> I don't consider E cores and asymmetry to be a particular problem any more than SMT has been.
> If the algorithm scales, then good. Just run on every thread.

In an OpenMP program, unless the P-cores are given more work to do than the E-cores, the performance would be the same as just having E-cores. When a parallel for is used to compute the elements of an array across multiple cores, OpenMP has no way of knowing during compile time how many more array indices (loop iterations) to give to the P-cores compared to the E-cores. A runtime work queue could be used (called dynamic scheduling in OpenMP), but that has more overhead than static scheduling. As far as I could determine with a few internet searches, it is not possible to measure the performance of the P-cores and E-cores on iterations of a loop and statically schedule more loop iterations to the P-cores, at least with OpenMP. Does anyone know how to use heterogeneous CPU cores for parallel for loops without the overhead of dynamic loop scheduling?