[fpc-pascal] Crypt++ 5.5.2 Tiger::Transform converted toDCPCrypt2.Compress, AV: Compiler Optimization Bug with SSE2 ? Or just programming bug ?HELP NEEDED :) ?!

[Skybuck Flying]

> FPC needs to preserve ebx, esi and edi.

So it's  push ebx + push esi + push edi vs pushad. And same for pop.

According to my AMD optimization manual.

pushad requires 6 latency.
push register requires 3 latency.

So pushad should be faster.

Bye,
  Skybuck.