| Issue |
52870
|
| Summary |
Optimization: Prefer 32-bit integers over 64-bit if they are in range
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
easyaspi314
|
If LLVM knows that an integer value can fit in a 32-bit integer, it should probably do that.
On 32-bit targets it avoids expensive multiword arithmetic, and on x86_64, it can significantly reduce the number of REX prefixes resulting in smaller codegen.
For example:
```c
#include <stdint.h>
void side_effect();
void foo()
{
for (int64_t i = 0; i < 999; i++) {
side_effect();
}
}
void bar()
{
for (int32_t i = 0; i < 999; i++) {
side_effect();
}
}
```
Clang x86 codegen: https://godbolt.org/z/T8fo4Yare
GCC x86 codegen: https://godbolt.org/z/rsWx7Mv6M - note that GCC eliminates the 64-bit arithmetic
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
