jpekkila 20138263f4 The previous attempt (dsl_feature_completeness_2019-08-23) to enable arbitrary kernel functions was a failure: we get significant performance loss (25-100%) if step_number is not passed as a template parameter to the integration kernel. Apparently the CUDA compiler cannot perform some optimizations if there is a if/else construct in a performance-critical part which cannot be evaluated at compile time. This branch keeps step_number as a template parameter but takes rest of the user parameters as uniforms (dt is no longer passed as a function parameter but as an uniform with the DSL instead).

2019-08-27 17:36:33 +03:00

..

.gitignore

Moved STENCIL_ORDER and NGHOST out of user-defined parameter as these are actually internal defines used to configure the built-in functions. Additionally, renamed all explicitly declared uniforms from dsx -> AC_dsx in the DSL in preparation for having clear connection between DSL uniforms and the library parameter handles created by the user (AcRealParam etc)

2019-08-19 16:40:47 +03:00

stencil_assembly.sas

Autoformatted the DSL files

2019-08-20 18:41:26 +03:00

stencil_definition.sdh

The previous attempt (dsl_feature_completeness_2019-08-23) to enable arbitrary kernel functions was a failure: we get significant performance loss (25-100%) if step_number is not passed as a template parameter to the integration kernel. Apparently the CUDA compiler cannot perform some optimizations if there is a if/else construct in a performance-critical part which cannot be evaluated at compile time. This branch keeps step_number as a template parameter but takes rest of the user parameters as uniforms (dt is no longer passed as a function parameter but as an uniform with the DSL instead).

2019-08-27 17:36:33 +03:00

stencil_process.sps

The previous attempt (dsl_feature_completeness_2019-08-23) to enable arbitrary kernel functions was a failure: we get significant performance loss (25-100%) if step_number is not passed as a template parameter to the integration kernel. Apparently the CUDA compiler cannot perform some optimizations if there is a if/else construct in a performance-critical part which cannot be evaluated at compile time. This branch keeps step_number as a template parameter but takes rest of the user parameters as uniforms (dt is no longer passed as a function parameter but as an uniform with the DSL instead).

2019-08-27 17:36:33 +03:00