SWDEV-334152 - Set release as systemscope

Set release scope as system for dispatch AQL when events are passed to
hip*LaunchKernelGGL*

Change-Id: I93b91591e0ab023f1ecc5247f7905eca26147358


[ROCm/clr commit: 02566677cf]
This commit is contained in:
Saleel Kudchadker
2022-04-25 15:29:25 -07:00
parent 3a0462c981
commit d9c2aee526
2 ha cambiato i file con 9 aggiunte e 7 eliminazioni
@@ -1081,14 +1081,12 @@ VirtualGPU::VirtualGPU(Device& device, bool profiling, bool cooperative,
dispatchPacketHeaderNoSync_ =
(HSA_PACKET_TYPE_KERNEL_DISPATCH << HSA_PACKET_HEADER_TYPE) |
(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE) |
(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE) |
(HSA_FENCE_SCOPE_NONE << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE);
(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE);
dispatchPacketHeader_=
(HSA_PACKET_TYPE_KERNEL_DISPATCH << HSA_PACKET_HEADER_TYPE) |
(1 << HSA_PACKET_HEADER_BARRIER) |
(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE) |
(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE) |
(HSA_FENCE_SCOPE_NONE << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE);
(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE);
} else {
dispatchPacketHeaderNoSync_ =
(HSA_PACKET_TYPE_KERNEL_DISPATCH << HSA_PACKET_HEADER_TYPE) |
@@ -2971,9 +2969,12 @@ bool VirtualGPU::submitKernelInternal(const amd::NDRangeContainer& sizes, const
aqlHeaderWithOrder &= kAqlHeaderMask;
}
if (addSystemScope_) {
aqlHeaderWithOrder &= ~(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE);
aqlHeaderWithOrder |= (HSA_FENCE_SCOPE_SYSTEM << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE);
if (addSystemScope_ || (vcmd != nullptr &&
vcmd->getEventScope() == amd::Device::kCacheStateSystem)) {
aqlHeaderWithOrder &= ~(HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE |
HSA_FENCE_SCOPE_AGENT << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE);
aqlHeaderWithOrder |= (HSA_FENCE_SCOPE_SYSTEM << HSA_PACKET_HEADER_ACQUIRE_FENCE_SCOPE |
HSA_FENCE_SCOPE_SYSTEM << HSA_PACKET_HEADER_RELEASE_FENCE_SCOPE);
addSystemScope_ = false;
}
@@ -432,6 +432,7 @@ NDRangeKernelCommand::NDRangeKernelCommand(HostQueue& queue, const EventWaitList
profilingInfo_.clear();
profilingInfo_.callback_ = nullptr;
profilingInfo_.marker_ts_ = true;
setEventScope(amd::Device::kCacheStateSystem);
}
kernel_.retain();
}