---
libavcodec/synth_filter.c | 116 +++++++++++++++++++++++++++++++++++++++++++++-
libavcodec/synth_filter.h | 13 ++++++
2 files changed, 128 insertions(+), 1 deletion(-)
diff --git a/libavcodec/synth_filter.c b/libavcodec/synth_filter.c
index 9eab9c5..1c5dab5 100644
--- a/libavcodec/synth_filter.c
+++ b/libavcodec/synth_filter.c
@@ -1,5 +1,6 @@
/*
* copyright (c) 2008 Michael Niedermayer <michae...@gmx.at>
+ * Copyright (C) 2016 foo86
*
* This file is part of FFmpeg.
*
@@ -19,6 +20,8 @@
*/
#include "fft.h"
+#include "dcadct.h"
+#include "dcamath.h"
#include "synth_filter.h"
static void synth_filter_float(FFTContext *imdct,
@@ -57,9 +60,120 @@ static void synth_filter_float(FFTContext *imdct,
*synth_buf_offset = (*synth_buf_offset - 32) & 511;
}
+static void synth_filter_float_64(FFTContext *imdct,
+ float *synth_buf_ptr, int *synth_buf_offset,
+ float synth_buf2[64], const float
window[1024],
+ float out[64], const float in[64], float
scale)
+{
+ float *synth_buf = synth_buf_ptr + *synth_buf_offset;
+ int i, j;
+
+ imdct->imdct_half(imdct, synth_buf, in);
+
+ for (i = 0; i < 32; i++) {
+ float a = synth_buf2[i ];
+ float b = synth_buf2[i + 32];
+ float c = 0;
+ float d = 0;
+ for (j = 0; j < 1024 - *synth_buf_offset; j += 128) {
+ a += window[i + j ] * (-synth_buf[31 - i + j ]);
+ b += window[i + j + 32] * ( synth_buf[ i + j ]);
+ c += window[i + j + 64] * ( synth_buf[32 + i + j ]);
+ d += window[i + j + 96] * ( synth_buf[63 - i + j ]);
+ }
+ for ( ; j < 1024; j += 128) {
+ a += window[i + j ] * (-synth_buf[31 - i + j - 1024]);
+ b += window[i + j + 32] * ( synth_buf[ i + j - 1024]);
+ c += window[i + j + 64] * ( synth_buf[32 + i + j - 1024]);
+ d += window[i + j + 96] * ( synth_buf[63 - i + j - 1024]);
+ }
+ out[i ] = a * scale;
+ out[i + 32] = b * scale;
+ synth_buf2[i ] = c;
+ synth_buf2[i + 32] = d;
+ }
+
+ *synth_buf_offset = (*synth_buf_offset - 64) & 1023;
+}
+
+static void synth_filter_fixed(DCADCTContext *imdct,
+ int32_t *synth_buf_ptr, int *synth_buf_offset,
+ int32_t synth_buf2[32], const int32_t
window[512],
+ int32_t out[32], const int32_t in[32])
+{
+ int32_t *synth_buf = synth_buf_ptr + *synth_buf_offset;
+ int i, j;
+
+ imdct->imdct_half[0](synth_buf, in);
+
+ for (i = 0; i < 16; i++) {
+ int64_t a = synth_buf2[i ] * (INT64_C(1) << 21);
+ int64_t b = synth_buf2[i + 16] * (INT64_C(1) << 21);
+ int64_t c = 0;
+ int64_t d = 0;
+ for (j = 0; j < 512 - *synth_buf_offset; j += 64) {
+ a += (int64_t)window[i + j ] * synth_buf[ i + j ];
+ b += (int64_t)window[i + j + 16] * synth_buf[15 - i + j ];
+ c += (int64_t)window[i + j + 32] * synth_buf[16 + i + j ];
+ d += (int64_t)window[i + j + 48] * synth_buf[31 - i + j ];
+ }
+ for ( ; j < 512; j += 64) {
+ a += (int64_t)window[i + j ] * synth_buf[ i + j - 512];
+ b += (int64_t)window[i + j + 16] * synth_buf[15 - i + j - 512];
+ c += (int64_t)window[i + j + 32] * synth_buf[16 + i + j - 512];
+ d += (int64_t)window[i + j + 48] * synth_buf[31 - i + j - 512];
+ }
+ out[i ] = clip23(norm21(a));
