| Issue |
138465
|
| Summary |
Optimization pass to improve performance of interpreters, and other recursive programs
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
the-ssd
|
Currently, LLVM generates an additional jump for interpreters, which could be inclined for better performance. (which is how interpreters written in Assembly like LuaJIT's interpreter work).
Example:
```llvm
define i32 @example() {
dispatch:
; dispatch logic
br {instruction}
instruction1:
; instruction logic
br label %dispatch
instruction2:
; instruction logic
br label %dispatch
}
```
But a faster version inlines logic from dispatch into instructions, and usually runs faster.
```llvm
define i32 @example() {
; dispatch logic
br {instruction}
instruction1:
; instruction logic
; dispatch logic
br {instruction}
instruction2:
; instruction logic
; dispatch logic
br {instruction}
}
```
But this is also a small optimization for a small amount of programs, and maybe better suited for a plugin.
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
