Development BOINC 7.0.62


log in

Advanced search

Message boards : Number crunching : Development BOINC 7.0.62

Previous · 1 · 2 · 3 · 4 · 5 · Next
Author Message
Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1273527 - Posted: 20 Aug 2012, 21:58:52 UTC - in response to Message 1271264.

I had the same.
Over 30 computation errors in one day with some completing but most GPU apps error out.
I removed .33 and went back to .31 and not one error
even the tasks that downloaded while I was running .33 error out


As DA would say "I''m not seeing that".

I did however flush my cache of GPU work before I did the upgrade. Seti (opti) and Einstein (stock) cuda apps appear to running fine. Its been running for over a week. Do you have a <flops> entry in app_info? I don't.

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8760
Credit: 52,709,504
RAC: 25,984
United Kingdom
Message 1273547 - Posted: 20 Aug 2012, 23:44:48 UTC - in response to Message 1271264.

even the tasks that downloaded while I was running .33 error out

With what error number or description? Any still in your list?

Jord has posted the v7.0.34 change list for Mac OS X

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1273949 - Posted: 22 Aug 2012, 12:58:36 UTC
Last modified: 22 Aug 2012, 13:02:18 UTC

7.0.34 change log (Mac only)

- API, SCR: Switch from *.txf fonts to TrueType fonts in graphics applications, ensuring that all related files have no licensing issues.

- Mac: Update scripts for building branded installers to allow building with Xcode 4.3 under OS 10.7.x Lion; change names of branded installers from "*macOSX_universal" to "*macOSX_i686" since we no longer support PowerPC Macs.

- SS: Update project files to use FreeType and FTGL on Windows.

- Sample Graphics app: Switch from *.txf fonts to TrueType fonts.

- Mac: Update build instructions and build script.

- WIN: Make the example graphics application build again.

- WIN: Get rid of the linker warnings for both the default screensaver and the example graphics application

- API: remove support for BMP and TGA image files; the code to parse these had no license info. Only JPEG is supported now.

- client: fix error in runtime estimation for active tasks

- MGR: Enforce the 0..10 day limit on the connect interval for the advanced preferences dialog. At some point we should rename it.

- API: Remove ttfont.cpp from Makefile.am. Apps should include it in there Makefile as needed.

- WINSCR: Fix problem with the screensaver needlessly cycling in preview mode. Ugh.

- WIN: Remove reference to deleted tgalib.cpp file.

- Client/manager: there was a bug because some code was writing "cpu" in XML, and other code was looking for "CPU". To fix this and prevent similar problems, processor type names are now encapsulated in proc_type_name_xml(). Code should use this rather than having hard-wired names. Redefine: GPU_TYPE_* as macros that call proc_type_name_xml().

- MGR: Fix build break.

- client: when we're making a scheduler RPC for a reason other than work fetch, and we're deciding whether to piggyback a work request, skip the checks for hysteresis (buffer < min) and for per-resource backoff time. These checks are there only to limit the rate of RPCs, which is not relevant since we're doing one anyway. This fixes a bug where a project w/ sporadic jobs specifies a next_rpc_delay to ensure regular polling from clients. When these polls occur they should request work regardless of backoff.

- client: tweak to the above: never ask for work if buffer > max. This is needed to prevent projects that use next_rpc_delay from queuing unbounded work.

- client: treat all 4xx HTTP errors as permanent

- code cleanup

- WINSETUP: Provide the ability to create the acct_mgr_login.xml from the command line. Useful for mass deployments within companies and charities.

- Mac installer: Fixes for short (posix) user names containing spaces.

Profile Michael W.F. Miles
Avatar
Send message
Joined: 24 Mar 07
Posts: 249
Credit: 29,166,576
RAC: 7,615
Canada
Message 1274017 - Posted: 22 Aug 2012, 17:02:11 UTC - in response to Message 1273547.

No there is none still in list I reverted back to 7.0.31 and no problem.
I think that was a time exceeded error as they would take exactly the same time as other wu but when they finished they erred.

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8760
Credit: 52,709,504
RAC: 25,984
United Kingdom
Message 1274096 - Posted: 22 Aug 2012, 19:56:57 UTC - in response to Message 1274017.
Last modified: 22 Aug 2012, 20:00:29 UTC

No there is none still in list I reverted back to 7.0.31 and no problem.
I think that was a time exceeded error as they would take exactly the same time as other wu but when they finished they erred.

And, for the second time of asking, with what error number or description? Any still in your list?

Edit - I see you're getting

Exit status -1073741819 (0xffffffffc0000005) Unknown error number

with BOINC v7.0.31 - but I was more interested in what you had with v7.0.33: 'time exceeded' would match the warning I posted when .33 was first released.

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1286986 - Posted: 23 Sep 2012, 8:15:18 UTC
Last modified: 23 Sep 2012, 8:17:08 UTC

7.0.36 change log

- client: If OpenCL detection gets an error for a platform or device, finish detection of the remaining platforms and / or devices. Move OpenCL memory size workaround back into OpenCL source file.

- client: put some casts to double in NVIDIA detect code. Shouldn't make any difference.

- client: work around a nasty bug which crashes OSX (!) on some dual-gpu MacBooks with CUDA installed if we call cuInit() unless we force use of the discrete (NVIDIA) GPU.

- client: initialize memfree and memtotal before use for Nvidia cards. It appears that the Nvidia API was only setting 32-bits of the 64-bit value. The remaining 32-bits were whatever was on the stack.

- client: Don't support CUDA on dual-GPU MacBooks with automatic GPU switching but continue to support OpenCL for NVIDIA GPUs. This prevents forcing use of the discrete (NVIDIA) GPU, which greatly decreases time the computer can run on battery power.

- client: if a project has excluded GPUs of a given type, allow it to fetch work of that type if the # of runnable jobs it <= the # of non-excluded instances (rather than 0).

- client: fix bug that broke file signing with X.509 certificates. From matszpk. Fixes #1168.

- client: print log msgs (enabled by task_debug) if the client times out on quitting or aborting a task, and has to kill it.

- API, SCR: Switch from *.txf fonts to TrueType fonts in graphics applications, ensuring that all related files have no licensing issues.

- Mac: Update scripts for building branded installers to allow building with Xcode 4.3 under OS 10.7.x Lion; change names of branded installers from "*macOSX_universal" to "*macOSX_i686" since we no longer support PowerPC Macs.

- SS: Update project files to use FreeType and FTGL on Windows.

- Sample Graphics app: Switch from *.txf fonts to TrueType fonts.

- Mac: Update build instructions and build script.

- WIN: Make the example graphics application build again.

- WIN: Get rid of the linker warnings for both the default screensaver and the example graphics application

- API: remove support for BMP and TGA image files; the code to parse these had no license info. Only JPEG is supported now.

- client: fix error in runtime estimation for active tasks

- MGR: Enforce the 0..10 day limit on the connect interval for the advanced preferences dialog. At some point we should rename it.

- API: Remove ttfont.cpp from Makefile.am. Apps should include it in there Makefile as needed.

- WINSCR: Fix problem with the screensaver needlessly cycling in preview mode. Ugh.

- WIN: Remove reference to deleted tgalib.cpp file.

- Client/manager: there was a bug because some code was writing "cpu" in XML, and other code was looking for "CPU". To fix this and prevent similar problems, processor type names are now encapsulated in proc_type_name_xml(). Code should use this rather than having hard-wired names. Redefine: GPU_TYPE_* as macros that call proc_type_name_xml().

- MGR: Fix build break.

- client: when we're making a scheduler RPC for a reason other than work fetch, and we're deciding whether to piggyback a work request, skip the checks for hysteresis (buffer < min) and for per-resource backoff time. These checks are there only to limit the rate of RPCs, which is not relevant since we're doing one any. This fixes a bug where a project w/ sporadic jobs specifies a next_rpc_delay to ensure regular polling from clients. When these polls occur they should request work regardless of backoff.

- client: tweak to the above: never ask for work if buffer > max. This is needed to prevent projects that use next_rpc_delay from queuing unbounded work.

- client: treat all 4xx HTTP errors as permanent

- code cleanup

- WINSETUP: Provide the ability to create the acct_mgr_login.xml from the command line. Useful for mass deployments within companies and charities.

- Mac installer: Fixes for short (posix) user names containing spaces.

Profile Jesiros
Send message
Joined: 3 Aug 08
Posts: 50
Credit: 414,467
RAC: 705
Russia
Message 1289064 - Posted: 28 Sep 2012, 22:12:22 UTC

А вот мне интересно: кто же пишет сам BOINC-клиент? Ради бога, верните вкладку 'Сообщения' на место, где она была в версии 6.10.60. Ведь там же вся нужная техническая информация, крайне полезно. А вместо неё поставили какие-то 'Комментарии' что-ли. Совершенно бессмысленная и ненужная вещь. А для того чтобы посмотреть информацию нужно каждый раз лезть в меню мышкой или заучивать комбинацию клавиш. Неудобно, крайне неудобно. Только это и удерживает меня от перехода с 6.10.60 на новую версию.

