A list of nominal CUDA instructions throughput

Accelerated Computing CUDA CUDA Programming and Performance
1

This presentation:

explains that instruction throughput depends on “Nominal instruction throughput”. It later says that arithmetic instruction throughput for an integer add, for example, is 4 cycles/warp. Is there any list of nominal throughputs for most/all of CUDA instructions?

EDIT:
I found

But it doesn’t have modulo operation for example. Also, how about atomic functions?

2

Table 2 in Section 5.4.1 of the CUDA C Programming Guide (CUDA 6 release candidate) gives the throughput for different categories of instruction on each of the compute capabilities.

3

The reason the modulo operation is not listed in that table is because there is no modulo instruction (or division instruction, for that matter) provided by the hardware.

4

As for atomic functions, the throughput depends on the access pattern, so it can be quite variable. Best to benchmark it on your hardware with realistic data.