floatdsp: move butterflies_float from dsputil to avfloatdsp.

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

libavcodec/aacdec.c

@@ -1693,9 +1693,9 @@ static void apply_mid_side_stereo(AACContext *ac, ChannelElement *cpe)
             if (cpe->ms_mask[idx] &&
                     cpe->ch[0].band_type[idx] < NOISE_BT && cpe->ch[1].band_type[idx] < NOISE_BT) {
                 for (group = 0; group < ics->group_len[g]; group++) {
-                    ac->dsp.butterflies_float(ch0 + group * 128 + offsets[i],
+                    ac->fdsp.butterflies_float(ch0 + group * 128 + offsets[i],
-                                              ch1 + group * 128 + offsets[i],
+                                               ch1 + group * 128 + offsets[i],
-                                              offsets[i+1] - offsets[i]);
+                                               offsets[i+1] - offsets[i]);
                 }
             }
         }

libavcodec/arm/dsputil_init_neon.c

@@ -142,7 +142,6 @@ void ff_avg_h264_chroma_mc8_neon(uint8_t *, uint8_t *, int, int, int, int);
 void ff_avg_h264_chroma_mc4_neon(uint8_t *, uint8_t *, int, int, int, int);
 void ff_avg_h264_chroma_mc2_neon(uint8_t *, uint8_t *, int, int, int, int);
 
-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_vector_clipf_neon(float *dst, const float *src, float min, float max,
@@ -294,7 +293,6 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
         c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
     }
 
-    c->butterflies_float          = ff_butterflies_float_neon;
     c->scalarproduct_float        = ff_scalarproduct_float_neon;
     c->vector_clipf               = ff_vector_clipf_neon;
     c->vector_clip_int32          = ff_vector_clip_int32_neon;

libavcodec/arm/dsputil_neon.S

@@ -531,18 +531,6 @@ function ff_add_pixels_clamped_neon, export=1
         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
-
 function ff_scalarproduct_float_neon, export=1
         vmov.f32        q2,  #0.0
 1:      vld1.32         {q0},[r0,:128]!

libavcodec/dsputil.c

@@ -2353,17 +2353,6 @@ WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)
 WRAPPER8_16_SQ(rd8x8_c, rd16_c)
 WRAPPER8_16_SQ(bit8x8_c, bit16_c)
 
-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;
-    }
-}
-
 float ff_scalarproduct_float_c(const float *v1, const float *v2, int len)
 {
     float p = 0.0;
@@ -2706,7 +2695,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->apply_window_int16 = apply_window_int16_c;
     c->vector_clip_int32 = vector_clip_int32_c;
     c->scalarproduct_float = ff_scalarproduct_float_c;
-    c->butterflies_float = butterflies_float_c;
 
     c->shrink[0]= av_image_copy_plane;
     c->shrink[1]= ff_shrink22;

libavcodec/dsputil.h

@@ -349,13 +349,6 @@ typedef struct DSPContext {
      * @param len length of vectors, multiple of 4
      */
     float (*scalarproduct_float)(const float *v1, const float *v2, 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);
 
     /* (I)DCT */
     void (*fdct)(DCTELEM *block/* align 16*/);

libavcodec/imc.c

@@ -36,6 +36,7 @@
 #include <stdio.h>
 
 #include "libavutil/channel_layout.h"
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
@@ -95,6 +96,7 @@ typedef struct {
     GetBitContext gb;
 
     DSPContext dsp;
+    AVFloatDSPContext fdsp;
     FFTContext fft;
     DECLARE_ALIGNED(32, FFTComplex, samples)[COEFFS / 2];
     float *out_samples;
@@ -244,6 +246,7 @@ static av_cold int imc_decode_init(AVCodecContext *avctx)
         return ret;
     }
     ff_dsputil_init(&q->dsp, avctx);
+    avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     avctx->sample_fmt     = AV_SAMPLE_FMT_FLTP;
     avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
                                                  : AV_CH_LAYOUT_STEREO;
@@ -959,8 +962,8 @@ static int imc_decode_frame(AVCodecContext *avctx, void *data,
     }
 
     if (avctx->channels == 2) {
-        q->dsp.butterflies_float((float *)q->frame.extended_data[0],
+        q->fdsp.butterflies_float((float *)q->frame.extended_data[0],
-                                 (float *)q->frame.extended_data[1], COEFFS);
+                                  (float *)q->frame.extended_data[1], COEFFS);
     }
 
     *got_frame_ptr   = 1;

libavcodec/mpegaudiodec.c

@@ -25,6 +25,7 @@
  */
 
 #include "libavutil/channel_layout.h"
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 #include "internal.h"
@@ -82,7 +83,7 @@ typedef struct MPADecodeContext {
     int err_recognition;
     AVCodecContext* avctx;
     MPADSPContext mpadsp;
-    DSPContext dsp;
+    AVFloatDSPContext fdsp;
     AVFrame frame;
 } MPADecodeContext;
 
@@ -434,8 +435,8 @@ static av_cold int decode_init(AVCodecContext * avctx)
 
     s->avctx = avctx;
 
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     ff_mpadsp_init(&s->mpadsp);
-    ff_dsputil_init(&s->dsp, avctx);
 
     if (avctx->request_sample_fmt == OUT_FMT &&
         avctx->codec_id != AV_CODEC_ID_MP3ON4)
@@ -1157,7 +1158,7 @@ found2:
         /* NOTE: the 1/sqrt(2) normalization factor is included in the
            global gain */
 #if CONFIG_FLOAT
-       s-> dsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
+       s->fdsp.butterflies_float(g0->sb_hybrid, g1->sb_hybrid, 576);
 #else
         tab0 = g0->sb_hybrid;
         tab1 = g1->sb_hybrid;

libavcodec/twinvq.c

@@ -178,7 +178,6 @@ static const ModeTab mode_44_48 = {
 typedef struct TwinContext {
     AVCodecContext *avctx;
     AVFrame frame;
-    DSPContext      dsp;
     AVFloatDSPContext fdsp;
     FFTContext mdct_ctx[3];
 
@@ -693,7 +692,7 @@ static void imdct_output(TwinContext *tctx, enum FrameType ftype, int wtype,
     if (tctx->avctx->channels == 2) {
         memcpy(&out[1][0],     &prev_buf[2*mtab->size],         size1 * sizeof(out[1][0]));
         memcpy(&out[1][size1], &tctx->curr_frame[2*mtab->size], size2 * sizeof(out[1][0]));
-        tctx->dsp.butterflies_float(out[0], out[1], mtab->size);
+        tctx->fdsp.butterflies_float(out[0], out[1], mtab->size);
     }
 }
 
@@ -1157,7 +1156,6 @@ static av_cold int twin_decode_init(AVCodecContext *avctx)
         return -1;
     }
 
-    ff_dsputil_init(&tctx->dsp, avctx);
     avpriv_float_dsp_init(&tctx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     if ((ret = init_mdct_win(tctx))) {
         av_log(avctx, AV_LOG_ERROR, "Error initializing MDCT\n");

libavcodec/wma.c

@@ -82,7 +82,6 @@ int ff_wma_init(AVCodecContext *avctx, int flags2)
         || avctx->bit_rate    <= 0)
         return -1;
 
-    ff_dsputil_init(&s->dsp, avctx);
     ff_fmt_convert_init(&s->fmt_conv, avctx);
     avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
 

libavcodec/wma.h

@@ -132,7 +132,6 @@ typedef struct WMACodecContext {
     float lsp_pow_e_table[256];
     float lsp_pow_m_table1[(1 << LSP_POW_BITS)];
     float lsp_pow_m_table2[(1 << LSP_POW_BITS)];
-    DSPContext dsp;
     FmtConvertContext fmt_conv;
     AVFloatDSPContext fdsp;
 

libavcodec/wmadec.c

@@ -719,7 +719,7 @@ static int wma_decode_block(WMACodecContext *s)
             s->channel_coded[0] = 1;
         }
 
-        s->dsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
+        s->fdsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len);
     }
 
 next:

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;
 }

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

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);

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;
 
 /**