dsputil.h

手机端的H264源码

     * block -> idct -> clip to unsigned 8 bit -> dest.
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
     * @param line_size size in bytes of a horizotal line of dest
     */
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
    
    /**
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
     * @param line_size size in bytes of a horizotal line of dest
     */
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
    
    /**
     * idct input permutation.
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
     * IDCT)
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
     * with the zigzag/alternate scan<br>
     * an example to avoid confusion:
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
     * - (x -> referece dct -> reference idct -> x)
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
     */
    uint8_t idct_permutation[64];
    int idct_permutation_type;
#define FF_NO_IDCT_PERM 1
#define FF_LIBMPEG2_IDCT_PERM 2
#define FF_SIMPLE_IDCT_PERM 3
#define FF_TRANSPOSE_IDCT_PERM 4    

} 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__ */

#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;\
}


#endif