Does NC profile 【tcgen05.ld.red】 instructions?

Developer Tools Nsight Compute
1

I need to measure tcgen05.ld.red instruction counts for performance analysis, but I do not currently have access to supported hardware. So I want to confirm the profiling capability before requesting access to the hardware.

  1. Does Nsight Compute capture and report tcgen05.ld.red instruction counts？

  2. Can the instruction mix be differentiated between tcgen05.ld and tcgen05.ld.red in the profiling output?

Thanks for your help !!

2

Nsight Compute doesn’t directly provide metrics for instruction counts of PTX instructions. It does provide detailed metrics for any SASS instruction generated from PTX, both in the CLI, UI and programmatically. It does also show per-line instructions counts of various types for PTX, assuming your compilation toolchain retains the PTX-to-SASS mappings in the compiled binary. You can however write a Python script to compute the overall aggregates yourself.