X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fdsputil.h;h=f54d74d285cf7c4695b359e762d72c916557b737;hb=f988ce6cad8a98af1eaf260fb6995deac65542e5;hp=80bcd62dc965f7c20f41d572e33cb2a082718f9a;hpb=ddeaf723c5e97d1ae81a4827b73f99f3bc3d86f3;p=ffmpeg diff --git a/libavcodec/dsputil.h b/libavcodec/dsputil.h index 80bcd62dc96..f54d74d285c 100644 --- a/libavcodec/dsputil.h +++ b/libavcodec/dsputil.h @@ -1,6 +1,6 @@ /* * DSP utils - * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. + * Copyright (c) 2000, 2001, 2002 Fabrice Bellard * Copyright (c) 2002-2004 Michael Niedermayer * * This file is part of FFmpeg. @@ -21,15 +21,16 @@ */ /** - * @file dsputil.h + * @file libavcodec/dsputil.h * DSP utils. * note, many functions in here may use MMX which trashes the FPU state, it is * absolutely necessary to call emms_c() between dsp & float/double code */ -#ifndef DSPUTIL_H -#define DSPUTIL_H +#ifndef AVCODEC_DSPUTIL_H +#define AVCODEC_DSPUTIL_H +#include "libavutil/intreadwrite.h" #include "avcodec.h" @@ -37,6 +38,7 @@ /* dct code */ typedef short DCTELEM; typedef int DWTELEM; +typedef short IDWTELEM; void fdct_ifast (DCTELEM *data); void fdct_ifast248 (DCTELEM *data); @@ -47,7 +49,7 @@ void j_rev_dct (DCTELEM *data); void j_rev_dct4 (DCTELEM *data); void j_rev_dct2 (DCTELEM *data); void j_rev_dct1 (DCTELEM *data); -void ff_mpeg_idct_c(DCTELEM *data); +void ff_wmv2_idct_c(DCTELEM *data); void ff_fdct_mmx(DCTELEM *block); void ff_fdct_mmx2(DCTELEM *block); @@ -59,10 +61,17 @@ void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride); void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block); void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block); +void ff_h264_idct_add16_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); +void ff_h264_idct_add16intra_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); +void ff_h264_idct8_add4_c(uint8_t *dst, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); +void ff_h264_idct_add8_c(uint8_t **dest, const int *blockoffset, DCTELEM *block, int stride, const uint8_t nnzc[6*8]); void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1, const float *src2, int src3, int blocksize, int step); -void ff_float_to_int16_c(int16_t *dst, const float *src, int len); +void ff_vector_fmul_window_c(float *dst, const float *src0, const float *src1, + const float *win, float add_bias, int len); +void ff_float_to_int16_c(int16_t *dst, const float *src, long len); +void ff_float_to_int16_interleave_c(int16_t *dst, const float **src, long len, int channels); /* encoding scans */ extern const uint8_t ff_alternate_horizontal_scan[64]; @@ -82,6 +91,13 @@ void ff_vp3_idct_c(DCTELEM *block/* align 16*/); void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); +void ff_vp3_v_loop_filter_c(uint8_t *src, int stride, int *bounding_values); +void ff_vp3_h_loop_filter_c(uint8_t *src, int stride, int *bounding_values); + +/* VP6 DSP functions */ +void ff_vp6_filter_diag4_c(uint8_t *dst, uint8_t *src, int stride, + const int16_t *h_weights, const int16_t *v_weights); + /* 1/2^n downscaling functions from imgconvert.c */ void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); @@ -92,11 +108,14 @@ void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy, int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); /* minimum alignment rules ;) -if u notice errors in the align stuff, need more alignment for some asm code for some cpu -or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... +If you notice errors in the align stuff, need more alignment for some ASM code +for some CPU or need to use a function with less aligned data then send a mail +to the ffmpeg-devel mailing list, ... -!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) -i (michael) didnt check them, these are just the alignents which i think could be reached easily ... +!warning These alignments might not match reality, (missing attribute((align)) +stuff somewhere possible). +I (Michael) did not check them, these are just the alignments which I think +could be reached easily ... !future video codecs might need functions with less strict alignment */ @@ -152,6 +171,25 @@ typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align widt // for snow slices typedef struct slice_buffer_s slice_buffer; +/** + * Scantable. + */ +typedef struct ScanTable{ + const uint8_t *scantable; + uint8_t permutated[64]; + uint8_t raster_end[64]; +#if ARCH_PPC + /** Used by dct_quantize_altivec to find last-non-zero */ + DECLARE_ALIGNED(16, uint8_t, inverse[64]); +#endif +} ScanTable; + +void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable); + +void