shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[OpenMP][Offload] Update the Logic for Configuring Auto Zero-Copy (#143638)

Browse Source 
Summary:

Currently the Auto Zero-Copy is enabled by checking every initialized
device to ensure that no dGPU is attached to an APU. However, an APU is
designed to comprise a homogeneous set of GPUs, therefore, it should be
sufficient to check any device for configuring Auto Zero-Copy. In this
PR, it checks the first initialized device in the list.

The changes in this PR are to clearly reflect the design and logic of
enabling the feature for further improving the readibility.
This commit is contained in:
Kewen12 2025-06-11 11:12:54 -07:00 committed by GitHub
parent c2f0af514b
commit bbe59e19b6
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 7 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
offload/libomptarget/PluginManager.cpp
View File

@ -286,16 +286,16 @@ void PluginManager::registerLib(__tgt_bin_desc *Desc) {
}
PM->RTLsMtx.unlock();
bool UseAutoZeroCopy = Plugins.size() > 0;
bool UseAutoZeroCopy = false;
auto ExclusiveDevicesAccessor = getExclusiveDevicesAccessor();
for (const auto &Device : *ExclusiveDevicesAccessor)
UseAutoZeroCopy &= Device->useAutoZeroCopy();
// APUs are homogeneous set of GPUs. Check the first device for
// configuring Auto Zero-Copy.
if (ExclusiveDevicesAccessor->size() > 0) {
auto &Device = *(*ExclusiveDevicesAccessor)[0];
UseAutoZeroCopy = Device.useAutoZeroCopy();
}
// Auto Zero-Copy can only be currently triggered when the system is an
// homogeneous APU architecture without attached discrete GPUs.
// If all devices suggest to use it, change requirement flags to trigger
// zero-copy behavior when mapping memory.
if (UseAutoZeroCopy)
addRequirements(OMPX_REQ_AUTO_ZERO_COPY);