[llvm-branch-commits] [llvm] release/21.x: [SelectOptimize] Fix incorrect -1 immediate for large integers (#170860) (PR #171596)

[Yingwei Zheng via llvm-branch-commits]

dtcxzyw wrote:

> ; CHECK-NEXT: [[CMP1_FROZEN:%.*]] = freeze i1 [[CMP1]]
              ^
<stdin>:569:32: note: scanning from here
 %sext1 = sext i1 %cmp1 to i128
                               ^
<stdin>:569:32: note: with "CMP1" equal to "%cmp1"
 %sext1 = sext i1 %cmp1 to i128
                               ^
<stdin>:570:2: note: possible intended match here
 %add = add i128 %sext1, 0
 ^

Input file: <stdin>
Check file: /home/gha/actions-runner/_work/llvm-project/llvm-project/llvm/test/CodeGen/AArch64/selectopt-cast.ll

-dump-input=help explains the following input dump.

Input was:
<<<<<<
            .
            .
            .
          564:  br label %loop 
          565:  
          566: loop: ; preds = %loop, %entry 
          567:  %load = load i128, ptr %a, align 16 
          568:  %cmp1 = icmp slt i128 %load, 0 
          569:  %sext1 = sext i1 %cmp1 to i128 
next:915'0                                    X error: no match found
next:915'1                                      with "CMP1" equal to "%cmp1"
          570:  %add = add i128 %sext1, 0 
next:915'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~
next:915'2      ?                          possible intended match
          571:  %cmp2 = icmp slt i128 %add, 0 
next:915'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          572:  br i1 %cmp2, label %loop, label %exit 
next:915'0     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
          573:  
next:915'0     ~
          574: exit: ; preds = %loop 
next:915'0     ~~~~~~~~~~~~~~~~~~~~~~
          575:  ret void 
next:915'0     ~~~~~~~~~~
          576: } 
next:915'0     ~~
>>>>>>

--

Looks like this transformation doesn't work on 21.x?

https://github.com/llvm/llvm-project/pull/171596