On Dec 12, 2005, at 21:55, Joshua Isom wrote:
Because the random benchmark is used for a few other tests, I figured
I'd do it next. To do it the same way in pir is slow because it
requires putting a variable into a global register and retrieving it
900,000 times. I have three versions written up. One that's split
over two files, with one file including main, and the other including
an init section and the actual needed sub. Another that's the same
but just one file, but a tad uglier to me(7 extra lines just for error
checking). And the final cream of the crop(i.e. about twice as fast
as the pir), a pasm version, which takes advantage of the fact that
the registers stay the same the whole time unless modified.
Hmm. I think we just submit 2 versions: random.pasm and the the split
one with randomd_lib.pir.
The global is ugly, yes. We could use a closure, but that would
restrict the usage to a fixed :outer, which isn't really good. Well, we
dont't have a C-like static construct.
But before doing that I'd appreciate, if you cleanup the rand algorithm
a bit: IM, IA, IC, and last are integers only. The part of the
calculation doesn't involve floats. E.g.:
# random-2.pasm
# I0 = N (counter)
# I1 = last
# N3 = the argument for gen_random
# N2 = random result
main:
get_params "(0)", P0
elements I0, P0
eq I0, 2, arg
set I0, 900000
branch arg_ok
arg:
set S0, P0[1]
set I0, S0
arg_ok:
set I1, 42
set N3, 100.0
while_1:
bsr gen_random
dec I0
if I0, while_1
new P0, .FixedFloatArray
set P0, 1
set P0[0], N2
sprintf S0, "%.9f\n", P0
print S0
end
.constant IM 139968
.constant IA 3877
.constant IC 29573
gen_random:
mul I2, I1, .IA
add I2, .IC
mod I1, I2, .IM
set N2, I1
mul N2, N3
div N2, .IM
ret
