Host Implementation of Histogram APIs #1974
Conversation
danhoeflinger
changed the title
danhoeflinger
requested review from
reble,
timmiesmith,
akukanov,
MikeDvorskiy,
SergeyKopienko,
mmichel11,
adamfidel and
dmitriy-sobolev
| @@ -4289,6 +4289,86 @@ __pattern_shift_right(_Tag __tag, _ExecutionPolicy&& __exec, _BidirectionalItera | |||
| return __res.base(); | |||
| } | |||
|
|
|||
| template <typename _ForwardIterator, typename _IdxHashFunc, typename _RandomAccessIterator, class _IsVector> | |||
There was a problem hiding this comment.
Is there a reason why _IsVector is a class when the rest are typenames?
There was a problem hiding this comment.
No, it seems the convention in this file is generally for everything to be class (though its not fully consistent). I'll adjust it to the norm.
include/oneapi/dpl/pstl/omp/util.h
Outdated
| std::uint32_t __count = 0; | ||
| std::uint32_t __j = 0; | ||
|
|
||
| for (; __j < __thread_specific_storage.size() && __count <= __i; ++__j) |
There was a problem hiding this comment.
std::uint32_t __count = 0 could probably be moved here to the initialization expression of this loop.
There was a problem hiding this comment.
thanks, done.
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
| for (; __first != __last; ++__first) | ||
| { | ||
| std::int32_t __bin = __func.get_bin(*__first); | ||
| if (__bin >= 0) |
There was a problem hiding this comment.
- Can
get_binreturn negative value? - If yes, what is correct behavior for
__brick_histogram?
There was a problem hiding this comment.
Yes, -1 is returned when the input element does not fall within any histogram bin. The correct behavior is to do nothing and skip this input element.
Specification
"Input values that do not map to a defined bin are skipped silently."
I recently looked into expanding the bin helper interface to include a separate function to check bounds, and another to get the bin which assumes it is in bounds. I thought this might provide benefit by reducing the number of branches by 1, but I saw no performance benefit from this change for CPU or GPU. It is still something we could pursue in the future.
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
| namespace __detail | ||
| { | ||
|
|
||
| template <typename _ValueType, typename... _Args> | ||
| struct __enumerable_thread_local_storage | ||
| { |
There was a problem hiding this comment.
Even though these class templates are now internal, the API description for them is still a part of the backend API - you should know what can be done with the object obtained from the make function. We will therefore need to document it somewhere in sufficient detail - perhaps in the RFC for a start.
There was a problem hiding this comment.
Yes, I believe this interface is described in the RFC to some extent already. I can make that more specific and explicit.
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Signed-off-by: Dan Hoeflinger <[email protected]>
Implementation of histogram APIs for host backends.
Implementations are provided for
serial,tbb, andopenMPbackends. We add a generic__thread_enumerable_storagestruct to add a generic thread local storage for our host backends. We use the new TLS (Thread local storage) withparallel_forto implement histogram. Testing is also added, and some minor adjustments are made to cmake.Please see the RFC documentation / discussion here for more details.