%endmacro%define PALIGNR PALIGNR_MMXPREDICT_FILTER mmxext%define PALIGNR PALIGNR_SSSE3PREDICT_FILTER ssse3;-----------------------------------------------------------------------------; void predict_8x8_v_mmxext( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_v_mmxext, 2,2    movq        mm0, [r1+16]    STORE8x8    mm0, mm0    RET;-----------------------------------------------------------------------------; void predict_8x8_h_mmxext( uint8_t *src, uint8_t edge[33] );-----------------------------------------------------------------------------INIT_MMXcglobal predict_8x8_h_mmxext, 2,2    movu   m3, [r1+7]    mova   m7, m3    punpckhbw m3, m3    punpcklbw m7, m7    pshufw m0, m3, 0xff    pshufw m1, m3, 0xaa    pshufw m2, m3, 0x55    pshufw m3, m3, 0x00    pshufw m4, m7, 0xff    pshufw m5, m7, 0xaa    pshufw m6, m7, 0x55    pshufw m7, m7, 0x00%assign n 0%rep 8    mova [r0+n*FDEC_STRIDE], m %+ n%assign n n+1%endrep    RET;-----------------------------------------------------------------------------; void predict_8x8_dc_mmxext( uint8_t *src, uint8_t *edge );;-----------------------------------------------------------------------------cglobal predict_8x8_dc_mmxext, 2,2    pxor        mm0, mm0    pxor        mm1, mm1    psadbw      mm0, [r1+7]    psadbw      mm1, [r1+16]    paddw       mm0, [pw_8 GLOBAL]    paddw       mm0, mm1    psrlw       mm0, 4    pshufw      mm0, mm0, 0    packuswb    mm0, mm0    STORE8x8    mm0, mm0    RET;-----------------------------------------------------------------------------; void predict_8x8_dc_top_mmxext( uint8_t *src, uint8_t *edge );;-----------------------------------------------------------------------------%macro PRED8x8_DC 2cglobal %1, 2,2    pxor        mm0, mm0    psadbw      mm0, [r1+%2]    paddw       mm0, [pw_4 GLOBAL]    psrlw       mm0, 3    pshufw      mm0, mm0, 0    packuswb    mm0, mm0    STORE8x8    mm0, mm0    RET%endmacroPRED8x8_DC predict_8x8_dc_top_mmxext, 16PRED8x8_DC predict_8x8_dc_left_mmxext, 7%ifndef ARCH_X86_64; sse2 is faster even on amd, so there's no sense in spending exe size on these; functions if we know sse2 is available.;-----------------------------------------------------------------------------; void predict_8x8_ddl_mmxext( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_ddl_mmxext, 2,2    movq        mm5, [r1+16]    movq        mm2, [r1+17]    movq        mm3, [r1+23]    movq        mm4, [r1+25]    movq        mm1, mm5    psllq       mm1, 8    PRED8x8_LOWPASS mm0, mm1, mm2, mm5, mm7    PRED8x8_LOWPASS mm1, mm3, mm4, [r1+24], mm6%assign Y 7%rep 6    movq        [r0+Y*FDEC_STRIDE], mm1    movq        mm2, mm0    psllq       mm1, 8    psrlq       mm2, 56    psllq       mm0, 8    por         mm1, mm2%assign Y (Y-1)%endrep    movq        [r0+Y*FDEC_STRIDE], mm1    psllq       mm1, 8    psrlq       mm0, 56    por         mm1, mm0%assign Y (Y-1)    movq        [r0+Y*FDEC_STRIDE], mm1    RET;-----------------------------------------------------------------------------; void predict_8x8_ddr_mmxext( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_ddr_mmxext, 2,2    movq        mm1, [r1+7]    movq        mm2, [r1+9]    movq        mm3, [r1+15]    movq        mm4, [r1+17]    PRED8x8_LOWPASS mm0, mm1, mm2, [r1+8], mm7    PRED8x8_LOWPASS mm1, mm3, mm4, [r1+16], mm6%assign Y 7%rep 6    movq        [r0+Y*FDEC_STRIDE], mm0    movq        mm2, mm1    psrlq       mm0, 8    psllq       mm2, 56    psrlq       mm1, 8    por         mm0, mm2%assign Y (Y-1)%endrep    movq        [r0+Y*FDEC_STRIDE], mm0    psrlq       mm0, 8    psllq       mm1, 56    por         mm0, mm1%assign Y (Y-1)    movq        [r0+Y*FDEC_STRIDE], mm0    RET;-----------------------------------------------------------------------------; void predict_8x8_hu_mmxext( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------%define PALIGNR PALIGNR_MMXcglobal predict_8x8_hu_mmxext, 2,2    movq    mm1, [r1+7]         ; l0 l1 l2 l3 l4 l5 l6 l7    add      r0, 4*FDEC_STRIDE    pshufw  mm0, mm1, 00011011b ; l6 l7 l4 l5 l2 l3 l0 l1    psllq   mm1, 56             ; l7 .. .. .. .. .. .. ..    