Issue with building reference models with xPC target

I have a model that requires concurrent execution (multiple processors) to run effectively. This means that the top-level model consists of several sub-blocks that are all model references (each referenced block can be assigned to a processor).

I am experiencing strange behavior during the build process. For one of these referenced models, I can compile it on its own and compilation will take < 3 minutes. However, if I build the top-level model, it always recompiles the referenced model, and this compilation can take upwards of 20 minutes.

I find it strange that the compilation on the referenced model (the sub-block) is being triggered even though I haven't modified the block at all. E.g. if the sub-block is 'B' and the top-level model is 'A', then I do

rtwbuild('B') rtwbuild('A')

The first build of B goes quickly. Even though the build of A is happening immediately afterwards, a recompile of B is triggered, and this process takes ~20 minutes. I see no reason why any of the model's dependencies should be modified in between the two commands...?

When the build of A gets to model B, the message I get is

### Checking the status of model reference RTW target for model 'B' used in 'A' ### Model reference RTW target (B.c) for model B is out of date because B.c does not exist.

Any thoughts on why this might be happening?

Best, Chethan

Answer 1

[Chethan Pandarinath]

Just for everyone's future reference, the solution to this problem was achieved in the following way:

In Model Configuration Parameters, under Code Generation, I ran "Check model..."

This made a couple specific recommendations, which I followed: Under "Check model configuration", it was recommended to turn on Signal storage reuse, and other sub-options. This was done. Under "Check for optimal bus virtuality", it recommended converting a couple buses to output as non-virtual. This was also followed.

After these adjustments, compile times dropped down from ~25 minutes to ~2.

Answer 2

[Rajiv Ghosh-Roy]

Just building model B is a different build format from building A (which in turn builds B). In the first case, you are building B for use as a standalone model. In the second case, A needs B to run, and this builds it differently. Hence the dependency build.

[Chethan Pandarinath]

I see. Any thoughts on why a model that compiles in 3 minutes standalone would take ~20 minutes to compile as part of the larger build? (I.e. 20 minutes just for the specific compile of sub-block B, not the full model A)

[Rajiv Ghosh-Roy]

I'm not sure why this is happening; your best bet would probably be to contact Support since you will be able to share your model. The builds "should" take roughly the same time. A wild guess would be that you are getting close to maxing out memory during the concurrent build.

[Chethan Pandarinath]

Thanks Rajiv. Yeah, unfortunately it's not a memory issues (several gigs of memory free; the compile only takes <100MB). The CPU usage stays at 100% (on a single processor) throughout the compile.