fab format question #473
-
I see that when writing the checkpoint file we set fab format to native and then after writing the checkpoint we set it back to the previous format (specified by the user for e.g. single precision). Is there a data loss in the conversion? Can you point me to where do we ensure that the calculations are still being done in double precision? Thank you |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
We will always output the checkpoint files in whatever format the host computer is using, at full precision. The fab format will allow you to swap the endianness of the plotfile or output it in reduced (i.e. 32-bit instead of 64-bit) precision. In some setups you'll see:
this makes plotfiles single precision, mainly just to save space. Most output tools can manage the byte-swapping if the endianness of the machine you are doing analysis on is different than the one that you are doing the simulations on, so you don't need to worry about that generally. |
Beta Was this translation helpful? Give feedback.
We will always output the checkpoint files in whatever format the host computer is using, at full precision. The fab format will allow you to swap the endianness of the plotfile or output it in reduced (i.e. 32-bit instead of 64-bit) precision.
In some setups you'll see:
this makes plotfiles single precision, mainly just to save space. Most output tools can manage the byte-swapping if the endianness of the machine you are doing analysis on is different than the one that you are doing the simulations on, so you don't need to worry about that generally.