Add math special functions: erf, erfcx, expint1, Bessel functions, Hankel functions #3920

vqd8a · 2021-04-07T16:04:26Z

This PR will add mathematical special functions needed by GEMMA's deep slot code.

vqd8a · 2021-05-18T20:10:49Z

I have added the implementations of these functions and their unit tests. All checks passed except HIP tests in jenkin-ci check, where all 0s are returned for the Bessel functions and Hankel functions.
I don't know what happened since these unit tests are going well on Caraway and Tulip.

Rombur · 2021-05-18T20:24:34Z

Which compiler are you using on caraway and tulip? There are know bugs in the compiler used for CI. Once we merged #4025, the CI should used a better compiler.

vqd8a · 2021-05-18T20:40:28Z

@Rombur rocm/4.1.0 was used on caraway and rocm/4.2.0 was used on tulip. Thanks for letting me know. I will wait for #4025 got merged.

dalg24 · 2021-05-13T21:54:16Z

core/src/Kokkos_MathematicalSpecialFunctions.hpp

+    e1 = -infinity<RealType>::value;
+  }
+  else if (x == 0.0) {
+    e1 = infinity<RealType>::value;


Shouldn't it return -infty for x=0 ?

@dalg24 I use the definition of E1(x) while std::expint uses the definition of Ei(x). So I changed the function name to expint1 to avoid confusion.

core/src/Kokkos_Core.hpp

dalg24 · 2021-05-25T12:29:23Z

core/src/Kokkos_MathematicalSpecialFunctions.hpp

+
+//! Compute exponential integral E1(x) (x > 0).
+template <class RealType>
+KOKKOS_INLINE_FUNCTION RealType expint(const RealType& x) {


Why not taking argument by value

@dalg24 I see in Kokkos_Complex.hpp, taking argument by reference is used. Please explain why we should take argument by value.

Because this overload is for floating point numbers.
https://en.cppreference.com/w/cpp/numeric/special_functions/expint

@dalg24 I see. Thanks.

@dalg24 I have changed it to taking argument by value.

dalg24 · 2021-05-25T12:38:08Z

core/src/Kokkos_MathematicalSpecialFunctions.hpp

+
+//! Compute error function erf(z) for z=cmplx(x,y).
+template <class CmplxType>
+KOKKOS_INLINE_FUNCTION CmplxType cerf(const CmplxType& z) {


Consider constraining the overload set for complex numbers.

Suggested change

KOKKOS_INLINE_FUNCTION CmplxType cerf(const CmplxType& z) {

template <class T>

KOKKOS_INLINE_FUNCTION Kokkos::complex<T> cerf(const Kokkos::complex<T>& z) {

Why did you prefix the name with "c"?

@dalg24 The names are now erf and erfcx.

dalg24 · 2021-06-01T21:13:43Z

core/src/Kokkos_MathematicalSpecialFunctions.hpp

+//! Compute scaled complementary error function erfcx(z)=exp(z^2)*erfc(z)
+//! for z=cmplx(x,y).
+template <class CmplxType>
+KOKKOS_INLINE_FUNCTION CmplxType cerfcx(const CmplxType& z) {


Ok to add erfcx but we need to provide an implementation for scalar types as well
erfcx(FloatingPoint)

@dalg24 I have added an implementation for scalar type. For now, I do not use the std::erfc() for erfcx()=exp(x^2)*erfc(x) because it is going to overflow quickly (e.g. erfcx(35)=NaN). Instead, I use erfcx(Kokkos::complex<RealType>(x, 0.0)) for now. I will have an implementation of erfcx(real) in the future.

dalg24 · 2021-06-01T21:28:24Z

core/src/Kokkos_MathematicalSpecialFunctions.hpp

+//! Compute Bessel function J0(z) of the first kind of order zero
+//! for a complex argument
+template <class CmplxType, class RealType, class IntType>
+KOKKOS_INLINE_FUNCTION CmplxType cbesselj0(const CmplxType& z,


Implement cyl_bessel_j and error out if the order argument is not 0 (or 1)

@dalg24 Since we decided to use cyl_bessel_j0 and cyl_bessel_j1 for order 0 and order 1, respectively, so I do not have cyl_bessel_j for now.

dalg24 · 2021-06-01T21:34:54Z

core/src/Kokkos_MathematicalSpecialFunctions.hpp

+//! Compute Bessel function Y1(z) of the second kind of order one
+//! for a complex argument
+template <class CmplxType, class RealType, class IntType>
+KOKKOS_INLINE_FUNCTION CmplxType cbessely1(const CmplxType& z,


Rename cyl_bessel_y1 for consistency with the C++ math functions

@dalg24 I have changed all the Bessel function names accordingly.

vqd8a · 2021-06-29T14:19:55Z

@dalg24 I have addressed all your comments. Please let me know if you have additional comments. Thanks.

crtrott

I'd love for Damien to approve but not sure he got access right now.

vqd8a · 2021-07-08T18:12:32Z

@crtrott Thanks, Christian. Should we wait for Damien to merge it?

