ARM: NEON optimised scalarproduct_float

Originally committed as revision 19971 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/arm/dsputil_neon.c

@@ -168,6 +168,7 @@ void ff_sv_fmul_scalar_2_neon(float *dst, const float **vp, float mul,
 void ff_sv_fmul_scalar_4_neon(float *dst, const float **vp, float mul,
                               int len);
 void ff_butterflies_float_neon(float *v1, float *v2, int len);
+float ff_scalarproduct_float_neon(const float *v1, const float *v2, int len);
 
 void ff_float_to_int16_neon(int16_t *, const float *, long);
 void ff_float_to_int16_interleave_neon(int16_t *, const float **, long, int);
@@ -282,6 +283,7 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
     c->vector_fmul_window = ff_vector_fmul_window_neon;
     c->vector_fmul_scalar = ff_vector_fmul_scalar_neon;
     c->butterflies_float = ff_butterflies_float_neon;
+    c->scalarproduct_float = ff_scalarproduct_float_neon;
 
     c->vector_fmul_sv_scalar[0] = ff_vector_fmul_sv_scalar_2_neon;
     c->vector_fmul_sv_scalar[1] = ff_vector_fmul_sv_scalar_4_neon;

libavcodec/arm/dsputil_neon_s.S

@@ -1010,3 +1010,16 @@ function ff_butterflies_float_neon, export=1
         bgt             1b
         bx              lr
         .endfunc
+
+function ff_scalarproduct_float_neon, export=1
+        vmov.f32        q2,  #0.0
+1:      vld1.32         {q0},[r0,:128]!
+        vld1.32         {q1},[r1,:128]!
+        vmla.f32        q2,  q0,  q1
+        subs            r2,  r2,  #4
+        bgt             1b
+        vadd.f32        d0,  d4,  d5
+        vpadd.f32       d0,  d0,  d0
+NOVFP   vmov.32         r0,  d0[0]
+        bx              lr
+        .endfunc