3 ms·
GNU GCC 10 to Implement MMX Intrinsics with SSE
- rurban 8y agoI don't get it. Why emulating MMX with slower SSE ops (due to needed translations) when the CPU can run it natively? The internal CPU microops do it much better than the assembler code. I don't see the benefit at all. Maybe because the kernel doesn't need to safe the MMX refs at context switches anymore. But that's kernel-only, and gcc is mostly for userland.