From: Tom Rini Date: Thu, 27 Feb 2014 14:51:25 -0500 Subject: [U-Boot] [PATCH 0/4] Big acceleration of Kbuild performance In-Reply-To: <1393511304-9432-1-git-send-email-yamada.m@jp.panasonic.com> References: <1393511304-9432-1-git-send-email-yamada.m@jp.panasonic.com> Message-ID: <20140227195125.GO16805@bill-the-cat> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: u-boot@lists.denx.de On Thu, Feb 27, 2014 at 11:28:20PM +0900, Masahiro Yamada wrote: > Kbuild brought about many advantages for us but > a significant performance regression was reported by > Simon Glass. > > After some discussions and analysis, it turned out > its main cause is in $(call cc-option,...). > > Historically, U-Boot parses all config.mk > (arch/*/config.mk and board/*/config.mk) > every time descending into subdirectories. > That means cc-options are evaluated over and over again. > > $(call cc-option,...) is useful but costly. > So we want to evaluate them only in ./Makefile > and spl/Makefile and export compiler flags. > > 1/4 through 3/4 are preparation for 4/4. > > 4/4 is what we really want to do. > By applying it, I think Kbuild will get much faster. > > Simon, > Could you post the result of "time" command? > I want to know how fast Kbuild will be on your computer. Locally, I see am335x_evm down to about 9 seconds, and it was 11 seconds or so before Kbuild and 23s on master. I think this speed up over pre-Kbuild is quite real as we've gone from ~2000 cc-option execs to 90. I'd really love to see some run-time sanity check from the Tegra folks, even tho I did see Stephen ack that patch. Thanks for getting this done so quick! -- Tom -------------- next part -------------- A non-text attachment was scrubbed... Name: not available Type: application/pgp-signature Size: 836 bytes Desc: Digital signature URL: