Does anyone have an idea how to resolve the AGI in the following piece of assembly code? shr edx, 16 ; 1 lea eax, [eax+esp+02000h] ; 1 - 1 lea ebx, [ebx+ebp+02000h] ; 1 and eax, 0F8077FFh ; 1 - 1 and ebx, 007F87FFh ; 1 mov ecx, [esi+edx*4] ; 1 - 1 lea edx, [eax+ebx] ; 2 ; Exp_AGI_U mov [edi+0], ecx ; 1 - 1 It''s currently a 5-cycle code for Pentium, but I really would like to make it 4-cycle. It''s for an affine texture mapper, and making it one cycle faster would mean a big speedup. I already have a 4-cycle inner pixel loop, but it doesn''t use texture tiling and it''s not free to do Unreal style dithering: add ebx, ebp ; 1 - 1 sbb edx, edx ; 1 add eax, esp ; 1 - 1 mov [edi+0], ecx ; 1 mov ecx, [esi*4] ; 1 - 1 adc esi, [VStep+4+edx*4] ; 2 Theoretically this is the fastest Pentium version of this method. Does anyone know good resources about fast inner pixel loops, like the one from Unreal? Thanks, Nick