;----------------------------------------------------------------------------- ; ; Monolith 14. Perspective Correct Bi-linear gouraud modulated ; 565 input texture NO Z buffered 565 output. ; ; ; Exactly the same as monolith 7 except Z buffer code removed. ; ;----------------------------------------------------------------------------- INCLUDE iammx.inc INCLUDE offs_acp.inc ; Names are read LSB to MSB, so B5G6R5 means five bits of blue starting ; at the LSB, then six bits of green, then five bits of red. ;TBD check to see if this value is correct. COLOR_SHIFT equ 8 .586 .model flat ; Big seperating lines seperate code into span code ; and loop code. If span and loop are not going to ; end up being combined then it will be easy to ; seperate the code. .data ; Need externs for all of the variables that are needed for various beads EXTERN IncHighandLow16:MMWORD EXTERN UFracVFracMask:MMWORD EXTERN UV32to15Mask:MMWORD EXTERN Makelow16one:MMWORD EXTERN MaskKeepUValues:MMWORD EXTERN MaskKeepVValues:MMWORD EXTERN UFrac:MMWORD EXTERN VFrac:MMWORD EXTERN Zero:MMWORD EXTERN memD3DTFG_POINT:MMWORD EXTERN GiveUp:MMWORD EXTERN LastW:MMWORD EXTERN Val0x000a000a:MMWORD EXTERN Val0xffff:MMWORD EXTERN Val0x0000002000000020:MMWORD EXTERN Val0x0000ffff0000ffff:MMWORD EXTERN MaskRed565to888:MMWORD EXTERN MaskGreen565to888:MMWORD EXTERN MaskBlue565to888:MMWORD EXTERN MaskRed555to888:MMWORD EXTERN MaskGreen555to888:MMWORD EXTERN MaskBlue555to888:MMWORD EXTERN MaskAlpha1555to8888:MMWORD EXTERN MaskRed1555to8888:MMWORD EXTERN MaskGreen1555to8888:MMWORD EXTERN MaskBlue1555to8888:MMWORD ; TBD. I think that I want to do 0xffff instead of 0xff. This will ; have to be checked. There is a value very similiar to this in ; buf write. EXTERN SetAlphato0xffff:MMWORD EXTERN SetAlphato0xff:MMWORD ; TODO This equate are identical to the ones in texread.mas. Maybe they should be in a common .inc file. RedShift565to888 equ 8 GreenShift565to888 equ 5 BlueShift565to888 equ 3 RedShift555to888 equ 9 GreenShift555to888 equ 6 BlueShift555to888 equ 3 AlphaShift1555to8888 equ 16 RedShift1555to8888 equ 9 GreenShift1555to8888 equ 6 BlueShift1555to8888 equ 3 EXTERN BilinearMaskRed565to888:MMWORD EXTERN BilinearMaskGreen565to888:MMWORD EXTERN BilinearMaskBlue565to888:MMWORD EXTERN BilinearShiftRed565to888:MMWORD EXTERN BilinearShiftGreen565to888:MMWORD EXTERN BilinearShiftBlue565to888:MMWORD EXTERN Zero:MMWORD EXTERN DW_One_One:MMWORD EXTERN MaskOffAlpha:MMWORD EXTERN ShiftTA:MMWORD EXTERN Val0x00ff00ff00ff00ff:MMWORD EXTERN Val0x000000ff00ff00ff:MMWORD EXTERN Val0X0000000001000000:MMWORD EXTERN AlphaVal128:MMWORD EXTERN RGBVal128:MMWORD EXTERN g_uDitherValue:MMWORD EXTERN SetAlphato0xff:MMWORD EXTERN u888to565RedBlueMask:MMWORD EXTERN u888to565GreenMask:MMWORD EXTERN u888to565Multiplier:MMWORD EXTERN uVal0x000007ff03ff07ff:MMWORD EXTERN uVal0x0000078003c00780:MMWORD