[aarch64-port-dev ] Errors in aarch64 C1 caught by the Mauve test suite

[Andy Johnson]

This is the way I set up the Mauve test runs:
1. Create a test directory on the build system (I use Linux-x86-64)
2. cd <test-directory>
3. export CVSROOT=:pserver:anoncvs at sourceware.org:/cvs/mauve
4. cvs login {password anoncvs}
5. cvs -z 9 co mauve
6. cd mauve
7. ./configure
8. make
    Note: if "make" fails due to an autoconfig version error, change the
version in the Makefile to match the version installed on your system.  I
changed 1.10 to 1.11.
9. java Harness -vm <path to JVM under test> -showpasses java.lang.Integer
    This should report 0 failures out of 31 tests run.
10. xvfb-run java Harness -vm <path to JVM under test> java
11. xvfb-run java Harness -vm <path to JVM under test> javax
     These last two steps create the class files for the java and javax
tests, so that they don't have to be compiled on the aarch64 target.  I do
this because I am running on
     the Foundation_v8 emulator.  Some errors will be reported.  You may
wish to record these failures, so that you can ignore the same failures on
aarch64.
12. Create a tar ball of the mauve directory.
13. Copy the tar ball to the aarch64 image.
      I do this by mounting the linaro-image-lamp-genericarmv8.ext3 image
as /media/usb and then copying the tar ball to /media/usb/home/root.
14. Launch the Foundation model running openembedded and linaro-image-lamp.
15. Unwind the mauve tarball.
16. cd mauve
17. Edit gnu/testlet/config.java to change the path to mauve. I change it
to /home/root/mauve.
18. make {This re-builds the test harness so that the newly edited path is
used}
19. java Harness -vm "<path to JVM under test> -Xcomp" -showpasses
java.lang.Integer
     This verifies that the test harness works on aarch64.
20. java Harness -vm "<path to JVM under test> -Xcomp" <any of the reported
test failures> (e.g. java.util.LinkedList.AcuniaLinkedListTest)




On 28 July 2013 08:46, Andy Johnson <andy.johnson at linaro.org> wrote:

> I just use CVS to download the Mauve test suite.  The site contains a link
> for downloading a nightly build, but it's empty, so CVS is your only
> option. Then I configure and build it on my build system.  Then I do a
> complete run using:
> java Harness -vm "<path to java command> -Xcomp"
> You can also use:
> java Harness -vm "<path to java command> -Xcomp" java.util (or
> java.util.<specific test, as noted in the report>
> if you only want to try out the "util" subsection, where most of the
> SIGSEGV errors are reported.
>
> I then create a tarball of the mauve suite and copy it over to my
> openembedded image, re-configure and re-build the Harness (since it creates
> absolute paths to the java/class files to be tested) and repeat the process
> above.  I create the tarball from the x86-64 runs so that the class files
> already exist (they are build on-demand the first time a test is run),
> since running javac on openembedded takes forever.  If I understand your
> most recent comments correctly, it looks like you may have access to real
> aarch64 hardware.  Lucky you.
>
> If it would be easier for you, I can simply email you the tarball that I
> use.  It GPL, so there's no issue with sharing it publicly.
>
> -AndyJ
>
>
>
> On 28 July 2013 04:23, Andrew Haley <aph at redhat.com> wrote:
>
>> On 07/26/2013 08:35 PM, Andy Johnson wrote:
>> > I have been running the Mauve test suite on openembedded using the ARM
>> > Foundation_v8 model and the aarch64 port of OpenJDK8.  With a few
>> > exceptions, the results from running the template interpreter (-Xint)
>> match
>> > those from the baseline xi6-64 OpenJDK8 port.  However, I am getting a
>> > number of failures using -Xcomp (C! compiler) that I don't get using
>> > -Xint.  I am excluding test failures that are generic to Mauve (i.e.
>> they
>> > fail on all OpenJDK8 ports, including x86-64).
>>
>> Great, thanks for that.
>>
>> Can you please give me detailed instructions about how to reproduce one
>> or two of these?
>>
>> Thanks,
>> Andrew.
>>
>>
>