From owner-sc22wg14+sc22wg14-domo2=www.open-std.org@open-std.org Sat Oct 26 00:18:37 2013 Return-Path: X-Original-To: sc22wg14-domo2 Delivered-To: sc22wg14-domo2@www.open-std.org Received: by www.open-std.org (Postfix, from userid 521) id 81188358233; Sat, 26 Oct 2013 00:18:37 +0200 (CEST) Delivered-To: sc22wg14@open-std.org Received: from outbound-queue-2.mail.thdo.gradwell.net (outbound-queue-2.mail.thdo.gradwell.net [212.11.70.35]) by www.open-std.org (Postfix) with ESMTP id 39819358230 for ; Sat, 26 Oct 2013 00:18:22 +0200 (CEST) Received: from outbound-edge-1.mail.thdo.gradwell.net (bonnie.gradwell.net [212.11.70.2]) by outbound-queue-2.mail.thdo.gradwell.net (Postfix) with ESMTP id 51F1E2310B for ; Fri, 25 Oct 2013 23:18:21 +0100 (BST) Received: from digraph.polyomino.org.uk (HELO digraph.polyomino.org.uk) (81.187.227.50) (smtp-auth username postmaster%pop3.polyomino.org.uk, mechanism cram-md5) by outbound-edge-1.mail.thdo.gradwell.net (qpsmtpd/0.83) with (AES256-SHA encrypted) ESMTPSA; Fri, 25 Oct 2013 23:18:21 +0100 Received: from jsm28 (helo=localhost) by digraph.polyomino.org.uk with local-esmtp (Exim 4.76) (envelope-from ) id 1VZpi0-0005WK-7N for sc22wg14@open-std.org; Fri, 25 Oct 2013 22:18:20 +0000 Date: Fri, 25 Oct 2013 22:18:20 +0000 (UTC) From: "Joseph S. Myers" X-X-Sender: jsm28@digraph.polyomino.org.uk To: sc22wg14@open-std.org Subject: Atomic compound assignment and exact floating-point underflow Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII X-Gradwell-MongoId: 526aee2d.7a01-50c-1 X-Gradwell-Auth-Method: mailbox X-Gradwell-Auth-Credentials: postmaster@pop3.polyomino.org.uk Sender: owner-sc22wg14@open-std.org Precedence: bulk The following is really an issue to be considered for the floating-point TS part 5 rather than an issue with C11 as it stands. C11 gives a sequence of feholdexcept / atomic_compare_exchange_strong / feclearexcept / feupdateenv intended to be equivalent to atomic compound assignment when floating point is involved. The semantics differ from those of a true atomic read-modify-write, however, where a floating-point value is computed that is tiny but exact, where the underflow exception is signaled but the corresponding flag is not raised; in that case, feupdateenv will not signal the exception at all. This difference is detectable if an alternate exception handling attribute is in effect that permits exact underflow to be detected. Thus, if the TS part 5 provides C bindings to such alternate exception handling, it will need to address the code sequence for compound assignment in footnote 113. -- Joseph S. Myers joseph@codesourcery.com