aarch64: hpeldsp NEON optimizations

Ported from ARMv7 NEON.

4 files changed, 528 insertions, 5 deletions

diff --git a/libavcodec/aarch64/Makefile b/libavcodec/aarch64/Makefile
index 1d80d9a268..aa1dd4e073 100644
--- a/libavcodec/aarch64/Makefile
+++ b/libavcodec/aarch64/Makefile
@@ -1,9 +1,12 @@
 OBJS-$(CONFIG_H264CHROMA)               += aarch64/h264chroma_init_aarch64.o
 OBJS-$(CONFIG_H264DSP)                  += aarch64/h264dsp_init_aarch64.o
 OBJS-$(CONFIG_H264QPEL)                 += aarch64/h264qpel_init_aarch64.o
+OBJS-$(CONFIG_HPELDSP)                  += aarch64/hpeldsp_init_aarch64.o
 OBJS-$(CONFIG_RV40_DECODER)             += aarch64/rv40dsp_init_aarch64.o
 OBJS-$(CONFIG_VC1_DECODER)              += aarch64/vc1dsp_init_aarch64.o
 
 NEON-OBJS-$(CONFIG_H264CHROMA)          += aarch64/h264cmc_neon.o
 NEON-OBJS-$(CONFIG_H264DSP)             += aarch64/h264idct_neon.o
-NEON-OBJS-$(CONFIG_H264QPEL)            += aarch64/h264qpel_neon.o
+NEON-OBJS-$(CONFIG_H264QPEL)            += aarch64/h264qpel_neon.o             \
+                                           aarch64/hpeldsp_neon.o
+NEON-OBJS-$(CONFIG_HPELDSP)             += aarch64/hpeldsp_neon.o
diff --git a/libavcodec/aarch64/h264qpel_init_aarch64.c b/libavcodec/aarch64/h264qpel_init_aarch64.c
index 11611df241..4beb11b7b5 100644
--- a/libavcodec/aarch64/h264qpel_init_aarch64.c
+++ b/libavcodec/aarch64/h264qpel_init_aarch64.c
@@ -101,7 +101,7 @@ av_cold void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth)
     int cpu_flags = av_get_cpu_flags();
 
     if (have_neon(cpu_flags) && !high_bit_depth) {
-        /* c->put_h264_qpel_pixels_tab[0][ 0] = ff_put_h264_qpel16_mc00_neon; */
+        c->put_h264_qpel_pixels_tab[0][ 0] = ff_put_h264_qpel16_mc00_neon;
         c->put_h264_qpel_pixels_tab[0][ 1] = ff_put_h264_qpel16_mc10_neon;
         c->put_h264_qpel_pixels_tab[0][ 2] = ff_put_h264_qpel16_mc20_neon;
         c->put_h264_qpel_pixels_tab[0][ 3] = ff_put_h264_qpel16_mc30_neon;
@@ -118,7 +118,7 @@ av_cold void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth)
         c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_neon;
         c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_neon;
 
-        /* c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon; */
+        c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon;
         c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_neon;
         c->put_h264_qpel_pixels_tab[1][ 2] = ff_put_h264_qpel8_mc20_neon;
         c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_neon;
@@ -135,7 +135,7 @@ av_cold void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth)
         c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_neon;
         c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_neon;
 
-        /* c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon; */
+        c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon;
         c->avg_h264_qpel_pixels_tab[0][ 1] = ff_avg_h264_qpel16_mc10_neon;
         c->avg_h264_qpel_pixels_tab[0][ 2] = ff_avg_h264_qpel16_mc20_neon;
         c->avg_h264_qpel_pixels_tab[0][ 3] = ff_avg_h264_qpel16_mc30_neon;
@@ -152,7 +152,7 @@ av_cold void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth)
         c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_neon;
         c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_neon;
 
-        /* c->avg_h264_qpel_pixels_tab[1][ 0] = ff_avg_h264_qpel8_mc00_neon; */
+        c->avg_h264_qpel_pixels_tab[1][ 0] = ff_avg_h264_qpel8_mc00_neon;
         c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_neon;
         c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_neon;