crtrott · 2021-07-08T18:13:25Z

I will check if he is able to review if not i will go ahead

Initial placeholder for math special functions

c9cae21

vqd8a closed this Apr 7, 2021

vqd8a reopened this Apr 7, 2021

vqd8a marked this pull request as draft April 7, 2021 16:07

vqd8a added the [WIP] label Apr 7, 2021

vqd8a changed the title ~~DRAFT: Add math special functions: erf, erfcx, expint, bessel functions, hankel functions~~ DRAFT: Add math special functions: erf, erfcx, expint, Bessel functions, Hankel functions Apr 7, 2021

vqd8a added 21 commits May 5, 2021 10:38

Merge branch 'develop' into mathspecialfunctions

f57a70d

Merge branch 'develop' into mathspecialfunctions

b33f134

Add expint

707e209

Add expint and unit test

e80e2ae

Delete cpp file

226bd09

Add expint, cerf, cerfcx, cbesselj0, cbessely0 and unit tests

d9f8a3e

Add cbesselj1, cbessely1 and unit tests

82a6021

Add cbesseli0, cbesselk0, cbesseli1, cbesselk1, and unit tests

09ee16f

Merge branch 'develop' into mathspecialfunctions

6b8b671

Add cbesselh10, cbesselh11, cbesselh20, cbesselh21, and unit tests

e5549ce

Fix unused variable warnings

0d33639

Test nan

ff0d78b

Test nan

c6a7813

Test nan

6f8e1fb

Test nan

0e1bf38

Test nan

a6691b2

Test nan

73e2eef

Remove nan test

1f75e6e

Apply clang-format

15634e8

Fix PI and NAN issues

171e094

Modify Hankel function for z=0

1c91029

vqd8a changed the title ~~DRAFT: Add math special functions: erf, erfcx, expint, Bessel functions, Hankel functions~~ Add math special functions: erf, erfcx, expint, Bessel functions, Hankel functions May 18, 2021

vqd8a marked this pull request as ready for review May 18, 2021 20:11

vqd8a requested a review from dalg24 May 18, 2021 20:11

vqd8a removed the [WIP] label May 18, 2021

Merge branch 'develop' into mathspecialfunctions

1683f46

dalg24 reviewed Jun 1, 2021

View reviewed changes

vqd8a added 7 commits June 9, 2021 10:23

Merge branch 'develop' into mathspecialfunctions

687b178

Move Kokkos_MathematicalSpecialFunctions.hpp up

1642118

Update function names consistent with the C++ math functions

c8c7491

Update CMakeLists.txt

7288961

Update

783b969

Update CMakeLists

257af56

Add implementation of erfcx(Real)

6cca3db

vqd8a changed the title ~~Add math special functions: erf, erfcx, expint, Bessel functions, Hankel functions~~ Add math special functions: erf, erfcx, expint1, Bessel functions, Hankel functions Jun 29, 2021

vqd8a requested a review from crtrott June 30, 2021 18:29

crtrott approved these changes Jul 8, 2021

View reviewed changes

dalg24 merged commit 7c5ee65 into kokkos:develop Jul 12, 2021

vqd8a added InDevelop Enhancement, fix, etc. has been merged into the develop branch; Feature Request Create new capability; will potentially require voting labels Sep 15, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add math special functions: erf, erfcx, expint1, Bessel functions, Hankel functions #3920

Add math special functions: erf, erfcx, expint1, Bessel functions, Hankel functions #3920

vqd8a commented Apr 7, 2021

vqd8a commented May 18, 2021

Rombur commented May 18, 2021

vqd8a commented May 18, 2021

dalg24 May 13, 2021

vqd8a Jun 29, 2021

dalg24 May 25, 2021

vqd8a Jun 9, 2021

dalg24 Jun 9, 2021

vqd8a Jun 9, 2021

vqd8a Jun 29, 2021

dalg24 May 25, 2021

dalg24 May 25, 2021

vqd8a Jun 29, 2021

dalg24 Jun 1, 2021

vqd8a Jun 29, 2021

dalg24 Jun 1, 2021

vqd8a Jun 29, 2021

dalg24 Jun 1, 2021

vqd8a Jun 29, 2021

vqd8a commented Jun 29, 2021

crtrott left a comment

vqd8a commented Jul 8, 2021

crtrott commented Jul 8, 2021

	KOKKOS_INLINE_FUNCTION CmplxType cerf(const CmplxType& z) {
	template <class T>
	KOKKOS_INLINE_FUNCTION Kokkos::complex<T> cerf(const Kokkos::complex<T>& z) {

Add math special functions: erf, erfcx, expint1, Bessel functions, Hankel functions #3920

Add math special functions: erf, erfcx, expint1, Bessel functions, Hankel functions #3920

Conversation

vqd8a commented Apr 7, 2021

vqd8a commented May 18, 2021

Rombur commented May 18, 2021

vqd8a commented May 18, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

vqd8a commented Jun 29, 2021

crtrott left a comment

Choose a reason for hiding this comment

vqd8a commented Jul 8, 2021

crtrott commented Jul 8, 2021