https://bugs.llvm.org/show_bug.cgi?id=43609
Bug ID: 43609
Summary: Miscompile with -ffast-math and -ftree-vectorize on
arithmetic involving doubles and unsigned longs
Product: clang
Version: 8.0
Hardware: PC
OS: All
Status: NEW
Severity: normal
Priority: P
Component: C
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected], [email protected],
[email protected], [email protected],
[email protected]
The following program:
#include <stdlib.h>
#include <stdio.h>
__attribute((noinline)) void fill_vector(double * x, const double dx, const
size_t X) {
/* fill vector with uniformly spaced values */
for (size_t ix = 0; ix < X; ix++)
x[ix] = (ix + 0.5) * dx;
}
int main(void) {
const size_t X = 19;
double * x = malloc(sizeof(double) * X);
fill_vector(x, 1.0, X);
/* print the values */
for (size_t ix = 0; ix < X; ix++)
fprintf(stdout, "x[%zu] = %g\n", ix, x[ix]);
}
produces the following output in Clang 8 and newer with -Ofast (or -O1
-ffast-math -ftree-vectorize):
x[0] = 0
x[1] = 2
x[2] = 2
x[3] = 4
x[4] = 4
x[5] = 6
x[6] = 6
x[7] = 8
x[8] = 8
x[9] = 10
x[10] = 10
x[11] = 12
x[12] = 12
x[13] = 14
x[14] = 14
x[15] = 16
x[16] = 16.5
x[17] = 17.5
x[18] = 18.5
Note the change from incorrect to correct behaviour between the groups-of-4 and
the last three entries.
Changing the loop variable from "size_t" to "unsigned int" makes the problem go
away. Compiling with "-fno-associative-math" or, strangely,
"-fno-reciprocal-math" also makes it go away.
--
You are receiving this mail because:
You are on the CC list for the bug._______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
