On 64bit architectures V8 promotes at 32bits. (and if we talk about optimizing compilation things are getting more complicated as there int31, int32 and float6

On 64bit architectures V8 promotes at 32bits. (and if we talk about optimizing compilation things are getting more complicated as there int31, int32 and float64 can all coexist and you can have 31-bit integer stored in a float64 value - if operation was specialized for floating point values)