This enables us to actually use / test the reaction sorting scheme (e.g., via the command line interface or via the PYJAC_RXN_SORT environment variable for unit-tests).
In addition, fixes a performance degradation in OpenCL code (compared to paper version) -- the Intel pown function is much slower than our own internal (unrolled) power function, so switch back.
This enables us to actually use / test the reaction sorting scheme (e.g., via the command line interface or via the
PYJAC_RXN_SORTenvironment variable for unit-tests).In addition, fixes a performance degradation in OpenCL code (compared to paper version) -- the Intel
pownfunction is much slower than our own internal (unrolled) power function, so switch back.