Add a float DSP framework to libavutil

Move vector_fmul() from DSPContext to AVFloatDSPContext.

39 files changed, 602 insertions, 188 deletions

diff --git a/libavcodec/aac.h b/libavcodec/aac.h
index 1b6013e458..45aacd1071 100644
--- a/libavcodec/aac.h
+++ b/libavcodec/aac.h
@@ -30,6 +30,7 @@
 #ifndef AVCODEC_AAC_H
 #define AVCODEC_AAC_H
 
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "dsputil.h"
 #include "fft.h"
@@ -292,6 +293,7 @@ typedef struct {
     FFTContext mdct_ltp;
     DSPContext dsp;
     FmtConvertContext fmt_conv;
+    AVFloatDSPContext fdsp;
     int random_state;
     /** @} */
 
diff --git a/libavcodec/aacdec.c b/libavcodec/aacdec.c
index ddc7eef138..3cf4a7fe18 100644
--- a/libavcodec/aacdec.c
+++ b/libavcodec/aacdec.c
@@ -79,7 +79,7 @@
            Parametric Stereo.
  */
 
-
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "internal.h"
 #include "get_bits.h"
@@ -867,6 +867,7 @@ static av_cold int aac_decode_init(AVCodecContext *avctx)
 
     ff_dsputil_init(&ac->dsp, avctx);
     ff_fmt_convert_init(&ac->fmt_conv, avctx);
+    avpriv_float_dsp_init(&ac->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
 
     ac->random_state = 0x1f2e3d4c;
 
@@ -2032,10 +2033,10 @@ static void windowing_and_mdct_ltp(AACContext *ac, float *out,
     const float *swindow_prev = ics->use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
 
     if (ics->window_sequence[0] != LONG_STOP_SEQUENCE) {
-        ac->dsp.vector_fmul(in, in, lwindow_prev, 1024);
+        ac->fdsp.vector_fmul(in, in, lwindow_prev, 1024);
     } else {
         memset(in, 0, 448 * sizeof(float));
-        ac->dsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
+        ac->fdsp.vector_fmul(in + 448, in + 448, swindow_prev, 128);
     }
     if (ics->window_sequence[0] != LONG_START_SEQUENCE) {
         ac->dsp.vector_fmul_reverse(in + 1024, in + 1024, lwindow, 1024);
diff --git a/libavcodec/aacenc.c b/libavcodec/aacenc.c
index 6021c375bb..0186011136 100644
--- a/libavcodec/aacenc.c
+++ b/libavcodec/aacenc.c
@@ -30,6 +30,7 @@
  * add temporal noise shaping
  ***********************************/
 
+#include "libavutil/float_dsp.h"
 #include "libavutil/opt.h"
 #include "avcodec.h"
 #include "put_bits.h"
@@ -182,7 +183,9 @@ static void put_audio_specific_config(AVCodecContext *avctx)
 }
 
 #define WINDOW_FUNC(type) \
-static void apply_ ##type ##_window(DSPContext *dsp, SingleChannelElement *sce, const float *audio)
+static void apply_ ##type ##_window(DSPContext *dsp, AVFloatDSPContext *fdsp, \
+                                    SingleChannelElement *sce, \
+                                    const float *audio)
 
 WINDOW_FUNC(only_long)
 {
@@ -190,7 +193,7 @@ WINDOW_FUNC(only_long)
     const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
     float *out = sce->ret;
 
-    dsp->vector_fmul        (out,        audio,        lwindow, 1024);
+    fdsp->vector_fmul       (out,        audio,        lwindow, 1024);
     dsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
 }
 
@@ -200,7 +203,7 @@ WINDOW_FUNC(long_start)
     const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
     float *out = sce->ret;
 
-    dsp->vector_fmul(out, audio, lwindow, 1024);
+    fdsp->vector_fmul(out, audio, lwindow, 1024);
     memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
     dsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
     memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
@@ -213,7 +216,7 @@ WINDOW_FUNC(long_stop)
     float *out = sce->ret;
 
     memset(out, 0, sizeof(out[0]) * 448);
-    dsp->vector_fmul(out + 448, audio + 448, swindow, 128);
+    fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
     memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
     dsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
 }
@@ -227,7 +230,7 @@ WINDOW_FUNC(eight_short)
     int w;
 
     for (w = 0; w < 8; w++) {
-        dsp->vector_fmul        (out, in, w ? pwindow : swindow, 128);
+        fdsp->vector_fmul       (out, in, w ? pwindow : swindow, 128);
         out += 128;
         in  += 128;
         dsp->vector_fmul_reverse(out, in, swindow, 128);
@@ -235,7 +238,9 @@ WINDOW_FUNC(eight_short)
     }
 }
 
-static void (*const apply_window[4])(DSPContext *dsp, SingleChannelElement *sce, const float *audio) = {
+static void (*const apply_window[4])(DSPContext *dsp, AVFloatDSPContext *fdsp,
+                                     SingleChannelElement *sce,
+                                     const float *audio) = {
     [ONLY_LONG_SEQUENCE]   = apply_only_long_window,
     [LONG_START_SEQUENCE]  = apply_long_start_window,
     [EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
@@ -248,7 +253,7 @@ static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
     int i;
     float *output = sce->ret;
 
-    apply_window[sce->ics.window_sequence[0]](&s->dsp, sce, audio);
+    apply_window[sce->ics.window_sequence[0]](&s->dsp, &s->fdsp, sce, audio);
 
     if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
         s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
@@ -694,6 +699,7 @@ static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
     int ret = 0;
 
     ff_dsputil_init(&s->dsp, avctx);
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
 
     // window init
     ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
diff --git a/libavcodec/aacenc.h b/libavcodec/aacenc.h
index f4e531a9a6..81ffb973d6 100644
--- a/libavcodec/aacenc.h
+++ b/libavcodec/aacenc.h
@@ -22,6 +22,7 @@
 #ifndef AVCODEC_AACENC_H
 #define AVCODEC_AACENC_H
 
