For numerical chunks, each sample has a constant size, so this change allocates and reuses a read buffer to avoid one allocation per sample. With a large sample file (64 double channels, 1M samples) the total runtime decreases by 11% and the context switches by 98%.
For numerical chunks, each sample has a constant size, so this change allocates and reuses a read buffer to avoid one allocation per sample. With a large sample file (64 double channels, 1M samples) the total runtime decreases by 11% and the context switches by 98%.