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_cpu.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 LOAD4(stride,in,a,b,c,d)\
00070     "movq 0*"#stride"+"#in", "#a"\n\t"\
00071     "movq 1*"#stride"+"#in", "#b"\n\t"\
00072     "movq 2*"#stride"+"#in", "#c"\n\t"\
00073     "movq 3*"#stride"+"#in", "#d"\n\t"
00074 
00075 #define STORE4(stride,out,a,b,c,d)\
00076     "movq "#a", 0*"#stride"+"#out"\n\t"\
00077     "movq "#b", 1*"#stride"+"#out"\n\t"\
00078     "movq "#c", 2*"#stride"+"#out"\n\t"\
00079     "movq "#d", 3*"#stride"+"#out"\n\t"
00080 
00081 /* in/out: mma=mma+mmb, mmb=mmb-mma */
00082 #define SUMSUB_BA( a, b ) \
00083     "paddw "#b", "#a" \n\t"\
00084     "paddw "#b", "#b" \n\t"\
00085     "psubw "#a", "#b" \n\t"
00086 
00087 #define SBUTTERFLY(a,b,t,n,m)\
00088     "mov" #m " " #a ", " #t "         \n\t" /* abcd */\
00089     "punpckl" #n " " #b ", " #a "     \n\t" /* aebf */\
00090     "punpckh" #n " " #b ", " #t "     \n\t" /* cgdh */\
00091 
00092 #define TRANSPOSE4(a,b,c,d,t)\
00093     SBUTTERFLY(a,b,t,wd,q) /* a=aebf t=cgdh */\
00094     SBUTTERFLY(c,d,b,wd,q) /* c=imjn b=kolp */\
00095     SBUTTERFLY(a,c,d,dq,q) /* a=aeim d=bfjn */\
00096     SBUTTERFLY(t,b,c,dq,q) /* t=cgko c=dhlp */
00097 
00098 static inline void transpose4x4(uint8_t *dst, uint8_t *src, x86_reg dst_stride, x86_reg src_stride){
00099     __asm__ volatile( //FIXME could save 1 instruction if done as 8x4 ...
00100         "movd  (%1), %%mm0              \n\t"
00101         "add   %3, %1                   \n\t"
00102         "movd  (%1), %%mm1              \n\t"
00103         "movd  (%1,%3,1), %%mm2         \n\t"
00104         "movd  (%1,%3,2), %%mm3         \n\t"
00105         "punpcklbw %%mm1, %%mm0         \n\t"
00106         "punpcklbw %%mm3, %%mm2         \n\t"
00107         "movq %%mm0, %%mm1              \n\t"
00108         "punpcklwd %%mm2, %%mm0         \n\t"
00109         "punpckhwd %%mm2, %%mm1         \n\t"
00110         "movd  %%mm0, (%0)              \n\t"
00111         "add   %2, %0                   \n\t"
00112         "punpckhdq %%mm0, %%mm0         \n\t"
00113         "movd  %%mm0, (%0)              \n\t"
00114         "movd  %%mm1, (%0,%2,1)         \n\t"
00115         "punpckhdq %%mm1, %%mm1         \n\t"
00116         "movd  %%mm1, (%0,%2,2)         \n\t"
00117 
00118         :  "+&r" (dst),
00119            "+&r" (src)
00120         :  "r" (dst_stride),
00121            "r" (src_stride)
00122         :  "memory"
00123     );
00124 }
00125 
00126 // e,f,g,h can be memory
00127 // out: a,d,t,c
00128 #define TRANSPOSE8x4(a,b,c,d,e,f,g,h,t)\
00129     "punpcklbw " #e ", " #a " \n\t" /* a0 e0 a1 e1 a2 e2 a3 e3 */\
00130     "punpcklbw " #f ", " #b " \n\t" /* b0 f0 b1 f1 b2 f2 b3 f3 */\
00131     "punpcklbw " #g ", " #c " \n\t" /* c0 g0 c1 g1 c2 g2 d3 g3 */\
00132     "punpcklbw " #h ", " #d " \n\t" /* d0 h0 d1 h1 d2 h2 d3 h3 */\
00133     SBUTTERFLY(a, b, t, bw, q)   /* a= a0 b0 e0 f0 a1 b1 e1 f1 */\
00134                                  /* t= a2 b2 e2 f2 a3 b3 e3 f3 */\
00135     SBUTTERFLY(c, d, b, bw, q)   /* c= c0 d0 g0 h0 c1 d1 g1 h1 */\
00136                                  /* b= c2 d2 g2 h2 c3 d3 g3 h3 */\
00137     SBUTTERFLY(a, c, d, wd, q)   /* a= a0 b0 c0 d0 e0 f0 g0 h0 */\
00138                                  /* d= a1 b1 c1 d1 e1 f1 g1 h1 */\
00139     SBUTTERFLY(t, b, c, wd, q)   /* t= a2 b2 c2 d2 e2 f2 g2 h2 */\
00140                                  /* c= a3 b3 c3 d3 e3 f3 g3 h3 */