ff_emulated_edge_mc(uint8_t *buf, uint8_t *src, int linesize, + int block_w, int block_h, + int src_x, int src_y, int w, int h); + /** * DSPContext. */ @@ -174,32 +212,33 @@ typedef struct DSPContext { */ void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); + void (*clear_block)(DCTELEM *block/*align 16*/); void (*clear_blocks)(DCTELEM *blocks/*align 16*/); int (*pix_sum)(uint8_t * pix, int line_size); int (*pix_norm1)(uint8_t * pix, int line_size); // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4 - me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */ - me_cmp_func sse[5]; - me_cmp_func hadamard8_diff[5]; - me_cmp_func dct_sad[5]; - me_cmp_func quant_psnr[5]; - me_cmp_func bit[5]; - me_cmp_func rd[5]; - me_cmp_func vsad[5]; - me_cmp_func vsse[5]; - me_cmp_func nsse[5]; - me_cmp_func w53[5]; - me_cmp_func w97[5]; - me_cmp_func dct_max[5]; - me_cmp_func dct264_sad[5]; - - me_cmp_func me_pre_cmp[5]; - me_cmp_func me_cmp[5]; - me_cmp_func me_sub_cmp[5]; - me_cmp_func mb_cmp[5]; - me_cmp_func ildct_cmp[5]; //only width 16 used - me_cmp_func frame_skip_cmp[5]; //only width 8 used + me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */ + me_cmp_func sse[6]; + me_cmp_func hadamard8_diff[6]; + me_cmp_func dct_sad[6]; + me_cmp_func quant_psnr[6]; + me_cmp_func bit[6]; + me_cmp_func rd[6]; + me_cmp_func vsad[6]; + me_cmp_func vsse[6]; + me_cmp_func nsse[6]; + me_cmp_func w53[6]; + me_cmp_func w97[6]; + me_cmp_func dct_max[6]; + me_cmp_func dct264_sad[6]; + + me_cmp_func me_pre_cmp[6]; + me_cmp_func me_cmp[6]; + me_cmp_func me_sub_cmp[6]; + me_cmp_func mb_cmp[6]; + me_cmp_func ildct_cmp[6]; //only width 16 used + me_cmp_func frame_skip_cmp[6]; //only width 8 used int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2, int size); @@ -277,9 +316,10 @@ typedef struct DSPContext { * h264 Chroma MC */ h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; - /* This is really one func used in VC-1 decoding */ - h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3]; h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; + /* This is really one func used in VC-1 decoding */ + h264_chroma_mc_func put_no_rnd_vc1_chroma_pixels_tab[3]; + h264_chroma_mc_func avg_no_rnd_vc1_chroma_pixels_tab[3]; qpel_mc_func put_h264_qpel_pixels_tab[4][16]; qpel_mc_func avg_h264_qpel_pixels_tab[4][16]; @@ -303,40 +343,64 @@ typedef struct DSPContext { /* huffyuv specific */ void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); + void (*add_bytes_l2)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/, int w); void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); /** * subtract huffyuv's variant of median prediction * note, this might read from src1[-1], src2[-1] */ void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top); - void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w); - - void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0); - void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); - void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta); + void (*add_hfyu_median_prediction)(uint8_t *dst, uint8_t *top, uint8_t *diff, int w, int *left, int *left_top); + /* this might write to dst[w] */ + void (*add_png_paeth_prediction)(uint8_t *dst, uint8_t *src, uint8_t *top, int w, int bpp); + void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w); + + void (*h264_v_loop_filter_luma)(uint8_t *pix/*align 16*/, int stride, int alpha, int beta, int8_t *tc0); + void (*h264_h_loop_filter_luma)(uint8_t *pix/*align 4 */, int stride, int alpha, int beta, int8_t *tc0); + /* v/h_loop_filter_luma_intra: align 16 */ + void (*h264_v_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); + void (*h264_h_loop_filter_luma_intra)(uint8_t *pix, int stride, int alpha, int beta); + void (*h264_v_loop_filter_chroma)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta, int8_t *tc0); + void (*h264_h_loop_filter_chroma)(uint8_t *pix/*align 4*/, int stride, int alpha, int beta, int8_t *tc0); + void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); + void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix/*align 8*/, int stride, int alpha, int beta); // h264_loop_filter_strength: simd only. the C version is inlined in h264.c void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2], - int bidir, int edges, int step, int mask_mv0, int mask_mv1); + int bidir, int edges, int step, int mask_mv0, int mask_mv1, int field); void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale); void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale); void (*h261_loop_filter)(uint8_t *src, int stride); + void (*x8_v_loop_filter)(uint8_t *src, int stride, int