yading@10: /* yading@10: * DSP utils yading@10: * Copyright (c) 2000, 2001, 2002 Fabrice Bellard yading@10: * Copyright (c) 2002-2004 Michael Niedermayer yading@10: * yading@10: * This file is part of FFmpeg. yading@10: * yading@10: * FFmpeg is free software; you can redistribute it and/or yading@10: * modify it under the terms of the GNU Lesser General Public yading@10: * License as published by the Free Software Foundation; either yading@10: * version 2.1 of the License, or (at your option) any later version. yading@10: * yading@10: * FFmpeg is distributed in the hope that it will be useful, yading@10: * but WITHOUT ANY WARRANTY; without even the implied warranty of yading@10: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU yading@10: * Lesser General Public License for more details. yading@10: * yading@10: * You should have received a copy of the GNU Lesser General Public yading@10: * License along with FFmpeg; if not, write to the Free Software yading@10: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA yading@10: */ yading@10: yading@10: /** yading@10: * @file yading@10: * DSP utils. yading@10: * note, many functions in here may use MMX which trashes the FPU state, it is yading@10: * absolutely necessary to call emms_c() between dsp & float/double code yading@10: */ yading@10: yading@10: #ifndef AVCODEC_DSPUTIL_H yading@10: #define AVCODEC_DSPUTIL_H yading@10: yading@10: #include "libavutil/intreadwrite.h" yading@10: #include "avcodec.h" yading@10: #include "rnd_avg.h" yading@10: yading@10: yading@10: //#define DEBUG yading@10: yading@10: /* encoding scans */ yading@10: extern const uint8_t ff_alternate_horizontal_scan[64]; yading@10: extern const uint8_t ff_alternate_vertical_scan[64]; yading@10: extern const uint8_t ff_zigzag_direct[64]; yading@10: extern const uint8_t ff_zigzag248_direct[64]; yading@10: yading@10: /* pixel operations */ yading@10: #define MAX_NEG_CROP 1024 yading@10: yading@10: /* temporary */ yading@10: extern uint32_t ff_squareTbl[512]; yading@10: extern const uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP]; yading@10: yading@10: void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: yading@10: /* RV40 functions */ yading@10: void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride); yading@10: yading@10: void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, yading@10: int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); yading@10: yading@10: /* minimum alignment rules ;) yading@10: If you notice errors in the align stuff, need more alignment for some ASM code yading@10: for some CPU or need to use a function with less aligned data then send a mail yading@10: to the ffmpeg-devel mailing list, ... yading@10: yading@10: !warning These alignments might not match reality, (missing attribute((align)) yading@10: stuff somewhere possible). yading@10: I (Michael) did not check them, these are just the alignments which I think yading@10: could be reached easily ... yading@10: yading@10: !future video codecs might need functions with less strict alignment yading@10: */ yading@10: yading@10: /* add and put pixel (decoding) */ yading@10: // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 yading@10: //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4 yading@10: typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h); yading@10: typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride); yading@10: yading@10: typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h); yading@10: yading@10: #define DEF_OLD_QPEL(name)\ yading@10: void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);\ yading@10: void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);\ yading@10: void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride); yading@10: yading@10: DEF_OLD_QPEL(qpel16_mc11_old_c) yading@10: DEF_OLD_QPEL(qpel16_mc31_old_c) yading@10: DEF_OLD_QPEL(qpel16_mc12_old_c) yading@10: DEF_OLD_QPEL(qpel16_mc32_old_c) yading@10: DEF_OLD_QPEL(qpel16_mc13_old_c) yading@10: DEF_OLD_QPEL(qpel16_mc33_old_c) yading@10: DEF_OLD_QPEL(qpel8_mc11_old_c) yading@10: DEF_OLD_QPEL(qpel8_mc31_old_c) yading@10: DEF_OLD_QPEL(qpel8_mc12_old_c) yading@10: DEF_OLD_QPEL(qpel8_mc32_old_c) yading@10: DEF_OLD_QPEL(qpel8_mc13_old_c) yading@10: DEF_OLD_QPEL(qpel8_mc33_old_c) yading@10: yading@10: /* motion estimation */ yading@10: // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2 yading@10: // although currently h<4 is not used as functions with width <8 are neither used nor implemented yading@10: typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/; yading@10: yading@10: /** yading@10: * Scantable. yading@10: */ yading@10: typedef struct ScanTable{ yading@10: const uint8_t *scantable; yading@10: uint8_t permutated[64]; yading@10: uint8_t raster_end[64]; yading@10: } ScanTable; yading@10: yading@10: void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); yading@10: