================
@@ -292,6 +292,11 @@ bool AddSubMulHelper(InterpState &S, CodePtr OpPC,
unsigned Bits, const T &LHS,
// If for some reason evaluation continues, use the truncated results.
S.Stk.push<T>(Result);
+ // If wrapping is enabled, the new value is fine.
+ // FIXME: Pass this to the add/sub/mul op instead?
----------------
tbaederr wrote:
I was mainly thinking about performance implications.
For
```c++
constexpr bool inc2() {
unsigned a = 0;
for (unsigned i = 0; i != 1'000'000; ++i) {
++a;++a;++a;++a;
}
return true;
}
constexpr bool inc() {
for (unsigned i = 0; i != 1'0; ++i) {
inc2();
}
return true;
}
static_assert(inc());
```
Before:
```
Benchmark 1: bin/clang++ -c -std=c++23 ./inc.cpp -fconstexpr-steps=0
Time (mean ± σ): 3.764 s ± 0.076 s [User: 3.732 s, System: 0.015 s]
Range (min … max): 3.705 s … 4.050 s 30 runs
```
After:
```
Benchmark 1: bin/clang++ -c -std=c++23 ./inc.cpp -fconstexpr-steps=0
Time (mean ± σ): 3.828 s ± 0.115 s [User: 3.796 s, System: 0.015 s]
Range (min … max): 3.729 s … 4.337 s 30 runs
```
That is just a quick check though, I haven't measured what impact passing it as
a flag would have.
https://github.com/llvm/llvm-project/pull/183761
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
