[FFmpeg-devel] [libav-devel] [PATCH] x86inc: Avoid using eax/rax for storing the stack pointer

[Henrik Gramner]

On Mon, Dec 26, 2016 at 2:32 AM, Ronald S. Bultje <rsbultje at gmail.com> wrote:
> I know I'm terribly nitpicking here for the limited scope of the comment,
> but this only matters for functions that have a return value. Do you think
> it makes sense to allow functions to opt out of this requirement if they
> explicitly state to not have a return value?

An opt-out would only be relevant on 64-bit Windows when the following
criteria are true for a function:

* Reserves exactly 6 registers
* Reserves stack space with the original stack pointer stored in a
register (as opposed to the stack)
* Requires >16 byte stack alignment (e.g. spilling ymm registers to the stack)
* Does not have a return value

If and only if all of those are true this would result in one register
being unnecessarily saved (the cost of which would likely be hidden by
OoE). On other systems than WIN64 or if any of the conditions above is
false an opt-out doesn't make any sense.

Considering how rare that corner case is in combination with how
fairly insignificant the downside is I'm not sure it makes that much
sense to complicate the x86inc API further with an opt-out just for
that specific scenario.

[I accidentally only sent this reply to libav-devel, so resending it
to ffmpeg-devel as well]