; .. p11 p10 p9    movq    mm7, mm3    punpcklbw mm3, mm0          ; p4 p3 p2 p1    punpckhbw mm7, mm0          ; p8 p7 p6 p5    movq    mm1, mm7    movq    mm0, mm7    movq    mm4, mm7    movq   [r0+3*FDEC_STRIDE], mm3    PALIGNR mm7, mm3, 2, mm5    movq   [r0+2*FDEC_STRIDE], mm7    PALIGNR mm1, mm3, 4, mm5    movq   [r0+1*FDEC_STRIDE], mm1    PALIGNR mm0, mm3, 6, mm3    movq    [r0+0*FDEC_STRIDE], mm0    movq    mm2, mm6    movq    mm3, mm6    movq   [r0-1*FDEC_STRIDE], mm4    PALIGNR mm6, mm4, 2, mm5    movq   [r0-2*FDEC_STRIDE], mm6    PALIGNR mm2, mm4, 4, mm5    movq   [r0-3*FDEC_STRIDE], mm2    PALIGNR mm3, mm4, 6, mm4    movq   [r0-4*FDEC_STRIDE], mm3    RET;-----------------------------------------------------------------------------; void predict_8x8_hd_ssse3( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------%macro PREDICT_8x8_HD 1cglobal predict_8x8_hd_%1, 2,2    add       r0, 4*FDEC_STRIDE    movdqa  xmm0, [r1]    movdqa  xmm1, [r1+16]    movdqa  xmm2, xmm1    movdqa  xmm3, xmm1    PALIGNR xmm1, xmm0, 7, xmm4    PALIGNR xmm2, xmm0, 9, xmm5    PALIGNR xmm3, xmm0, 8, xmm0    movdqa  xmm4, xmm1    pavgb   xmm4, xmm3    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm5    punpcklbw xmm4, xmm0    movhlps xmm0, xmm4%assign Y 3%rep 3    movq   [r0+(Y)*FDEC_STRIDE], xmm4    movq   [r0+(Y-4)*FDEC_STRIDE], xmm0    psrldq xmm4, 2    psrldq xmm0, 2%assign Y (Y-1)%endrep    movq   [r0+(Y)*FDEC_STRIDE], xmm4    movq   [r0+(Y-4)*FDEC_STRIDE], xmm0    RET%endmacroINIT_XMMPREDICT_8x8_HD sse2%define PALIGNR PALIGNR_SSSE3PREDICT_8x8_HD ssse3INIT_MMX%define PALIGNR PALIGNR_MMX;-----------------------------------------------------------------------------; void predict_8x8_hu_sse2( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------%macro PREDICT_8x8_HU 1cglobal predict_8x8_hu_%1, 2,2    add        r0, 4*FDEC_STRIDE%ifidn %1, ssse3    movq      mm5, [r1+7]    movq      mm6, [pb_reverse GLOBAL]    movq      mm1, mm5    movq      mm2, mm5    movq      mm3, mm5    pshufb    mm5, mm6    psrlq     mm6, 8    pshufb    mm2, mm6    psrlq     mm6, 8    pshufb    mm3, mm6    movq      mm4, mm5%else    movq      mm1, [r1+7]           ; l0 l1 l2 l3 l4 l5 l6 l7    pshufw    mm0, mm1, 00011011b   ; l6 l7 l4 l5 l2 l3 l0 l1    movq      mm2, mm0    psllw     mm0, 8    psrlw     mm2, 8    por       mm2, mm0              ; l7 l6 l5 l4 l3 l2 l1 l0    psllq     mm1, 56               ; l7 .. .. .. .. .. .. ..    movq      mm3, mm2    movq      mm4, mm2    movq      mm5, mm2    psrlq     mm2, 8    psrlq     mm3, 16    por       mm2, mm1              ; l7 l7 l6 l5 l4 l3 l2 l1    punpckhbw mm1, mm1    por       mm3, mm1              ; l7 l7 l7 l6 l5 l4 l3 l2%endif    pavgb     mm4, mm2    PRED8x8_LOWPASS mm1, mm3, mm5, mm2, mm6    movq2dq   xmm0, mm4    movq2dq   xmm1, mm1    punpcklbw xmm0, xmm1    punpckhbw  mm4, mm1%assign Y -4%rep 3    movq     [r0+Y*FDEC_STRIDE], xmm0    