         c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_neon;
diff --git a/libavcodec/aarch64/hpeldsp_init_aarch64.c b/libavcodec/aarch64/hpeldsp_init_aarch64.c
new file mode 100644
index 0000000000..6bc4c09f6c
--- /dev/null
+++ b/libavcodec/aarch64/hpeldsp_init_aarch64.c
@@ -0,0 +1,123 @@
+/*
+ * ARM NEON optimised DSP functions
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stddef.h>
+#include <stdint.h>
+
+#include "config.h"
+
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/aarch64/cpu.h"
+#include "libavcodec/hpeldsp.h"
+
+void     ff_put_pixels16_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void  ff_put_pixels16_x2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void  ff_put_pixels16_y2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void ff_put_pixels16_xy2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void      ff_put_pixels8_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void   ff_put_pixels8_x2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void   ff_put_pixels8_y2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void  ff_put_pixels8_xy2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+
+void  ff_put_pixels16_x2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void  ff_put_pixels16_y2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void ff_put_pixels16_xy2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void   ff_put_pixels8_x2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void   ff_put_pixels8_y2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void  ff_put_pixels8_xy2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+
+void     ff_avg_pixels16_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void  ff_avg_pixels16_x2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void  ff_avg_pixels16_y2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void ff_avg_pixels16_xy2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void      ff_avg_pixels8_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void   ff_avg_pixels8_x2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void   ff_avg_pixels8_y2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+void  ff_avg_pixels8_xy2_neon(uint8_t *block, const uint8_t *pixels,
+                              ptrdiff_t line_size, int h);
+
+void  ff_avg_pixels16_x2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void  ff_avg_pixels16_y2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+void ff_avg_pixels16_xy2_no_rnd_neon(uint8_t *block, const uint8_t *pixels,
+                                     ptrdiff_t line_size, int h);
+
+av_cold void ff_hpeldsp_init_aarch64(HpelDSPContext *c, int flags)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_neon(cpu_flags)) {
+        c->put_pixels_tab[0][0] = ff_put_pixels16_neon;
+        c->put_pixels_tab[0][1] = ff_put_pixels16_x2_neon;
+        c->put_pixels_tab[0][2] = ff_put_pixels16_y2_neon;
+        c->put_pixels_tab[0][3] = ff_put_pixels16_xy2_neon;
+        c->put_pixels_tab[1][0] = ff_put_pixels8_neon;
+        c->put_pixels_tab[1][1] = ff_put_pixels8_x2_neon;
+        c->put_pixels_tab[1][2] = ff_put_pixels8_y2_neon;
+        c->put_pixels_tab[1][3] = ff_put_pixels8_xy2_neon;
+
+        c->put_no_rnd_pixels_tab[0][0] = ff_put_pixels16_neon;
+        c->put_no_rnd_pixels_tab[0][1] = ff_put_pixels16_x2_no_rnd_neon;
+        c->put_no_rnd_pixels_tab[0][2] = ff_put_pixels16_y2_no_rnd_neon;
+        c->put_no_rnd_pixels_tab[0][3] = ff_put_pixels16_xy2_no_rnd_neon;
+        c->put_no_rnd_pixels_tab[1][0] = ff_put_pixels8_neon;
+        c->put_no_rnd_pixels_tab[1][1] = ff_put_pixels8_x2_no_rnd_neon;
