lrwiman <[EMAIL PROTECTED]> writes:
> It has been mentioned several times recently that factoring is all integer
> work, and LL testing is nearly all floating point.
>
> It is my understanding that on intel CPU's, these are done on separate parts of
> the CPU. Would it increase net performance to do factoring and LL assignments
> at the same time?
There is _some_ scope here, but I think we need to be _very_
careful. Actually, the Intel CPU is a bit of a nightmare from this
point of view.
a) The integer multiplier and the floating-point multiplier share
common circuitry, therefore a (I)MUL and a FMUL cannot execute
in parallel. Since division is (only very partially) pipelined & takes
"forever", we avoid using it where practical. Therefore, the time-
dominant code in (integer) factorization is actually multiplication.
b) Some CPU types - including the Intel P6 family, for which the
potential gain would be greatest - already use the FPU (at least to
some extent) to do factorization - because can be faster to use the
FPU to do integer multiplies than it is to do them in the integer
ALU, despite the conversions - doing this reduces the demand on
the (very limited number of) (named) integer registers, too.
c) If you add extra integer mode instructions into the opcode
stream, you'd have to be very careful that these do not cause
opcode fetch, decoder or data bottlenecks, even when they don't
depend on shared execution units. We really _don't_ want to slow
down the LL code _at all_!
Looking at George's macros yesterday, I was struck by how
efficient they are in terms of usage of FPU registers & L1 data
cache lines. Really you couldn't squeeze anything else in there,
unless it would run entirely in (integer) CPU registers, or you were
prepared to take the "hit" involved in L1 data cache line misses
quite frequently.
> Also, I'm going to quit first time LL testing. Call me impatient,
but I don't
> want to wait until early July for my exponent to finish, thus I'm going to
> switch to double-checking.
Hi, Mr. Impatient!
Seriously, though, I wonder how many others will do this now that
it looks like the chance of grabbing the $50K has gone?
Also, how much latent interest might there be in firing straight at
the $100K prize, even with run lengths > 1 year?
Regards
Brian Beesley
________________________________________________________________
Unsubscribe & list info -- http://www.scruz.net/~luke/signup.htm
