diff options
| author | Lynne <dev@lynne.ee> | 2022-09-19 05:53:01 +0200 |
|---|---|---|
| committer | Lynne <dev@lynne.ee> | 2022-09-23 12:35:27 +0200 |
| commit | ace42cf581f8c06872bfb58cf575d9e8bd398c0a (patch) | |
| tree | 217d6653d5664d47f95c327fdb09d63e01dffcb3 /tests | |
| parent | 3241e9225c7adfb2d8d24cfd05a7a8db8ddbd023 (diff) | |
| download | ffmpeg-ace42cf581f8c06872bfb58cf575d9e8bd398c0a.tar.gz | |
x86/tx_float: add 15xN PFA FFT AVX SIMD
~4x faster than the C version.
The shuffles in the 15pt dim1 are seriously expensive. Not happy with it,
but I'm contempt.
Can be easily converted to pure AVX by removing all vpermpd/vpermps
instructions.
Diffstat (limited to 'tests')
| -rw-r--r-- | tests/checkasm/av_tx.c | 4 |
1 files changed, 2 insertions, 2 deletions
diff --git a/tests/checkasm/av_tx.c b/tests/checkasm/av_tx.c index 1fa6da45ac..aa8fc6b4e9 100644 --- a/tests/checkasm/av_tx.c +++ b/tests/checkasm/av_tx.c @@ -24,7 +24,7 @@ #include <stdlib.h> -#define EPS 0.00005 +#define EPS 0.0005 #define SCALE_NOOP(x) (x) #define SCALE_INT20(x) (av_clip64(lrintf((x) * 2147483648.0), INT32_MIN, INT32_MAX) >> 12) @@ -40,7 +40,7 @@ } while (0) static const int check_lens[] = { - 2, 4, 8, 16, 32, 64, 1024, 16384, + 2, 4, 8, 16, 32, 64, 120, 960, 1024, 1920, 16384, }; static AVTXContext *tx_refs[AV_TX_NB][2 /* Direction */][FF_ARRAY_ELEMS(check_lens)] = { 0 }; |