The type bool is a 32-bit type, which makes it inefficient for use in arrays of flags. Ideally, it would be an 8-bit type.
In our benchmarks, we have been working around this manually to reduce memory footprints; see for example primes which on input size n needs an array of n flags. Using Word8.word array achieves 4x smaller memory footprint in comparison to bool array. This is significantly faster--I've measured as much as 2-3x time improvement on this particular benchmark on high core counts.
I haven't looked closely yet at what changes to the compiler would be needed to achieve 8-bit booleans, but my guess is that it might be a little tricky. It would certainly impact the FFI (which, following MLton, specifies 32-bit bools; see ForeignFunctionInterfaceTypes).
Some preliminary digging:
The type
boolis a 32-bit type, which makes it inefficient for use in arrays of flags. Ideally, it would be an 8-bit type.In our benchmarks, we have been working around this manually to reduce memory footprints; see for example
primeswhich on input sizenneeds an array ofnflags. UsingWord8.word arrayachieves 4x smaller memory footprint in comparison tobool array. This is significantly faster--I've measured as much as 2-3x time improvement on this particular benchmark on high core counts.I haven't looked closely yet at what changes to the compiler would be needed to achieve 8-bit booleans, but my guess is that it might be a little tricky. It would certainly impact the FFI (which, following MLton, specifies 32-bit bools; see ForeignFunctionInterfaceTypes).
Some preliminary digging:
WordSize.boolis set to 32 here:mpl/mlton/atoms/word-size.fun
Line 54 in a71659b