EXTERN u888to555RedBlueMask:MMWORD EXTERN u888to555GreenMask:MMWORD EXTERN u888to555Multiplier:MMWORD EXTERN uVal0x000007ff07ff07ff:MMWORD EXTERN uVal0x0000078007800780:MMWORD ;----------------------------------------------------------------------------- ; Span Variables StackPos dd ? uSpans dd ? ;----------------------------------------------------------------------------- ;----------------------------------------------------------------------------- ; Loop Variables iSurfaceStep dd ? uPix dd ? ;----------------------------------------------------------------------------- .code PUBLIC _MMXMLRast_14 _MMXMLRast_14: push ebp mov StackPos, esp mov eax, esp sub esp, 0Ch ; This will need to change if stack frame size changes. push ebx push esi push edi ; Put pCtx into ebx mov ebx, [eax+8] ;PD3DI_RASTPRIM pP = pCtx->pPrim; mov ecx, [ebx+RASTCTX_pPrim] ;while (pP) ;{ PrimLoop: cmp ecx, 0 je ExitPrimLoop ;UINT16 uSpans = pP->uSpans; movzx eax, word ptr [ecx+RASTPRIM_uSpans] mov uSpans, eax ;PD3DI_RASTSPAN pS = (PD3DI_RASTSPAN)(pP + 1); mov ebp, ecx add ebp, SIZEOF_RASTPRIM ;while (uSpans-- > 0) ;{ SpanLoop: mov edx, uSpans mov eax, edx dec eax mov uSpans, eax test edx, edx jle ExitSpanLoop ;pCtx->pfnBegin(pCtx, pP, pS); ;----------------------------------------------------------------------------- ; LoopAny code inserted here. This is to get rid of an extra ; jump. ;----------------------------------------------------------------------------- ; Setup Code begins ; get values to iterate ;uPix = pS->uPix; movzx eax, word ptr [ebp+RASTSPAN_uPix] mov uPix, eax ;pCtx->SI.iDW = 0x0; mov dword ptr [ebx+RASTCTX_SI+SPANITER_iDW], 0 mov esi, [ebp+RASTSPAN_iW] movq mm5, MMWORD PTR [ebp+RASTSPAN_iUoW1] ;pCtx->SI.iUd_TexNum = d_WTimesUVoW(pS->iW,pS->iUoW1); ;pCtx->SI.iVd_TexNum = d_WTimesUVoW(pS->iW,pS->iVoW1); pslld mm5, 8 shl esi, 4 movd eax, mm5 psrlq mm5, 32 imul esi mov [ebx+RASTCTX_SI+SPANITER_iU1], edx movd eax, mm5 imul esi mov [ebx+RASTCTX_SI+SPANITER_iV1], edx ;if (pP->iDOoWDX > 0) ;{ cmp dword ptr [ecx+RASTPRIM_iDOoWDX], 0 jg SpecialWLastMonTest ;// iSpecialW should be negative for the first 3 pixels of span ;pCtx->SI.iSpecialW = -3; mov word ptr [ebx+RASTCTX_SI+SPANITER_iSpecialW], -3 jmp DoneSpecialWifMonTest ;} ;else ;{ SpecialWLastMonTest: ;// iSpecialW should be negative for the last 3 pixels of span ;pCtx->SI.iSpecialW = 0x7fff - uPix; mov eax, 07fffh sub eax, uPix ;pCtx->SI.iSpecialW += 5; // this may wrap, but it should add eax, 5 mov [ebx+RASTCTX_SI+SPANITER_iSpecialW], eax ;} DoneSpecialWifMonTest: ;if (pP->uFlags & D3DI_RASTPRIM_X_DEC) ;{ mov eax, [ecx+RASTPRIM_uFlags] and eax, D3DI_RASTPRIM_X_DEC test eax, eax jz LeftToRightSpan ;iSurfaceStep = -pCtx->iSurfaceStep; mov eax, [ebx+RASTCTX_iSurfaceStep] neg eax mov iSurfaceStep, eax ;} jmp DoneSpanDirif ;else ;{ LeftToRightSpan: ;iSurfaceStep = pCtx->iSurfaceStep; mov