libavcodec: Implementation of 32 bit fixed point FFT

Iterative implementation of 32 bit fixed point split-radix FFT.
Max FFT that can be calculated currently is 2^12.
Signed-off-by: Nedeljko Babic <nbabic@mips.com>
Signed-off-by: Michael Niedermayer <michaelni@gmx.at>

doc/mips.txt

@@ -47,6 +47,11 @@ Files that have MIPS copyright notice in them:
 * libavutil/mips/
       float_dsp_mips.c
       libm_mips.h
+* libavcodec/
+      fft_fixed_32.c
+      fft_init_table.c
+      fft_table.h
+      mdct_fixed_32.c
 * libavcodec/mips/
       aaccoder_mips.c
       aacpsy_mips.h

libavcodec/Makefile

@@ -44,6 +44,7 @@ OBJS-$(CONFIG_ENCODERS)                += faandct.o jfdctfst.o jfdctint.o
 OBJS-$(CONFIG_ERROR_RESILIENCE)        += error_resilience.o
 FFT-OBJS-$(CONFIG_HARDCODED_TABLES)    += cos_tables.o cos_fixed_tables.o
 OBJS-$(CONFIG_FFT)                     += avfft.o fft_fixed.o fft_float.o \
+                                          fft_fixed_32.o fft_init_table.o \
                                           $(FFT-OBJS-yes)
 OBJS-$(CONFIG_GOLOMB)                  += golomb.o
 OBJS-$(CONFIG_H264CHROMA)              += h264chroma.o
@@ -55,7 +56,7 @@ OBJS-$(CONFIG_HUFFMAN)                 += huffman.o
 OBJS-$(CONFIG_LIBXVID)                 += libxvid_rc.o
 OBJS-$(CONFIG_LPC)                     += lpc.o
 OBJS-$(CONFIG_LSP)                     += lsp.o
-OBJS-$(CONFIG_MDCT)                    += mdct_fixed.o mdct_float.o
+OBJS-$(CONFIG_MDCT)                    += mdct_fixed.o mdct_float.o mdct_fixed_32.o
 OBJS-$(CONFIG_MPEGAUDIO)               += mpegaudio.o mpegaudiodata.o   \
                                           mpegaudiodecheader.o
 OBJS-$(CONFIG_MPEGAUDIODSP)            += mpegaudiodsp.o                \
@@ -811,6 +812,7 @@ TESTPROGS = cabac                                                       \
             dct                                                         \
             fft                                                         \
             fft-fixed                                                   \
+            fft-fixed32                                                 \
             golomb                                                      \
             iirfilter                                                   \
             imgconvert                                                  \

libavcodec/fft-fixed32-test.c

+/*
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#define CONFIG_FFT_FLOAT 0
+#define CONFIG_FFT_FIXED_32 1
+#include "fft-test.c"

libavcodec/fft-internal.h

@@ -36,12 +36,29 @@
 
 #else
 
+#define SCALE_FLOAT(a, bits) lrint((a) * (double)(1 << (bits)))
+
+#if CONFIG_FFT_FIXED_32
+
+#define CMUL(dre, dim, are, aim, bre, bim) do {             \
+        int64_t accu;                                     \
+        (accu)  = (int64_t)(bre) * (are);                 \
+        (accu) -= (int64_t)(bim) * (aim);                 \
+        (dre)   = (int)(((accu) + 0x40000000) >> 31);       \
+        (accu)  = (int64_t)(bre) * (aim);                 \
+        (accu) += (int64_t)(bim) * (are);                 \
+        (dim)   = (int)(((accu) + 0x40000000) >> 31);       \
+    } while (0)
+
+#define FIX15(a) av_clip(SCALE_FLOAT(a, 31), -2147483647, 2147483647)
+
+#else /* CONFIG_FFT_FIXED_32 */
+
 #include "fft.h"
 #include "mathops.h"
 
 void ff_mdct_calcw_c(FFTContext *s, FFTDouble *output, const FFTSample *input);
 
-#define SCALE_FLOAT(a, bits) lrint((a) * (double)(1 << (bits)))
 #define FIX15(a) av_clip(SCALE_FLOAT(a, 15), -32767, 32767)
 
