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 107A338233DE; Thu,  1 Sep 2022 00:33:14 +0000 (GMT)
DKIM-Filter: OpenDKIM Filter v2.11.0 sourceware.org 107A338233DE
DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gcc.gnu.org;
	s=default; t=1661992394;
	bh=0Mi8/E9hdmJZj9ZQXS9yvlL2RSl5Dwi/XUcnCGw1Hx4=;
	h=From:To:Subject:Date:In-Reply-To:References:From;
	b=WyYsNVahwDEnmBvwEcM/kqtgqU8/J2V/FBuNs/Q4GvURobRbqvEOZ8fjZWlNGSZ9u
	 ZP+6/3TZximuZC8DuAf4bnOSw6l598SWQTVd4lmGjX4vZiU23en5w3PZjwsGLre422
	 KldMiBZ8B4LEzzqd9SSsZ9y6Oz9WswCu9Xo0F2Yg=
From: "georgepee at gmail dot com" <gcc-bugzilla@gcc.gnu.org>
To: gcc-bugs@gcc.gnu.org
Subject: [Bug target/106763] Armv8.2 vmov.f16 instruction sometimes causes
 SIGILL
Date: Thu, 01 Sep 2022 00:33:13 +0000
X-Bugzilla-Reason: CC
X-Bugzilla-Type: changed
X-Bugzilla-Watch-Reason: None
X-Bugzilla-Product: gcc
X-Bugzilla-Component: target
X-Bugzilla-Version: 12.2.1
X-Bugzilla-Keywords: 
X-Bugzilla-Severity: normal
X-Bugzilla-Who: georgepee at gmail dot com
X-Bugzilla-Status: RESOLVED
X-Bugzilla-Resolution: INVALID
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-106763-4-gYcskqMUqZ@http.gcc.gnu.org/bugzilla/>
In-Reply-To: <bug-106763-4@http.gcc.gnu.org/bugzilla/>
References: <bug-106763-4@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
List-Id: <gcc-bugs.sourceware.org>

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=3D106763

--- Comment #9 from George Pee <georgepee at gmail dot com> ---
Thank you for following up even after I closed the ticket.

Unfortunately, I'm unable to switch to a 64-bit kernel at the moment.

Using this works around the issue by treating it via a neon path and enabli=
ng
the vfp bit and retrying the instruction.

@@ -824,6 +824,9 @@ call_fpe:
        .align  6

 .LCneon_arm_opcodes:
+       .word   0xee000000                      @ mask
+       .word   0xee000000                      @ opcode
+
        .word   0xfe000000                      @ mask
        .word   0xf2000000                      @ opcode



I am now using this simplified case, which fails somewhere between 100,000 =
and
1,000,000 iterations:

    int c =3D 0;
    while(1)
    {
        c++;
        asm (=20
            "vmov.f16    r6, s18\n"
        );
        if (c % 100 =3D=3D 0)
            printf("%d\n",c);
    }

It's quite odd that it is intermittent.  After instrumenting vfp enable/dis=
able
in the kernel, it seems as though there is something disabling the vfp bit =
in
the fpexc register, but it doesn't seem to be the kernel.

I am able to reproduce this with other FP16 instructions, but not other
non-FP16 VFP instructions.=