[llvmg++] Code is generated for empty classes

[lattner]

Bugzilla Link	137
Resolution	FIXED
Resolved on	Jun 01, 2009 22:47
Version	1.0
OS	All
Depends On	llvm/llvm-bugzilla-archive#156
CC	@lattner

Extended Description

This testcase should not generate any stores in the function body:

---

struct X {};
X foo() { return X(); }

It currently generates:

---

    %struct.X = type { ubyte }

void %_Z3foov(%struct.X* %agg.result) {
entry:
%tmp = getelementptr %struct.X* %agg.result, long 0, ubyte 0
store ubyte 0, ubyte* %tmp
ret void
}

The type is correct (empty classes in C++ have one byte), but the store is not
needed. Empty classes come up a lot in C++ libraries, like the STL, these
stores are getting in the way.

Unfortunately this will require some changes in the front-end, so this will
probably happen after 1.1.

-Chris

[lattner]

Here's another case where we generate (in this case lots of) pointless code:

struct foo {
int array[100];
void bar();
};

void test() {
foo().bar();
}

[lattner]

Believe it or not, this bug isn't fixable until Bug 156 is.

-Chris

[lattner]

Changing all of these bugs who do not have people looking at them to be assigned
to "unassignedbugs", indicating that if someone is feeling ambitious, they can
take ownership of the bug.

If I stole your bug, and you still want it, feel free to take ownership back.

-Chris

[lattner]

Changing all of these bugs who do not have people looking at them to be assigned
to "unassignedbugs", indicating that if someone is feeling ambitious, they can
take ownership of the bug.

If I stole your bug, and you still want it, feel free to take ownership back.

-Chris

[lattner]

retargeting... probably won't be fixed for 1.2

[llvmbot]

FYI, head version of llvm-gcc4 still produces:

define void @​_Z3foov(%struct.X* %agg.result) {
entry:
%tmp = getelementptr %struct.X* %agg.result, i32 0, i32 0
; <i8*> [#uses=1]
store i8 0, i8* %tmp
ret void
}

Are we sure we care about this bug?

[lattner]

Yes, we care about it. That store is dead, and shouldn't exist.

-Chris

[llvmbot]

With TOT, we generate this now with optimization level > 0:

$ cat t.cpp
struct X {};
X foo() { return X(); }

$ llvm-g++ -o - -S -O1 t.cpp -emit-llvm
; ModuleID = 't.cpp'
target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:128:128"
target triple = "i386-apple-darwin9.7"

define i8 @​_Z3foov() nounwind readnone {
entry:
ret i8 0
}

Should we mark as fixed? Or should the -O0 case be handled as well?

[lattner]

We still generate useless code, but I think this is a failing of g++, and we're not going to fix it. For example, at -O0 this generates a useless i8 store:

struct x {};
void foo(x *P) {
*P = x();
}

Closing as fixed.

[llvmbot]

mentioned in issue llvm/llvm-bugzilla-archive#156