 #define sqrthalf ((int16_t)((1<<15)*M_SQRT1_2))
@@ -62,6 +79,8 @@ void ff_mdct_calcw_c(FFTContext *s, FFTDouble *output, const FFTSample *input);
 #define CMULL(dre, dim, are, aim, bre, bim)     \
     CMULS(dre, dim, are, aim, bre, bim, 0)
 
+#endif /* CONFIG_FFT_FIXED_32 */
+
 #endif /* CONFIG_FFT_FLOAT */
 
 #define ff_imdct_calc_c FFT_NAME(ff_imdct_calc_c)

libavcodec/fft-test.c

@@ -54,6 +54,10 @@
 #   define RANGE 1.0
 #   define REF_SCALE(x, bits)  (x)
 #   define FMT "%10.6f"
+#elif CONFIG_FFT_FIXED_32
+#   define RANGE 8388608
+#   define REF_SCALE(x, bits) (x)
+#   define FMT "%6d"
 #else
 #   define RANGE 16384
 #   define REF_SCALE(x, bits) ((x) / (1<<(bits)))

libavcodec/fft.c

@@ -32,6 +32,10 @@
 #include "fft.h"
 #include "fft-internal.h"
 
+#if CONFIG_FFT_FIXED_32
+#include "fft_table.h"
+#else /* CONFIG_FFT_FIXED_32 */
+
 /* cos(2*pi*x/n) for 0<=x<=n/4, followed by its reverse */
 #if !CONFIG_HARDCODED_TABLES
 COSTABLE(16);
@@ -65,6 +69,8 @@ COSTABLE_CONST FFTSample * const FFT_NAME(ff_cos_tabs)[] = {
     FFT_NAME(ff_cos_65536),
 };
 
+#endif /* CONFIG_FFT_FIXED_32 */
+
 static void fft_permute_c(FFTContext *s, FFTComplex *z);
 static void fft_calc_c(FFTContext *s, FFTComplex *z);
 
@@ -81,7 +87,7 @@ static int split_radix_permutation(int i, int n, int inverse)
 
 av_cold void ff_init_ff_cos_tabs(int index)
 {
-#if !CONFIG_HARDCODED_TABLES
+#if (!CONFIG_HARDCODED_TABLES) && (!CONFIG_FFT_FIXED_32)
     int i;
     int m = 1<<index;
     double freq = 2*M_PI/m;
@@ -157,6 +163,12 @@ av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
     s->mdct_calc   = ff_mdct_calc_c;
 #endif
 
+#if CONFIG_FFT_FIXED_32
+    {
+        int n=0;
+        ff_fft_lut_init(fft_offsets_lut, 0, 1 << 16, &n);
+    }
+#else /* CONFIG_FFT_FIXED_32 */
 #if CONFIG_FFT_FLOAT
     if (ARCH_ARM)     ff_fft_init_arm(s);
     if (ARCH_PPC)     ff_fft_init_ppc(s);
@@ -167,10 +179,11 @@ av_cold int ff_fft_init(FFTContext *s, int nbits, int inverse)
     if (CONFIG_MDCT)  s->mdct_calcw = ff_mdct_calcw_c;
     if (ARCH_ARM)     ff_fft_fixed_init_arm(s);
 #endif
-
     for(j=4; j<=nbits; j++) {
         ff_init_ff_cos_tabs(j);
     }
+#endif /* CONFIG_FFT_FIXED_32 */
+
 
     if (s->fft_permutation == FF_FFT_PERM_AVX) {
         fft_perm_avx(s);
@@ -206,6 +219,169 @@ av_cold void ff_fft_end(FFTContext *s)
     av_freep(&s->tmp_buf);
 }
 
+#if CONFIG_FFT_FIXED_32
+
+static void fft_calc_c(FFTContext *s, FFTComplex *z) {
+
+    int nbits, i, n, num_transforms, offset, step;
+    int n4, n2, n34;
+    FFTSample tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
+    FFTComplex *tmpz;
+    FFTSample w_re, w_im;
+    FFTSample *w_re_ptr, *w_im_ptr;
+    const int fft_size = (1 << s->nbits);
+    int64_t accu;
+
+    num_transforms = (0x2aab >> (16 - s->nbits)) | 1;
+
+    for (n=0; n<num_transforms; n++){
+        offset = fft_offsets_lut[n] << 2;
+        tmpz = z + offset;
+
+        tmp1 = tmpz[0].re + tmpz[1].re;
+        tmp5 = tmpz[2].re + tmpz[3].re;
+        tmp2 = tmpz[0].im + tmpz[1].im;
+        tmp6 = tmpz[2].im + tmpz[3].im;
+        tmp3 = tmpz[0].re - tmpz[1].re;
+        tmp8 = tmpz[2].im - tmpz[3].im;
+        tmp4 = tmpz[0].im - tmpz[1].im;
+        tmp7 = tmpz[2].re - tmpz[3].re;
+
+        tmpz[0].re = tmp1 + tmp5;
+        tmpz[2].re = tmp1 - tmp5;
+        tmpz[0].im = tmp2 + tmp6;
+        tmpz[2].im = tmp2 - tmp6;
+        tmpz[1].re = tmp3 + tmp8;
+        tmpz[3].re = tmp3 - tmp8;
+        tmpz[1].im = tmp4 - tmp7;
+        tmpz[3].im = tmp4 + tmp7;
+    }
+
+    if (fft_size < 8)
+        return;
+
+    num_transforms = (num_transforms >> 1) | 1;
+
+    for (n=0; n<num_transforms; n++){
+        offset = fft_offsets_lut[n] << 3;
+        tmpz = z + offset;
+
+        tmp1 = tmpz[4].re + tmpz[5].re;
+        tmp3 = tmpz[6].re + tmpz[7].re;
+        tmp2 = tmpz[4].im + tmpz[5].im;
+        tmp4 = tmpz[6].im + tmpz[7].im;
+        tmp5 = tmp1 + tmp3;
+        tmp7 = tmp1 - tmp3;
+        tmp6 = tmp2 + tmp4;
+        tmp8 = tmp2 - tmp4;
+
+        tmp1 = tmpz[4].re - tmpz[5].re;
+        tmp2 = tmpz[4].im - tmpz[5].im;
+        tmp3 = tmpz[6].re - tmpz[7].re;
+        tmp4 = tmpz[6].im - tmpz[7].im;
+
+        tmpz[4].re = tmpz[0].re - tmp5;
+        tmpz[0].re = tmpz[0].re + tmp5;
+        tmpz[4].im = tmpz[0].im - tmp6;
+        tmpz[0].im = tmpz[0].im + tmp6;
+        tmpz[6].re = tmpz[2].re - tmp8;
+        tmpz[2].re = tmpz[2].re + tmp8;
+        tmpz[6].im = tmpz[2].im + tmp7;
+        tmpz[2].im = tmpz[2].im - tmp7;
+
+        accu = (int64_t)Q31(M_SQRT1_2)*(tmp1 + tmp2);
+        tmp5 = (int32_t)((accu + 0x40000000) >> 31);
+        accu = (int64_t)Q31(M_SQRT1_2)*(tmp3 - tmp4);
+        tmp7 = (int32_t)((accu + 0x40000000) >> 31);
+        accu = (int64_t)Q31(M_SQRT1_2)*(tmp2 - tmp1);
+        tmp6 = (int32_t)((accu + 0x40000000) >> 31);
+        accu = (int64_t)Q31(M_SQRT1_2)*(tmp3 + tmp4);
+        tmp8 = (int32_t)((accu + 0x40000000) >> 31);
+        tmp1 = tmp5 + tmp7;
+        tmp3 = tmp5 - tmp7;
+        tmp2 = tmp6 + tmp8;
+        tmp4 = tmp6 - tmp8;
+
+        tmpz[5].re = tmpz[1].re - tmp1;
+        tmpz[1].re = tmpz[1].re + tmp1;
+        tmpz[5].im = tmpz[1].im - tmp2;
+        tmpz[1].im = tmpz[1].im + tmp2;
+        tmpz[7].re = tmpz[3].re - tmp4;
+        tmpz[3].re = tmpz[3].re + tmp4;
+        tmpz[7].im = tmpz[3].im + tmp3;
+        tmpz[3].im = tmpz[3].im - tmp3;
+    }
+
+    step = 1 << ((MAX_LOG2_NFFT-4) - 4);
+    n4 = 4;
+
+    for (nbits=4; nbits<=s->nbits; nbits++){
+        n2  = 2*n4;
+        n34 = 3*n4;
+        num_transforms = (num_transforms >> 1) | 1;
+
+        for (n=0; n<num_transforms; n++){
+            offset = fft_offsets_lut[n] << nbits;
+            tmpz = z + offset;
+
+            tmp5 = tmpz[ n2].re + tmpz[n34].re;
+            tmp1 = tmpz[ n2].re - tmpz[n34].re;
+            tmp6 = tmpz[ n2].im + tmpz[n34].im;
+            tmp2 = tmpz[ n2].im - tmpz[n34].im;
+
+            tmpz[ n2].re = tmpz[ 0].re - tmp5;
+            tmpz[  0].re = tmpz[ 0].re + tmp5;
+            tmpz[ n2].im = tmpz[ 0].im - tmp6;
+            tmpz[  0].im = tmpz[ 0].im + tmp6;
+            tmpz[n34].re = tmpz[n4].re - tmp2;
+            tmpz[ n4].re = tmpz[n4].re + tmp2;
+            tmpz[n34].im = tmpz[n4].im + tmp1;
+            tmpz[ n4].im = tmpz[n4].im - tmp1;
+
+            w_re_ptr = w_tab_sr + step;
+            w_im_ptr = w_tab_sr + MAX_FFT_SIZE/(4*16) - step;
+
+            for (i=1; i<n4; i++){
+                w_re = w_re_ptr[0];
+                w_im = w_im_ptr[0];
+                accu  = (int64_t)w_re*tmpz[ n2+i].re;
+                accu += (int64_t)w_im*tmpz[ n2+i].im;
+                tmp1 = (int32_t)((accu + 0x40000000) >> 31);
+                accu  = (int64_t)w_re*tmpz[ n2+i].im;
+                accu -= (int64_t)w_im*tmpz[ n2+i].re;
+                tmp2 = (int32_t)((accu + 0x40000000) >> 31);
+                accu  = (int64_t)w_re*tmpz[n34+i].re;
+                accu -= (int64_t)w_im*tmpz[n34+i].im;
+                tmp3 = (int32_t)((accu + 0x40000000) >> 31);
+                accu  = (int64_t)w_re*tmpz[n34+i].im;
+                accu += (int64_t)w_im*tmpz[n34+i].re;
+                tmp4 = (int32_t)((accu + 0x40000000) >> 31);
+
+                tmp5 = tmp1 + tmp3;
+                tmp1 = tmp1 - tmp3;
+                tmp6 = tmp2 + tmp4;
+                tmp2 = tmp2 - tmp4;
+
+                tmpz[ n2+i].re = tmpz[   i].re - tmp5;
+                tmpz[    i].re = tmpz[   i].re + tmp5;
+                tmpz[ n2+i].im = tmpz[   i].im - tmp6;
+                tmpz[    i].im = tmpz[   i].im + tmp6;
+                tmpz[n34+i].re = tmpz[n4+i].re - tmp2;
+                tmpz[ n4+i].re = tmpz[n4+i].re + tmp2;
+                tmpz[n34+i].im = tmpz[n4+i].im + tmp1;
+                tmpz[ n4+i].im = tmpz[n4+i].im - tmp1;
+
+                w_re_ptr += step;
+                w_im_ptr -= step;
+            }
+        }
+        step >>= 1;
+        n4   <<= 1;
+    }
+}
+
+#else /* CONFIG_FFT_FIXED_32 */
+
 #define BUTTERFLIES(a0,a1,a2,a3) {\
     BF(t3, t5, t5, t1);\
     BF(a2.re, a0.re, a0.re, t5);\
@@ -351,3 +527,4 @@ static void fft_calc_c(FFTContext *s, FFTComplex *z)
 {
     fft_dispatch[s->nbits-2](z);
 }
+#endif /* CONFIG_FFT_FIXED_32 */