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "put_bits.h"
 #include "dsputil.h"
@@ -58,6 +59,7 @@ typedef struct AACEncContext {
     FFTContext mdct1024;                         ///< long (1024 samples) frame transform context
     FFTContext mdct128;                          ///< short (128 samples) frame transform context
     DSPContext  dsp;
+    AVFloatDSPContext fdsp;
     float *planar_samples[6];                    ///< saved preprocessed input
 
     int samplerate_index;                        ///< MPEG-4 samplerate index
diff --git a/libavcodec/ac3enc.c b/libavcodec/ac3enc.c
index 3962fab748..50d7f704fa 100644
--- a/libavcodec/ac3enc.c
+++ b/libavcodec/ac3enc.c
@@ -2494,6 +2494,7 @@ av_cold int ff_ac3_encode_init(AVCodecContext *avctx)
 #endif
 
     ff_dsputil_init(&s->dsp, avctx);
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     ff_ac3dsp_init(&s->ac3dsp, avctx->flags & CODEC_FLAG_BITEXACT);
 
     dprint_options(s);
diff --git a/libavcodec/ac3enc.h b/libavcodec/ac3enc.h
index e8415a2d69..be9dcf21c1 100644
--- a/libavcodec/ac3enc.h
+++ b/libavcodec/ac3enc.h
@@ -29,6 +29,8 @@
 #define AVCODEC_AC3ENC_H
 
 #include <stdint.h>
+
+#include "libavutil/float_dsp.h"
 #include "ac3.h"
 #include "ac3dsp.h"
 #include "avcodec.h"
@@ -158,6 +160,7 @@ typedef struct AC3EncodeContext {
     AVCodecContext *avctx;                  ///< parent AVCodecContext
     PutBitContext pb;                       ///< bitstream writer context
     DSPContext dsp;
+    AVFloatDSPContext fdsp;
     AC3DSPContext ac3dsp;                   ///< AC-3 optimized functions
     FFTContext mdct;                        ///< FFT context for MDCT calculation
     const SampleType *mdct_window;          ///< MDCT window function array
diff --git a/libavcodec/ac3enc_fixed.c b/libavcodec/ac3enc_fixed.c
index 8d75ee303b..cc8f158c9d 100644
--- a/libavcodec/ac3enc_fixed.c
+++ b/libavcodec/ac3enc_fixed.c
@@ -68,10 +68,11 @@ av_cold int AC3_NAME(mdct_init)(AC3EncodeContext *s)
 /*
  * Apply KBD window to input samples prior to MDCT.
  */
-static void apply_window(DSPContext *dsp, int16_t *output, const int16_t *input,
+static void apply_window(void *dsp, int16_t *output, const int16_t *input,
                          const int16_t *window, unsigned int len)
 {
-    dsp->apply_window_int16(output, input, window, len);
+    DSPContext *dsp0 = dsp;
+    dsp0->apply_window_int16(output, input, window, len);
 }
 
 
diff --git a/libavcodec/ac3enc_float.c b/libavcodec/ac3enc_float.c
index 0b9e1de298..9eff5c5b6c 100644
--- a/libavcodec/ac3enc_float.c
+++ b/libavcodec/ac3enc_float.c
@@ -86,10 +86,12 @@ av_cold int ff_ac3_float_mdct_init(AC3EncodeContext *s)
 /*
  * Apply KBD window to input samples prior to MDCT.
  */
-static void apply_window(DSPContext *dsp, float *output, const float *input,
-                         const float *window, unsigned int len)
+static void apply_window(void *dsp, float *output,
+                         const float *input, const float *window,
+                         unsigned int len)
 {
-    dsp->vector_fmul(output, input, window, len);
+    AVFloatDSPContext *fdsp = dsp;
+    fdsp->vector_fmul(output, input, window, len);
 }
 
 
diff --git a/libavcodec/ac3enc_template.c b/libavcodec/ac3enc_template.c
index 9427cfe971..6e0a2b6f34 100644
--- a/libavcodec/ac3enc_template.c
+++ b/libavcodec/ac3enc_template.c
@@ -33,7 +33,7 @@
 
 static void scale_coefficients(AC3EncodeContext *s);
 
-static void apply_window(DSPContext *dsp, SampleType *output,
+static void apply_window(void *dsp, SampleType *output,
                          const SampleType *input, const SampleType *window,
                          unsigned int len);
 
@@ -107,8 +107,13 @@ static void apply_mdct(AC3EncodeContext *s)
             AC3Block *block = &s->blocks[blk];
             const SampleType *input_samples = &s->planar_samples[ch][blk * AC3_BLOCK_SIZE];
 
+#if CONFIG_AC3ENC_FLOAT
+            apply_window(&s->fdsp, s->windowed_samples, input_samples,
+                         s->mdct_window, AC3_WINDOW_SIZE);
+#else
             apply_window(&s->dsp, s->windowed_samples, input_samples,
                          s->mdct_window, AC3_WINDOW_SIZE);
+#endif
 
             if (s->fixed_point)
                 block->coeff_shift[ch+1] = normalize_samples(s);
diff --git a/libavcodec/arm/dsputil_init_neon.c b/libavcodec/arm/dsputil_init_neon.c
index d3ef85048f..398326c8c8 100644
--- a/libavcodec/arm/dsputil_init_neon.c
+++ b/libavcodec/arm/dsputil_init_neon.c
@@ -150,7 +150,6 @@ void ff_avg_h264_chroma_mc2_neon(uint8_t *, uint8_t *, int, int, int, int);
 void ff_vp3_v_loop_filter_neon(uint8_t *, int, int *);
 void ff_vp3_h_loop_filter_neon(uint8_t *, int, int *);
 
-void ff_vector_fmul_neon(float *dst, const float *src0, const float *src1, int len);
 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,
@@ -328,7 +327,6 @@ void ff_dsputil_init_neon(DSPContext *c, AVCodecContext *avctx)
         c->vp3_idct_dc_add   = ff_vp3_idct_dc_add_neon;
     }
 
-    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->vector_fmac_scalar         = ff_vector_fmac_scalar_neon;
diff --git a/libavcodec/arm/dsputil_init_vfp.c b/libavcodec/arm/dsputil_init_vfp.c
index d5e2d3b4ef..d77d686578 100644
--- a/libavcodec/arm/dsputil_init_vfp.c
+++ b/libavcodec/arm/dsputil_init_vfp.c
@@ -18,20 +18,13 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-#include "libavutil/arm/cpu.h"
 #include "libavcodec/dsputil.h"
 #include "dsputil_arm.h"
 
-void ff_vector_fmul_vfp(float *dst, const float *src0,
-                        const float *src1, int len);
 void ff_vector_fmul_reverse_vfp(float *dst, const float *src0,
                                 const float *src1, int len);
 
 void ff_dsputil_init_vfp(DSPContext* c, AVCodecContext *avctx)
 {
-    int cpu_flags = av_get_cpu_flags();
-
-    if (!have_vfpv3(cpu_flags))
-        c->vector_fmul = ff_vector_fmul_vfp;
     c->vector_fmul_reverse = ff_vector_fmul_reverse_vfp;
 }
diff --git a/libavcodec/arm/dsputil_neon.S b/libavcodec/arm/dsputil_neon.S
index b2243c8ccb..9a5a40d6ac 100644
--- a/libavcodec/arm/dsputil_neon.S
+++ b/libavcodec/arm/dsputil_neon.S
@@ -534,45 +534,6 @@ function ff_add_pixels_clamped_neon, export=1
         bx              lr
 endfunc
 
-function ff_vector_fmul_neon, export=1
-        subs            r3,  r3,  #8
-        vld1.32         {d0-d3},  [r1,:128]!
-        vld1.32         {d4-d7},  [r2,:128]!
-        vmul.f32        q8,  q0,  q2
-        vmul.f32        q9,  q1,  q3
-        beq             3f
-        bics            ip,  r3,  #15
-        beq             2f
-1:      subs            ip,  ip,  #16
-        vld1.32         {d0-d1},  [r1,:128]!
-        vld1.32         {d4-d5},  [r2,:128]!
-        vmul.f32        q10, q0,  q2
-        vld1.32         {d2-d3},  [r1,:128]!
-        vld1.32         {d6-d7},  [r2,:128]!
-        vmul.f32        q11, q1,  q3
-        vst1.32         {d16-d19},[r0,:128]!
-        vld1.32         {d0-d1},  [r1,:128]!
-        vld1.32         {d4-d5},  [r2,:128]!
-        vmul.f32        q8,  q0,  q2
-        vld1.32         {d2-d3},  [r1,:128]!
-        vld1.32         {d6-d7},  [r2,:128]!
-        vmul.f32        q9,  q1,  q3
-        vst1.32         {d20-d23},[r0,:128]!
-        bne             1b
-        ands            r3,  r3,  #15
-        beq             3f
-2:      vld1.32         {d0-d1},  [r1,:128]!
-        vld1.32         {d4-d5},  [r2,:128]!
-        vst1.32         {d16-d17},[r0,:128]!
-        vmul.f32        q8,  q0,  q2
-        vld1.32         {d2-d3},  [r1,:128]!
-        vld1.32         {d6-d7},  [r2,:128]!
-        vst1.32         {d18-d19},[r0,:128]!
-        vmul.f32        q9,  q1,  q3
-3:      vst1.32         {d16-d19},[r0,:128]!
-        bx              lr
-endfunc
-
 function ff_vector_fmul_window_neon, export=1
         push            {r4,r5,lr}
         ldr             lr,  [sp, #12]
diff --git a/libavcodec/arm/dsputil_vfp.S b/libavcodec/arm/dsputil_vfp.S
index 8d385c7e42..9df955dbf9 100644
--- a/libavcodec/arm/dsputil_vfp.S
+++ b/libavcodec/arm/dsputil_vfp.S
@@ -37,53 +37,6 @@
  */
 
 /**
- * ARM VFP optimized implementation of 'vector_fmul_c' function.
- * Assume that len is a positive number and is multiple of 8
- */
-@ void ff_vector_fmul_vfp(float *dst, const float *src0, const float *src1, int len)
-function ff_vector_fmul_vfp, export=1
-        vpush           {d8-d15}
-        fmrx            r12, fpscr
-        orr             r12, r12, #(3 << 16) /* set vector size to 4 */
-        fmxr            fpscr, r12
-
-        vldmia          r1!, {s0-s3}
-        vldmia          r2!, {s8-s11}
-        vldmia          r1!, {s4-s7}
-        vldmia          r2!, {s12-s15}
-        vmul.f32        s8,  s0,  s8
-1:
-        subs            r3,  r3,  #16
-        vmul.f32        s12, s4,  s12
-        itttt           ge
-        vldmiage        r1!, {s16-s19}
-        vldmiage        r2!, {s24-s27}
-        vldmiage        r1!, {s20-s23}
-        vldmiage        r2!, {s28-s31}
-        it              ge
-        vmulge.f32      s24, s16, s24
-        vstmia          r0!, {s8-s11}
-        vstmia          r0!, {s12-s15}
-        it              ge
-        vmulge.f32      s28, s20, s28
-        itttt           gt
-        vldmiagt        r1!, {s0-s3}
-        vldmiagt        r2!, {s8-s11}
-        vldmiagt        r1!, {s4-s7}
-        vldmiagt        r2!, {s12-s15}
-        ittt            ge
-        vmulge.f32      s8,  s0,  s8
-        vstmiage        r0!, {s24-s27}
-        vstmiage        r0!, {s28-s31}
-        bgt             1b
-
-        bic             r12, r12, #(7 << 16) /* set vector size back to 1 */
-        fmxr            fpscr, r12
-        vpop            {d8-d15}
-        bx              lr
-endfunc
-
-/**
  * ARM VFP optimized implementation of 'vector_fmul_reverse_c' function.
  * Assume that len is a positive number and is multiple of 8
  */
diff --git a/libavcodec/atrac3.c b/libavcodec/atrac3.c
index f468f11065..638889b32b 100644
--- a/libavcodec/atrac3.c
+++ b/libavcodec/atrac3.c
@@ -36,9 +36,9 @@
 #include <stddef.h>
 #include <stdio.h>
 
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
-#include "dsputil.h"
 #include "bytestream.h"
 #include "fft.h"
 #include "fmtconvert.h"
@@ -125,13 +125,13 @@ typedef struct {
 
     FFTContext          mdct_ctx;
     FmtConvertContext   fmt_conv;
+    AVFloatDSPContext   fdsp;
 } ATRAC3Context;
 
 static DECLARE_ALIGNED(32, float, mdct_window)[MDCT_SIZE];
 static VLC              spectral_coeff_tab[7];
 static float            gain_tab1[16];
 static float            gain_tab2[31];
-static DSPContext       dsp;
 
 
 /**
@@ -164,7 +164,7 @@ static void IMLT(ATRAC3Context *q, float *pInput, float *pOutput, int odd_band)
     q->mdct_ctx.imdct_calc(&q->mdct_ctx,pOutput,pInput);
 
     /* Perform windowing on the output. */
-    dsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
+    q->fdsp.vector_fmul(pOutput, pOutput, mdct_window, MDCT_SIZE);
 
 }
 
@@ -1039,7 +1039,7 @@ static av_cold int atrac3_decode_init(AVCodecContext *avctx)
         q->matrix_coeff_index_next[i] = 3;
     }
 
-    ff_dsputil_init(&dsp, avctx);
+    avpriv_float_dsp_init(&q->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
     ff_fmt_convert_init(&q->fmt_conv, avctx);
 
     q->pUnits = av_mallocz(sizeof(channel_unit)*q->channels);
diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c
index f42c58516f..942f606ea8 100644
--- a/libavcodec/dsputil.c
+++ b/libavcodec/dsputil.c
@@ -2363,12 +2363,6 @@ 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 vector_fmul_c(float *dst, const float *src0, const float *src1, int len){
-    int i;
-    for(i=0; i<len; i++)
-        dst[i] = src0[i] * src1[i];
-}
-
 static void vector_fmul_reverse_c(float *dst, const float *src0, const float *src1, int len){
     int i;
     src1 += len-1;
@@ -2898,7 +2892,6 @@ av_cold void ff_dsputil_init(DSPContext* c, AVCodecContext *avctx)
 #if CONFIG_AC3_DECODER
     c->ac3_downmix = ff_ac3_downmix_c;
 #endif
-    c->vector_fmul = vector_fmul_c;
     c->vector_fmul_reverse = vector_fmul_reverse_c;
     c->vector_fmul_add = vector_fmul_add_c;
     c->vector_fmul_window = vector_fmul_window_c;
diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h
index e0e5083d99..ec3d7ee007 100644
--- a/libavcodec/dsputil.h
+++ b/libavcodec/dsputil.h
@@ -399,7 +399,6 @@ typedef struct DSPContext {
     void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
     void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len);
     /* assume len is a multiple of 16, and arrays are 32-byte aligned */
-    void (*vector_fmul)(float *dst, const float *src0, const float *src1, int len);
     void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
     /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
     void (*vector_fmul_add)(float *dst, const float *src0, const float *src1, const float *src2, int len);
diff --git a/libavcodec/nellymoserenc.c b/libavcodec/nellymoserenc.c
index e1b1d793a4..18ea6846c2 100644
--- a/libavcodec/nellymoserenc.c
+++ b/libavcodec/nellymoserenc.c
@@ -35,6 +35,7 @@
  * http://wiki.multimedia.cx/index.php?title=Nellymoser
  */
 
+#include "libavutil/float_dsp.h"
 #include "libavutil/mathematics.h"
 #include "nellymoser.h"
 #include "avcodec.h"
@@ -55,6 +56,7 @@ typedef struct NellyMoserEncodeContext {
     AVCodecContext  *avctx;
     int             last_frame;
     DSPContext      dsp;
+    AVFloatDSPContext fdsp;
     FFTContext      mdct_ctx;
     AudioFrameQueue afq;
     DECLARE_ALIGNED(32, float, mdct_out)[NELLY_SAMPLES];
@@ -120,11 +122,11 @@ static void apply_mdct(NellyMoserEncodeContext *s)
     float *in1 = s->buf + NELLY_BUF_LEN;
     float *in2 = s->buf + 2 * NELLY_BUF_LEN;
 
-    s->dsp.vector_fmul        (s->in_buff,                 in0, ff_sine_128, NELLY_BUF_LEN);
+    s->fdsp.vector_fmul       (s->in_buff,                 in0, ff_sine_128, NELLY_BUF_LEN);
     s->dsp.vector_fmul_reverse(s->in_buff + NELLY_BUF_LEN, in1, ff_sine_128, NELLY_BUF_LEN);
     s->mdct_ctx.mdct_calc(&s->mdct_ctx, s->mdct_out, s->in_buff);
 
-    s->dsp.vector_fmul        (s->in_buff,                 in1, ff_sine_128, NELLY_BUF_LEN);
+    s->fdsp.vector_fmul       (s->in_buff,                 in1, ff_sine_128, NELLY_BUF_LEN);
     s->dsp.vector_fmul_reverse(s->in_buff + NELLY_BUF_LEN, in2, ff_sine_128, NELLY_BUF_LEN);
     s->mdct_ctx.mdct_calc(&s->mdct_ctx, s->mdct_out + NELLY_BUF_LEN, s->in_buff);
 }
@@ -172,6 +174,7 @@ static av_cold int encode_init(AVCodecContext *avctx)
     if ((ret = ff_mdct_init(&s->mdct_ctx, 8, 0, 32768.0)) < 0)
         goto error;
     ff_dsputil_init(&s->dsp, avctx);
+    avpriv_float_dsp_init(&s->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
 
     /* Generate overlap window */
     ff_sine_window_init(ff_sine_128, 128);
diff --git a/libavcodec/ppc/float_altivec.c b/libavcodec/ppc/float_altivec.c
index 6e0556ec42..5068fd4eb8 100644
--- a/libavcodec/ppc/float_altivec.c
+++ b/libavcodec/ppc/float_altivec.c
@@ -23,21 +23,6 @@
 
 #include "dsputil_altivec.h"
 
-static void vector_fmul_altivec(float *dst, const float *src0, const float *src1, int len)
-{
-    int i;
-    vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
-    for(i=0; i<len-7; i+=8) {
-        d0 = vec_ld(0, src0+i);
-        s = vec_ld(0, src1+i);
-        d1 = vec_ld(16, src0+i);
-        d0 = vec_madd(d0, s, zero);
-        d1 = vec_madd(d1, vec_ld(16,src1+i), zero);
-        vec_st(d0, 0, dst+i);
-        vec_st(d1, 16, dst+i);
-    }
-}
-
 static void vector_fmul_reverse_altivec(float *dst, const float *src0,
                                         const float *src1, int len)
 {
@@ -124,7 +109,6 @@ static void vector_fmul_window_altivec(float *dst, const float *src0, const floa
 
 void ff_float_init_altivec(DSPContext* c, AVCodecContext *avctx)
 {
-    c->vector_fmul = vector_fmul_altivec;
     c->vector_fmul_reverse = vector_fmul_reverse_altivec;
     c->vector_fmul_add = vector_fmul_add_altivec;
     if(!(avctx->flags & CODEC_FLAG_BITEXACT)) {
diff --git a/libavcodec/ra288.c b/libavcodec/ra288.c
index d7cda3ccf8..aa7ad33312 100644
--- a/libavcodec/ra288.c
+++ b/libavcodec/ra288.c
@@ -19,6 +19,7 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #define BITSTREAM_READER_LE
 #include "get_bits.h"
@@ -26,7 +27,6 @@
 #include "lpc.h"
 #include "celp_math.h"
 #include "celp_filters.h"
-#include "dsputil.h"
 