eax, [ebx+RASTCTX_iSurfaceStep] mov iSurfaceStep, eax ;} DoneSpanDirif: ; Setup Code Ends ; ---------------------------------------------------------------------------------------------------------------- ; Loop Code Begins ;//while (1) ;//{ PixelLoop: ; texturecode ;--------------------------------------------------------------------------- ;void Tex1Addr_TexAddrWrapMirror_Persp_Bilinear_NoMip(PD3DI_RASTCTX pCtx, PD3DI_RASTPRIM pP, ; PD3DI_RASTSPAN pS) ;{ ;PD3DI_SPANTEX pTex = &pCtx->Texture[0]; mov esi, [ebx+RASTCTX_pTexture] ; ---------------------------------------- ; Doing UV calculation a little more accurate ; Exactly like C code. ; I shift iU and iV to the right not by (TEX_FINAL_SHIFT - iShiftU0) but by ; (TEX_FINAL_SHIFT - iShiftU0 - 6). iShiftU0 = pTex->iShiftU - iLOD0 ; (TEX_FINAL_SHIFT - (pTex->iShiftU - iLOD0)) ; (TEX_FINAL_SHIFT + iLOD0 - pTex->iShiftU) ; COMMENT1** ; If textures have a max of 1024 then shiftU0 would be at most 10 which would ; make (TEXT_FINAL_SHIFT - iShiftU - 6) at most zero. This is why I choose 6 ; It will also give bi-linear 6 bits of precision I think it was said that ; only five was needed. ;INT16 iShiftU0 = pTex->iShiftU - iLOD0; ;INT16 iShiftV0 = pTex->iShiftV - iLOD0; movq mm5, MMWORD PTR Val0x000a000a ; This is TEX_FINAL_SHIFT - 6 = 10. ;iLOD0 is zero in monolithic case so no subtraction needed. movd mm4, [esi+SPANTEX_iShiftU] psubw mm5, mm4 movq mm4, mm5 pand mm5, MMWORD PTR Val0xffff psrld mm4, 16 movd mm1, [ebx+RASTCTX_SI+SPANITER_iU1] psrad mm1, mm5 movd mm2, [ebx+RASTCTX_SI+SPANITER_iV1] psrad mm2, mm4 punpckldq mm1, mm2 psubd mm1, MMWORD PTR Val0x0000002000000020 ; Texture Pitch cannot be calculated so it must be looked up in the iShiftPitch table ; ----------------- Start of hack ; ATTENTION This is really hacked right now. Just to get it working ; Pitch would be better for me, instead of shift pitch. ; With actual pitch, this would be two moves and a shift. ;shl eax, 1 movzx edx, word ptr [esi+SPANTEX_iShiftPitch] add edx, 16 movd mm2, edx movq mm5, MMWORD ptr Makelow16one pslld mm5, mm2 ;pslld mm5, 16 ;. Use this after hack. ; not needed in hacked version since i add to shifted value. ; ----------------- End of hack por mm5, MMWORD ptr Makelow16one ; Make the low 16 bits of dword one ; This helps in calculating texture address. ; Gets U and V value into mm1 so that it can be mirrored, wrapped or ; clamped. This can be done for two values in the point case ; or four values in the bilinear case. ;INT32 iUFrac = iU00 & 0x03f; ;INT32 iVFrac = iV00 & 0x03f; ;iU00 >>= 6; ;iV00 >>= 6; movq mm2, mm1 psrad mm1, 6 ;pand mm1, MMWORD PTR Val0x0000ffff0000ffff pand mm2, dword ptr UFracVFracMask ; UFracVFracMask = 0x0000003f0000003f ; Going to use only 8 bits for bi-linear so that I can do a pmullw. ; Currently at 6 bits so shift up by 2. psllw mm2, 2 movq mm0, mm2 ; Replicate