Misleading definition in the cuda-memcheck docs


I found that there is an ambiguity in the cuda-memcheck documentation, racecheck section.
The descriptions of the RAW and WAR hazards seem swapped.
Is it possible that the author intended to say that the behavior in these definitions
is the desired one (and not the wrong one)?


I agree the descriptions seem to be listed under the incorrect headings, based on how I learned the terms:

WAR hazard: data is overwritten before it can be read; incorrectly retrieves new data
RAW hazard: data is read before it is written/updated; incorrectly retrieves old data

I will bring this to the attention of the debugger team so this section can be clarified in future versions of the documentation. Thanks for bringing this to our attention.