Merge remote branch 'qatar/master'

* qatar/master:
  ac3enc: ARM optimised ac3_compute_matissa_size
  ac3: armv6 optimised bit_alloc_calc_bap
  fate: simplify fft test rules
  avio: document avio_alloc_context.
  lavf: make compute_chapters_end less picky.
  sierravmd: fix Indeo3 videos
  FFT: simplify fft8()
  fate: add fixed-point fft/mdct tests
  Fixed-point support in fft-test
  ape: check that number of seektable entries is equal to number of frames

Merged-by: Michael Niedermayer <michaelni@gmx.at>

8 files changed, 229 insertions, 33 deletions

diff --git a/libavcodec/Makefile b/libavcodec/Makefile
index 7f771be2b1..0730124870 100644
--- a/libavcodec/Makefile
+++ b/libavcodec/Makefile
@@ -661,7 +661,7 @@ SKIPHEADERS                            += mpegaudio3.h
 
 EXAMPLES = api
 
-TESTPROGS = cabac dct eval fft h264 iirfilter rangecoder snow
+TESTPROGS = cabac dct eval fft fft-fixed h264 iirfilter rangecoder snow
 TESTPROGS-$(HAVE_MMX) += motion
 TESTOBJS = dctref.o
 
diff --git a/libavcodec/arm/Makefile b/libavcodec/arm/Makefile
index 08697da29b..3b77a5548d 100644
--- a/libavcodec/arm/Makefile
+++ b/libavcodec/arm/Makefile
@@ -1,6 +1,10 @@
-OBJS-$(CONFIG_AC3DSP)                  += arm/ac3dsp_init_arm.o
+OBJS-$(CONFIG_AC3DSP)                  += arm/ac3dsp_init_arm.o         \
+                                          arm/ac3dsp_arm.o
+
 OBJS-$(CONFIG_DCA_DECODER)             += arm/dcadsp_init_arm.o         \
 
+ARMV6-OBJS-$(CONFIG_AC3DSP)            += arm/ac3dsp_armv6.o
+
 OBJS-$(CONFIG_VP5_DECODER)             += arm/vp56dsp_init_arm.o
 OBJS-$(CONFIG_VP6_DECODER)             += arm/vp56dsp_init_arm.o
 OBJS-$(CONFIG_VP8_DECODER)             += arm/vp8dsp_init_arm.o
diff --git a/libavcodec/arm/ac3dsp_arm.S b/libavcodec/arm/ac3dsp_arm.S
new file mode 100644
index 0000000000..d7d498e41f
--- /dev/null
+++ b/libavcodec/arm/ac3dsp_arm.S
@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2011 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 "asm.S"
+
+function ff_ac3_compute_mantissa_size_arm, export=1
+        push            {r4-r8,lr}
+        ldm             r0,  {r4-r8}
+        mov             r3,  r0
+        mov             r0,  #0
+1:
+        ldrb            lr,  [r1], #1
+        subs            r2,  r2,  #1
+        blt             2f
+        cmp             lr,  #4
+        bgt             3f
+        subs            lr,  lr,  #1
+        addlt           r4,  r4,  #1
+        addeq           r5,  r5,  #1
+        ble             1b
+        subs            lr,  lr,  #2
+        addlt           r6,  r6,  #1
+        addeq           r7,  r7,  #1
+        addgt           r8,  r8,  #1
+        b               1b
+3:
+        cmp             lr,  #14
+        sublt           lr,  lr,  #1
+        addgt           r0,  r0,  #16
+        addle           r0,  r0,  lr
+        b               1b
+2:
+        stm             r3,  {r4-r8}
+        pop             {r4-r8,pc}
+endfunc
diff --git a/libavcodec/arm/ac3dsp_armv6.S b/libavcodec/arm/ac3dsp_armv6.S
new file mode 100644
index 0000000000..7f01addbde
--- /dev/null
+++ b/libavcodec/arm/ac3dsp_armv6.S
@@ -0,0 +1,83 @@
+/*
+ * Copyright (c) 2011 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 "asm.S"
+
+function ff_ac3_bit_alloc_calc_bap_armv6, export=1
+        ldr             r12, [sp]
+        cmp             r12, #-960
+        beq             4f
+        push            {r4-r11,lr}
+        add             r5,  sp,  #40
+        movrel          r4,  X(ff_ac3_bin_to_band_tab)
+        movrel          lr,  X(ff_ac3_band_start_tab)
+        ldm             r5,  {r5-r7}
+        ldrb            r4,  [r4, r2]
+        add             r1,  r1,  r2,  lsl #1           @ psd + start
+        add             r0,  r0,  r4,  lsl #1           @ mask + band
+        add             r4,  lr,  r4
+        add             r7,  r7,  r2                    @ bap + start
+        ldrb            r10, [r4], #1
+1:
+        ldrsh           r9,  [r0], #2                   @ mask[band]
+        movw            r8,  #0x1fe0
+        sub             r9,  r9,  r12                   @   - snr_offset
+        mov             r11, r10
+        ldrb            r10, [r4], #1                   @ band_start_tab[band++]
+        subs            r9,  r9,  r5                    @   - floor
+        movlt           r9,  #0
+        cmp             r10, r3                         @   - end
+        and             r9,  r9,  r8                    @   & 0x1fe0
+        subgt           r8,  r3,  r11
+        suble           r8,  r10, r11
+        add             r9,  r9,  r5                    @   + floor => m
+        tst             r8,  #1
+        add             r2,  r7,  r8
+        bne             3f
+        b               5f
+2:
+        ldrsh           r8,  [r1], #2
+        ldrsh           lr,  [r1], #2
+        sub             r8,  r8,  r9
+        sub             lr,  lr,  r9
+        usat            r8,  #6,  r8,  asr #5           @ address
+        usat            lr,  #6,  lr,  asr #5
+        ldrb            r8,  [r6, r8]                   @ bap_tab[address]
+        ldrb            lr,  [r6, lr]
+        strb            r8,  [r7], #1                   @ bap[bin]
+        strb            lr,  [r7], #1
+5:      cmp             r7,  r2
+        blo             2b
+        cmp             r3,  r11
+        bgt             1b
+        pop             {r4-r11,pc}
+3:
+        ldrsh           r8,  [r1], #2                   @ psd[bin]
+        sub             r8,  r8,  r9                    @   - m
+        usat            r8,  #6,  r8,  asr #5           @ address
+        ldrb            r8,  [r6, r8]                   @ bap_tab[address]
+        strb            r8,  [r7], #1                   @ bap[bin]
+        b               5b
+4:
+        ldr             r0,  [sp, #12]
+        mov             r1,  #0
+        mov             r2,  #256
+        b               memset
+endfunc
diff --git a/libavcodec/arm/ac3dsp_init_arm.c b/libavcodec/arm/ac3dsp_init_arm.c
index 0dfeaa9c80..8534c9b97c 100644
--- a/libavcodec/arm/ac3dsp_init_arm.c
+++ b/libavcodec/arm/ac3dsp_init_arm.c
@@ -29,8 +29,21 @@ void ff_ac3_lshift_int16_neon(int16_t *src, unsigned len, unsigned shift);
 void ff_ac3_rshift_int32_neon(int32_t *src, unsigned len, unsigned shift);
 void ff_float_to_fixed24_neon(int32_t *dst, const float *src, unsigned int len);
 
