virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2024-12-23 12:43:46 +02:00
Code Issues Releases Activity

Merge FFTContext and MDCTContext

Browse Source 
Originally committed as revision 19931 to svn://svn.ffmpeg.org/ffmpeg/trunk
This commit is contained in:
Måns Rullgård 2009-09-20 17:30:20 +00:00
parent ec129499b8
commit 01b2214758
22 changed files with 91 additions and 99 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
libavcodec/aac.h
View File

@ -257,8 +257,8 @@ typedef struct {
* @defgroup tables Computed / set up during initialization.
* @{
*/
MDCTContext mdct;
MDCTContext mdct_small;
FFTContext mdct;
FFTContext mdct_small;
DSPContext dsp;
int random_state;
/** @} */

4
libavcodec/aacenc.h
View File

@ -49,8 +49,8 @@ extern AACCoefficientsEncoder ff_aac_coders[];
*/
typedef struct AACEncContext {
PutBitContext pb;
MDCTContext mdct1024; ///< long (1024 samples) frame transform context
MDCTContext mdct128; ///< short (128 samples) frame transform context
FFTContext mdct1024; ///< long (1024 samples) frame transform context
FFTContext mdct128; ///< short (128 samples) frame transform context
DSPContext dsp;
DECLARE_ALIGNED_16(FFTSample, output[2048]); ///< temporary buffer for MDCT input coefficients
int16_t* samples; ///< saved preprocessed input

4
libavcodec/ac3dec.h
View File

@ -146,8 +146,8 @@ typedef struct {
///@defgroup imdct IMDCT
int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags (blksw)
MDCTContext imdct_512; ///< for 512 sample IMDCT
MDCTContext imdct_256; ///< for 256 sample IMDCT
FFTContext imdct_512; ///< for 512 sample IMDCT
FFTContext imdct_256; ///< for 256 sample IMDCT
///@}
///@defgroup opt optimization

6
libavcodec/arm/fft_init_arm.c
View File

@ -23,9 +23,9 @@
void ff_fft_permute_neon(FFTContext *s, FFTComplex *z);
void ff_fft_calc_neon(FFTContext *s, FFTComplex *z);
void ff_imdct_calc_neon(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_neon(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_mdct_calc_neon(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_mdct_calc_neon(FFTContext *s, FFTSample *output, const FFTSample *input);
av_cold void ff_fft_init_arm(FFTContext *s)
{

34
libavcodec/arm/mdct_neon.S
View File

@ -28,10 +28,10 @@ function ff_imdct_half_neon, export=1
push {r4-r8,lr}
mov r12, #1
ldr lr, [r0, #4] @ nbits
ldr r4, [r0, #8] @ tcos
ldr r5, [r0, #12] @ tsin
ldr r3, [r0, #24] @ revtab
ldr lr, [r0, #28] @ mdct_bits
ldr r4, [r0, #32] @ tcos
ldr r5, [r0, #36] @ tsin
ldr r3, [r0, #8] @ revtab
lsl r12, r12, lr @ n = 1 << nbits
lsr lr, r12, #2 @ n4 = n >> 2
add r7, r2, r12, lsl #1
@ -73,13 +73,12 @@ function ff_imdct_half_neon, export=1
mov r4, r0
mov r6, r1
add r0, r0, #16
bl ff_fft_calc_neon
mov r12, #1
ldr lr, [r4, #4] @ nbits
ldr r5, [r4, #12] @ tsin
ldr r4, [r4, #8] @ tcos
ldr lr, [r4, #28] @ mdct_bits
ldr r5, [r4, #36] @ tsin
ldr r4, [r4, #32] @ tcos
lsl r12, r12, lr @ n = 1 << nbits
lsr lr, r12, #3 @ n8 = n >> 3
@ -134,7 +133,7 @@ function ff_imdct_half_neon, export=1
function ff_imdct_calc_neon, export=1
push {r4-r6,lr}
ldr r3, [r0, #4]
ldr r3, [r0, #28]
mov r4, #1
mov r5, r1
lsl r4, r4, r3
@ -171,10 +170,10 @@ function ff_mdct_calc_neon, export=1
push {r4-r10,lr}
mov r12, #1
ldr lr, [r0, #4] @ nbits
ldr r4, [r0, #8] @ tcos
ldr r5, [r0, #12] @ tsin
ldr r3, [r0, #24] @ revtab
ldr lr, [r0, #28] @ mdct_bits
ldr r4, [r0, #32] @ tcos
ldr r5, [r0, #36] @ tsin
ldr r3, [r0, #8] @ revtab
lsl lr, r12, lr @ n = 1 << nbits
add r7, r2, lr @ in4u
sub r9, r7, #16 @ in4d
@ -224,7 +223,7 @@ function ff_mdct_calc_neon, export=1
vst2.32 {d6[1],d7[1]}, [r10,:64]
mov r12, #1
ldr lr, [r0, #4] @ nbits
ldr lr, [r0, #28] @ mdct_bits
lsl lr, r12, lr @ n = 1 << nbits
sub r8, r2, #16 @ in1d
add r2, r9, #16 @ in0u
@ -272,13 +271,12 @@ function ff_mdct_calc_neon, export=1
mov r4, r0
mov r6, r1
add r0, r0, #16
bl ff_fft_calc_neon
mov r12, #1
ldr lr, [r4, #4] @ nbits
ldr r5, [r4, #12] @ tsin
ldr r4, [r4, #8] @ tcos
ldr lr, [r4, #28] @ mdct_bits
ldr r5, [r4, #36] @ tsin
ldr r4, [r4, #32] @ tcos
lsl r12, r12, lr @ n = 1 << nbits
lsr lr, r12, #3 @ n8 = n >> 3