qscale); + void (*x8_h_loop_filter)(uint8_t *src, int stride, int qscale); + + void (*vp3_v_loop_filter)(uint8_t *src, int stride, int *bounding_values); + void (*vp3_h_loop_filter)(uint8_t *src, int stride, int *bounding_values); + + void (*vp6_filter_diag4)(uint8_t *dst, uint8_t *src, int stride, + const int16_t *h_weights,const int16_t *v_weights); + /* assume len is a multiple of 4, and arrays are 16-byte aligned */ void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize); + void (*ac3_downmix)(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len); + /* no alignment needed */ + void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc); /* assume len is a multiple of 8, and arrays are 16-byte aligned */ void (*vector_fmul)(float *dst, const float *src, int len); void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len); /* assume len is a multiple of 8, and src arrays are 16-byte aligned */ void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step); + /* assume len is a multiple of 4, and arrays are 16-byte aligned */ + void (*vector_fmul_window)(float *dst, const float *src0, const float *src1, const float *win, float add_bias, int len); + /* assume len is a multiple of 8, and arrays are 16-byte aligned */ + void (*int32_to_float_fmul_scalar)(float *dst, const int *src, float mul, int len); /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767] * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */ - void (*float_to_int16)(int16_t *dst, const float *src, int len); + void (*float_to_int16)(int16_t *dst, const float *src, long len); + void (*float_to_int16_interleave)(int16_t *dst, const float **src, long len, int channels); /* (I)DCT */ void (*fdct)(DCTELEM *block/* align 16*/); @@ -348,13 +412,13 @@ typedef struct DSPContext { /** * 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 + * @param line_size size in bytes of a horizontal 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 + * @param line_size size in bytes of a horizontal line of dest */ void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); @@ -377,39 +441,97 @@ typedef struct DSPContext { #define FF_SIMPLE_IDCT_PERM 3 #define FF_TRANSPOSE_IDCT_PERM 4 #define FF_PARTTRANS_IDCT_PERM 5 +#define FF_SSE2_IDCT_PERM 6 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale); void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale); #define BASIS_SHIFT 16 #define RECON_SHIFT 6 + void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w); +#define EDGE_WIDTH 16 + /* h264 functions */ - void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride); - void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride); - void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride); - void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride); + /* NOTE!!! if you implement any of h264_idct8_add, h264_idct8_add4 then you must implement all of them + NOTE!!! if you implement any of h264_idct_add, h264_idct_add16, h264_idct_add16intra, h264_idct_add8 then you must implement all of them + The reason for above, is that no 2 out of one list may use a different permutation. + */ + void (*h264_idct_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); + void (*h264_idct8_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); + void (*h264_idct_dc_add)(uint8_t *dst/*align 4*/, DCTELEM *block/*align 16*/, int stride); + void (*h264_idct8_dc_add)(uint8_t *dst/*align 8*/, DCTELEM *block/*align 16*/, int stride); void (*h264_dct)(DCTELEM block[4][4]); + void (*h264_idct_add16)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); + void (*h264_idct8_add4)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); + void (*h264_idct_add8)(uint8_t **dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); + void (*h264_idct_add16intra)(uint8_t *dst/*align 16*/, const int *blockoffset, DCTELEM *block/*align 16*/, int stride, const uint8_t nnzc[6*8]); /* snow wavelet */ - void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width); - void (*horizontal_compose97i)(DWTELEM *b, int width); + void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width); + void (*horizontal_compose97i)(IDWTELEM *b, int width); void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8); void (*prefetch)(void *mem, int stride, int h); void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height); + /* mlp/truehd functions */ + void (*mlp_filter_channel)(int32_t *state, const int32_t *coeff, + int firorder, int iirorder, + unsigned int filter_shift, int32_t mask, int blocksize, + int32_t *sample_buffer); + /* vc1 functions */ void (*vc1_inv_trans_8x8)(DCTELEM *b); - void (*vc1_inv_trans_8x4)(DCTELEM *b, int n); - void (*vc1_inv_trans_4x8)(DCTELEM *b, int n); - void (*vc1_inv_trans_4x4)(DCTELEM *b, int n); + void (*vc1_inv_trans_8x4)(uint8_t *dest, int line_size, DCTELEM *block); + void (*vc1_inv_trans_4x8)(uint8_t *dest, int line_size, DCTELEM *block); + void (*vc1_inv_trans_4x4)(uint8_t *dest, int line_size, DCTELEM *block); void (*vc1_v_overlap)(uint8_t* src, int stride); void (*vc1_h_overlap)(uint8_t* src, int stride); + void (*vc1_v_loop_filter4)(uint8_t *src, int stride, int pq); + void (*vc1_h_loop_filter4)(uint8_t *src, int stride, int pq); + void (*vc1_v_loop_filter8)(uint8_t *src, int stride, int pq); + void (*vc1_h_loop_filter8)(uint8_t *src, int stride, int pq); + void (*vc1_v_loop_filter16)(uint8_t *src, int stride, int pq); + void (*vc1_h_loop_filter16)(uint8_t *src, int stride, int pq); /* put 8x8 block with bicubic interpolation and quarterpel precision * last argument is actually round value instead of height */ op_pixels_func put_vc1_mspel_pixels_tab[16]; + op_pixels_func avg_vc1_mspel_pixels_tab[16]; + + /* intrax8 functions */ + void (*x8_spatial_compensation[12])(uint8_t *src , uint8_t *dst, int linesize); + void (*x8_setup_spatial_compensation)(uint8_t *src, uint8_t *dst, int linesize, + int * range, int * sum, int edges); + + /* ape functions */ + /** + * Add contents of the second vector to the first one. + * @param len length of vectors, should be multiple of 16 + */ + void (*add_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len); + /** + * Add contents of the second vector to the first one. + * @param len length of vectors, should be multiple of 16 + */ + void (*sub_int16)(int16_t *v1/*align 16*/, int16_t *v2, int len); + /** + * Calculate scalar product of two vectors. + * @param len length of vectors, should be multiple of 16 + * @param shift number of bits to discard from product + */ + int32_t (*scalarproduct_int16)(int16_t *v1, int16_t *v2/*align 16*/, int len, int shift); + + /* rv30 functions */ + qpel_mc_func put_rv30_tpel_pixels_tab[4][16]; + qpel_mc_func avg_rv30_tpel_pixels_tab[4][16]; + + /* rv40 functions */ + qpel_mc_func put_rv40_qpel_pixels_tab[4][16]; + qpel_mc_func avg_rv40_qpel_pixels_tab[4][16]; + h264_chroma_mc_func put_rv40_chroma_pixels_tab[3]; + h264_chroma_mc_func avg_rv40_chroma_pixels_tab[3]; } DSPContext; void dsputil_static_init(void); @@ -473,7 +595,7 @@ static inline int get_penalty_factor(int lambda, int lambda2, int type){ int mm_support(void); void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); -void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); +void dsputil_init_arm(DSPContext* c, AVCodecContext *avctx); void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx); void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); @@ -484,19 +606,10 @@ void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx); #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v) -#if defined(HAVE_MMX) +#if 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 */ -#define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */ -#define MM_SSE3 0x0040 /* Prescott SSE3 functions */ -#define MM_SSSE3 0x0080 /* Conroe SSSE3 functions */ - extern int mm_flags; void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); @@ -505,85 +618,52 @@ void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int li static inline void emms(void) { - __asm __volatile ("emms;":::"memory"); + __asm__ volatile ("emms;":::"memory"); } #define emms_c() \ {\ - if (mm_flags & MM_MMX)\ + if (mm_flags & FF_MM_MMX)\ emms();\ } -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) - -#define STRIDE_ALIGN 8 - 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 optimizations */ -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(4, t, v) -#define STRIDE_ALIGN 4 - -#define MM_IWMMXT 0x0100 /* XScale IWMMXT */ +#elif ARCH_ARM extern int mm_flags; -#elif defined(HAVE_MLIB) - -/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) -#define STRIDE_ALIGN 8 - -#elif defined(ARCH_SPARC) - -/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) -#define STRIDE_ALIGN 8 - -#elif defined(ARCH_ALPHA) - -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) -#define STRIDE_ALIGN 8 - -#elif defined(ARCH_POWERPC) +#if HAVE_NEON +# define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) +# define STRIDE_ALIGN 16 +#endif -#define MM_ALTIVEC 0x0001 /* standard AltiVec */ +#elif ARCH_PPC extern int mm_flags; -#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) -#define pixel altivec_pixel -#include -#undef pixel -#endif - #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) #define STRIDE_ALIGN 16 -#elif defined(HAVE_MMI) +#elif HAVE_MMI #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v) #define STRIDE_ALIGN 16 -#elif defined(ARCH_SH4) - -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) -#define STRIDE_ALIGN 8 - -#elif defined(ARCH_BFIN) +#else -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) -#define STRIDE_ALIGN 8 +#define mm_flags 0 +#define mm_support() 0 -#else +#endif -#define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) -#define STRIDE_ALIGN 8 +#ifndef DECLARE_ALIGNED_8 +# define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v) +#endif +#ifndef STRIDE_ALIGN +# define STRIDE_ALIGN 8 #endif /* PSNR */ @@ -609,19 +689,40 @@ typedef struct FFTContext { uint16_t *revtab; FFTComplex *exptab; FFTComplex *exptab1; /* only used by SSE code */ + FFTComplex *tmp_buf; + 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, FFTSample *tmp); + void (*imdct_calc)(struct MDCTContext *s, FFTSample *output, const FFTSample *input); + void (*imdct_half)(struct MDCTContext *s, FFTSample *output, const FFTSample *input); } FFTContext; +extern FFTSample* ff_cos_tabs[13]; + +/** + * Sets up a complex FFT. + * @param nbits log2 of the length of the input array + * @param inverse if 0 perform the forward transform, if 1 perform the inverse + */ int ff_fft_init(FFTContext *s, int nbits, int inverse); -void ff_fft_permute(FFTContext *s, FFTComplex *z); +void ff_fft_permute_c(FFTContext *s, FFTComplex *z); +void ff_fft_permute_sse(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_3dn(FFTContext *s, FFTComplex *z); void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z); void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z); +/** + * Do the permutation needed BEFORE calling ff_fft_calc(). + */ +static inline void ff_fft_permute(FFTContext *s, FFTComplex *z) +{ + s->fft_permute(s, z); +} +/** + * Do a complex FFT with the parameters defined in ff_fft_init(). The + * input data must be permuted before. No 1.0/sqrt(n) normalization is done. + */ static inline void ff_fft_calc(FFTContext *s, FFTComplex *z) { s->fft_calc(s, z); @@ -639,24 +740,85 @@ typedef struct MDCTContext { 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_imdct_calc_3dn2(MDCTContext *s, FFTSample *output, - const FFTSample *input, FFTSample *tmp); -void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output, - const FFTSample *input, FFTSample *tmp); -void ff_mdct_calc(MDCTContext *s, FFTSample *out, - const FFTSample *input, FFTSample *tmp); +static inline void ff_imdct_calc(MDCTContext *s, FFTSample *output, const FFTSample *input) +{ + s->fft.imdct_calc(s, output, input); +} +static inline void ff_imdct_half(MDCTContext *s, FFTSample *output, const FFTSample *input) +{ + s->fft.imdct_half(s, output, input); +} + +/** + * Generate a Kaiser-Bessel Derived Window. + * @param window pointer to half window + * @param alpha determines window shape + * @param n size of half window + */ +void ff_kbd_window_init(float *window, float alpha, int n); + +/** + * Generate a sine window. + * @param window pointer to half window + * @param n size of half window + */ +void ff_sine_window_init(float *window, int n); +extern float ff_sine_128 [ 128]; +extern float ff_sine_256 [ 256]; +extern float ff_sine_512 [ 512]; +extern float ff_sine_1024[1024]; +extern float ff_sine_2048[2048]; +extern float ff_sine_4096[4096]; +extern float *ff_sine_windows[6]; + +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_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_mdct_calc(MDCTContext *s, FFTSample *out, const FFTSample *input); void ff_mdct_end(MDCTContext *s); -#define WARPER8_16(name8, name16)\ +/* Real Discrete Fourier Transform */ + +enum RDFTransformType { + RDFT, + IRDFT, + RIDFT, + IRIDFT, +}; + +typedef struct { + int nbits; + int inverse; + int sign_convention; + + /* pre/post rotation tables */ + FFTSample *tcos; + FFTSample *tsin; + FFTContext fft; +} RDFTContext; + +/** + * Sets up a real FFT. + * @param nbits log2 of the length of the input array + * @param trans the type of transform + */ +int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans); +void ff_rdft_calc(RDFTContext *s, FFTSample *data); +void ff_rdft_end(RDFTContext *s); + +#define WRAPPER8_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)\ +#define WRAPPER8_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);\ @@ -671,80 +833,80 @@ static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int st } -static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h) +static inline void copy_block2(uint8_t *dst, const uint8_t *src, int dstStride, int srcStride, int h) { int i; for(i=0; i