+        c->put_no_rnd_pixels_tab[1][2] = ff_put_pixels8_y2_no_rnd_neon;
+        c->put_no_rnd_pixels_tab[1][3] = ff_put_pixels8_xy2_no_rnd_neon;
+
+        c->avg_pixels_tab[0][0] = ff_avg_pixels16_neon;
+        c->avg_pixels_tab[0][1] = ff_avg_pixels16_x2_neon;
+        c->avg_pixels_tab[0][2] = ff_avg_pixels16_y2_neon;
+        c->avg_pixels_tab[0][3] = ff_avg_pixels16_xy2_neon;
+        c->avg_pixels_tab[1][0] = ff_avg_pixels8_neon;
+        c->avg_pixels_tab[1][1] = ff_avg_pixels8_x2_neon;
+        c->avg_pixels_tab[1][2] = ff_avg_pixels8_y2_neon;
+        c->avg_pixels_tab[1][3] = ff_avg_pixels8_xy2_neon;
+
+        c->avg_no_rnd_pixels_tab[0] = ff_avg_pixels16_neon;
+        c->avg_no_rnd_pixels_tab[1] = ff_avg_pixels16_x2_no_rnd_neon;
+        c->avg_no_rnd_pixels_tab[2] = ff_avg_pixels16_y2_no_rnd_neon;
+        c->avg_no_rnd_pixels_tab[3] = ff_avg_pixels16_xy2_no_rnd_neon;
+    }
+}
diff --git a/libavcodec/aarch64/hpeldsp_neon.S b/libavcodec/aarch64/hpeldsp_neon.S
new file mode 100644
index 0000000000..29782908f8
--- /dev/null
+++ b/libavcodec/aarch64/hpeldsp_neon.S
@@ -0,0 +1,397 @@
+/*
+ * ARM NEON optimised DSP functions
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ * Copyright (c) 2013 Janne Grunau <janne-libav@jannau.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/aarch64/asm.S"
+
+.macro  pixels16        rnd=1, avg=0
+  .if \avg
+        mov             x12, x0
+  .endif
+1:      ld1             {v0.16B},  [x1], x2
+        ld1             {v1.16B},  [x1], x2
+        ld1             {v2.16B},  [x1], x2
+        ld1             {v3.16B},  [x1], x2
+  .if \avg
+        ld1             {v4.16B},  [x12], x2
+        urhadd          v0.16B,  v0.16B,  v4.16B
+        ld1             {v5.16B},  [x12], x2
+        urhadd          v1.16B,  v1.16B,  v5.16B
+        ld1             {v6.16B},  [x12], x2
+        urhadd          v2.16B,  v2.16B,  v6.16B
+        ld1             {v7.16B},  [x12], x2
+        urhadd          v3.16B,  v3.16B,  v7.16B
+  .endif
+        subs            w3,  w3,  #4
+        st1             {v0.16B},  [x0], x2
+        st1             {v1.16B},  [x0], x2
+        st1             {v2.16B},  [x0], x2
+        st1             {v3.16B},  [x0], x2
+        b.ne            1b
+        ret
+.endm
+
+.macro  pixels16_x2     rnd=1, avg=0
+1:      ld1             {v0.16B, v1.16B}, [x1], x2
+        ld1             {v2.16B, v3.16B}, [x1], x2
+        subs            w3,  w3,  #2
+        ext             v1.16B,  v0.16B,  v1.16B,  #1
+        avg             v0.16B,  v0.16B,  v1.16B
+        ext             v3.16B,  v2.16B,  v3.16B,  #1
+        avg             v2.16B,  v2.16B,  v3.16B
+  .if \avg
+        ld1             {v1.16B}, [x0], x2
+        ld1             {v3.16B}, [x0]
+        urhadd          v0.16B,  v0.16B,  v1.16B
+        urhadd          v2.16B,  v2.16B,  v3.16B
+        sub             x0,  x0,  x2
+  .endif
+        st1             {v0.16B}, [x0], x2
+        st1             {v2.16B}, [x0], x2
+        b.ne            1b
+        ret
+.endm
+
+.macro  pixels16_y2     rnd=1, avg=0
+        sub             w3,  w3,  #2
+        ld1             {v0.16B}, [x1], x2
+        ld1             {v1.16B}, [x1], x2
+1:      subs            w3,  w3,  #2
+        avg             v2.16B,  v0.16B,  v1.16B
+        ld1             {v0.16B}, [x1], x2
+        avg             v3.16B,  v0.16B,  v1.16B
+        ld1             {v1.16B}, [x1], x2
+  .if \avg
+        ld1             {v4.16B}, [x0], x2
+        ld1             {v5.16B}, [x0]
+        urhadd          v2.16B,  v2.16B,  v4.16B
+        urhadd          v3.16B,  v3.16B,  v5.16B
+        sub             x0,  x0,  x2
+  .endif
