Obvious optimizations for future commit
-- This capability is essential to provide an ordered set of used reads to downstream users of LIBS, such as ART, who want an efficient way to get the reads used in LIBS
-- Vastly expanded the multi-read, multi-sample LIBS unit tests to make sure this capability is working
-- Added createReadStream to ArtificialSAMUtils that makes it relatively easy to create multi-read, multi-sample read streams for testing
-- Split out all of the inner classes of LIBS into separate independent classes
-- Split / add unit tests for many of these components.
-- Radically expand unit tests for SAMRecordAlignmentState (the lowest level piece of code) making sure at least some of it works
-- No need to change unit tests or integration tests. No change in functionality.
-- Added (currently disabled) code to track all submitted reads to LIBS, but this isn't accessible or tested
Instead of the GATK Engine creating a new BaseRecalibrator (not clean), it just keeps track of the arguments (clean).
There are still some dependency issues, but it looks like they are related to Ami's code. Need to look into it further.
-- Added unit tests for combining RecalibrationTables. As a side effect now has serious tests for incrementDatumOrPutIfNecessary
-- Removed unnecessary enum.index system from RecalibrationTables.
-- Moved what were really static utility methods out of RecalibrationEngine and into RecalUtils.
-- Added unit tests for EventType and ReadRecalibrationInfo
-- Simplified interface of EventType. Previously this enum carried an index with it, but this is redundant with the enum.ordinal function. Now just using that function instead.
-- With the newer, faster BQSR, scaling was limited by the NestedIntegerArray. The solution to this is to make the entire table thread-local, so that each nct thread has its own data and doesn't have any collisions.
-- Removed the previous partial solution of having a thread-local quality score table
-- Added a new argument -lowMemory
- Made few small modifications to code
- Replaced the two arguments in GATKReportTable constructor with an enum used to specify way of sorting the table
This isn't hooked up yet with BQSR; it's just a static method used in my testing walker. I'll hook this into BQSR after more testing and the addition of unit tests.
Most of the changes in this commit are actually documentation-related.
-- Underlying system now uses long nano times to be more consistent with standard java practice
-- Updated a few places in the code that were converting from nanoseconds to double seconds to use the new nanoseconds interface directly
-- Bringing us to 100% test coverage with clover with AutoFormattingTimeUnitTest
-- Intermediate commit on the way to archiving SomaticIndelDetector and other tools.
-- SomaticIndelDetector, PairMaker and RemapAlignments tools have been refactored into the private andrey package. All utility classes refactored into here as well. At this point, the SomaticIndelDetector builds in this version of the GATK.
-- Subsequent commit will put this code into the archive so it no longer builds in the GATK
-- AdvancedRecalibrationEngine now uses a thread-local table for the quality score table, and in finalizeData merges these thread-local tables into the final table. Radically reduces the contention for RecalDatum in this very highly used table
-- Refactored the utility function to combine two tables into RecalUtils, and created UnitTests for this function, as well as all of RecalibrationTables. Updated combine in RecalibrationReport to use this table combiner function
-- Made several core functions in RecalDatum into final methods for performance
-- Added RecalibrationTestUtils, a home for recalibration testing utilities
-- The previous model was to enqueue individual map jobs (with a resolution of 1 map job per map call), to track the number of map calls submitted via a counter and a semaphore, and to use this information in each map job and reduce to control the number of map jobs, when reduce was complete, etc. All hideously complex.
-- This new model is vastly simply. The reducer basically knows nothing about the control mechanisms in the NanoScheduler. It just supports multi-threaded reduce. The NanoScheduler enqueues exactly nThread jobs to be run, which continually loop reading, mapping, and reducing until they run out of material to read, when they shut down. The master thread of the NS just holds a CountDownLatch, initialized to nThreads, and when each thread exits it reduces the latch by 1. The master thread gets the final reduce result when its free by the latch reaching 0. It's all super super simple.
-- Because this model uses vastly fewer synchronization primitives within the NS itself, it's naturally much faster at getting things done, without any of the overhead obvious in profiles of BQSR -nct 2.
-- reduceAsMuchAsPossible no longer blocks threads via synchronization, but instead uses an explicit lock to manage access. If the lock is already held (because some thread is doing reduce) then the thread attempting to reduce immediately exits the call and continues doing productive work. They removes one major source of blocking contention in the NanoScheduler
-- Created a separate, limited interface MapResultsQueue object that previously was set to the PriorityBlockingQueue.
-- The MapResultsQueue is now backed by a synchronized ExpandingArrayList, since job ids are integers incrementing from 0 to N. This means we avoid the n log n sort in the priority queue which was generating a lot of cost in the reduce step
-- Had to update ReducerUnitTest because the test itself was brittle, and broken when I changed the underlying code.
-- A few bits of minor code cleanup through the system (removing unused constructors, local variables, etc)
-- ExpandingArrayList called ensureCapacity so that we increase the size of the arraylist once to accommodate the upcoming size needs
-- Pre-read MapData into a list, which is actually faster than dealing with future lock contention issues with lots of map threads
-- Increase the ReadShard default size to 100K reads by default
-- Created a ReadRecalibrationInfo class that holds all of the information (read, base quality vectors, error vectors) for a read for the call to updateDataForRead in RecalibrationEngine. This object has a restrictive interface to just get information about specific qual and error values at offset and for event type. This restrict allows us to avoid creating an vector of byte 45 for each read to represent BI and BD values not in the reads. Shaves 5% of the runtime off the entire code.
