[ardour-dev] AMD64 SSE optimisation

[John Rigg]

On Wed, Dec 28, 2005 at 08:36:28PM +0000, John Rigg wrote:
> Now I'm finding that my original patch
> was indeed an over-simplification. On single stepping through the code
> with gdb to diagnose some strange behaviour, I've found that x86_64 uses
> different registers for some variables compared to i386, so there's a bit
> more work to do yet.

I've worked out what's happening. The variables get optimised out,
even with only -O (this is with gcc-4.0), and are stored in registers.

For example, in x86_compute_peak in the sandbox tester:
%rdi	&buf[0] (also stored at 32(%rbp))
%rsi	nframes
%rbx	ITERATIONS
%xmm0	peak

This should make the code simpler and faster. I've started reworking it
and hope to have some results in the next week or so.

John