From bbefc3dad687f4370afd5bc85386d983201cb321 Mon Sep 17 00:00:00 2001 From: MaxThevenet Date: Mon, 24 Feb 2020 15:54:50 -0800 Subject: Profiler wrapper to allow for cudaDeviceSynchronize (#738) * add warpx profiler wrapper * call WX profiler wrapper instead of BL * EOL and cleaning * add doc on profile syncs * do not use PROFILE macros in .H files * replace WX_PROFILE with WARPX_PROFILE and use bool instead of int * typo * this file shouldn't be there --- Docs/source/running_cpp/parameters.rst | 4 ++++ 1 file changed, 4 insertions(+) (limited to 'Docs') diff --git a/Docs/source/running_cpp/parameters.rst b/Docs/source/running_cpp/parameters.rst index 0121e7bf1..936dd262a 100644 --- a/Docs/source/running_cpp/parameters.rst +++ b/Docs/source/running_cpp/parameters.rst @@ -974,6 +974,10 @@ Numerics and algorithms value here will make the simulation unphysical, but will allow QED effects to become more apparent. Note that this option will only have an effect if the warpx.use_Hybrid_QED flag is also triggered. + * ``warpx.do_device_synchronize_before_profile`` (`bool`) optional (default `1`) + When running in an accelerated platform, whether to call a deviceSynchronize around profiling regions. + This allows the profiler to give meaningful timers, but (hardly) slows down the simulation. + * ``warpx.sort_int`` (`int`) optional (defaults: ``-1`` on CPU; ``4`` on GPU) If ``<=0``, do not sort particles. If ``>0``, sort particles by bin every ``sort_int`` iteration. It is turned on on GPUs for performance reasons (to improve memory locality). -- cgit v1.2.3