h264/aarch64: add intra loop filter neon asm

Add my neon asm from x264 relicensed under the LGPL 2.1 or later. Ported
(x264 uses nv12 chroma) and optimized.

Cycle count for checkasm --bench on a Snapdragon 820e:
h264_h_loop_filter_luma_intra_8bpp_c: 60.0
h264_h_loop_filter_luma_intra_8bpp_neon: 54.2
h264_v_loop_filter_luma_intra_8bpp_c: 148.3
h264_v_loop_filter_luma_intra_8bpp_neon: 73.8
h264_h_loop_filter_chroma_intra_8bpp_c: 27.8
h264_h_loop_filter_chroma_intra_8bpp_neon: 21.4
h264_h_loop_filter_chroma_mbaff_intra_8bpp_c: 15.8
h264_h_loop_filter_chroma_mbaff_intra_8bpp_neon: 15.7
h264_v_loop_filter_chroma_intra_8bpp_c: 45.8
h264_v_loop_filter_chroma_intra_8bpp_neon: 17.3

2 files changed, 313 insertions, 0 deletions

diff --git a/libavcodec/aarch64/h264dsp_init_aarch64.c b/libavcodec/aarch64/h264dsp_init_aarch64.c
index b106f11..07bda2f 100644
--- a/libavcodec/aarch64/h264dsp_init_aarch64.c
+++ b/libavcodec/aarch64/h264dsp_init_aarch64.c
@@ -29,10 +29,20 @@ void ff_h264_v_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
                                      int beta, int8_t *tc0);
 void ff_h264_h_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
                                      int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_luma_intra_neon(uint8_t *pix, int stride, int alpha,
+                                           int beta);
+void ff_h264_h_loop_filter_luma_intra_neon(uint8_t *pix, int stride, int alpha,
+                                           int beta);
 void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
                                        int beta, int8_t *tc0);
 void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
                                        int beta, int8_t *tc0);
+void ff_h264_v_loop_filter_chroma_intra_neon(uint8_t *pix, int stride,
+                                             int alpha, int beta);
+void ff_h264_h_loop_filter_chroma_intra_neon(uint8_t *pix, int stride,
+                                             int alpha, int beta);
+void ff_h264_h_loop_filter_chroma_mbaff_intra_neon(uint8_t *pix, int stride,
+                                                   int alpha, int beta);
 
 void ff_weight_h264_pixels_16_neon(uint8_t *dst, int stride, int height,
                                    int log2_den, int weight, int offset);
@@ -77,8 +87,14 @@ av_cold void ff_h264dsp_init_aarch64(H264DSPContext *c, const int bit_depth,
     if (have_neon(cpu_flags) && bit_depth == 8) {
         c->h264_v_loop_filter_luma   = ff_h264_v_loop_filter_luma_neon;
         c->h264_h_loop_filter_luma   = ff_h264_h_loop_filter_luma_neon;
+        c->h264_v_loop_filter_luma_intra= ff_h264_v_loop_filter_luma_intra_neon;
+        c->h264_h_loop_filter_luma_intra= ff_h264_h_loop_filter_luma_intra_neon;
+
         c->h264_v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_neon;
         c->h264_h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
+        c->h264_v_loop_filter_chroma_intra = ff_h264_v_loop_filter_chroma_intra_neon;
+        c->h264_h_loop_filter_chroma_intra = ff_h264_h_loop_filter_chroma_intra_neon;
+        c->h264_h_loop_filter_chroma_mbaff_intra = ff_h264_h_loop_filter_chroma_mbaff_intra_neon;
 
         c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels_16_neon;
         c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels_8_neon;
diff --git a/libavcodec/aarch64/h264dsp_neon.S b/libavcodec/aarch64/h264dsp_neon.S
index b649f1d..448e575 100644
--- a/libavcodec/aarch64/h264dsp_neon.S
+++ b/libavcodec/aarch64/h264dsp_neon.S
@@ -1,6 +1,7 @@
 /*
  * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
  * Copyright (c) 2013 Janne Grunau <janne-libav@jannau.net>
+ * Copyright (c) 2014 Janne Grunau <janne-libav@jannau.net>
  *
  * This file is part of Libav.