+void ff_ac3_bit_alloc_calc_bap_armv6(int16_t *mask, int16_t *psd,
+                                     int start, int end,
+                                     int snr_offset, int floor,
+                                     const uint8_t *bap_tab, uint8_t *bap);
+
+int ff_ac3_compute_mantissa_size_arm(int cnt[5], uint8_t *bap, int nb_coefs);
+
 av_cold void ff_ac3dsp_init_arm(AC3DSPContext *c, int bit_exact)
 {
+    c->compute_mantissa_size     = ff_ac3_compute_mantissa_size_arm;
+
+    if (HAVE_ARMV6) {
+        c->bit_alloc_calc_bap    = ff_ac3_bit_alloc_calc_bap_armv6;
+    }
+
     if (HAVE_NEON) {
         c->ac3_exponent_min      = ff_ac3_exponent_min_neon;
         c->ac3_max_msb_abs_int16 = ff_ac3_max_msb_abs_int16_neon;
diff --git a/libavcodec/fft-fixed-test.c b/libavcodec/fft-fixed-test.c
new file mode 100644
index 0000000000..fa750b6326
--- /dev/null
+++ b/libavcodec/fft-fixed-test.c
@@ -0,0 +1,20 @@
+/*
+ * 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
+#include "fft-test.c"
diff --git a/libavcodec/fft-test.c b/libavcodec/fft-test.c
index 9ad8793c2b..a676627de2 100644
--- a/libavcodec/fft-test.c
+++ b/libavcodec/fft-test.c
@@ -27,8 +27,10 @@
 #include "libavutil/lfg.h"
 #include "libavutil/log.h"
 #include "fft.h"
+#if CONFIG_FFT_FLOAT
 #include "dct.h"
 #include "rdft.h"
+#endif
 #include <math.h>
 #include <unistd.h>
 #include <sys/time.h>
@@ -47,7 +49,19 @@
    pim += (MUL16(are, bim) + MUL16(bre, aim));\
 }
 
-FFTComplex *exptab;
+#if CONFIG_FFT_FLOAT
+#   define RANGE 1.0
+#   define REF_SCALE(x, bits)  (x)
+#   define FMT "%10.6f"
+#else
+#   define RANGE 16384
+#   define REF_SCALE(x, bits) ((x) / (1<<(bits)))
+#   define FMT "%6d"
+#endif
+
+struct {
+    float re, im;
+} *exptab;
 
 static void fft_ref_init(int nbits, int inverse)
 {
@@ -55,7 +69,7 @@ static void fft_ref_init(int nbits, int inverse)
     double c1, s1, alpha;
 
     n = 1 << nbits;
-    exptab = av_malloc((n / 2) * sizeof(FFTComplex));
+    exptab = av_malloc((n / 2) * sizeof(*exptab));
 
     for (i = 0; i < (n/2); i++) {
         alpha = 2 * M_PI * (float)i / (float)n;
@@ -92,12 +106,12 @@ static void fft_ref(FFTComplex *tabr, FFTComplex *tab, int nbits)
             CMAC(tmp_re, tmp_im, c, s, q->re, q->im);
             q++;
         }
-        tabr[i].re = tmp_re;
-        tabr[i].im = tmp_im;
+        tabr[i].re = REF_SCALE(tmp_re, nbits);
+        tabr[i].im = REF_SCALE(tmp_im, nbits);
     }
 }
 
-static void imdct_ref(float *out, float *in, int nbits)
+static void imdct_ref(FFTSample *out, FFTSample *in, int nbits)
 {
     int n = 1<<nbits;
     int k, i, a;
@@ -110,12 +124,12 @@ static void imdct_ref(float *out, float *in, int nbits)
             f = cos(M_PI * a / (double)(2 * n));
             sum += f * in[k];
         }
-        out[i] = -sum;
+        out[i] = REF_SCALE(-sum, nbits - 2);
     }
 }
 
 /* NOTE: no normalisation by 1 / N is done */
-static void mdct_ref(float *output, float *input, int nbits)
+static void mdct_ref(FFTSample *output, FFTSample *input, int nbits)
 {
     int n = 1<<nbits;
     int k, i;
@@ -128,10 +142,11 @@ static void mdct_ref(float *output, float *input, int nbits)
             a = (2*M_PI*(2*i+1+n/2)*(2*k+1) / (4 * n));
             s += input[i] * cos(a);
         }
-        output[k] = s;
+        output[k] = REF_SCALE(s, nbits - 1);
     }
 }
 
