[x264-devel] [PATCH 2/2] arm: implement x264_pixel_var2_8x16_neon

[Janne Grunau]

checkasm --bench on a cortex-a9:
var2_8x16_c: 5677
var2_8x16_neon: 1421
---
 common/arm/pixel-a.S | 43 +++++++++++++++++++++++++++++++++++++++++++
 common/arm/pixel.h   |  3 ++-
 common/pixel.c       |  1 +
 3 files changed, 46 insertions(+), 1 deletion(-)

diff --git a/common/arm/pixel-a.S b/common/arm/pixel-a.S
index 6bc904d..0b996a8 100644
--- a/common/arm/pixel-a.S
+++ b/common/arm/pixel-a.S
@@ -635,6 +635,49 @@ function x264_pixel_var2_8x8_neon
     bx              lr
 .endfunc
 
+function x264_pixel_var2_8x16_neon
+    vld1.64         {d16}, [r0,:64], r1
+    vld1.64         {d17}, [r2,:64], r3
+    vld1.64         {d18}, [r0,:64], r1
+    vld1.64         {d19}, [r2,:64], r3
+    vsubl.u8        q10, d16, d17
+    vsubl.u8        q11, d18, d19
+    SQR_ACC         q1,  d20, d21,  vmull.s16
+    vld1.64         {d16}, [r0,:64], r1
+    vadd.s16        q0,  q10, q11
+    vld1.64         {d17}, [r2,:64], r3
+    SQR_ACC         q2,  d22, d23,  vmull.s16
+    mov             ip,  #14
+1:  subs            ip,  ip,  #2
+    vld1.64         {d18}, [r0,:64], r1
+    vsubl.u8        q10, d16, d17
+    vld1.64         {d19}, [r2,:64], r3
+    vadd.s16        q0,  q0,  q10
+    SQR_ACC         q1,  d20, d21
+    vsubl.u8        q11, d18, d19
+    beq             2f
+    vld1.64         {d16}, [r0,:64], r1
+    vadd.s16        q0,  q0,  q11
+    vld1.64         {d17}, [r2,:64], r3
+    SQR_ACC         q2,  d22, d23
+    b               1b
+2:
+    vadd.s16        q0,  q0,  q11
+    SQR_ACC         q2,  d22, d23
+
+    ldr             ip,  [sp]
+    vadd.s16        d0,  d0,  d1
+    vadd.s32        q1,  q1,  q2
+    vpaddl.s16      d0,  d0
+    vadd.s32        d1,  d2,  d3
+    vpadd.s32       d0,  d0,  d1
+
+    vmov            r0,  r1,  d0
+    vst1.32         {d0[1]}, [ip,:32]
+    mul             r0,  r0,  r0
+    sub             r0,  r1,  r0,  lsr #7
+    bx              lr
+.endfunc
 
 .macro LOAD_DIFF_8x4 q0 q1 q2 q3
     vld1.32     {d1}, [r2], r3
diff --git a/common/arm/pixel.h b/common/arm/pixel.h
index acd517b..c55ed9a 100644
--- a/common/arm/pixel.h
+++ b/common/arm/pixel.h
@@ -58,7 +58,8 @@ int x264_pixel_sa8d_16x16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t );
 uint64_t x264_pixel_var_8x8_neon  ( uint8_t *, intptr_t );
 uint64_t x264_pixel_var_8x16_neon ( uint8_t *, intptr_t );
 uint64_t x264_pixel_var_16x16_neon( uint8_t *, intptr_t );
-int x264_pixel_var2_8x8_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
+int x264_pixel_var2_8x8_neon ( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
+int x264_pixel_var2_8x16_neon( uint8_t *, intptr_t, uint8_t *, intptr_t, int * );
 
 uint64_t x264_pixel_hadamard_ac_8x8_neon  ( uint8_t *, intptr_t );
 uint64_t x264_pixel_hadamard_ac_8x16_neon ( uint8_t *, intptr_t );
diff --git a/common/pixel.c b/common/pixel.c
index d362879..e16f292 100644
--- a/common/pixel.c
+++ b/common/pixel.c
@@ -1350,6 +1350,7 @@ void x264_pixel_init( int cpu, x264_pixel_function_t *pixf )
         pixf->var[PIXEL_8x16]   = x264_pixel_var_8x16_neon;
         pixf->var[PIXEL_16x16]  = x264_pixel_var_16x16_neon;
         pixf->var2[PIXEL_8x8]   = x264_pixel_var2_8x8_neon;
+        pixf->var2[PIXEL_8x16]  = x264_pixel_var2_8x16_neon;
 
         pixf->ssim_4x4x2_core   = x264_pixel_ssim_4x4x2_core_neon;
         pixf->ssim_end4         = x264_pixel_ssim_end4_neon;
-- 
1.9.0