Re: Caller/Callee saved Registers

>Anthony L Kimball writes......


>Contra M. Beusterien, I feel it is an architectural issue in as
>much as it is a problem most appropriately solved by throwing
>hardware at it: The obvious architectural solution seems to be for
>the caller to set a use mask, for the callee to set another
>use mask, and for the processor to do a save of the intersection,
>with special-purpose logic. Anything like this out there?




The study of architectural support to reduce RSR traffic across function
calls is presented in the following paper:


Architectural Support for Reduced Register Saving/Restoring in
Single-Window Register Files. Miquel Huguet and Tomas Lang. ACM
Transactions on Computer Systems. Volume 9, No 1, Feb 1991.


A more detailed study can be found in Huguet's PhD thesis.


Architectural and Compiler Support for Efficient Function Calls. Ph.D.
Dissertation. Computer Science Dept, Univ. of California at Los Angeles,
Sep 1989.


They study the impact of hardware support (use masks) and compiler
optimizations, individually, and also when put together. The experimental
results are very interesting (difficult to summarize them here, though!).


Rakesh Ghiya.
School of Computer Science,
McGill University,
Montreal, Canada.
--

Return to the comp.compilers page.
Search the comp.compilers archives again.