Re: Pentium memcpy patch. Unintended effect?

David Hinds (dhinds@hyper.stanford.edu)
8 Feb 1997 17:21:06 GMT


Chad Page (cpage@comp3244.resnet.sjsu.edu) wrote:

: On Fri, 7 Feb 1997 aidas@ixsrs4.ix.netcom.com wrote:

: >
: > I patched and compiled 2.1.25 with the Pentium memcpy patch. Everything
: > works fine, except for Quake. Now, Quake runs a fair bit choppier.
: >
: > I'm not really concerned about Quake. I'm more concerned whether this is
: > some kind of defect in my hardware, or simply a side effect of the design
: > of Quake. The latter seems more likely.

: Quake *heavily* uses the FPU, and from what Ted just said it would
: sound like the pentium memcpy patch would affect Quake since the floating
: point registers would be swapped every time pentium memcpy is used. This
: is also similar to why Carmack won't use MMX instructions in quake :)

Shouldn't the threshold for when the FPU registers will be used for
mempcy() be tuneable, then, so that on a system that runs FP-intensive
code, the fancy memcpy will only be done when it is faster *including
the extra context switch time*?

-- Dave Hinds