From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-bugzilla@gcc.gnu.org>
Received: by sourceware.org (Postfix, from userid 48)
 id A47E83851C29; Tue, 23 Mar 2021 14:08:16 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org A47E83851C29
From: "kretz at kde dot org" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug c++/99728] code pessimization when using wrapper classes around
 SIMD types
Date: Tue, 23 Mar 2021 14:08:16 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: c++
X-Bugzilla-Version: 10.2.1
X-Bugzilla-Keywords: missed-optimization
X-Bugzilla-Severity: normal
X-Bugzilla-Who: kretz at kde dot org
X-Bugzilla-Status: UNCONFIRMED
X-Bugzilla-Resolution: 
X-Bugzilla-Priority: P3
X-Bugzilla-Assigned-To: unassigned at gcc dot gnu.org
X-Bugzilla-Target-Milestone: ---
X-Bugzilla-Flags: 
X-Bugzilla-Changed-Fields: 
Message-ID: <bug-99728-4-vnTt4IXbj0@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-99728-4@http.gcc.gnu.org/bugzilla/>
References: <bug-99728-4@http.gcc.gnu.org/bugzilla/>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
X-Bugzilla-URL: http://gcc.gnu.org/bugzilla/
Auto-Submitted: auto-generated
MIME-Version: 1.0
X-BeenThere: gcc-bugs@gcc.gnu.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: Gcc-bugs mailing list <gcc-bugs.gcc.gnu.org>
List-Unsubscribe: <https://gcc.gnu.org/mailman/options/gcc-bugs>,
 <mailto:gcc-bugs-request@gcc.gnu.org?subject=unsubscribe>
List-Archive: <https://gcc.gnu.org/pipermail/gcc-bugs/>
List-Post: <mailto:gcc-bugs@gcc.gnu.org>
List-Help: <mailto:gcc-bugs-request@gcc.gnu.org?subject=help>
List-Subscribe: <https://gcc.gnu.org/mailman/listinfo/gcc-bugs>,
 <mailto:gcc-bugs-request@gcc.gnu.org?subject=subscribe>
X-List-Received-Date: Tue, 23 Mar 2021 14:08:16 -0000

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D99728

--- Comment #4 from Matthias Kretz (Vir) <kretz at kde dot org> ---
FWIW, using std::experimental::native_simd<double> also does not hoist the
stores out of the loop. However, if you pass d by value and return d, the i=
ssue
goes away. So I guess this is an aliasing pessimization. Even though you ad=
ded
__restrict__. In any case __m256 has the problem that it is declared with t=
he
may_alias attribute. I recommend to just never use __m256 unless you have no
other choice. Note that `s0data_s<__m256d>` warns "ignoring attributes on
template argument '__m256d' [-Wignored-attributes]", meaning it drops the
may_alias attribute.

Reduced test case (https://godbolt.org/z/PW98Wsfoj):
#include <immintrin.h>

struct Tvsimple {
  __m256d v;
};

struct s0data_s {
  Tvsimple lam1, lam2;
};
struct s0data_s_intrin {
  __m256d lam1, lam2;
};

template <typename T>
void foo(T &__restrict__ d, size_t l, size_t lmax) {
  while (l <=3D lmax) {
    d.lam1 =3D d.lam2;
    l +=3D 2;
  }
}

// hoists load out of the loop but loops over the store
template void foo<>(s0data_s &__restrict__ d, size_t l, size_t lmax);

// turns loop into a single branch
template void foo<>(s0data_s_intrin &__restrict__ d, size_t l, size_t lmax)=
;=