[ros-dev] RE: [ion] 19742: -
ImplementRtlPrefectMemoryNonTemporal.Patch by Patrick
Baggett<email@example.com> and myself.
ionucu at videotron.ca
Wed Nov 30 04:16:09 CET 2005
Thomas Weidenmueller wrote:
>Alex Ionescu wrote:
>>(To be honest, my original idea was to
>>use "nop" and overwrite it with "ret", but both Patrick (who wrote the
>>patch) and MS were right in making it a "ret" by default and using "nop"
>>to overwrite, since that's even faster).
>I don't understand how a modified nop to a ret can be faster than a ret
>generated at compile time, or vice versa. However, having a ret by
>default makes more sense ;)
If you have a "nop" by default, then your job is as follows:
- Detect support for the instruction
- If it's NOT supported, overwrite with ret.
Additionnally, the default state of the function is to WORK, which
fundamentally speaking is wrong because you're assuming that the
advanced functionality exists, and then you're disabling it. Since
you're trying to be safe, then the whole point is to protect from the
However, with a "ret" by default, you only overwrite with nop if the
instruction IS supported. Apart from the design issue itself, I think
the premise here is that more CPUs don't have the support for the
function, then the ones that do.
More information about the Ros-dev