| Issue |
76128
|
| Summary |
[InstCombine] Missing optimization: fold `div(v, a) * b + rem(v, a)` to `div(v, a) * (b - a) + v`
|
| Labels |
new issue
|
| Assignees |
|
| Reporter |
XChy
|
Alive2 proof: https://alive2.llvm.org/ce/z/7SnGmc
### Description:
```llvm
define i32 @src(i32 noundef %val) {
entry:
%div = udiv i32 %val, 10
%shl = shl i32 %div, 4
%rem = urem i32 %val, 10
%add = or disjoint i32 %shl, %rem
ret i32 %add
}
```
can be folded to:
```llvm
define i32 @tgt(i32 noundef %val) {
entry:
%div = udiv i32 %val, 10
%reass.mul = mul nuw i32 %div, 6
%i = add i32 %reass.mul, %val
ret i32 %i
}
```
This is a simple example. `or disjoint` can be replaced with `add nsw nuw`, `shl 4` can be replaced with equivalent `mul 16`, and constants here can be replaced with other simple constants: https://alive2.llvm.org/ce/z/qmYLyJ
I don't replace constants with arguments because I haven't figure out completetly what condition is needed for this fold yet.
### Real-world motivation
This snippet of IR is derived from [linux/lib/bcd.c](https://github.com/torvalds/linux/blob/master/lib/bcd.c#L11C17-L11C17) (after O3 pipeline).
The example above is a reduced version. If you're interested in the original suboptimal IR and optimal IR, see also:https://godbolt.org/z/hra9qYbY7
**Let me know if you can confirm that it's an optimization opportunity, thanks.**
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