VFrac value for bilinear punpckhwd mm2, mm2 punpcklwd mm2, mm2 ; Replicate UFrac Value for bilinear punpcklwd mm0, mm0 punpcklwd mm0, mm0 movq dword ptr VFrac, mm2 movq dword ptr UFrac, mm0 ;INT32 iU01 = iU00 + 1; ;INT32 iV01 = iV00 + 1; packssdw mm1, mm1 ; replicate U and V value to upper 16 bit locations paddw mm1, dword ptr IncHighandLow16 ; This will make texture values be (High word to low word): ; iV01, iU00, iV00, iU01 ; Need to do this to make texture look up for bilinear easier. ; I have to combine to get all combinations anyway. It just ; happens to be better for me to have iV00, iU01 pair first. ;UINT16 uMaskU0 = pTex->uMaskU >> iLOD0; UINT16 uMaskV0 = pTex->uMaskV >> iLOD0; ; put mask in mm3 and replicate to match location for wrap/mirror/clamp movd mm0, [esi+SPANTEX_uMaskU] ; Load U and V mask ; replicate mask if doing bilinear punpckldq mm0, mm0 ; Monolith cases assumed that iLOD0 was zero so no shift needed. ;INT16 iFlip; ; MM1 should contain 16 bit iU and iV for both texture locations ; End Result is MM1 value wrapped or mirrored ; in Bilinear Case, four values can be done ; iU00, iV00, iU01, iV01 ; This code really does alot for the bilinear case and is kinda wasteful ; in the normal mode. ;iFlip1 = iU00 & pTex->iFlipMaskU; ;iFlip2 = iV00 & pTex->iFlipMaskV; ;iFlip3 = iU01 & pTex->iFlipMaskU; ;iFlip4 = iV01 & pTex->iFlipMaskV; movq mm7, mm1 ; Point doesnt need replication movd mm4, [esi+SPANTEX_iFlipMaskU] ; if bilinear replicate values together, Point doesnt need this. punpckldq mm4, mm4 ; Monolith cases assumed that iLOD0 was zero so no shift needed. pand mm7, mm4 ;iFlip1 = MMX_cmpeqw(iFlip1, 0); ;iFlip2 = MMX_cmpeqw(iFlip2, 0); ;iFlip3 = MMX_cmpeqw(iFlip3, 0); ;iFlip4 = MMX_cmpeqw(iFlip4, 0); pcmpeqw mm7, MMWORD PTR Zero ;iFlip1 = uMaskU0 & ~ iFlip1; ;iFlip2 = uMaskV0 & ~ iFlip2; ;iFlip3 = uMaskU0 & ~ iFlip3; ;iFlip4 = uMaskV0 & ~ iFlip4; pandn mm7, mm0 ;iU00 &= uMaskU0; ;iV00 &= uMaskV0; ;iU01 &= uMaskU0; ;iV01 &= uMaskV0; pand mm1, mm0 ;iU00 ^= iFlip1; ;iV00 ^= iFlip2; ;iU01 ^= iFlip3; ;iV01 ^= iFlip4; pxor mm1, mm7 ; Result in mm1 now since TexAddrAll ends up that way. ; Making other two cases for texture addressing has to be simplier than ; this and not use so many registers. Puts U1 V0 U0 V1 into mm3. ; TBD Make this better. ; values are still stored as iV01, iU00, iV00, iU01 movq mm2, mm1 movq mm3, mm1 ; Calculate address for 1st and 3rd texels pmaddwd mm1, mm5 ; Throw in first address calculation. ; Just to get it started. Calculate ; iU0+iV1*iShiftU0 and iU1+iV0*iShiftU0 ; values are being changed to iV01, iU01, iV00, iU00 ; seven instructions for this seems excessive. pand mm2, MMWORD ptr MaskKeepUValues pand mm3, MMWORD ptr MaskKeepVValues movq mm4, mm2 psllq mm2, 32 psrlq mm4, 32 por mm3, mm2 por mm3, mm4 ; Calculate address for 2nd and 4th texels pmaddwd mm3, mm5 ; Calculates iU1+iV0*iShiftU0 