 #define MAX_BACKWARD_FILTER_ORDER  36
 #define MAX_BACKWARD_FILTER_LEN    40
@@ -38,6 +38,7 @@
 typedef struct {
     AVFrame frame;
     DSPContext dsp;
+    AVFloatDSPContext fdsp;
     DECLARE_ALIGNED(32, float,   sp_lpc)[FFALIGN(36, 16)];   ///< LPC coefficients for speech data (spec: A)
     DECLARE_ALIGNED(32, float, gain_lpc)[FFALIGN(10, 16)];   ///< LPC coefficients for gain        (spec: GB)
 
@@ -62,7 +63,7 @@ static av_cold int ra288_decode_init(AVCodecContext *avctx)
 {
     RA288Context *ractx = avctx->priv_data;
     avctx->sample_fmt = AV_SAMPLE_FMT_FLT;
-    ff_dsputil_init(&ractx->dsp, avctx);
+    avpriv_float_dsp_init(&ractx->fdsp, avctx->flags & CODEC_FLAG_BITEXACT);
 
     avcodec_get_frame_defaults(&ractx->frame);
     avctx->coded_frame = &ractx->frame;
@@ -137,7 +138,7 @@ static void do_hybrid_window(RA288Context *ractx,
                                             MAX_BACKWARD_FILTER_LEN   +
                                             MAX_BACKWARD_FILTER_NONREC, 16)]);
 
-    ractx->dsp.vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
+    ractx->fdsp.vector_fmul(work, window, hist, FFALIGN(order + n + non_rec, 16));
 
     convolve(buffer1, work + order    , n      , order);
     convolve(buffer2, work + order + n, non_rec, order);
@@ -164,7 +165,7 @@ static void backward_filter(RA288Context *ractx,
     do_hybrid_window(ractx, order, n, non_rec, temp, hist, rec, window);
 
     if (!compute_lpc_coefs(temp, order, lpc, 0, 1, 1))
-        ractx->dsp.vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
+        ractx->fdsp.vector_fmul(lpc, lpc, tab, FFALIGN(order, 16));
 
     memmove(hist, hist + n, move_size*sizeof(*hist));
 }
diff --git a/libavcodec/twinvq.c b/libavcodec/twinvq.c
index 67bc16088e..4421074d80 100644
--- a/libavcodec/twinvq.c
+++ b/libavcodec/twinvq.c
@@ -19,6 +19,7 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
@@ -176,6 +177,7 @@ typedef struct TwinContext {
     AVCodecContext *avctx;
     AVFrame frame;
     DSPContext      dsp;
+    AVFloatDSPContext fdsp;
     FFTContext mdct_ctx[3];
 
     const ModeTab *mtab;
@@ -787,8 +789,8 @@ static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb,
             dec_bark_env(tctx, bark1[i][j], bark_use_hist[i][j], i,
                          tctx->tmp_buf, gain[sub*i+j], ftype);
 
-            tctx->dsp.vector_fmul(chunk + block_size*j, chunk + block_size*j, tctx->tmp_buf,
-                                  block_size);
+            tctx->fdsp.vector_fmul(chunk + block_size*j, chunk + block_size*j,
+                                   tctx->tmp_buf, block_size);
 
         }
 
@@ -809,7 +811,7 @@ static void read_and_decode_spectrum(TwinContext *tctx, GetBitContext *gb,
         dec_lpc_spectrum_inv(tctx, lsp, ftype, tctx->tmp_buf);
 
         for (j = 0; j < mtab->fmode[ftype].sub; j++) {
-            tctx->dsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
+            tctx->fdsp.vector_fmul(chunk, chunk, tctx->tmp_buf, block_size);
             chunk += block_size;
         }
     }
@@ -1156,6 +1158,7 @@ static av_cold int twin_decode_init(AVCodecContext *avctx)
     }
 
     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");
         twin_decode_close(avctx);
diff --git a/libavcodec/vorbisdec.c b/libavcodec/vorbisdec.c
index 3489b8f029..08198d732a 100644
--- a/libavcodec/vorbisdec.c
+++ b/libavcodec/vorbisdec.c
@@ -26,6 +26,7 @@
 #include <math.h>
 
 #define BITSTREAM_READER_LE
+#include "libavutil/float_dsp.h"
 #include "avcodec.h"
 #include "get_bits.h"
 #include "dsputil.h"
@@ -124,6 +125,7 @@ typedef struct vorbis_context_s {
     AVFrame frame;
     GetBitContext gb;
     DSPContext dsp;
+    AVFloatDSPContext fdsp;
     FmtConvertContext fmt_conv;
 
     FFTContext mdct[2];
@@ -983,6 +985,7 @@ static av_cold int vorbis_decode_init(AVCodecContext *avccontext)
 
     vc->avccontext = avccontext;
     ff_dsputil_init(&vc->dsp, avccontext);
+    avpriv_float_dsp_init(&vc->fdsp, avccontext->flags & CODEC_FLAG_BITEXACT);
     ff_fmt_convert_init(&vc->fmt_conv, avccontext);
 
     if (avccontext->request_sample_fmt == AV_SAMPLE_FMT_FLT) {
@@ -1605,7 +1608,7 @@ static int vorbis_parse_audio_packet(vorbis_context *vc)
     for (j = vc->audio_channels-1;j >= 0; j--) {
         ch_floor_ptr = vc->channel_floors   + j           * blocksize / 2;
         ch_res_ptr   = vc->channel_residues + res_chan[j] * blocksize / 2;
-        vc->dsp.vector_fmul(ch_floor_ptr, ch_floor_ptr, ch_res_ptr, blocksize / 2);
+        vc->fdsp.vector_fmul(ch_floor_ptr, ch_floor_ptr, ch_res_ptr, blocksize / 2);
         mdct->imdct_half(mdct, ch_res_ptr, ch_floor_ptr);
     }
 
diff --git a/libavcodec/x86/dsputil_mmx.c b/libavcodec/x86/dsputil_mmx.c
index fca09b4e65..434d1859ec 100644
--- a/libavcodec/x86/dsputil_mmx.c
+++ b/libavcodec/x86/dsputil_mmx.c
@@ -2502,11 +2502,6 @@ int  ff_add_hfyu_left_prediction_sse4(uint8_t *dst, const uint8_t *src,
 
 float ff_scalarproduct_float_sse(const float *v1, const float *v2, int order);
 
