Re-enable CFGSimplifier #16459
Re-enable CFGSimplifier #16459
Conversation
|
FYI @gita-omr |
Signed-off-by: Akira Saitoh <saiaki@jp.ibm.com>
Akira1Saitoh
force-pushed
the
enableCFG
branch
from
88401f3
to
b9d5fd5
Compare
|
Jenkins test sanity,extended all jdk8,jdk11,jdk19 |
|
Build_JDK19_x86-64_mac_Personal and Test_openjdk11_j9_extended.functional_x86-64_windows_Personal seem to be env issues... |
|
In Test_openjdk19_j9_sanity.functional_x86-64_windows_Personal, cmdLineTester_jvmtitests_hcr_3 failed due to a timeout in one test: rc021. |
|
10x Grinder runs of |
|
|
|
Tried 15x Grinder runs of cmdLineTester_jvmtitests_hcr_3. All passed. Re-run of failed external Jenkins job also passed. |
|
Tried sanity.functional on internal Grinder. I was not able to produce the problem so far. There were some infra errors, though.
I also asked @knn-k to launch re-runs of failed external Jenkins job several times. The problem did not occur.
|
Re-enable CFGSimplifier which has been disabled by [1].
The motivation for re-enabling it is that we want to use it for removing boundary check code from the loop when using Vector JEP.
When vectorizing library calls of
jdk.incubator.vectorpackage, there are many calls toPreconditions.checkIndex()[2] left in the loop. It checks if index is in the range and throws an OOB exception when the check fails. We want to take this boundary check code out of the loop.CFGSimplifier should be able to transform the boundary check pattern into a BNDCHK node [3] and we can let loopVersioner remove the BNDCHK from the loop.
The original issue with CFGSimplifier [4] seems to be already fixed by [5]. I am not able to reproduce it with multiple runs of system test using this commit. It seems that the most recent problem with CFGSimplifier [6] was seen in sanity.functional but I am not able to reproduce it either.
[1] #9783
[2] https://github.com/ibmruntimes/openj9-openjdk-jdk19/blob/d78ba9640fed4fc0b5cda490a2ac8c2f21622059/src/java.base/share/classes/jdk/internal/util/Preconditions.java#L299
[3] eclipse/omr#5085
[4] #9754
[5] #10224
[6] #10961 (comment)
Signed-off-by: Akira Saitoh saiaki@jp.ibm.com