aarch64: h264 loop filter NEON optimizations

Ported from ARMv7 NEON.

libavcodec/aarch64/Makefile

@@ -6,7 +6,8 @@ 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-$(CONFIG_H264DSP)             += aarch64/h264dsp_neon.o              \
+                                           aarch64/h264idct_neon.o
 NEON-OBJS-$(CONFIG_H264QPEL)            += aarch64/h264qpel_neon.o             \
                                            aarch64/hpeldsp_neon.o
 NEON-OBJS-$(CONFIG_HPELDSP)             += aarch64/hpeldsp_neon.o

libavcodec/aarch64/h264dsp_init_aarch64.c

@@ -25,6 +25,15 @@
 #include "libavutil/aarch64/cpu.h"
 #include "libavcodec/h264dsp.h"
 
+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_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_idct_add_neon(uint8_t *dst, int16_t *block, int stride);
 void ff_h264_idct_dc_add_neon(uint8_t *dst, int16_t *block, int stride);
 void ff_h264_idct_add16_neon(uint8_t *dst, const int *block_offset,
@@ -49,6 +58,11 @@ av_cold void ff_h264dsp_init_aarch64(H264DSPContext *c, const int bit_depth,
     int cpu_flags = av_get_cpu_flags();
 
     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_chroma = ff_h264_v_loop_filter_chroma_neon;
+        c->h264_h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
+
         c->h264_idct_add        = ff_h264_idct_add_neon;
         c->h264_idct_dc_add     = ff_h264_idct_dc_add_neon;
         c->h264_idct_add16      = ff_h264_idct_add16_neon;

libavcodec/aarch64/neon.S

@@ -80,6 +80,30 @@
         trn2            \r7\().4S,  \t1\().4S,  \r7\().4S
 .endm
 
+.macro  transpose_4x16B r0, r1, r2, r3, t4, t5, t6, t7
+        trn1            \t4\().16B, \r0\().16B,  \r1\().16B
+        trn2            \t5\().16B, \r0\().16B,  \r1\().16B
+        trn1            \t6\().16B, \r2\().16B,  \r3\().16B
+        trn2            \t7\().16B, \r2\().16B,  \r3\().16B
+
+        trn1            \r0\().8H,  \t4\().8H,  \t6\().8H
+        trn2            \r2\().8H,  \t4\().8H,  \t6\().8H
+        trn1            \r1\().8H,  \t5\().8H,  \t7\().8H
+        trn2            \r3\().8H,  \t5\().8H,  \t7\().8H
+.endm
+
+.macro  transpose_4x8B  r0, r1, r2, r3, t4, t5, t6, t7
+        trn1            \t4\().8B,  \r0\().8B,  \r1\().8B
+        trn2            \t5\().8B,  \r0\().8B,  \r1\().8B
+        trn1            \t6\().8B,  \r2\().8B,  \r3\().8B
+        trn2            \t7\().8B,  \r2\().8B,  \r3\().8B
+
+        trn1            \r0\().4H,  \t4\().4H,  \t6\().4H
+        trn2            \r2\().4H,  \t4\().4H,  \t6\().4H
+        trn1            \r1\().4H,  \t5\().4H,  \t7\().4H
+        trn2            \r3\().4H,  \t5\().4H,  \t7\().4H
+.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