Open mattkretz opened 7 years ago
Current masked load/store uses "holes" in the accessed memory. The alternative is to load/store N (= popcount(mask)) consecutive values from/to memory starting from the given address.
https://isocpp.org/files/papers/P2664R2.html#permute_by_mask
danieltowner
commented
1 year ago
Current masked load/store uses "holes" in the accessed memory. The alternative is to load/store N (= popcount(mask)) consecutive values from/to memory starting from the given address.