FFTW Logo
Go to the previous or top benchmark results.

Results on a Cray C90:

You should note that the C90 is an exotic machine. It is a vector processor, which means that it has special instructions for doing operations on multiple array elements simultaneously. In theory, the compiler automatically vectorizes your code where possible. In practice, however, you have to play with the structure of your code to get the compiler to vectorize it. The only FFT subroutine that was optimized for vectorization on the C90 was Cray's SCILIB routine (this is obvious from the graph below).

1D Benchmark for Array Size = Power of 2

3D Benchmark

Go to the previous or top benchmark results. hat was optimized for vectorization on the C90 was Cray's SCILIB routine (this is obvious from the graph below).

1D Benchmark for Array Size = Power of 2

3D Benchmark

Go to the previous or top benchmark results. gif" WIDTnp` 29" ALT="FFTiPgo">
Go to the previous or top benchmark results.

Results on a Cray C90:

You should note that the C90 is an exotic machine. It is a vector processor, which means that it has special instructions for doing operations on multiple array elements simultaneously. In theory, the compiler automatically vectorizes your code where possible. In practice, however, you have to play with the structure of your code to get the compiler to vectorize it. The only FFT subroutine tFFT subroutine t