[hashcash] Speed problem with 1.03 on Mac G4
- From: hal@xxxxxxxxxx ("Hal Finney")
- To: hashcash@xxxxxxxxxxxxx
- Date: Sun, 15 Aug 2004 09:42:16 -0700 (PDT)
I tried the latest version (1.03) on an iBook with a 1 GHz G4.
First, I had never heard the term "altivec" before. I knew enough to
disable the -mmmx and such in the Makefile, but it would be nice to
have some comments there, or in the README, to hint about what people on
other architectures should do. I did "man gcc" and searched for altivec,
and discovered the -faltivec flag. So I put that in place of the -mmmx
and other Intel stuff.
The program still did not build, apparently due to a compiler bug.
The error was:
cc -O -O3 -funroll-loops -faltivec -DREGEXP_POSIX -DCHROMATIX -c -o
fastmint_altivec_compact_2.o fastmint_altivec_compact_2.c
/var/tmp//ccsrKO5q.s:758:FATAL:Symbol myword12 already defined.
I took out the -O3 and left the -O and then it compiled OK.
cc --version prints:
cc (GCC) 3.3 20030304 (Apple Computer, Inc. build 1495)
fastmint_benchtest produced the following edited speed report:
Rate Name (* machine default)
1092927 ANSI Compact 1-pipe
1003708 ANSI Standard 1-pipe
933830 ANSI Ultra-Compact 1-pipe
1239875 ANSI Compact 2-pipe
905185 ANSI Standard 2-pipe
4098477 PowerPC Altivec Standard 1x4-pipe
4215577 PowerPC Altivec Compact 2x4-pipe
4339564 PowerPC Altivec Standard 2x4-pipe *
Nice and fast. Now, the problem is that hashcash -m takes 8-10
seconds before printing the estimated time. And then after that
it takes about another 5-7 seconds longer than estimated. Here is
a typical run:
$ time ./hashcash -m -b 20 -r test
time estimate: 0 seconds (242 milli-seconds)
hashcash token: 1:20:040815:test::3Y5ZCsKkkYUATBfw:000032Rr
real 0m17.015s
user 0m16.620s
sys 0m0.200s
I did this four times and the time taken varied from 16.9 to 17.5 seconds.
I did some 23-bit trials and it estimated 2 seconds but took from 17 to 22
seconds.
It appears that there is about a 17 second overhead for, presumably,
choosing the best architecture. That's OK if you're generating a big
stamp, but it is probably not acceptable for applications which rely on
small stamps.
Hal Finney
Other related posts:
