dsputil.h

wince 平台下的h264 压缩代码程序

    void (*vp3_dsp_init)(void);

    /** 
     * This function is responsible for taking a block of zigzag'd,
     * quantized DCT coefficients, reconstructing the original block of
     * samples, and placing it into the output.
     * @param input_data 64 zigzag'd, quantized DCT coefficients
     * @param dequant_matrix 64 zigzag'd quantizer coefficients
     * @param coeff_count index of the last coefficient
     * @param dest the final output location where the transformed samples
     * are to be placed
     * @param stride the width in 8-bit samples of a line on this plane
     */
    void (*vp3_idct_put)(int16_t *input_data, int16_t *dequant_matrix,
        int coeff_count, uint8_t *dest, int stride);

    /** 
     * This function is responsible for taking a block of zigzag'd,
     * quantized DCT coefficients, reconstructing the original block of
     * samples, and adding the transformed samples to an existing block of
     * samples in the output.
     * @param input_data 64 zigzag'd, quantized DCT coefficients
     * @param dequant_matrix 64 zigzag'd quantizer coefficients
     * @param coeff_count index of the last coefficient
     * @param dest the final output location where the transformed samples
     * are to be placed
     * @param stride the width in 8-bit samples of a line on this plane
     */
    void (*vp3_idct_add)(int16_t *input_data, int16_t *dequant_matrix,
        int coeff_count, uint8_t *dest, int stride);

} DSPContext;

void dsputil_static_init(void);
void dsputil_init(DSPContext* p, AVCodecContext *avctx);

/**
 * permute block according to permuatation.
 * @param last last non zero element in scantable order
 */
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);

void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);

#define	BYTE_VEC32(c)	((c)*0x01010101UL)

static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
{
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
}

static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
{
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
}

/**
 * Empty mmx state.
 * this must be called between any dsp function and float/double code.
 * for example sin(); dsp->idct_put(); emms_c(); cos()
 */
#define emms_c()

/* should be defined by architectures supporting
   one or more MultiMedia extension */
int mm_support(void);

#if defined(HAVE_MMX)

#undef emms_c

#define MM_MMX    0x0001 /* standard MMX */
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
#define MM_SSE    0x0008 /* SSE functions */
#define MM_SSE2   0x0010 /* PIV SSE2 functions */

extern int mm_flags;

void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);

static inline void emms(void)
{
    __asm __volatile ("emms;":::"memory");
}


#define emms_c() \
{\
    if (mm_flags & MM_MMX)\
        emms();\
}

#define __align8 __attribute__ ((aligned (8)))

void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);

#elif defined(ARCH_ARMV4L)

/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
   line ptimizations */
#ifdef WINCE
#define __align8
#else
#define __align8 __attribute__ ((aligned (4)))
#endif

void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);

#elif defined(HAVE_MLIB)

/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
#define __align8 __attribute__ ((aligned (8)))

void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);

#elif defined(ARCH_ALPHA)

#define __align8 __attribute__ ((aligned (8)))

void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);

#elif defined(ARCH_POWERPC)

#define MM_ALTIVEC    0x0001 /* standard AltiVec */

extern int mm_flags;

#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
#define pixel altivec_pixel
#include <altivec.h>
#undef pixel
#endif

#define __align8 __attribute__ ((aligned (16)))

void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);

#elif defined(HAVE_MMI)

#define __align8 __attribute__ ((aligned (16)))

void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);

#elif defined(ARCH_SH4)

#define __align8 __attribute__ ((aligned (8)))

void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);

#else

#ifdef WINCE
#define __align8
#else
#define __align8
#endif

#endif

#ifdef __GNUC__

struct unaligned_64 { uint64_t l; } __attribute__((packed));
struct unaligned_32 { uint32_t l; } __attribute__((packed));
struct unaligned_16 { uint16_t l; } __attribute__((packed));

#define LD16(a) (((const struct unaligned_16 *) (a))->l)
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
#define LD64(a) (((const struct unaligned_64 *) (a))->l)

#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)

#else /* __GNUC__ */

#ifdef WINCE
/*#define LD16(a) ( *(a) * 0x0100 + *(a+1) )
#define LD32(a) ( *(a) * 0x01000000 + *(a+1) * 0x00010000 + *(a+2) * 0x00000100 + *(a+3) )
#define LD64(a) ( *(a) * 0x0100000000000000 + *(a+1) * 0x0001000000000000 + *(a+2) * 0x0000010000000000 + *(a+3) * 0x0000000100000000 + *(a+4) * 0x0000000001000000 + *(a+5) * 0x0000000000010000 + *(a+6) * 0x0000000000000100 + *(a+7) )

#define ST32(a, b) *(a) = (uint8_t)b & 0xFF000000;\
				   *(a+1) = (uint8_t)b & 0x00FF0000;\
				   *(a+2) = (uint8_t)b & 0x0000FF00;\
				   *(a+3) = (uint8_t)b & 0x000000FF;*/
// SMALL ENDIAN
#define LD16(a) ( *(a) | ((uint16_t)*(a+1) << 8) )
#define LD32(a) ( ((*(a))) | ((*(a+1))<< 8)\
                | ((*(a+2))<< 16) | (*(a+3) <<24) )
#define LD64(a) ( *(a) | ((uint64_t)*(a+1) << 8)\
                | ((uint64_t)*(a+2) << 16) | ((uint64_t)*(a+3) << 24)\
                | ((uint64_t)*(a+4) << 32) | ((uint64_t)*(a+5) << 40)\
                | ((uint64_t)*(a+6) << 48) | ((uint64_t)*(a+7) << 56) )
#define ST32(a, b) *(a) = (uint8_t)b & 0xFF000000;\
				   *(a+1) = (uint8_t)b & 0x00FF0000;\
				   *(a+2) = (uint8_t)b & 0x0000FF00;\
				   *(a+3) = (uint8_t)b & 0x000000FF;
/* MAUVAISE SOLUTION
#define LD16(a) ( ((uint16_t)*(a) << 8) | *(a+1) )
#define LD32(a) ( ((uint32_t)*(a) << 24) | ((uint32_t)*(a+1) << 16)\
                | ((uint32_t)*(a+2) << 8) | *(a+3) )
#define LD64(a) ( ((uint64_t)*(a) << 56) | ((uint64_t)*(a+1) << 48)\
                | ((uint64_t)*(a+2) << 40) | ((uint64_t)*(a+3) << 32)\
                | ((uint64_t)*(a+4) << 24) | ((uint64_t)*(a+5) << 16)\
                | ((uint64_t)*(a+6) << 8) | *(a+7) )
#define ST32(a, b) *(a) = (uint8_t)b & 0xFF000000;\
				   *(a+1) = (uint8_t)b & 0x00FF0000;\
				   *(a+2) = (uint8_t)b & 0x0000FF00;\
				   *(a+3) = (uint8_t)b & 0x000000FF;*/
#else
#define LD16(a) (*((uint16_t*)(a)))
#define LD32(a) (*((uint32_t*)(a)))
#define LD64(a) (*((uint64_t*)(a)))

#define ST32(a, b) *((uint32_t*)(a)) = (b)
#endif

#endif /* !__GNUC__ */

/* PSNR */
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
              int orig_linesize[3], int coded_linesize,
              AVCodecContext *avctx);

/* FFT computation */

/* NOTE: soon integer code will be added, so you must use the
   FFTSample type */
typedef float FFTSample;

typedef struct FFTComplex {
    FFTSample re, im;
} FFTComplex;

typedef struct FFTContext {
    int nbits;
    int inverse;
    uint16_t *revtab;
    FFTComplex *exptab;
    FFTComplex *exptab1; /* only used by SSE code */
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
} FFTContext;

int ff_fft_init(FFTContext *s, int nbits, int inverse);
void ff_fft_permute(FFTContext *s, FFTComplex *z);
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);

static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
{
    s->fft_calc(s, z);
}
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;

int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
                const FFTSample *input, FFTSample *tmp);
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
               const FFTSample *input, FFTSample *tmp);
void ff_mdct_end(MDCTContext *s);

#define WARPER8_16(name8, name16)\
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
    return name8(s, dst           , src           , stride, h)\
          +name8(s, dst+8         , src+8         , stride, h);\
}

#define WARPER8_16_SQ(name8, name16)\
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
    int score=0;\
    score +=name8(s, dst           , src           , stride, 8);\
    score +=name8(s, dst+8         , src+8         , stride, 8);\
    if(h==16){\
        dst += 8*stride;\
        src += 8*stride;\
        score +=name8(s, dst           , src           , stride, 8);\
        score +=name8(s, dst+8         , src+8         , stride, 8);\
    }\
    return score;\
}

#ifndef HAVE_LRINTF
/* XXX: add ISOC specific test to avoid specific BSD testing. */
/* better than nothing implementation. */
/* btw, rintf() is existing on fbsd too -- alex */
static inline long int lrintf(float x)
{
#ifdef CONFIG_WIN32
    /* XXX: incorrect, but make it compile */
    return (int)(x);
#else
    return (int)(rint(x));
#endif
}
#endif

#endif