From 7f0ad39bc78ec8ba6adec7f804f33dba96084051 Mon Sep 17 00:00:00 2001 From: Wayne Franz Date: Tue, 9 Apr 2024 08:36:46 -0400 Subject: [PATCH] Update ChangeLog with note about hipGraphs (#354) Now that most rocPRIM algorithms work within hipGraphs, most hipCUB algorithms will too, when using the rocPRIM backend. There are a couple of exceptions to this. This change adds a note about them to the changelog. --- CHANGELOG.md | 8 ++++++++ 1 file changed, 8 insertions(+) diff --git a/CHANGELOG.md b/CHANGELOG.md index 8848d42e..43e746d8 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -16,6 +16,14 @@ Documentation for hipCUB is available at type (`key_type`) can be sorted via these overloads, if an appropriate decomposer is passed. The decomposer has to implement `operator(const key_type&)` which returns a `hipcub::tuple` of references pointing to members of `key_type`. +* On AMD GPUs (using the HIP backend), it is possible to issue hipCUB API calls inside of + hipGraphs, with several exceptions: + * CachingDeviceAllocator + * GridBarrierLifetime + * DeviceSegmentedRadixSort + * DeviceRunLengthEncode + Currently, these classes rely on one or more synchronous calls to function correctly. Because of this, they cannot be used inside of hipGraphs. + ### Changed * The NVIDIA backend now requires CUB, Thrust and libcu++ 2.2.0. If it is not found it will be downloaded from the NVIDIA CCCL repository.