dmeoli / ADS / 39327307091

98%

push

github

fft() is one unrolled codelet up to 64 points and above it the four steps of Bailey, with no reversal of the bits: the columns, then the rows, transformed 4 at a time in the lanes of the vectors of AVX (2 on SSE2 and NEON, by the vector extensions of GCC and Clang; arrays of doubles elsewhere or with ADS_FFT_SIMD=0), by codelets of up to 64 points and, above them, by the rounds of Stockham on 4 columns or rows gathered at a time with the next rows prefetched; the blocks between the steps are transposed in registers, those of the outputs start on a line of the cache whatever the alignment of the vector, and the powers of the root of unity between the steps come from one table, or from two from 2^16 points on; the tests of every power of 2 up to 2^20 against radix 2 and the definition, of every width of the lanes, of the points at every offset of a line, and the CI on arrays of doubles and on AVX

318 of 323 new or added lines in 1 file covered. (98.45%)

23629 of 24088 relevant lines covered (98.09%)

8652144.94 hits per line