On 10/19/2012 06:23 AM, Christophe Gisquet wrote: > Hi, > > 2012/10/19 Justin Ruggles <justin.rugg...@gmail.com>: >> - q->fdsp.vector_fmul(output, output, mdct_window, MDCT_SIZE); >> + q->fdsp.vector_fmul(output, output, mdct_window, 256); >> + q->dsp.vector_fmul_reverse(output + 256, output + 256, mdct_window, >> 256); > > What's the benefit? It looks like the same amount of computations. Is > it worthwhile to benchmark that? Locally (START/STOP_TIMER) or > globally (-benchmark) ?
Benchmarked locally, just using the single vector_fmul() is about 20% faster than the 2 separate calls. So I'm sending a new patch to just clean up the window init function. -Justin _______________________________________________ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel