Provide cuda::ptr_in_range
#6086
Conversation
|
next step is to aggregate memory/pointer utilities in |
This comment has been minimized.
This comment has been minimized.
|
@davebayer @miscco open question, we don't do anything special in |
|
I was also tempted to optimize the function in the same way of |
|
pre-commit.ci autofix |
|
/ok to test 829f9e0 |
This comment has been minimized.
This comment has been minimized.
|
/ok to test a506e84 |
This comment has been minimized.
This comment has been minimized.
|
I got scared enough by https://quuxplusone.github.io/blog/2019/01/20/std-less-nightmare/ to use a safer version of the function. |
| [[nodiscard]] _CCCL_API bool __ptr_in_range_host(_Tp* __ptr, _Tp* __start, _Tp* __end) noexcept | ||
| { | ||
| _CCCL_ASSERT(::std::greater_equal<>{}(__end, __start), "__ptr_in_range_host: __end must be greater than __start"); | ||
| return ::std::greater_equal<>{}(__ptr, __start) && ::std::less<>{}(__ptr, __end); |
There was a problem hiding this comment.
just out of curiosity, is our common practice to use ::std::* within HOST_COMPILATION instead of universally relying on ::cuda::std::less (in this case) to do the host/device dispatching?
There was a problem hiding this comment.
In this case we really rely on a special blessing of std::greater_equal etc. in the C++ standard that allows pointers to be compared and establishes a total order. operator< on raw pointers only yields a partial order, which the compiler abuses in some situations. E.g.:
int a[10];
int b[10];
int* p = a +5;
if (a < p && p < a + 10) // this is legal
if (a < p && p < b) // this is NOT legal, pointers from different arrays cannot be compared. invokes UB
if (std::less{}(a, p) && std::less{}(p, b)) // this IS legal, pointer addresses are numerically comparedDon't ask me why :D I really don't know.
There was a problem hiding this comment.
Here is the wording: https://eel.is/c++draft/comparisons#general-2
There was a problem hiding this comment.
I read about it before reviewing here. My questions was why we don't use cuda::std::* instead of just std::. Not why we do not use the < operator. But I guess you answered by saying "special blessing of std::greater_equal" which kinda implies that the cuda::std counterpart does not partake of that blessing?
There was a problem hiding this comment.
yes, correct. The host compiler recognizes std::greater_equal but not cuda::std::greater_equal
Co-authored-by: Giannis Gonidelis <[email protected]>
|
/ok to test ddd33e5 |
|
/ok to test d164dd5 |
This comment has been minimized.
This comment has been minimized.
🥳 CI Workflow Results🟩 Finished in 3h 43m: Pass: 100%/242 | Total: 3d 02h | Max: 2h 00m | Hits: 97%/304639See results here. |
5 participants
Description
Checking if a pointer is in a given range is very common. The utility has already been proposed in WG21 P3234. It is already provided by Boost Core and part of libc++/libstdc++ as internal function.