+#if CONFIG_FFT_FLOAT
 static void idct_ref(float *output, float *input, int nbits)
 {
     int n = 1<<nbits;
@@ -164,11 +179,12 @@ static void dct_ref(float *output, float *input, int nbits)
         output[k] = s;
     }
 }
+#endif
 
 
-static float frandom(AVLFG *prng)
+static FFTSample frandom(AVLFG *prng)
 {
-    return (int16_t)av_lfg_get(prng) / 32768.0;
+    return (int16_t)av_lfg_get(prng) / 32768.0 * RANGE;
 }
 
 static int64_t gettime(void)
@@ -178,7 +194,7 @@ static int64_t gettime(void)
     return (int64_t)tv.tv_sec * 1000000 + tv.tv_usec;
 }
 
-static int check_diff(float *tab1, float *tab2, int n, double scale)
+static int check_diff(FFTSample *tab1, FFTSample *tab2, int n, double scale)
 {
     int i;
     double max= 0;
@@ -186,9 +202,9 @@ static int check_diff(float *tab1, float *tab2, int n, double scale)
     int err = 0;
 
     for (i = 0; i < n; i++) {
-        double e= fabsf(tab1[i] - (tab2[i] / scale));
+        double e = fabsf(tab1[i] - (tab2[i] / scale)) / RANGE;
         if (e >= 1e-3) {
-            av_log(NULL, AV_LOG_ERROR, "ERROR %5d: %10.6f %10.6f\n",
+            av_log(NULL, AV_LOG_ERROR, "ERROR %5d: "FMT" "FMT"\n",
                    i, tab1[i], tab2[i]);
             err = 1;
         }
@@ -233,8 +249,10 @@ int main(int argc, char **argv)
     int do_inverse = 0;
     FFTContext s1, *s = &s1;
     FFTContext m1, *m = &m1;
+#if CONFIG_FFT_FLOAT
     RDFTContext r1, *r = &r1;
     DCTContext d1, *d = &d1;
+#endif
     int fft_nbits, fft_size, fft_size_2;
     double scale = 1.0;
     AVLFG prng;
@@ -297,6 +315,7 @@ int main(int argc, char **argv)
         ff_fft_init(s, fft_nbits, do_inverse);
         fft_ref_init(fft_nbits, do_inverse);
         break;
+#if CONFIG_FFT_FLOAT
     case TRANSFORM_RDFT:
         if (do_inverse)
             av_log(NULL, AV_LOG_INFO,"IDFT_C2R");
@@ -312,6 +331,10 @@ int main(int argc, char **argv)
             av_log(NULL, AV_LOG_INFO,"DCT_II");
         ff_dct_init(d, fft_nbits, do_inverse ? DCT_III : DCT_II);
         break;
+#endif
+    default:
+        av_log(NULL, AV_LOG_ERROR, "Requested transform not supported\n");
+        return 1;
     }
     av_log(NULL, AV_LOG_INFO," %d test\n", fft_size);
 
@@ -328,15 +351,15 @@ int main(int argc, char **argv)
     switch (transform) {
     case TRANSFORM_MDCT:
         if (do_inverse) {
-            imdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
-            m->imdct_calc(m, tab2, (float *)tab1);
-            err = check_diff((float *)tab_ref, tab2, fft_size, scale);
+            imdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
+            m->imdct_calc(m, tab2, (FFTSample *)tab1);
+            err = check_diff((FFTSample *)tab_ref, tab2, fft_size, scale);
         } else {
-            mdct_ref((float *)tab_ref, (float *)tab1, fft_nbits);
+            mdct_ref((FFTSample *)tab_ref, (FFTSample *)tab1, fft_nbits);
 
-            m->mdct_calc(m, tab2, (float *)tab1);
+            m->mdct_calc(m, tab2, (FFTSample *)tab1);
 
