[cfe-dev] [RFC] Setting dereferenceable flag on the implicit this parameter for non-static member functions

Hubert Tong via cfe-dev cfe-dev at lists.llvm.org
Wed Dec 6 15:09:07 PST 2017


I am not sure that there is a requirement that *this remains a valid object.
Experimenting with the -O3 behaviour of the following program (with a
powerpc64le-unknown-linux-gnu target), it seems the dereferenceable
attribute is scope based?

int a[1], b[1], *cp;
void dobadstuff() { delete cp; }

__attribute__((__noinline__))
void foo(int * __restrict__ a, int * __restrict__ b, int &c, int n) {
  dobadstuff();
  for (int i = 0; i < n; ++i)
    if (a[i] > 0)
      a[i] = c * b[i];
}

int main(void) {
  cp = new int;
  foo(a, b, *cp, 1);
}

Valgrind reports an invalid read.


On Wed, Dec 6, 2017 at 8:06 AM, Hal Finkel via cfe-dev <
cfe-dev at lists.llvm.org> wrote:

>
> On 12/05/2017 01:47 PM, Lei Huang via cfe-dev wrote:
>
>
> Hello,
>
> In the discussion on bugzilla 30729, it is mentioned that the 'this' pointer
> needs to be valid upon entry to a non-static method.  Does the standard
> guarantee this is non-null on entry?
>
>
> Yes. You have to call a non-static member function on a valid object.
>
>  If so, is there a reason we can't use that fact to mark 'this' as '
> dereferenceable(sizeof(*this))'?
>
>
> Yes, this seems like a good idea.
>
>  -Hal
>
>
> There are LICM optimizations we can do based on the knowledge that 'this' is
> non-null on entry to a non-static member function.
>
> eg.  For the following IR,  the two highlighted loads are not being
> hoisted out of the for loop because we are not able to guarantee that the
> pointer is non-null.  If the 'this' pointer is guaranteed to be non-null
> on entry and we mark it thus,  then the 2 loads within the for-loop body
> can then be hoisted out into the loop preheader.
>
> This is of course just one example of an optimization we could perform
> based on this knowledge, but there are probably a number of others (i.e.
> anything that relies on a pointer being 'dereferenceable(N)').
>
> *$ cat a.ll*
> target datalayout = "e-m:e-i64:64-n32:64"
> target triple = "powerpc64le-unknown-linux-gnu"
> %struct.S = type { <4 x i32>, <4 x i32> }
> ; Function Attrs: norecurse nounwind readonly
> define <4 x i32> @_ZNK1S20constShouldBeHoistedEmDv4_i(%struct.S*
> nocapture readonly %this, i64 %n, <4 x
> i32> %x) align 2 {
> entry:
>  %tobool9 = icmp eq i64 %n, 0
>  br i1 %tobool9, label %for.end, label %for.body.lr.ph
> for.body.lr.ph:                                   ; preds = %entry
>  %k1 = getelementptr inbounds %struct.S, %struct.S* %this, i64 0, i32 0
>  %k2 = getelementptr inbounds %struct.S, %struct.S* %this, i64 0, i32 1
>  br label %for.body
> for.body:                                         ; preds = %
> for.body.lr.ph, %if.end
>  %n.addr.011 = phi i64 [ %n, %for.body.lr.ph ], [ %div, %if.end ]
>  %x.addr.010 = phi <4 x i32> [ %x, %for.body.lr.ph ], [ %x.addr.1,
> %if.end ]
>  %rem = and i64 %n.addr.011, 15
>  %cmp = icmp eq i64 %rem, 0
>  br i1 %cmp, label %if.end, label %if.then
> if.then:                                          ; preds = %for.body
>  *%0 = load <4 x i32>, <4 x i32>* %k1, align 16*
>  %add = add <4 x i32> %0, %x.addr.010
>  *%1 = load <4 x i32>, <4 x i32>* %k2, align 16*
>  %xor = xor <4 x i32> %add, %1
>  br label %if.end
> if.end:                                           ; preds = %for.body,
> %if.then
>  %x.addr.1 = phi <4 x i32> [ %xor, %if.then ], [ %x.addr.010, %for.body ]
>  %div = lshr i64 %n.addr.011, 4
>  %tobool = icmp eq i64 %div, 0
>  br i1 %tobool, label %for.end, label %for.body
> for.end:                                          ; preds = %if.end, %entry
>  %x.addr.0.lcssa = phi <4 x i32> [ %x, %entry ], [ %x.addr.1, %if.end ]
>  ret <4 x i32> %x.addr.0.lcssa
> }
>
>
>
> Regards,
> Lei Huang
>
>
> LLVM Development on POWER
> Internal mail: C2/YGK/8200/MKM
> Phone: (905) 413-4419
> TieLine: 969-4419
> E-mail: lei at ca.ibm.com
>
>
>
> _______________________________________________
> cfe-dev mailing listcfe-dev at lists.llvm.orghttp://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>
>
> --
> Hal Finkel
> Lead, Compiler Technology and Programming Languages
> Leadership Computing Facility
> Argonne National Laboratory
>
>
> _______________________________________________
> cfe-dev mailing list
> cfe-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-dev
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20171206/41978970/attachment-0001.html>


More information about the cfe-dev mailing list