File Pro And Bus Errors

[Walter Vaughan]

Jim Asman wrote:
> At 05:02P Fri Jul 08 2005, Walter Vaughan wrote:
>>the rest of the OS compiled with GCC 3.X or if they have even tested ICC. 
> Now, that's the "talk" I was (ph|f)ishing for! :-)

Can't let this thread die can I.

One thing that seems to me is quite evident is that *any* benchmark I have seen 
beyond being useless, now they are doubly useless.

I don't know how many of you are following the AMD lawsuit and European 
investigation regarding the ICC compilier, but the short version is that the 
compiler makes code that detects an AMD CPU, and if so purposefully inserts code 
that slows processing down by a significant amount even if the chip has all the 
needed micro-code to process identical to Intel's CPU. I've read that the 
"penalty" that Intel imposes is about 20%.

It is possible to force in the "C" code a compiler hack does not detect anything 
at runtime, but it's not well known. I also believe sometime last year someone 
figured out a way to strip the test from the binary file, but probably no 
commercial software on the planet uses either of those methods.

Of course the solution is GCC, but it does not generate binaries that execute in 
the X86 environment as fast as ICC.

So what would be interesting would be to see if KB has a Athlon machine and 
compiled filePro in a non-broken-AMD-works-for-both-same binary mode to see if 
there is any speedups available to filePro beyond disk subsystem limitations.