3DNow! — Extended Instruction Set

 

3DNow! — Extensions

When AMD released their Athlon processor in 1999, they also took the liberty to extend their 3DNow! instruction set with a few new instructions. They basically provide a few more conversion operations and a few accumulation operations. These instructions have many names, including 3DNow!+, 3DNow!ext, 3DNow!2, 3DNow! Professional, etc.

3DNow! — Conversions Instructions

Extending the pf2id and pi2fd instructions, 3DNow!2 gives us 2 more type conversion instructions, for converting to and from 16-bit words.

pf2iw converts a floating point value to a 16-bit signed integer. To top 16 bits of value are copies of the sign bit.

pi2fw converts a 16-bit signed integer to a 32-bit floating point value. This instruction throws out data in the top 16-bits of each half, which makes it different from pi2fd.

While not strictly a conversion instruction, I couldn't justify a whole section for the pswapd instruction.

pswapd stores the top and bottom 32-bits of a register or memory location in reverse order in a destination register. The top 32-bits are stored in the bottom half of the destination, and the bottom 32-bits are stored in the top half.

3DNow! — Accumulation Extensions

3DNow!2 gives us 2 more pfacc-like instructions as well.

pfnacc behaves just like pfacc, except instead of summing the pairs it subtracts them.

pfpnacc is a bit more tricky. This instruction subtracts the top half of a register from the bottom half, and stores the result in the bottom half, and sums the top and bottom halves of a source register or memory location, storing the result in the top of the first register.