dsputil: move vector_fmul_scalar() to AVFloatDSPContext in libavutil

libavcodec/aacdec.c

@@ -1360,7 +1360,7 @@ static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
 
                     band_energy = ac->dsp.scalarproduct_float(cfo, cfo, off_len);
                     scale = sf[idx] / sqrtf(band_energy);
-                    ac->dsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
+                    ac->fdsp.vector_fmul_scalar(cfo, cfo, scale, off_len);
                 }
             } else {
                 const float *vq = ff_aac_codebook_vector_vals[cbt_m1];
@@ -1506,7 +1506,7 @@ static int decode_spectrum_and_dequant(AACContext *ac, float coef[1024],
                             }
                         } while (len -= 2);
 
-                        ac->dsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
+                        ac->fdsp.vector_fmul_scalar(cfo, cfo, sf[idx], off_len);
                     }
                 }
 
@@ -1730,10 +1730,10 @@ static void apply_intensity_stereo(AACContext *ac, ChannelElement *cpe, int ms_p
                         c *= 1 - 2 * cpe->ms_mask[idx];
                     scale = c * sce1->sf[idx];
                     for (group = 0; group < ics->group_len[g]; group++)
-                        ac->dsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
-                                                   coef0 + group * 128 + offsets[i],
-                                                   scale,
-                                                   offsets[i + 1] - offsets[i]);
+                        ac->fdsp.vector_fmul_scalar(coef1 + group * 128 + offsets[i],
+                                                    coef0 + group * 128 + offsets[i],
+                                                    scale,
+                                                    offsets[i + 1] - offsets[i]);
                 }
             } else {
                 int bt_run_end = sce1->band_type_run_end[idx];

libavcodec/arm/dsputil_init_neon.c

@@ -144,8 +144,6 @@ void ff_avg_h264_chroma_mc2_neon(uint8_t *, uint8_t *, int, int, int, int);
 
 void ff_vector_fmul_window_neon(float *dst, const float *src0,
                                 const float *src1, const float *win, int len);
-void ff_vector_fmul_scalar_neon(float *dst, const float *src, 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_vector_fmul_reverse_neon(float *dst, const float *src0,
@@ -305,7 +303,6 @@ 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_reverse        = ff_vector_fmul_reverse_neon;

libavcodec/arm/dsputil_neon.S

@@ -642,44 +642,6 @@ function ff_vorbis_inverse_coupling_neon, export=1
 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
-        it              eq
-        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_butterflies_float_neon, export=1
 1:      vld1.32         {q0},[r0,:128]
         vld1.32         {q1},[r1,:128]

libavcodec/dsputil.c

@@ -2392,14 +2392,6 @@ static void vector_fmul_window_c(float *dst, const float *src0,
     }
 }
 
-static void vector_fmul_scalar_c(float *dst, const float *src, float mul,
-                                 int len)
-{
-    int i;
-    for (i = 0; i < len; i++)
-        dst[i] = src[i] * mul;
-}
-
 static void butterflies_float_c(float *restrict v1, float *restrict v2,
                                 int len)
 {
@@ -2869,7 +2861,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
     c->scalarproduct_float = ff_scalarproduct_float_c;
     c->butterflies_float = butterflies_float_c;
     c->butterflies_float_interleave = butterflies_float_interleave_c;
-    c->vector_fmul_scalar = vector_fmul_scalar_c;
 
     c->shrink[0]= av_image_copy_plane;
     c->shrink[1]= ff_shrink22;

libavcodec/dsputil.h

@@ -382,16 +382,6 @@ typedef struct DSPContext {
     void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, int len);
     /* assume len is a multiple of 8, and arrays are 16-byte aligned */
     void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
-    /**
-     * Multiply a vector of floats by a scalar float.  Source and
-     * destination vectors must overlap exactly or not at all.
-     * @param dst result vector, 16-byte aligned
-     * @param src input vector, 16-byte aligned
-     * @param mul scalar value
-     * @param len length of vector, multiple of 4
-     */
-    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
-                               int len);
     /**
      * Calculate the scalar product of two vectors of floats.
      * @param v1  first vector, 16-byte aligned

libavcodec/libmp3lame.c

@@ -28,12 +28,12 @@
 
 #include "libavutil/channel_layout.h"
 #include "libavutil/common.h"
+#include "libavutil/float_dsp.h"
 #include "libavutil/intreadwrite.h"
 #include "libavutil/log.h"
 #include "libavutil/opt.h"
 #include "avcodec.h"
 #include "audio_frame_queue.h"
-#include "dsputil.h"
 #include "internal.h"
 #include "mpegaudio.h"
 #include "mpegaudiodecheader.h"
@@ -50,7 +50,7 @@ typedef struct LAMEContext {
     int reservoir;
     float *samples_flt[2];
     AudioFrameQueue afq;
-    DSPContext dsp;
+    AVFloatDSPContext fdsp;
 } LAMEContext;
 
 
@@ -167,7 +167,7 @@ static av_cold int mp3lame_encode_init(AVCodecContext *avctx)
     if (ret < 0)
         goto error;
 
-    ff_dsputil_init(&s->dsp, avctx);
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
 
     return 0;
 error:
@@ -205,10 +205,10 @@ static int mp3lame_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
                 return AVERROR(EINVAL);
             }
             for (ch = 0; ch < avctx->channels; ch++) {
-                s->dsp.vector_fmul_scalar(s->samples_flt[ch],
-                                          (const float *)frame->data[ch],
-                                          32768.0f,
-                                          FFALIGN(frame->nb_samples, 8));
+                s->fdsp.vector_fmul_scalar(s->samples_flt[ch],
+                                           (const float *)frame->data[ch],
+                                           32768.0f,
+                                           FFALIGN(frame->nb_samples, 8));
             }
             ENCODE_BUFFER(lame_encode_buffer_float, float, s->samples_flt);
             break;

libavcodec/wmaenc.c

@@ -111,7 +111,7 @@ static void apply_window_and_mdct(AVCodecContext * avctx, const AVFrame *frame)
 
     for (ch = 0; ch < avctx->channels; ch++) {
         memcpy(s->output, s->frame_out[ch], window_len * sizeof(*s->output));
-        s->dsp.vector_fmul_scalar(s->frame_out[ch], audio[ch], n, len);
+        s->fdsp.vector_fmul_scalar(s->frame_out[ch], audio[ch], n, len);
         s->dsp.vector_fmul_reverse(&s->output[window_len], s->frame_out[ch], win, len);
         s->fdsp.vector_fmul(s->frame_out[ch], s->frame_out[ch], win, len);
         mdct->mdct_calc(mdct, s->coefs[ch], s->output);

libavcodec/wmaprodec.c

@@ -86,6 +86,7 @@
  * subframe in order to reconstruct the output samples.
  */
 
+#include "libavutil/float_dsp.h"
 #include "libavutil/intfloat.h"
 #include "libavutil/intreadwrite.h"
 #include "avcodec.h"
@@ -170,6 +171,7 @@ typedef struct WMAProDecodeCtx {
     AVCodecContext*  avctx;                         ///< codec context for av_log
     AVFrame          frame;                         ///< AVFrame for decoded output
     DSPContext       dsp;                           ///< accelerated DSP functions
+    AVFloatDSPContext fdsp;
     uint8_t          frame_data[MAX_FRAMESIZE +
                       FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
     PutBitContext    pb;                            ///< context for filling the frame_data buffer
@@ -280,6 +282,8 @@ static av_cold int decode_init(AVCodecContext *avctx)
 
     s->avctx = avctx;
     ff_dsputil_init(&s->dsp, avctx);
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
+
     init_put_bits(&s->pb, s->frame_data, MAX_FRAMESIZE);
 
     avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
@@ -1008,12 +1012,12 @@ static void inverse_channel_transform(WMAProDecodeCtx *s)
                     }
                 } else if (s->avctx->channels == 2) {
                     int len = FFMIN(sfb[1], s->subframe_len) - sfb[0];
-                    s->dsp.vector_fmul_scalar(ch_data[0] + sfb[0],
-                                              ch_data[0] + sfb[0],
-                                              181.0 / 128, len);
-                    s->dsp.vector_fmul_scalar(ch_data[1] + sfb[0],
-                                              ch_data[1] + sfb[0],
-                                              181.0 / 128, len);
+                    s->fdsp.vector_fmul_scalar(ch_data[0] + sfb[0],
+                                               ch_data[0] + sfb[0],
+                                               181.0 / 128, len);
+                    s->fdsp.vector_fmul_scalar(ch_data[1] + sfb[0],
+                                               ch_data[1] + sfb[0],
+                                               181.0 / 128, len);
                 }
             }
         }
@@ -1259,9 +1263,9 @@ static int decode_subframe(WMAProDecodeCtx *s)
                             s->channel[c].scale_factor_step;
                 const float quant = pow(10.0, exp / 20.0);
                 int start = s->cur_sfb_offsets[b];
-                s->dsp.vector_fmul_scalar(s->tmp + start,
-                                          s->channel[c].coeffs + start,
-                                          quant, end - start);
+                s->fdsp.vector_fmul_scalar(s->tmp + start,
+                                           s->channel[c].coeffs + start,
+                                           quant, end - start);
             }
 
             /** apply imdct (imdct_half == DCTIV with reverse) */

libavutil/arm/float_dsp_init_neon.c

@@ -29,8 +29,12 @@ void ff_vector_fmul_neon(float *dst, const float *src0, const float *src1, int l
 void ff_vector_fmac_scalar_neon(float *dst, const float *src, float mul,
                                 int len);
 
+void ff_vector_fmul_scalar_neon(float *dst, const float *src, float mul,
+                                int len);
+
 void ff_float_dsp_init_neon(AVFloatDSPContext *fdsp)
 {
     fdsp->vector_fmul = ff_vector_fmul_neon;
     fdsp->vector_fmac_scalar = ff_vector_fmac_scalar_neon;
+    fdsp->vector_fmul_scalar = ff_vector_fmul_scalar_neon;
 }

libavutil/arm/float_dsp_neon.S

@@ -108,3 +108,41 @@ NOVFP   vdup.32         q15, r2
         bx              lr
         .unreq          len
 endfunc
+
+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
+        it              eq
+        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

libavutil/float_dsp.c

@@ -36,10 +36,19 @@ static void vector_fmac_scalar_c(float *dst, const float *src, float mul,
         dst[i] += src[i] * mul;
 }
 
+static void vector_fmul_scalar_c(float *dst, const float *src, float mul,
+                                 int len)
+{
+    int i;
+    for (i = 0; i < len; i++)
+        dst[i] = src[i] * mul;
+}
+
 void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
 {
     fdsp->vector_fmul = vector_fmul_c;
     fdsp->vector_fmac_scalar = vector_fmac_scalar_c;
+    fdsp->vector_fmul_scalar = vector_fmul_scalar_c;
 
 #if ARCH_ARM
     ff_float_dsp_init_arm(fdsp);

libavutil/float_dsp.h

@@ -51,6 +51,21 @@ typedef struct AVFloatDSPContext {
      */
     void (*vector_fmac_scalar)(float *dst, const float *src, float mul,
                                int len);
+
+    /**
+     * Multiply a vector of floats by a scalar float.  Source and
+     * destination vectors must overlap exactly or not at all.
+     *
+     * @param dst result vector
+     *            constraints: 16-byte aligned
+     * @param src input vector
+     *            constraints: 16-byte aligned
+     * @param mul scalar value
+     * @param len length of vector
+     *            constraints: multiple of 4
+     */
+    void (*vector_fmul_scalar)(float *dst, const float *src, float mul,
+                               int len);
 } AVFloatDSPContext;
 
 /**