and iU0+iV1*iShiftU0 mov edi, [esi+SPANTEX_pBits] ;UINT32 uTex00 = pCtx->pfnTexRead(iU00, iV00, pTex->iShiftU, ; pTex->pBits[iLOD0], &pCtx->Texture[0]); ; Combine U and V values before making call. ; -------------------- In Monolithic version calls are inlined. ;D3DCOLOR TexRead_B5G6R5_NoBorder(INT32 iU, INT32 iV, INT32 iShiftU, PUINT8 pBits, PD3DI_SPANTEX pTex) ;{ ; Added code for color converting 2 pixels at a time ; movq mm2, MMWORD PTR Zero pxor mm2, mm2 ; 1st (mm1) and 2nd (mm3) texel movd eax, mm3 ; load 2nd texel address movzx eax, word ptr [edi+2*eax] movd mm4, eax ; mm4 = 2nd texel movd eax, mm1 ; load 1st texel address movzx eax, word ptr [edi+2*eax] movd mm7, eax ; mm7 = 1st texel ; mm7 = 2nd texel (high 32 bits), 1st texel (low 32 bits) punpckldq mm7, mm4 movq mm5, mm7 movq mm4, mm7 pand mm5, MMWORD PTR BilinearMaskRed565to888 ; = 0x0000f8000000f800 pand mm7, MMWORD PTR BilinearMaskGreen565to888 ; = 0x000007e0000007e0 pand mm4, MMWORD PTR BilinearMaskBlue565to888 ; = 0x0000001f0000001f pslld mm5, MMWORD PTR BilinearShiftRed565to888 ; = 8 pslld mm7, MMWORD PTR BilinearShiftGreen565to888 ; = 5 pslld mm4, MMWORD PTR BilinearShiftBlue565to888 ; = 3 por mm7, mm5 ; combine R+G por mm7, mm4 ; combine (R+G) + B movq mm4, mm7 ; copy 1st and 2nd texels ; mm4 calculated from high 32 bits of mm3 (2nd texel) ; pad high 8 bits of each component with zeros because D3DCOLOR has 16 ; bits for each color component punpckhbw mm4, mm2 ; mm7 calculated from low 32 bits of mm1 (1st texel) ; pad high 8 bits of each component with zeros because D3DCOLOR has 16 ; bits for each color component punpcklbw mm7, mm2 psrlq mm3, 32 ; shift 4th texel address into low 32 bits ; mm7 = final calc on 1st and 2nd texel psubw mm7, mm4 psllw mm4, 8 pmullw mm7, dword ptr UFrac paddw mm7, mm4 ; 3rd (mm1) and 4th (mm3) texel movd eax, mm3 ; load 4th texel address psrlq mm1, 32 ; shift 3rd texel address into low 32 bits movzx eax, word ptr [edi+2*eax] movd mm6, eax ; mm6 = 4th texel movd eax, mm1 ; load 3rd texel address movzx eax, word ptr [edi+2*eax] movd mm4, eax ; mm4 = 3rd texel ; mm6 = 4th texel (high 32 bits), 3rd texel (low 32 bits) punpckldq mm6, mm4 movq mm5, mm6 movq mm4, mm6 pand mm5, MMWORD PTR BilinearMaskRed565to888 ; = 0x0000f8000000f800 pand mm6, MMWORD PTR BilinearMaskGreen565to888 ; = 0x000007e0000007e0 pand mm4, MMWORD PTR BilinearMaskBlue565to888 ; = 0x0000001f0000001f pslld mm5, MMWORD PTR BilinearShiftRed565to888 ; = 8 pslld mm6, MMWORD PTR BilinearShiftGreen565to888 ; = 5 pslld mm4, MMWORD PTR BilinearShiftBlue565to888 ; = 3 por mm6, mm5 ; combine R+G por mm6, mm4 ; combine (R+G) + B movq mm4, mm6 ; copy 3rd and 4th texels ; mm4 calculated from high 32 bits of mm3 (4th texel) ; pad high 8 bits of each component with zeros because D3DCOLOR has 16 ; bits for each color component punpckhbw mm4, mm2 ; mm6 calculated from