movq mm6, [Up13]  movq mm7, [Up31]  COL03_SSE mm0, mm1, 0  MUL_PACK mm0,mm1, mm6, mm7  movq mm4, mm0  movq mm5, mm1  STORE_1 mm4, mm5  add ecx, eax  COL03_SSE mm2, mm3, 1  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  COL03_SSE mm0, mm1, 2  MUL_PACK mm0,mm1, mm6, mm7  MIX_ROWS mm2, mm3, mm0, mm1  STORE_2 mm2, mm3  COL03_SSE mm2, mm3, 3  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  COL03_SSE mm0, mm1, 4  MUL_PACK mm0,mm1, mm6, mm7  MIX_ROWS mm2, mm3, mm0, mm1  STORE_2 mm2, mm3  COL03_SSE mm2, mm3, 5  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  COL03_SSE mm0, mm1, 6  MUL_PACK mm0,mm1, mm6, mm7  MIX_ROWS mm2, mm3, mm0, mm1  STORE_2 mm2, mm3  COL03_SSE mm2, mm3, 7  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  STORE_1 mm2, mm3  mov ecx, [esp+4]  add ecx, 8  COL47_SSE mm0, mm1, 0  MUL_PACK mm0,mm1, mm6, mm7  movq mm4, mm0  movq mm5, mm1  STORE_1 mm4, mm5  add ecx, eax  COL47_SSE mm2, mm3, 1  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  COL47_SSE mm0, mm1, 2  MUL_PACK mm0,mm1, mm6, mm7  MIX_ROWS mm2, mm3, mm0, mm1  STORE_2 mm2, mm3  COL47_SSE mm2, mm3, 3  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  COL47_SSE mm0, mm1, 4  MUL_PACK mm0,mm1, mm6, mm7  MIX_ROWS mm2, mm3, mm0, mm1  STORE_2 mm2, mm3  COL47_SSE mm2, mm3, 5  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  COL47_SSE mm0, mm1, 6  MUL_PACK mm0,mm1, mm6, mm7  MIX_ROWS mm2, mm3, mm0, mm1  STORE_2 mm2, mm3  COL47_SSE mm2, mm3, 7  MUL_PACK mm2,mm3, mm6, mm7  MIX_ROWS mm0, mm1, mm2, mm3  STORE_2 mm0, mm1  STORE_1 mm2, mm3  ret;===========================================================================;; void xvid_Add_Upsampled_8x8_16To8_xmm(uint8_t *Dst,;                                       const int16_t *Src, const int BpS);;;===========================================================================align 16xvid_Add_Upsampled_8x8_16To8_xmm:  ; 549c  mov ecx, [esp+4]  ; Dst  mov edx, [esp+8]  ; Src  mov eax, [esp+12] ; BpS  COL03_SSE mm0, mm1, 0  MUL_PACK mm0,mm1, [Up13], [Up31]  movq mm4, mm0  movq mm5, mm1  STORE_ADD_1 mm4, mm5  add ecx, eax  COL03_SSE mm2, mm3, 1  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  COL03_SSE mm0, mm1, 2  MUL_PACK mm0,mm1, [Up13], [Up31]  MIX_ROWS mm2, mm3, mm0, mm1  STORE_ADD_2 mm2, mm3  COL03_SSE mm2, mm3, 3  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  COL03_SSE mm0, mm1, 4  MUL_PACK mm0,mm1, [Up13], [Up31]  MIX_ROWS mm2, mm3, mm0, mm1  STORE_ADD_2 mm2, mm3  COL03_SSE mm2, mm3, 5  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  COL03_SSE mm0, mm1, 6  MUL_PACK mm0,mm1, [Up13], [Up31]  MIX_ROWS mm2, mm3, mm0, mm1  STORE_ADD_2 mm2, mm3  COL03_SSE mm2, mm3, 7  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  STORE_ADD_1 mm2, mm3  mov ecx, [esp+4]  add ecx, 8  COL47_SSE mm0, mm1, 0  MUL_PACK mm0,mm1, [Up13], [Up31]  movq mm4, mm0  movq mm5, mm1  STORE_ADD_1 mm4, mm5  add ecx, eax  COL47_SSE mm2, mm3, 1  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  COL47_SSE mm0, mm1, 2  MUL_PACK mm0,mm1, [Up13], [Up31]  MIX_ROWS mm2, mm3, mm0, mm1  STORE_ADD_2 mm2, mm3  COL47_SSE mm2, mm3, 3  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  COL47_SSE mm0, mm1, 4  MUL_PACK mm0,mm1, [Up13], [Up31]  MIX_ROWS