-- Cleaned up code and added lots more docs
-- With this commit we no longer have much in the way of low-hanging fruit left in the optimization of BQSR. 95% of the runtime is spent in BAQing the read, and updating the RecalData in the NestedIntegerArrays.
-- Update SAMDataSource so that the merged header contains GATKSAMReadGroupRecord
-- Now getting the NGSPlatform for a GATKSAMRecord is actually efficient, instead of computing the NGS platform over and over from the PL string
-- Updated a few places in the code where the input argument is actually a GATKSAMRecord, not a SAMRecord for type safety
- Added an optional argument to BaseRecalibrator to produce sorted GATKReport Tables
- Modified BSQR Integration Tests to include the optional argument. Tests now produce sorted tables
the function ls_getLicenseUsage() is not supported by LSF v8.x, comment the line:
public static native lsfLicUsage.ByReference ls_getLicenseUsage()
Signed-off-by: Eric Banks <ebanks@broadinstitute.org>
This is an intermediate commit so that there is a record of these changes in our
commit history. Next step is to isolate the test classes as well, and then move
the entire package to the Picard repository and replace it with a jar in our repo.
-Removed all dependencies on org.broadinstitute.sting (still need to do the test classes,
though)
-Had to split some of the utility classes into "GATK-specific" vs generic methods
(eg., GATKVCFUtils vs. VCFUtils)
-Placement of some methods and choice of exception classes to replace the StingExceptions
and UserExceptions may need to be tweaked until everyone is happy, but this can be
done after the move.
-- Now each map job reads a value, performs map, and does as much reducing as possible. This ensures that we scale performance with the nct value, so -nct 2 should result in 2x performance, -nct 3 3x, etc. All of this is accomplished using exactly NCT% of the CPU of the machine.
-- Has the additional value of actually simplifying the code
-- Resolves a long-standing annoyance with the nano scheduler.
-- Don't just read all inputs into a list, and then provide an iterator to that list, actually make a real iterator so NanoScheduler input thread can contribute meaningfully to the work load
-- Use NanoScheduler progress function, instead of home-grown updater
-- Refactor calculation so that upfront constant values are pre-computed, and cached, and their values just looked up during application
-- Trivial comment on how we might use BAQ better in BaseRecalibrator
-- Cleaned up code in updateDataForRead so that constant values where not computed in inner loops
-- BaseRecalibrator doesn't create it's own fasta index reader, it just piggy backs on the GATK one
-- ReadCovariates <init> now uses a thread local cache for it's int[][][] keys member variable. This stops us from recreating an expensive array over and over. In order to make this really work had to update recordValues in ContextCovariate so it writes 0s over base values its skipping because of low quality base clipping. Previously the values in the ReadCovariates keys were 0 because they were never modified by ContextCovariates. Now these values are actually zero'd out explicitly by the covariates.
-- No longer computes at each update the overall read group table. Now computes this derived table only at the end of the computation, using the ByQual table as input. Reduces BQSR runtime by 1/3 in my test
The indels are still annotated as before, but now all other variant types are annotated too.
I'm doing this because of requests on the forum but am not making it standard. If we find it to be useful we can turn it on by default later.
-- Uses high-performance local writer backed by byte array that writes the entire VCF line in some write operation to the underlying output stream.
-- Fixes problems with indexing of unflushed writes while still allowing efficient block zipping
-- Same (or better) IO performance as previous implementation
-- IndexingVariantContextWriter now properly closes the underlying output stream when it's closed
-- Updated compressed VCF output file
this introduced a bug in reduce reads by de-activating it's hard clipping of the out of bounds soft-clips (specially in the MT).
DEV-322 #resolve #time 4m
This reverts commit 42acfd9d0bccfc0411944c342a5b889f5feae736.
-Switch back to the old implementation, if needed, with --use_legacy_downsampler
-LocusIteratorByStateExperimental becomes the new LocusIteratorByState, and
the original LocusIteratorByState becomes LegacyLocusIteratorByState
-Similarly, the ExperimentalReadShardBalancer becomes the new ReadShardBalancer,
with the old one renamed to LegacyReadShardBalancer
-Performance improvements: locus traversals used to be 20% slower in the new
downsampling implementation, now they are roughly the same speed.
-Tests show a very high level of concordance with UG calls from the previous
implementation, with some new calls and edge cases that still require more examination.
-With the new implementation, can now use -dcov with ReadWalkers to set a limit
on the max # of reads per alignment start position per sample. Appropriate value
for ReadWalker dcov may be in the single digits for some tools, but this too
requires more investigation.
Mark DePristo
Mauricio Carneiro
Eric Banks
Ami Levy-Moonshine
Ryan Poplin
Tad Jordan
Chris Hartl
Joel Thibault
Chris Hartl
xingwei2012
David Roazen
Yossi Farjoun