Intel(R) TBB 4.2 update 4 is released and available for download

[Kirill_R_Intel]

Intel(R) Threading Building Blocks 4.2 update 4 is now released and available for download. Check the What's new article for list of changes.

You can download Intel TBB 4.2 update 4 from commercial and open source sites.

[RafSchietekat]

On OS X, dining_philosophers doesn't build: #include <unistd.h> to use sleep().

(Added) seismic is still invisible with argument 300... or maybe I just blinked at the wrong time.

[Kirill_R_Intel]

Thanks Raf for reporting this. We'll double check the samples and fix them if needed.

[RafSchietekat]

Not (yet?) in this release: Clang as the default compiler on OS X, X Window System support on recent Debian-derived Linux distributions.

[Christophe_H_Intel]

Hello, Raf,

About Dining Philosophers, could you tell me which OS/X and clang versions you have problems with?  I've tried two different versions (

Darwin 13.1.0 Darwin Kernel Version 13.1.0: Thu Jan 16 19:40:37 PST 2014; root:xnu-2422.90.20~2/RELEASE_X86_64 x86_64; Apple LLVM version 5.1 (clang-503.0.38) (based on LLVM 3.4svn)

Darwin 12.5.0 Darwin Kernel Version 12.5.0: Sun Sep 29 13:33:47 PDT 2013; root:xnu-2050.48.12~1/RELEASE_X86_64 x86_64; Apple LLVM version 5.1 (clang-503.0.40) (based on LLVM 3.4svn

and haven't found the error.

Thanks and Regards,
Chris

[RafSchietekat]

I'm seeing "Kernel Version: Darwin 13.1.0" (as in your first version) and "clang-503.0.40" (as in your second version).

"man 3 sleep" mentions "#include <unistd.h>", and if I add that to examples/graph/dining_philosophers/dining_philosophers.cpp it solves the problem "dining_philosophers.cpp:219:5: error: use of undeclared identifier 'sleep'".

(Added) Note that include/tbb/machine/gcc_generic.h includes this header (not immediately obvious why?), but it's not used with Clang as the compiler (verified for "make compiler=clang examples" by adding "#ifndef __GNUC__ / #error / #endif"). I couldn't immediately find the cause of the problem or of its sudden manifestation, but it (now?) also occurs for tbb42_20140122oss (previous update?), which I presume I would also have tested on OS X (or not?), so this may be caused by a recent change in the environment rather than by a change in TBB (meaning I should probably have reported it in a different thread). But it also seems better to just add "#include <unistd.h>" immediately before "#define SLEEP(a) sleep(a)" and reserve one's curiosity for more interesting subjects.

[Christophe_H_Intel]

Hi, Raf,

Thanks for the extra information.  I will fix it by using the tbb_thread method for sleep; seems like it will reduce the complexity of the test and remove the need to update it further.

Regards,
Chris

[RafSchietekat]

If you want to go there, you could do even better by reviewing all uses of sleeping. For example, there are 4 independent but (nearly?) identical definitions of a function MilliSleep(), 3 of rt_sleep(), and also Harness::Sleep(), that are comparable to SLEEP(): functions instead of a macro, more precise by a factor 1000, but all with similar or even greater complexity for the same simple functionality. It's hard to keep a DRY eye looking at that situation... :-)

(2014-05-20 Edited)