Localized Register Spilling - GNU Project - Free Software Foundation (FSF)

Bernd Schmidt has contributed code to implement localized register spilling for GCC. Bernd's code complements the major improvements Joern Rennecke has made to reload inheritance over the last few months. Together, they can greatly decrease the costs associated with spilling registers due to the compiler running out of machine registers.

GCC's register allocators attempt to pack pseudo registers, which are potentially unbounded in number, into the register set provided by a particular target machine. Pseudo registers which have non-overlapping lifetimes may be allocated to the same machine register. Furthermore, each pseudo has exactly one home, either a machine register or a stack slot.

If the register spilling/reloading pass of the compiler needs a particular machine register for a reload, then pseudos allocated that that machine register may need to be spilled (ie, deallocated from that machine register and either reassigned to another free register or moved into the stack).

Previously, GCC would spill every pseudo which has allocated to that machine register. Even though only one such pseudo could ever hold a live value at the location where the reload was needed.

With Bernd's work, GCC is able to only spill the pseudo that is live at the point where the reload is needed. This significantly reduces the number of pseudos which get spilled into the stack and as a result can significantly improve code for register poor machines such as the x86.

One side effect of Bernd's work is that incorrect asms which explicitly clobber a register that is also mentioned in the inputs/outputs for the asm will always generate an error on all ports. This is going to cause linux kernels to fail to build on x86 processors until the linux kernel developers fix the asms in the linux kernel.

Joern's reload inheritance improvements are designed to optimize away redundant loads, stores, copies and address computations that are created due to register reloading and spilling. Thus, Bernd's patches reduce the amount of spilling and reloading that is performed and Joern's patches try to optimize any spills and reloads that Bernd's patches could not avoid.

Please send FSF & GNU inquiries & questions to gnu@gnu.org. There are also other ways to contact the FSF.

These pages are maintained by the GCC team.

For questions related to the use of GCC, please consult these web pages and the GCC manuals. If that fails, the gcc-help@gcc.gnu.org mailing list might help.
Please send comments on these web pages and the development of GCC to our developer mailing list at gcc@gnu.org or gcc@gcc.gnu.org. All of our lists have public archives.

Verbatim copying and distribution of this entire article is permitted in any medium, provided this notice is preserved.

Last modified 2006-06-21