libavcodec/fft.h

@@ -26,6 +26,10 @@
 #define CONFIG_FFT_FLOAT 1
 #endif
 
+#ifndef CONFIG_FFT_FIXED_32
+#define CONFIG_FFT_FIXED_32 0
+#endif
+
 #include <stdint.h>
 #include "config.h"
 #include "libavutil/mem.h"
@@ -40,15 +44,26 @@ typedef float FFTDouble;
 
 #else
 
+#if CONFIG_FFT_FIXED_32
+
+#define Q31(x) (int)((x)*2147483648.0 + 0.5)
+#define FFT_NAME(x) x ## _fixed_32
+
+typedef int32_t FFTSample;
+
+#else /* CONFIG_FFT_FIXED_32 */
+
 #define FFT_NAME(x) x ## _fixed
 
 typedef int16_t FFTSample;
-typedef int     FFTDouble;
+
+#endif /* CONFIG_FFT_FIXED_32 */
 
 typedef struct FFTComplex {
-    int16_t re, im;
+    FFTSample re, im;
 } FFTComplex;
 
+typedef int    FFTDouble;
 typedef struct FFTContext FFTContext;
 
 #endif /* CONFIG_FFT_FLOAT */

libavcodec/fft_fixed.c

@@ -17,4 +17,5 @@
  */
 
 #define CONFIG_FFT_FLOAT 0
+#define CONFIG_FFT_FIXED_32 0
 #include "fft.c"

libavcodec/mips/fft_init_table.c → libavcodec/fft_fixed_32.c

@@ -26,7 +26,9 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- * Author:  Stanislav Ocovaj (socovaj@mips.com)
+ * Authors:  Stanislav Ocovaj (socovaj@mips.com)
+ *           Goran Cordasic   (goran@mips.com)
+ *           Djordje Pesut    (djordje@mips.com)
  *
  * This file is part of FFmpeg.
  *
@@ -45,23 +47,6 @@
  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  */
 
-/**
- * @file
- * definitions and initialization of LUT table for MIPS FFT
- */
-#include "fft_table.h"
-
-uint16_t fft_offsets_lut[0x2aab];
-
-void ff_fft_lut_init(uint16_t *table, int off, int size, int *index)
-{
-    if (size < 16) {
-        table[*index] = off >> 2;
-        (*index)++;
-    }
-    else {
-        ff_fft_lut_init(table, off, size>>1, index);
-        ff_fft_lut_init(table, off+(size>>1), size>>2, index);
-        ff_fft_lut_init(table, off+3*(size>>2), size>>2, index);
-    }
-}
+#define CONFIG_FFT_FLOAT 0
+#define CONFIG_FFT_FIXED_32 1
+#include "fft.c"

libavcodec/fft_float.c

@@ -17,4 +17,5 @@
  */
 
 #define CONFIG_FFT_FLOAT 1
+#define CONFIG_FFT_FIXED_32 0
 #include "fft.c"

libavcodec/mips/fft_table.h → libavcodec/fft_table.h

@@ -26,7 +26,9 @@
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  *
- * Author:  Stanislav Ocovaj (socovaj@mips.com)
+ * Authors:  Stanislav Ocovaj (socovaj@mips.com)
+ *           Goran Cordasic   (goran@mips.com)
+ *           Djordje Pesut    (djordje@mips.com)
  *
  * This file is part of FFmpeg.
  *
@@ -47,17 +49,18 @@
 
 /**
  * @file
- * definitions and LUT table for MIPS FFT
+ * definitions and tables for FFT
  */
-#ifndef AVCODEC_MIPS_FFT_TABLE_H
-#define AVCODEC_MIPS_FFT_TABLE_H
+#ifndef AVCODEC_FFT_TABLE_H
+#define AVCODEC_FFT_TABLE_H
 
 #include "libavcodec/fft.h"
 
 #define MAX_LOG2_NFFT 16 //!< Specifies maxiumum allowed fft size
 #define MAX_FFT_SIZE (1 << MAX_LOG2_NFFT)
 
+extern int32_t w_tab_sr[];
 extern uint16_t fft_offsets_lut[];
 void ff_fft_lut_init(uint16_t *table, int off, int size, int *index);
 
-#endif /* AVCODEC_MIPS_FFT_TABLE_H */
+#endif /* AVCODEC_FFT_TABLE_H */

libavcodec/mdct.c

@@ -34,7 +34,11 @@
 #if CONFIG_FFT_FLOAT
 #   define RSCALE(x) (x)
 #else
+#if CONFIG_FFT_FIXED_32
+#   define RSCALE(x) (((x) + 32) >> 6)
+#else /* CONFIG_FFT_FIXED_32 */
 #   define RSCALE(x) ((x) >> 1)
+#endif /* CONFIG_FFT_FIXED_32 */
 #endif
 
 /**

libavcodec/mdct_fixed.c

@@ -17,6 +17,7 @@
  */
 
 #define CONFIG_FFT_FLOAT 0
+#define CONFIG_FFT_FIXED_32 0
 #include "mdct.c"
 
