Closed tueda closed 3 months ago
In the change of 1d4b775, why was the code introduced in 0b8f3d7 (with another line) commented out? Why the number of memcpy
was increased from 2 to 3? Was it just a merge failure with 652c715?
@vermaseren Do you remember this?
I would bet this is the right fix. (At least "phenomenologically" it works.)
As written in https://github.com/vermaseren/form/issues/7#issuecomment-1677266031, this reverts a part of 1d4b775, and somehow fixes the optimization regression.
Actually, I understand neither why it fixes the regression nor what was the intention of the change.