[jdk8u-dev] RFR: 8261235: C1 compilation fails with assert(res->vreg_number() == index) failed: conversion check [v2]

[Chuanqi Zang]

On Mon, 19 May 2025 03:11:01 GMT, Chuanqi Zang <duke at openjdk.org> wrote:

>> Why is this an important fix to have in OpenJDK 8u? Why does the risk of adding it at this stage of the lifecycle of 8u not outweigh the benefits?
>
>> Why is this an important fix to have in OpenJDK 8u? Why does the risk of adding it at this stage of the lifecycle of 8u not outweigh the benefits?
> 
> Thanks for your reply @jerboaa . 
> 
> 1.  As I mentioned in "C1 compilation failed in HugeMethods", I explained the background of our specific case. In our Flink datastream program, we discovered that numerous hot methods generated by jsoniter or fastjson were being interpreted. The bytecode of these methods was incorrectly generated and exceeded the compilation threshold. When we added the -XX:-DontCompileHugeMethods option to the command line, the JDK 8u consistently crashed. To reproduce the identical stack backtrace, we built the fastdebug version, and with this pull request (pr), the crash disappear. This is why we believe this issue is of significant importance to OpenJDK 8u.​
> 
> 2.  this pr only to check  vreg_number in linear scan stage to bailout the compile task.  and there is a follow fix  https://bugs.openjdk.org/browse/JDK-8287396 . because 8u and 17u with different c++ standards, the follow pr build failed in OpenJDK 8u.

> @zangcq Please enable GHA on your fork.

done

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

PR Comment: https://git.openjdk.org/jdk8u-dev/pull/622#issuecomment-2889772559