has anyone in the last few days even checked if the MMX instructions do
generate
the expected (and needed) illegal instruction trap for examination on
non-MMX
capable processors. If I recall correctly the docs that came with the
Intel
optimizing compiler say something to the contrary (either the IP being
wrong on
the trap or not generating a trap at all). If this piece of
documentation
is correct the whole issue could be dropped. If we ever need to do MMX
I would like to see an extension to the binary format much the same as
on the Power-MAC which had so called fat binaries that held code for
680xx and power-PC
so the OS could pick the native code for the current machine. In our
case the
Program would contain both functions for both cases and the dynamic
linker could
decide which would be used. On both types of machines the result would
be optimal,
with no necessary emulation, which - if at all possible - would be far
from useable
on non-MMX machines.
regards
Karl Günter Wünsch
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu