internal error when running JuMP's tests #28444
Labels
bug
Indicates an unexpected problem or unintended behavior
compiler:optimizer
Optimization passes (mostly in base/compiler/ssair/)
Comments
JeffBezanson
added
the
compiler:optimizer
|
This would happen if a use is discovered to be unused during the pass. I think it would be correct just to ignore the statement in that case: and similarly for the loop before that. Does need a properly reduced test case though. |
|
This is a fun one. There's a second bug here that makes this transformation apply even though it's not legal in the JuMP example. After fixing that, the following simple example reproduces the reported behavior: (doesn't fail on master though, because of the same legality check bug working in the opposite direction). |
Keno
added a commit
that referenced
this issue
Aug 6, 2018
The legality check was using use counts after `finish(compact)` got to delete, which made them inaccurate. Instead, take a copy of the use counts before. Additionally, ignore any uses that got deleted during `finish(compact)`. Fixes #28444.
Keno
added a commit
that referenced
this issue
Aug 6, 2018
The legality check was using use counts after `finish(compact)` got to delete, which made them inaccurate. Instead, take a copy of the use counts before. Additionally, ignore any uses that got deleted during `finish(compact)`. Fixes #28444.
Keno
added a commit
that referenced
this issue
Jan 5, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce.
Keno
added a commit
that referenced
this issue
Jan 5, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce.
Keno
added a commit
that referenced
this issue
Jan 6, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce.
KristofferC
pushed a commit
that referenced
this issue
Jan 9, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Jan 11, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Feb 4, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Feb 4, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Feb 11, 2019
The legality check was using use counts after `finish(compact)` got to delete, which made them inaccurate. Instead, take a copy of the use counts before. Additionally, ignore any uses that got deleted during `finish(compact)`. Fixes #28444.
KristofferC
pushed a commit
that referenced
this issue
Feb 11, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Feb 11, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Apr 20, 2019
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
KristofferC
pushed a commit
that referenced
this issue
Feb 20, 2020
PR #28478 moved the computation of the use counts before the finish call. to fix #28444. However, the early parts of the finish call fixes up phi node arguments, which fail to get counted if we look at use counts before that fixup is performed. This causes #30594 where the only non-trivial use is on the backedge of the phi and would thus incorrectly fail to get accounted for. Fix that by taking the use count after phi fixup but before dce. (cherry picked from commit f8f2045)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
When running JuMP's tests (current
master, commitcd6457b6c0c82de339b30bbd747a3f1bac135b8b), the following internal error is printed:This happens locally and also on travis (https://travis-ci.org/JuliaOpt/JuMP.jl/jobs/412077553#L796). The error appears to be ignored, it doesn't crash julia. (The tests proceed to fail at a later point.)
The text was updated successfully, but these errors were encountered: