[x264-devel] [PATCH 04/11] aarch64: implement x264_pixel_vsad_neon

[Janne Grunau]

35 times faster than C.
---
 common/aarch64/pixel-a.S | 24 ++++++++++++++++++++++++
 common/aarch64/pixel.h   |  2 ++
 common/pixel.c           |  1 +
 3 files changed, 27 insertions(+)

diff --git a/common/aarch64/pixel-a.S b/common/aarch64/pixel-a.S
index 8c7b927..e3ca63d 100644
--- a/common/aarch64/pixel-a.S
+++ b/common/aarch64/pixel-a.S
@@ -248,6 +248,30 @@ SAD_X_FUNC  4, 16, 8
 SAD_X_FUNC  4, 16, 16
 
 
+function x264_pixel_vsad_neon, export=1
+    subs        w2,  w2,  #2
+    ld1        {v0.16b},  [x0],  x1
+    ld1        {v1.16b},  [x0],  x1
+    uabdl       v6.8h,  v0.8b,  v1.8b
+    uabdl2      v7.8h,  v0.16b, v1.16b
+    b.le        2f
+1:
+    subs        w2,  w2,  #2
+    ld1        {v0.16b},  [x0],  x1
+    uabal       v6.8h,  v1.8b,  v0.8b
+    uabal2      v7.8h,  v1.16b, v0.16b
+    ld1        {v1.16b},  [x0],  x1
+    b.lt        2f
+    uabal       v6.8h,  v0.8b,  v1.8b
+    uabal2      v7.8h,  v0.16b, v1.16b
+    b.gt        1b
+2:
+    add         v5.8h,  v6.8h,  v7.8h
+    uaddlv      s0,  v5.8h
+    fmov        w0,  s0
+    ret
+endfunc
+
 .macro SSD_START_4
     ld1        {v16.s}[0], [x0], x1
     ld1        {v17.s}[0], [x2], x3
diff --git a/common/aarch64/pixel.h b/common/aarch64/pixel.h
index 99ab16d..e1f1b7c 100644
--- a/common/aarch64/pixel.h
+++ b/common/aarch64/pixel.h
@@ -47,6 +47,8 @@ DECL_X4( sad, neon )
 DECL_X1( satd, neon )
 DECL_X1( ssd, neon )
 
+int x264_pixel_vsad_neon( uint8_t *, intptr_t, int );
+
 int x264_pixel_sa8d_8x8_neon  ( uint8_t *, intptr_t, uint8_t *, intptr_t );
 int x264_pixel_sa8d_16x16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t );
 uint64_t x264_pixel_sa8d_satd_16x16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t );
diff --git a/common/pixel.c b/common/pixel.c
index 5842aed..cd6fa1c 100644
--- a/common/pixel.c
+++ b/common/pixel.c
@@ -1428,6 +1428,7 @@ void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
         pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
+        pixf->vsad = x264_pixel_vsad_neon;
 
         pixf->intra_sad_x3_4x4    = x264_intra_sad_x3_4x4_neon;
         pixf->intra_satd_x3_4x4   = x264_intra_satd_x3_4x4_neon;
-- 
2.0.4