-            err = check_diff((float *)tab_ref, tab2, fft_size / 2, scale);
+            err = check_diff((FFTSample *)tab_ref, tab2, fft_size / 2, scale);
         }
         break;
     case TRANSFORM_FFT:
@@ -345,8 +368,9 @@ int main(int argc, char **argv)
         s->fft_calc(s, tab);
 
         fft_ref(tab_ref, tab1, fft_nbits);
-        err = check_diff((float *)tab_ref, (float *)tab, fft_size * 2, 1.0);
+        err = check_diff((FFTSample *)tab_ref, (FFTSample *)tab, fft_size * 2, 1.0);
         break;
+#if CONFIG_FFT_FLOAT
     case TRANSFORM_RDFT:
         if (do_inverse) {
             tab1[         0].im = 0;
@@ -387,6 +411,7 @@ int main(int argc, char **argv)
         }
         err = check_diff((float *)tab_ref, (float *)tab, fft_size, 1.0);
         break;
+#endif
     }
 
     /* do a speed test */
@@ -404,15 +429,16 @@ int main(int argc, char **argv)
                 switch (transform) {
                 case TRANSFORM_MDCT:
                     if (do_inverse) {
-                        m->imdct_calc(m, (float *)tab, (float *)tab1);
+                        m->imdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
                     } else {
-                        m->mdct_calc(m, (float *)tab, (float *)tab1);
+                        m->mdct_calc(m, (FFTSample *)tab, (FFTSample *)tab1);
                     }
                     break;
                 case TRANSFORM_FFT:
                     memcpy(tab, tab1, fft_size * sizeof(FFTComplex));
                     s->fft_calc(s, tab);
                     break;
+#if CONFIG_FFT_FLOAT
                 case TRANSFORM_RDFT:
                     memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
                     r->rdft_calc(r, tab2);
@@ -421,6 +447,7 @@ int main(int argc, char **argv)
                     memcpy(tab2, tab1, fft_size * sizeof(FFTSample));
                     d->dct_calc(d, tab2);
                     break;
+#endif
                 }
             }
             duration = gettime() - time_start;
@@ -441,12 +468,14 @@ int main(int argc, char **argv)
     case TRANSFORM_FFT:
         ff_fft_end(s);
         break;
+#if CONFIG_FFT_FLOAT
     case TRANSFORM_RDFT:
         ff_rdft_end(r);
         break;
     case TRANSFORM_DCT:
         ff_dct_end(d);
         break;
+#endif
     }
 
     av_free(tab);
diff --git a/libavcodec/fft.c b/libavcodec/fft.c
index 6f08662ddc..d12d9f7f99 100644
--- a/libavcodec/fft.c
+++ b/libavcodec/fft.c
@@ -246,21 +246,16 @@ static void fft4(FFTComplex *z)
 
 static void fft8(FFTComplex *z)
 {
-    FFTDouble t1, t2, t3, t4, t5, t6, t7, t8;
+    FFTDouble t1, t2, t3, t4, t5, t6;
 
     fft4(z);
 
     BF(t1, z[5].re, z[4].re, -z[5].re);
     BF(t2, z[5].im, z[4].im, -z[5].im);
-    BF(t3, z[7].re, z[6].re, -z[7].re);
-    BF(t4, z[7].im, z[6].im, -z[7].im);
-    BF(t8, t1, t3, t1);
-    BF(t7, t2, t2, t4);
-    BF(z[4].re, z[0].re, z[0].re, t1);
-    BF(z[4].im, z[0].im, z[0].im, t2);
-    BF(z[6].re, z[2].re, z[2].re, t7);
-    BF(z[6].im, z[2].im, z[2].im, t8);
+    BF(t5, z[7].re, z[6].re, -z[7].re);
+    BF(t6, z[7].im, z[6].im, -z[7].im);
 
+    BUTTERFLIES(z[0],z[2],z[4],z[6]);
     TRANSFORM(z[1],z[3],z[5],z[7],sqrthalf,sqrthalf);
 }