at http://www.digitalmischief.co.uk/fruitbowl/
is a gmp with a new mul/addmul and basecase running at 2.5c/l
On my machine a 1800Mhz sempron , I get a benchmark score of 8209 which when
scaled to a 2600Mhz cpu (for example) would give a score of 11860.
The new mul_basecase is actually slower than the old one until you hit about 8
limbs, so obviously some more work needs to be done for the smaller sizes.
I rewrote mpn_mul_1 to use a varient which did not need pipelining , and
compairing this to mpn_addmul_1 which does use pipelining , at small sizes we
get
./speed -r -c -s 1-40 mpn_mul_1.23 mpn_addmul_1.23
overhead 6.04 cycles, precision 10000 units of 5.53e-10 secs, CPU freq 1808.24
MHz
mpn_mul_1.23 mpn_addmul_1.23
1 15.11 #0.5334
2 #17.12 1.3529
3 #18.14 1.2774
4 #22.16 1.1364
5 #25.19 1.1999
6 #26.20 1.3074
7 #30.22 1.3001
8 #33.25 1.1826
9 #35.27 1.2290
10 #37.28 1.2428
11 #41.31 1.1464
12 #44.33 1.0916
13 #45.34 1.1780
14 #48.36 1.1663
15 #51.38 1.1177
16 #55.40 1.0555
17 #55.42 1.1456
18 #59.44 1.1184
19 #61.46 1.0992
20 #66.47 1.0312
21 #65.49 1.1233
22 #70.52 1.0856
23 #71.53 1.0851
24 #77.54 1.0139
25 #75.57 1.1071
26 #81.59 1.0616
27 #81.61 1.0746
28 88.72 #0.9995
29 #85.79 1.0926
30 #92.76 1.0437
31 #91.81 1.0664
32 99.78 #0.9911
33 #95.90 1.0833
34 #103.92 1.0292
35 #101.90 1.0598
36 124.72 #0.8738
37 119.83 #0.9510
38 #127.80 1.0157
39 #125.87 1.0395
40 #131.88 1.0148
So perhaps to get better speed for small sizes , we may have to use
non-pipelined version , anyway there are still more overheads that can be
optimized first.
The new mul_basecase is also a "cut and paste" of mul_1/addmul_1 so I wouldn't
bother to look at it. I'll make a neat version when I get faster version of
mul/addmul
--~--~---------~--~----~------------~-------~--~----~
You received this message because you are subscribed to the Google Groups
"mpir-devel" group.
To post to this group, send email to mpir-devel@googlegroups.com
To unsubscribe from this group, send email to [EMAIL PROTECTED]
For more options, visit this group at
http://groups.google.com/group/mpir-devel?hl=en
-~----------~----~----~----~------~----~------~--~---
