[FFmpeg-cvslog] avfilter/vf_interlace : add AVX2 for lowpass_line 8 and 16

[Martin Vignali]

ffmpeg | branch: master | Martin Vignali <martin.vignali at gmail.com> | Mon Dec 18 11:19:17 2017 +0100| [1a5865b6dcc97754a1d7eedc130fb58237d2a715] | committer: Martin Vignali

avfilter/vf_interlace : add AVX2 for lowpass_line 8 and 16

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=1a5865b6dcc97754a1d7eedc130fb58237d2a715
---

 libavfilter/x86/vf_interlace.asm    |  5 +++++
 libavfilter/x86/vf_interlace_init.c | 12 ++++++++++++
 2 files changed, 17 insertions(+)

diff --git a/libavfilter/x86/vf_interlace.asm b/libavfilter/x86/vf_interlace.asm
index 06b269828a..99ec9e0542 100644
--- a/libavfilter/x86/vf_interlace.asm
+++ b/libavfilter/x86/vf_interlace.asm
@@ -201,5 +201,10 @@ LOWPASS_LINE
 INIT_XMM avx
 LOWPASS_LINE
 
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+LOWPASS_LINE
+%endif
+
 INIT_XMM sse2
 LOWPASS_LINE_COMPLEX
diff --git a/libavfilter/x86/vf_interlace_init.c b/libavfilter/x86/vf_interlace_init.c
index b024b61735..0de0fea382 100644
--- a/libavfilter/x86/vf_interlace_init.c
+++ b/libavfilter/x86/vf_interlace_init.c
@@ -32,6 +32,9 @@ void ff_lowpass_line_sse2(uint8_t *dstp, ptrdiff_t linesize,
 void ff_lowpass_line_avx (uint8_t *dstp, ptrdiff_t linesize,
                           const uint8_t *srcp, ptrdiff_t mref,
                           ptrdiff_t pref, int clip_max);
+void ff_lowpass_line_avx2 (uint8_t *dstp, ptrdiff_t linesize,
+                          const uint8_t *srcp, ptrdiff_t mref,
+                          ptrdiff_t pref, int clip_max);
 
 void ff_lowpass_line_16_sse2(uint8_t *dstp, ptrdiff_t linesize,
                              const uint8_t *srcp, ptrdiff_t mref,
@@ -39,6 +42,9 @@ void ff_lowpass_line_16_sse2(uint8_t *dstp, ptrdiff_t linesize,
 void ff_lowpass_line_16_avx (uint8_t *dstp, ptrdiff_t linesize,
                              const uint8_t *srcp, ptrdiff_t mref,
                              ptrdiff_t pref, int clip_max);
+void ff_lowpass_line_16_avx2 (uint8_t *dstp, ptrdiff_t linesize,
+                             const uint8_t *srcp, ptrdiff_t mref,
+                             ptrdiff_t pref, int clip_max);
 
 void ff_lowpass_line_complex_sse2(uint8_t *dstp, ptrdiff_t linesize,
                                   const uint8_t *srcp, ptrdiff_t mref,
@@ -62,6 +68,9 @@ av_cold void ff_interlace_init_x86(InterlaceContext *s, int depth)
         if (EXTERNAL_AVX(cpu_flags))
             if (s->lowpass == VLPF_LIN)
                 s->lowpass_line = ff_lowpass_line_16_avx;
+        if (EXTERNAL_AVX2_FAST(cpu_flags))
+            if (s->lowpass == VLPF_LIN)
+                s->lowpass_line = ff_lowpass_line_16_avx2;
     } else {
         if (EXTERNAL_SSE2(cpu_flags)) {
             if (s->lowpass == VLPF_LIN)
@@ -72,5 +81,8 @@ av_cold void ff_interlace_init_x86(InterlaceContext *s, int depth)
         if (EXTERNAL_AVX(cpu_flags))
             if (s->lowpass == VLPF_LIN)
                 s->lowpass_line = ff_lowpass_line_avx;
+        if (EXTERNAL_AVX2_FAST(cpu_flags))
+            if (s->lowpass == VLPF_LIN)
+                s->lowpass_line = ff_lowpass_line_avx2;
     }
 }