On Wed, Jan 18, 2012 at 2:27 PM, Alex Converse <[email protected]> wrote: > On Wed, Jan 18, 2012 at 12:39 AM, Nathan Caldwell <[email protected]> wrote: >> >> --- >> libavcodec/aacpsy.c | 3 +-- >> 1 files changed, 1 insertions(+), 2 deletions(-) >> >> diff --git a/libavcodec/aacpsy.c b/libavcodec/aacpsy.c >> index 04feea3..577d8fb 100644 >> --- a/libavcodec/aacpsy.c >> +++ b/libavcodec/aacpsy.c >> @@ -823,8 +823,7 @@ static FFPsyWindowInfo psy_lame_window(FFPsyContext *ctx, >> float const *const pfe = pf + AAC_BLOCK_SIZE_LONG / >> (AAC_NUM_BLOCKS_SHORT * PSY_LAME_NUM_SUBBLOCKS); >> float p = 1.0f; >> for (; pf < pfe; pf++) >> - if (p < fabsf(*pf)) >> - p = fabsf(*pf); >> + p = FFMAX(p, fabsf(*pf)); > > Do we double evaluate fabsf(*pf) here? Maybe using an explicit temp > would be better? This isn't where we spend most of our cycles IIRC so > I guess this is OK as-is.
IIRC, most of the time is spent in a couple of the quantize_and_encode_cost functions, followed by analysis, followed by another quantize_and_encode_cost function. -- -Nathan Caldwell _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
