ARM: NEON optimisations for some dsputil functions

NEON versions of the following functions are added:

  vector_fmul_scalar
  vector_fmul_sv_scalar
  sv_fmul_scalar
  butterflies_float

Originally committed as revision 19957 to svn://svn.ffmpeg.org/ffmpeg/trunk

libavcodec/arm/dsputil_neon.c

@@ -157,6 +157,17 @@ void ff_vector_fmul_neon(float *dst, const float *src, int len);
 void ff_vector_fmul_window_neon(float *dst, const float *src0,
                                 const float *src1, const float *win,
                                 float add_bias, int len);
+void ff_vector_fmul_scalar_neon(float *dst, const float *src, float mul,
+                                int len);
+void ff_vector_fmul_sv_scalar_2_neon(float *dst, const float *src,
+                                     const float **vp, float mul, int len);
+void ff_vector_fmul_sv_scalar_4_neon(float *dst, const float *src,
+                                     const float **vp, float mul, int len);
+void ff_sv_fmul_scalar_2_neon(float *dst, const float **vp, float mul,
+                              int len);
+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);
 
 void ff_float_to_int16_neon(int16_t *, const float *, long);
 void ff_float_to_int16_interleave_neon(int16_t *, const float **, long, int);
@@ -269,6 +280,14 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
 
     c->vector_fmul = ff_vector_fmul_neon;
     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->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;
+
+    c->sv_fmul_scalar[0] = ff_sv_fmul_scalar_2_neon;
+    c->sv_fmul_scalar[1] = ff_sv_fmul_scalar_4_neon;
 
     if (!(avctx->flags & CODEC_FLAG_BITEXACT)) {
         c->float_to_int16 = ff_float_to_int16_neon;