-void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
-                        int len);
-void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
-                        int len);
-
 void ff_vector_fmul_reverse_sse(float *dst, const float *src0,
                                 const float *src1, int len);
 void ff_vector_fmul_reverse_avx(float *dst, const float *src0,
@@ -2833,7 +2828,6 @@ static void dsputil_init_sse(DSPContext *c, AVCodecContext *avctx, int mm_flags)
     c->vorbis_inverse_coupling = vorbis_inverse_coupling_sse;
     c->ac3_downmix             = ac3_downmix_sse;
 #if HAVE_YASM
-    c->vector_fmul         = ff_vector_fmul_sse;
     c->vector_fmul_reverse = ff_vector_fmul_reverse_sse;
     c->vector_fmul_add     = ff_vector_fmul_add_sse;
 #endif
@@ -2995,7 +2989,6 @@ static void dsputil_init_avx(DSPContext *c, AVCodecContext *avctx, int mm_flags)
         }
     }
     c->butterflies_float_interleave = ff_butterflies_float_interleave_avx;
-    c->vector_fmul = ff_vector_fmul_avx;
     c->vector_fmul_reverse = ff_vector_fmul_reverse_avx;
     c->vector_fmul_add = ff_vector_fmul_add_avx;
 #endif
diff --git a/libavcodec/x86/dsputil_yasm.asm b/libavcodec/x86/dsputil_yasm.asm
index c6ef834503..313e774669 100644
--- a/libavcodec/x86/dsputil_yasm.asm
+++ b/libavcodec/x86/dsputil_yasm.asm
@@ -1130,38 +1130,6 @@ VECTOR_CLIP_INT32 6, 1, 0, 0
 %endif
 
