Hi Moritz

> On 1 Aug 2026, at 21:15, Moritz Hoffmann <[email protected]> wrote:
> 
> Hi Jean,
> 
> I tried enabling IPO/LTO, and it seems to work fine on all supported 
> platforms. OpenBSD does not seem to support LTO, but all other platforms show 
> no difference in undefined behavior compared to non-LTO builds. I think we 
> should try enabling it again once the undefined behavior fixes are 
> implemented.
> 
> You can check the test runs here:
> 
>   - https://github.com/antiguru/ooRexx/actions/runs/30713617550
>   - https://github.com/antiguru/ooRexx/actions/runs/30713617554
>   - https://github.com/antiguru/ooRexx/actions/runs/30713617566


Thank you very much!

Other success: The test suite with IPO no longer crashes on WSL for Windows ARM.


> 
> I plan to merge the fixes sometime next week.

+1



rexxcps on macOS, 5 runs:

ci/platforms with IPO/LTO 26885093 26816314 26924713 26646605 26625486 REXX 
clauses per second

It's better than the previous results I communicated a few days ago:
rexxcps on macOS, 5 runs:
oorexx 5        25248593 25152888 25126680 25316819 25216759 REXX clauses per 
second
concurrency-dispatch-fixes 25058219 25080047 25053259 25104624 25087157 REXX 
clauses per second

6.0% faster than oorexx 5 without your fixes.
6.8% faster than concurrency-dispatch-fixes.
This is within the range of improvements that Enrico observed in 2020.
https://sourceforge.net/p/oorexx/mailman/oorexx-devel/thread/EF50446F-5B4F-442D-BCD2-CE2CFEDE8AC7%40mac.com/#msg37080771



This improvement is not visible in the test suite duration.
Sometimes ooRexx 5 without IPO is faster; sometimes ci/platforms with IPO is 
faster.
For the same executable, I can see variations of up to ±10 seconds from one run 
to another.
So, no reliable conclusion can be drawn.





_______________________________________________
Oorexx-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/oorexx-devel

Reply via email to