void ff_init_scantable_permutation(uint8_t *idct_permutation, yading@10: int idct_permutation_type); yading@10: yading@10: /** yading@10: * DSPContext. yading@10: */ yading@10: typedef struct DSPContext { yading@10: /* pixel ops : interface with DCT */ yading@10: void (*get_pixels)(int16_t *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); yading@10: void (*diff_pixels)(int16_t *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); yading@10: void (*put_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); yading@10: void (*put_signed_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); yading@10: void (*add_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); yading@10: void (*add_pixels8)(uint8_t *pixels, int16_t *block, int line_size); yading@10: int (*sum_abs_dctelem)(int16_t *block/*align 16*/); yading@10: /** yading@10: * translational global motion compensation. yading@10: */ yading@10: void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); yading@10: /** yading@10: * global motion compensation. yading@10: */ yading@10: void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, yading@10: int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); yading@10: void (*clear_block)(int16_t *block/*align 16*/); yading@10: void (*clear_blocks)(int16_t *blocks/*align 16*/); yading@10: int (*pix_sum)(uint8_t * pix, int line_size); yading@10: int (*pix_norm1)(uint8_t * pix, int line_size); yading@10: // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 yading@10: yading@10: me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */ yading@10: me_cmp_func sse[6]; yading@10: me_cmp_func hadamard8_diff[6]; yading@10: me_cmp_func dct_sad[6]; yading@10: me_cmp_func quant_psnr[6]; yading@10: me_cmp_func bit[6]; yading@10: me_cmp_func rd[6]; yading@10: me_cmp_func vsad[6]; yading@10: me_cmp_func vsse[6]; yading@10: me_cmp_func nsse[6]; yading@10: me_cmp_func w53[6]; yading@10: me_cmp_func w97[6]; yading@10: me_cmp_func dct_max[6]; yading@10: me_cmp_func dct264_sad[6]; yading@10: yading@10: me_cmp_func me_pre_cmp[6]; yading@10: me_cmp_func me_cmp[6]; yading@10: me_cmp_func me_sub_cmp[6]; yading@10: me_cmp_func mb_cmp[6]; yading@10: me_cmp_func ildct_cmp[6]; //only width 16 used yading@10: me_cmp_func frame_skip_cmp[6]; //only width 8 used yading@10: yading@10: int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, yading@10: int size); yading@10: yading@10: /** yading@10: * Thirdpel motion compensation with rounding (a+b+1)>>1. yading@10: * this is an array[12] of motion compensation functions for the 9 thirdpe yading@10: * positions
yading@10: * *pixels_tab[ xthirdpel + 4*ythirdpel ] yading@10: * @param block destination where the result is stored yading@10: * @param pixels source yading@10: * @param line_size number of bytes in a horizontal line of block yading@10: * @param h height yading@10: */ yading@10: tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width? yading@10: tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width? yading@10: yading@10: qpel_mc_func put_qpel_pixels_tab[2][16]; yading@10: qpel_mc_func avg_qpel_pixels_tab[2][16]; yading@10: qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; yading@10: qpel_mc_func put_mspel_pixels_tab[8]; yading@10: yading@10: me_cmp_func pix_abs[2][4]; yading@10: yading@10: /* huffyuv specific */ yading@10: void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); yading@10: void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const uint8_t *src2/*align 1*/,int w); yading@10: /** yading@10: * subtract huffyuv's variant of median prediction yading@10: * note, this might read from src1[-1], src2[-1] yading@10: */ yading@10: void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top); yading@10: void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top); yading@10: int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left); yading@10: void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha); yading@10: /* this might write to dst[w] */ yading@10: void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); yading@10: void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len); yading@10: yading@10: void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); yading@10: void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); yading@10: yading@10: /* assume len is a multiple of 8, and arrays are 16-byte aligned */ yading@10: void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */); yading@10: yading@10: /* (I)DCT */ yading@10: void (*fdct)(int16_t *block/* align 16*/); yading@10: void (*fdct248)(int16_t *block/* align 16*/); yading@10: yading@10: /* IDCT really*/ yading@10: void (*idct)(int16_t *block/* align 16*/); yading@10: yading@10: /** yading@10: * block -> idct -> clip to unsigned 8 bit -> dest. yading@10: * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) yading@10: * @param line_size size in bytes of a