[LLVMbugs] [Bug 23077] New: LICM seems unnecessarily slow
bugzilla-daemon at llvm.org
bugzilla-daemon at llvm.org
Mon Mar 30 15:42:15 PDT 2015
https://llvm.org/bugs/show_bug.cgi?id=23077
Bug ID: 23077
Summary: LICM seems unnecessarily slow
Product: libraries
Version: trunk
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: Scalar Optimizations
Assignee: unassignedbugs at nondot.org
Reporter: tobias at grosser.es
CC: llvmbugs at cs.uiuc.edu
Classification: Unclassified
Created attachment 14121
--> https://llvm.org/bugs/attachment.cgi?id=14121&action=edit
Test case
This simple file takes with 'opt -basicaa -licm out.opt.ll' 130ms to run. This
is suprisingly slow.
If I comment out the following lines:
// If there are any call sites in the alias set, add them to this AST.
- for (unsigned i = 0, e = AS.UnknownInsts.size(); i != e; ++i)
- add(AS.UnknownInsts[i]);
+ //for (unsigned i = 0, e = AS.UnknownInsts.size(); i != e; ++i)
+ // add(AS.UnknownInsts[i]);
Compile time is reduced to 20ms.
Surisingly only a single test case fails:
Transforms/LICM/PR21582.ll
I do not yet understand well what these Unknown instructions are suppose to be
used for, but I somehow have the feeling there is an obvious inefficiency here.
If I dump the AST:
+++ b/lib/Transforms/Scalar/LICM.cpp
@@ -200,6 +200,7 @@ bool LICM::runOnLoop(Loop *L, LPPassManager &LPM) {
delete InnerAST;
LoopToAliasSetMap.erase(InnerL);
}
+ CurAST->dump();
I get(after some delay) the following output:
Alias Set Tracker: 128 alias sets for 0 pointer values.
AliasSet[0x454de10, 1] may alias, Mod/Ref
1 Unknown instructions: void <badref>
AliasSet[0x454de60, 1] may alias, Mod/Ref
1 Unknown instructions: void <badref>
AliasSet[0x454deb0, 1] may alias, Mod/Ref
1 Unknown instructions: void <badref>
AliasSet[0x454df20, 1] may alias, Mod/Ref
1 Unknown instructions: void <badref>
AliasSet[0x454df90, 1] may alias, Mod/Ref
1 Unknown instructions: void <badref>
AliasSet[0x454e000, 1] may alias, Mod/Ref
1 Unknown instructions: void <badref>
...
Do we really need 128 alias sets here?
--
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/20150330/a92e18e4/attachment.html>
More information about the llvm-bugs
mailing list