writeallocate

Writeallocate, or write-allocate, is a cache write policy used in computer memory hierarchies to determine how a write miss is handled. It specifies whether a missing cache line should be brought into the cache (allocated) before the write is performed.

Under the write-allocate policy, when a write miss occurs, the cache line containing the target address is fetched from a lower level of the memory hierarchy into the cache, and the write is then carried out on that cached copy. The updated line may later be written back to memory depending on the cache’s write-back or write-through strategy. On subsequent writes to the same line, the cache can serve them quickly from the cache, leveraging locality.

The alternative is write-no-allocate (no-write-allocate), where a write miss does not load the line into the cache; the write is sent directly to memory, and the line is not placed in the cache unless a subsequent read or a future write brings it in. This can reduce cache pollution for streaming or one-off writes that are unlikely to be reused soon.

The choice between write-allocate and write-no-allocate affects performance and bandwidth. Write-allocate tends to improve performance for temporal and spatially localized writes by keeping the data close to the processor, which is beneficial in caches with write-back policies. Write-no-allocate can be advantageous for streaming workloads that write large blocks with little reuse, reducing unnecessary cache traffic.

In practice, many processors adopt write-allocate for data caches, often in combination with write-back strategies, while some subsystems or specific workloads may opt for no-write-allocate. The policy can influence effective memory bandwidth and cache efficiency and is a key consideration in memory hierarchy design.