diff options
| author | Szabolcs Nagy <szabolcs.nagy@arm.com> | 2018-12-07 13:47:14 +0000 |
|---|---|---|
| committer | Szabolcs Nagy <szabolcs.nagy@arm.com> | 2018-12-07 14:45:20 +0000 |
| commit | 75b8d8c6c12b37a38211defcc5b941adb7de121f (patch) | |
| tree | afbc1a9594f364a3f74ab0b38383f10bb04376b1 /math/powf.c | |
| parent | 11253b0b9d6ba14e5a609ff7436db1079c3f9bb8 (diff) | |
| download | arm-optimized-routines-75b8d8c6c12b37a38211defcc5b941adb7de121f.tar.gz | |
Fix powf overflow handling in non-nearest rounding mode
The threshold value at which powf overflows depends on the rounding mode
and the current check did not take this into account. So when the result
was rounded away from zero it could become infinity without setting
errno to ERANGE.
Example: pow(0x1.7ac7cp+5, 23) is 0x1.fffffep+127 + 0.1633ulp
If the result goes above 0x1.fffffep+127 + 0.5ulp then errno is set,
which is fine in nearest rounding mode, but
powf(0x1.7ac7cp+5, 23) is inf in upward rounding mode
powf(-0x1.7ac7cp+5, 23) is -inf in downward rounding mode
and the previous implementation did not set errno in these cases.
This special case is fixed without affecting the common code path by
checking the rounding mode and using appropriate threshold value.
Arithmetics is used to check the rounding mode to avoid introducing a
stack frame when calling fegetround.
Unfortunately the current test system does not support rounding modes.
(Found by comparing against musl powf behaviour on the libc-test test
suite, unfortunately that test system does not support errno.)
Diffstat (limited to 'math/powf.c')
| -rw-r--r-- | math/powf.c | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/math/powf.c b/math/powf.c index af51adb..06116f0 100644 --- a/math/powf.c +++ b/math/powf.c @@ -199,7 +199,14 @@ powf (float x, float y) { /* |y*log(x)| >= 126. */ if (ylogx > 0x1.fffffffd1d571p+6 * POWF_SCALE) + /* |x^y| > 0x1.ffffffp127. */ return __math_oflowf (sign_bias); + if (WANT_ROUNDING && WANT_ERRNO + && ylogx > 0x1.fffffffa3aae2p+6 * POWF_SCALE) + /* |x^y| > 0x1.fffffep127, check if we round away from 0. */ + if ((!sign_bias && opt_barrier_float (-ylogx) != -128 * POWF_SCALE) + || (sign_bias && -opt_barrier_float (ylogx) != -128 * POWF_SCALE)) + return __math_oflowf (sign_bias); if (ylogx <= -150.0 * POWF_SCALE) return __math_uflowf (sign_bias); #if WANT_ERRNO_UFLOW |