  *
@@ -181,6 +182,203 @@ function ff_h264_h_loop_filter_luma_neon, export=1
         ret
 endfunc
 
+
+.macro h264_loop_filter_start_intra
+    orr             w4,  w2,  w3
+    cbnz            w4,  1f
+    ret
+1:
+    sxtw            x1,  w1
+    dup             v30.16b, w2                // alpha
+    dup             v31.16b, w3                // beta
+.endm
+
+.macro h264_loop_filter_luma_intra
+    uabd            v16.16b, v7.16b,  v0.16b        // abs(p0 - q0)
+    uabd            v17.16b, v6.16b,  v7.16b        // abs(p1 - p0)
+    uabd            v18.16b, v1.16b,  v0.16b        // abs(q1 - q0)
+    cmhi            v19.16b, v30.16b, v16.16b       // < alpha
+    cmhi            v17.16b, v31.16b, v17.16b       // < beta
+    cmhi            v18.16b, v31.16b, v18.16b       // < beta
+
+    movi            v29.16b, #2
+    ushr            v30.16b, v30.16b, #2            // alpha >> 2
+    add             v30.16b, v30.16b, v29.16b       // (alpha >> 2) + 2
+    cmhi            v16.16b, v30.16b, v16.16b       // < (alpha >> 2) + 2
+
+    and             v19.16b, v19.16b, v17.16b
+    and             v19.16b, v19.16b, v18.16b
+    shrn            v20.8b,  v19.8h,  #4
+    mov             x4, v20.d[0]
+    cbz             x4, 9f
+
+    ushll           v20.8h,  v6.8b,   #1
+    ushll           v22.8h,  v1.8b,   #1
+    ushll2          v21.8h,  v6.16b,  #1
+    ushll2          v23.8h,  v1.16b,  #1
+    uaddw           v20.8h,  v20.8h,  v7.8b
+    uaddw           v22.8h,  v22.8h,  v0.8b
+    uaddw2          v21.8h,  v21.8h,  v7.16b
+    uaddw2          v23.8h,  v23.8h,  v0.16b
+    uaddw           v20.8h,  v20.8h,  v1.8b
+    uaddw           v22.8h,  v22.8h,  v6.8b
+    uaddw2          v21.8h,  v21.8h,  v1.16b
+    uaddw2          v23.8h,  v23.8h,  v6.16b
+
+    rshrn           v24.8b,  v20.8h,  #2 // p0'_1
+    rshrn           v25.8b,  v22.8h,  #2 // q0'_1
+    rshrn2          v24.16b, v21.8h,  #2 // p0'_1
+    rshrn2          v25.16b, v23.8h,  #2 // q0'_1
+
+    uabd            v17.16b, v5.16b,  v7.16b        // abs(p2 - p0)
+    uabd            v18.16b, v2.16b,  v0.16b        // abs(q2 - q0)
+    cmhi            v17.16b, v31.16b, v17.16b       // < beta
+    cmhi            v18.16b, v31.16b, v18.16b       // < beta
+
+    and             v17.16b, v16.16b, v17.16b  // if_2 && if_3
+    and             v18.16b, v16.16b, v18.16b  // if_2 && if_4
+
+    not             v30.16b, v17.16b
+    not             v31.16b, v18.16b
+
+    and             v30.16b, v30.16b, v19.16b  // if_1 && !(if_2 && if_3)
+    and             v31.16b, v31.16b, v19.16b  // if_1 && !(if_2 && if_4)
+
+    and             v17.16b, v19.16b, v17.16b  // if_1 && if_2 && if_3
+    and             v18.16b, v19.16b, v18.16b  // if_1 && if_2 && if_4
+
+    //calc            p, v7, v6, v5, v4, v17, v7, v6, v5, v4
+    uaddl           v26.8h,  v5.8b,   v7.8b