4
libavcodec/atrac1.c
View File

@ -79,7 +79,7 @@ typedef struct {
DECLARE_ALIGNED_16(float, high[512]);
float* bands[3];
DECLARE_ALIGNED_16(float, out_samples[AT1_MAX_CHANNELS][AT1_SU_SAMPLES]);
MDCTContext mdct_ctx[3];
FFTContext mdct_ctx[3];
int channels;
DSPContext dsp;
} AT1Ctx;
@ -94,7 +94,7 @@ static const uint8_t mdct_long_nbits[3] = {7, 7, 8};
static void at1_imdct(AT1Ctx *q, float *spec, float *out, int nbits,
int rev_spec)
{
MDCTContext* mdct_context;
FFTContext* mdct_context;
int transf_size = 1 << nbits;
mdct_context = &q->mdct_ctx[nbits - 5 - (nbits > 6)];

2
libavcodec/atrac3.c
View File

@ -123,7 +123,7 @@ static DECLARE_ALIGNED_16(float,mdct_window[512]);
static VLC spectral_coeff_tab[7];
static float gain_tab1[16];
static float gain_tab2[31];
static MDCTContext mdct_ctx;
static FFTContext mdct_ctx;
static DSPContext dsp;

2
libavcodec/cook.c
View File

@ -136,7 +136,7 @@ typedef struct cook {
AVLFG random_state;
/* transform data */
MDCTContext mdct_ctx;
FFTContext mdct_ctx;
float* mlt_window;
/* VLC data */

2
libavcodec/dca.c
View File

@ -250,7 +250,7 @@ typedef struct {
int debug_flag; ///< used for suppressing repeated error messages output
DSPContext dsp;
MDCTContext imdct;
FFTContext imdct;
} DCAContext;
static const uint16_t dca_vlc_offs[] = {

