JNI. See copied text from email below.
2. This commit contains all the code used in profiling, detecting FP
exceptions, dumping intermediate results. All flagged off using ifdefs,
but it's there.
--------------Text from email
As we discussed before, it's the denormal numbers that are causing the
slowdown - the core executes some microcode uops (called FP assists)
when denormal numbers are detected for FP operations (even un-vectorized
code).
The C++ compiler by default enables flush to zero (FTZ) - when set, the
hardware simply converts denormal numbers to 0. The Java binary
(executable provided by Oracle, not the native library) seems to be
compiled without FTZ (sensible choice, they want to be conservative).
Hence, the JNI invocation sees a large slowdown. Disabling FTZ in C++
slows down the C++ sandbox performance to the JNI version (fortunately,
the reverse also holds :)).
Not sure how to show the overhead for these FP assists easily - measured
a couple of counters.
FP_ASSISTS:ANY - shows number of uops executed as part of the FP
assists. When FTZ is enabled, this is 0 (both C++ and JNI), when FTZ is
disabled this value is around 203540557 (both C++ and JNI)
IDQ:MS_UOPS_CYCLES - shows the number of cycles the decoder was issuing
uops when the microcode sequencing engine was busy. When FTZ is enabled,
this is around 1.77M cycles (both C++ and JNI), when FTZ is disabled
this value is around 4.31B cycles (both C++ and JNI). This number is
still small with respect to total cycles (~40B), but it only reflects
the cycles in the decode stage. The total overhead of the microcode
assist ops could be larger.
As suggested by Mustafa, I compared intermediate values (matrices M,X,Y)
and final output of compute_full_prob. The values produced by C++ and
Java are identical to the last bit (as long as both use FTZ or no-FTZ).
Comparing the outputs of compute_full_prob for the cases no-FTZ and FTZ,
there are differences for very small values (denormal numbers).
Examples:
Diff values 1.952970E-33 1.952967E-33
Diff values 1.135071E-32 1.135070E-32
Diff values 1.135071E-32 1.135070E-32
Diff values 1.135071E-32 1.135070E-32
For this test case (low coverage NA12878), all these values would be
recomputed using the double precision version. Enabling FTZ should be
fine.
-------------------End text from email
These changes happened in Tribble, but Joel clobbered them with his commit.
We can now change the logging priority on failures to validate the sequence dictionary to WARN.
Thanks to Tim F for indirectly pointing this out.
1. Throw a user error when the input data for a given genotype does not contain PLs.
2. Add VCF header line for --dbsnp input
3. Need to check that the UG result is not null
4. Don't error out at positions with no gVCFs (which is possible when using a dbSNP rod)
Joel is working on these failures in a separate branch. Since
maven (currently! we're working on this..) won't run the whole
test suite to completion if there's a failure early on, we need
to temporarily disable these tests in order to allow group members
to run tests on their branches again.
Added pom.xml workarounds for duplicate classpath error, due to gatk-framework dependency containing required BaseTest, and jarred *UnitTest/*IntegrationTest classes that also exist as files under target/test-classes.
Here are the git moved directories in case other files need to be moved during a merge:
git-mv private/java/src/ private/gatk-private/src/main/java/
git-mv private/R/scripts/ private/gatk-private/src/main/resources/
git-mv private/java/test/ private/gatk-private/src/test/java/
git-mv private/testdata/ private/gatk-private/src/test/resources/
git-mv private/scala/qscript/ private/queue-private/src/main/qscripts/
git-mv private/scala/src/ private/queue-private/src/main/scala/
git-mv protected/java/src/ protected/gatk-protected/src/main/java/
git-mv protected/java/test/ protected/gatk-protected/src/test/java/
git-mv public/java/src/ public/gatk-framework/src/main/java/
git-mv public/java/test/ public/gatk-framework/src/test/java/
git-mv public/testdata/ public/gatk-framework/src/test/resources/
git-mv public/scala/qscript/ public/queue-framework/src/main/qscripts/
git-mv public/scala/src/ public/queue-framework/src/main/scala/
git-mv public/scala/test/ public/queue-framework/src/test/scala/
Changes:
-------
<NON_REF> likelihood in variant sites is calculated as the maximum possible likelihood for an unseen alternative allele: for reach read is calculated as the second best likelihood amongst the reported alleles.
When –ERC gVCF, stand_conf_emit and stand_conf_call are forcefully set to 0. Also dontGenotype is set to false for consistency sake.
Integration test MD5 have been changed accordingly.
Additional fix:
--------------
Specially after adding the <NON_REF> allele, but also happened without that, QUAL values tend to go to 0 (very large integer number in log 10) due to underflow when combining GLs (GenotypingEngine.combineGLs). To fix that combineGLs has been substituted by combineGLsPrecise that uses the log-sum-exp trick.
In just a few cases this change results in genotype changes in integration tests but after double-checking using unit-test and difference between combineGLs and combineGLsPrecise in the affected integration test, the previous GT calls were either border-line cases and or due to the underflow.
Merge branch 'master' of /home/mghodrat/PairHMM/shared-repository into intel_pairhmm
Conflicts:
PairHMM_JNI/org_broadinstitute_sting_utils_pairhmm_VectorLoglessPairHMM.cc
2. Split into DebugJNILoglessPairHMM and VectorLoglessPairHMM with base
class JNILoglessPairHMM. DebugJNILoglessPairHMM can, in principle,
invoke any other child class of JNILoglessPairHMM.
3. Added more profiling code for Java parts of LoglessPairHMM
Problem:
matchToMatch transition calculation was wrong resulting in transition probabilites coming out of the Match state that added more than 1.
Reports:
https://www.pivotaltracker.com/s/projects/793457/stories/62471780https://www.pivotaltracker.com/s/projects/793457/stories/61082450
Changes:
The transition matrix update code has been moved to a common place in PairHMMModel to dry out its multiple copies.
MatchToMatch transtion calculation has been fixed and implemented in PairHMMModel.
Affected integration test md5 have been updated, there were no differences in GT fields and example differences always implied
small changes in likelihoods that is what is expected.
Karthik Gururaj
Ryan Poplin
Eric Banks
droazen
Joel Thibault
Khalid Shakir
Valentin Ruano-Rubio
mghodrat