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8760
Credit: 52,709,504
RAC: 25,984
United Kingdom
Message 1289066 - Posted: 28 Sep 2012, 22:16:03 UTC - in response to Message 1289064.

But I wonder: who writes himself BOINC-client? For God's sake, return the tab 'View' to the place where it was in version 6.10.60. After all, there is all the necessary technical information is very useful. And instead put some 'Comments', that is. Completely pointless and useless thing. And in order to see the information you need each time to go into the menu with the mouse or the keyboard shortcut to memorize. Uncomfortable, very uncomfortable. The only thing that keeps me from switching to the new version 6.10.60.

David Anderson

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1301251 - Posted: 2 Nov 2012, 13:09:24 UTC
Last modified: 2 Nov 2012, 13:10:29 UTC

7.0.38 change log

- client: Initialize the total mem size_t.

- client: enforce excluded app at startup.

- win compile fix.

- scheduler: fix bug that cause NCI to be sent even when client is not requesting work.

- client: parse the <vbox_window> option in cc_config.xml

- Client/manager: if a GPU app is suspended because a GPU-exclusive app is running, show an appropriate message.

- client: message tweak.

- A bunch of tweaks from Steffen Moller, e.g. using MAXPATHLEN.

- Mac: Fix build breaks.

- client: MAXPATHLEN tweak.

- Manager: fix "exclusive GPU app running" message logic.

- client and API: improve the way an app checks for the death of the client.

Old: heartbeat mechanism
Problem: if the client is blocked for > 30 secs (e.g. because it takes a long time to write the state file, or because it's stopped in a debugger) then apps exit. This is bad if the app doesn't checkpoint and has been running for a long time.

New: the client passes its PID to the app.
The app periodically (10 sec) checks that the process still exists.

Notes:
For backward compatibility (e.g. new API w/ old client, or vice versa) the client still sends heartbeats, and the API checks heartbeats if the client doesn't pass a PID. The new mechanism works only if the client's PID isn't assigned to a new process within 10 secs of the client exiting. Windows 2000 reuses PIDs immediately, so check for Win2K and don't use this mechanism if so.

TODO: For Unix multithread apps, critical sections aren't currently being enforced. Need to fix this by masking signals.


- Win compile fixes. For now, include psapi.lib in various project properties. Try to figure out a different way.

- MGR: We don't save Simple View's width & height since its window is not resizable, so don't try to read them back.

- lib: don't clear entire APP_VERSION struct in APP_VERSION::parse_coproc()

- MGR: Fix references to the taskbar events which are now included in the stock wxWidget library,

- MGR: Fix the logging class which changed for the new wxWidgets.

- MGR: Add missing references winspool.lib in the linker.

- Add missing references to psapi.lib for various build configurations.

- client: change work fetch policy to avoid starving GPUs in situations where GPU exclusions are used.

- client: fix bug in round-robin simulation when GPU exclusions are used.

Note: this fixes a major problem (starvation) with project-level GPU exclusion. However, project-level GPU exclusion interferes with most of the client's scheduling policies. E.g., round-robin simulation doesn't take GPU exclusion into account, and the resulting completion estimates and device shortfalls can be wrong by an order of magnitude.

The only way I can see to fix this would be to model each GPU instance as a separate resource, and to associate each job with a particular GPU instance. This would be a sweeping change in both client and server.

- SCR: Fix bug in X11 screensaver.

- Mac: Fix icons in Simple View Task and Project selection combo boxes.

- Mac: Fix Quit command in Simple View and a crash bug with Quit AppleEvent.

- WINSETUP: Add the account manager auth logic from the 6.8 installer to current generation installers.

- WINSETUP: Make whether or not to create start menu items configurable via the command line. Useful for mass deployment installs.

- Add missing files and fix project files.

- Slight tweaks to the installer. Installer tried picking up the wrong font file.

- Fix assert in Simple Preferences: use window foreground color for CTransparentStaticLine if skin does not specify <static_line_color>

- Fix assert in SkinManager when file referenced by <application_logo> tag is missing.

- Mac: Work around wxMemoryDC::Clear() bug in drawing Simple View background.

- Mac: Work around apparent bug in wxWidgets 2.9.4 which failed to render wxNoteBook pages.