+        st1             {v2.16B}, [x0], x2
+        st1             {v3.16B}, [x0], x2
+        b.ne            1b
+
+        avg             v2.16B,  v0.16B,  v1.16B
+        ld1             {v0.16B}, [x1], x2
+        avg             v3.16B,  v0.16B,  v1.16B
+  .if \avg
+        ld1             {v4.16B}, [x0], x2
+        ld1             {v5.16B}, [x0]
+        urhadd          v2.16B,  v2.16B,  v4.16B
+        urhadd          v3.16B,  v3.16B,  v5.16B
+        sub             x0,  x0,  x2
+  .endif
+        st1             {v2.16B},     [x0], x2
+        st1             {v3.16B},     [x0], x2
+
+        ret
+.endm
+
+.macro  pixels16_xy2    rnd=1, avg=0
+        sub             w3,  w3,  #2
+        ld1             {v0.16B, v1.16B}, [x1], x2
+        ld1             {v4.16B, v5.16B}, [x1], x2
+NRND    movi            v26.8H, #1
+        ext             v1.16B,  v0.16B,  v1.16B,  #1
+        ext             v5.16B,  v4.16B,  v5.16B,  #1
+        uaddl           v16.8H,  v0.8B,   v1.8B
+        uaddl2          v20.8H,  v0.16B,  v1.16B
+        uaddl           v18.8H,  v4.8B,   v5.8B
+        uaddl2          v22.8H,  v4.16B,  v5.16B
+1:      subs            w3,  w3,  #2
+        ld1             {v0.16B, v1.16B}, [x1], x2
+        add             v24.8H,  v16.8H,  v18.8H
+NRND    add             v24.8H,  v24.8H,  v26.8H
+        ext             v30.16B, v0.16B,  v1.16B,  #1
+        add             v1.8H,   v20.8H,  v22.8H
+        mshrn           v28.8B,  v24.8H,  #2
+NRND    add             v1.8H,   v1.8H,   v26.8H
+        mshrn2          v28.16B, v1.8H,   #2
+  .if \avg
+        ld1             {v16.16B},        [x0]
+        urhadd          v28.16B, v28.16B, v16.16B
+  .endif
+        uaddl           v16.8H,  v0.8B,   v30.8B
+        ld1             {v2.16B, v3.16B}, [x1], x2
+        uaddl2          v20.8H,  v0.16B,  v30.16B
+        st1             {v28.16B},        [x0], x2
+        add             v24.8H,  v16.8H,  v18.8H
+NRND    add             v24.8H,  v24.8H,  v26.8H
+        ext             v3.16B,  v2.16B,  v3.16B,  #1
+        add             v0.8H,   v20.8H,  v22.8H
+        mshrn           v30.8B,  v24.8H,  #2
+NRND    add             v0.8H,   v0.8H,   v26.8H
+        mshrn2          v30.16B, v0.8H,   #2
+  .if \avg
+        ld1             {v18.16B},        [x0]
+        urhadd          v30.16B, v30.16B, v18.16B
+  .endif
+        uaddl           v18.8H,   v2.8B,  v3.8B
+        uaddl2          v22.8H,   v2.16B, v3.16B
+        st1             {v30.16B},        [x0], x2
+        b.gt            1b
+
+        ld1             {v0.16B, v1.16B}, [x1], x2
+        add             v24.8H,  v16.8H,  v18.8H
+NRND    add             v24.8H,  v24.8H,  v26.8H
+        ext             v30.16B, v0.16B,  v1.16B,  #1
+        add             v1.8H,   v20.8H,  v22.8H
+        mshrn           v28.8B,  v24.8H,  #2
+NRND    add             v1.8H,   v1.8H,   v26.8H
+        mshrn2          v28.16B, v1.8H,   #2
+  .if \avg
+        ld1             {v16.16B},        [x0]
+        urhadd          v28.16B, v28.16B, v16.16B
+  .endif
+        uaddl           v16.8H,  v0.8B,   v30.8B
+        uaddl2          v20.8H,  v0.16B,  v30.16B
+        st1             {v28.16B},        [x0], x2
+        add             v24.8H,  v16.8H,  v18.8H
+NRND    add             v24.8H,  v24.8H,  v26.8H
+        add             v0.8H,   v20.8H,  v22.8H
+        mshrn           v30.8B,  v24.8H,  #2
+NRND    add             v0.8H,   v0.8H,   v26.8H
+        mshrn2          v30.16B, v0.8H,   #2
+  .if \avg
+        ld1             {v18.16B},        [x0]
+        urhadd          v30.16B, v30.16B, v18.16B
+  .endif
+        st1             {v30.16B},        [x0], x2
