FFmpeg: libavcodec/x86/dsputil

00001 /*
00002  * MMX optimized DSP utils
00003  * Copyright (c) 2007  Aurelien Jacobs <aurel@gnuage.org>
00004  *
00005  * This file is part of FFmpeg.
00006  *
00007  * FFmpeg is free software; you can redistribute it and/or
00008  * modify it under the terms of the GNU Lesser General Public
00009  * License as published by the Free Software Foundation; either
00010  * version 2.1 of the License, or (at your option) any later version.
00011  *
00012  * FFmpeg is distributed in the hope that it will be useful,
00013  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00014  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00015  * Lesser General Public License for more details.
00016  *
00017  * You should have received a copy of the GNU Lesser General Public
00018  * License along with FFmpeg; if not, write to the Free Software
00019  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00020  */
00021 
00022 #ifndef AVCODEC_X86_DSPUTIL_MMX_H
00023 #define AVCODEC_X86_DSPUTIL_MMX_H
00024 
00025 #include <stdint.h>
00026 #include "libavcodec/dsputil.h"
00027 #include "libavutil/x86/asm.h"
00028 
00029 typedef struct { uint64_t a, b; } xmm_reg;
00030 
00031 extern const uint64_t ff_bone;
00032 extern const uint64_t ff_wtwo;
00033 
00034 extern const uint64_t ff_pdw_80000000[2];
00035 
00036 extern const xmm_reg  ff_pw_3;
00037 extern const xmm_reg  ff_pw_4;
00038 extern const xmm_reg  ff_pw_5;
00039 extern const xmm_reg  ff_pw_8;
00040 extern const uint64_t ff_pw_15;
00041 extern const xmm_reg  ff_pw_16;
00042 extern const xmm_reg  ff_pw_18;
00043 extern const uint64_t ff_pw_20;
00044 extern const xmm_reg  ff_pw_27;
00045 extern const xmm_reg  ff_pw_28;
00046 extern const xmm_reg  ff_pw_32;
00047 extern const uint64_t ff_pw_42;
00048 extern const uint64_t ff_pw_53;
00049 extern const xmm_reg  ff_pw_63;
00050 extern const xmm_reg  ff_pw_64;
00051 extern const uint64_t ff_pw_96;
00052 extern const uint64_t ff_pw_128;
00053 extern const uint64_t ff_pw_255;
00054 
00055 extern const xmm_reg  ff_pb_1;
00056 extern const xmm_reg  ff_pb_3;
00057 extern const uint64_t ff_pb_7;
00058 extern const uint64_t ff_pb_1F;
00059 extern const uint64_t ff_pb_3F;
00060 extern const uint64_t ff_pb_81;
00061 extern const xmm_reg  ff_pb_A1;
00062 extern const xmm_reg  ff_pb_F8;
00063 extern const uint64_t ff_pb_FC;
00064 extern const xmm_reg  ff_pb_FE;
00065 
00066 extern const double ff_pd_1[2];
00067 extern const double ff_pd_2[2];
00068 
00069 #define SBUTTERFLY(a,b,t,n,m)\
00070     "mov" #m " " #a ", " #t "         \n\t" /* abcd */\
00071     "punpckl" #n " " #b ", " #a "     \n\t" /* aebf */\
00072     "punpckh" #n " " #b ", " #t "     \n\t" /* cgdh */\
00073 
00074 #define TRANSPOSE4(a,b,c,d,t)\
00075     SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\
00076     SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\
00077     SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\
00078     SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */
00079 
00080 #define MOVQ_WONE(regd) \
00081     __asm__ volatile ( \
00082     "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
00083     "psrlw $15, %%" #regd ::)
00084 
00085 void ff_dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx);
00086 void ff_dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
00087 
00088 void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00089 void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00090 void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00091 
00092 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00093 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00094 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00095 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00096 
00097 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd);
00098 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd);
00099 
00100 void ff_put_rv40_qpel8_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size);
00101 void ff_put_rv40_qpel16_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size);
00102 void ff_avg_rv40_qpel8_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size);
00103 void ff_avg_rv40_qpel16_mc33_mmx(uint8_t *block, uint8_t *pixels, int line_size);
00104 
00105 void ff_mmx_idct(DCTELEM *block);
00106 void ff_mmxext_idct(DCTELEM *block);
00107 
00108 
00109 void ff_deinterlace_line_mmx(uint8_t *dst,
00110                              const uint8_t *lum_m4, const uint8_t *lum_m3,
00111                              const uint8_t *lum_m2, const uint8_t *lum_m1,
00112                              const uint8_t *lum,
00113                              int size);
00114 
00115 void ff_deinterlace_line_inplace_mmx(const uint8_t *lum_m4,
00116                                      const uint8_t *lum_m3,
00117                                      const uint8_t *lum_m2,
00118                                      const uint8_t *lum_m1,
00119                                      const uint8_t *lum, int size);
00120 
00121 #endif /* AVCODEC_X86_DSPUTIL_MMX_H */
libavcodec/x86/dsputil_mmx.h
