>memcpy appears to be an inline...
>I have a installable driver (ntfs) which wants the function
>memcpy...and we don't have it...
>The non-inline version is very simple...
>Would it be reasonable to have a non-inline version that's highly
>optimized assembler?
The inline version *is* highly optimized assembler (although the way
the trailing bytes are handled could be tested for optimalness). I
put some instrumentation in the kernel a while back to test alignment
of src and dest in memcpy() invocations, and in almost all cases they
were 4-byte aligned, which the inline version handles very well.
I guess my point is, a non-inline version should look just like the
inline version.
Tom.