On Mon, 12 May 2014, Janne Grunau wrote:
Opus celt decoding 11% faster and the iMDCT over 2.5 times faster on
Apple's A7.
---
Untangled the 3 * 5 * 2^x imdct/fft from opus. Only file, struct and
function names are still referencing opus but that is just cosmetic.
libavcodec/aarch64/Makefile | 2 +
libavcodec/aarch64/asm-offsets.h | 30 ++
libavcodec/aarch64/opus_imdct_init.c | 45 +++
libavcodec/aarch64/opus_imdct_neon.S | 647 +++++++++++++++++++++++++++++++++++
libavcodec/opus.h | 18 -
libavcodec/opus_celt.c | 5 +-
libavcodec/opus_imdct.c | 42 ++-
libavcodec/opus_imdct.h | 59 ++++
8 files changed, 809 insertions(+), 39 deletions(-)
create mode 100644 libavcodec/aarch64/asm-offsets.h
create mode 100644 libavcodec/aarch64/opus_imdct_init.c
create mode 100644 libavcodec/aarch64/opus_imdct_neon.S
create mode 100644 libavcodec/opus_imdct.h
Looks ok to me. What about doing the same for arm as well?
// Martin
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