+
+        ret
+.endm
+
+.macro  pixels8         rnd=1, avg=0
+1:      ld1             {v0.8B}, [x1], x2
+        ld1             {v1.8B}, [x1], x2
+        ld1             {v2.8B}, [x1], x2
+        ld1             {v3.8B}, [x1], x2
+  .if \avg
+        ld1             {v4.8B}, [x0], x2
+        urhadd          v0.8B,  v0.8B,  v4.8B
+        ld1             {v5.8B}, [x0], x2
+        urhadd          v1.8B,  v1.8B,  v5.8B
+        ld1             {v6.8B}, [x0], x2
+        urhadd          v2.8B,  v2.8B,  v6.8B
+        ld1             {v7.8B}, [x0], x2
+        urhadd          v3.8B,  v3.8B,  v7.8B
+        sub             x0,  x0,  x2,  lsl #2
+  .endif
+        subs            w3,  w3,  #4
+        st1             {v0.8B}, [x0], x2
+        st1             {v1.8B}, [x0], x2
+        st1             {v2.8B}, [x0], x2
+        st1             {v3.8B}, [x0], x2
+        b.ne            1b
+        ret
+.endm
+
+.macro  pixels8_x2      rnd=1, avg=0
+1:      ld1             {v0.8B, v1.8B}, [x1], x2
+        ext             v1.8B,  v0.8B,  v1.8B,  #1
+        ld1             {v2.8B, v3.8B}, [x1], x2
+        ext             v3.8B,  v2.8B,  v3.8B,  #1
+        subs            w3,  w3,  #2
+        avg             v0.8B,   v0.8B,   v1.8B
+        avg             v2.8B,   v2.8B,   v3.8B
+  .if \avg
+        ld1             {v4.8B},     [x0], x2
+        ld1             {v5.8B},     [x0]
+        urhadd          v0.8B,   v0.8B,   v4.8B
+        urhadd          v2.8B,   v2.8B,   v5.8B
+        sub             x0,  x0,  x2
+  .endif
+        st1             {v0.8B}, [x0], x2
+        st1             {v2.8B}, [x0], x2
+        b.ne            1b
+        ret
+.endm
+
+.macro  pixels8_y2      rnd=1, avg=0
+        sub             w3,  w3,  #2
+        ld1             {v0.8B},  [x1], x2
+        ld1             {v1.8B},  [x1], x2
+1:      subs            w3,  w3,  #2
+        avg             v4.8B,  v0.8B,  v1.8B
+        ld1             {v0.8B},  [x1], x2
+        avg             v5.8B,  v0.8B,  v1.8B
+        ld1             {v1.8B},  [x1], x2
+  .if \avg
+        ld1             {v2.8B},     [x0], x2
+        ld1             {v3.8B},     [x0]
+        urhadd          v4.8B,  v4.8B,  v2.8B
+        urhadd          v5.8B,  v5.8B,  v3.8B
+        sub             x0,  x0,  x2
+  .endif
+        st1             {v4.8B},     [x0], x2
+        st1             {v5.8B},     [x0], x2
+        b.ne            1b
+
+        avg             v4.8B,  v0.8B,  v1.8B
+        ld1             {v0.8B},  [x1], x2
+        avg             v5.8B,  v0.8B,  v1.8B
+  .if \avg
+        ld1             {v2.8B},     [x0], x2
+        ld1             {v3.8B},     [x0]
+        urhadd          v4.8B,  v4.8B,  v2.8B
+        urhadd          v5.8B,  v5.8B,  v3.8B
+        sub             x0,  x0,  x2
+  .endif
+        st1             {v4.8B},     [x0], x2
+        st1             {v5.8B},     [x0], x2
+
+        ret
+.endm
+
+.macro  pixels8_xy2     rnd=1, avg=0
+        sub             w3,  w3,  #2
+        ld1             {v0.16B},     [x1], x2
+        ld1             {v1.16B},     [x1], x2
+NRND    movi            v19.8H, #1
+        ext             v4.16B,  v0.16B,  v4.16B,  #1
+        ext             v6.16B,  v1.16B,  v6.16B,  #1
+        uaddl           v16.8H,  v0.8B,  v4.8B
+        uaddl           v17.8H,  v1.8B,  v6.8B
+1:      subs            w3,  w3,  #2
+        ld1             {v0.16B},     [x1], x2
+        add             v18.8H, v16.8H,  v17.8H
+        ext             v4.16B,  v0.16B,  v4.16B,  #1
+NRND    add             v18.8H, v18.8H, v19.8H
+        uaddl           v16.8H,  v0.8B,  v4.8B
