Status of SSE support in current gcc mainline

[Jan Hubicka <jh at suse dot cz>]

Hi,
Since several people has asked me, here is small update to my previous mail
concerning SSE.

Current mailine tree should be finally able to produce SSE code without any
additional patches.  The code quality should be considerably better compared to
one after my initial merger due to better caller save sequences, conditional move
support and several other updates.

The feature is enabled using -msse or -msse2 switch and you will need CVS version
of binutils to get gas accept the SSE2 instructions.

There are still problems remaining.  You will probably notice problems with the
SSE builtins, since I didn't tested them yet and the code for forcing stack
alignment is not merged yet, so unless your runtime is compiled properly to
preserve the alignment, you will get crash as soon as gcc will try to operate
with 128bit data on the stack.

Also there is still open question whether it is safe to enable SSE support
by default.

Beside that you should get considerable speedup on any float/double intensive
code on Pentium3/Pentium4 CPUs.  Please let me know how it works. I am interested
especially in testcase where performance is considerably lower.

Honza