[SYCL] Move an accessor pointer to global_device space

[MrSidims]

With this patch an accessor pointer to global buffer is moved from
global space to global_device space. That is done to distinguish this
pointer from those USM pointers, that are allocated global space or
global_host space, in compile time.

In addition to this change there are added explicit conversion operator
from global_device to global space for multi_ptr class and implicit
convertion for atomic class from global_device for global space.
The last change isn't covered by specification published here:
#1840 , but is required to pass
atomic_api CTS.

[MrSidims]

@intel/llvm-reviewers-runtime ping

[v-klochkov]

Not a request to change, but it would be nice to fix it in my opinion.
The 'atomic' class seems to be ripped by those 'private', 'public' keywords already and your fix added more to it.

With your fix the class has: private: public: private: public: private.
I suggest moving those new private getter methods to the line 336

[MrSidims]

Not a request to change, but it would be nice to fix it in my opinion.
The 'atomic' class seems to be ripped by those 'private', 'public' keywords already and your fix added more to it.

With your fix the class has: private: public: private: public: private.
I suggest moving those new private getter methods to the line 336

I don't mind to change it, here I was just following Robert C. Martin "Clean code" advise to order methods in a way so readers have no need to scroll the code too much. What do you think?

[MrSidims]

Not a request to change, but it would be nice to fix it in my opinion.
The 'atomic' class seems to be ripped by those 'private', 'public' keywords already and your fix added more to it.
With your fix the class has: private: public: private: public: private.
I suggest moving those new private getter methods to the line 336

I don't mind to change it, here I was just following Robert C. Martin "Clean code" advise to order methods in a way so readers have no need to scroll the code too much. What do you think?

Well, actually, now I can even remove this method.

[v-klochkov]

I don't mind to change it, here I was just following Robert C. Martin "Clean code" advise to order methods in a way so readers have no need to scroll the code too much. What do you think?

In the process of answer to this question I was going to still suggest moving the methods to the place where the field declared. Then I realized that atomic::getPtr() is used only inside atomic class. This makes that getter method redundant, in my opinion.

I suppose you added it to public section before adding that "friend" class and that was the purpose of the getter method. After adding "friend" statement there is no big need in that getter.
In the constructor you can just write: "Ptr = RHS.Ptr;" instead of "Ptr = RHS.getPtr();"

[MrSidims]

I don't mind to change it, here I was just following Robert C. Martin "Clean code" advise to order methods in a way so readers have no need to scroll the code too much. What do you think?

In the process of answer to this question I was going to still suggest moving the methods to the place where the field declared. Then I realized that atomic::getPtr() is used only inside atomic class. This makes that getter method redundant, in my opinion.

I suppose you added it to public section before adding that "friend" class and that was the purpose of the getter method. After adding "friend" statement there is no big need in that getter.
In the constructor you can just write: "Ptr = RHS.Ptr;" instead of "Ptr = RHS.getPtr();"

Yeap, I've just removed it :)