shithub: opus

Info  •  Files  •  Log  •  Branches

Unifying scaling of fixed-point and float FFT

--- a/celt/dump_modes/dump_modes.c

+++ b/celt/dump_modes/dump_modes.c

@@ -172,8 +172,9 @@

          fprintf (file, "static const kiss_fft_state fft_state%d_%d_%d = {\n",

                mode->Fs, mdctSize, k);

          fprintf (file, "%d,\t/* nfft */\n", mode->mdct.kfft[k]->nfft);

-#ifndef FIXED_POINT

-         fprintf (file, "%0.9ff,\t/* scale */\n", mode->mdct.kfft[k]->scale);

+         fprintf (file, WORD16 ",\t/* scale */\n", mode->mdct.kfft[k]->scale);

+#ifdef FIXED_POINT

+         fprintf (file, "%d,\t/* scale_shift */\n", mode->mdct.kfft[k]->scale_shift);

 #endif

          fprintf (file, "%d,\t/* shift */\n", mode->mdct.kfft[k]->shift);

          fprintf (file, "{");

--- a/celt/kiss_fft.c

+++ b/celt/kiss_fft.c

@@ -405,7 +405,13 @@

         kiss_twiddle_cpx *twiddles;

         st->nfft=nfft;

-#ifndef FIXED_POINT

+#ifdef FIXED_POINT

+        st->scale_shift = celt_ilog2(st->nfft);

+        if (st->nfft == 1<<st->scale_shift)

+           st->scale = Q15ONE;

+        else

+           st->scale = (1073741824+st->nfft/2)/st->nfft>>(15-st->scale_shift);

+#else

         st->scale = 1.f/nfft;

 #endif

         if (base != NULL)

@@ -507,16 +513,14 @@

 void opus_fft(const kiss_fft_state *st,const kiss_fft_cpx *fin,kiss_fft_cpx *fout)

 {

    int i;

-#ifdef FIXED_POINT

-   /* FIXME: This should eventually just go in the state. */

    opus_val16 scale;

-   int scale_shift;

-   scale_shift = celt_ilog2(st->nfft);

-   if (st->nfft == 1<<scale_shift)

-      scale = Q15ONE;

-   else

-      scale = (1073741824+st->nfft/2)/st->nfft>>(15-scale_shift);

+#ifdef FIXED_POINT

+   /* Allows us to scale with MULT16_32_Q16(), which is faster than

+      MULT16_32_Q15() on ARM. */

+   int scale_shift = st->scale_shift-1;

 #endif

+   SAVE_STACK;

+   scale = st->scale;

    celt_assert2 (fin != fout, "In-place FFT not supported");

    /* Bit-reverse the input */

@@ -523,13 +527,8 @@

    for (i=0;i<st->nfft;i++)

    {

       kiss_fft_cpx x = fin[i];

-#ifdef FIXED_POINT

-      fout[st->bitrev[i]].r = SHR32(MULT16_32_Q15(scale, x.r), scale_shift);

-      fout[st->bitrev[i]].i = SHR32(MULT16_32_Q15(scale, x.i), scale_shift);

-#else

-      fout[st->bitrev[i]].r = st->scale*x.r;

-      fout[st->bitrev[i]].i = st->scale*x.i;

-#endif

+      fout[st->bitrev[i]].r = SHR32(MULT16_32_Q16(scale, x.r), scale_shift);

+      fout[st->bitrev[i]].i = SHR32(MULT16_32_Q16(scale, x.i), scale_shift);

    }

    opus_fft_impl(st, fout);

 }

--- a/celt/kiss_fft.h

+++ b/celt/kiss_fft.h

@@ -79,8 +79,9 @@

 typedef struct kiss_fft_state{

     int nfft;

-#ifndef FIXED_POINT

-    kiss_fft_scalar scale;

+    opus_val16 scale;

+#ifdef FIXED_POINT

+    int scale_shift;

 #endif

     int shift;

     opus_int16 factors[2*MAXFACTORS];

--- a/celt/mdct.c

+++ b/celt/mdct.c

@@ -121,22 +121,12 @@

    const kiss_twiddle_scalar *trig;

    opus_val16 scale;

 #ifdef FIXED_POINT

-   int scale_shift;

-#endif

-   SAVE_STACK;

-#ifdef FIXED_POINT

-   /* FIXME: This should eventually just go in the state. */

-   scale_shift = celt_ilog2(st->nfft);

-   if (st->nfft == 1<<scale_shift)

-      scale = Q15ONE;

-   else

-      scale = (1073741824+st->nfft/2)/st->nfft>>(15-scale_shift);

    /* Allows us to scale with MULT16_32_Q16(), which is faster than

       MULT16_32_Q15() on ARM. */

-   scale_shift--;

-#else

-   scale = st->scale;

+   int scale_shift = st->scale_shift-1;

 #endif

+   SAVE_STACK;

+   scale = st->scale;

    N = l->n;

    trig = l->trig;

--- a/celt/static_modes_fixed.h

+++ b/celt/static_modes_fixed.h

@@ -426,6 +426,8 @@

 #define FFT_STATE48000_960_0

 static const kiss_fft_state fft_state48000_960_0 = {

 480,    /* nfft */

+17476,	/* scale */

+8,      /* scale_shift */

 -1,     /* shift */

 {5, 96, 3, 32, 2, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, },	/* factors */

 fft_bitrev480,  /* bitrev */

@@ -437,6 +439,8 @@

 #define FFT_STATE48000_960_1

 static const kiss_fft_state fft_state48000_960_1 = {

 240,    /* nfft */

+17476,	/* scale */

+7,      /* scale_shift */

 1,      /* shift */

 {5, 48, 3, 16, 4, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, },	/* factors */

 fft_bitrev240,  /* bitrev */

@@ -448,6 +452,8 @@

 #define FFT_STATE48000_960_2

 static const kiss_fft_state fft_state48000_960_2 = {

 120,    /* nfft */

+17476,	/* scale */

+6,      /* scale_shift */

 2,      /* shift */

 {5, 24, 3, 8, 2, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, },	/* factors */

 fft_bitrev120,  /* bitrev */

@@ -459,6 +465,8 @@

 #define FFT_STATE48000_960_3

 static const kiss_fft_state fft_state48000_960_3 = {

 60,     /* nfft */

+17476,	/* scale */

+5,      /* scale_shift */

 3,      /* shift */

 {5, 12, 3, 4, 4, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, },	/* factors */

 fft_bitrev60,   /* bitrev */