mm2, mm3, mm0, mm1  STORE_ADD_2 mm2, mm3  COL47_SSE mm2, mm3, 5  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  COL47_SSE mm0, mm1, 6  MUL_PACK mm0,mm1, [Up13], [Up31]  MIX_ROWS mm2, mm3, mm0, mm1  STORE_ADD_2 mm2, mm3  COL47_SSE mm2, mm3, 7  MUL_PACK mm2,mm3, [Up13], [Up31]  MIX_ROWS mm0, mm1, mm2, mm3  STORE_ADD_2 mm0, mm1  STORE_ADD_1 mm2, mm3  ret;===========================================================================;; void xvid_HFilter_31_mmx(uint8_t *Src1, uint8_t *Src2, int Nb_Blks);; void xvid_VFilter_31_x86(uint8_t *Src1, uint8_t *Src2, const int BpS, int Nb_Blks);; void xvid_HFilter_31_x86(uint8_t *Src1, uint8_t *Src2, int Nb_Blks);;;===========================================================================;//////////////////////////////////////////////////////////////////////;// horizontal/vertical filtering: [x,y] -> [ (3x+y+2)>>2, (x+3y+2)>>2 ];//;// We use the trick: tmp = (x+y+2) -> [x = (tmp+2x)>>2, y = (tmp+2y)>>2];//////////////////////////////////////////////////////////////////////align 16xvid_HFilter_31_mmx:  push esi  push edi  mov esi, [esp+4  +8]  ; Src1  mov edi, [esp+8  +8]  ; Src2  mov eax, [esp+12 +8] ; Nb_Blks  lea eax,[eax*2]  movq mm5, [Cst2]  pxor mm7, mm7  lea esi, [esi+eax*4]  lea edi, [edi+eax*4]  neg eax.Loop:  ;12c  movd mm0, [esi+eax*4]  movd mm1, [edi+eax*4]  movq mm2, mm5  punpcklbw mm0, mm7  punpcklbw mm1, mm7  paddsw mm2, mm0  paddsw mm0, mm0  paddsw mm2, mm1  paddsw mm1, mm1  paddsw mm0, mm2  paddsw mm1, mm2  psraw mm0, 2  psraw mm1, 2  packuswb mm0, mm7  packuswb mm1, mm7  movd [esi+eax*4], mm0  movd [edi+eax*4], mm1  add eax,1  jl .Loop  pop edi  pop esi  ret  ; mmx is of no use here. Better use plain ASM. Moreover,  ; this is for the fun of ASM coding, coz' every modern compiler can  ; end up with a code that looks very much like this one...align 16xvid_VFilter_31_x86:  push esi  push edi  push ebx  push ebp  mov esi, [esp+4  +16]  ; Src1  mov edi, [esp+8  +16]  ; Src2  mov ebp, [esp+12 +16]  ; BpS  mov eax, [esp+16 +16]  ; Nb_Blks  lea eax,[eax*8].Loop:  ;7c  movzx ecx, byte [esi]  movzx edx, byte [edi]  lea ebx, [ecx+edx+2]  lea ecx,[ebx+2*ecx]  lea edx,[ebx+2*edx]  shr ecx,2  shr edx,2  mov [esi], cl  mov [edi], dl  lea esi, [esi+ebp]  lea edi, [edi+ebp]  dec eax  jg .Loop  pop ebp  pop ebx  pop edi  pop esi  ret  ; this one's just a little faster than gcc's code. Very little.align 16xvid_HFilter_31_x86:  push esi  push edi  push ebx  mov esi, [esp+4  +12]  ; Src1  mov edi, [esp+8  +12]  ; Src2  mov eax, [esp+12 +12]  ; Nb_Blks  lea eax,[eax*8]  lea esi, [esi+eax]  lea edi, [esi+eax]  neg eax.Loop:  ; 6c  movzx ecx, byte [esi+eax]  movzx edx, byte [edi+eax]  lea ebx, [ecx+edx+2]  lea ecx,[ebx+2*ecx]  lea edx,[ebx+2*edx]  shr ecx,2  shr edx,2  mov [esi+eax], cl  mov [edi+eax], dl  inc eax  jl .Loop  pop ebx  pop edi  pop esi  ret;//////////////////////////////////////////////////////////////////////;// 16b downsampling 16x16 -> 8x8;//////////////////////////////////////////////////////////////////////%macro HFILTER_1331 2  ;%1:src %2:dst reg. -trashes mm0/mm1/mm2  movq mm2, [Mask_ff]  movq %2,  [%1-1]    ;-10123456  movq mm0, [%1]      ; 01234567  movq mm1, [%1+1]    ; 12345678  pand  %2, mm2       ;-1|1|3|5  pand