From gcarr@ucar.edu Thu Jun 3 08:21:51 2004 Received: from tablemtn.cgd.ucar.edu (tablemtn.cgd.ucar.edu [128.117.21.12]) by echorock.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i53ELptT023960 for ; Thu, 3 Jun 2004 08:21:51 -0600 Received: from [192.168.0.7] (vpn19.ucar.edu [128.117.65.19]) by tablemtn.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i53EMFOn026066 for ; Thu, 3 Jun 2004 08:22:15 -0600 (MDT) Mime-Version: 1.0 X-Sender: gcarr@mailhub.cgd.ucar.edu (Unverified) Message-Id: Date: Thu, 3 Jun 2004 08:22:57 -0600 To: Ccsm-x1@echorock.cgd.ucar.edu From: George R Carr Jr Content-Type: text/plain; charset="us-ascii" ; format="flowed" Subject: [Ccsm-x1] Status X-BeenThere: ccsm-x1@mailman.cgd.ucar.edu X-Mailman-Version: 2.1 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 03 Jun 2004 14:21:51 -0000 OK. If this gets out it means that the email list is now running. Please use it to communicate to the group (includes NCAR, ORNL, and Cray recipients). Note: The ccsm3_0_rel02 tag has a problem that should be fixed by tomorrow if all goes well. I am prepared to begin a validation run on the x1 as soon as we determine we should start. Should we wait for rel03 to start things going? The script to collect the runtime statistics got sidelined but should be completed today or tomorrow. We can run with any of the recent tags to begin the load balancing exercise. Until the script is ready I can look directly at the various files directly on Phoenix if you let me know where to look. Thanks. -- =========================== George R Carr Jr UCAR/NCAR/CGD 303-497-1764 From gcarr@ucar.edu Thu Jun 3 14:46:37 2004 Received: from tablemtn.cgd.ucar.edu (tablemtn.cgd.ucar.edu [128.117.21.12]) by echorock.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i53KkbtT028366 for ; Thu, 3 Jun 2004 14:46:37 -0600 Received: from [192.168.0.7] (vpn18.ucar.edu [128.117.65.18]) by tablemtn.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i53Kl1On001344 for ; Thu, 3 Jun 2004 14:47:01 -0600 (MDT) Mime-Version: 1.0 X-Sender: gcarr@mailhub.cgd.ucar.edu (Unverified) Message-Id: In-Reply-To: <40BF674D.6030506@cray.com> References: <40BF674D.6030506@cray.com> Date: Thu, 3 Jun 2004 14:47:44 -0600 To: Ccsm-x1@echorock.cgd.ucar.edu From: George R Carr Jr Subject: Re: [Ccsm-x1] Status Content-Type: text/plain; charset="us-ascii" ; format="flowed" X-BeenThere: ccsm-x1@mailman.cgd.ucar.edu X-Mailman-Version: 2.1 Precedence: list List-Id: CCSM: Cray X1 Activities List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 03 Jun 2004 20:46:38 -0000 >George R Carr Jr wrote: > >>OK. If this gets out it means that the email list is now running. >>Please use it to communicate >>to the group (includes NCAR, ORNL, and Cray recipients). > > >Just wanted to let you know that I did get this. > >> >> >>Note: The ccsm3_0_rel02 tag has a problem that should be fixed by >>tomorrow if all >>goes well. >> >>I am prepared to begin a validation run on the x1 as soon as we >>determine we should >>start. Should we wait for rel03 to start things going? > >I think Matt hasn't gotten the last version to work again since CUG. >He's debugging >now. > >> >>The script to collect the runtime statistics got sidelined but >>should be completed >>today or tomorrow. We can run with any of the recent tags to begin >>the load balancing >>exercise. Until the script is ready I can look directly at the >>various files directly on >>Phoenix if you let me know where to look. >> >>Thanks. Let me know if there is something I can do to help. Thanks. -- =========================== George R Carr Jr UCAR/NCAR/CGD 303-497-1764 From gcarr@ucar.edu Fri Aug 13 09:08:46 2004 Received: from tablemtn.cgd.ucar.edu (tablemtn.cgd.ucar.edu [128.117.21.12]) by echorock.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i7DF8ko1019708 for ; Fri, 13 Aug 2004 09:08:46 -0600 Received: from [128.117.22.180] (hasui.cgd.ucar.edu [128.117.22.180]) by tablemtn.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i7DF9F6I010522 for ; Fri, 13 Aug 2004 09:09:15 -0600 (MDT) Mime-Version: 1.0 X-Sender: gcarr@mailhub.cgd.ucar.edu (Unverified) Message-Id: In-Reply-To: References: Date: Fri, 13 Aug 2004 09:09:54 -0600 To: Ccsm-x1@echorock.cgd.ucar.edu From: George R Carr Jr Content-Type: text/plain; charset="us-ascii" ; format="flowed" Subject: [Ccsm-x1] Re: X-BeenThere: ccsm-x1@mailman.cgd.ucar.edu X-Mailman-Version: 2.1 Precedence: list Reply-To: "CCSM: Cray X1 Activities" List-Id: CCSM: Cray X1 Activities List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 13 Aug 2004 15:08:46 -0000 At 6:40 PM -0700 8/12/04, Matthew Cordery wrote: >Ah, the loops aren't the problem. This is the problem: > > enforce = .false. > if (present(enforce_mask) .and. enforce_mask) enforce = .true. > if (present(enforce_all ) .and. enforce_all ) enforce = .true. > >I've seen this sort of thing before on another compiler. The problem is >that things like enforce_mask and enforce_all are optional arguments. >The problem is that if, say, enforce_mask does not exist, the >present(enforce_mask) is false. However, the compiler also has to test the >second part of the if statement and if the optional argument does not >exist then you get a memory fault. The proper logic is: > >if( present( enforce_mask ) )then > if( enforce_mask ) enforce = .true. >end if > >Apparently, the IBM compiler makes an assumption for you here about >whether or not you need to evaluate the rest of the if statement. It's >debatable whether or not this is the correct thing to do or not. In any >event, I'll flag it for the compiler tech people and see what they think. > I've sent off this email together with your two previous emails to get them looked at. I'll let you know what we find. Thanks. -- =========================== George R Carr Jr UCAR/NCAR/CGD 303-497-1764 From gcarr@ucar.edu Fri Aug 13 11:58:43 2004 Received: from tablemtn.cgd.ucar.edu (tablemtn.cgd.ucar.edu [128.117.21.12]) by echorock.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i7DHwho1021732 for ; Fri, 13 Aug 2004 11:58:43 -0600 Received: from [128.117.22.180] (hasui.cgd.ucar.edu [128.117.22.180]) by tablemtn.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i7DHxC6I021867 for ; Fri, 13 Aug 2004 11:59:12 -0600 (MDT) Mime-Version: 1.0 X-Sender: gcarr@mailhub.cgd.ucar.edu (Unverified) Message-Id: Date: Fri, 13 Aug 2004 11:59:52 -0600 To: Ccsm-x1@echorock.cgd.ucar.edu From: George R Carr Jr Content-Type: text/plain; charset="us-ascii" ; format="flowed" Subject: [Ccsm-x1] For your performance tests X-BeenThere: ccsm-x1@mailman.cgd.ucar.edu X-Mailman-Version: 2.1 Precedence: list Reply-To: "CCSM: Cray X1 Activities" List-Id: CCSM: Cray X1 Activities List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 13 Aug 2004 17:58:43 -0000 Matthew/Ilene, An idea that might help you when looking at the performance of a single component. While working on pop (or other) component, if you reduce the number of processors for that component, you will reach the threshhold where it is now the performance limitation. At this point profiling it will be more useful, because the mpi wait times should be minimized. George -- =========================== George R Carr Jr UCAR/NCAR/CGD 303-497-1764 From gcarr@ucar.edu Fri Aug 13 13:39:57 2004 Received: from tablemtn.cgd.ucar.edu (tablemtn.cgd.ucar.edu [128.117.21.12]) by echorock.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i7DJdvo1023108 for ; Fri, 13 Aug 2004 13:39:57 -0600 Received: from [128.117.22.180] (hasui.cgd.ucar.edu [128.117.22.180]) by tablemtn.cgd.ucar.edu (8.12.8/8.12.8) with ESMTP id i7DJeQ6I028532 for ; Fri, 13 Aug 2004 13:40:27 -0600 (MDT) Mime-Version: 1.0 X-Sender: gcarr@mailhub.cgd.ucar.edu (Unverified) Message-Id: Date: Fri, 13 Aug 2004 13:41:06 -0600 To: Ccsm-x1@echorock.cgd.ucar.edu From: George R Carr Jr Content-Type: text/plain; charset="us-ascii" ; format="flowed" Subject: [Ccsm-x1] Re: X-BeenThere: ccsm-x1@mailman.cgd.ucar.edu X-Mailman-Version: 2.1 Precedence: list Reply-To: "CCSM: Cray X1 Activities" List-Id: CCSM: Cray X1 Activities List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Fri, 13 Aug 2004 19:39:57 -0000 At 6:40 PM -0700 8/12/04, Matthew Cordery wrote: >Ah, the loops aren't the problem. This is the problem: > > enforce = .false. > if (present(enforce_mask) .and. enforce_mask) enforce = .true. > if (present(enforce_all ) .and. enforce_all ) enforce = .true. > >I've seen this sort of thing before on another compiler. The problem is >that things like enforce_mask and enforce_all are optional arguments. >The problem is that if, say, enforce_mask does not exist, the >present(enforce_mask) is false. However, the compiler also has to test the >second part of the if statement and if the optional argument does not >exist then you get a memory fault. The proper logic is: > >if( present( enforce_mask ) )then > if( enforce_mask ) enforce = .true. >end if > >Apparently, the IBM compiler makes an assumption for you here about >whether or not you need to evaluate the rest of the if statement. It's >debatable whether or not this is the correct thing to do or not. In any >event, I'll flag it for the compiler tech people and see what they think. > I've sent off this email together with your two previous emails to get them looked at. Brian Kauffman believes your analysis is correct. He will add it to our list of fixes to test and roll into the next patch. In the meantime, we should add this to our local baseline. Matthew, would you put together a "patched" couple file??? Thanks! -- =========================== George R Carr Jr UCAR/NCAR/CGD 303-497-1764