 /* same as ff_mdct_calcw_c with double-width unscaled output */

libavcodec/mdct_fixed_32.c

+/*
+ * Copyright (c) 2012
+ *      MIPS Technologies, Inc., California.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. Neither the name of the MIPS Technologies, Inc., nor the names of its
+ *    contributors may be used to endorse or promote products derived from
+ *    this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE MIPS TECHNOLOGIES, INC. ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE MIPS TECHNOLOGIES, INC. BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ * Authors:  Stanislav Ocovaj (socovaj@mips.com)
+ *           Goran Cordasic   (goran@mips.com)
+ *           Djordje Pesut    (djordje@mips.com)
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg 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.
+ *
+ * FFmpeg 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 FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#define CONFIG_FFT_FLOAT 0
+#define CONFIG_FFT_FIXED_32 1
+#include "mdct.c"

libavcodec/mdct_float.c

@@ -17,4 +17,5 @@
  */
 
 #define CONFIG_FFT_FLOAT 1
+#define CONFIG_FFT_FIXED_32 0
 #include "mdct.c"

libavcodec/mips/Makefile

@@ -9,7 +9,6 @@ MIPSFPU-OBJS-$(CONFIG_AMRWB_DECODER)      += mips/acelp_filters_mips.o     \
                                              mips/acelp_vectors_mips.o
 MIPSFPU-OBJS-$(CONFIG_MPEGAUDIODSP)       += mips/mpegaudiodsp_mips_float.o
 MIPSDSPR1-OBJS-$(CONFIG_MPEGAUDIODSP)     += mips/mpegaudiodsp_mips_fixed.o
-OBJS-$(CONFIG_FFT)                        += mips/fft_init_table.o
 MIPSFPU-OBJS-$(CONFIG_FFT)                += mips/fft_mips.o
 MIPSFPU-OBJS                              += mips/fmtconvert_mips.o
 OBJS-$(CONFIG_AC3DSP)                     += mips/ac3dsp_mips.o

libavcodec/mips/fft_mips.c

@@ -49,7 +49,7 @@
  */
 #include "config.h"
 #include "libavcodec/fft.h"
-#include "fft_table.h"
+#include "libavcodec/fft_table.h"
 
 /**
  * FFT transform

tests/fate/fft.mak

@@ -38,5 +38,22 @@ $(FATE_FFT_FIXED): libavcodec/fft-fixed-test$(EXESUF)
 $(FATE_FFT_FIXED): CMD = run libavcodec/fft-fixed-test $(CPUFLAGS:%=-c%) $(ARGS)
 $(FATE_FFT_FIXED): REF = /dev/null
 
-FATE-$(call ALLYES, AVCODEC FFT) += $(FATE_FFT) $(FATE_FFT_FIXED)
-fate-fft: $(FATE_FFT) $(FATE_FFT_FIXED)
+define DEF_FFT_FIXED32
+FATE_FFT_FIXED32 += fate-fft-fixed32-$(1)   fate-ifft-fixed32-$(1)  \
+                  fate-mdct-fixed32-$(1) fate-imdct-fixed32-$(1)
+
+fate-fft-fixed32-$(1):   ARGS = -n$(1)
+fate-ifft-fixed32-$(1):  ARGS = -n$(1) -i
+#fate-mdct-fixed32-$(1):  ARGS = -n$(1) -m
+fate-imdct-fixed32-$(1): ARGS = -n$(1) -m -i
+endef
+
+$(foreach N, 4 5 6 7 8 9 10 11 12, $(eval $(call DEF_FFT_FIXED32,$(N))))
+
+fate-fft-fixed32-test: $(FATE_FFT_FIXED32)
+$(FATE_FFT_FIXED32): libavcodec/fft-fixed32-test$(EXESUF)
+$(FATE_FFT_FIXED32): CMD = run libavcodec/fft-fixed32-test $(CPUFLAGS:%=-c%) $(ARGS)
+$(FATE_FFT_FIXED32): REF = /dev/null
+
+FATE-$(call ALLYES, AVCODEC FFT) += $(FATE_FFT) $(FATE_FFT_FIXED) $(FATE_FFT_FIXED32)
+fate-fft: $(FATE_FFT) $(FATE_FFT_FIXED) $(FATE_FFT_FIXED32)