AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SWDEV-467102 - Hidden heap init for graph capture

Browse Source 
If the graph has kernels that does device side allocation,  during packet capture, heap is
allocated because heap pointer has to be added to the AQL packet, and initialized during
graph launch.

Handle race with wait when 2 kernels with device heap are enqueued on multiple streams.

Change-Id: I45933b77fcaf7bc8fdf1bc906462e32b5d8d3688


[ROCm/clr commit: 57156c524d]
This commit is contained in:
Anusha GodavarthySurya
2024-06-11 11:59:05 +00:00
committed by Anusha Godavarthy Surya
parent c8bc9e3f2e
commit 291f079669
9 changed files with 57 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/clr/hipamd/src/hip_graph_internal.cpp
+19 -6
View File
@@ -358,8 +358,7 @@ void GetKernelArgSizeForGraph(std::vector<std::vector<Node>>& parallelLists,
// arg size required for all graph kernel nodes to allocate
for (const auto& list : parallelLists) {
for (auto& node : list) {
if (node->GetType() == hipGraphNodeTypeKernel &&
!reinterpret_cast<hip::GraphKernelNode*>(node)->HasHiddenHeap()) {
if (node->GetType() == hipGraphNodeTypeKernel) {
kernArgSizeForGraph += reinterpret_cast<hip::GraphKernelNode*>(node)->GetKerArgSize();
} else if (node->GetType() == hipGraphNodeTypeGraph) {
auto& childParallelLists = reinterpret_cast<hip::ChildGraphNode*>(node)->GetParallelLists();
@@ -375,8 +374,13 @@ hipError_t AllocKernelArgForGraph(std::vector<hip::Node>& topoOrder, hip::Stream
hip::GraphExec* graphExec) {
hipError_t status = hipSuccess;
for (auto& node : topoOrder) {
if (node->GetType() == hipGraphNodeTypeKernel &&
!reinterpret_cast<hip::GraphKernelNode*>(node)->HasHiddenHeap()) {
if (node->GetType() == hipGraphNodeTypeKernel) {
// Check if graph requires hidden heap and set as part of graphExec param.
static bool initialized = false;
if (!initialized && reinterpret_cast<hip::GraphKernelNode*>(node)->HasHiddenHeap()) {
graphExec->SetHiddenHeap();
initialized = true;
}
auto kernelNode = reinterpret_cast<hip::GraphKernelNode*>(node);
// From the kernel pool allocate the kern arg size required for the current kernel node.
address kernArgOffset = nullptr;
@@ -591,8 +595,7 @@ hipError_t EnqueueGraphWithSingleList(std::vector<hip::Node>& topoOrder, hip::St
accumulate = new amd::AccumulateCommand(*hip_stream, {}, nullptr);
}
for (int i = 0; i < topoOrder.size(); i++) {
if (DEBUG_CLR_GRAPH_PACKET_CAPTURE && topoOrder[i]->GetType() == hipGraphNodeTypeKernel &&
!reinterpret_cast<hip::GraphKernelNode*>(topoOrder[i])->HasHiddenHeap()) {
if (DEBUG_CLR_GRAPH_PACKET_CAPTURE && topoOrder[i]->GetType() == hipGraphNodeTypeKernel) {
if (topoOrder[i]->GetEnabled()) {
hip_stream->vdev()->dispatchAqlPacket(topoOrder[i]->GetAqlPacket(),
topoOrder[i]->GetKernelName(),
@@ -640,6 +643,16 @@ hipError_t GraphExec::Run(hipStream_t stream) {
if (parallelLists_.size() == 1 &&
instantiateDeviceId_ == hip_stream->DeviceId()) {
if (DEBUG_CLR_GRAPH_PACKET_CAPTURE) {
// If the graph has kernels that does device side allocation, during packet capture, heap is
// allocated because heap pointer has to be added to the AQL packet, and initialized during
// graph launch.
static bool initialized = false;
if (!initialized && HasHiddenHeap()) {
hip_stream->vdev()->HiddenHeapInit();
initialized = true;
}
}
status = EnqueueGraphWithSingleList(topoOrder_, hip_stream, this);
} else if (parallelLists_.size() == 1 &&
instantiateDeviceId_ != hip_stream->DeviceId()) {
projects/clr/hipamd/src/hip_graph_internal.hpp
+5
View File
@@ -569,6 +569,7 @@ struct GraphExec : public amd::ReferenceCountedObject {
uint32_t kernarg_graph_cur_offset_ = 0;
uint32_t kernarg_graph_size_ = 128 * Ki;
int instantiateDeviceId_ = -1;
bool hasHiddenHeap_ = false; //!< Kernel has hidden heap(device side allocation)
public:
GraphExec(std::vector<Node>& topoOrder, std::vector<std::vector<Node>>& lists,
@@ -619,6 +620,10 @@ struct GraphExec : public amd::ReferenceCountedObject {
}
return clonedNode;
}
// returns if graph has nodes that require hidden heap/not
bool HasHiddenHeap() const { return hasHiddenHeap_; }
// Graph has nodes that require hidden heap.
void SetHiddenHeap() { hasHiddenHeap_ = true; }
address allocKernArg(size_t size, size_t alignment) {
assert(alignment != 0);