[x264-devel] ppc: Rework satd_4* likewise

[Luca Barbato]

x264 | branch: master | Luca Barbato <lu_zero at gentoo.org> | Sun Aug 19 09:28:42 2018 +0200| [83acefef8990302caf962c77e5a8189bb620ca6f] | committer: Anton Mitrofanov

ppc: Rework satd_4* likewise

Now 4x4 is as slow as C and 4x8 is a 2% faster than before.

> http://git.videolan.org/gitweb.cgi/x264.git/?a=commit;h=83acefef8990302caf962c77e5a8189bb620ca6f
---

 common/ppc/pixel.c | 34 ++++++++++++++++++++++------------
 1 file changed, 22 insertions(+), 12 deletions(-)

diff --git a/common/ppc/pixel.c b/common/ppc/pixel.c
index de1877a0..5bb2bec6 100644
--- a/common/ppc/pixel.c
+++ b/common/ppc/pixel.c
@@ -114,6 +114,23 @@ PIXEL_SAD_ALTIVEC( pixel_sad_8x8_altivec,   8,  8,  2s, 1 )
     VEC_ABS( a );          \
     c = vec_sum4s( a, b )
 
+static ALWAYS_INLINE vec_s32_t add_abs_4( vec_s16_t a, vec_s16_t b,
+                                          vec_s16_t c, vec_s16_t d )
+{
+    vec_s16_t t0 = vec_abs( a );
+    vec_s16_t t1 = vec_abs( b );
+    vec_s16_t t2 = vec_abs( c );
+    vec_s16_t t3 = vec_abs( d );
+
+    vec_s16_t s0 = vec_adds( t0, t1 );
+    vec_s16_t s1 = vec_adds( t2, t3 );
+
+    vec_s32_t s01 = vec_sum4s( s0, vec_splat_s32( 0 ) );
+    vec_s32_t s23 = vec_sum4s( s1, vec_splat_s32( 0 ) );
+
+    return vec_add( s01, s23 );
+}
+
 /***********************************************************************
  * SATD 4x4
  **********************************************************************/
@@ -142,10 +159,7 @@ static int pixel_satd_4x4_altivec( uint8_t *pix1, intptr_t i_pix1,
     VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                  temp0v, temp1v, temp2v, temp3v );
 
-    VEC_ADD_ABS( temp0v, zero_s32v, satdv );
-    VEC_ADD_ABS( temp1v, satdv,     satdv );
-    VEC_ADD_ABS( temp2v, satdv,     satdv );
-    VEC_ADD_ABS( temp3v, satdv,     satdv );
+    satdv = add_abs_4( temp0v, temp1v, temp2v, temp3v );
 
     satdv = vec_sum2s( satdv, zero_s32v );
     satdv = vec_splat( satdv, 1 );
@@ -177,10 +191,8 @@ static int pixel_satd_4x8_altivec( uint8_t *pix1, intptr_t i_pix1,
                      diff0v, diff1v, diff2v, diff3v );
     VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                  temp0v, temp1v, temp2v, temp3v );
-    VEC_ADD_ABS( temp0v, zero_s32v, satdv );
-    VEC_ADD_ABS( temp1v, satdv,     satdv );
-    VEC_ADD_ABS( temp2v, satdv,     satdv );
-    VEC_ADD_ABS( temp3v, satdv,     satdv );
+
+    satdv = add_abs_4( temp0v, temp1v, temp2v, temp3v );
 
     VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 4, diff0v );
     VEC_DIFF_H( pix1, i_pix1, pix2, i_pix2, 4, diff1v );
@@ -192,10 +204,8 @@ static int pixel_satd_4x8_altivec( uint8_t *pix1, intptr_t i_pix1,
                      diff0v, diff1v, diff2v, diff3v );
     VEC_HADAMAR( diff0v, diff1v, diff2v, diff3v,
                  temp0v, temp1v, temp2v, temp3v );
-    VEC_ADD_ABS( temp0v, satdv,     satdv );
-    VEC_ADD_ABS( temp1v, satdv,     satdv );
-    VEC_ADD_ABS( temp2v, satdv,     satdv );
-    VEC_ADD_ABS( temp3v, satdv,     satdv );
+
+    satdv = vec_add( satdv, add_abs_4( temp0v, temp1v, temp2v, temp3v ) );
 
     satdv = vec_sum2s( satdv, zero_s32v );
     satdv = vec_splat( satdv, 1 );