diff --git a/libavcodec/dsputil.c b/libavcodec/dsputil.c index 44b48996ed..e960696e1a 100644 --- a/libavcodec/dsputil.c +++ b/libavcodec/dsputil.c @@ -1822,7 +1822,7 @@ static int dct_sad8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2 return sum; } -void simple_idct(INT16 *block); //FIXME +void simple_idct(DCTELEM *block); //FIXME static int quant_psnr8x8_c(/*MpegEncContext*/ void *c, uint8_t *src1, uint8_t *src2, int stride){ MpegEncContext * const s= (MpegEncContext *)c; diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index 23c4228258..d885c4aca1 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -25,6 +25,7 @@ //#define DEBUG /* dct code */ typedef short DCTELEM; +//typedef int DCTELEM; void fdct_ifast (DCTELEM *data); void ff_jpeg_fdct_islow (DCTELEM *data); @@ -156,7 +157,7 @@ void dsputil_init(DSPContext* p, unsigned mask); * permute block according to permuatation. * @param last last non zero element in scantable order */ -void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last); +void ff_block_permute(DCTELEM *block, UINT8 *permutation, const UINT8 *scantable, int last); #define emms_c() diff --git a/libavcodec/dv.c b/libavcodec/dv.c index f436caf128..404f018f14 100644 --- a/libavcodec/dv.c +++ b/libavcodec/dv.c @@ -157,7 +157,7 @@ static const UINT16 block_sizes[6] = { /* decode ac coefs */ static void dv_decode_ac(DVVideoDecodeContext *s, - BlockInfo *mb, INT16 *block, int last_index) + BlockInfo *mb, DCTELEM *block, int last_index) { int last_re_index; int shift_offset = mb->shift_offset; diff --git a/libavcodec/h263.c b/libavcodec/h263.c index 64b2e18bda..025288b03f 100644 --- a/libavcodec/h263.c +++ b/libavcodec/h263.c @@ -67,7 +67,7 @@ static inline int mpeg4_decode_dc(MpegEncContext * s, int n, int *dir_ptr); static inline int mpeg4_decode_block(MpegEncContext * s, DCTELEM * block, int n, int coded, int intra); static int h263_pred_dc(MpegEncContext * s, int n, UINT16 **dc_val_ptr); -static void mpeg4_inv_pred_ac(MpegEncContext * s, INT16 *block, int n, +static void mpeg4_inv_pred_ac(MpegEncContext * s, DCTELEM *block, int n, int dir); static void mpeg4_decode_sprite_trajectory(MpegEncContext * s); static inline int ff_mpeg4_pred_dc(MpegEncContext * s, int n, UINT16 **dc_val_ptr, int *dir_ptr); @@ -999,7 +999,7 @@ static int h263_pred_dc(MpegEncContext * s, int n, UINT16 **dc_val_ptr) } -void h263_pred_acdc(MpegEncContext * s, INT16 *block, int n) +void h263_pred_acdc(MpegEncContext * s, DCTELEM *block, int n) { int x, y, wrap, a, c, pred_dc, scale, i; INT16 *dc_val, *ac_val, *ac_val1; @@ -1852,7 +1852,7 @@ static inline int ff_mpeg4_pred_dc(MpegEncContext * s, int n, UINT16 **dc_val_pt return pred; } -void mpeg4_pred_ac(MpegEncContext * s, INT16 *block, int n, +void mpeg4_pred_ac(MpegEncContext * s, DCTELEM *block, int n, int dir) { int i; @@ -1907,7 +1907,7 @@ void mpeg4_pred_ac(MpegEncContext * s, INT16 *block, int n, } -static void mpeg4_inv_pred_ac(MpegEncContext * s, INT16 *block, int n, +static void mpeg4_inv_pred_ac(MpegEncContext * s, DCTELEM *block, int n, int dir) { int i; diff --git a/libavcodec/mpegvideo.h b/libavcodec/mpegvideo.h index 6f9d5b6989..2198429ffb 100644 --- a/libavcodec/mpegvideo.h +++ b/libavcodec/mpegvideo.h @@ -692,7 +692,7 @@ void h263_encode_picture_header(MpegEncContext *s, int picture_number); int h263_encode_gob_header(MpegEncContext * s, int mb_line); INT16 *h263_pred_motion(MpegEncContext * s, int block, int *px, int *py); -void mpeg4_pred_ac(MpegEncContext * s, INT16 *block, int n, +void mpeg4_pred_ac(MpegEncContext * s, DCTELEM *block, int n, int dir); void ff_set_mpeg4_time(MpegEncContext * s, int picture_number); void mpeg4_encode_picture_header(MpegEncContext *s, int picture_number); diff --git a/libavcodec/simple_idct.c b/libavcodec/simple_idct.c index 8c9ce7b936..703e94f213 100644 --- a/libavcodec/simple_idct.c +++ b/libavcodec/simple_idct.c @@ -67,7 +67,7 @@ #endif -static inline void idctRowCondDC (int16_t * row) +static inline void idctRowCondDC (DCTELEM * row) { int a0, a1, a2, a3, b0, b1, b2, b3; #ifdef FAST_64BIT @@ -82,26 +82,40 @@ static inline void idctRowCondDC (int16_t * row) #else #define ROW0_MASK 0xffffLL #endif - if ( ((((uint64_t *)row)[0] & ~ROW0_MASK) | - ((uint64_t *)row)[1]) == 0) { - temp = (row[0] << 3) & 0xffff; - temp += temp << 16; - temp += temp << 32; - ((uint64_t *)row)[0] = temp; - ((uint64_t *)row)[1] = temp; - return; - } + if(sizeof(DCTELEM)==2){ + if ( ((((uint64_t *)row)[0] & ~ROW0_MASK) | + ((uint64_t *)row)[1]) == 0) { + temp = (row[0] << 3) & 0xffff; + temp += temp << 16; + temp += temp << 32; + ((uint64_t *)row)[0] = temp; + ((uint64_t *)row)[1] = temp; + return; + } + }else{ + if (!(row[1]|row[2]|row[3]|row[4]|row[5]|row[6]|row[7])) { + row[0]=row[1]=row[2]=row[3]=row[4]=row[5]=row[6]=row[7]= row[0] << 3; + return; + } + } #else - if (!(((uint32_t*)row)[1] | - ((uint32_t*)row)[2] | - ((uint32_t*)row)[3] | - row[1])) { - temp = (row[0] << 3) & 0xffff; - temp += temp << 16; - ((uint32_t*)row)[0]=((uint32_t*)row)[1] = - ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp; - return; - } + if(sizeof(DCTELEM)==2){ + if (!(((uint32_t*)row)[1] | + ((uint32_t*)row)[2] | + ((uint32_t*)row)[3] | + row[1])) { + temp = (row[0] << 3) & 0xffff; + temp += temp << 16; + ((uint32_t*)row)[0]=((uint32_t*)row)[1] = + ((uint32_t*)row)[2]=((uint32_t*)row)[3] = temp; + return; + } + }else{ + if (!(row[1]|row[2]|row[3]|row[4]|row[5]|row[6]|row[7])) { + row[0]=row[1]=row[2]=row[3]=row[4]=row[5]=row[6]=row[7]= row[0] << 3; + return; + } + } #endif a0 = (W4 * row[0]) + (1 << (ROW_SHIFT - 1)); @@ -159,7 +173,7 @@ static inline void idctRowCondDC (int16_t * row) } static inline void idctSparseColPut (UINT8 *dest, int line_size, - int16_t * col) + DCTELEM * col) { int a0, a1, a2, a3, b0, b1, b2, b3; UINT8 *cm = cropTbl + MAX_NEG_CROP; @@ -231,7 +245,7 @@ static inline void idctSparseColPut (UINT8 *dest, int line_size, } static inline void idctSparseColAdd (UINT8 *dest, int line_size, - int16_t * col) + DCTELEM * col) { int a0, a1, a2, a3, b0, b1, b2, b3; UINT8 *cm = cropTbl + MAX_NEG_CROP; @@ -302,7 +316,7 @@ static inline void idctSparseColAdd (UINT8 *dest, int line_size, dest[0] = cm[dest[0] + ((a0 - b0) >> COL_SHIFT)]; } -static inline void idctSparseCol (int16_t * col) +static inline void idctSparseCol (DCTELEM * col) { int a0, a1, a2, a3, b0, b1, b2, b3; @@ -365,7 +379,7 @@ static inline void idctSparseCol (int16_t * col) col[56] = ((a0 - b0) >> COL_SHIFT); } -void simple_idct_put(UINT8 *dest, int line_size, INT16 *block) +void simple_idct_put(UINT8 *dest, int line_size, DCTELEM *block) { int i; for(i=0; i<8; i++) @@ -375,7 +389,7 @@ void simple_idct_put(UINT8 *dest, int line_size, INT16 *block) idctSparseColPut(dest + i, line_size, block + i); } -void simple_idct_add(UINT8 *dest, int line_size, INT16 *block) +void simple_idct_add(UINT8 *dest, int line_size, DCTELEM *block) { int i; for(i=0; i<8; i++) @@ -385,7 +399,7 @@ void simple_idct_add(UINT8 *dest, int line_size, INT16 *block) idctSparseColAdd(dest + i, line_size, block + i); } -void simple_idct(INT16 *block) +void simple_idct(DCTELEM *block) { int i; for(i=0; i<8; i++) @@ -406,7 +420,7 @@ void simple_idct(INT16 *block) and the butterfly must be multiplied by 0.5 * sqrt(2.0) */ #define C_SHIFT (4+1+12) -static inline void idct4col(UINT8 *dest, int line_size, const INT16 *col) +static inline void idct4col(UINT8 *dest, int line_size, const DCTELEM *col) { int c0, c1, c2, c3, a0, a1, a2, a3; const UINT8 *cm = cropTbl + MAX_NEG_CROP; @@ -443,10 +457,10 @@ static inline void idct4col(UINT8 *dest, int line_size, const INT16 *col) /* XXX: I think a 1.0/sqrt(2) normalization should be needed to compensate the extra butterfly stage - I don't have the full DV specification */ -void simple_idct248_put(UINT8 *dest, int line_size, INT16 *block) +void simple_idct248_put(UINT8 *dest, int line_size, DCTELEM *block) { int i; - INT16 *ptr; + DCTELEM *ptr; /* butterfly */ ptr = block; @@ -486,7 +500,7 @@ void simple_idct248_put(UINT8 *dest, int line_size, INT16 *block) #define C2 C_FIX(0.2705980501) #define C3 C_FIX(0.5) #define C_SHIFT (4+1+12) -static inline void idct4col_add(UINT8 *dest, int line_size, const INT16 *col) +static inline void idct4col_add(UINT8 *dest, int line_size, const DCTELEM *col) { int c0, c1, c2, c3, a0, a1, a2, a3; const UINT8 *cm = cropTbl + MAX_NEG_CROP; @@ -514,7 +528,7 @@ static inline void idct4col_add(UINT8 *dest, int line_size, const INT16 *col) #define R2 R_FIX(0.2705980501) #define R3 R_FIX(0.5) #define R_SHIFT 11 -static inline void idct4row(INT16 *row) +static inline void idct4row(DCTELEM *row) { int c0, c1, c2, c3, a0, a1, a2, a3; const UINT8 *cm = cropTbl + MAX_NEG_CROP; @@ -533,7 +547,7 @@ static inline void idct4row(INT16 *row) row[3]= (c0 - c1) >> R_SHIFT; } -void simple_idct84_add(UINT8 *dest, int line_size, INT16 *block) +void simple_idct84_add(UINT8 *dest, int line_size, DCTELEM *block) { int i; @@ -548,7 +562,7 @@ void simple_idct84_add(UINT8 *dest, int line_size, INT16 *block) } } -void simple_idct48_add(UINT8 *dest, int line_size, INT16 *block) +void simple_idct48_add(UINT8 *dest, int line_size, DCTELEM *block) { int i; diff --git a/libavcodec/simple_idct.h b/libavcodec/simple_idct.h index 428c6072c8..0ee1e05ed3 100644 --- a/libavcodec/simple_idct.h +++ b/libavcodec/simple_idct.h @@ -18,14 +18,14 @@ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ -void simple_idct_put(UINT8 *dest, int line_size, INT16 *block); -void simple_idct_add(UINT8 *dest, int line_size, INT16 *block); -void ff_simple_idct_mmx(short *block); -void ff_simple_idct_add_mmx(UINT8 *dest, int line_size, INT16 *block); -void ff_simple_idct_put_mmx(UINT8 *dest, int line_size, INT16 *block); -void simple_idct(short *block); +void simple_idct_put(UINT8 *dest, int line_size, DCTELEM *block); +void simple_idct_add(UINT8 *dest, int line_size, DCTELEM *block); +void ff_simple_idct_mmx(int16_t *block); +void ff_simple_idct_add_mmx(UINT8 *dest, int line_size, int16_t *block); +void ff_simple_idct_put_mmx(UINT8 *dest, int line_size, int16_t *block); +void simple_idct(DCTELEM *block); -void simple_idct248_put(UINT8 *dest, int line_size, INT16 *block); +void simple_idct248_put(UINT8 *dest, int line_size, DCTELEM *block); -void simple_idct84_add(UINT8 *dest, int line_size, INT16 *block); -void simple_idct48_add(UINT8 *dest, int line_size, INT16 *block); +void simple_idct84_add(UINT8 *dest, int line_size, DCTELEM *block); +void simple_idct48_add(UINT8 *dest, int line_size, DCTELEM *block);