tege@sics.se (Torbj|rn Granlund) (11/29/89)
Are there any XP systems out there? I consider doing an extension to GCC to generate the three address instructions, supported by these systems. To do this, I need a net connected machine to test on. I you can do the testing, it's even better.
csg@pyramid.pyramid.com (Carl S. Gutekunst) (11/29/89)
>Are there any XP systems out there? Sure. That's the MIServer CPU. >I consider doing an extension to GCC to generate the three address >instructions, supported by these systems. Alas, these were dropped. It was one of those things that seemed like a pretty neat idea at the time, and the CPU people thought it would be trivial to add them to the new pipeline of the XP CPU. But a lot of people (including some of the compiler folks) weren't convinced they were that useful, and would have created binary incompatibility between the machines as well. When it appeared that release schedules might slip, work on the three operand instructions was quietly dropped. (Note that it was three *operand*, not three *address*.) There are some new instructions in the XP CPU, but they are applicable to the kernel only -- extending the architecture out to >8 CPUs, for example. <csg>
wendyt@cs.washington.edu (Wendy Thrash) (11/30/89)
In article <92827@pyramid.pyramid.com> csg@pyramid.pyramid.com (Carl S. Gutekunst) writes: >But a lot of people (including some of >the compiler folks) weren't convinced they were that useful, and would have >created binary incompatibility between the machines as well. When it appeared >that release schedules might slip, work on the three operand instructions was >quietly dropped. Maybe I'm just getting senile in my old age, but my memory is that there were some pretty good reasons not to be convinced. We didn't just sit around scratching ourselves and saying "Whaddaya think? Useful?" We knew just about how much performance improvement you could get from the things, and apparently somebody decided (quite rightly, I'd say) that (as Carl suggests) they weren't worth the price of incompatability.