Merge branch 'cpu/opt' [PR #82]

* cpu/opt:
[Opt] Add first version of opt backend, use colocated grad
Add backend prioritization
Add t17-basis to test BasisApply for GRAD and TRANSPOSE

[Opt] Add first version of opt backend, use colocated grad

Merge branch 'master' into evector_support

Merge branch 'template-backend' [PR #63]

* template-backend:
Add template backend, requested in issue #38
Split Ref into multiple files, update style

Add template backend, requested in issue #38

Merge pull request #36 from CEED/magma-dev

[WIP] Magma backend [magma-dev]

Merge branch 'magma-dev' of https://github.com/CEED/libCEED into magma-dev

Additional tweaks for building using IBM XL Fortran.

In Makefile, use CUDA headers/libs/compiler from the same location.
Tweak the MAGMA link flags.

make: handle opt/debug in consistent way

I'm not wild about this "make NDEBUG=" interface for specifying an
optimization preference, but don't want to change it right now for the
release.

Merge branch 'master' into magma-dev

Conflicts:
Makefile
tests/tap.sh

Merge branch 'master' into run_nek_fix

Merge branch 'jed/FC-environment' [PR #55]

* jed/FC-environment:
makefile: only replace CC and FC when they are default (or undefined)

makefile: only replace CC and FC when they are default (or undefined)

Unconditional use of = overrides the environment, but spack would prefer
to just export CC and FC rather than pass as command-li

makefile: only replace CC and FC when they are default (or undefined)

Unconditional use of = overrides the environment, but spack would prefer
to just export CC and FC rather than pass as command-line arguments.

Reported-by: Veselin Dobrev <dobrev1@llnl.gov>

show more ...

Merge branch 'master' into nek_gpu

Makefile: use FC=gfortran instead of ?=

Make predefines FC=f77 so our FC?=gfortran is skipped. Note that
command line and environment variables still take precedent, so this
change won't affect any

Makefile: use FC=gfortran instead of ?=

Make predefines FC=f77 so our FC?=gfortran is skipped. Note that
command line and environment variables still take precedent, so this
change won't affect any users that have these variables set.

show more ...

Merge branch 'unify-examples-dev' [PR #52]

* unify-examples-dev:
examples/petsc: work around offensive astyle formatting
Several improvements/unification between the libCEED examples

Several small updates in the Nek scripts and documentations.

With these, I can run all libCEED examples with the /gpu/occa spec on my Linux
box!!

Merge branch 'master' into nek_gpu

Merge pull request #53 from thilinarmtb/install_ceedf_h

install: Add ceedf.h to install

install: Add ceedf.h to install

Several improvements/unification between the libCEED examples

- All examples except Nek support '-ceed' now as command-line parameter for the Ceed spec
- The MFEM and PETSc examples were renamed to

Several improvements/unification between the libCEED examples

- All examples except Nek support '-ceed' now as command-line parameter for the Ceed spec
- The MFEM and PETSc examples were renamed to bp?
- The libceed example (ex1-volume-annulus) was moved to examples/ceed/ex1
- For clarity, the libCEED portion of the MFEM examples were moved to header files
- Documentation + sample runs were added

show more ...

Merge branch 'master' into nek5k_updates

Merge branch 'occa-dev' [PR #42]

* occa-dev: (70 commits)
Makefile: explicitly use bash because shell script does not have conforming shebang
make okl-cache: simplify caching call and fix to fin

Merge branch 'occa-dev' [PR #42]

* occa-dev: (70 commits)
Makefile: explicitly use bash because shell script does not have conforming shebang
make okl-cache: simplify caching call and fix to find OCCA executable
Cache OCCA kernels in PETSc examples makefile
[occa] PETSc example with ctx inout, qfdatasize & offsets
[occa] device from ressource, OpenCL TILE size tied to 1
[occa] fixed /ocl/occa
Couple more OKL kernel fixes in example Makefiles
Comment-out the restrict kernels using atomicAdd(double), so we can build the OCCA backends on GPU cards with lower SMs.
[occa] occaPropertiesSet to match new C wrapper API
[occa] dbg macros + ceed argument moved down to occa backend
[occa] Makefile targets comments, occa source debug cleanup
[test] t30 output removed, added V vector 0.0 check
[occa] examples Makefile occa cache
[occa] basis & restrict stat logic to test out of source examples
[occa] pushing backends/occa/*.okl to install prefix/lib/okl and use them if needed
[occa] style
[occa] CeedDebug output when CEED_DEBUG or DBG are non null
[occa] Updated to 1.0 C API
[occa] make style
[occa] cleanup + OCL but tests that use ex1-volume-annulus.okl
...

show more ...

Makefile: explicitly use bash because shell script does not have conforming shebang

make okl-cache: simplify caching call and fix to find OCCA executable

The occa command does not give the user a choice about where the cached
files go, so don't define variables that could be mislea

make okl-cache: simplify caching call and fix to find OCCA executable

The occa command does not give the user a choice about where the cached
files go, so don't define variables that could be misleading.

show more ...