floatdsp: move butterflies_float from dsputil to avfloatdsp.

This makes wmadec/enc, twinvq and mpegaudiodec (i.e. mp2/mp3)
independent of dsputil.

4 files changed, 37 insertions, 0 deletions

diff --git a/libavutil/arm/float_dsp_init_neon.c b/libavutil/arm/float_dsp_init_neon.c
index c6f02bd2c5..b3644e82a2 100644
--- a/libavutil/arm/float_dsp_init_neon.c
+++ b/libavutil/arm/float_dsp_init_neon.c
@@ -41,6 +41,8 @@ void ff_vector_fmul_add_neon(float *dst, const float *src0, const float *src1,
 void ff_vector_fmul_reverse_neon(float *dst, const float *src0,
                                  const float *src1, int len);
 
+void ff_butterflies_float_neon(float *v1, float *v2, int len);
+
 void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp)
 {
     fdsp->vector_fmul = ff_vector_fmul_neon;
@@ -49,4 +51,5 @@ void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp)
     fdsp->vector_fmul_window = ff_vector_fmul_window_neon;
     fdsp->vector_fmul_add    = ff_vector_fmul_add_neon;
     fdsp->vector_fmul_reverse = ff_vector_fmul_reverse_neon;
+    fdsp->butterflies_float = ff_butterflies_float_neon;
 }
diff --git a/libavutil/arm/float_dsp_neon.S b/libavutil/arm/float_dsp_neon.S
index d00e59de8f..4acc406d33 100644
--- a/libavutil/arm/float_dsp_neon.S
+++ b/libavutil/arm/float_dsp_neon.S
@@ -244,3 +244,15 @@ function ff_vector_fmul_reverse_neon, export=1
 2:      vst1.32         {q8-q9},  [r0,:128]!
         bx              lr
 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
diff --git a/libavutil/float_dsp.c b/libavutil/float_dsp.c
index aa5293710b..c6e2b41d66 100644
--- a/libavutil/float_dsp.c
+++ b/libavutil/float_dsp.c
@@ -89,6 +89,18 @@ static void vector_fmul_reverse_c(float *dst, const float *src0,
         dst[i] = src0[i] * src1[-i];
 }
 
+static void butterflies_float_c(float *restrict v1, float *restrict v2,
+                                int len)
+{
+    int i;
+
+    for (i = 0; i < len; i++) {
+        float t = v1[i] - v2[i];
+        v1[i] += v2[i];
+        v2[i] = t;
+    }
+}
+
 void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
 {
     fdsp->vector_fmul = vector_fmul_c;
@@ -98,6 +110,7 @@ void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
     fdsp->vector_fmul_window = vector_fmul_window_c;
     fdsp->vector_fmul_add = vector_fmul_add_c;
     fdsp->vector_fmul_reverse = vector_fmul_reverse_c;
+    fdsp->butterflies_float = butterflies_float_c;
 
 #if ARCH_ARM
     ff_float_dsp_init_arm(fdsp);
diff --git a/libavutil/float_dsp.h b/libavutil/float_dsp.h
index 96fcdab1e6..ec57b36f79 100644
--- a/libavutil/float_dsp.h
+++ b/libavutil/float_dsp.h
@@ -137,6 +137,15 @@ typedef struct AVFloatDSPContext {
      */
     void (*vector_fmul_reverse)(float *dst, const float *src0,
                                 const float *src1, int len);
+
+    /**
+     * Calculate the sum and difference of two vectors of floats.
+     *
+     * @param v1  first input vector, sum output, 16-byte aligned
+     * @param v2  second input vector, difference output, 16-byte aligned
+     * @param len length of vectors, multiple of 4
+     */
+    void (*butterflies_float)(float *restrict v1, float *restrict v2, int len);
 } AVFloatDSPContext;
 
 /**