44
libavcodec/dsputil.h
View File

@ -665,8 +665,6 @@ void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
FFTSample type */
typedef float FFTSample;
struct MDCTContext;
typedef struct FFTComplex {
FFTSample re, im;
} FFTComplex;
@ -678,11 +676,16 @@ typedef struct FFTContext {
FFTComplex *exptab;
FFTComplex *exptab1; /* only used by SSE code */
FFTComplex *tmp_buf;
int mdct_size; /* size of MDCT (i.e. number of input data * 2) */
int mdct_bits; /* n = 2^nbits */
/* pre/post rotation tables */
FFTSample *tcos;
FFTSample *tsin;
void (*fft_permute)(struct FFTContext *s, FFTComplex *z);
void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
void (*mdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input);
void (*imdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
void (*imdct_half)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
void (*mdct_calc)(struct FFTContext *s, FFTSample *output, const FFTSample *input);
int split_radix;
} FFTContext;
@ -720,28 +723,19 @@ void ff_fft_end(FFTContext *s);
/* MDCT computation */
typedef struct MDCTContext {
int n; /* size of MDCT (i.e. number of input data * 2) */
int nbits; /* n = 2^nbits */
/* pre/post rotation tables */
FFTSample *tcos;
FFTSample *tsin;
FFTContext fft;
} MDCTContext;
static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input)
static inline void ff_imdct_calc(FFTContext *s, FFTSample *output, const FFTSample *input)
{
s->fft.imdct_calc(s, output, input);
s->imdct_calc(s, output, input);
}
static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input)
static inline void ff_imdct_half(FFTContext *s, FFTSample *output, const FFTSample *input)
{
s->fft.imdct_half(s, output, input);
s->imdct_half(s, output, input);
}
static inline void ff_mdct_calc(MDCTContext *s, FFTSample *output,
static inline void ff_mdct_calc(FFTContext *s, FFTSample *output,
const FFTSample *input)
{
s->fft.mdct_calc(s, output, input);
s->mdct_calc(s, output, input);
}
/**
@ -768,11 +762,11 @@ extern float ff_sine_2048[2048];
extern float ff_sine_4096[4096];
extern float * const ff_sine_windows[13];
int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale);
void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_mdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_mdct_end(MDCTContext *s);
int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale);
void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_mdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_mdct_end(FFTContext *s);
/* Real Discrete Fourier Transform */

2
libavcodec/fft-test.c
View File

@ -184,7 +184,7 @@ int main(int argc, char **argv)
int do_mdct = 0;
int do_inverse = 0;
FFTContext s1, *s = &s1;
MDCTContext m1, *m = &m1;
FFTContext m1, *m = &m1;
int fft_nbits, fft_size;
double scale = 1.0;
AVLFG prng;

32
libavcodec/mdct.c
View File

@ -72,15 +72,15 @@ av_cold void ff_sine_window_init(float *window, int n) {
/**
* init MDCT or IMDCT computation.
*/
av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale)
av_cold int ff_mdct_init(FFTContext *s, int nbits, int inverse, double scale)
{
int n, n4, i;
double alpha, theta;
memset(s, 0, sizeof(*s));
n = 1 << nbits;
s->nbits = nbits;
s->n = n;
s->mdct_bits = nbits;
s->mdct_size = n;
n4 = n >> 2;
s->tcos = av_malloc(n4 * sizeof(FFTSample));
if (!s->tcos)
@ -96,7 +96,7 @@ av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale)
s->tcos[i] = -cos(alpha) * scale;
s->tsin[i] = -sin(alpha) * scale;
}
if (ff_fft_init(&s->fft, s->nbits - 2, inverse) < 0)
if (ff_fft_init(s, s->mdct_bits - 2, inverse) < 0)
goto fail;
return 0;
fail:
@ -122,16 +122,16 @@ av_cold int ff_mdct_init(MDCTContext *s, int nbits, int inverse, double scale)
* @param output N/2 samples
* @param input N/2 samples
*/
void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
void ff_imdct_half_c(FFTContext *s, FFTSample *output, const FFTSample *input)
{
int k, n8, n4, n2, n, j;
const uint16_t *revtab = s->fft.revtab;
const uint16_t *revtab = s->revtab;
const FFTSample *tcos = s->tcos;
const FFTSample *tsin = s->tsin;
const FFTSample *in1, *in2;
FFTComplex *z = (FFTComplex *)output;
n = 1 << s->nbits;
n = 1 << s->mdct_bits;
n2 = n >> 1;
n4 = n >> 2;
n8 = n >> 3;
@ -145,7 +145,7 @@ void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
in1 += 2;
in2 -= 2;
}
ff_fft_calc(&s->fft, z);
ff_fft_calc(s, z);
/* post rotation + reordering */
for(k = 0; k < n8; k++) {
@ -164,10 +164,10 @@ void ff_imdct_half_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
* @param output N samples
* @param input N/2 samples
*/
void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
void ff_imdct_calc_c(FFTContext *s, FFTSample *output, const FFTSample *input)
{
int k;
int n = 1 << s->nbits;
int n = 1 << s->mdct_bits;
int n2 = n >> 1;
int n4 = n >> 2;
@ -184,16 +184,16 @@ void ff_imdct_calc_c(MDCTContext *s, FFTSample *output, const FFTSample *input)
* @param input N samples
* @param out N/2 samples
*/
void ff_mdct_calc_c(MDCTContext *s, FFTSample *out, const FFTSample *input)
void ff_mdct_calc_c(FFTContext *s, FFTSample *out, const FFTSample *input)
{
int i, j, n, n8, n4, n2, n3;
FFTSample re, im;
const uint16_t *revtab = s->fft.revtab;
const uint16_t *revtab = s->revtab;
const FFTSample *tcos = s->tcos;
const FFTSample *tsin = s->tsin;
FFTComplex *x = (FFTComplex *)out;
n = 1 << s->nbits;
n = 1 << s->mdct_bits;
n2 = n >> 1;
n4 = n >> 2;
n8 = n >> 3;
@ -212,7 +212,7 @@ void ff_mdct_calc_c(MDCTContext *s, FFTSample *out, const FFTSample *input)
CMUL(x[j].re, x[j].im, re, im, -tcos[n8 + i], tsin[n8 + i]);
}
ff_fft_calc(&s->fft, x);
ff_fft_calc(s, x);
/* post rotation */
for(i=0;i<n8;i++) {
@ -226,9 +226,9 @@ void ff_mdct_calc_c(MDCTContext *s, FFTSample *out, const FFTSample *input)
}
}
av_cold void ff_mdct_end(MDCTContext *s)
av_cold void ff_mdct_end(FFTContext *s)
{
av_freep(&s->tcos);
av_freep(&s->tsin);
ff_fft_end(&s->fft);
ff_fft_end(s);
}

2
libavcodec/nellymoserdec.c
View File

@ -50,7 +50,7 @@ typedef struct NellyMoserDecodeContext {
int add_bias;
float scale_bias;
DSPContext dsp;
MDCTContext imdct_ctx;
FFTContext imdct_ctx;
DECLARE_ALIGNED_16(float,imdct_out[NELLY_BUF_LEN * 2]);
} NellyMoserDecodeContext;

2
libavcodec/nellymoserenc.c
View File

@ -52,7 +52,7 @@ typedef struct NellyMoserEncodeContext {
int bufsel;
int have_saved;
DSPContext dsp;
MDCTContext mdct_ctx;
FFTContext mdct_ctx;
DECLARE_ALIGNED_16(float, mdct_out[NELLY_SAMPLES]);
DECLARE_ALIGNED_16(float, in_buff[NELLY_SAMPLES]);
DECLARE_ALIGNED_16(float, buf[2][3 * NELLY_BUF_LEN]); ///< sample buffer

2
libavcodec/twinvq.c
View File

@ -172,7 +172,7 @@ static const ModeTab mode_44_48 = {
typedef struct TwinContext {
AVCodecContext *avctx;
DSPContext dsp;
MDCTContext mdct_ctx[3];
FFTContext mdct_ctx[3];
const ModeTab *mtab;

2
libavcodec/vorbis_dec.c
View File

@ -128,7 +128,7 @@ typedef struct vorbis_context_s {
GetBitContext gb;
DSPContext dsp;
MDCTContext mdct[2];
FFTContext mdct[2];
uint_fast8_t first_frame;
uint_fast32_t version;
uint_fast8_t audio_channels;

2
libavcodec/vorbis_enc.c
View File

@ -98,7 +98,7 @@ typedef struct {
int channels;
int sample_rate;
int log2_blocksize[2];
MDCTContext mdct[2];
FFTContext mdct[2];
const float * win[2];
int have_saved;
float * saved;

2
libavcodec/wma.h
View File

@ -116,7 +116,7 @@ typedef struct WMACodecContext {
WMACoef coefs1[MAX_CHANNELS][BLOCK_MAX_SIZE];
DECLARE_ALIGNED_16(float, coefs[MAX_CHANNELS][BLOCK_MAX_SIZE]);
DECLARE_ALIGNED_16(FFTSample, output[BLOCK_MAX_SIZE * 2]);
MDCTContext mdct_ctx[BLOCK_NB_SIZES];
FFTContext mdct_ctx[BLOCK_NB_SIZES];
float *windows[BLOCK_NB_SIZES];
/* output buffer for one frame and the last for IMDCT windowing */
DECLARE_ALIGNED_16(float, frame_out[MAX_CHANNELS][BLOCK_MAX_SIZE * 2]);

2
libavcodec/wmaprodec.c
View File

@ -166,7 +166,7 @@ typedef struct WMAProDecodeCtx {
uint8_t frame_data[MAX_FRAMESIZE +
FF_INPUT_BUFFER_PADDING_SIZE];///< compressed frame data
PutBitContext pb; ///< context for filling the frame_data buffer
MDCTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
FFTContext mdct_ctx[WMAPRO_BLOCK_SIZES]; ///< MDCT context per block size
DECLARE_ALIGNED_16(float, tmp[WMAPRO_BLOCK_MAX_SIZE]); ///< IMDCT output buffer
float* windows[WMAPRO_BLOCK_SIZES]; ///< windows for the different block sizes

12
libavcodec/x86/fft.h
View File

@ -26,11 +26,11 @@ void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
void ff_imdct_calc_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_3dn(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_calc_3dn(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_3dn(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_calc_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample *input);
#endif

12
libavcodec/x86/fft_3dn2.c
View File

@ -53,14 +53,14 @@ void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z)
FFSWAP(FFTSample, z[i].im, z[i+1].re);
}
void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input)
void ff_imdct_half_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input)
{
x86_reg j, k;
long n = 1 << s->nbits;
long n = 1 << s->mdct_bits;
long n2 = n >> 1;
long n4 = n >> 2;
long n8 = n >> 3;
const uint16_t *revtab = s->fft.revtab;
const uint16_t *revtab = s->revtab;
const FFTSample *tcos = s->tcos;
const FFTSample *tsin = s->tsin;
const FFTSample *in1, *in2;
@ -101,7 +101,7 @@ void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *inpu
);
}
ff_fft_dispatch_3dn2(z, s->fft.nbits);
ff_fft_dispatch_3dn2(z, s->nbits);
#define CMUL(j,mm0,mm1)\
"movq (%2,"#j",2), %%mm6 \n"\
@ -144,10 +144,10 @@ void ff_imdct_half_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *inpu
__asm__ volatile("femms");
}
void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, const FFTSample *input)
void ff_imdct_calc_3dn2(FFTContext *s, FFTSample *output, const FFTSample *input)
{
x86_reg j, k;
long n = 1 << s->nbits;
long n = 1 << s->mdct_bits;
long n4 = n >> 2;
ff_imdct_half_3dn2(s, output+n4, input);

12
libavcodec/x86/fft_sse.c
View File

@ -71,14 +71,14 @@ void ff_fft_permute_sse(FFTContext *s, FFTComplex *z)
memcpy(z, s->tmp_buf, n*sizeof(FFTComplex));
}
void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input)
void ff_imdct_half_sse(FFTContext *s, FFTSample *output, const FFTSample *input)
{
av_unused x86_reg i, j, k, l;
long n = 1 << s->nbits;
long n = 1 << s->mdct_bits;
long n2 = n >> 1;
long n4 = n >> 2;
long n8 = n >> 3;
const uint16_t *revtab = s->fft.revtab + n8;
const uint16_t *revtab = s->revtab + n8;
const FFTSample *tcos = s->tcos;
const FFTSample *tsin = s->tsin;
FFTComplex *z = (FFTComplex *)output;
@ -129,7 +129,7 @@ void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input
#endif
}
ff_fft_dispatch_sse(z, s->fft.nbits);
ff_fft_dispatch_sse(z, s->nbits);
/* post rotation + reinterleave + reorder */
@ -172,10 +172,10 @@ void ff_imdct_half_sse(MDCTContext *s, FFTSample *output, const FFTSample *input
);
}
void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, const FFTSample *input)
void ff_imdct_calc_sse(FFTContext *s, FFTSample *output, const FFTSample *input)
{
x86_reg j, k;
long n = 1 << s->nbits;
long n = 1 << s->mdct_bits;
long n4 = n >> 2;
ff_imdct_half_sse(s, output+n4, input);