movq    mm2, mm0    psllw   mm0, 8    psrlw   mm2, 8    por     mm2, mm0            ; l7 l6 l5 l4 l3 l2 l1 l0    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    pavgb   mm4, mm2    PRED8x8_LOWPASS mm1, mm3, mm5, mm2, mm6    movq    mm5, mm4    punpcklbw mm4, mm1          ; p4 p3 p2 p1    punpckhbw mm5, mm1          ; p8 p7 p6 p5    movq    mm6, mm5    movq    mm7, mm5    movq    mm0, mm5    PALIGNR mm5, mm4, 2, mm1    pshufw  mm1, mm6, 11111001b    PALIGNR mm6, mm4, 4, mm2    pshufw  mm2, mm7, 11111110b    PALIGNR mm7, mm4, 6, mm3    pshufw  mm3, mm0, 11111111b    movq   [r0-4*FDEC_STRIDE], mm4    movq   [r0-3*FDEC_STRIDE], mm5    movq   [r0-2*FDEC_STRIDE], mm6    movq   [r0-1*FDEC_STRIDE], mm7    movq   [r0+0*FDEC_STRIDE], mm0    movq   [r0+1*FDEC_STRIDE], mm1    movq   [r0+2*FDEC_STRIDE], mm2    movq   [r0+3*FDEC_STRIDE], mm3    RET;-----------------------------------------------------------------------------; void predict_8x8_vr_core_mmxext( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------; fills only some pixels:; f01234567; 0........; 1,,,,,,,,; 2 .......; 3 ,,,,,,,; 4  ......; 5  ,,,,,,; 6   .....; 7   ,,,,,cglobal predict_8x8_vr_core_mmxext, 2,2    movq        mm2, [r1+16]    movq        mm3, [r1+15]    movq        mm1, [r1+14]    movq        mm4, mm3    pavgb       mm3, mm2    PRED8x8_LOWPASS mm0, mm1, mm2, mm4, mm7%assign Y 0%rep 3    movq        [r0+ Y   *FDEC_STRIDE], mm3    movq        [r0+(Y+1)*FDEC_STRIDE], mm0    psllq       mm3, 8    psllq       mm0, 8%assign Y (Y+2)%endrep    movq        [r0+ Y   *FDEC_STRIDE], mm3    movq        [r0+(Y+1)*FDEC_STRIDE], mm0    RET;-----------------------------------------------------------------------------; void predict_8x8c_p_core_mmxext( uint8_t *src, int i00, int b, int c );-----------------------------------------------------------------------------cglobal predict_8x8c_p_core_mmxext, 1,2    LOAD_PLANE_ARGS    movq        mm1, mm2    pmullw      mm2, [pw_3210 GLOBAL]    psllw       mm1, 2    paddsw      mm0, mm2        ; mm0 = {i+0*b, i+1*b, i+2*b, i+3*b}    paddsw      mm1, mm0        ; mm1 = {i+4*b, i+5*b, i+6*b, i+7*b}    mov         r1d, 8ALIGN 4.loop:    movq        mm5, mm0    movq        mm6, mm1    psraw       mm5, 5    psraw       mm6, 5    packuswb    mm5, mm6    movq        [r0], mm5    paddsw      mm0, mm4    paddsw      mm1, mm4    add         r0, FDEC_STRIDE    dec         r1d    jg          .loop    REP_RET;-----------------------------------------------------------------------------; void predict_16x16_p_core_mmxext( uint8_t *src, int i00, int b, int c );-----------------------------------------------------------------------------cglobal predict_16x16_p_core_mmxext, 1,2    LOAD_PLANE_ARGS    movq        mm5, mm2    movq        mm1, mm2    pmullw      mm5, [pw_3210 GLOBAL]    psllw       mm2, 3    psllw       mm1, 2    movq        mm3, mm2    paddsw      mm0, mm5        ; mm0 = {i+ 0*b, i+ 1*b, i+ 2*b, i+ 3*b}    paddsw      mm1, mm0        ; mm1 = {i+ 4*b, i+ 5*b, i+ 6*b, i+ 7*b}    paddsw      mm2, mm0        ; mm2 = {i+ 8*b, i+ 9*b, i+10*b, i+11*b}    paddsw      mm3, mm1        ; mm3 = {i+12*b, i+13*b, i+14*b, i+15*b}    mov         r1d, 16ALIGN 4.loop:    movq        mm5, mm0    movq        mm6, mm1    psraw       mm5, 5    psraw       mm6, 5    packuswb    mm5, mm6    movq        [r0], mm5    movq        mm5, mm2    movq        mm6, mm3    psraw       mm5, 5    psraw       mm6, 5    packuswb    mm5, mm6    movq        [r0+8], mm5    paddsw      mm0, mm4    paddsw      mm1, mm4    paddsw      mm2, mm4    paddsw      mm3, mm4    add         r0, FDEC_STRIDE    dec         r1d    jg          .loop    REP_RET%endif ; !ARCH_X86_64;-----------------------------------------------------------------------------; void predict_8x8_ddl_sse2( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_ddl_sse2, 2,2    movdqa      xmm3, [r1+16]    movdqu      xmm2, [r1+17]    movdqa      xmm1, xmm3    pslldq      xmm1, 1    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm4%assign Y 0%rep 8    psrldq      xmm0, 1    movq        [r0+Y*FDEC_STRIDE], xmm0%assign Y (Y+1)%endrep    RET;-----------------------------------------------------------------------------; void predict_8x8_ddr_sse2( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_ddr_sse2, 2,2    movdqu      xmm3, [r1+8]    movdqu      xmm1, [r1+7]    movdqa      xmm2, xmm3    psrldq      xmm2, 1    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm3, xmm4    movdqa      xmm1, xmm0    psrldq      xmm1, 1%assign Y 7%rep 3    movq        [r0+Y*FDEC_STRIDE], xmm0    movq        [r0+(Y-1)*FDEC_STRIDE], xmm1    psrldq      xmm0, 2    psrldq      xmm1, 2%assign Y (Y-2)%endrep    movq        [r0+1*FDEC_STRIDE], xmm0    movq        [r0+0*FDEC_STRIDE], xmm1    RET;-----------------------------------------------------------------------------; void predict_8x8_vl_sse2( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_vl_sse2, 2,2    movdqa      xmm4, [r1+16]    movdqa      xmm2, xmm4    movdqa      xmm1, xmm4    movdqa      xmm3, xmm4    psrldq      xmm2, 1    pslldq      xmm1, 1    pavgb       xmm3, xmm2    PRED8x8_LOWPASS_XMM xmm0, xmm1, xmm2, xmm4, xmm5; xmm0: (t0 + 2*t1 + t2 + 2) >> 2; xmm3: (t0 + t1 + 1) >> 1%assign Y 0%rep 3    psrldq      xmm0, 1    movq        [r0+ Y   *FDEC_STRIDE], xmm3    movq        [r0+(Y+1)*FDEC_STRIDE], xmm0    psrldq      xmm3, 1%assign Y (Y+2)%endrep    psrldq      xmm0, 1    movq        [r0+ Y   *FDEC_STRIDE], xmm3    movq        [r0+(Y+1)*FDEC_STRIDE], xmm0    RET;-----------------------------------------------------------------------------; void predict_8x8_vr_sse2( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------cglobal predict_8x8_vr_sse2, 2,2,7    movdqu      xmm0, [r1+8]    movdqa      xmm6, [pw_ff00 GLOBAL]    add         r0, 4*FDEC_STRIDE    movdqa      xmm1, xmm0    movdqa      xmm2, xmm0    movdqa      xmm3, xmm0    pslldq      xmm0, 1    pslldq      xmm1, 2    pavgb       xmm2, xmm0    PRED8x8_LOWPASS_XMM xmm4, xmm3, xmm1, xmm0, xmm5    pandn       xmm6, xmm4    movdqa      xmm5, xmm4    psrlw       xmm4, 8    packuswb    xmm6, xmm4    movhlps     xmm4, xmm6    movhps [r0-3*FDEC_STRIDE], xmm5    movhps [r0-4*FDEC_STRIDE], xmm2    psrldq      xmm5, 4    movss       xmm5, xmm6    psrldq      xmm2, 4    movss       xmm2, xmm4%assign Y 3%rep 3    psrldq      xmm5, 1    psrldq      xmm2, 1    movq        [r0+Y*FDEC_STRIDE], xmm5    movq        [r0+(Y-1)*FDEC_STRIDE], xmm2%assign Y (Y-2)%endrep    RET;-----------------------------------------------------------------------------; void predict_8x8_hd_mmxext( uint8_t *src, uint8_t *edge );-----------------------------------------------------------------------------%define PALIGNR PALIGNR_MMXcglobal predict_8x8_hd_mmxext, 2,2    add     r0, 4*FDEC_STRIDE    movq    mm0, [r1]           ; l7 .. .. .. .. .. .. ..    movq    mm1, [r1+8]         ; lt l0 l1 l2 l3 l4 l5 l6    movq    mm2, [r1+16]        ; t7 t6 t5 t4 t3 t2 t1 t0    movq    mm3, mm1            ; lt l0 l1 l2 l3 l4 l5 l6    movq    mm4, mm2            ; t7 t6 t5 t4 t3 t2 t1 t0    PALIGNR mm2, mm1, 7, mm5    ; t6 t5 t4 t3 t2 t1 t0 lt    PALIGNR mm1, mm0, 7, mm6    ; l0 l1 l2 l3 l4 l5 l6 l7    PALIGNR mm4, mm3, 1, mm7    ; t0 lt l0 l1 l2 l3 l4 l5    movq    mm5, mm3    pavgb   mm3, mm1    PRED8x8_LOWPASS mm0, mm4, mm1, mm5, mm7    movq    mm4, mm2    movq    mm1, mm2            ; t6 t5 t4 t3 t2 t1 t0 lt    psrlq   mm4, 16             ; .. .. t6 t5 t4 t3 t2 t1    psrlq   mm1, 8              ; .. t6 t5 t4 t3 t2 t1 t0    PRED8x8_LOWPASS mm6, mm4, mm2, mm1, mm5