[FFmpeg-cvslog] avcodec/vp9: avx2 implementation of ipred_dl_16x16_16

[Ilia]

ffmpeg | branch: master | Ilia <zakne0ne at gmail.com> | Mon Mar 13 05:06:26 2017 +0700| [2f3d10a01ac5f613f80db8542bf3ecda1dd40d79] | committer: Ronald S. Bultje

avcodec/vp9: avx2 implementation of ipred_dl_16x16_16

vp9_diag_downleft_16x16_10bpp_c: 263.0
vp9_diag_downleft_16x16_10bpp_sse2: 44.7
vp9_diag_downleft_16x16_10bpp_ssse3: 32.5
vp9_diag_downleft_16x16_10bpp_avx: 31.9
vp9_diag_downleft_16x16_10bpp_avx2: 25.7
vp9_diag_downleft_16x16_12bpp_c: 264.7
vp9_diag_downleft_16x16_12bpp_sse2: 44.4
vp9_diag_downleft_16x16_12bpp_ssse3: 32.0
vp9_diag_downleft_16x16_12bpp_avx: 32.4
vp9_diag_downleft_16x16_12bpp_avx2: 25.5

Benchmarked with 10000 runs

Signed-off-by: Ilia <zakne0ne at gmail.com>
Signed-off-by: Ronald S. Bultje <rsbultje at gmail.com>

> http://git.videolan.org/gitweb.cgi/ffmpeg.git/?a=commit;h=2f3d10a01ac5f613f80db8542bf3ecda1dd40d79
---

 libavcodec/x86/vp9dsp_init_16bpp.c    |  2 ++
 libavcodec/x86/vp9intrapred_16bpp.asm | 39 +++++++++++++++++++++++++++++++++++
 2 files changed, 41 insertions(+)

diff --git a/libavcodec/x86/vp9dsp_init_16bpp.c b/libavcodec/x86/vp9dsp_init_16bpp.c
index eb67499..4576ff1 100644
--- a/libavcodec/x86/vp9dsp_init_16bpp.c
+++ b/libavcodec/x86/vp9dsp_init_16bpp.c
@@ -51,6 +51,7 @@ decl_ipred_fns(h,       16, mmxext, sse2);
 decl_ipred_fns(dc,      16, mmxext, sse2);
 decl_ipred_fns(dc_top,  16, mmxext, sse2);
 decl_ipred_fns(dc_left, 16, mmxext, sse2);
+decl_ipred_fn(dl,       16,     16, avx2);
 
 #define decl_ipred_dir_funcs(type) \
 decl_ipred_fns(type, 16, sse2,  sse2); \
@@ -133,6 +134,7 @@ av_cold void ff_vp9dsp_init_16bpp_x86(VP9DSPContext *dsp)
         init_fpel_func(2, 1,  32, avg, _16, avx2);
         init_fpel_func(1, 1,  64, avg, _16, avx2);
         init_fpel_func(0, 1, 128, avg, _16, avx2);
+        init_ipred_func(dl, DIAG_DOWN_LEFT, 16, 16, avx2);
     }
 
 #endif /* HAVE_YASM */
diff --git a/libavcodec/x86/vp9intrapred_16bpp.asm b/libavcodec/x86/vp9intrapred_16bpp.asm
index c0ac16d..212e413 100644
--- a/libavcodec/x86/vp9intrapred_16bpp.asm
+++ b/libavcodec/x86/vp9intrapred_16bpp.asm
@@ -847,6 +847,45 @@ DL_FUNCS
 INIT_XMM avx
 DL_FUNCS
 
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+cglobal vp9_ipred_dl_16x16_16, 2, 4, 5, dst, stride, l, a
+    movifnidn               aq, amp
+    mova                    m0, [aq]                   ; abcdefghijklmnop
+    vpbroadcastw           xm1, [aq+30]                ; pppppppp
+    vperm2i128              m2, m0, m1, q0201          ; ijklmnoppppppppp
+    vpalignr                m3, m2, m0, 2              ; bcdefghijklmnopp
+    vpalignr                m4, m2, m0, 4              ; cdefghijklmnoppp
+    LOWPASS                  0,  3,  4                 ; BCDEFGHIJKLMNOPp
+    vperm2i128              m2, m0, m1, q0201          ; JKLMNOPppppppppp
+    DEFINE_ARGS dst, stride, stride3, cnt
+    mov                   cntd, 2
+    lea               stride3q, [strideq*3]
+.loop:
+    mova      [dstq+strideq*0], m0
+    vpalignr                m3, m2, m0, 2
+    vpalignr                m4, m2, m0, 4
+    mova      [dstq+strideq*1], m3
+    mova      [dstq+strideq*2], m4
+    vpalignr                m3, m2, m0, 6
+    vpalignr                m4, m2, m0, 8
+    mova      [dstq+stride3q ], m3
+    lea                   dstq, [dstq+strideq*4]
+    mova      [dstq+strideq*0], m4
+    vpalignr                m3, m2, m0, 10
+    vpalignr                m4, m2, m0, 12
+    mova      [dstq+strideq*1], m3
+    mova      [dstq+strideq*2], m4
+    vpalignr                m3, m2, m0, 14
+    mova      [dstq+stride3q ], m3
+    lea                   dstq, [dstq+strideq*4]
+    mova                    m0, m2
+    vperm2i128              m2, m2, m2, q0101          ; pppppppppppppppp
+    dec                   cntd
+    jg .loop
+    RET
+%endif
+
 %macro DR_FUNCS 1 ; stack_mem_for_32x32_32bit_function
 cglobal vp9_ipred_dr_4x4_16, 4, 4, 3, dst, stride, l, a
     movh                    m0, [lq]                ; wxyz....