[llvm-bugs] [Bug 153991] Missed optimization: fold x*a == b -> x == b/a

[LLVM Bugs via llvm-bugs]

Issue	153991
Summary	Missed optimization: fold x*a == b -> x == b/a
Labels	new issue
Assignees
Reporter	zxt5

    ```
define i1 @src(i64 %arg0, ptr %arg1) {
 %v0 = getelementptr inbounds nuw float, ptr %arg1, i64 %arg0
  %v1 = load float, ptr %v0, align 4
  %v2 = fmul nsz float %v1, 5.000000e-01
  %v3 = fcmp nsz olt float %v2, 8.000000e+00
  ret i1 %v3
}

define i1 @tgt(i64 %arg0, ptr %arg1) {
  %v0 = getelementptr inbounds nuw float, ptr %arg1, i64 %arg0
  %v1 = load float, ptr %v0, align 4
  %v2 = fcmp nsz olt float %v1, 1.600000e+01
  ret i1 %v2
}
```

Alive2: https://alive2.llvm.org/ce/z/LcL7YT
Godbolt: https://godbolt.org/z/WPMbTh7GP

Pattern found in https://github.com/dtcxzyw/llvm-opt-benchmark/blob/main/bench/ffmpeg/optimized/af_arnndn.ll

_______________________________________________
llvm-bugs mailing list
llvm-bugs@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs