Florian Klaempfl pisze:
DarekM schrieb:
Hi
I've notice two improvement
1. equation in SSE
# Var Zr located in register mreg0md
# Var Zi located in register mreg1md
# Var Ti located in register mreg3md
# Var Tr located in register mreg2md
# [39] Tr := Zr * Zr;
movsd %xmm0,%xmm4
mulsd %xmm0,%xmm4
movsd %xmm4,%xmm2
it can be changed to:
# [39] Tr := Zr * Zr;
movsd %xmm0,%xmm2
mulsd %xmm0,%xmm2
Yes, but it's not easy to implement ;)
I think so, but its worth of effort. That are very popular equation
(the same with A=B+C) . And second: can cpu then make own
optimalizations (next line use different registers).
my stupid suggest:
after translate of equation test of last line:
movsd %xmmA,%xmmB
if %xmmB is not use before change every %xmmA to %xmmB and remove last line.
but sorry it I don't understand some of it.
Darek
_______________________________________________
fpc-devel maillist - fpc-devel@lists.freepascal.org
http://lists.freepascal.org/mailman/listinfo/fpc-devel
