1. 7
  1. 7

    Had to read half of the article before it finally sunk in that the headline is about elimination of stores (memory writes) by computer hardware, and not some kind of metaphor for the place down the street where you buy nuts and botls and paint.

    1. 3

      I had to read your comment before realizing that eliminating memory writes is not the only possible interpretation ;)

    2. 1

      Perhaps this predictor also considers factors such as the occupancy of outbound queues: when the bus is near capacity, searching for eliminating redundant writes might be more worth the power or latency penalty compared to the case when there is little apparent pressure on the bus.

      I think the outcome of Predictor Test is actually consistent with this theory: in the alt01 case, it checks for redundancy ~60% of the time, and since only half of the data to be written is zero, eventually ~30% of the store operations are eliminated.