SIMD¶

Category: Parallelism · Status: stub

One line: One CPU instruction applied to several numbers at once: parallelism inside a single core.

Also called: vectorization, single instruction, multiple data.

How it connects¶

flowchart LR
  n_data_parallelism["Data parallelism"]
  n_simd["SIMD"]
  n_simd -->|is a| n_data_parallelism
  classDef center stroke-width:3px
  class n_simd center
  classDef outside stroke-dasharray: 4 3
  class n_data_parallelism outside

Is a kind of: Data parallelism

In each language¶


Rust	Platform intrinsics in `std::arch` ↗; the portable `std::simd` ↗ is nightly-only
C++	`std::simd` ↗ in C++26, after `std::experimental::simd` ↗
Java	The Vector API ↗, still an incubator module in JDK 25
C#	`Vector<T>` ↗, which supports hardware acceleration, and the instruction-set register types of `System.Runtime.Intrinsics` ↗
Swift	`SIMD` ↗ vector types such as `SIMD4<Float>` in the standard library

Where to read more¶

In the books: Modern Parallel Programming with C++ and Assembly Language, Daniel Kusswurm — ch. 1, 'SIMD Fundamentals'
In the books: Pro TBB, Michael Voss, Rafael Asenjo, James Reinders — ch. 4, 'TBB and the Parallel Algorithms of the C++ Standard Template Library' → 'Other Ways to Introduce SIMD Parallelism'
In the books: Data Parallel C++, James Reinders, Ben Ashbaugh, James Brodman, Michael Kinsner, John Pennycook, Xinmin Tian — ch. 16, 'Programming for CPUs' → 'The Basics of SIMD Hardware'
In the books: Parallel Programming with Intel Parallel Studio XE, Stephen Blair-Chappell, Andrew Stokes — ch. 4, 'Producing Optimized Code' → 'More on Auto-Vectorization'
Reference: Wikipedia: Single instruction, multiple data ↗