Hello everyone: I made some changes to the Hugs sources which give small runtime speed gains.
I think you're the first person to look at optimizing Hugs in quite a while. Great stuff!
I toyed with changing whatIs a few years ago. Instead of your table-based approach, I was thinking of using a bitwise encoding. My idea was to use one bit to select int/not int and that everything else would be interpreted as a triple of the form:
(WhatisTag,ModuleId,Index)
Where, for example, the WhatisTag is 5 bits, the ModuleId is 10 bits and the Index is 16 bits (the numbers will probably need tweaked somewhat). So, each module would have its own set of tables for storing names, types, etc.
In the Hugs-alike compiler I was working on a few years ago (never got around to finishing it) I used the following representation: a 'cell' is either boxed or unboxed, according to bit zero (0 == a heap reference), and an unboxed word consists of 7 bits of "tag" (bits 1-8), and 24 bits of data. Applications and list cells are just pairs (like Hugs), but everything else has a "box" tag in the first word. The heap stored variable-width objects unlike Hugs which only has pairs, but I don't think that's relevant. I think this is pretty similar to what you suggested above, Alastair. The appropriate macros are: #define IsRef(c) (((c) & 1) == 0) #define IsBox(c) (((c) & 1) && TagOf(c) >= BoxedThings) #define IsApp(c) (IsRef(c) && !IsBox(Fst(c))) #define MkValue(c) ((uint)(c) << 8) #define ValueOf(c) ((uint)(c) >> 8) #define TagOf(c) ((c) & 0xff) #define whatIs(c) (IsRef(c) ? (IsBox(Fst(c)) ? Fst(c) : Ap) : TagOf(c)) I remember testing various representations at the time, and this one turned out to be pretty good. Cheers, Simon