Arseny Kapoulkine 8c0115c03a CodeGen: Rempve tempVectorPure for now ... On Apple Mn, it looks like denormals do not have an extra cost at least on the benchmarks tested. On old ARM CPUs it looks like denormals are flushed to zero; there might still be some CPUs that would benefit from zeroing out the 4th component, but for now let's assume that they are all well behaved. Removing zero-ins increases performance of some code by ~20% on Apple Mn.		2024-02-13 17:24:26 -08:00
..
include	CodeGen: Implement SIMD math and vector moves	2024-02-13 15:35:17 -08:00
src	CodeGen: Rempve tempVectorPure for now	2024-02-13 17:24:26 -08:00