Errata

Codeplex: Niels_Dekker wrote Sep 1, 2011 at 8:15 AM

Chapter 2, Parallel Loops, shows four calls to Concurrency::parallel_for<_Index_type,_Function>, each of them having an unsigned int ('0u') as first argument, and either a size_t or a vector::size_type as second argument. This causes compile errors when the target platform is x64:

error C2782: 'void Concurrency::parallel_for(_Index_type,_Index_type,const _Function &)' : template parameter '_Index_type' is ambiguous

I would suggest to explicitly specify the first template argument:
Find what: 'parallel_for(0u'
Replace with: 'parallel_for<size_t>(0u'

Codeplex: jsh003guardz63 wrote Jul 19, 2011 at 9:20 AM

The explanation of performance characteristics of pipelines on p96 needs clarification and possibly correction.

In particular, ?If there are enough available cores, and if all stages of a pipeline take an equal amount of time, the execution time for the pipeline as a whole is the same as the time for just one stage?, would appear to be incorrect.

If:
S = the number of stages in figure 4 (= 4)
N = the number of images in figure 4 (= 6)
T = the duration to process one stage for one image (= t1 - t0 = t2 ? t1)
Then:
The elapsed time for a completely parallel and completely efficient pipeline is (S+N-1) * T.
The elapsed time taken for one stage = N * T.
But:
(S+N-1) x T is only equal to N * T when S = 1.

If the sidebar on p96 is referring to the case where S = 1, the phrasing of that sidebar needs changing, as it currently implies that (S+N-1) x T = N * T for any S. If the sidebar isn't referring to the case where S = 1, it needs re-wording as its intended meaning is unclear and/or incorrect. Possibly, it's trying to say that each core has an equal execution time, but I don't think that's the intended meaning... it?s hard to be sure what it is trying to say.