low 32 bits of mm1 (3rd texel) ; pad high 8 bits of each component with zeros because D3DCOLOR has 16 ; bits for each color component punpcklbw mm6, mm2 psubw mm6, mm4 psllw mm4, 8 pmullw mm6, dword ptr UFrac movq mm1, mm7 ; mm6 = final calc on 3rd and 4th texel paddw mm6, mm4 ; mm1 = final calc on 1st+2nd texel and 3rd+4th texel psrlw mm6, 8 psrlw mm7, 8 psubw mm6, mm7 pmullw mm6, dword ptr VFrac paddw mm1, mm6 psrlw mm1, 8 ;modulate ;UINT16 uB = pS->uB>>COLOR_SHIFT; ;UINT16 uG = pS->uG>>COLOR_SHIFT; ;UINT16 uR = pS->uR>>COLOR_SHIFT; movq mm4, [ebp+RASTSPAN_uB] psrlw mm4, COLOR_SHIFT ; COLOR_SHIFT is set to 8. ; Alpha not needed in this monolith ;UINT16 uTB = (UINT16)(RGBA_GETBLUE(pCtx->SI.TexCol[0])); ;UINT16 uTG = (UINT16)(RGBA_GETGREEN(pCtx->SI.TexCol[0])); ;UINT16 uTR = (UINT16)(RGBA_GETRED(pCtx->SI.TexCol[0])); ;UINT16 uTA = (UINT16)(RGBA_GETALPHA(pCtx->SI.TexCol[0])); ; this is a PMULLW, which works on unsigned 16 bit quantities ;pCtx->SI.uBB = uB*uTB; ;pCtx->SI.uBG = uG*uTG; ;pCtx->SI.uBR = uR*uTR; ;pCtx->SI.uBA = uTA<uB += pP->iDBDX; pS->uG += pP->iDGDX; ;pS->uR += pP->iDRDX; pS->uA += pP->iDADX; movq mm1, [ebp+RASTSPAN_uB] paddw mm1, [ecx+RASTPRIM_iDBDX] movq [ebp+RASTSPAN_uB], mm1 ;pS->iUoW1 += pP->iDUoW1DX; ;pS->iVoW1 += pP->iDVoW1DX; movq mm5, [ebp+RASTSPAN_iUoW1] paddd mm5, [ecx+RASTPRIM_iDUoW1DX] movq [ebp+RASTSPAN_iUoW1], mm5 ;pS->iOoW += pP->iDOoWDX; mov eax, [ebp+RASTSPAN_iOoW] add eax, [ecx+RASTPRIM_iDOoWDX] mov [ebp+RASTSPAN_iOoW], eax ;INT32 iWn0 = pS->iW + pCtx->SI.iDW; // 1.15.16 ; TODO Could do this and OoW Add at same time with MMX. mov edx, [ebp+RASTSPAN_iW] mov LastW, edx ; Save iW to calc iDW for next time. add edx, [ebx+RASTCTX_SI+SPANITER_iDW] ;if (pCtx->SI.iSpecialW < 0) ;{ xor edi, edi cmp di, word ptr [ebx+RASTCTX_SI+SPANITER_iSpecialW] jle DontDoSpecialW1 ;DoSpecialW1: ; This label is a left over from when ;if (iWn0 < 0) ;{ cmp edx, edi jl WOutOfRange1 ;iWn0 = pS->iW >> 1; // use iW/2 as a guess, instead mov edx, LastW sar edx, 1 ;} WOutOfRange1: ;VAL32 iWn1; ;INT16 iWnOld = iWn0 + 0x100; // make sure while fails first time ; Dont need to make sure it fails. I do a post test which guarentees it will execute once. ;INT32 iGiveUp = 7; mov GiveUp, 8 ; Pre decrementing instead of post decrementing. ;while((abs(iWnOld - iWn0) > 0x20) && (iGiveUp-- > 0)) ;{ SpecW1Loop1: ; Could move this to bottom of loop and combine results somehow. ; TBD look at it more. dec GiveUp jz ExitSpecWLoop1 ; Shift iOoW by one since imul cannot have sign bit set ; OoW cannot reach one, only 0x7fffffff ;shr eax, 1 ; 1.31 >> 1 = 1.30 ; Get ready to do Two minus iOoW*iW mov esi, (1 SHL 16) ;iWnOld = iWn0; mov edi, edx ; Result should be close to one so we want most of the ; precision in the low bits. Need to give more bits ; leaway since these are the bad cases. ; iWn1 = imul32h(pS->iOoW, iWn0); // 1.31*1.15.16 = 