+    uaddl2          v27.8h,  v5.16b,  v7.16b
+    uaddw           v26.8h,  v26.8h,  v0.8b
+    uaddw2          v27.8h,  v27.8h,  v0.16b
+    add             v20.8h,  v20.8h,  v26.8h
+    add             v21.8h,  v21.8h,  v27.8h
+    uaddw           v20.8h,  v20.8h,  v0.8b
+    uaddw2          v21.8h,  v21.8h,  v0.16b
+    rshrn           v20.8b,  v20.8h,  #3 // p0'_2
+    rshrn2          v20.16b, v21.8h,  #3 // p0'_2
+    uaddw           v26.8h,  v26.8h,  v6.8b
+    uaddw2          v27.8h,  v27.8h,  v6.16b
+    rshrn           v21.8b,  v26.8h,  #2 // p1'_2
+    rshrn2          v21.16b, v27.8h,  #2 // p1'_2
+    uaddl           v28.8h,  v4.8b,   v5.8b
+    uaddl2          v29.8h,  v4.16b,  v5.16b
+    shl             v28.8h,  v28.8h,  #1
+    shl             v29.8h,  v29.8h,  #1
+    add             v28.8h,  v28.8h,  v26.8h
+    add             v29.8h,  v29.8h,  v27.8h
+    rshrn           v19.8b,  v28.8h,  #3 // p2'_2
+    rshrn2          v19.16b, v29.8h,  #3 // p2'_2
+
+    //calc            q, v0, v1, v2, v3, v18, v0, v1, v2, v3
+    uaddl           v26.8h,  v2.8b,   v0.8b
+    uaddl2          v27.8h,  v2.16b,  v0.16b
+    uaddw           v26.8h,  v26.8h,  v7.8b
+    uaddw2          v27.8h,  v27.8h,  v7.16b
+    add             v22.8h,  v22.8h,  v26.8h
+    add             v23.8h,  v23.8h,  v27.8h
+    uaddw           v22.8h,  v22.8h,  v7.8b
+    uaddw2          v23.8h,  v23.8h,  v7.16b
+    rshrn           v22.8b,  v22.8h,  #3 // q0'_2
+    rshrn2          v22.16b, v23.8h,  #3 // q0'_2
+    uaddw           v26.8h,  v26.8h,  v1.8b
+    uaddw2          v27.8h,  v27.8h,  v1.16b
+    rshrn           v23.8b,  v26.8h,  #2 // q1'_2
+    rshrn2          v23.16b, v27.8h,  #2 // q1'_2
+    uaddl           v28.8h,  v2.8b,   v3.8b
+    uaddl2          v29.8h,  v2.16b,  v3.16b
+    shl             v28.8h,  v28.8h,  #1
+    shl             v29.8h,  v29.8h,  #1
+    add             v28.8h,  v28.8h,  v26.8h
+    add             v29.8h,  v29.8h,  v27.8h
+    rshrn           v26.8b,  v28.8h,  #3 // q2'_2
+    rshrn2          v26.16b, v29.8h,  #3 // q2'_2
+
+    bit             v7.16b,  v24.16b, v30.16b  // p0'_1
+    bit             v0.16b,  v25.16b, v31.16b  // q0'_1
+    bit             v7.16b, v20.16b,  v17.16b  // p0'_2
+    bit             v6.16b, v21.16b,  v17.16b  // p1'_2
+    bit             v5.16b, v19.16b,  v17.16b  // p2'_2
+    bit             v0.16b, v22.16b,  v18.16b  // q0'_2
+    bit             v1.16b, v23.16b,  v18.16b  // q1'_2
+    bit             v2.16b, v26.16b,  v18.16b  // q2'_2
+.endm
+
+function ff_h264_v_loop_filter_luma_intra_neon, export=1
+    h264_loop_filter_start_intra
+
+    ld1             {v0.16b},  [x0], x1 // q0
+    ld1             {v1.16b},  [x0], x1 // q1
+    ld1             {v2.16b},  [x0], x1 // q2
+    ld1             {v3.16b},  [x0], x1 // q3
+    sub             x0,  x0,  x1, lsl #3