+        mshrn           v5.8B,  v18.8H, #2
+        ld1             {v1.16B},     [x1], x2
+        add             v18.8H, v16.8H,  v17.8H
+  .if \avg
+        ld1             {v7.8B},     [x0]
+        urhadd          v5.8B,  v5.8B,  v7.8B
+  .endif
+NRND    add             v18.8H, v18.8H, v19.8H
+        st1             {v5.8B},     [x0], x2
+        mshrn           v7.8B,  v18.8H, #2
+  .if \avg
+        ld1             {v5.8B},     [x0]
+        urhadd          v7.8B,  v7.8B,  v5.8B
+  .endif
+        ext             v6.16B,  v1.16B,  v6.16B,  #1
+        uaddl           v17.8H,  v1.8B,   v6.8B
+        st1             {v7.8B},     [x0], x2
+        b.gt            1b
+
+        ld1             {v0.16B},     [x1], x2
+        add             v18.8H, v16.8H, v17.8H
+        ext             v4.16B, v0.16B, v4.16B,  #1
+NRND    add             v18.8H, v18.8H, v19.8H
+        uaddl           v16.8H,  v0.8B, v4.8B
+        mshrn           v5.8B,  v18.8H, #2
+        add             v18.8H, v16.8H, v17.8H
+  .if \avg
+        ld1             {v7.8B},     [x0]
+        urhadd          v5.8B,  v5.8B,  v7.8B
+  .endif
+NRND    add             v18.8H, v18.8H, v19.8H
+        st1             {v5.8B},     [x0], x2
+        mshrn           v7.8B,  v18.8H, #2
+  .if \avg
+        ld1             {v5.8B},     [x0]
+        urhadd          v7.8B,  v7.8B,  v5.8B
+  .endif
+        st1             {v7.8B},     [x0], x2
+
+        ret
+.endm
+
+.macro  pixfunc         pfx, name, suf, rnd=1, avg=0
+  .if \rnd
+    .macro avg  rd, rn, rm
+        urhadd          \rd, \rn, \rm
+    .endm
+    .macro mshrn rd, rn, rm
+        rshrn           \rd, \rn, \rm
+    .endm
+    .macro mshrn2 rd, rn, rm
+        rshrn2          \rd, \rn, \rm
+    .endm
+    .macro NRND insn:vararg
+    .endm
+  .else
+    .macro avg  rd, rn, rm
+        uhadd           \rd, \rn, \rm
+    .endm
+    .macro mshrn rd, rn, rm
+        shrn            \rd, \rn, \rm
+    .endm
+    .macro mshrn2 rd, rn, rm
+        shrn2           \rd, \rn, \rm
+    .endm
+    .macro NRND insn:vararg
+        \insn
+    .endm
+  .endif
+function ff_\pfx\name\suf\()_neon, export=1
+        \name           \rnd, \avg
+endfunc
+        .purgem         avg
+        .purgem         mshrn
+        .purgem         mshrn2
+        .purgem         NRND
+.endm
+
+.macro  pixfunc2        pfx, name, avg=0
+        pixfunc         \pfx, \name,          rnd=1, avg=\avg
+        pixfunc         \pfx, \name, _no_rnd, rnd=0, avg=\avg
+.endm
+
+function ff_put_h264_qpel16_mc00_neon, export=1
+        mov             w3,  #16
+endfunc
+
+        pixfunc         put_, pixels16,     avg=0
+        pixfunc2        put_, pixels16_x2,  avg=0
+        pixfunc2        put_, pixels16_y2,  avg=0
+        pixfunc2        put_, pixels16_xy2, avg=0
+
+function ff_avg_h264_qpel16_mc00_neon, export=1
+        mov             w3,  #16
+endfunc
+
+        pixfunc         avg_, pixels16,     avg=1
+        pixfunc2        avg_, pixels16_x2,  avg=1
+        pixfunc2        avg_, pixels16_y2,  avg=1
+        pixfunc2        avg_, pixels16_xy2, avg=1
+
+function ff_put_h264_qpel8_mc00_neon, export=1
+        mov             w3,  #8
+endfunc
+
+        pixfunc         put_, pixels8,     avg=0
+        pixfunc2        put_, pixels8_x2,  avg=0
+        pixfunc2        put_, pixels8_y2,  avg=0
+        pixfunc2        put_, pixels8_xy2, avg=0
+
+function ff_avg_h264_qpel8_mc00_neon, export=1
+        mov             w3,  #8
+endfunc
+
+        pixfunc         avg_, pixels8,     avg=1
+        pixfunc         avg_, pixels8_x2,  avg=1
+        pixfunc         avg_, pixels8_y2,  avg=1
+        pixfunc         avg_, pixels8_xy2, avg=1