[6476] in Athena Bugs
likely rt compiler bug... (2.1y)
daemon@ATHENA.MIT.EDU (daemon@ATHENA.MIT.EDU)
Wed Nov 28 03:22:34 1990
Date: Wed, 28 Nov 90 03:22:19 -0500
From: Ken Raeburn <Raeburn@MIT.Edu>
To: bugs@ATHENA.MIT.EDU
If I'm reading it right, the assembly output from the following code,
when compiled without optimization, is incorrect. Optimizing "fixes"
the problem.
#include <stdio.h>
#include <math.h>
#define sqr(x) ((x)*(x))
void icd(n, nx, a, l)
int n, nx;
double **a, **l;
{
int i;
double liisq;
for (i=0; i<n; i++) {
liisq= a[0][i];
if (i > 0) liisq -= sqr(l[1][i-1]);
if (i-nx >=0) liisq -= sqr(l[2][i-nx]);
printf("before line\n");
l[0][i] = sqrt(liisq);
printf("line passed\n");
l[1][i] = a[1][i]/l[0][i];
l[2][i] = a[2][i]/l[0][i];
}
}
The code storing the result of `sqrt' is:
balix r15,_.sqrt
l r0,LF2-_icd(r14)
ls r0,0(r13) # get l[0]
cas r4,r0,r11 # get an address
mr r11,r12 # *now* compute the offset needed
shl r11,3 # for the above address calculation?
sts r2,0(r4) # store...
sts r3,4(r4)
No other instructions in this function set r11.
This came from an OLC user this evening....
