https://llvm.org/bugs/show_bug.cgi?id=23845
Bug ID: 23845
Summary: Apparent miscompilation with BBVectorize and sqrt
Product: new-bugs
Version: 3.6
Hardware: PC
OS: Windows NT
Status: NEW
Severity: normal
Priority: P
Component: new bugs
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected]
Classification: Unclassified
Created attachment 14470
--> https://llvm.org/bugs/attachment.cgi?id=14470&action=edit
unoptimised IR
In the unoptimised IR there are two calls to wombat326, each of which does a
sqrt. (see wobble211)
In the optimised code there is only one sqrt call.
One sqrt call seems to have been erroneously removed.
Disabling BBVectorize seems to remove the miscompilation.
Optimisation passes are basically the standard -03, with vectorisation passes
enabled.
Platform: LLVM 3.6, windows 64-bit. Using LLVM for jitting.
Unoptimised IR:
https://gist.github.com/anonymous/55f29c2b165efd66f0a3#file-gistfile1-ll-L98
Optimised IR: https://gist.github.com/anonymous/5029a142fcc946d708a8
--
You are receiving this mail because:
You are on the CC list for the bug.
_______________________________________________
LLVMbugs mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/llvmbugs
