RFR: 8254317: C2: Resource consumption of ConvI2LNode::Ideal() grows exponentially [v2]

[Roberto Castañeda Lozano]

On Thu, 22 Oct 2020 08:43:09 GMT, Roberto Castañeda Lozano <rcastanedalo at openjdk.org> wrote:

>> test/hotspot/jtreg/compiler/conversions/TestMoveConvI2LThroughAddIs.java line 41:
>> 
>>> 39:  *      -XX:CompileOnly=::testChain,::testTree,::testDAG
>>> 40:  *      compiler.conversions.TestMoveConvI2LThroughAddIs basic
>>> 41:  * @run main/othervm/timeout=1 -Xcomp -XX:-TieredCompilation -XX:-Inline
>> 
>> By default timeout is 120 sec. Why you set it to 1 sec?
>> The same for second @run.
>
> These two runs (`stress1` and `stress2`) test that compilation time does not explode. When it does (before the proposed fix), the runs take > 10s to execute (but less than 120s, at least `stress1`), while after the fix both take just a few ms (all measurements from my local and our CI machines). I just deemed 1s to be a good value to balance the risk of false positives (reporting an nonexistent bug because the timeout is too short) and false negatives (not reporting a real bug because the timeout is too long). I will explore further increasing the load of these runs to see if we can increase the timeout without increasing the risk of false negatives.

I have addressed this in the latest update.

-------------

PR: https://git.openjdk.java.net/jdk/pull/727