+    ld1             {v4.16b},  [x0], x1 // p3
+    ld1             {v5.16b},  [x0], x1 // p2
+    ld1             {v6.16b},  [x0], x1 // p1
+    ld1             {v7.16b},  [x0]     // p0
+
+    h264_loop_filter_luma_intra
+
+    sub             x0,  x0,  x1, lsl #1
+    st1             {v5.16b}, [x0], x1  // p2
+    st1             {v6.16b}, [x0], x1  // p1
+    st1             {v7.16b}, [x0], x1  // p0
+    st1             {v0.16b}, [x0], x1  // q0
+    st1             {v1.16b}, [x0], x1  // q1
+    st1             {v2.16b}, [x0]      // q2
+9:
+    ret
+endfunc
+
+function ff_h264_h_loop_filter_luma_intra_neon, export=1
+    h264_loop_filter_start_intra
+
+    sub             x0,  x0,  #4
+    ld1             {v4.8b},  [x0], x1
+    ld1             {v5.8b},  [x0], x1
+    ld1             {v6.8b},  [x0], x1
+    ld1             {v7.8b},  [x0], x1
+    ld1             {v0.8b},  [x0], x1
+    ld1             {v1.8b},  [x0], x1
+    ld1             {v2.8b},  [x0], x1
+    ld1             {v3.8b},  [x0], x1
+    ld1             {v4.d}[1],  [x0], x1
+    ld1             {v5.d}[1],  [x0], x1
+    ld1             {v6.d}[1],  [x0], x1
+    ld1             {v7.d}[1],  [x0], x1
+    ld1             {v0.d}[1],  [x0], x1
+    ld1             {v1.d}[1],  [x0], x1
+    ld1             {v2.d}[1],  [x0], x1
+    ld1             {v3.d}[1],  [x0], x1
+
+    transpose_8x16B v4, v5, v6, v7, v0, v1, v2, v3, v21, v23
+
+    h264_loop_filter_luma_intra
+
+    transpose_8x16B v4, v5, v6, v7, v0, v1, v2, v3, v21, v23
+
+    sub             x0,  x0,  x1, lsl #4
+    st1             {v4.8b},  [x0], x1
+    st1             {v5.8b},  [x0], x1
+    st1             {v6.8b},  [x0], x1
+    st1             {v7.8b},  [x0], x1
+    st1             {v0.8b},  [x0], x1
+    st1             {v1.8b},  [x0], x1
+    st1             {v2.8b},  [x0], x1
+    st1             {v3.8b},  [x0], x1
+    st1             {v4.d}[1],  [x0], x1
+    st1             {v5.d}[1],  [x0], x1
+    st1             {v6.d}[1],  [x0], x1
+    st1             {v7.d}[1],  [x0], x1
+    st1             {v0.d}[1],  [x0], x1
+    st1             {v1.d}[1],  [x0], x1
+    st1             {v2.d}[1],  [x0], x1
+    st1             {v3.d}[1],  [x0], x1
+9:
+    ret
+endfunc
+
 .macro  h264_loop_filter_chroma
         dup             v22.8B, w2              // alpha
         dup             v23.8B, w3              // beta
@@ -266,6 +464,105 @@ function ff_h264_h_loop_filter_chroma_neon, export=1
         ret
 endfunc
 
+
+.macro h264_loop_filter_chroma_intra
+    uabd            v26.8b, v16.8b, v17.8b  // abs(p0 - q0)
+    uabd            v27.8b, v18.8b, v16.8b  // abs(p1 - p0)
+    uabd            v28.8b, v19.8b, v17.8b  // abs(q1 - q0)
+    cmhi            v26.8b, v30.8b, v26.8b  // < alpha
+    cmhi            v27.8b, v31.8b, v27.8b  // < beta
+    cmhi            v28.8b, v31.8b, v28.8b  // < beta
