Overview

Single Instruction, Multiple Data (SIMD) is a set of instructions to perform vectorized computations.

MMX

• Introduced in 1997.

• 57 instructions.

• 2-operand instructions.

• 8 registers: MM0–MM7.
  • 64-bit registers.

A single instruction can be applied to:

• one 64-bit integer or

• two 32-bit integers or

• four 16-bit integers or

• eight 8-bit integers.

Issues:

• Re-used existing x87 floating point registers.
  • Unable to work on both floating point and SIMD data at the same time.

• Only supported operations on integers.

SSE

Streaming SIMD Extensions

• Introduced in 1999.

• 70 instructions.

• 8 new registers: XMM0–XMM7
  • 128-bit registers.
  • four 32-bit single-precision floating point numbers.

A single instruction can be applied to:

• two 64-bit double-precision floating point numbers or

• two 64-bit integers or

• four 32-bit integers or

• eight 16-bit short integers or

• sixteen 8-bit bytes.

SSE2

• Introduced in 2000.

• 144 instructions.

• 8 registers: XMM0–XMM7
  • 16 registers in x86-64 mode: XMM0–XMM15.
  • 128-bit registers.
  • Double-precision floating point operations.

• MMX integer operations on 128-bit XMM registers.

Advantages:

• MMX and x87 register do not alias one another.

Issues:

• Slow access to data in memory not aligned to a 16-byte boundary.

SSE3

• Introduced in 2004.

• 13 new instructions over SS2.

Advantages:

• Unaligned load instructions are faster.

• Horizontal instructions to speed up the several DSP and 3D operations.

SSE4

• Introduced in 2006.

• 54 instructions.

Advantages:

• Unaligned load instructions are as fast as aligned versions.

• Dot product instruction.

• Additional integer instructions.

AVX

Advanced Vector Extensions (AVX)

• Introduced in 2008.

• 8 registers.
  • Renamed XMM0–XMM7 to YMM0–YMM7.
  • 16 registers in x86-64 mode: YMM0–YMM15.
  • 256-bit registers.

• 3-operand instructions.

A single instruction can be applied to:

• eight 32-bit single-precision floating point numbers or

• four 64-bit double-precision floating point numbers.

Advantages:

• Supports 128-bit and 256-bit SIMD in AVX-128 mode.

AVX2

• Introduced in 2013.

• Fused Multiply-Accumulate (FMA) instructions.

• 3-operand FMA (FMA3) instructions.

AVX-512 (or AVX3)

• Introduced in 2013.

• 512-bit registers.

• 32 registers in x86-64 mode: ZMM0-ZMM31.

• 4-operand instructions.