Why does the PPC kernel architecture not use the builtin

[Robert Kavaler]

functions such as memcpy?  Are these broken?  I notice that the user
level code uses the builtin, but the kernel specifies -fno-builtin.
This seems like a big hit to efficiency.

In the same vein, the code (.h) to do byte swapping has a comment that
the ppc byteswap instruction should not be used, so instead a less
effecient code sequence is used.  This is supposed to be a problem with
a specific rev. of the compiler.  Does anyone know if this has been
fixed in the newest gcc or egcs (soon to be merged) compilers?

Thank you,

Robert Kavaler


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/