 ;-----------------------------------------------------------------------------
-; void vector_fmul(float *dst, const float *src0, const float *src1, int len)
-;-----------------------------------------------------------------------------
-%macro VECTOR_FMUL 0
-cglobal vector_fmul, 4,4,2, dst, src0, src1, len
-    lea       lenq, [lend*4 - 2*mmsize]
-ALIGN 16
-.loop
-    mova      m0,   [src0q + lenq]
-    mova      m1,   [src0q + lenq + mmsize]
-    mulps     m0, m0, [src1q + lenq]
-    mulps     m1, m1, [src1q + lenq + mmsize]
-    mova      [dstq + lenq], m0
-    mova      [dstq + lenq + mmsize], m1
-
-    sub       lenq, 2*mmsize
-    jge       .loop
-%if mmsize == 32
-    vzeroupper
-    RET
-%else
-    REP_RET
-%endif
-%endmacro
-
-INIT_XMM sse
-VECTOR_FMUL
-%if HAVE_AVX
-INIT_YMM avx
-VECTOR_FMUL
-%endif
-
-;-----------------------------------------------------------------------------
 ; void vector_fmul_reverse(float *dst, const float *src0, const float *src1,
 ;                          int len)
 ;-----------------------------------------------------------------------------
diff --git a/libavutil/Makefile b/libavutil/Makefile
index fb19ebf1e2..56ccd04dff 100644
--- a/libavutil/Makefile
+++ b/libavutil/Makefile
@@ -57,6 +57,7 @@ OBJS = adler32.o                                                        \
        eval.o                                                           \
        fifo.o                                                           \
        file.o                                                           \
+       float_dsp.o                                                      \
        imgutils.o                                                       \
        intfloat_readwrite.o                                             \
        inverse.o                                                        \
diff --git a/libavutil/arm/Makefile b/libavutil/arm/Makefile
index e600383b9c..ac7eca6751 100644
--- a/libavutil/arm/Makefile
+++ b/libavutil/arm/Makefile
@@ -1 +1,8 @@
 OBJS += arm/cpu.o                                                       \
+        arm/float_dsp_init_arm.o                                        \
+
+ARMVFP-OBJS += arm/float_dsp_init_vfp.o                                 \
+               arm/float_dsp_vfp.o                                      \
+
+NEON-OBJS += arm/float_dsp_init_neon.o                                  \
+             arm/float_dsp_neon.o                                       \
diff --git a/libavutil/arm/float_dsp_arm.h b/libavutil/arm/float_dsp_arm.h
new file mode 100644
index 0000000000..976e5a0a24
--- /dev/null
+++ b/libavutil/arm/float_dsp_arm.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVUTIL_ARM_FLOAT_DSP_ARM_H
+#define AVUTIL_ARM_FLOAT_DSP_ARM_H
+
+#include "libavutil/float_dsp.h"
+
+void ff_dsputil_init_vfp (AVFloatDSPContext *fdsp);
+void ff_dsputil_init_neon(AVFloatDSPContext *fdsp);
+
+#endif /* AVUTIL_ARM_FLOAT_DSP_ARM_H */
diff --git a/libavutil/arm/float_dsp_init_arm.c b/libavutil/arm/float_dsp_init_arm.c
new file mode 100644
index 0000000000..0fee8611c4
--- /dev/null
+++ b/libavutil/arm/float_dsp_init_arm.c
@@ -0,0 +1,33 @@
+/*
+ * ARM optimized DSP utils
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/arm/cpu.h"
+#include "libavutil/float_dsp.h"
+#include "float_dsp_arm.h"
+
+void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (have_vfp(cpu_flags))
+        ff_dsputil_init_vfp(fdsp);
+    if (have_neon(cpu_flags))
+        ff_dsputil_init_neon(fdsp);
+}
diff --git a/libavutil/arm/float_dsp_init_neon.c b/libavutil/arm/float_dsp_init_neon.c
new file mode 100644
index 0000000000..1e732db9df
--- /dev/null
+++ b/libavutil/arm/float_dsp_init_neon.c
@@ -0,0 +1,32 @@
+/*
+ * ARM NEON optimised Float DSP functions
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <stdint.h>
+
+#include "libavutil/float_dsp.h"
+#include "float_dsp_arm.h"
+
+void ff_vector_fmul_neon(float *dst, const float *src0, const float *src1, int len);
+
+void ff_dsputil_init_neon(AVFloatDSPContext *fdsp)
+{
+    c->vector_fmul = ff_vector_fmul_neon;
+}
diff --git a/libavutil/arm/float_dsp_init_vfp.c b/libavutil/arm/float_dsp_init_vfp.c
new file mode 100644
index 0000000000..d58dbded3a
--- /dev/null
+++ b/libavutil/arm/float_dsp_init_vfp.c
@@ -0,0 +1,34 @@
+/*
+ * Copyright (c) 2008 Siarhei Siamashka <ssvb@users.sourceforge.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/arm/cpu.h"
+#include "libavutil/float_dsp.h"
+#include "float_dsp_arm.h"
+
+void ff_vector_fmul_vfp(float *dst, const float *src0, const float *src1,
+                        int len);
+
+void ff_dsputil_init_vfp(DSPContext* c, AVCodecContext *avctx)
+{
+    int cpu_flags = av_get_cpu_flags();
+
+    if (!have_vfpv3(cpu_flags))
+        c->vector_fmul = ff_vector_fmul_vfp;
+}
diff --git a/libavutil/arm/float_dsp_neon.S b/libavutil/arm/float_dsp_neon.S
new file mode 100644
index 0000000000..d66fa09424
--- /dev/null
+++ b/libavutil/arm/float_dsp_neon.S
@@ -0,0 +1,64 @@
+/*
+ * ARM NEON optimised Float DSP functions
+ * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "asm.S"
+
+        preserve8
+
+function ff_vector_fmul_neon, export=1
+        subs            r3,  r3,  #8
+        vld1.32         {d0-d3},  [r1,:128]!
+        vld1.32         {d4-d7},  [r2,:128]!
+        vmul.f32        q8,  q0,  q2
+        vmul.f32        q9,  q1,  q3
+        beq             3f
+        bics            ip,  r3,  #15
+        beq             2f
+1:      subs            ip,  ip,  #16
+        vld1.32         {d0-d1},  [r1,:128]!
+        vld1.32         {d4-d5},  [r2,:128]!
+        vmul.f32        q10, q0,  q2
+        vld1.32         {d2-d3},  [r1,:128]!
+        vld1.32         {d6-d7},  [r2,:128]!
+        vmul.f32        q11, q1,  q3
+        vst1.32         {d16-d19},[r0,:128]!
+        vld1.32         {d0-d1},  [r1,:128]!
+        vld1.32         {d4-d5},  [r2,:128]!
+        vmul.f32        q8,  q0,  q2
+        vld1.32         {d2-d3},  [r1,:128]!
+        vld1.32         {d6-d7},  [r2,:128]!
+        vmul.f32        q9,  q1,  q3
+        vst1.32         {d20-d23},[r0,:128]!
+        bne             1b
+        ands            r3,  r3,  #15
+        beq             3f
+2:      vld1.32         {d0-d1},  [r1,:128]!
+        vld1.32         {d4-d5},  [r2,:128]!
+        vst1.32         {d16-d17},[r0,:128]!
+        vmul.f32        q8,  q0,  q2
+        vld1.32         {d2-d3},  [r1,:128]!
+        vld1.32         {d6-d7},  [r2,:128]!
+        vst1.32         {d18-d19},[r0,:128]!
+        vmul.f32        q9,  q1,  q3
+3:      vst1.32         {d16-d19},[r0,:128]!
+        bx              lr
+endfunc
diff --git a/libavutil/arm/float_dsp_vfp.S b/libavutil/arm/float_dsp_vfp.S
new file mode 100644
index 0000000000..3931828381
--- /dev/null
+++ b/libavutil/arm/float_dsp_vfp.S
@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) 2008 Siarhei Siamashka <ssvb@users.sourceforge.net>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+#include "asm.S"
+
+/**
+ * Assume that len is a positive number and is multiple of 8
+ */
+@ void ff_vector_fmul_vfp(float *dst, const float *src0, const float *src1, int len)
+function ff_vector_fmul_vfp, export=1
+        vpush           {d8-d15}
+        fmrx            r12, fpscr
+        orr             r12, r12, #(3 << 16) /* set vector size to 4 */
+        fmxr            fpscr, r12
+
+        vldmia          r1!, {s0-s3}
+        vldmia          r2!, {s8-s11}
+        vldmia          r1!, {s4-s7}
+        vldmia          r2!, {s12-s15}
+        vmul.f32        s8,  s0,  s8
+1:
+        subs            r3,  r3,  #16
+        vmul.f32        s12, s4,  s12
+        itttt           ge
+        vldmiage        r1!, {s16-s19}
+        vldmiage        r2!, {s24-s27}
