here''s my problem: I''ve been using an MMX alpha blend blitter routine written in MMX by some-else for a while now. But, what I have found out is that it doesn''t handle surface widths which are not multiple of 4 properly. Here is the code it uses: i = height; do { j = width/4; // If you see two lines together this means that I am guessing that these two // instructions will pair in the Pentium U & V pipes. _asm { push edi ;Save off to restore later push esi ;Save off to restore later mov edi,lpDest ; Move the destination pointer into edi mov esi,lpSprite ; Move the source pointer into esi SPAN_RUN_565: movq mm7,[edi] ; Copy the 8 bytes pointed to by edi into mm7 movq mm6,[esi] ; Copy the 8 bytes pointed to by esi into mm6 movq mm2,ALPHA64 ; Copy ALPHA64 into mm2 movq mm0,mm7 ; RED - Copy mm7 to mm0, destination pixels pand mm0,MASKRED ; RED - and mm0 with the red channel mask [0r00 0r00 0r00 0r00] movq mm1,mm6 ; RED - Copy mm6 to mm1, source pixels pand mm1,MASKRED ; RED - and mm1 with red channel mask [0r00 0r00 0r00 0r00] psrlw mm0,11 ; RED - shift each pixel to the right by 11 [000r 000r 000r 000r] movq mm5,mm7 ; GRN - Copy destination pixels (mm7) to mm5 psrlw mm1,11 ; RED - shift each pixel to the right by 11 [000r 000r 000r 000r] paddw mm1, ADD64 movq mm3,mm6 ; GRN - Copy source pixels (mm6) to mm3 psubsw mm1,mm0 ; RED - Subtract the destination from the source pand mm5,MASKGREEN ; GRN - and mm5 with the green channel mask [00g0 00g0 00g0 00g0] pmullw mm1,mm2 ; RED - multiple the subtraction result by the ALPHA value pand mm3,MASKGREEN ; GRN - and mm3 with the green channel mask [00g0 00g0 00g0 00g0] psrlw mm5,5 ; GRN - shift each pixel to the right by 5 [000g 000g 000g 000g] psrlw mm3,5 ; GRN - Shift green source to the right by 5 nop ; Do nothing paddw mm3, ADD64 psrlw mm1,8 ; RED - Divide by 256 this is done to avoid floating point math psubsw mm3,mm5 ; GRN - Subtract the green source and destination components pmullw mm3,mm2 ; GRN - Multiple the result from the subtraction by the alpha value paddw mm1,mm0 ; RED - Add the destination red component value psubw mm1, ALPHABY4 psllw mm1,11 ; RED - Shift the red component back into place, 11 to the left movq mm0,mm7 ; BLU - Copy the destination pixels to mm0 pand mm0, MASKBLUE ; BLU - ''and'' mm0 with the blue channel mask [000b 000b 000b 000b] psrlw mm3,8 ; GRN - Divide by 256 this is done to avoid floating point math paddw mm3,mm5 ; GRN - Add the destination green component value movq mm4, mm6 ; BLU - Copy the source into mm4 psubw mm3, ALPHABY4 pand mm4, MASKBLUE ; BLU - ''and'' mm0 with the blue channel mask [000b 000b 000b 000b] psllw mm3,5 ; GRN - Shift the green component back into place, 5 to the left paddw mm4, ADD64 psubsw mm4,mm0 ; BLU - Subtract the blue source and destination components por mm1,mm3 ; GRN - ''or'' the red and green components results back together pmullw mm4,mm2 ; BLU - Multiple the result from the subtraction by the alpha value movq mm3,COLORKEY64 ; Load COLORKEY64 into mm4 psrlw mm4,8 ; BLU - Divide by 256 this is done to avoid floating point math pcmpeqw mm3,mm6 ; Compare colorKey to original source paddw mm4,mm0 ; BLU - Add the destination blue component value movq mm5,mm3 ; Copy mm3 to mm5 psubw mm4, ALPHABY4 por mm1,mm4 ; BLU - ''or'' the blue result to the red and green results pand mm5,mm7 ; ''and'' mm5 and the original destination pixels pandn mm3,mm1 ; not''s mm3 then performs and "AND" with mm1 por mm3,mm5 ; And mm3 and mm5 movq [edi],mm3 ; Copy the 4 alpha blended pixels to the destination add edi,8 ; Add 8 bytes to the destination pointer mov eax,dword ptr [j] ; Load the value of j into eax add esi,8 ; Add 8 bytes to the source pointer sub eax,1 ; subtract 1 from eax (j--) mov dword ptr [lpDest], edi ; Save the new destination pointer value mov dword ptr [j],eax ; save the new value back to j mov dword ptr [lpSprite], esi ; Save the new source pointer value cmp eax,0 ; If eax = 0 we have set the flag jg SPAN_RUN_565 ; if flag is zero finish else loop back for more emms ; Clean up the MMX registers pop esi ; Restore esi pop edi ; Restore edi } lpDest += dbuf; lpSprite += sbuf; }while (--i > 0); okay so I need a clean up routine after every line which will handle the 1, 2, or 3 pixels left over buy the move 4 at a time functions. something like this: if (j*4 != width) { handle the pixels here } can anybody give me a hand in filling in the blank?

I am the author of that code. I''m surprised I did not code for that case. My bad. If you look in the source file where you got that code from you find something like:

if (oddWidth)
{
sTemp = *((WORD*)lpSprite);

if (sTemp != ColorKey)
{
dTemp = *((WORD*)lpDest);
sb = sTemp & 0x1f;
db = dTemp & 0x1f;
sg = (sTemp >> 5) & 0x3f;
dg = (dTemp >> 5) & 0x3f;
sr = (sTemp >> 11) & 0x1f;
dr = (dTemp >> 11) & 0x1f;

*((WORD*)lpDest) = (WORD)((ALPHA * (sb - db) >> 8) + db /
((ALPHA * (sg - dg) >> 8) + dg) << 5 /
((ALPHA * (sr - dr) >> 8) + dr) << 11);
}

lpDest += 2;
lpSprite += 2;
}

This is not 100% what you need but you should be able to figure it out from there.

John