lavc/aarch64: add pred16x16 10-bit functions

Benchmarks:                      A53     A72
pred16x16_dc_10_c:              136.0   124.0
pred16x16_dc_10_neon:           121.2   106.0
pred16x16_horizontal_10_c:      155.0    73.2
pred16x16_horizontal_10_neon:    82.2    67.7
pred16x16_top_dc_10_c:          106.0    93.7
pred16x16_top_dc_10_neon:        87.7    77.2
pred16x16_vertical_10_c:         83.0    67.7
pred16x16_vertical_10_neon:      54.2    61.7

Some functions work slower than C and are left commented out.

libavcodec/aarch64/h264pred_init.c

@@ -45,6 +45,11 @@ void ff_pred8x8_0lt_dc_neon(uint8_t *src, ptrdiff_t stride);
 void ff_pred8x8_l00_dc_neon(uint8_t *src, ptrdiff_t stride);
 void ff_pred8x8_0l0_dc_neon(uint8_t *src, ptrdiff_t stride);
 
+void ff_pred16x16_top_dc_neon_10(uint8_t *src, ptrdiff_t stride);
+void ff_pred16x16_dc_neon_10(uint8_t *src, ptrdiff_t stride);
+void ff_pred16x16_hor_neon_10(uint8_t *src, ptrdiff_t stride);
+void ff_pred16x16_vert_neon_10(uint8_t *src, ptrdiff_t stride);
+
 static av_cold void h264_pred_init_neon(H264PredContext *h, int codec_id,
                                         const int bit_depth,
                                         const int chroma_format_idc)
@@ -78,6 +83,12 @@ static av_cold void h264_pred_init_neon(H264PredContext *h, int codec_id,
             codec_id != AV_CODEC_ID_VP7 && codec_id != AV_CODEC_ID_VP8)
             h->pred16x16[PLANE_PRED8x8  ] = ff_pred16x16_plane_neon;
     }
+    if (bit_depth == 10) {
+        h->pred16x16[DC_PRED8x8     ] = ff_pred16x16_dc_neon_10;
+        h->pred16x16[VERT_PRED8x8   ] = ff_pred16x16_vert_neon_10;
+        h->pred16x16[HOR_PRED8x8    ] = ff_pred16x16_hor_neon_10;
+        h->pred16x16[TOP_DC_PRED8x8 ] = ff_pred16x16_top_dc_neon_10;
+    }
 }
 
 av_cold void ff_h264_pred_init_aarch64(H264PredContext *h, int codec_id,

libavcodec/aarch64/h264pred_neon.S

@@ -359,3 +359,111 @@ function ff_pred8x8_0l0_dc_neon, export=1
         dup             v1.8b,  v1.b[0]
         b               .L_pred8x8_dc_end
 endfunc
+
+.macro ldcol.16  rd,  rs,  rt,  n=4,  hi=0
+.if \n >= 4 || \hi == 0
+        ld1             {\rd\().h}[0],  [\rs], \rt
+        ld1             {\rd\().h}[1],  [\rs], \rt
+.endif
+.if \n >= 4 || \hi == 1
+        ld1             {\rd\().h}[2],  [\rs], \rt
+        ld1             {\rd\().h}[3],  [\rs], \rt
+.endif
+.if \n == 8
+        ld1             {\rd\().h}[4],  [\rs], \rt
+        ld1             {\rd\().h}[5],  [\rs], \rt
+        ld1             {\rd\().h}[6],  [\rs], \rt
+        ld1             {\rd\().h}[7],  [\rs], \rt
+.endif
+.endm
+
+// slower than C
+/*
+function ff_pred16x16_128_dc_neon_10, export=1
+        movi            v0.8h, #2, lsl #8 // 512, 1 << (bit_depth - 1)
+
+        b               .L_pred16x16_dc_10_end
+endfunc
+*/
+
+function ff_pred16x16_top_dc_neon_10, export=1
+        sub             x2,  x0,  x1
+
+        ld1             {v0.8h, v1.8h}, [x2]
+
+        add             v0.8h, v0.8h, v1.8h
+        addv            h0, v0.8h
+
+        urshr           v0.4h,  v0.4h,  #4
+        dup             v0.8h, v0.h[0]
+        b               .L_pred16x16_dc_10_end
+endfunc
+
+// slower than C
+/*
+function ff_pred16x16_left_dc_neon_10, export=1
+        sub             x2,  x0,  #2 // access to the "left" column
+        ldcol.16        v0,  x2,  x1,  8
+        ldcol.16        v1,  x2,  x1,  8 // load "left" column
+
+        add             v0.8h, v0.8h, v1.8h
+        addv            h0,  v0.8h
+
+        urshr           v0.4h,  v0.4h,  #4
+        dup             v0.8h, v0.h[0]
+        b               .L_pred16x16_dc_10_end
+endfunc
+*/
+
+function ff_pred16x16_dc_neon_10, export=1
+        sub             x2,  x0,  x1 // access to the "top" row
+        sub             x3,  x0,  #2 // access to the "left" column
+
+        ld1             {v0.8h, v1.8h}, [x2]
+        ldcol.16        v2,  x3,  x1,  8
+        ldcol.16        v3,  x3,  x1,  8 // load pixels in "top" row and "left" col
+
+        add             v0.8h, v0.8h, v1.8h
+        add             v2.8h, v2.8h, v3.8h
+        add             v0.8h, v0.8h, v2.8h
+        addv            h0, v0.8h
+
+        urshr           v0.4h,  v0.4h,  #5
+        dup             v0.8h,  v0.h[0]
+.L_pred16x16_dc_10_end:
+        mov             v1.16b,  v0.16b
+        mov             w3,  #8
+6:      st1             {v0.8h, v1.8h}, [x0], x1
+        subs            w3,  w3,  #1
+        st1             {v0.8h, v1.8h}, [x0], x1
+        b.ne            6b
+        ret
+endfunc
+
+function ff_pred16x16_hor_neon_10, export=1
+        sub             x2,  x0,  #2
+        add             x3,  x0,  #16
+
+        mov             w4,  #16
+1:      ld1r            {v0.8h},  [x2],  x1
+        subs            w4,  w4,  #1
+        st1             {v0.8h},  [x0],  x1
+        st1             {v0.8h},  [x3],  x1
+        b.ne            1b
+        ret
+endfunc
+
+function ff_pred16x16_vert_neon_10, export=1
+        sub             x2,  x0,  x1
+        add             x1,  x1,  x1
+
+        ld1             {v0.8h, v1.8h},  [x2],  x1
+
+        mov             w3,  #8
+1:      subs            w3,  w3,  #1
+        st1             {v0.8h, v1.8h},  [x0],  x1
+        st1             {v0.8h, v1.8h},  [x2],  x1
+
+        b.ne            1b
+        ret
+endfunc