Re: [RFC PATCH] Detecting lifetime-dse issues via Valgrind

[Alexander Monakov <amonakov@ispras.ru>]

[-- Attachment #1: Type: text/plain, Size: 1273 bytes --]


On Mon, 13 Nov 2023, Richard Biener wrote:

> > Ideally we'd position it such that more locals are put in SSA form,
> > but not too late to miss some UB, right? Perhaps after first pass_ccp?
> 
> I guess it’s worth experimenting.  Even doing it right before RTL expansion
> might work.  Note if you pick ccp you have to use a separate place for -O0

While Daniil is experimenting with this, I want to raise my concern about
attempting this instrumentation too late. Consider the main thing we are
trying to catch:

	// inlined operator new:
	this->foo = 42;
	// inlined constructor:
	*this = { CLOBBER };
	// caller:
	int tmp = this->foo;
	return tmp;

Our instrumentation adds

	__valgrind_make_mem_undefined(this, sizeof *this);

immediately after the clobber.

I am concerned that if GCC ever learns to leave out the following access
to 'this->foo', leaving tmp uninitialized, we will end up with:

	this->foo = 42;
	*this = { CLOBBER };
	__valgrind_make_mem_undefined(this, sizeof *this);
	int tmp(D);
	return tmp(D); // uninitialized

and Valgrind will not report anything since the invalid load is optimized out.

With early instrumentation such optimization is not going to happen, since the
builtin may modify *this.

Is my concern reasonable?

Thanks.
Alexander