aarch64: h264 qpel NEON optimizations

Ported from ARMv7 NEON.

libavcodec/aarch64/Makefile

 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_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

libavcodec/aarch64/h264qpel_init_aarch64.c

+/*
+ * 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 <stdint.h>
+
+#include "config.h"
+#include "libavutil/attributes.h"
+#include "libavutil/cpu.h"
+#include "libavutil/aarch64/cpu.h"
+#include "libavcodec/h264qpel.h"
+
+void ff_put_h264_qpel16_mc00_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc10_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc20_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc30_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc01_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc11_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc21_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc31_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc02_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc12_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc22_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc32_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc03_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc13_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc23_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel16_mc33_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+
+void ff_put_h264_qpel8_mc00_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc10_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc20_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc30_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc01_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc11_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc21_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc31_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc02_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc12_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc22_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc32_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc03_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc13_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc23_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_put_h264_qpel8_mc33_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+
+void ff_avg_h264_qpel16_mc00_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc10_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc20_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc30_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc01_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc11_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc21_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc31_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc02_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc12_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc22_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc32_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc03_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc13_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc23_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel16_mc33_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+
+void ff_avg_h264_qpel8_mc00_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc10_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc20_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc30_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc01_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc11_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc21_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc31_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc02_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc12_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc22_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc32_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc03_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc13_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc23_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+void ff_avg_h264_qpel8_mc33_neon(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
+
+av_cold void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth)
+{
+    const int high_bit_depth = bit_depth > 8;
+    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][ 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;
+        c->put_h264_qpel_pixels_tab[0][ 4] = ff_put_h264_qpel16_mc01_neon;
+        c->put_h264_qpel_pixels_tab[0][ 5] = ff_put_h264_qpel16_mc11_neon;
+        c->put_h264_qpel_pixels_tab[0][ 6] = ff_put_h264_qpel16_mc21_neon;
+        c->put_h264_qpel_pixels_tab[0][ 7] = ff_put_h264_qpel16_mc31_neon;
+        c->put_h264_qpel_pixels_tab[0][ 8] = ff_put_h264_qpel16_mc02_neon;
+        c->put_h264_qpel_pixels_tab[0][ 9] = ff_put_h264_qpel16_mc12_neon;
+        c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_neon;
+        c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_neon;
+        c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_neon;
+        c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_neon;
+        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][ 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;
+        c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_neon;
+        c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_neon;
+        c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_neon;
+        c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_neon;
+        c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_neon;
+        c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_neon;
+        c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_neon;
+        c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_neon;
+        c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_neon;
+        c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_neon;
+        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][ 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;
+        c->avg_h264_qpel_pixels_tab[0][ 4] = ff_avg_h264_qpel16_mc01_neon;
+        c->avg_h264_qpel_pixels_tab[0][ 5] = ff_avg_h264_qpel16_mc11_neon;
+        c->avg_h264_qpel_pixels_tab[0][ 6] = ff_avg_h264_qpel16_mc21_neon;
+        c->avg_h264_qpel_pixels_tab[0][ 7] = ff_avg_h264_qpel16_mc31_neon;
+        c->avg_h264_qpel_pixels_tab[0][ 8] = ff_avg_h264_qpel16_mc02_neon;
+        c->avg_h264_qpel_pixels_tab[0][ 9] = ff_avg_h264_qpel16_mc12_neon;
+        c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_neon;
+        c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_neon;
+        c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_neon;
+        c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_neon;
+        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][ 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;
+        c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_neon;
+        c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_neon;
+        c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_neon;
+        c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_neon;
+        c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_neon;
+        c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_neon;
+        c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_neon;
+        c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_neon;
+        c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_neon;
+        c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_neon;
+        c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_neon;
+        c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
+    }
+}

