[flang-commits] [flang] [mlir] Reland "[flang][OpenMP] Fix lowering of LINEAR iteration variables (#188851)" (PR #194623)

[Leandro Lupori via flang-commits]

luporl wrote:

OpenMP V&V test results, with and without this PR applied:
`  .f90 pass rate: 68/129 (52%) [ 53 build failures ]`
The issues reported in #189959 don't occur.

Regarding Fujitsu testsuite, my test environment had some issues previously, which caused some OpenMP tests to fail with and without this PR applied, leading to the conclusion that there were no regressions.
These issues were fixed and the results I get now mostly match those in Linaro's CI (http://54.172.246.49:9090/jobs/tcwg_flang_test--main-aarch64-O0-debug-build/builds/10160/archive/artifacts/03-test_llvm/). There are a few C and C++ tests and 3 Fortran tests that fail only on my end. The Fortran tests seem to be flaky.

These are the results of an April 21 run from Linaro's CI:
```
Total Discovered Tests: 92831
  Passed            : 92664 (99.82%)
  Failed            :   103 (0.11%)
  Executable Missing:    64 (0.07%)
```

Local results before applying this PR:
```
Total Discovered Tests: 92831
  Passed            : 92652 (99.81%)
  Failed            :   113 (0.12%)
  Executable Missing:    66 (0.07%)
```
Local results after applying this PR:
```
Total Discovered Tests: 92831
  Passed            : 92651 (99.81%)
  Failed            :   114 (0.12%)
  Executable Missing:    66 (0.07%)
```
Failed tests difference, between results before and after applying this PR:
```
<   test-suite :: Fujitsu/Fortran/0660/Fujitsu-Fortran-0660_0097.test
>   test-suite :: Fujitsu/Fortran/1052/Fujitsu-Fortran-1052_0108.test
>   test-suite :: Fujitsu/Fortran/1052/Fujitsu-Fortran-1052_0112.test
```

0660_0097 seems to be flaky. 1052_0108 and 1052_0112 are tagged as flaky in Linaro's CI and they pass when run manually.

https://github.com/llvm/llvm-project/pull/194623