horizontal line of dest yading@10: */ yading@10: void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, int16_t *block/*align 16*/); yading@10: yading@10: /** yading@10: * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. yading@10: * @param line_size size in bytes of a horizontal line of dest yading@10: */ yading@10: void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, int16_t *block/*align 16*/); yading@10: yading@10: /** yading@10: * idct input permutation. yading@10: * several optimized IDCTs need a permutated input (relative to the normal order of the reference yading@10: * IDCT) yading@10: * this permutation must be performed before the idct_put/add, note, normally this can be merged yading@10: * with the zigzag/alternate scan
yading@10: * an example to avoid confusion: yading@10: * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) yading@10: * - (x -> reference dct -> reference idct -> x) yading@10: * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) yading@10: * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) yading@10: */ yading@10: uint8_t idct_permutation[64]; yading@10: int idct_permutation_type; yading@10: #define FF_NO_IDCT_PERM 1 yading@10: #define FF_LIBMPEG2_IDCT_PERM 2 yading@10: #define FF_SIMPLE_IDCT_PERM 3 yading@10: #define FF_TRANSPOSE_IDCT_PERM 4 yading@10: #define FF_PARTTRANS_IDCT_PERM 5 yading@10: #define FF_SSE2_IDCT_PERM 6 yading@10: yading@10: int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); yading@10: void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); yading@10: #define BASIS_SHIFT 16 yading@10: #define RECON_SHIFT 6 yading@10: yading@10: void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides); yading@10: #define EDGE_WIDTH 16 yading@10: #define EDGE_TOP 1 yading@10: #define EDGE_BOTTOM 2 yading@10: yading@10: void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); yading@10: yading@10: /** yading@10: * Calculate scalar product of two vectors. yading@10: * @param len length of vectors, should be multiple of 16 yading@10: */ yading@10: int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len); yading@10: /* ape functions */ yading@10: /** yading@10: * Calculate scalar product of v1 and v2, yading@10: * and v1[i] += v3[i] * mul yading@10: * @param len length of vectors, should be multiple of 16 yading@10: */ yading@10: int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul); yading@10: yading@10: /** yading@10: * Apply symmetric window in 16-bit fixed-point. yading@10: * @param output destination array yading@10: * constraints: 16-byte aligned yading@10: * @param input source array yading@10: * constraints: 16-byte aligned yading@10: * @param window window array yading@10: * constraints: 16-byte aligned, at least len/2 elements yading@10: * @param len full window length yading@10: * constraints: multiple of ? greater than zero yading@10: */ yading@10: void (*apply_window_int16)(int16_t *output, const int16_t *input, yading@10: const int16_t *window, unsigned int len); yading@10: yading@10: /** yading@10: * Clip each element in an array of int32_t to a given minimum and maximum value. yading@10: * @param dst destination array yading@10: * constraints: 16-byte aligned yading@10: * @param src source array yading@10: * constraints: 16-byte aligned yading@10: * @param min minimum value yading@10: * constraints: must be in the range [-(1 << 24), 1 << 24] yading@10: * @param max maximum value yading@10: * constraints: must be in the range [-(1 << 24), 1 << 24] yading@10: * @param len number of elements in the array yading@10: * constraints: multiple of 32 greater than zero yading@10: */ yading@10: void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min, yading@10: int32_t max, unsigned int len); yading@10: yading@10: op_fill_func fill_block_tab[2]; yading@10: } DSPContext; yading@10: yading@10: void ff_dsputil_static_init(void); yading@10: void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx); yading@10: void avpriv_dsputil_init(DSPContext* p, AVCodecContext *avctx); yading@10: attribute_deprecated void dsputil_init(DSPContext* c, AVCodecContext *avctx); yading@10: yading@10: int ff_check_alignment(void); yading@10: yading@10: void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type); yading@10: yading@10: void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); yading@10: void ff_dsputil_init_arm(DSPContext* c, AVCodecContext *avctx); yading@10: void ff_dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); yading@10: void ff_dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); yading@10: void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); yading@10: void ff_dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx); yading@10: void ff_dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); yading@10: yading@10: void ff_dsputil_init_dwt(DSPContext *c); yading@10: yading@10: #endif /* AVCODEC_DSPUTIL_H */