From mboxrd@z Thu Jan 1 00:00:00 1970 Return-path: Received: from mx1.redhat.com ([209.132.183.28]:33441 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752667AbaCaLTY (ORCPT ); Mon, 31 Mar 2014 07:19:24 -0400 Message-ID: <53394F23.6080805@redhat.com> (sfid-20140331_131946_703598_3B2083C5) Date: Mon, 31 Mar 2014 13:18:59 +0200 From: Denys Vlasenko MIME-Version: 1.0 To: Arend van Spriel , Franky Lin , Hante Meuleman , "John W. Linville" , linux-kernel@vger.kernel.org, linux-wireless@vger.kernel.org Subject: Re: [PATCH 1/4] brcm80211: deinline brcmf_chip_cr4_enterdl, save 440 bytes References: <1396215079-7541-1-git-send-email-dvlasenk@redhat.com> <53391B89.7010304@broadcom.com> In-Reply-To: <53391B89.7010304@broadcom.com> Content-Type: text/plain; charset=UTF-8 Sender: linux-wireless-owner@vger.kernel.org List-ID: On 03/31/2014 09:38 AM, Arend van Spriel wrote: > On 30/03/14 23:31, Denys Vlasenko wrote: >> Automated script discovered that without forced inlining, >> gcc-4.7 generates smaller code for this function. >> >> There is no need to declare static functions inline anyway: >> nowadays gcc detects single-callsite static functions >> which benefit from inlining. > > These patches look awfully familiar. I tend to object, but I don't know the details of this automated script. The script removes "static" keyword, recompiles the .c file, compares the sizes, and if code size went down, creates a patch > How about execution time or is this only compile tested? The change adds one pair of call/return instructions - probably around 5-10 CPU cycles. The function in question is a part of firmware download logic, which is nowhere near being hot path/. > The other thing is that you seem to rely on a specific gcc version. > What about pre-4.7? How about different architectures. > Was this determined on x86, arm, sparc, mips. > All these questions make me say 'nay'. Not making functions inline unless there is a good reason is a general good coding practice. It is not a compiler- or architecture-specific optimization.