SWDEV-508776 - support VGPRs validation (#274)
Clarify some VGPRs terms description.
Fix some wrong query logics of availableVGPRs_ and
availableRegistersPerCU_ in device info.
Add hipDeviceAttributeMaxAvailableVgprsPerThread
attribute query.
Remove hardcoding of following
info_.vgprAllocGranularity_
info_.vgprsPerSimd_
[ROCm/clr commit: 397f303d97]
Этот коммит содержится в:
@@ -619,7 +619,7 @@ Kernel::Kernel(const amd::Device& dev, const std::string& name, const Program& p
|
||||
workGroupInfo_.usedGPRs_ = 0;
|
||||
workGroupInfo_.availableSGPRs_ = 0;
|
||||
workGroupInfo_.usedSGPRs_ = 0;
|
||||
workGroupInfo_.availableVGPRs_ = 0;
|
||||
workGroupInfo_.availableVGPRs_ = dev.info().availableVGPRs_;
|
||||
workGroupInfo_.usedVGPRs_ = 0;
|
||||
workGroupInfo_.availableLDSSize_ = 0;
|
||||
workGroupInfo_.usedLDSSize_ = 0;
|
||||
|
||||
Ссылка в новой задаче
Block a user