I've been struggling with a problem for a few days that I think this might be great for. I'm trying to figure out how many AVX multiply-accumulate ops (vfmadd231ps) are being executed for a particular workload so I can compare two different implementations (going a bit deeper: I'm trying to figure out why there are substantial performance differences between two programs. Is it because one of them is doing a lot more multiply-accumulates (expensive ops) or is is because one is just much more cache efficient). I've got assembler for both, but one of them is macro-assembler so it's tough to actually reason about how many of these ops are being run.