- VBOX: Change the way elapsed_time is calculated in vboxwrapper.

Previously: elapsed_time was just incremented with the value of the polling period each iteraction through the main loop. This introduced issues when vboxmanage lagged for whatever reason. This lag could go as high as 5 seconds. Over the timespan of a day this could increase the wall clock time of a task a great deal.

Now: elapsed_time is incremented with the time it took to execute the main loop.

- VBOX: Increase elapsed_time if we were forced to sleep after executing the main loop.

- Mac: Work around apparent refresh bug in wxCarbon 2.9.4 wxGenericListCtrl

- Win: Fix rendering of CBOINCTaskCtrl buttons on Windows. On Windows with wxWidgets 2.9.4, buttons don't refresh properly unless they are children of the wxStaticBox, but on Mac the layout is wrong unless the buttons are children of the parent of the wxStaticBox.

- Fix more bad wxFlexGridSizers, etc. in Attach Wizard, Advanced Preferences. Fix assert caused by Clear() of empty Project Categories wxComboBox.

- Fix bad wxFlexGridSizer in CDlgGenericMessage.

- Fix assert due to missing wxOK calling SafeMessageBox() in ProjectInfoPage.

- VBOX: Remove timeouts for creating and deleting snapshots.

- client: fix bug in handling proxy info in cc_config.file.

- client: if exiting because of exit_after_finished flag, write state file before exiting so we don't restart the job later.

- client, Unix: make Curl sockets close-on-exec, so that app processes don't inherit them.

- client: fix bug that makes client exit if a slot dir contains finish file on startup.

- GUI RPC: expose TIME_STATS info (e.g. on_frac) in the binding of the get_state() RPC.

- client: move client_start_time and previous_uptime from CLIENT_STATE to TIME_STATS, so that these are also visible in GUI RPC

- scheduler RPC: move uptime and previous_uptime into <time_stats>

- client: condition an RR simulation message on <rrsim_detail>

- boinccmd: show TIME_STATS info in --get_state

VBOX: Don't report any error messages when unregistering the VM.

VirtualBox 4.2 no longer requires certain commands to be executed to unregister a VM while older versions do. Just ignore any error codes, if it becomes a problem we can always make it conditional on what version of VirtualBox is installed.

- VBOX: Make deleting a snapshot a non-terminal event.

Deleting stale snapshots now consists of enumerating the existing snapshots and deleting all but the most recent snapshot. This fixes the problem where, for one reason or another, a snapshot could not be deleted at one point in time. The wrapper would just continue to create new ones and was unable to successfully deal with the older ones. Disk space would eventually run out.

- client: message tweaks for failed app startup.

- Fix MinGW build: generate_svn_version.sh only works on top level.

- VBOX: Cleanup log a little bit.

- VBOX: Cleanup all the snapshots before unregistering the VM.

- VBOX: Make adjustments to the clean-up procedure for Vbox 4.2.

- client/manager: move the conditional define of MAXPATHLEN from boinc_win.h to filesys.h, so that the client will hopefully build on Hurd, which doesn't define MAXPATHLEN.

- client: in checking reasons for not requesting work, look at backoff last. Otherwise the user can get a misleading message if they update a project that's backed off.

Profile Khangollo
Avatar
Send message
Joined: 1 Aug 00
Posts: 245
Credit: 36,410,524
RAC: 0
Slovenia
Message 1301270 - Posted: 2 Nov 2012, 14:52:45 UTC
Last modified: 2 Nov 2012, 15:46:11 UTC

There's no 7.0.38 in official svn (boinc.berkeley.edu/svn/tags/). Has the repository changed?
Edit: Sorry, nevermind. I just noticed it moved to git.
____________

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8760
Credit: 52,709,504
RAC: 25,984
United Kingdom
Message 1301283 - Posted: 2 Nov 2012, 15:48:23 UTC - in response to Message 1301270.

There's no 7.0.38 in official svn (boinc.berkeley.edu/svn/tags/). Has the repository changed?

Yes, they're using GIT now - which brings its own problems.

v7.0.38 appears to work once installed, but there's an installer bug which fails to re-create the program launch icons - I doubt this build will even make it into official alpha testing.

