Performance and Accuracy

Performance and Accuracy

The vector intrinsics are optimized and software pipelined to take advantage of the R8000's performance features. Throughput is several times greater than that of repeatedly calling the corresponding scalar function though the result may not necessarily agree to the last bit. Also, the results from vsin(), vcos() and vtan() are valid only for argument values between 228 and -228. Similarly, Vlog(0.0) and vlog(+inf) return NaN.