1.16.47 >> 32 = 1.16.15 imul edx ;iWn1 = (1L<<16) - iWn1; // 2.0 - iWn1 sub esi, edx ;while(iWn1.i < 0) ;{ SpecW1Loop2: test esi, esi jns SpecW1ExitLoop2 ; This jump should be predicted correctly most of the time. ;iWn1=(iWn1+(1L<<15))>>1; // iWn1 = (iWn1 + 1.0)/2 add esi, (1 SHL 15) sar esi, 1 jmp SpecW1Loop2 ;} SpecW1ExitLoop2: ;iWn1 <<= 15; // 1.16.15 << 15 = 1.1.30 mov eax, edi shl eax, 5 ; 1.15.16 << 5 = 1.10.21 TBD Can I shift off upper bits?? shl esi, 12 ; 4.15 << 12 = 4.27 ; ;iWn0 = imul32h(iWn1, iWn0)<<2; // 1.1.30 * 1.15.16 = 1.17.46 >> 32 = 1.17.14 << 2 = 1.15.16 ; Actually 4.27 * 1.10.21 = 1.14.48 >> 32 = 1.14.16. No need for post shift. mul esi ; Have to do (abs(iWnOld - iWn0) > 0x20) code here. sub edi, edx ; These four lines are abs code. mov eax, edi sar eax, 31 xor edi, eax sub edi, eax cmp edi, 020h ;Assuming that loop will only happen once. jbe ExitSpecWLoop1 ; Reload eax with iOoW. mov eax, [ebp+RASTSPAN_iOoW] jmp SpecW1Loop1 ;} ;else ;{ DontDoSpecialW1: ; Everything should be positive in Non-SpecialW case. ;INT32 iWn1; mov esi, (1 SHL 16) mov edi, edx ; This should be close to one so Low bits are most important. ;iWn1 = (iOoW*iWn0)>>15; // 1.31*0.15.16 == 0.16.47 >> 32 = 0.16.15 mul edx ;iWn1 = (1L<<16) - iWn1; // 2.0 - iWn1 sub esi, edx ;iWn1 <<= 15; // 1.16.15 << 15 = 1.1.30 shl esi, 15 ; 0.16.15 << 15 = 0.2.30 mov eax, esi ;iWn0 = imul32h(iWn1, iWn0)<<2; // 1.1.30 * 1.15.16 = 1.17.46 >> 32 = 1.17.14 << 2 = 1.15.16 mul edi ; 0.2.30 * 1.15.16 = 1.17.46 >> 32 = 1.17.14 shl edx, 2 ; 1.17.14 << 2 = 1.15.16 ;} ;} ExitSpecWLoop1: ;pCtx->SI.iDW = iWn0 - (UINT16)pS->iW; ;pS->iW = iWn0; mov [ebp+RASTSPAN_iW], edx mov esi, edx ; Save W for multiplying by UoW and VoW sub edx, LastW mov [ebx+RASTCTX_SI+SPANITER_iDW], edx ;pCtx->SI.iSpecialW += 1; // this is supposed to wrap past 0x7fff sometimes inc word ptr [ebx+RASTCTX_SI+SPANITER_iSpecialW] ;pCtx->SI.iU1 = d_WTimesUVoW(pS->iW,pS->iUoW1); ;pCtx->SI.iV1 = d_WTimesUVoW(pS->iW,pS->iVoW1); pslld mm5, 8 shl esi, 4 movd eax, mm5 psrlq mm5, 32 imul esi mov [ebx+RASTCTX_SI+SPANITER_iU1], edx movd eax, mm5 imul esi mov [ebx+RASTCTX_SI+SPANITER_iV1], edx ;//pS->pSurface += iSurfaceStep; mov edx, dword ptr [ebp+RASTSPAN_pSurface] add edx, iSurfaceStep mov dword ptr [ebp+RASTSPAN_pSurface], edx ;#ifdef DBG ;// handy for debug to see where we are ;//pS->uX += (INT16)pCtx->SI.iXStep; ;#endif ;// } // while jmp PixelLoop ExitPixelLoop: ; Loop code ends ;----------------------------------------------------------------------------- ; LoopAny code ends here ;----------------------------------------------------------------------------- ;pS++; add ebp, SIZEOF_RASTSPAN ;} jmp SpanLoop ExitSpanLoop: ;pP = pP->pNext; mov ecx, [ecx+RASTPRIM_pNext] ;} jmp PrimLoop ExitPrimLoop: ;_asm{ emms ;} ;return S_OK; xor eax, eax ;} pop edi pop esi pop ebx mov esp, StackPos pop ebp ret END