[llvm-bugs] [Bug 46651] New: Fix all opt tests to work under NPM

via llvm-bugs llvm-bugs at lists.llvm.org
Wed Jul 8 15:33:07 PDT 2020


https://bugs.llvm.org/show_bug.cgi?id=46651

            Bug ID: 46651
           Summary: Fix all opt tests to work under NPM
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: aeubanks at google.com
                CC: alina.sbirlea at gmail.com, hans at chromium.org,
                    htmldeveloper at gmail.com, llvm-bugs at lists.llvm.org
            Blocks: 46649

Many check-llvm tests use something like "opt -foo-pass". Previously this would
always use the legacy pass manager. I've added the "enable-new-pm" opt flag to
use the new pass manager instead even for "-foo-pass". (previously you would
use "-passes=foo" to use NPM) An alternative is to add the corresponding
"-passes=foo" RUN line for a test that has "-foo-pass", but I'd rather not
touch that many tests, and I think maintaining compatibility between the two is
a better guarantee of feature parity. And we'd still end up doing most of the
remaining work anyway.

To repro, set the "enable-new-pm" flag in llvm/tools/opt/opt.cpp to true and
run check-llvm. (maybe there should be a CMake flag for this? I considered
adding this onto -DENABLE_EXPERIMENTAL_NEW_PASS_MANAGER=ON but that would break
check-llvm for people already using that flag for Clang.)

Right now there are around ~1600 check-llvm failures on Linux with
"enable-new-pm" set to true.
Some of these are pass names being inconsistent between the legacy pass manager
and NPM.
Some of these are opt features that the NPM opt path doesn't handle.
Some of these are real bugs in NPM passes.


Referenced Bugs:

https://bugs.llvm.org/show_bug.cgi?id=46649
[Bug 46649] Change default pass manager to new pass manager
-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20200708/80cf8566/attachment.html>


More information about the llvm-bugs mailing list