Hi team, I've identified a potential optimization opportunity to eliminate concat calls for constant strings and replace them directly with a LEAQ straddr instruction, similar to what the Java JIT does.
This is a cheap optimization that can enable further optimizations down the line, such as compile-time evaluation of len(const_string).
A local run of ./make.bash shows 200+ potential sites for this optimization.
Here is an example:
func ff(a, b int) string {
y := "abc"
t := y + "def"
if t == "abcdef" {
return t + "ghi"
}
return t
}before:
after applying opts:
For implementation, we could either
- add a dedicated stringopt pass before expand_call (which centralizes the code for future optimizations like strcmp) or
- integrate it into the sccp pass after expand_call (which uses its framework but disperses the code) or
- add some generic rules
Do you have a preference on the approach? Would the community be interested in this?
Hi team, I've identified a potential optimization opportunity to eliminate concat calls for constant strings and replace them directly with a LEAQ straddr instruction, similar to what the Java JIT does.
This is a cheap optimization that can enable further optimizations down the line, such as compile-time evaluation of len(const_string).
A local run of .
/make.bashshows 200+ potential sites for this optimization.Here is an example:
before:
after applying opts:
For implementation, we could either
Do you have a preference on the approach? Would the community be interested in this?