psrldq    xmm0, 2%assign Y (Y+1)%endrep    pshufw     mm5, mm4, 11111001b    pshufw     mm6, mm4, 11111110b    pshufw     mm7, mm4, 11111111b    movq     [r0+Y*FDEC_STRIDE], xmm0    movq     [r0+0*FDEC_STRIDE], mm4    movq     [r0+1*FDEC_STRIDE], mm5    movq     [r0+2*FDEC_STRIDE], mm6    movq     [r0+3*FDEC_STRIDE], mm7    RET%endmacroPREDICT_8x8_HU sse2PREDICT_8x8_HU ssse3;-----------------------------------------------------------------------------; void predict_8x8c_v_mmx( uint8_t *src );-----------------------------------------------------------------------------cglobal predict_8x8c_v_mmx, 1,1    movq        mm0, [r0 - FDEC_STRIDE]    STORE8x8    mm0, mm0    RET;-----------------------------------------------------------------------------; void predict_8x8c_h_mmxext( uint8_t *src );-----------------------------------------------------------------------------%macro PRED_8x8C_H 1cglobal predict_8x8c_h_%1, 1,1%ifidn %1, ssse3    mova   m1, [pb_3 GLOBAL]%endif%assign n 0%rep 8    SPLATB m0, r0+FDEC_STRIDE*n-1, m1    mova [r0+FDEC_STRIDE*n], m0%assign n n+1%endrep    RET%endmacroINIT_MMX%define SPLATB SPLATB_MMXPRED_8x8C_H mmxext%define SPLATB SPLATB_SSSE3PRED_8x8C_H ssse3;-----------------------------------------------------------------------------; void predict_8x8c_dc_core_mmxext( uint8_t *src, int s2, int s3 );-----------------------------------------------------------------------------cglobal predict_8x8c_dc_core_mmxext, 1,1    movq        mm0, [r0 - FDEC_STRIDE]    pxor        mm1, mm1    pxor        mm2, mm2    punpckhbw   mm1, mm0    punpcklbw   mm0, mm2    psadbw      mm1, mm2        ; s1    psadbw      mm0, mm2        ; s0%ifdef ARCH_X86_64    movd        mm4, r1d    movd        mm5, r2d    paddw       mm0, mm4    pshufw      mm2, mm5, 0%else    paddw       mm0, r1m    pshufw      mm2, r2m, 0%endif    psrlw       mm0, 3    paddw       mm1, [pw_2 GLOBAL]    movq        mm3, mm2    pshufw      mm1, mm1, 0    pshufw      mm0, mm0, 0     ; dc0 (w)    paddw       mm3, mm1    psrlw       mm3, 3          ; dc3 (w)    psrlw       mm2, 2          ; dc2 (w)    psrlw       mm1, 2          ; dc1 (w)    packuswb    mm0, mm1        ; dc0,dc1 (b)    packuswb    mm2, mm3        ; dc2,dc3 (b)    STORE8x8    mm0, mm2    RETcglobal predict_8x8c_dc_top_mmxext, 1,1    movq        mm0, [r0 - FDEC_STRIDE]    pxor        mm1, mm1    pxor        mm2, mm2    punpckhbw   mm1, mm0    punpcklbw   mm0, mm2    psadbw      mm1, mm2        ; s1    psadbw      mm0, mm2        ; s0    psrlw       mm1, 1    psrlw       mm0, 1    pavgw       mm1, mm2    pavgw       mm0, mm2    pshufw      mm1, mm1, 0    pshufw      mm0, mm0, 0     ; dc0 (w)    packuswb    mm0, mm1        ; dc0,dc1 (b)    STORE8x8    mm0, mm0    RET;-----------------------------------------------------------------------------; void predict_8x8c_p_core_sse2( uint8_t *src, int i00, int b, int c );-----------------------------------------------------------------------------cglobal predict_8x8c_p_core_sse2, 1,1    movd        xmm0, r1m    movd        xmm2, r2m    movd        xmm4, r3m    pshuflw     xmm0, xmm0, 0    pshuflw     xmm2, xmm2, 0    pshuflw     xmm4, xmm4, 0    punpcklqdq  xmm0, xmm0    punpcklqdq  xmm2, xmm2    punpcklqdq  xmm4, xmm4    pmullw      xmm2, [pw_76543210 GLOBAL]    paddsw      xmm0, xmm2        ; xmm0 = {i+0*b, i+1*b, i+2*b, i+3*b, i+4*b, i+5*b, i+6*b, i+7*b}    movdqa      xmm3, xmm0    paddsw      xmm3, xmm4    paddsw      xmm4, xmm4call .loop    add           r0, FDEC_STRIDE*4.loop:    movdqa      xmm5, xmm0    movdqa      xmm1, xmm3    psraw       xmm0, 5    psraw       xmm3, 5    packuswb    xmm0, xmm3    movq        [r0+FDEC_STRIDE*0], xmm0    movhps      [r0+FDEC_STRIDE*1], xmm0    paddsw      xmm5, xmm4    paddsw      xmm1, xmm4    movdqa      xmm0, xmm5    movdqa      xmm3, xmm1    psraw       xmm5, 5    psraw       xmm1, 5    packuswb    xmm5, xmm1    movq        [r0+FDEC_STRIDE*2], xmm5    movhps      [r0+FDEC_STRIDE*3], xmm5    paddsw      xmm0, xmm4    paddsw      xmm3, xmm4    RET;-----------------------------------------------------------------------------; void predict_16x16_p_core_sse2( uint8_t *src, int i00, int b, int c );-----------------------------------------------------------------------------cglobal predict_16x16_p_core_sse2, 1,2,8    movd        xmm0, r1m    movd        xmm1, r2m    movd        xmm2, r3m    pshuflw     xmm0, xmm0, 0    pshuflw     xmm1, xmm1, 0    pshuflw     xmm2, xmm2, 0    punpcklqdq  xmm0, xmm0    punpcklqdq  xmm1, xmm1    punpcklqdq  xmm2, xmm2    movdqa      xmm3, xmm1    pmullw      xmm3, [pw_76543210 GLOBAL]    psllw       xmm1, 3    paddsw      xmm0, xmm3  ; xmm0 = {i+ 0*b, i+ 1*b, i+ 2*b, i+ 3*b, i+ 4*b, i+ 5*b, i+ 6*b, i+ 7*b}    paddsw      xmm1, xmm0  ; xmm1 = {i+ 8*b, i+ 9*b, i+10*b, i+11*b, i+12*b, i+13*b, i+14*b, i+15*b}    movdqa      xmm7, xmm2    paddsw      xmm7, xmm7    mov         r1d, 8ALIGN 4.loop:    movdqa      xmm3, xmm0    movdqa      xmm4, xmm1    movdqa      xmm5, xmm0    movdqa      xmm6, xmm1    psraw       xmm3, 5    psraw       xmm4, 5    paddsw      xmm5, xmm2    paddsw      xmm6, xmm2    psraw       xmm5, 5    psraw       xmm6, 5    packuswb    xmm3, xmm4    packuswb    xmm5, xmm6    movdqa      [r0+FDEC_STRIDE*0], xmm3    movdqa      [r0+FDEC_STRIDE*1], xmm5    paddsw      xmm0, xmm7    paddsw      xmm1, xmm7    add         r0, FDEC_STRIDE*2    dec         r1d    jg          .loop    REP_RET;-----------------------------------------------------------------------------; void predict_16x16_v_mmx( uint8_t *src );-----------------------------------------------------------------------------cglobal predict_16x16_v_mmx, 1,2    movq        mm0, [r0 - FDEC_STRIDE]    movq        mm1, [r0 - FDEC_STRIDE + 8]    STORE16x16  mm0, mm1    