LinearDiag - fix bug, comp i only contributes to the diagonal for comp i

MAGMA - update MAGMA backend to use new imode parameter

ElemRestriction - make lmode a separate enum

ElemRestriction - remove lmode from backend apply signature

ElemRestriction - move lmode to constructor over apply

Enable MAGMA non-tensor basis (#424)

* update magma backend files from magma-dev to new branch

* add skeleton for elem restrictions

* start putting code and files for nontensor case

* more

Enable MAGMA non-tensor basis (#424)

* update magma backend files from magma-dev to new branch

* add skeleton for elem restrictions

* start putting code and files for nontensor case

* more framework for adding magma elem restrictions

* typo/old code error fixes for interface and header file, etc

* add the writedofs kernels

* fix nonconstants - template them for now

* fix bugs in non-tensor basis apply (interp and weight)

* update magma backend files from magma-dev to new branch

* add skeleton for elem restrictions

* start putting code and files for nontensor case

* more framework for adding magma elem restrictions

* typo/old code error fixes for interface and header file, etc

* add the writedofs kernels

* fix nonconstants - template them for now

* fix bugs in non-tensor basis apply (interp and weight)

* fix incorrect merge conflict resolution of header file

* fix bugs in lmode=notranspose elem restrictions
and in copying of indices to device

* test simpler kernels for lmode=transpose elem restrict

* swap element and component ordering in tensor basis actions

* update comments in restriction kernels to match swapped ordering

* fix if statement to work with CEED_VECTOR_NONE instead of NULL

* minor code cleanup

* skip t204-7 for magma after changing E-vector layout

* remove commented old calls for now

* use magma set/get vectors instead of cuda memcopies

* remove dead code

* make style changes

Co-authored-by: Stan Tomov <tomov@eecs.utk.edu>
Co-authored-by: abdelfattah83 <36712794+abdelfattah83@users.noreply.github.com>

show more ...

Operator - use threshold based on max norm for diag/fdm assembly

Operator - only assemble diag for nonzero QF entries

Operator - fix fallback ceed mem leak

CPU - update lcov markers

Ref - reorganize AssembleLinearDiagonal loop for performance

Opt - clean up Operator create code

XSMM - release JiT kernels

CPU - add comment blocks at top of all backend functions

Basis - update documentation for CeedMatrixMultiply

Operator - Switch CreateFDMElementInverse to fallback mechanism

PETSc - remove extra memcpy in Multigrid example

Basis - drop BasisGetValue in favor of BasisGetInterp/Grad

Operator - move AssembleDiagonal to fallback

Style - rename files vec->vector for consistency

make style

Ref - small restriction changes

Ref - use CeedPragmaSIMD in rstr

Ref - make ncomp const for rstr as well

Ref - make restriction block size const