+        vldmiage        r1!, {s20-s23}
+        vldmiage        r2!, {s28-s31}
+        it              ge
+        vmulge.f32      s24, s16, s24
+        vstmia          r0!, {s8-s11}
+        vstmia          r0!, {s12-s15}
+        it              ge
+        vmulge.f32      s28, s20, s28
+        itttt           gt
+        vldmiagt        r1!, {s0-s3}
+        vldmiagt        r2!, {s8-s11}
+        vldmiagt        r1!, {s4-s7}
+        vldmiagt        r2!, {s12-s15}
+        ittt            ge
+        vmulge.f32      s8,  s0,  s8
+        vstmiage        r0!, {s24-s27}
+        vstmiage        r0!, {s28-s31}
+        bgt             1b
+
+        bic             r12, r12, #(7 << 16) /* set vector size back to 1 */
+        fmxr            fpscr, r12
+        vpop            {d8-d15}
+        bx              lr
+endfunc
diff --git a/libavutil/float_dsp.c b/libavutil/float_dsp.c
new file mode 100644
index 0000000000..039dd07d36
--- /dev/null
+++ b/libavutil/float_dsp.c
@@ -0,0 +1,42 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+
+#include "float_dsp.h"
+
+static void vector_fmul_c(float *dst, const float *src0, const float *src1,
+                          int len)
+{
+    int i;
+    for (i = 0; i < len; i++)
+        dst[i] = src0[i] * src1[i];
+}
+
+void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int bit_exact)
+{
+    fdsp->vector_fmul = vector_fmul_c;
+
+#if ARCH_ARM
+    ff_float_dsp_init_arm(fdsp);
+#elif ARCH_PPC
+    ff_float_dsp_init_ppc(fdsp, bit_exact);
+#elif ARCH_X86
+    ff_float_dsp_init_x86(fdsp);
+#endif
+}
diff --git a/libavutil/float_dsp.h b/libavutil/float_dsp.h
new file mode 100644
index 0000000000..30161a252b
--- /dev/null
+++ b/libavutil/float_dsp.h
@@ -0,0 +1,53 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVUTIL_FLOAT_DSP_H
+#define AVUTIL_FLOAT_DSP_H
+
+typedef struct AVFloatDSPContext {
+    /**
+     * Calculate the product of two vectors of floats and store the result in
+     * a vector of floats.
+     *
+     * @param dst  output vector
+     *             constraints: 32-byte aligned
+     * @param src0 first input vector
+     *             constraints: 32-byte aligned
+     * @param src1 second input vector
+     *             constraints: 32-byte aligned
+     * @param len  number of elements in the input
+     *             constraints: multiple of 16
+     */
+    void (*vector_fmul)(float *dst, const float *src0, const float *src1,
+                        int len);
+} AVFloatDSPContext;
+
+/**
+ * Initialize a float DSP context.
+ *
+ * @param fdsp    float DSP context
+ * @param strict  setting to non-zero avoids using functions which may not be IEEE-754 compliant
+ */
+void avpriv_float_dsp_init(AVFloatDSPContext *fdsp, int strict);
+
+
+void ff_float_dsp_init_arm(AVFloatDSPContext *fdsp);
+void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int strict);
+void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp);
+
+#endif /* AVUTIL_FLOAT_DSP_H */
diff --git a/libavutil/ppc/Makefile b/libavutil/ppc/Makefile
index 5b18b08957..4fd8d6d57e 100644
--- a/libavutil/ppc/Makefile
+++ b/libavutil/ppc/Makefile
@@ -1 +1,4 @@
 OBJS += ppc/cpu.o                                                       \
+        ppc/float_dsp_init.o                                            \
+
+ALTIVEC-OBJS += ppc/float_dsp_altivec.o                                 \
diff --git a/libavutil/ppc/float_dsp_altivec.c b/libavutil/ppc/float_dsp_altivec.c
new file mode 100644
index 0000000000..f08abc8d6f
--- /dev/null
+++ b/libavutil/ppc/float_dsp_altivec.c
@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "util_altivec.h"
+
+void ff_vector_fmul_altivec(float *dst, const float *src0, const float *src1,
+                            int len)
+{
+    int i;
+    vector float d0, d1, s, zero = (vector float)vec_splat_u32(0);
+    for (i = 0; i < len - 7; i += 8) {
+        d0 = vec_ld( 0, src0 + i);
+        s  = vec_ld( 0, src1 + i);
+        d1 = vec_ld(16, src0 + i);
+        d0 = vec_madd(d0, s, zero);
+        d1 = vec_madd(d1, vec_ld(16, src1 + i), zero);
+        vec_st(d0,  0, dst + i);
+        vec_st(d1, 16, dst + i);
+    }
+}
diff --git a/libavutil/ppc/float_dsp_init.c b/libavutil/ppc/float_dsp_init.c
new file mode 100644
index 0000000000..0477749e83
--- /dev/null
+++ b/libavutil/ppc/float_dsp_init.c
@@ -0,0 +1,37 @@
+/*
+ * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org>
+ *
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/cpu.h"
+#include "libavutil/float_dsp.h"
+
+extern void ff_vector_fmul_altivec(float *dst, const float *src0,
+                                   const float *src1, int len);
+
+void ff_float_dsp_init_ppc(AVFloatDSPContext *fdsp, int bit_exact)
+{
+#if HAVE_ALTIVEC
+    int mm_flags = av_get_cpu_flags();
+
+    if (!(mm_flags & AV_CPU_FLAG_ALTIVEC))
+        return;
+
+    fdsp->vector_fmul = ff_vector_fmul_altivec;
+#endif
+}
diff --git a/libavutil/x86/Makefile b/libavutil/x86/Makefile
index de8a341442..4546353669 100644
--- a/libavutil/x86/Makefile
+++ b/libavutil/x86/Makefile
@@ -1 +1,4 @@
 OBJS += x86/cpu.o                                                       \
+        x86/float_dsp_init.o                                            \
+
+YASM-OBJS += x86/float_dsp.o                                            \
diff --git a/libavutil/x86/float_dsp.asm b/libavutil/x86/float_dsp.asm
new file mode 100644
index 0000000000..53be7ab99a
--- /dev/null
+++ b/libavutil/x86/float_dsp.asm
@@ -0,0 +1,55 @@
+;*****************************************************************************
+;* x86-optimized Float DSP functions
+;*
+;* This file is part of Libav.
+;*
+;* Libav is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* Libav is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with Libav; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "x86inc.asm"
+
+SECTION .text
+
+;-----------------------------------------------------------------------------
+; void vector_fmul(float *dst, const float *src0, const float *src1, int len)
+;-----------------------------------------------------------------------------
+%macro VECTOR_FMUL 0
+cglobal vector_fmul, 4,4,2, dst, src0, src1, len
+    lea       lenq, [lend*4 - 2*mmsize]
+ALIGN 16
+.loop
+    mova      m0,   [src0q + lenq]
+    mova      m1,   [src0q + lenq + mmsize]
+    mulps     m0, m0, [src1q + lenq]
+    mulps     m1, m1, [src1q + lenq + mmsize]
+    mova      [dstq + lenq], m0
+    mova      [dstq + lenq + mmsize], m1
+
+    sub       lenq, 2*mmsize
+    jge       .loop
+%if mmsize == 32
+    vzeroupper
+    RET
+%else
+    REP_RET
+%endif
+%endmacro
+
+INIT_XMM sse
+VECTOR_FMUL
+%if HAVE_AVX
+INIT_YMM avx
+VECTOR_FMUL
+%endif
diff --git a/libavutil/x86/float_dsp_init.c b/libavutil/x86/float_dsp_init.c
new file mode 100644
index 0000000000..10bb226f23
--- /dev/null
+++ b/libavutil/x86/float_dsp_init.c
@@ -0,0 +1,41 @@
+/*
+ * This file is part of Libav.
+ *
+ * Libav is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * Libav is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with Libav; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "config.h"
+
+#include "libavutil/cpu.h"
+#include "libavutil/float_dsp.h"
+
+extern void ff_vector_fmul_sse(float *dst, const float *src0, const float *src1,
+                               int len);
+extern void ff_vector_fmul_avx(float *dst, const float *src0, const float *src1,
+                               int len);
+
+void ff_float_dsp_init_x86(AVFloatDSPContext *fdsp)
+{
+#if HAVE_YASM
+    int mm_flags = av_get_cpu_flags();
+
+    if (mm_flags & AV_CPU_FLAG_SSE && HAVE_SSE) {
+        fdsp->vector_fmul = ff_vector_fmul_sse;
+    }
+    if (mm_flags & AV_CPU_FLAG_AVX && HAVE_AVX) {
+        fdsp->vector_fmul = ff_vector_fmul_avx;
+    }
+#endif
+}