mysterious local memory usage in my kernel

I have a cuda kernel which is not delivering the performance I expected. I checked the cubin file for the memory usage (56 bytes of local memory is used even though I didnt force a register cap on the kernel):

lmem = 56
smem = 144
reg  = 21
bar  = 0

My kernel is listed below. Can anyone let me know why lmem is used in my kernel? Thanks,

It’s the calls to cosf and sinf. Can you use __cosf and __sinf? They don’t require local memory.

That solves my problem. Thank you,