12:31:51stassatswonder if it'll be useful to fold (gcd ... 1) to 1
13:11:56stassatsnow i wonder if it's worth deriving (gcd ... prime) as (or (eql 1) (eql prime))
13:19:38stassatsi don't have anything better to do, so, i'll add it too
15:43:11stassatswhat if the bignums were laid out in the opposite order?
15:43:48stassatsjust produced some vops that compare bignums against zero, having to load length to get the last digit, could just load the first one instead
15:46:07stassats(unsigned-byte 64)-p vop would be simpler too, check the length, load the first digit
15:47:06stassatsthough still need to check for different flags based on the length
15:55:33pkhuongsaturn2: common lisp compilers can't do backward propagation because the standard mandates errors. type mismatches have fully defined error behaviour.
15:57:31pkhuongeven then i don't think we're allowed to fail upstream of the type mismatch just because we know it's reachable?
15:59:03pkhuongI also personally don't care enough about safety-0 to add a pass that only works for unsafe code.
15:59:11stassatscan reduce some operations, e.g. (truly-the fixnum (truncate double-float)) just check for a fitting float and return garbage otherwise