diff options
| author | Martin Vignali <martin.vignali@gmail.com> | 2017-12-03 17:02:44 +0100 |
|---|---|---|
| committer | Martin Vignali <martin.vignali@gmail.com> | 2017-12-03 19:17:23 +0100 |
| commit | 9719d57b34f2156dc26d7a46ba809b1f9427011c (patch) | |
| tree | 653863d0cbb2abc7964d9fd6e112ee5a035a4854 /libavfilter | |
| parent | 51345cb1d5539027f2faa45b920cf97bbd9d5ff1 (diff) | |
| download | ffmpeg-9719d57b34f2156dc26d7a46ba809b1f9427011c.tar.gz | |
avfilter/x86/vf_threshold : add avx2 version for threshold 8
Diffstat (limited to 'libavfilter')
| -rw-r--r-- | libavfilter/x86/vf_threshold.asm | 7 | ||||
| -rw-r--r-- | libavfilter/x86/vf_threshold_init.c | 18 |
2 files changed, 22 insertions, 3 deletions
diff --git a/libavfilter/x86/vf_threshold.asm b/libavfilter/x86/vf_threshold.asm index 38499dcd04..117cc98fe2 100644 --- a/libavfilter/x86/vf_threshold.asm +++ b/libavfilter/x86/vf_threshold.asm @@ -34,7 +34,7 @@ SECTION .text cglobal threshold8, 10, 13, 5, in, threshold, min, max, out, ilinesize, tlinesize, flinesize, slinesize, olinesize, w, h, x mov wd, dword wm mov hd, dword hm - mova m4, [pb_128] + VBROADCASTI128 m4, [pb_128] add inq, wq add thresholdq, wq add minq, wq @@ -70,4 +70,9 @@ RET INIT_XMM sse4 THRESHOLD_8 +%if HAVE_AVX2_EXTERNAL +INIT_YMM avx2 +THRESHOLD_8 +%endif + %endif diff --git a/libavfilter/x86/vf_threshold_init.c b/libavfilter/x86/vf_threshold_init.c index e2bbae11d5..f448cb4b39 100644 --- a/libavfilter/x86/vf_threshold_init.c +++ b/libavfilter/x86/vf_threshold_init.c @@ -30,12 +30,26 @@ void ff_threshold8_sse4(const uint8_t *in, const uint8_t *threshold, ptrdiff_t flinesize, ptrdiff_t slinesize, ptrdiff_t olinesize, int w, int h); +void ff_threshold8_avx2(const uint8_t *in, const uint8_t *threshold, + const uint8_t *min, const uint8_t *max, + uint8_t *out, + ptrdiff_t ilinesize, ptrdiff_t tlinesize, + ptrdiff_t flinesize, ptrdiff_t slinesize, + ptrdiff_t olinesize, + int w, int h); av_cold void ff_threshold_init_x86(ThresholdContext *s) { int cpu_flags = av_get_cpu_flags(); - if (ARCH_X86_64 && EXTERNAL_SSE4(cpu_flags) && s->depth == 8) { - s->threshold = ff_threshold8_sse4; + if (ARCH_X86_64) { + if (s->depth == 8) { + if (EXTERNAL_SSE4(cpu_flags)) { + s->threshold = ff_threshold8_sse4; + } + if (EXTERNAL_AVX2_FAST(cpu_flags)) { + s->threshold = ff_threshold8_avx2; + } + } } } |