On 9/17/19 6:50 PM, Richard Sandiford wrote: [ ... ]
This patch tries to avoid the problem by not adding register shuffle copies if there appears to be no chance that the two operands could be allocated to the same register.
The table below summarises the tests that had more or fewer assembly lines after the patch (always a bad metric, but it's just to get a flavour): Target Tests Delta Best Worst Median ====== ===== ===== ==== ===== ======
x86_64-linux-gnu 39 -577 -164 23 -1
Hmmm, this sounds certainly interesting enough to try on its own merits, even if it's not committed by tomorrow morning ...
Fascinating analysis - thanks ! -- Toon Moene - e-mail: t...@moene.org - phone: +31 346 214290 Saturnushof 14, 3738 XG Maartensdijk, The Netherlands At home: http://moene.org/~toon/; weather: http://moene.org/~hirlam/ Progress of GNU Fortran: http://gcc.gnu.org/wiki/GFortran#news