On 2月5日, 下午6時01分, Kevin Millikin <[email protected]> wrote: > On ARM, we prefer compact code and to a lesser extent, quick compilation. > For that reason, we do not inline as aggressively as on other platforms, > making the register allocation benefit less dramatic.
Hi, I understand the provisions for code size on ARM; it's a very reasonable priority on that target. Still, I believe the benefits of register allocation can also be found there too. Currently, V8 generates multiple individual loads/stores for each compiled statement. This creates a lot of code size that could be shaved off by implementing register allocation for ARM. As for the problem of callee-save/restore overhead, ARM can save/ restore almost all registers using STM/LDM, which is just two instructions; not much to worry in terms of code size. So I was wondering if the V8 developers can reconsider register allocation on ARM? It should be better for both code size and performance. :) -- v8-users mailing list [email protected] http://groups.google.com/group/v8-users
