diff options
| author | Paul Zimmermann <Paul.Zimmermann@inria.fr> | 2021-04-01 08:14:10 +0200 |
|---|---|---|
| committer | Paul Zimmermann <Paul.Zimmermann@inria.fr> | 2021-04-02 06:15:48 +0200 |
| commit | 9acda61d94acc5348c2330f2519a14d1a4a37e73 (patch) | |
| tree | dcad90e95870279c37b5be7c646b3a3f6edc15cb /math/auto-libm-test-in | |
| parent | 595c22ecd8e87a27fd19270ed30fdbae9ad25426 (diff) | |
| download | glibc-9acda61d94acc5348c2330f2519a14d1a4a37e73.tar.gz glibc-9acda61d94acc5348c2330f2519a14d1a4a37e73.tar.xz glibc-9acda61d94acc5348c2330f2519a14d1a4a37e73.zip | |
Fix the inaccuracy of j0f/j1f/y0f/y1f [BZ #14469, #14470, #14471, #14472]
For j0f/j1f/y0f/y1f, the largest error for all binary32
inputs is reduced to at most 9 ulps for all rounding modes.
The new code is enabled only when there is a cancellation at the very end of
the j0f/j1f/y0f/y1f computation, or for very large inputs, thus should not
give any visible slowdown on average. Two different algorithms are used:
* around the first 64 zeros of j0/j1/y0/y1, approximation polynomials of
degree 3 are used, computed using the Sollya tool (https://www.sollya.org/)
* for large inputs, an asymptotic formula from [1] is used
[1] Fast and Accurate Bessel Function Computation,
John Harrison, Proceedings of Arith 19, 2009.
Inputs yielding the new largest errors are added to auto-libm-test-in,
and ulps are regenerated for various targets (thanks Adhemerval Zanella).
Tested on x86_64 with --disable-multi-arch and on powerpc64le-linux-gnu.
Reviewed-by: Adhemerval Zanella <adhemerval.zanella@linaro.org>
Diffstat (limited to 'math/auto-libm-test-in')
| -rw-r--r-- | math/auto-libm-test-in | 20 |
1 files changed, 17 insertions, 3 deletions
diff --git a/math/auto-libm-test-in b/math/auto-libm-test-in index 4edaaa8ee1..9fbd0c626b 100644 --- a/math/auto-libm-test-in +++ b/math/auto-libm-test-in @@ -5790,8 +5790,11 @@ j0 -0x1.001000001p+593 j0 0x1p1023 j0 0x1p16382 j0 0x1p16383 -# the next value generates larger error bounds on x86_64 (binary32) +# the next values yield large errors for binary32 +# (cf BZ #27670 for the xfail entry) j0 0x2.602774p+0 xfail-rounding:ibm128-libgcc +j0 0x1.04c39cp+6 +j0 0x1.4b7066p+7 # the next value exercises the flt-32 code path for x >= 2^127 j0 0x8.2f4ecp+124 @@ -5825,8 +5828,11 @@ j1 0x1p-60 j1 0x1p-100 j1 0x1p-600 j1 0x1p-10000 -# the next value generates larger error bounds on x86_64 (binary32) +# the next values yield large errors in the binary32 format +# (cf BZ #27670 for the xfail entries) j1 0x3.ae4b2p+0 xfail-rounding:ibm128-libgcc +j1 0x1.2f28eap+7 xfail-rounding:binary64 xfail-rounding:binary128 xfail-rounding:intel96 xfail-rounding:ibm128-libgcc +j1 0x1.a1d20ap+6 xfail-rounding:binary128 xfail-rounding:intel96 xfail-rounding:ibm128-libgcc j1 min j1 -min j1 min_subnorm @@ -8273,8 +8279,11 @@ y0 0x1p-100 y0 0x1p-110 y0 0x1p-600 y0 0x1p-10000 -# the next value generates larger error bounds on x86_64 (binary32) +# the next values yield large errors for binary32 +# (cf BZ #16492 for the xfail entries) y0 0xd.3432bp-4 +y0 0x1.33eaacp+5 xfail:binary64 xfail:intel96 xfail-rounding:ibm128-libgcc +y0 0x1.a681cep-1 xfail-rounding:ibm128-libgcc y0 min y0 min_subnorm @@ -8303,6 +8312,11 @@ y1 0x1p-100 y1 0x1p-110 y1 0x1p-600 y1 0x1p-10000 +# the next three values yield the largest error in the binary32 format +# (cf BZ #27670 for the xfail entries) +y1 0x1.065194p+7 xfail-rounding:binary64 xfail-rounding:intel96 xfail-rounding:ibm128-libgcc +y1 0x1.c1badep+0 xfail-rounding:ibm128-libgcc +y1 0x1.c1bc2ep+0 y1 min y1 min_subnorm |