diff options
| author | Joseph Myers <joseph@codesourcery.com> | 2021-09-21 21:54:37 +0000 |
|---|---|---|
| committer | Joseph Myers <joseph@codesourcery.com> | 2021-09-21 21:54:37 +0000 |
| commit | 1356f38df5be0776823eb2c40cc4e607b86b9680 (patch) | |
| tree | 52590f89365894844fac4bb98898e4725445c844 /math/auto-libm-test-in | |
| parent | 0b5ca7c3e551e5502f3be3b06453324fe8604e82 (diff) | |
| download | glibc-1356f38df5be0776823eb2c40cc4e607b86b9680.tar.gz glibc-1356f38df5be0776823eb2c40cc4e607b86b9680.tar.xz glibc-1356f38df5be0776823eb2c40cc4e607b86b9680.zip | |
Fix f64xdivf128, f64xmulf128 spurious underflows (bug 28358)
As described in bug 28358, the round-to-odd computations used in the libm functions that round their results to a narrower format can yield spurious underflow exceptions in the following circumstances: the narrowing only narrows the precision of the type and not the exponent range (i.e., it's narrowing _Float128 to _Float64x on x86_64, x86 or ia64), the architecture does after-rounding tininess detection (which applies to all those architectures), the result is inexact, tiny before rounding but not tiny after rounding (with the chosen rounding mode) for _Float64x (which is possible for narrowing mul, div and fma, not for narrowing add, sub or sqrt), so the underflow exception resulting from the toward-zero computation in _Float128 is spurious for _Float64x. Fixed by making ROUND_TO_ODD call feclearexcept (FE_UNDERFLOW) in the problem cases (as indicated by an extra argument to the macro); there is never any need to preserve underflow exceptions from this part of the computation, because the conversion of the round-to-odd value to the narrower type will underflow in exactly the cases in which the function should raise that exception, but it may be more efficient to avoid the extra manipulation of the floating-point environment when not needed. Tested for x86_64 and x86, and with build-many-glibcs.py.
Diffstat (limited to 'math/auto-libm-test-in')
| -rw-r--r-- | math/auto-libm-test-in | 30 |
1 files changed, 30 insertions, 0 deletions
diff --git a/math/auto-libm-test-in b/math/auto-libm-test-in index 2045a0f8d1..119e2ecb2a 100644 --- a/math/auto-libm-test-in +++ b/math/auto-libm-test-in @@ -4870,6 +4870,21 @@ div 0x1.0000000000008001000000000001p0 0x1.0000000000008p0 # Similar, for double rounding to 64-bit of a division of 53-bit values. div 0x1ffe1p0 0xfffp0 +# Cases where there is underflow before rounding (for some format) but +# might not be after rounding, depending on the rounding mode. +div 0x1p-126 0x1.0000001p0 +div 0x1p-126 -0x1.0000001p0 +div -0x1p-126 0x1.0000001p0 +div -0x1p-126 -0x1.0000001p0 +div 0x1p-1022 0x1.00000000000001p0 +div 0x1p-1022 -0x1.00000000000001p0 +div -0x1p-1022 0x1.00000000000001p0 +div -0x1p-1022 -0x1.00000000000001p0 +div 0x1p-16382 0x1.00000000000000001p0 +div 0x1p-16382 -0x1.00000000000000001p0 +div -0x1p-16382 0x1.00000000000000001p0 +div -0x1p-16382 -0x1.00000000000000001p0 + erf 0 erf -0 erf 0.125 @@ -6645,6 +6660,21 @@ mul 0x50000000000000005p-64 0xcccccccccccccccccccccccccccdp-114 # This product equals 2^64 + 2^11 + 1. mul 97689974585 188829449 +# Cases where there is underflow before rounding (for some format) but +# might not be after rounding, depending on the rounding mode. +mul 0x0.ffffff8p-126 0x1.0000001p0 +mul 0x0.ffffff8p-126 -0x1.0000001p0 +mul -0x0.ffffff8p-126 0x1.0000001p0 +mul -0x0.ffffff8p-126 -0x1.0000001p0 +mul 0x0.fffffffffffffcp-1022 0x1.00000000000001p0 +mul 0x0.fffffffffffffcp-1022 -0x1.00000000000001p0 +mul -0x0.fffffffffffffcp-1022 0x1.00000000000001p0 +mul -0x0.fffffffffffffcp-1022 -0x1.00000000000001p0 +mul 0x0.ffffffffffffffff8p-16382 0x1.00000000000000001p0 +mul 0x0.ffffffffffffffff8p-16382 -0x1.00000000000000001p0 +mul -0x0.ffffffffffffffff8p-16382 0x1.00000000000000001p0 +mul -0x0.ffffffffffffffff8p-16382 -0x1.00000000000000001p0 + pow 0 0 pow 0 -0 pow -0 0 |