* Add Estrin's method for polynomial evaluation
N.B.: This is a slightly modified version of the code provided by Thomas Dybdahl Ahle in a github issue.
[CI SKIP] [ci skip]
* Add comparisons to Horner with std::array.
[CI SKIP]
* Add Estrin's method for polynomial evaluation
N.B.: This is a slightly modified version of the code provided by Thomas Dybdahl Ahle in a github issue.
[CI SKIP] [ci skip]
* Fix hang in n=0 case.
* Fix out of bounds access in test.
* Fix endsect for estrin.qbk.
* Apply clang-format to make the 'inspect' stage happy.
* Add type_traits header to includes.
* Add ulp plot.
* Document decreased accuracy of Estrin's method.
* Add assertion for size of scratch pad
* Remove std::size since it is C++17
* Add C++14 testing
* estrin -> evaluate_polynomial_estrin.
---------
Co-authored-by: jzmaddock <john@johnmaddock.co.uk>
Co-authored-by: Matt Borland <matt@mattborland.com>
* Update tanh_sinh handling of boundaries.
So that we don't accidentality end up at an end point, even when arithmetic is inexact, and the FP type has no denomrms.
Fixes: https://github.com/boostorg/math/issues/893
* Add licence/copyright to new file.
* Remove tabs.
* Remove tests we don't need right now.
!!!REVERT THIS COMMIT BEFORE MERGING!!!
* Add s390x testing to drone.
* Correct drone file.
* Correct drone file (again)
* Prevent complete cancellation in bessel_jy logic.
* Correct testing for 128-bit floats.
* Make some more tests 128-bit long double safe.
* Make more tests 128-bit float safe.
* Fix some more 128-bit testing issues.
* More 128-bit float fixes.
* Make more tests 128-bit float safe.
* Fix up remaining tests for 128-bit floats.
* Yet more 128-bit float test case fixes.
* Fix up more tests for 128-bit floats and non-intel platforms.
* Fix up more tests to be 128-bit long double safe.
* More test case adjustments.
* More 128-bit float error rate adjustments.
* Fixes for autodiff tests
* Two more test fixes.
* Fix up daubechies_scaling_test.cpp and reinstate full CI.
Co-authored-by: Matt Borland <matt@mattborland.com>
where max_left_index/max_right_index may go to zero if the function being integrated overflows the FP type being used across nearly all of it's range. Fixes https://github.com/boostorg/math/issues/898
xref scipy/scipy#14901
Splits single if statement into two to avoid apparent clang short-circuit bug on M1 Mac platforms.
Co-authored-by: Matt Borland <matt@mattborland.com>
* Correct 1F1[-n, -n, x]
To match what Mathematica does.
Fixes: https://github.com/boostorg/math/issues/829
* Pass policy in forwarding call to gamma_q.
* Fix bug in 1F1 recurrence relations when one term goes to zero.
Fixes issue with 1F1[-n, n, n].
Also reduce cancellation error in a = -1 special case.
Fixes https://github.com/boostorg/math/issues/833.
