Hi all,
I have made some more minor improvements to division on penryn.
Here was the dismal state of affairs according to mpir_bench_two for m x n
divisions:
MPIR 2.6.0:
m bits x n bits
8192 32 => 926962
8192 64 => 803137
8192 128 => 534988
8192 4096 => 182048
131072 65536 => 1832
8388608 4194304 => 7.92
8192 8064 => 7061119
16777216 262144 => 2.63
MPIR new: (higher numbers are better)
m bits x n bits
8192 32 => 926962
8192 64 => 803137
8192 128 => 534988
8192 4096 => 189139
131072 65536 => 1992
8388608 4194304 => 8.18
8192 8064 => 8090867
16777216 262144 => 5.50
For comparison, here is GMP 5.1.3 (higher numbers are better):
m bits x n bits
8192 32 => 835918
8192 64 => 845628
8192 128 => 588428
8192 4096 => 194702
131072 65536 => 1863
8388608 4194304 => 9.90
8192 8064 => 7614933
16777216 262144 => 5.83
Unfortunately that 8192 x 8064 division is slower than GMP on K10. I have
no idea why. When I time the relevant mpn_sb_divappr_q, MPIR is quite a bit
faster. I'm putting it down to compiler issues because I've ruled out just
about everything else, with the exception of perhaps a slowdown in our
mpn_copy or mpn_lshift or something basic like that.
Bill.
--
You received this message because you are subscribed to the Google Groups
"mpir-devel" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to mpir-devel+unsubscr...@googlegroups.com.
To post to this group, send email to mpir-devel@googlegroups.com.
Visit this group at http://groups.google.com/group/mpir-devel.
For more options, visit https://groups.google.com/groups/opt_out.
