Identify the tasks to transfer linear algebra used in DDC splines to kokkos kernel

[jbigot]

No description provided.

[cedricchevalier19]

Meeting 5/13.

[blegouix]

Should I join ? Give me the time of the meeting if so

[yasahi-hpc]

kokkos/kokkos-kernels#2202

[tpadioleau]

General dense

LAPACKE_dgetrs
LAPACKE_dgetrf

General banded

LAPACKE_dgbtrs
LAPACKE_dgbtrf

PDS banded

LAPACKE_dpbtrs
LAPACKE_dpbtrf

PDS tridiagonal

LAPACKE_dpttrs
LAPACKE_dpttrf

[yasahi-hpc]

I have made a small investigation.

Function	dgetrs	dgbtrs	dpbtrs	dpttrs
Brief	general N-by-N matrix solver. A is factorized by dgetrf	general band matrix solver. A is factorized by dgbtrf	symmetric positive definite band matrix solver. A is factorized by dpbtrf	tridiagonal system solver. A is factorized by dpttrf
Dependency	dlaswp, dtrsm	dgemv, dger, dswap, dtbsv	dtbsv	dptts2
In Kokkos kernels	-	-	-	-

These functions depend on following subroutines.

Subroutines	dlaswp	dtrsm	dgemv	dger	dswap	dtbsv	dptts2	dscal
Brief	performs series of row interchanges on the matrix A	solves one of the matrix equations op( A )*X = alpha*B, or X*op( A ) = alpha*B	performs one of the matrix-vector operations y := alpha*A*x + beta*y, or y := alpha*A**T*x + beta*y	performs the rank 1 operation A := alpha*x*y**T + A	interchanges two vectors	solves one of the systems of equations A*x = b, or A**T*x = b	solves a tridiagonal system of the form A * X = B using the L*D*L' factorization of A computed by dpttrf	scales a vector by a constant
Serial	-	x	x	-	-	x	-	-
Team	-	x	x	-	-	-	-	-
Team Vector	-	x	-	-	-	-	-	-
Note				blas implementation available		In review	Depends on dscal	blas implementation available

[yasahi-hpc]

General dense

• Implement batched serial getrf (factorization) #2331: in review
• Implement batched serial iamax (helper) #2399: merged
• Implement batched serial laswp (helper) #2395: merged
• Implement batched serial getrs (solver)

General banded

• Implement batched serial gbtrf (factorization)
• Implement batched serial gbtrs (solver)

PDS banded

• Implement batched serial pbtrf (factorization) #2322: merged
• Implement batched serial pbtrs (solver) #2330: merged
• Add batched serial tbsv (solver) #2202: merged

PDS tridiagonal

• Implement batched serial pttrf (factorization) #2256: merged
• Implement batched serial pttrs (solver) #2277: merged

[jbigot]

Great news!