ace8e64850
At least SSE and ARM have opcodes that handle float-to-int conversions well enough. Also, Clang doesn't inline lrintf, incurring function call overhead for what should be a single opcode.