vaddb

vaddb is a vector instruction used in ARM’s NEON (Advanced SIMD) extension to perform element-wise addition of 8-bit integers. It operates on two source vectors and writes the result to a destination vector, with each corresponding pair of 8-bit lanes added independently.

In typical NEON usage, a 128-bit vector contains 16 lanes of 8-bit elements, so vaddb adds 16 pairs in parallel. The arithmetic is wraparound by default, meaning sums are computed modulo 256. The hardware does not apply saturation unless a separate saturating instruction is used.

Variants and syntax tend to vary by assembler and language binding. In assembly, you may see forms like vaddb.8 or simply vaddb, depending on the syntax and operand specification. In C or C++ via NEON intrinsics, the operation is commonly expressed with intrinsics such as vaddq_u8, which corresponds to a 128-bit, 16-lane unsigned 8-bit addition. The exact intrinsic name can differ by compiler and header file, but the semantic effect is the same.

Related instructions include saturating and wider variants, such as vqaddb (saturating addition of 8-bit integers) and other vector add/subtract operations for different bit-widths or signed interpretations. In other ISAs, a similar packed 8-bit addition exists under names like VPADDB in x86 SIMD extensions, reflecting the common data-parallel operation across architectures.

Use cases for vaddb include multimedia processing, image and video processing, and other data-parallel workloads where byte-wise addition across large data sets can be executed efficiently via SIMD. It is a fundamental primitive for algorithms that operate on packed 8-bit data.