[x264-devel] commit: Add AltiVec implementation of predict_8x8c_p. 2. 6x faster than scalar C. (Guillaume Poirier )

[git version control]

x264 | branch: master | Guillaume Poirier <gpoirier at mplayerhq.hu> | Sun Jan 18 22:44:14 2009 +0100| [09e76c903d3419619ed326a4dd114369a55bdd6e] | committer: Guillaume Poirier 

Add AltiVec implementation of predict_8x8c_p. 2.6x faster than scalar C.

> http://git.videolan.org/gitweb.cgi/x264.git/?a=commit;h=09e76c903d3419619ed326a4dd114369a55bdd6e
---

 common/ppc/predict.c |   56 +++++++++++++++++++++++++++++++++++++++++++++++++-
 common/ppc/predict.h |    1 +
 common/predict.c     |    7 ++++++
 3 files changed, 63 insertions(+), 1 deletions(-)

diff --git a/common/ppc/predict.c b/common/ppc/predict.c
index 589b00c..9b38fbb 100644
--- a/common/ppc/predict.c
+++ b/common/ppc/predict.c
@@ -23,6 +23,56 @@
 #include "pixel.h"
 #include "ppccommon.h"
 
+static void predict_8x8c_p_altivec( uint8_t *src )
+{
+    int i;
+    int a, b, c;
+    int H = 0;
+    int V = 0;
+    int i00;
+
+    for( i = 0; i < 4; i++ )
+    {
+        H += ( i + 1 ) * ( src[4+i - FDEC_STRIDE] - src[2 - i -FDEC_STRIDE] );
+        V += ( i + 1 ) * ( src[-1 +(i+4)*FDEC_STRIDE] - src[-1+(2-i)*FDEC_STRIDE] );
+    }
+
+    a = 16 * ( src[-1+7*FDEC_STRIDE] + src[7 - FDEC_STRIDE] );
+    b = ( 17 * H + 16 ) >> 5;
+    c = ( 17 * V + 16 ) >> 5;
+    i00 = a -3*b -3*c + 16;
+
+    vec_s16_u i00_u, b_u, c_u;
+    i00_u.s[0] = i00;
+    b_u.s[0]   = b;
+    c_u.s[0]   = c;
+
+    vec_u16_t val5_v = vec_splat_u16(5);
+    vec_s16_t i00_v, b_v, c_v;
+    i00_v = vec_splat(i00_u.v, 0);
+    b_v = vec_splat(b_u.v, 0);
+    c_v = vec_splat(c_u.v, 0);
+
+    vec_s16_t induc_v  = (vec_s16_t) CV(0, 1, 2, 3, 4, 5, 6, 7);
+    vec_s32_t mule_b_v = vec_mule(induc_v, b_v);
+    vec_s32_t mulo_b_v = vec_mulo(induc_v, b_v);
+    vec_s16_t mul_b_induc0_v = vec_pack(vec_mergeh(mule_b_v, mulo_b_v), vec_mergel(mule_b_v, mulo_b_v));
+    vec_s16_t add_i0_b_0v = vec_adds(i00_v, mul_b_induc0_v);
+
+    PREP_STORE8;
+
+    for( i = 0; i < 8; ++i )
+    {
+        vec_s16_t shift_0_v = vec_sra(add_i0_b_0v, val5_v);
+        vec_u8_t com_sat_v = vec_packsu(shift_0_v, shift_0_v);
+        VEC_STORE8(com_sat_v, &src[0]);
+        src += FDEC_STRIDE;
+        add_i0_b_0v = vec_adds(add_i0_b_0v, c_v);
+
+    }
+}
+
+
 /****************************************************************************
  * 16x16 prediction for intra luma block
  ****************************************************************************/
@@ -72,7 +122,6 @@ static void predict_16x16_p_altivec( uint8_t *src )
         vec_u8_t com_sat_v = vec_packsu(shift_0_v, shift_8_v);
         vec_st( com_sat_v, 0, &src[0]);
         src += FDEC_STRIDE;
-        i00 += c;
         add_i0_b_0v = vec_adds(add_i0_b_0v, c_v);
         add_i0_b_8v = vec_adds(add_i0_b_8v, c_v);
     }
@@ -187,3 +236,8 @@ void x264_predict_16x16_init_altivec( x264_predict_t pf[7] )
     pf[I_PRED_16x16_DC_TOP ] = predict_16x16_dc_top_altivec;
     pf[I_PRED_16x16_DC_128 ] = predict_16x16_dc_128_altivec;
 }
+
+void x264_predict_8x8c_init_altivec( x264_predict_t pf[7] )
+{
+    pf[I_PRED_CHROMA_P]       = predict_8x8c_p_altivec;
+}
diff --git a/common/ppc/predict.h b/common/ppc/predict.h
index 90863b8..29488aa 100644
--- a/common/ppc/predict.h
+++ b/common/ppc/predict.h
@@ -22,5 +22,6 @@
 #define X264_PPC_PREDICT_H
 
 void x264_predict_16x16_init_altivec ( x264_predict_t pf[7] );
+void x264_predict_8x8c_init_altivec( x264_predict_t pf[7] );
 
 #endif /* X264_PPC_PREDICT_H */
diff --git a/common/predict.c b/common/predict.c
index 9a66ed9..4fd4f06 100644
--- a/common/predict.c
+++ b/common/predict.c
@@ -786,6 +786,13 @@ void x264_predict_8x8c_init( int cpu, x264_predict_t pf[7] )
 #ifdef HAVE_MMX
     x264_predict_8x8c_init_mmx( cpu, pf );
 #endif
+
+#ifdef ARCH_PPC
+    if( cpu&X264_CPU_ALTIVEC )
+    {
+        x264_predict_8x8c_init_altivec( pf );
+    }
+#endif
 }
 
 void x264_predict_8x8_init( int cpu, x264_predict8x8_t pf[12] )