libavcodec/aarch64/neon.S

@@ -16,6 +16,70 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+.macro  transpose_8x8B  r0, r1, r2, r3, r4, r5, r6, r7, r8, r9
+        trn1            \r8\().8B,  \r0\().8B,  \r1\().8B
+        trn2            \r9\().8B,  \r0\().8B,  \r1\().8B
+        trn1            \r1\().8B,  \r2\().8B,  \r3\().8B
+        trn2            \r3\().8B,  \r2\().8B,  \r3\().8B
+        trn1            \r0\().8B,  \r4\().8B,  \r5\().8B
+        trn2            \r5\().8B,  \r4\().8B,  \r5\().8B
+        trn1            \r2\().8B,  \r6\().8B,  \r7\().8B
+        trn2            \r7\().8B,  \r6\().8B,  \r7\().8B
+
+        trn1            \r4\().4H,  \r0\().4H,  \r2\().4H
+        trn2            \r2\().4H,  \r0\().4H,  \r2\().4H
+        trn1            \r6\().4H,  \r5\().4H,  \r7\().4H
+        trn2            \r7\().4H,  \r5\().4H,  \r7\().4H
+        trn1            \r5\().4H,  \r9\().4H,  \r3\().4H
+        trn2            \r9\().4H,  \r9\().4H,  \r3\().4H
+        trn1            \r3\().4H,  \r8\().4H,  \r1\().4H
+        trn2            \r8\().4H,  \r8\().4H,  \r1\().4H
+
+        trn1            \r0\().2S,  \r3\().2S,  \r4\().2S
+        trn2            \r4\().2S,  \r3\().2S,  \r4\().2S
+
+        trn1            \r1\().2S,  \r5\().2S,  \r6\().2S
+        trn2            \r5\().2S,  \r5\().2S,  \r6\().2S
+
+        trn2            \r6\().2S,  \r8\().2S,  \r2\().2S
+        trn1            \r2\().2S,  \r8\().2S,  \r2\().2S
+
+        trn1            \r3\().2S,  \r9\().2S,  \r7\().2S
+        trn2            \r7\().2S,  \r9\().2S,  \r7\().2S
+.endm
+
+.macro  transpose_8x16B r0, r1, r2, r3, r4, r5, r6, r7, t0, t1
+        trn1            \t0\().16B, \r0\().16B, \r1\().16B
+        trn2            \t1\().16B, \r0\().16B, \r1\().16B
+        trn1            \r1\().16B, \r2\().16B, \r3\().16B
+        trn2            \r3\().16B, \r2\().16B, \r3\().16B
+        trn1            \r0\().16B, \r4\().16B, \r5\().16B
+        trn2            \r5\().16B, \r4\().16B, \r5\().16B
+        trn1            \r2\().16B, \r6\().16B, \r7\().16B
+        trn2            \r7\().16B, \r6\().16B, \r7\().16B
+
+        trn1            \r4\().8H,  \r0\().8H,  \r2\().8H
+        trn2            \r2\().8H,  \r0\().8H,  \r2\().8H
+        trn1            \r6\().8H,  \r5\().8H,  \r7\().8H
+        trn2            \r7\().8H,  \r5\().8H,  \r7\().8H
+        trn1            \r5\().8H,  \t1\().8H,  \r3\().8H
+        trn2            \t1\().8H,  \t1\().8H,  \r3\().8H
+        trn1            \r3\().8H,  \t0\().8H,  \r1\().8H
+        trn2            \t0\().8H,  \t0\().8H,  \r1\().8H
+
+        trn1            \r0\().4S,  \r3\().4S,  \r4\().4S
+        trn2            \r4\().4S,  \r3\().4S,  \r4\().4S
+
+        trn1            \r1\().4S,  \r5\().4S,  \r6\().4S
+        trn2            \r5\().4S,  \r5\().4S,  \r6\().4S
+
+        trn2            \r6\().4S,  \t0\().4S,  \r2\().4S
+        trn1            \r2\().4S,  \t0\().4S,  \r2\().4S
+
+        trn1            \r3\().4S,  \t1\().4S,  \r7\().4S
+        trn2            \r7\().4S,  \t1\().4S,  \r7\().4S
+.endm
+
 .macro  transpose_4x4H  r0, r1, r2, r3, r4, r5, r6, r7
         trn1            \r4\().4H,  \r0\().4H,  \r1\().4H
         trn2            \r5\().4H,  \r0\().4H,  \r1\().4H

libavcodec/h264qpel.c

@@ -78,6 +78,8 @@ av_cold void ff_h264qpel_init(H264QpelContext *c, int bit_depth)
         break;
     }
 
+    if (ARCH_AARCH64)
+        ff_h264qpel_init_aarch64(c, bit_depth);
     if (ARCH_ARM)
         ff_h264qpel_init_arm(c, bit_depth);
     if (ARCH_PPC)

libavcodec/h264qpel.h

@@ -31,6 +31,7 @@ typedef struct H264QpelContext {
 
 void ff_h264qpel_init(H264QpelContext *c, int bit_depth);
 
+void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth);
 void ff_h264qpel_init_arm(H264QpelContext *c, int bit_depth);
 void ff_h264qpel_init_ppc(H264QpelContext *c, int bit_depth);
 void ff_h264qpel_init_x86(H264QpelContext *c, int bit_depth);