rick@nsc.uucp (Rick Pelleg) (08/11/88)
We have written a Modula-2 compiler which produces code for the NS32000
family of CPUs. One problematic implementation issue is that of the size
of enumeration types and subrange types.
The two possibilities are:
1. An enumeration/subrange occupies the least possible amount of bytes (the
NS32000 are byte addressable machines).
2. An enumeration/subrange always occupies four bytes (a double-word). This
is the "natural" size on the NS32000.
The issue is the usual "speed vs. space" tradeoff. The first possibility of
course saves space, but is slower whenever doing mixed sized calculations
or array indexing, because of the conversions that must be done.
Any opinions about this issue? In particular, I am interested in the following
two questions:
1. How do Modula-2 compilers you know implement enumerated types and
subranges?
2. How would you like to see them implemented?
I'll summarize any interesting email I receive on the subject.
Thanks,
--
--- Rick Pelleg
National Semiconductor (Israel)
6 Maskit st. P.O.B. 3007, Herzlia 46104, Israel Tel. (972)52-522264
rick%taux01@nsc.com @{hplabs,pyramid,sun,decwrl}
--
Send compilers articles to ima!compilers or, in a pinch, to Levine@YALE.EDU
Plausible paths are { ihnp4 | decvax | cbosgd | harvard | yale | bbn}!ima
Please send responses to the originator of the message -- I cannot forward
mail accidentally sent back to compilers. Meta-mail to ima!compilers-requestrick@taux02.UUCP (Rick Pelleg ) (08/30/88)
Here is a summary of the responses received so far to my question on the net, regarding the size of enumeration and subrange types in Modula-2. I repeat the main points in my question at the end of this summary. ############################################################################### Stuart Lomas (sjl%myrias) (Compilers Group; Myrias Research Corporation; Edmonton, Alberta, Canada) If the time overhead is fairly small, use smallest amount of space. Best solution: Pick a default; allow user to override with compiler option. Default should be the fastest option. ken%cs.rochester: Make them full ints. The saving in pack/unpack code makes up for space wastage. Mick Jordan (mjj%acornrc): Many Modula-2 compilers implement enumerated types, subranges and booleans in one word, to avoid packing or using byte instructions. Would optimise for speed against space these days. The real pain in not having byte-sized quantities is in the interface to C. R. Tim Coslet (R_Tim_Coslet%cup.portal): Consider compiler options. Christopher Jewell (chrisj%cup.portal): Subranges usually occupy the same storage as the base type, but enums often occupy the minimum space required. Enumerations are often restricted to 256 elements. Pete Soper (soper%maxzilla - Encore Computer Corp): The 32k version of the ETH compiler - 1 byte for any enumeration, 4 bytes for integer/cardinal subranges. Except for memory tight environments or interface with other languages, allocate 4 bytes for every scalar type. Note that alignment causes waste of bytes anyway. Jeff Woolsey (woolsey%nsc) Would like to see enumerations implemented in minimum space. 32 bits is OVERKILL for enumerations. ############################################################################### My question was: > One problematic implementation issue in our NS32000 Modula2 compiler is that > of the size of enumeration types and subrange types: > 1. An enumeration/subrange occupies the least possible amount of bytes (the > NS32000 are byte addressable machines). > 2. An enumeration/subrange always occupies four bytes (a double-word). This > is the "natural" size on the NS32000. >Any opinions about this issue? In particular, I am interested in the >following two questions: > 1. How do Modula-2 compilers you know implement enumerated types and > subranges? > 2. How would you like to see them implemented? -- --- Rick Pelleg National Semiconductor (Israel) 6 Maskit st. P.O.B. 3007, Herzlia 46104, Israel Tel. (972)52-522264 rick%taux01@nsc.com @{hplabs,pyramid,sun,decwrl}