mm0, mm2       ; 0|2|4|6  pand mm1, mm2       ; 1|3|5|7  pand mm2, [%1+2]    ; 2|4|6|8  paddusw mm0, mm1  paddusw %2, mm2  pmullw mm0,  mm7  paddusw %2, mm0%endmacro%macro VFILTER_1331 4  ; %1-4: regs  %1-%2: trashed  paddsw %1, [Cst32]  paddsw %2, %3  pmullw %2, mm7  paddsw %1,%4  paddsw %1, %2  psraw %1, 6%endmacro;===========================================================================;; void xvid_Filter_18x18_To_8x8_mmx(int16_t *Dst,;                                   const uint8_t *Src, const int BpS);;;===========================================================================%macro COPY_TWO_LINES_1331 1     ; %1: dst  HFILTER_1331 edx    , mm5  HFILTER_1331 edx+eax, mm6  lea edx, [edx+2*eax]  VFILTER_1331 mm3,mm4,mm5, mm6  movq [%1], mm3  HFILTER_1331 edx    , mm3  HFILTER_1331 edx+eax, mm4  lea edx, [edx+2*eax]  VFILTER_1331 mm5,mm6,mm3,mm4  movq [%1+16], mm5%endmacroalign 16xvid_Filter_18x18_To_8x8_mmx:  ; 283c   (~4.4c per output pixel)  mov ecx, [esp+4]  ; Dst  mov edx, [esp+8]  ; Src  mov eax, [esp+12] ; BpS  movq mm7, [Cst3]  sub edx, eax    ; mm3/mm4/mm5/mm6 is used as a 4-samples delay line.      ; process columns 0-3  HFILTER_1331 edx    , mm3   ; pre-load mm3/mm4  HFILTER_1331 edx+eax, mm4  lea edx, [edx+2*eax]  COPY_TWO_LINES_1331 ecx + 0*16  COPY_TWO_LINES_1331 ecx + 2*16  COPY_TWO_LINES_1331 ecx + 4*16  COPY_TWO_LINES_1331 ecx + 6*16      ; process columns 4-7  mov edx, [esp+8]  sub edx, eax  add edx, 8  HFILTER_1331 edx    , mm3   ; pre-load mm3/mm4  HFILTER_1331 edx+eax, mm4  lea edx, [edx+2*eax]  COPY_TWO_LINES_1331 ecx + 0*16 +8  COPY_TWO_LINES_1331 ecx + 2*16 +8  COPY_TWO_LINES_1331 ecx + 4*16 +8  COPY_TWO_LINES_1331 ecx + 6*16 +8  ret;===========================================================================;; void xvid_Filter_Diff_18x18_To_8x8_mmx(int16_t *Dst,;                                        const uint8_t *Src, const int BpS);;;===========================================================================%macro DIFF_TWO_LINES_1331 1     ; %1: dst  HFILTER_1331 edx    , mm5  HFILTER_1331 edx+eax, mm6  lea edx, [edx+2*eax]  movq mm2, [%1]  VFILTER_1331 mm3,mm4,mm5, mm6  psubsw mm2, mm3  movq [%1], mm2  HFILTER_1331 edx    , mm3  HFILTER_1331 edx+eax, mm4  lea edx, [edx+2*eax]  movq mm2, [%1+16]  VFILTER_1331 mm5,mm6,mm3,mm4  psubsw mm2, mm5  movq [%1+16], mm2%endmacroalign 16xvid_Filter_Diff_18x18_To_8x8_mmx:  ; 302c  mov ecx, [esp+4]  ; Dst  mov edx, [esp+8]  ; Src  mov eax, [esp+12] ; BpS  movq mm7, [Cst3]  sub edx, eax    ; mm3/mm4/mm5/mm6 is used as a 4-samples delay line.      ; process columns 0-3  HFILTER_1331 edx    , mm3   ; pre-load mm3/mm4  HFILTER_1331 edx+eax, mm4  lea edx, [edx+2*eax]  DIFF_TWO_LINES_1331 ecx + 0*16  DIFF_TWO_LINES_1331 ecx + 2*16  DIFF_TWO_LINES_1331 ecx + 4*16  DIFF_TWO_LINES_1331 ecx + 6*16      ; process columns 4-7  mov edx, [esp+8]  sub edx, eax  add edx, 8  HFILTER_1331 edx    , mm3   ; pre-load mm3/mm4  HFILTER_1331 edx+eax, mm4  lea edx, [edx+2*eax]  DIFF_TWO_LINES_1331 ecx + 0*16 +8  DIFF_TWO_LINES_1331 ecx + 2*16 +8  DIFF_TWO_LINES_1331 ecx + 4*16 +8  DIFF_TWO_LINES_1331 ecx + 6*16 +8  ret;//////////////////////////////////////////////////////////////////////  ; pfeewwww... Never Do That On Stage Again. :)