+    and             v26.8b, v26.8b, v27.8b
+    and             v26.8b, v26.8b, v28.8b
+    mov             x2, v26.d[0]
+
+    ushll           v4.8h,   v18.8b,  #1
+    ushll           v6.8h,   v19.8b,  #1
+    cbz             x2, 9f
+    uaddl           v20.8h,  v16.8b,  v19.8b
+    uaddl           v22.8h,  v17.8b,  v18.8b
+    add             v20.8h,  v20.8h,  v4.8h
+    add             v22.8h,  v22.8h,  v6.8h
+    uqrshrn         v24.8b,  v20.8h,  #2
+    uqrshrn         v25.8b,  v22.8h,  #2
+    bit             v16.8b, v24.8b, v26.8b
+    bit             v17.8b, v25.8b, v26.8b
+.endm
+
+function ff_h264_v_loop_filter_chroma_intra_neon, export=1
+    h264_loop_filter_start_intra
+
+    sub             x0,  x0,  x1, lsl #1
+    ld1             {v18.8b}, [x0], x1
+    ld1             {v16.8b}, [x0], x1
+    ld1             {v17.8b}, [x0], x1
+    ld1             {v19.8b}, [x0]
+
+    h264_loop_filter_chroma_intra
+
+    sub             x0,  x0,  x1, lsl #1
+    st1             {v16.8b}, [x0], x1
+    st1             {v17.8b}, [x0], x1
+
+9:
+    ret
+endfunc
+
+function ff_h264_h_loop_filter_chroma_mbaff_intra_neon, export=1
+    h264_loop_filter_start_intra
+
+    sub             x4,  x0,  #2
+    sub             x0,  x0,  #1
+    ld1             {v18.8b}, [x4], x1
+    ld1             {v16.8b}, [x4], x1
+    ld1             {v17.8b}, [x4], x1
+    ld1             {v19.8b}, [x4], x1
+
+    transpose_4x8B v18, v16, v17, v19, v26, v27, v28, v29
+
+    h264_loop_filter_chroma_intra
+
+    st2             {v16.b,v17.b}[0], [x0], x1
+    st2             {v16.b,v17.b}[1], [x0], x1
+    st2             {v16.b,v17.b}[2], [x0], x1
+    st2             {v16.b,v17.b}[3], [x0], x1
+
+9:
+    ret
+endfunc
+
+function ff_h264_h_loop_filter_chroma_intra_neon, export=1
+    h264_loop_filter_start_intra
+
+    sub             x4,  x0,  #2
+    sub             x0,  x0,  #1
+    ld1             {v18.8b}, [x4], x1
+    ld1             {v16.8b}, [x4], x1
+    ld1             {v17.8b}, [x4], x1
+    ld1             {v19.8b}, [x4], x1
+    ld1             {v18.s}[1], [x4], x1
+    ld1             {v16.s}[1], [x4], x1
+    ld1             {v17.s}[1], [x4], x1
+    ld1             {v19.s}[1], [x4]
+
+    transpose_4x8B v18, v16, v17, v19, v26, v27, v28, v29
+
+    h264_loop_filter_chroma_intra
+
+    st2             {v16.b,v17.b}[0], [x0], x1
+    st2             {v16.b,v17.b}[1], [x0], x1
+    st2             {v16.b,v17.b}[2], [x0], x1
+    st2             {v16.b,v17.b}[3], [x0], x1
+    st2             {v16.b,v17.b}[4], [x0], x1
+    st2             {v16.b,v17.b}[5], [x0], x1
+    st2             {v16.b,v17.b}[6], [x0], x1
+    st2             {v16.b,v17.b}[7], [x0], x1
+
+9:
+    ret
+endfunc
+
+
 .macro  biweight_16     macs, macd
         dup             v0.16B,  w5
         dup             v1.16B,  w6