SIMD in the 90s: Programming Intel's Pentium MMX

In the late 1990s, the introduction of Intel's Pentium MMX processor brought SIMD (Single Instruction, Multiple Data) instructions to mainstream desktop CPUs, a major step forward in computer technology. SIMD is a parallel processing technique where a single instruction operates on multiple data points simultaneously. While modern CPUs and extensions like AVX-512, NEON, SVE, and GPUs have thousands of cores, the concept of SIMD dates back much further, with early examples like the ILLIAC IV supercomputer from the 1960s featuring hundreds of processing elements operating under a single instruction stream.

MMX, introduced with the Pentium MMX processor in 1997, provided 57 additional instructions for enhanced multimedia performance, including graphics, digital signal processing, audio, and modem software. It was a set of instructions built into the Pentium chip for improved performance. The CPU had to switch into MMX mode, which turned the first 64 bits of the x86's eight 80-bit floating-point registers into MMX registers.

The main idea behind SIMD is to process several integers packed inside a single 64-bit register simultaneously, rather than processing one integer at a time. For multimedia software, this could provide significant speedups. For example, instead of adding eight bytes individually, MMX allows all eight additions to be performed with a single instruction, demonstrating the concept of parallelism.

This SIMD-like behavior was not unique to MMX, as vector processing and SIMD have roots in earlier computer technologies. Vector processors, like the Cray-1 from 1976, featured multiple processing elements working on a sequence of elements over multiple cycles. Unlike SIMD, vector processors have a SET VECTOR SIZE instruction and process a fixed-width register, while SIMD operates on fixed-width registers.