NASM 2.05 based x86 Instruction Reference[ch232]
A.5.208 PDISTIB: MMX Packed Distance and Accumulate with Implied Register PDISTIB mm,m64 ; 0F 54 /r [CYRIX,MMX] PDISTIB, specific to the Cyrix MMX extensions, treats its two input operands as vectors of eight unsigned bytes. For each byte position, it finds the absolute difference between the bytes in that position in the two input operands, and adds that value to the byte in the same position in the implied output register. The addition is saturated to an unsigned byte in the same way as PADDUSB. To work out the implied register, invert the lowest bit in the register number. So PDISTIB MM0,M64 would put the result in MM1, but PDISTIB MM1,M64 would put the result in MM0. Note that PDISTIB cannot take a register as its second source operand. Operation: dstI[0-7] := dstI[0-7] + ABS(src0[0-7] - src1[0-7]), dstI[8-15] := dstI[8-15] + ABS(src0[8-15] - src1[8-15]), ....... ....... dstI[56-63] := dstI[56-63] + ABS(src0[56-63] - src1[56-63]).