rth at twiddle.net
Tue Feb 26 20:14:33 CET 2013
On 02/26/2013 10:41 AM, Torbjorn Granlund wrote:
> I'm not sure quite what's going on with the 3/4 issue rates. I really would
> have expected to see either exactly 1, or very nearly 1/2, especially for vadd.
> I think you mean 4/3. But also that is an underestimate. with 8-way
> unrolling I get a bit more, about 7/5.
I guess I should have written the units used: cyc/insn.
Which is why 5 makes sense in my table for chained multiplies, as opposed to
some bizzaro-land where the latency of a multiply is 1/5 or 2 tenths of a
cycle. I really did mean that I measured 4 adds issued per 3 cycles.
But you're also right that increased unrolling measures slightly better,
issuing 12 insns in 8.5 cycles, or < 1% difference from your 5/7.
More information about the gmp-devel