[Bug] [opt] Fix compilation crash when there is a container statement after an unconditional continue #1299
Conversation
… after an unconditional continue
| location = (int)statements.size() - 1; | ||
| location = (int)statements.size(); |
There was a problem hiding this comment.
(I wonder why this has been in the codebase for 5 months...)
So actually it should be
if a:
continue
if a:
if a:
continueinstead of
if a:
if a:
continue
if a:
continuewhich causes ControlFlowGraph::unreachable_code_elimination to produce malformed IR.
There was a problem hiding this comment.
Thanks for fixing this! It's surprising that we have this issue for so long time without noticing...
Codecov Report
@@ Coverage Diff @@
## master #1299 +/- ##
=======================================
Coverage 85.59% 85.59%
=======================================
Files 18 18
Lines 3283 3283
Branches 621 621
=======================================
Hits 2810 2810
Misses 345 345
Partials 128 128 Continue to review full report at Codecov.
|
| location = (int)statements.size() - 1; | ||
| location = (int)statements.size(); |
There was a problem hiding this comment.
Thanks for fixing this! It's surprising that we have this issue for so long time without noticing...
… after an unconditional continue (taichi-dev#1299) * [Bug] [opt] Fix compilation crash when there is a container statement after an unconditional continue * [skip ci] enforce code format * retrigger CI * retrigger CI Co-authored-by: Taichi Gardener <[email protected]>
Related issue = fix #1297
I moved the unreachable code elimination pass from
cfg_optimizationtosimplifybecause the former can't eliminate container statements properly. I tried writing something likestd::vector<Stmt *> not_in_any_nodesin the CFG, but it's hard to know whether aContinueStmtcan be eliminated in the CFG, so anIfStmtwith only oneContinueStmtin it can't be eliminated easily too.[Click here for the format server]