I've implemented my own RingBuffer, ModFreeRingBuffer, that uses conditional subtraction instead of expensive modulo operations. It maintains a base index that is kept within a fixed range of the buffer, and is conditionally subtracted as necessary. When used in a loop, the compiler is able to optimize the involved branches quite effectively.
ModFreeRingBuffer uses alloc, just like AllocRingBuffer; however, instead of using a Vec, I use a Box (the Vec stores an additional length argument that is unnecessary if we pre-allocate the capacity). I've modified AllocRingBuffer to have the same optimization.
I've also integrated ModFreeRingBuffer with the benchmarking system; it beats AllocRingBuffer<T, NonPowerOfTwo>, and is very competitive (within 25% of the time taken) for AllocRingBuffer<T, PowerOfTwo>. It can definitely be tuned further (for example, pointers can be used instead of indices, or indices can use units of bytes instead of elements).
I've implemented my own
RingBuffer,ModFreeRingBuffer, that uses conditional subtraction instead of expensive modulo operations. It maintains a base index that is kept within a fixed range of the buffer, and is conditionally subtracted as necessary. When used in a loop, the compiler is able to optimize the involved branches quite effectively.ModFreeRingBufferusesalloc, just likeAllocRingBuffer; however, instead of using aVec, I use aBox(theVecstores an additional length argument that is unnecessary if we pre-allocate the capacity). I've modifiedAllocRingBufferto have the same optimization.I've also integrated
ModFreeRingBufferwith the benchmarking system; it beatsAllocRingBuffer<T, NonPowerOfTwo>, and is very competitive (within 25% of the time taken) forAllocRingBuffer<T, PowerOfTwo>. It can definitely be tuned further (for example, pointers can be used instead of indices, or indices can use units of bytes instead of elements).