https://gcc.gnu.org/bugzilla/show_bug.cgi?id=69041
Andrew Pinski <pinskia at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Severity|normal |enhancement
Summary|Unnecessary push/pop of |Unnecessary push/pop of
|caller-save register (ecx) |caller-save register (ecx)
|on 32bit with vector |on 32bit with vector
|intrinsics. Sometimes |intrinsics.
|without the pop, clobbering |
|ebp (callee-save) |
Keywords|wrong-code |
--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
>clobbers the caller's ebp with the pushed value of ecx, but the esp=ebp part
>of leave cleans up after the mismatched push/pop
This is wrong. Leave does the pop.
Basically leave is `sp = ebp + 4; ebp = [ebp];`
So no wrong code. Just missed optimizations.
The removal of the stack frame happened in GCC 6 for dummy/dummy2.
The removal of vzeroupper for dummy/dummy2 happened in GCC 9.
Looks like there is still a frame pointer creation happening for add_pixdiff; I
have not looked into why though
