On 2020-08-23 21:19, Kevin Li wrote:
Alternatively, I could define another function in the C file:
void f_list(double *x, long len) {
for (long i = 0; i < len; i++) x[i] = f[i];
}
and expose this in the *.so. Then I could define another verb
fs =: 'path.to.library.so f_list > n *d x&cd
and call that instead. I find that fs is usually 2-3 times faster than
f,
presumably because of the overhead in calling into the shared library.
However, fs has two disadvantages:
1. It modifies the array in place, which makes my code less functional
Because fs passes a *d rather than a &d, f_list is modifying a copy in
place, and since it doesn't return anything the work is lost.
a =: i.5
fs a;#a
0
a
0 1 2 3 4
J arrays have value semantics, not reference semantics.
This a bit awkward, especially n=.
2. It makes it harder to work with atoms (I have to convert an atom
into a
1 dimensional array first).
fs2i =: './f_list.so f_list2 > x &d x'&cd
fs2 =: 3 : 0"1
n =. ,:`] @.(*@:#@:$) y
a =. fs2i n;#n
r =. memr a,0,(#n),8
memf a
r
)
Where f_list2 returns a malloc'd pointer as an integer:
extern(C) double f(double x) {
return x * 1.5;
}
extern(C) size_t f_list2(double* x, int len) {
import core.stdc.stdlib: malloc;
double* ret = cast(double*)malloc(double.sizeof * len);
foreach (i; 0 .. len) ret[i] = f(x[i]);
return cast(size_t)ret;
}
I also suspect there's a lot of unnecessary copying.
But usage is OK:
fs2 5
7.5
fs2 i.5
0 1.5 3 4.5 6
fs2 i.5 5
0 1.5 3 4.5 6
7.5 9 10.5 12 13.5
15 16.5 18 19.5 21
22.5 24 25.5 27 28.5
30 31.5 33 34.5 36
Alternately:
fs3 =: 3 : 0
n =. ,:y
a =. fs2i n;*/$n
r =. ($n) $ memr a,0,(*/$n),8
memf a
r
)
This has one FFI call regardless of y rank, as shown by
adding some noise to the library's functions:
fs2 5
f_list2 called
7.5
fs2 i.5
f_list2 called
0 1.5 3 4.5 6
fs2 i.5 5
f_list2 called
f_list2 called
f_list2 called
f_list2 called
f_list2 called
0 1.5 3 4.5 6
7.5 9 10.5 12 13.5
15 16.5 18 19.5 21
22.5 24 25.5 27 28.5
30 31.5 33 34.5 36
vs.
fs3 5
f_list2 called
7.5
fs3 i.5
f_list2 called
0 1.5 3 4.5 6
fs3 i.5 5
f_list2 called
0 1.5 3 4.5 6
7.5 9 10.5 12 13.5
15 16.5 18 19.5 21
22.5 24 25.5 27 28.5
30 31.5 33 34.5 36
Is there a way to get the best of both worlds?
Thanks!
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
