[llvm] Adding change to IndVarSimplify pass to optimize IVs stuck in trivial vector operations (PR #122248)

[via llvm-commits]

anilavakundu007 wrote:

> Can you share an IR sample before and after the pass, so it's easier to understand what it does?

Before IndVarSimplify Pass (in the -O2 pipeline): 

``` 
 %vecinit = insertelement <2 x i32> poison, i32 %u_xlati59.06, i64 0
 %vecinit1 = shufflevector <2 x i32> %vecinit, <2 x i32> poison, <2 x i32> zeroinitializer
 %add = add <2 x i32> %vecinit1, <i32 -1, i32 1>
 %0 = extractelement <2 x i32> %add, i64 0
 %add3 = add nsw i32 %add345, %0
 %1 = extractelement <2 x i32> %add, i64 1
 %cmp = icmp sgt i32 %1, 2
 ```
 After my patch in IndVarSimplify Pass: 
 
 ``` 
%vecinit = insertelement <2 x i32> poison, i32 %u_xlati59.06, i64 0                                                                 
%vecinit1 = shufflevector <2 x i32> %vecinit, <2 x i32> poison, <2 x i32> zeroinitializer                                            
%add = add <2 x i32> %vecinit1, <i32 -1, i32 1>                                                                                      
%0 = extractelement <2 x i32> %add, i64 0                                                                                            
%add3 = add nsw i32 %add345, %0                                                                                                      
%1 = add nsw i32 %u_xlati59.06, 1                                                                                                    
%cmp = icmp sgt i32 %1, 2 
```
Here the: 
```
%1 = extractelement <2 x i32> %add, i64 1 
%add = add <2 x i32> %vecinit1, <i32 -1, i32 1>
``` 
is broken to have a scalar bin op instruction: 
```
%1 = add nsw i32 %u_xlati59.06, 1 
```


https://github.com/llvm/llvm-project/pull/122248