arm: add missing macro instantiation for x264_pixel_avg_4x16_neon

checkasm --bench on a cortex-a9:
avg_4x16_c: 8910
avg_4x16_neon: 2091

diff --git a/common/arm/mc-a.S b/common/arm/mc-a.S
index 90f16116ebca1646352ca201e24f364e138bd1ef..6267e35010c24d745775df8e9679f248ff36c4bc 100644 (file)
--- a/common/arm/mc-a.S
+++ b/common/arm/mc-a.S
@@ -181,6 +181,7 @@ function x264_pixel_avg_\w\()x\h\()_neon
 AVGH  4, 2
 AVGH  4, 4
 AVGH  4, 8
+AVGH  4, 16
 AVGH  8, 4
 AVGH  8, 8
 AVGH  8, 16

diff --git a/common/arm/mc-c.c b/common/arm/mc-c.c
index 16ea1c5f893c7d908491c907dbed8dcc1ce5f0a6..d68369f75511c85ed1565dc2b51b1f2f52ab0579 100644 (file)
--- a/common/arm/mc-c.c
+++ b/common/arm/mc-c.c
@@ -37,6 +37,7 @@ void x264_pixel_avg_16x8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_
 void x264_pixel_avg_8x16_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
 void x264_pixel_avg_8x8_neon  ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
 void x264_pixel_avg_8x4_neon  ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
+void x264_pixel_avg_4x16_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
 void x264_pixel_avg_4x8_neon  ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
 void x264_pixel_avg_4x4_neon  ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
 void x264_pixel_avg_4x2_neon  ( uint8_t *, intptr_t, uint8_t *, intptr_t, uint8_t *, intptr_t, int );
@@ -229,6 +230,7 @@ void x264_mc_init_arm( int cpu, x264_mc_functions_t *pf )
     pf->avg[PIXEL_8x16]  = x264_pixel_avg_8x16_neon;
     pf->avg[PIXEL_8x8]   = x264_pixel_avg_8x8_neon;
     pf->avg[PIXEL_8x4]   = x264_pixel_avg_8x4_neon;
+    pf->avg[PIXEL_4x16]  = x264_pixel_avg_4x16_neon;
     pf->avg[PIXEL_4x8]   = x264_pixel_avg_4x8_neon;
     pf->avg[PIXEL_4x4]   = x264_pixel_avg_4x4_neon;
     pf->avg[PIXEL_4x2]   = x264_pixel_avg_4x2_neon;