caught segfault in [R] function - c

Question

I am very new to C but know my way around [R]. This error could be a very stupid mistake in C.
My C code does kernel smoothing.
*When I comment out the last line of code my function works: results[i] = v; *
This call kills R:
new.y<-zsmooth2( x=c(0:80000), xpts=dat$V2, ypts=dat$V4, h=10000)
* caught segfault *
address 0x1184f8000, cause 'memory not mapped'
Traceback:
1: .C("kernel_smooth", as.double(x), as.double(ypts), as.double(xpts), as.integer(n), as.integer(nxpts), as.double(h), result = double(length(xpts)))
2: zsmooth2(x = c(0:80000), xpts = dat$V2, ypts = dat$V4, h = 10000)
C-code:
#include <R.h>
#include <Rmath.h>
#include <stdio.h>
void kernel_smooth(double *x, double *ypts, double *xpts, int *n, int *nxpts, double *h, double *results){
int i, j;
for(i = 0; i < *n; i++){
double nsum = 0;
double dsum = 0;
double z = x[i] + *h;
double y = x[i] - *h;
for(j = 0; j < *nxpts; j++){
if(xpts[j] < y){
continue;
}
if(xpts[j] > z){
break;
}
double d = (xpts[j] - i) / *h;
double r = dnorm(d, 0, 1, 0);
nsum += r * ypts[j];
dsum += r;
}
Rprintf("test:i %d\n", i);
double v = nsum / dsum;
Rprintf("test:v %f\n", v);
results[i] = v;
}
}
R-code:
dyn.load("~/github/ZevRTricks/smoother1.so")
zsmooth2<-function(x, ypts, xpts, h){
n <- length(x)
nxpts <- length(xpts)
dens <- .C("kernel_smooth", as.double(x), as.double(ypts),
as.double(xpts), as.integer(n), as.integer(nxpts),
as.double(h), result = double(length(xpts)))
dens[["result"]]
}

Answer 1

xpts and ypts are vectors, and in your C code you are trying to access elements 1 to n in each of them. n is the length of x, which is 100 times longer in your second example than in your first example. Compare seq(from = 0, to = 80000 by = 100) to 0:80000, (and while you're at it you can drop the c() from around the 0:80000).
So I guess that xpts and ypts are at least 801 elements long, but less than 80001 elements. You've messed up your indexing somewhere.
Note also that you pass x to your C code, but don't actually use it for anything.