On Fri, 13 Feb 2026 12:55:47 GMT, Coleen Phillimore <[email protected]> wrote:

>> This commits does only what you say. 
>> https://github.com/openjdk/valhalla/pull/2068/changes/59c1fc2e7dedd079a61dced4401267dace7c642d
>>  However this still have to touch every line as each column is not wide 
>> enough to fit every entry, so when we now shift some by one and some by two 
>> alignment of all is required. 
>> 
>> @stefank thought, and I agree that if we have to touch every line anyways, 
>> let us just avoid this 4x64 grid and have one bytecode dispatch label per 
>> line, to not have to touch 64 lines in the future when making changes to one 
>> bytecode. (Someone might want to implement the fast bytecodes for zero in 
>> the future.)
>> 
>> I did push them as two separate commits so I can revert 
>> https://github.com/openjdk/valhalla/pull/2068/changes/342cd358eec59e1798ef62d03bbbd3a0e0f79d9d
>>  if you have really strong opinions on this. However I do think that the 
>> 4x64 is worse than the 1x256 solution.
>
> I actually look at this code periodically and don't really like scrolling 
> down 256 lines to get to interesting parts.  This code doesn't change so the 
> merge with valhalla would be zero if it was 4 bytecodes per line.  See if 
> @shipilev  has a differing opinion.  Sorry for only commenting on the style 
> nit but I found this jarring.

Actually I wonder if there's a performance benefit to this case label 
implementation anymore.  It was added when there was once a hope of replacing 
the template interpreter with this one.

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

PR Review Comment: 
https://git.openjdk.org/valhalla/pull/2068#discussion_r2804097218

Reply via email to