llvm dev - Jun 2008 - [LLVMdev] Instruction does not dominate all uses

Hi all,

I'm getting a weird error if I run llc on the attached module.
Did I do something wrong (llc says the module is broken, also
attached) or is this a bug?

Cheers,
Gary

-- 
http://gbenson.net/
-------------- next part --------------
A non-text attachment was scrubbed...
Name: test.bc
Type: application/octet-stream
Size: 2060 bytes
Desc: not available
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20080617/dc2fc531/attachment.obj>
-------------- next part --------------
Instruction does not dominate all uses!
	%constants10 = load [8 x i8]** %43		; <[8 x i8]*> [#uses=3]
	getelementptr [8 x i8]* %constants10, i32 0, i32 16		; <i8*>:61 [#uses=1]
Instruction does not dominate all uses!
	getelementptr [8 x i8]* %constants10, i32 0, i32 16		; <i8*>:61 [#uses=1]
	bitcast i8* %61 to [8 x i8]**		; <[8 x i8]**>:62 [#uses=1]
Instruction does not dominate all uses!
	bitcast i8* %61 to [8 x i8]**		; <[8 x i8]**>:62 [#uses=1]
	%cache = load [8 x i8]** %62		; <[8 x i8]*> [#uses=1]
Instruction does not dominate all uses!
	%cache = load [8 x i8]** %62		; <[8 x i8]*> [#uses=1]
	ptrtoint [8 x i8]* %cache to i32		; <i32>:63 [#uses=1]
Instruction does not dominate all uses!
	ptrtoint [8 x i8]* %cache to i32		; <i32>:63 [#uses=1]
	add i32 %63, 1808		; <i32>:64 [#uses=1]
Instruction does not dominate all uses!
	add i32 %63, 1808		; <i32>:64 [#uses=1]
	inttoptr i32 %64 to [16 x i8]*		; <[16 x i8]*>:65 [#uses=2]
Instruction does not dominate all uses!
	inttoptr i32 %64 to [16 x i8]*		; <[16 x i8]*>:65 [#uses=2]
	getelementptr [16 x i8]* %65, i32 0, i32 0		; <i8*>:66 [#uses=1]
Instruction does not dominate all uses!
	getelementptr [16 x i8]* %65, i32 0, i32 0		; <i8*>:66 [#uses=1]
	bitcast i8* %66 to i32*		; <i32*>:67 [#uses=1]
Instruction does not dominate all uses!
	bitcast i8* %66 to i32*		; <i32*>:67 [#uses=1]
	load i32* %67		; <i32>:68 [#uses=1]
Instruction does not dominate all uses!
	load i32* %67		; <i32>:68 [#uses=1]
	lshr i32 %68, 24		; <i32>:69 [#uses=1]
Instruction does not dominate all uses!
	lshr i32 %68, 24		; <i32>:69 [#uses=1]
	and i32 %69, 255		; <i32>:70 [#uses=1]
Instruction does not dominate all uses!
	and i32 %69, 255		; <i32>:70 [#uses=1]
	icmp eq i32 %70, 182		; <i1>:71 [#uses=1]
Instruction does not dominate all uses!
	icmp eq i32 %70, 182		; <i1>:71 [#uses=1]
	br i1 %71, label %resolved14, label %not_resolved
Broken module found, compilation aborted!
llc[0x109ef868]
llc[0x109efa38]
[0x100364]
llc[0x10101390]
/lib/libc.so.6(abort+0x240)[0xfe588c0]
llc[0x1096a4ac]
llc[0x1097a2c4]
llc(_ZN4llvm13FPPassManager13runOnFunctionERNS_8FunctionE+0x1c0)[0x109223e8]
llc(_ZN4llvm23FunctionPassManagerImpl3runERNS_8FunctionE+0xa4)[0x109227ac]
llc(_ZN4llvm19FunctionPassManager3runERNS_8FunctionE+0xc8)[0x109229ec]
llc(main+0xbf4)[0x1031ae18]
/lib/libc.so.6[0xfe3e56c]
/lib/libc.so.6[0xfe3e73c]

On Tue, Jun 17, 2008 at 8:29 AM, Gary Benson <gbenson at redhat.com>
wrote:
> Hi all,
>
> I'm getting a weird error if I run llc on the attached module.
> Did I do something wrong (llc says the module is broken, also
> attached) or is this a bug?
The module appears to be malformed.  Essentially, what the error
message means is that there is some path through the function that
reaches a use of a variable without passing through the definition,
which is illegal in SSA form.

http://llvm.org/docs/tutorial/LangImpl7.html might be useful for
understanding SSA form, and how to generate it easily.

-Eli

Gary Benson wrote:
> Hi all,
>
> I'm getting a weird error if I run llc on the attached module.
> Did I do something wrong (llc says the module is broken, also
> attached) or is this a bug?
>   
Are you running your own optimization passes, or just the standard LLVM 
passes?

If it's your own optimization pass, there might be a bug.  Otherwise, 
you might be running into a bug that I found last week in the standard 
LLVM optimization passes.  That bug has since been fixed 
(http://llvm.org/bugs/show_bug.cgi?id=2434); if you update your LLVM 
tree from SVN mainline and rebuild, it should fix the problem.

-- John T.
> Cheers,
> Gary
>
> --
> http://gbenson.net/
>