Removes unsafe implementations where in theory it should not affect perfoemance. I checked all implmenetations in compiler explorer, they give the equivalent code except for slice.fill() which is one zero check longer.
fastcmp is actually longer in assembly so i removed it as it contained unsafe too. the default cmp also calls the libc memcmp https://godbolt.org/z/83xTP74x1
Benchmarks are all over the place, so it's difficult to have a clear reading, it looks like a +-5% change in both ways with a slight bias for improvement.
If someone else can benchmark and corroborate my findings it would be great.
My benchmarks were done on a raid0 nvme windows machine.
Removes unsafe implementations where in theory it should not affect perfoemance. I checked all implmenetations in compiler explorer, they give the equivalent code except for
slice.fill()which is one zero check longer.fastcmp is actually longer in assembly so i removed it as it contained unsafe too. the default cmp also calls the libc memcmp https://godbolt.org/z/83xTP74x1
Benchmarks are all over the place, so it's difficult to have a clear reading, it looks like a +-5% change in both ways with a slight bias for improvement. If someone else can benchmark and corroborate my findings it would be great.
My benchmarks were done on a raid0 nvme windows machine.
Related to #1487