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12393
Credit: 2,666,323
RAC: 958
Netherlands
Message 1301286 - Posted: 2 Nov 2012, 15:52:45 UTC - in response to Message 1301270.
Last modified: 2 Nov 2012, 15:52:53 UTC

In the mean time, use http://boinc.berkeley.edu/gitweb/?p=boinc.git;a=tags. :-)
____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

Profile Ageless
Avatar
Send message
Joined: 9 Jun 99
Posts: 12393
Credit: 2,666,323
RAC: 958
Netherlands
Message 1305340 - Posted: 12 Nov 2012, 12:11:10 UTC

CTRL+SHIFT+E works also on simple view now. So for the helpers, from 7.0.38 onwards there's no need to point a novice user to advanced view and leave him there afterwards.

Apropos, 7.0.39 is available for testing for Linux and Macintosh users.

* VBOX: Do not count the sleep time as part of the elapsed time when the VM is supposed to be suspended. * VBOX: Ugh, if sleep_time is greater than zero, increment elapsed_time by the poll interval. * Some C++ files in client had execute permissions (??). Clear them. * a bunch of skin files had execute permissions (??). Clear them. * remove execute permissions from various files that shouldn't have them. * Linux: add slide-show screensaver (from David Coss). * Mac: update Xcode project for building boinc_zip * Mac: add slide-show screensaver to example_app Xcode project and Mac example_app build scripts example app: fix compiler warnings in slide_show.cpp

____________
Jord

Fighting for the correct use of the apostrophe, together with Weird Al Yankovic

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1326048 - Posted: 9 Jan 2013, 13:58:51 UTC

7.0.40 change log

- XML parser: make low-level functions inline, in an attempt (unsuccessful, as far as I can tell) to boost performance

- Some C++ files in client had execute permissions (??). Clear them.

- a bunch of skin files had execute permissions (??). Clear them.

- remove execute permissions from various files that shouldn't have them

- client: fix bug where, when updating a project, we fail to request work even though higher-priority projects are marked as no-new-tasks or are otherwise ineligible for work fetch.

- build fixes for OpenBSD, from rustyBSD

- client: TIME_STATS fields go in <time_stats> part of state file

- compile fix for FreeBSD

- remove deprecated MAC address code

- client: if a project has 64 or more ready-to-report tasks, report them. 64 is chosen a bit arbitrarily, but the idea is to limit the number of tasks reported per RPC, and to accelerate the reporting of small tasks.

- client: if a project is anonymous platform, don't send list of alternate platforms in scheduler RPC.

- client: fix uninitialized vars in CPU feature detection (from RustyBSD)

- client: when reset a project, clear its scheduler URL list (force a reload of the master page)

- allow config.xml to be a symlink

- Mgr: Fix Windows hangs and simplify code for notices display: Handle protocol-relative URLs which hung Manager on Windows. RSS feeds and web pages may use protocol-relative (scheme-relative) URLs, such as <img src="//sample.com/test.jpg"/> Since the html comes from a web server via http, the scheme is assumed to also be http. But we have cached the html in a local file, so it is no longer associated with the http protocol / scheme. Therefore all our URLs must explicity specify the http protocol.

- The Manager would hang when trying to get data from web servers if there was no Internet connection. This happened on Windows only, not on Mac. (I don't have the hardware to test on Linux, but I suspect it is OK as Mac uses UNIX.) I fixed this by implementing asynchronous network I/O in CBOINCHtmlListBox on Windows.

- Eliminate need for CBOINCVListBox class and greatly simplify CBOINCHtmlListBox class.

- client: add "client app configuration" feature; see http://boinc.berkeley.edu/trac/wiki/ClientAppConfig This lets users do the following:
1) limit the number of concurrent jobs of a given app (e.g. for WCG apps that are I/O-intensive)
2) Specify the CPU and GPU usage parameters of GPU versions of a given app. Implementation notes: max app concurrency is enforced in 2 places:
a) when building the initial job run list
b) when enforcing the final job run list.
Both are needed to avoid possible starvation. however, we don't enforce it during RR simulation. Doing so could cause erroneous shortfall and work fetch. This means, however, that work buffering will not work as expected if you're using max concurrency.

- client: add the above to VS project

- client: bug fixes for app config feature

- client: add new files app_config.cpp,h to Xcode project.

- client (Win) preallocate large files to avoid file fragmentation when you write to them incrementally (workaround for shoddy filesystem design)

- lib: Win compile fix for boinc_allocate_file()

- Mgr: Fix some rare issues on Windows (including a possible memory leak) when switching to a client on a different computer after trying to display notices without an Internet connection.

- client: if <dont_check_file_sizes> is set, don't check file existence at startup

- Mac installer: add missing pclose() (from RustyBSD)

- OpenCL: Add definition of GPU_TYPE_INTEL to match definitions of GPU_TYPE_ATI and GPU_TYPE_NVIDIA.

- OpenCL: First pass at adding support for Intel Ivy Bridge GPUs. The following files still need updating: ProjectInfoPage.cpp,.h, ProjectListCtrl.cpp,.h

- lib: add size info to messages when realloc() fails in MFILE

- MGR: Fix two potential security issues with browser.cpp where the query to the cookie database could have been abused. At present neither of the two parameters originate as user input so using it as an attack vector isn't very high. Prevent the functions from being exploited in the future in case the routines were ever used in a different way.

- client: strcasestr doesn't exist on Windows.

- client: strcasestr does exist on Windows in lib/str_replace.cpp, but I just needed to #include "str_replace.h". This is preferable because it handles INTEL as well as Intel and intel.

- fix typo in GUI RPC

- OpenCL: Add peak FLOPS computation for Intel Ivy Bridge GPUs.

- Mac: remove deleted files from Xcode project

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1326050 - Posted: 9 Jan 2013, 14:00:30 UTC

7.0.41 change log

- Manager: compile fixes for OpenBSD. From RustyBSD.

- client (unix): rename() doesn't work between filesystems. If the user has set things up so that slots/ is a symlink to a different filesystem, things won't work when the client moves output files from the slot to project dir. Solution: if rename() fails, try system("mv ...") since mv works across filesystems

- Manager: Fix Mac bug introduced by recent OpenBSD compile fixes.

- MGR: Minor fix to the OpenCL output for Intel GPUs.

- client: Hook up the XML portion of the Intel GPU detection code so the server scheduler knows about it.

- client: Print out the peak flops for the Intel GPU, the regular OpenCL descriptions do not show peak flops. NOTE: At this point we should be supporting Intel GPUs as far as detection and reporting its presence to the server goes. I don't know about scheduling though.

- client/server: fix build breaks I introduced last night with a variable rename.

- client/server: Make sure the GPU Type field is really classified as an Intel GPU.

- client/manager: tweaks to Intel GPU code

- client: win compile fix

- client: add new file gpu_intel.cpp to Xcode project.

- client: replace ignore_nvidia_dev etc. with an array. PRINCIPLE: AVOID PER-GPU-TYPE VARIABLES get rid of alloca() stuff in gutil.cpp; almost certainly not needed

- don't include malloc.h; it doesn't exist on BSD systems

- minor code cleanup from RustyBSD

- client: clean up redundant and confusing GPU descriptions. Problems: COPROC_NVIDIA::description() is really description of CUDA GPUs. COPROC_ATI::description() is really description of CAL GPUs. On Windows and Linux, AMD OpenCL and CAL always coexist,and NVIDIA OpenCL and CUDA always coexist. But on the Mac, AMD OpenCL always exists without CAL and NVIDIA OpenCL often exists without CUDA. Therefore, BOINC has never shown peak-flops on the Mac for AMD or for NVIDIA without CUDA because it does not display COPROC_ATI::description() and may not display COPROC_NVIDIA::description(). On all OS's, INTEL GPU OpenCL will always exist without any vendor-proprietary GPU computation framework, so COPROC_INTEL::description() would never show COPROC_USED. Solutions: [1] Instead of adding "(OpenCL only)" to these descriptions just to show peak-flops for these cases, show peak_flops in all OpenCL descriptions. [2] Preface COPROC_NVIDIA::description() with "CUDA: " and preface COPROC_ATI::description() with "CAL: " to match the "OpenCL: " preface in OPENCL_DEVICE_PROP::description(). [3] Eliminate useless COPROC_INTEL::description().

- client: COPROC_NVIDIA, COPROC_ATI and COPROC_INTEL constructors must initialize the entire struct, not just the GPU type field.

- client: re-apply app configuration after each scheduler RPC; otherwise the changes get overwritten

- lib: the XML tag for the description of an Intel GPU is <coproc_intel_gpu>, not <intel_gpu>

- lib: the XML type field for an Intel GPU is INTEL_GPU, not INTEL

- client: fix a few variable naming issue that cropped up during the merge to the 7.0.x branch.

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1326052 - Posted: 9 Jan 2013, 14:02:45 UTC

7.0.42 change log

- client: fix a few variable naming issue that cropped up during the merge to the 7.0.x branch.

- lib: the XML type field for an Intel GPU is INTEL_GPU, not INTEL.

- client: fix changes lost in previous merge from trunk: COPROCS::clear() should call intel_gpu.clear()

- client: fix changes lost in previous merge from trunk: set OPENCL_DEVICE_PROP::peak_flops field for COPROC_ATI when CAL is present for COPROC_NVIDIA when CUDA is present.

- Manager: remove sqlite3.c from Xcode project; add build script to build libsqlite3.a from source files downloaded from sqlite3.org

- lib: actually let's make it "intel_gpu".

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1326054 - Posted: 9 Jan 2013, 14:08:12 UTC

7.0.43 change log

- fix compile warning

- Git can ignore Unix compile outputs

- client: show Intel GPU memory size as integer

- Apply RustyBSD's patch for selectively disabling the detection of the X screensaver framework in configure.ac

- SCR: Include the compiled X11 screensaver app in the self extracting archive distro of the client software.

- Mgr: Fix delay displaying notices.

- OpenCL: Eliminate spurious error message when an OpenCL platform exists but the computer has no corresponding devices.

- Mac: Fix linker warning.

- client simulator: fix build errors

- Mgr: Fix potential crashes displaying notices asynchronously on Windows.

- Mgr: Fix bad line endings.

- Mgr: More robust way to display "No notices" message.

- Mgr: Shorten Internet timeout for notices on Windows to 5 seconds.

- Mgr: fix license declaration in file derived from wxWidgets.

- Mgr: Windows only: If OpenURL fails, we probably don't have an Internet connection so use a shorter timeout for subsequent calls to OpenURL until one succeeds; otherwise notices takes takes too long to display if there are multiple notices with images.

- Mgr: Display the fetching notices message until we have notices to display or have determined that there are no notices.

- Mgr: Don't call get_notices RPC before previous one is processed to prevent multiple display of notices.

- Mgr: Display "Fetching notices..." instead of "There are no notices" while getting notices. Display "There are no notices" only if we have actually determined there are no notices.

- Mgr: Add code to reload notices with images / items missing due to problems accessing their URLs.

- Mgr: If network activity is suspended, don't retrieve URL references within notices unless they are already in our cache or in the Windows cache.

- Mgr: Clear our internal Internet cache when selecting a different computer.

- Mgr: Display a message and "Retry now" button if some images or other items within notices failed to load from the Internet; if network activity is suspended, say so in the message.

- Mgr: When removing entries from our hash table, delete data to avoid memory leaks.

- Mgr: Allow existing to retrieve images or other items from the Internet even if network activity is suspended. (But client won't get new notices.)

- Mgr: Reset Internet timeout when retrying notices.

- Mgr: Bug fixes for Windows Async Internet access, including closing the WinINet handle when aborting due to timeout, etc. to prevent crashes.

- Mgr: Add all the new notices features to the Simple View notices dialog.

- client: improved log messages for work fetch

- Mgr: When closing Simple View notices, wait for all Async Internet access to complete before calling the notices panel destructor.
eliminate use of alloca() and strdupa().

- don't include malloc.h

- client: processor identification for ARM (from Evandro Menezes and Joachim Fritzsch)

- client: disable remote access if gui_rpc_auth.cfg is empty, or it doesn't exist and we can't open it for writing, or the write to it fails (from RustyBSD)

- client: don't crash if GUI RPC password is too long (from RustyBSD)

- client: Account for the display management power system when calculating idle time from the XSS system on Linux.

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1326055 - Posted: 9 Jan 2013, 14:09:43 UTC
Last modified: 9 Jan 2013, 14:10:03 UTC

7.0.44 change log

- Fix conflict resolution mistake

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1332147 - Posted: 28 Jan 2013, 3:31:39 UTC
Last modified: 28 Jan 2013, 3:31:57 UTC

7.0.45 change log

- client: show available disk space correctly on startup.

- client: add code for detecting running on batteries on OpenBSD. From RustyBSD.

- LIB: Standardize on using windows_format_error_string and drop windows_error_string. Move the windows_format_error_string function to win_util.cpp, .h instead of it being scattered between util.h and str_util.cpp. Convert the Windows error string into UTF8 before allowing it to be used by the caller. Remove windows_error_string from library.

- LIB: Call FormatMessageW directly and skip an extra string encoding conversion step.

- OpenCL: fix bug reported by Heinz-Bernd Eggenstein.

- Fix build break for boinc_opencl.cpp and move the include of <vector> below the pre-compiled header file for Windows builds in boinc_api.cpp

- client: make CPU throttling apply to GPU apps.

- MGR: when selecting a new client, clear Notices tab and don't display "Fetching notices" until connected to new client.

- MGR: clear Simple View Notices when disconnected and don't display "Fetching notices" until connected to client.

- client: get proxy info before attempting project list fetch or any other HTTP op.

- client: backoff message tweaks.

- Manager: fix case where the client has only an Intel GPU, and the manager was acting as if it had no GPUs.

- client: write GPU list in get_state() GUI RPC. Otherwise manager doesn't know what GPUs we have.

- client: the logic for work fetch in the presence of GPU-exclusions (especially per-app exclusions) was incomplete and buggy. Changes: make bitmaps of included instances per (app, resource type).

In round-robin simulation, we keep track of used instances (so that we know if there are instances that are idle because of exclusions). Do this based on app-level exclusions (previously it was done based on project-wide exclusions, which didn't include app-level exclusions).

Compute RSC_PROJECT_WORK_FETCH::non_excluded_instances as the logical OR of the per-app masks. I.e. if you exclude an instance for all apps separately, it's the same as excluding it for the project as a whole. (Note: this bitmap is used for only 1 purpose: if we have idle instances, don't request work from a project for which those instances are excluded.)

Define RSC_PROJECT_WORK_FETCH::ncoprocs_excluded as the # of instances excluded for *any* app, not the # excluded for all apps. This quantity is used in work fetch to make sure we don't unboundedly fetch jobs that turn out not to have a GPU to run on due to exclusions.

- MGR: Somehow we are receiving an WM_ACTIVATEAPP event before m_pFrame has been populated on Windows 8. If m_pFrame is NULL, ignore the event.

- client (Win): don't call msg_printf() from sysmon thread. Instead, put msg into a buffer and let main thread print it. This may fix crashes on system suspend/resume.

- Compile fix for non-Windows clients.

- OpenCL: remove dead code reported by Heinz-Bernd Eggenstein.

- WINSETUP: Update installer build files based on the 7.0.x branch

Profile MarkJProject donor
Volunteer tester
Avatar
Send message
Joined: 17 Feb 08
Posts: 944
Credit: 25,164,816
RAC: 2,408
Australia
Message 1333511 - Posted: 1 Feb 2013, 13:24:51 UTC

7.0.46 change log

- wrapper: use waitpid() instead of wait4(), which is missing on Android.

- client: if app does temporary exit, don't print premature exit warning

- client: show sysmon messages correctly. This was supposed to be in my 507cd79 commit, but it got botched somehow.

- client: the <task> debug flag enables suspend/resume messages for both CPU and GPU. Previously CPU messages were always shown,and GPU messages were shown if <cpu_sched_debug> was set.

- client: fix bug where reschedule wasn't being done on GPU suspend or resume.

- wrapper: fix CPU time accounting on Unix

- client: (android) add wifi location of Android 4 devices to detect wifi state properly.

- android: kill client process on startup to allow update

- android: bugfix in preferences tab

- android: rename of preferences

- android: deletion of unused components

- client: when formatting the OpenCL description field, use the human readable version of the GPU type. People were expecting Intel GPU instead of intel_gpu.

- android: Add environment variables for CC and CFLAGS. The regex stuff in the wrapper was using the Linux(x86) compiler instead of the Android (ARM) cross-compiler

- Wrapper: CPU time account was broken on Windows

- wrapper: remove unneeded debug code from the wrapper

- android: Add the NO_LCHOWN symbol to the list of symbols used to build boinc, boinc_zip, etc. Android doesn't support the lchown API.

- android/boinc_zip: Linux throws a warning about lchmod and on android it doesn't exist at all. Add -DNO_LCHOWN to Makefile.am so it applies to both environments

- boinc_zip: most of the zip code is C code

- boinc_zip: Add NO_LCHMOD for Android and Linux. Comment out duplicate definitions of gmtime() and localtime().

- Modifications to autoconf scripts for non-standard openssl installs

- Added boinc version checking to plan_class_spec. New plan_class_spec.xml tags are <max_core_client_version> and <min_core_client_version> which are integer values major*10000+minor*100+release

- lib: Only process FormatMessage output if the function succeeds. The CC calls windows_format_error_string() for all non-zero returns

Previous · 1 · 2 · 3 · 4 · 5 · Next

Message boards : Number crunching : Development BOINC 7.0.62

Copyright © 2014 University of California