REP_RET;-----------------------------------------------------------------------------; void predict_16x16_v_sse2( uint8_t *src );-----------------------------------------------------------------------------cglobal predict_16x16_v_sse2, 1,1    movdqa      xmm0, [r0 - FDEC_STRIDE]    STORE16x16_SSE2 xmm0    RET;-----------------------------------------------------------------------------; void predict_16x16_h_mmxext( uint8_t *src );-----------------------------------------------------------------------------%macro PRED_16x16_H 1cglobal predict_16x16_h_%1, 1,2    mov r1, FDEC_STRIDE*12%ifidn %1, ssse3    mova   m1, [pb_3 GLOBAL]%endif.vloop:%assign n 0%rep 4    SPLATB m0, r0+r1+FDEC_STRIDE*n-1, m1    mova [r0+r1+FDEC_STRIDE*n], m0%if mmsize==8    mova [r0+r1+FDEC_STRIDE*n+8], m0%endif%assign n n+1%endrep    add r1, -FDEC_STRIDE*4    jge .vloop    REP_RET%endmacro;no SSE2, its slower than MMX on all systems that don't support SSSE3INIT_MMX%define SPLATB SPLATB_MMXPRED_16x16_H mmxextINIT_XMM%define SPLATB SPLATB_SSSE3PRED_16x16_H ssse3;-----------------------------------------------------------------------------; void predict_16x16_dc_core_mmxext( uint8_t *src, int i_dc_left );-----------------------------------------------------------------------------%macro PRED16x16_DC 2    pxor        mm0, mm0    pxor        mm1, mm1    psadbw      mm0, [r0 - FDEC_STRIDE]    psadbw      mm1, [r0 - FDEC_STRIDE + 8]    paddusw     mm0, mm1    paddusw     mm0, %1    psrlw       mm0, %2                       ; dc    pshufw      mm0, mm0, 0    packuswb    mm0, mm0                      ; dc in bytes    STORE16x16  mm0, mm0%endmacrocglobal predict_16x16_dc_core_mmxext, 1,2%ifdef ARCH_X86_64    movd         mm2, r1d    PRED16x16_DC mm2, 5%else    PRED16x16_DC r1m, 5%endif    REP_RETcglobal predict_16x16_dc_top_mmxext, 1,2    PRED16x16_DC [pw_8 GLOBAL], 4    REP_RETcglobal predict_16x16_dc_left_core_mmxext, 1,1    movd       mm0, r1m    pshufw     mm0, mm0, 0    packuswb   mm0, mm0    STORE16x16 mm0, mm0    REP_RET;-----------------------------------------------------------------------------; void predict_16x16_dc_core_sse2( uint8_t *src, int i_dc_left );-----------------------------------------------------------------------------%macro PRED16x16_DC_SSE2 2    pxor        xmm0, xmm0    psadbw      xmm0, [r0 - FDEC_STRIDE]    movhlps     xmm1, xmm0    paddw       xmm0, xmm1    paddusw     xmm0, %1    psrlw       xmm0, %2                ; dc    pshuflw     xmm0, xmm0, 0    punpcklqdq  xmm0, xmm0    packuswb    xmm0, xmm0              ; dc in bytes    STORE16x16_SSE2 xmm0%endmacrocglobal predict_16x16_dc_core_sse2, 1,1    movd xmm2, r1m    PRED16x16_DC_SSE2 xmm2, 5    RETcglobal predict_16x16_dc_top_sse2, 1,1    PRED16x16_DC_SSE2 [pw_8 GLOBAL], 4    RETcglobal predict_16x16_dc_left_core_sse2, 1,1    movd       xmm0, r1m    pshuflw    xmm0, xmm0, 0    punpcklqdq xmm0, xmm0    packuswb   xmm0, xmm0    STORE16x16_SSE2 xmm0    RET