Re: Fastcode versus Highlander

Hi Dennis and John,

The problem with all of the 64-bit RTL functions I have proposed is that

You and me and Borland handled very well with Int64Div. Why should we not be
able to handle it one more time for the same function and also for Int64Mul?

I have no problems with the remaining proposals, because Fastcode has done
zero work on any of these.

Int64Mul is progressing "Fastcode way", new proposed function
Int64Mul_JOH_IA32_4 is in B&V 1.3.1. This new function is waiting
for four benchmark runs, AMD X2 and Presler runs show that it's very
fast and it passed validation on both.

If John prepares Int64Div version of new __lldiv, I can add that
to Int64Div B&V and that one starts going "Fastcode way" too, ok?


The Fastcode Project: