Evaluate the impact of adding FMA and/or increased optimization in CTSM on performance

From the NVIDIA/OpenACC hackathon we realized that we could likely increase the performance of CTSM by increasing the optimzation levels. FMA (Fused Multiply Adds) was turned off in CESM because of problems in CAM code. It seems unlikely that we'd have similar problems in CTSM code. Doing a little bit of performance testing to see what the improvement might be would be good to do. This was examined in CAM and the performance benefit wasn't worthwhile.

This was from the CSEG meeting.