+ out[i + 16] = clip23(norm21(b));
+ synth_buf2[i ] = norm21(c);
+ synth_buf2[i + 16] = norm21(d);
+ }
+
+ *synth_buf_offset = (*synth_buf_offset - 32) & 511;
+}
+
+static void synth_filter_fixed_64(DCADCTContext *imdct,
+ int32_t *synth_buf_ptr, int
*synth_buf_offset,
+ int32_t synth_buf2[64], const int32_t
window[1024],
+ int32_t out[64], const int32_t in[64])
+{
+ int32_t *synth_buf = synth_buf_ptr + *synth_buf_offset;
+ int i, j;
+
+ imdct->imdct_half[1](synth_buf, in);
+
+ for (i = 0; i < 32; i++) {
+ int64_t a = synth_buf2[i ] * (INT64_C(1) << 20);
+ int64_t b = synth_buf2[i + 32] * (INT64_C(1) << 20);
+ int64_t c = 0;
+ int64_t d = 0;
+ for (j = 0; j < 1024 - *synth_buf_offset; j += 128) {
+ a += (int64_t)window[i + j ] * synth_buf[ i + j ];
+ b += (int64_t)window[i + j + 32] * synth_buf[31 - i + j ];
+ c += (int64_t)window[i + j + 64] * synth_buf[32 + i + j ];
+ d += (int64_t)window[i + j + 96] * synth_buf[63 - i + j ];
+ }
+ for ( ; j < 1024; j += 128) {
+ a += (int64_t)window[i + j ] * synth_buf[ i + j - 1024];
+ b += (int64_t)window[i + j + 32] * synth_buf[31 - i + j - 1024];
+ c += (int64_t)window[i + j + 64] * synth_buf[32 + i + j - 1024];
+ d += (int64_t)window[i + j + 96] * synth_buf[63 - i + j - 1024];
+ }
+ out[i ] = clip23(norm20(a));
+ out[i + 32] = clip23(norm20(b));
+ synth_buf2[i ] = norm20(c);
+ synth_buf2[i + 32] = norm20(d);
+ }
+
+ *synth_buf_offset = (*synth_buf_offset - 64) & 1023;
+}
+
av_cold void ff_synth_filter_init(SynthFilterContext *c)
{
- c->synth_filter_float = synth_filter_float;
+ c->synth_filter_float = synth_filter_float;
+ c->synth_filter_float_64 = synth_filter_float_64;
+ c->synth_filter_fixed = synth_filter_fixed;
+ c->synth_filter_fixed_64 = synth_filter_fixed_64;
if (ARCH_AARCH64)
ff_synth_filter_init_aarch64(c);
diff --git a/libavcodec/synth_filter.h b/libavcodec/synth_filter.h
index e073f42..df3589a 100644
--- a/libavcodec/synth_filter.h
+++ b/libavcodec/synth_filter.h
@@ -22,6 +22,7 @@
#define AVCODEC_SYNTH_FILTER_H
#include "fft.h"
+#include "dcadct.h"
typedef struct SynthFilterContext {
void (*synth_filter_float)(FFTContext *imdct,
@@ -29,6 +30,18 @@ typedef struct SynthFilterContext {
float synth_buf2[32], const float window[512],
float out[32], const float in[32],
float scale);
+ void (*synth_filter_float_64)(FFTContext *imdct,
+ float *synth_buf_ptr, int *synth_buf_offset,
+ float synth_buf2[64], const float
window[1024],
+ float out[64], const float in[64], float
scale);
+ void (*synth_filter_fixed)(DCADCTContext *imdct,
+ int32_t *synth_buf_ptr, int *synth_buf_offset,
+ int32_t synth_buf2[32], const int32_t
window[512],
+ int32_t out[32], const int32_t in[32]);
+ void (*synth_filter_fixed_64)(DCADCTContext *imdct,
+ int32_t *synth_buf_ptr, int
*synth_buf_offset,
+ int32_t synth_buf2[64], const int32_t
window[1024],
+ int32_t out[64], const int32_t in[64]);
} SynthFilterContext;
void ff_synth_filter_init(SynthFilterContext *c);
--
2.1.4
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
