| Issue |
129615
|
| Summary |
Optimize chains of `memcmp`
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
Kmeakin
|
Often in C/C++, since there is no way to `switch` on strings, code will have big if-else chains of (the moral equivalent of) `memcp(s, "some string") == 0`. For example, [LLVM's own lexer](https://github.com/llvm/llvm-project/blob/main/llvm/lib/AsmParser/LLLexer.cpp#L493) compares every identifier against a few hundred different string literals to recognize keywords.
It would be nice if LLVM could optimize such chains into a more efficient method, eg a trie:
```c++
auto src(std::string_view s) {
if (s == "aaa") return 1;
if (s == "abb") return 2;
if (s == "bbb") return 3;
if (s == "baa") return 4;
...
return 0;
}
auto tgt(std::string_view s) {
if (s[0] == 'a') {
if (s[1] == 'a' && s[2] == 'a') return 1;
if (s[1] == 'b' && s[2] == 'b') return 2;
}
if (s[0] == 'b') {
if (s[1] == 'a' && s[2] == 'a') return 3;
if (s[1] == 'b' && s[2] == 'b') return 4;
}
...
return 0;
}
```
![]()
_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