libavcodec/arm/dsputil_neon_s.S

@@ -858,3 +858,155 @@ function ff_vorbis_inverse_coupling_neon, export=1
         bx              lr
         .endfunc
 #endif
+
+function ff_vector_fmul_scalar_neon, export=1
+VFP     len .req r2
+NOVFP   len .req r3
+VFP     vdup.32         q8,  d0[0]
+NOVFP   vdup.32         q8,  r2
+        bics            r12, len, #15
+        beq             3f
+        vld1.32         {q0},[r1,:128]!
+        vld1.32         {q1},[r1,:128]!
+1:      vmul.f32        q0,  q0,  q8
+        vld1.32         {q2},[r1,:128]!
+        vmul.f32        q1,  q1,  q8
+        vld1.32         {q3},[r1,:128]!
+        vmul.f32        q2,  q2,  q8
+        vst1.32         {q0},[r0,:128]!
+        vmul.f32        q3,  q3,  q8
+        vst1.32         {q1},[r0,:128]!
+        subs            r12, r12, #16
+        beq             2f
+        vld1.32         {q0},[r1,:128]!
+        vst1.32         {q2},[r0,:128]!
+        vld1.32         {q1},[r1,:128]!
+        vst1.32         {q3},[r0,:128]!
+        b               1b
+2:      vst1.32         {q2},[r0,:128]!
+        vst1.32         {q3},[r0,:128]!
+        ands            len, len, #15
+        bxeq            lr
+3:      vld1.32         {q0},[r1,:128]!
+        vmul.f32        q0,  q0,  q8
+        vst1.32         {q0},[r0,:128]!
+        subs            len, len, #4
+        bgt             3b
+        bx              lr
+        .unreq          len
+        .endfunc
+
+function ff_vector_fmul_sv_scalar_2_neon, export=1
+VFP     vdup.32         d16, d0[0]
+NOVFP   vdup.32         d16, r3
+NOVFP   ldr             r3,  [sp]
+        vld1.32         {d0},[r1,:64]!
+        vld1.32         {d1},[r1,:64]!
+1:      subs            r3,  r3,  #4
+        vmul.f32        d4,  d0,  d16
+        vmul.f32        d5,  d1,  d16
+        ldr             r12, [r2], #4
+        vld1.32         {d2},[r12,:64]
+        ldr             r12, [r2], #4
+        vld1.32         {d3},[r12,:64]
+        vmul.f32        d4,  d4,  d2
+        vmul.f32        d5,  d5,  d3
+        beq             2f
+        vld1.32         {d0},[r1,:64]!
+        vld1.32         {d1},[r1,:64]!
+        vst1.32         {d4},[r0,:64]!
+        vst1.32         {d5},[r0,:64]!
+        b               1b
+2:      vst1.32         {d4},[r0,:64]!
+        vst1.32         {d5},[r0,:64]!
+        bx              lr
+        .endfunc
+
+function ff_vector_fmul_sv_scalar_4_neon, export=1
+VFP     vdup.32         q10, d0[0]
+NOVFP   vdup.32         q10, r3
+NOVFP   ldr             r3,  [sp]
+        push            {lr}
+        bics            lr,  r3,  #7
+        beq             3f
+        vld1.32         {q0},[r1,:128]!
+        vld1.32         {q2},[r1,:128]!
+1:      ldr             r12, [r2], #4
+        vld1.32         {q1},[r12,:128]
+        ldr             r12, [r2], #4
+        vld1.32         {q3},[r12,:128]
+        vmul.f32        q8,  q0,  q10
+        vmul.f32        q8,  q8,  q1
+        vmul.f32        q9,  q2,  q10
+        vmul.f32        q9,  q9,  q3
+        subs            lr,  lr,  #8
+        beq             2f
+        vld1.32         {q0},[r1,:128]!
+        vld1.32         {q2},[r1,:128]!
+        vst1.32         {q8},[r0,:128]!
+        vst1.32         {q9},[r0,:128]!
+        b               1b
+2:      vst1.32         {q8},[r0,:128]!
+        vst1.32         {q9},[r0,:128]!
+        ands            r3,  r3,  #7
+        popeq           {pc}
+3:      vld1.32         {q0},[r1,:128]!
+        ldr             r12, [r2], #4
+        vld1.32         {q1},[r12,:128]
+        vmul.f32        q0,  q0,  q10
+        vmul.f32        q0,  q0,  q1
+        vst1.32         {q0},[r0,:128]!
+        subs            r3,  r3,  #4
+        bgt             3b
+        pop             {pc}
+        .endfunc
+
+function ff_sv_fmul_scalar_2_neon, export=1
+VFP     len .req r2
+NOVFP   len .req r3
+VFP     vdup.32         q8,  d0[0]
+NOVFP   vdup.32         q8,  r2
+        ldr             r12, [r1], #4
+        vld1.32         {d0},[r12,:64]
+        ldr             r12, [r1], #4
+        vld1.32         {d1},[r12,:64]
+1:      vmul.f32        q1,  q0,  q8
+        subs            len, len, #4
+        beq             2f
+        ldr             r12, [r1], #4
+        vld1.32         {d0},[r12,:64]
+        ldr             r12, [r1], #4
+        vld1.32         {d1},[r12,:64]
+        vst1.32         {q1},[r0,:128]!
+        b               1b
+2:      vst1.32         {q1},[r0,:128]!
+        bx              lr
+        .unreq          len
+        .endfunc
+
+function ff_sv_fmul_scalar_4_neon, export=1
+VFP     len .req r2
+NOVFP   len .req r3
+VFP     vdup.32         q8,  d0[0]
+NOVFP   vdup.32         q8,  r2
+1:      ldr             r12, [r1], #4
+        vld1.32         {q0},[r12,:128]
+        vmul.f32        q0,  q0,  q8
+        vst1.32         {q0},[r0,:128]!
+        subs            len, len, #4
+        bgt             1b
+        bx              lr
+        .unreq          len
+        .endfunc
+
+function ff_butterflies_float_neon, export=1
+1:      vld1.32         {q0},[r0,:128]
+        vld1.32         {q1},[r1,:128]
+        vsub.f32        q2,  q0,  q1
+        vadd.f32        q1,  q0,  q1
+        vst1.32         {q2},[r1,:128]!
+        vst1.32         {q1},[r0,:128]!
+        subs            r2,  r2,  #4
+        bgt             1b
+        bx              lr
+        .endfunc