00141 
00142 #if ARCH_X86_64
00143 // permutes 01234567 -> 05736421
00144 #define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\
00145     SBUTTERFLY(a,b,%%xmm8,wd,dqa)\
00146     SBUTTERFLY(c,d,b,wd,dqa)\
00147     SBUTTERFLY(e,f,d,wd,dqa)\
00148     SBUTTERFLY(g,h,f,wd,dqa)\
00149     SBUTTERFLY(a,c,h,dq,dqa)\
00150     SBUTTERFLY(%%xmm8,b,c,dq,dqa)\
00151     SBUTTERFLY(e,g,b,dq,dqa)\
00152     SBUTTERFLY(d,f,g,dq,dqa)\
00153     SBUTTERFLY(a,e,f,qdq,dqa)\
00154     SBUTTERFLY(%%xmm8,d,e,qdq,dqa)\
00155     SBUTTERFLY(h,b,d,qdq,dqa)\
00156     SBUTTERFLY(c,g,b,qdq,dqa)\
00157     "movdqa %%xmm8, "#g"              \n\t"
00158 #else
00159 #define TRANSPOSE8(a,b,c,d,e,f,g,h,t)\
00160     "movdqa "#h", "#t"                \n\t"\
00161     SBUTTERFLY(a,b,h,wd,dqa)\
00162     "movdqa "#h", 16"#t"              \n\t"\
00163     "movdqa "#t", "#h"                \n\t"\
00164     SBUTTERFLY(c,d,b,wd,dqa)\
00165     SBUTTERFLY(e,f,d,wd,dqa)\
00166     SBUTTERFLY(g,h,f,wd,dqa)\
00167     SBUTTERFLY(a,c,h,dq,dqa)\
00168     "movdqa "#h", "#t"                \n\t"\
00169     "movdqa 16"#t", "#h"              \n\t"\
00170     SBUTTERFLY(h,b,c,dq,dqa)\
00171     SBUTTERFLY(e,g,b,dq,dqa)\
00172     SBUTTERFLY(d,f,g,dq,dqa)\
00173     SBUTTERFLY(a,e,f,qdq,dqa)\
00174     SBUTTERFLY(h,d,e,qdq,dqa)\
00175     "movdqa "#h", 16"#t"              \n\t"\
00176     "movdqa "#t", "#h"                \n\t"\
00177     SBUTTERFLY(h,b,d,qdq,dqa)\
00178     SBUTTERFLY(c,g,b,qdq,dqa)\
00179     "movdqa 16"#t", "#g"              \n\t"
00180 #endif
00181 
00182 #define MOVQ_WONE(regd) \
00183     __asm__ volatile ( \
00184     "pcmpeqd %%" #regd ", %%" #regd " \n\t" \
00185     "psrlw $15, %%" #regd ::)
00186 
00187 void dsputilenc_init_mmx(DSPContext* c, AVCodecContext *avctx);
00188 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
00189 
00190 void ff_add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00191 void ff_put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00192 void ff_put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
00193 
00194 void ff_put_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00195 void ff_avg_cavs_qpel8_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00196 void ff_put_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00197 void ff_avg_cavs_qpel16_mc00_mmx2(uint8_t *dst, uint8_t *src, int stride);
00198 
00199 void ff_put_vc1_mspel_mc00_mmx(uint8_t *dst, const uint8_t *src, int stride, int rnd);
00200 void ff_avg_vc1_mspel_mc00_mmx2(uint8_t *dst, const uint8_t *src, int stride, int rnd);
00201 
00202 void ff_mmx_idct(DCTELEM *block);
00203 void ff_mmxext_idct(DCTELEM *block);
00204 
00205 
00206 void ff_deinterlace_line_mmx(uint8_t *dst,
00207                              const uint8_t *lum_m4, const uint8_t *lum_m3,
00208                              const uint8_t *lum_m2, const uint8_t *lum_m1,
00209                              const uint8_t *lum,
00210                              int size);
00211 
00212 void ff_deinterlace_line_inplace_mmx(const uint8_t *lum_m4,
00213                                      const uint8_t *lum_m3,
00214                                      const uint8_t *lum_m2,
00215                                      const uint8_t *lum_m1,
00216                                      const uint8_t *lum, int size);
00217 
00218 #endif /* AVCODEC_X86_DSPUTIL_MMX_H */
libavcodec/x86/dsputil_mmx.h
