Are the intrinsics listed anywhere?

Accelerated Computing CUDA CUDA Programming and Performance
1

Reading https://developer.nvidia.com/blog/cuda-pro-tip-optimized-filtering-warp-aggregated-atomics/, this refers to both functions __ffs(used once but not described in the programmers guide) and __popc (only mentioned that reduces to a single instruction).

These don’t seem to be described in the programmers guide. The PTX ISA documentation lists popc in the Reserved Instruction Set Table but not ffs.

Is there a comprehensive list of what intrinsics are available (perhaps platform-dependent), and how they function?

2

many intrinsics are listed in the CUDA math api docs including ffs and popc. There are some that are not there, including __dp4a().

3

Ah! Thank you!

Part of my problem was that the documentation search box on those pages doesn’t seem to pick them up. Even searching __ffs with the actual page open shows no results.

(although perhaps it’s a combination of problems, as even __syncthreads shows an empty box for me so maybe the search doesn’t like the leading underscores in combination with not liking the short length)

4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.