CCS 5.2 Parallel Builds option

Dave,

A general guideline is to never use more than the number of cores/processors you already have in your system... To know that, simply open a terminal/command prompt and type:

C:\>echo %NUMBER_OF_PROCESSORS%

user@host:~$ cat /proc/cpuinfo | grep processor | wc -l

This will display the number of physical cores reported to the operating system.

Certain systems can also run two threads per core (Intel's Hyper-thread or AMD's Bulldozer), but adding this to the number of parallel processing jobs ends up impacting the overall performance, as discussed in the very interesting article below:

http://bitsum.com/pl_when_hyperthreading_hurts.php

Another tip is to avoid max out the number of processors if your system is a server or is located remotely and is accessible by other users. This may cause build errors due to resource conflicts - I've seen that, especially in projects with linked files. 

Apart from this I can't think of any additional scenarios where the limitations may happen.

Hope this helps,

Rafael

So does this mean that if CCS is told to use two cores (for example) and if it any point during the build only one core is available, this resource conflict could result in a build error?  Is there any way to "lock down" cores for CCS to use during a build?  Otherwise, the build is at the mercy of the OS and any other processes that might be running on the build PC.

Hi,

No, there should not be a build error due to the use of parallel builds. CCS configures the project and its options and passes all of them to the gmake utility, which handles the build process itself - therefore the parallel build is managed by gmake but it still is bounded by the Operating System rules.

At the compile phase (or gmake compile rule), each build step runs 'n' instances of the compiler (each builds a single source file and uses its own process cl6x.exe) and only starts a new instance to compile the next source file if there are 'n-1" instances running - in other words, after one of the previous cl6x completes its job.

Close to the end of the gmake compile rule, no additional compiler instances are spawned waiting for the remaining source files to be compiled.

After the last source file is compiled, all required dependencies to run the linker are met; thus gmake runs the linker rule - typically a single process (cl6x -z).

The resource conflict should not result in a build error, since the build system opens a new compiler process (cl6x.exe) at every source file to be built and will wait until all dependencies are met before moving to the next rule. The only thing that may happen is that a loaded core will run the compiler at a slower speed, as it will preempt its operation to allow time for the other processes to run according to its queue rules. All modern Operating Systems are preemptive, which prevents a process to hijack a core.

Unless the Operating System allows statically assigning processes to specific cores - i.e., by the process name (cl6x.exe) - the build process is at its mercy. AFAIK this is true for Windows environments and I strongly suspect this is the case for Linux as well.

Regards,

Rafael