On Wed, Nov 06, 2019 at 12:12:54PM +0100, Jan Kratochvil wrote:
On Wed, 06 Nov 2019 11:49:18 +0100, Vít Ondruch wrote:
> > we can achieve a
> > performance gain of 5% to 27% depending on the workload.
>
> Where are these number coming from? And what is the reason for the
> performance hit for dynamically linked Python?
Yes, it looks suspicious. -fPIC was a performance hit for i686 but it no
longer affects x86_64.
No. It is a smaller performance hit on x86_64, no need to compute the PIC
base register at the start of functions and waste a register for it,
but still significant (mainly that indirect addressing is used for anything
not know to bind to variables in the current TU, while otherwise direct
addressing could be used).
Jakub