问题
In this question it was brought up that writing to two different offsets in a char array concurrently would imply a data race, since some processors such as Alpha don't have byte-wise addressing so it'd be hard to implement this.
I certainly see that this would very much slow down writing bytes on alpha processors (basically involving a LL/SC), but as I understand the C++ standard every field in an array is its own memory location (although from reading §1.7, I could also see the whole array as one memory location - that's probably what this question boils down to).
So basically is the following pseudo code
char arr[10]; // global field
Thread 1:
arr[1] = 0;
Thread 2:
arr[0] = 1;
well defined according to the C++14 standard or not?
回答1:
From the C++14 standard (1.7/3):
Two or more threads of execution (1.10) can update and access separate memory locations without interfering with each other.
Where it previously defines (emphasis mine)
A memory location is either an object of scalar type or a maximal sequence of adjacent bit-fields all having non-zero width.
So the chars of the array are memory locations, but the array itself is not; therefore, separate threads writing to different chars do not interfere with each other.
来源：https://stackoverflow.com/questions/34706564/writing-to-different-offsets-in-array-always-well-defined