Hi, all.
I am generating LLVM IR code for x64 platform using LLVM 64bit libraries ,
and for some reason I need to move to 32bit.
But after building 32bit libraries I found that IR code generates
incorrectly and it causes segfault.
For example, instead of generating function call it generates function phi.
64bit libs:
init: ; preds = %pre_entry
call void @Init()
br label %entry
32bit libs:
init:
phi void [ @Init(), <null operand!> ]
br label entry
Or other example. Instead of this:
lpad: ; preds = %entry
%lpad1 = landingpad { i8*, i32 } personality i32 (...)*
@__gxx_personality_v0
catch i8* bitcast (i8** @_ZTIPKc to i8*)
%exn = extractvalue { i8*, i32 } %lpad1, 0
<------
store i8* %exn, i8** %exn.slot
%ehselector = extractvalue { i8*, i32 } %lpad1, 1
store i32 %ehselector, i32* %ehselector.slot
br label %catch
catch: ; preds = %lpad
%exn2 = load i8** %exn.slot
%1 = call i8* @__cxa_begin_catch(i8* %exn2)
call void (i8*, ...)* @_printf(i8* %1)
<--------
br label %ret
I get this code:
lpad: ; preds = %entry
%lpad1 = landingpad { i8*, i32 } personality i32 (...)*
@__gxx_personality_v0
catch i8* bitcast (i8** @_ZTIPKc to i8*)
%exn = shufflevector { i8*, i32 } %lpad1
<---------
store i8* %exn, i8** %exn.slot
%ehselector = shufflevector { i8*, i32 } %lpad1
store i32 %ehselector, i32* %ehselector.slot
br label %catch
catch: ; preds = %lpad
%exn2 = load i8** %exn.slot
%1 = phi i8* [ %exn2, <null operand!> ], [ @__cxa_begin_catch,
Segmentation fault (core dumped)
Anyone can tell me what is it? Is it a bug of LLVM or our codegen?
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20140627/68dcf2db/attachment.html>