jwf@dewey.UUCP (Jim Franklin) (11/04/83)
Green Hills Software of Pasadena, CA claims to have an optimizing C compiler for the Motorola 68000. It runs as a cross-compiler on Berkeley 4.1 UNIX. From literature we received from Green Hills (GH), the compiler appears to generate excellent code when the optimizer switches are turned on. About a month ago we sent a tape to GH with some C benchmark programs and asked them to compile them (invoking the optimizer) and send us back the assembly language output (.s) files. GH compiled the programs and sent us back the .s files, but they didn't invoke their optimizer. They told us that they don't recommend invoking the optimizer unless the programs are loop bound. Incidentally, most of our C benchmarks were loop bound. If their optimizer really worked, then we assume that they would have compiled our benchmark using the optimizer. After all, they are trying to sell a product and they should show us the best they can do. We are suspicious that either (1) their compiler crashed with the optimizer turned on (e.g., bus error - core dumped), or (2) it ran fine but generated bogus code. At this point we don't know what to make of GH and their compiler. Does anyone out there know (1) if Green Hills is real -- i.e., does their compiler really work? (2) does their optimizer generate correct code in most cases? if it doesn't always work, does it know when it doesn't work or does it just generate mysteriously incorrect code? Can you tell the compiler not to optimize sections of a file? (3) is anyone using the GH VAX to 68000 cross-compiler for large projects (say 100,000 lines of C)?