SWDEV-498061 - Add capture support for hipModuleLaunchCooperativeKernel

Change-Id: I5ed188e046c680c2785b3952391f59ed1d0c21b8


[ROCm/clr commit: 30cb2d0e67]

projects/clr/hipamd/src/hip_graph.cpp

@@ -348,6 +348,40 @@ hipError_t capturehipModuleLaunchKernel(hipStream_t& stream, hipFunction_t& f, u
   return hipSuccess;
 }
 
+hipError_t capturehipModuleLaunchCooperativeKernel(hipStream_t& stream, hipFunction_t& f,
+                                                   uint32_t& gridDimX, uint32_t& gridDimY,
+                                                   uint32_t& gridDimZ, uint32_t& blockDimX,
+                                                   uint32_t& blockDimY, uint32_t& blockDimZ,
+                                                   uint32_t& sharedMemBytes, void**& kernelParams) {
+  ClPrint(amd::LOG_INFO, amd::LOG_API,
+          "[hipGraph] Current capture node ModuleLaunchCooperativeKernel on stream : %p", stream);
+
+  if (!hip::isValid(stream)) {
+    return hipErrorContextIsDestroyed;
+  }
+
+  hip::Stream* s = reinterpret_cast<hip::Stream*>(stream);
+  hipKernelNodeParams nodeParams;
+  nodeParams.func = f;
+  nodeParams.blockDim = {blockDimX, blockDimY, blockDimZ};
+  nodeParams.gridDim = {gridDimX, gridDimY, gridDimZ};
+  nodeParams.kernelParams = kernelParams;
+  nodeParams.sharedMemBytes = sharedMemBytes;
+  nodeParams.extra = nullptr;
+
+  hip::GraphNode* pGraphNode;
+  hipError_t status =
+      ihipGraphAddKernelNode(&pGraphNode, s->GetCaptureGraph(), s->GetLastCapturedNodes().data(),
+                             s->GetLastCapturedNodes().size(), &nodeParams, nullptr, true,
+                             amd::NDRangeKernelCommand::CooperativeGroups);
+  if (status != hipSuccess) {
+    return status;
+  }
+  s->SetLastCapturedNode(pGraphNode);
+
+  return hipSuccess;
+}
+
 hipError_t capturehipLaunchByPtr(hipStream_t& stream, hipFunction_t func, dim3 blockDim,
                                  dim3 gridDim, unsigned int sharedMemBytes, void** extra) {
   ClPrint(amd::LOG_INFO, amd::LOG_API, "[hipGraph] Current capture node LaunchByPtr on stream : %p",

projects/clr/hipamd/src/hip_graph_capture.hpp

@@ -42,6 +42,12 @@ hipError_t capturehipModuleLaunchKernel(hipStream_t& stream, hipFunction_t& f, u
                                         uint32_t& sharedMemBytes, void**& kernelParams,
                                         void**& extra);
 
+hipError_t capturehipModuleLaunchCooperativeKernel(hipStream_t& stream, hipFunction_t& f,
+                                                   uint32_t& gridDimX, uint32_t& gridDimY,
+                                                   uint32_t& gridDimZ, uint32_t& blockDimX,
+                                                   uint32_t& blockDimY, uint32_t& blockDimZ,
+                                                   uint32_t& sharedMemBytes, void**& kernelParams);
+
 hipError_t capturehipLaunchByPtr(hipStream_t& stream, hipFunction_t func, dim3 blockDim,
                                  dim3 gridDim, unsigned int sharedMemBytes, void** extra);
 

projects/clr/hipamd/src/hip_graph_internal.hpp

@@ -439,7 +439,6 @@ struct GraphNode : public hipGraphNodeDOTAttribute {
     bool isGraphCapture = false;
     if (DEBUG_CLR_GRAPH_PACKET_CAPTURE) {
       switch (GetType()) {
-        case hipGraphNodeTypeKernel:
         case hipGraphNodeTypeMemset:
           isGraphCapture = true;
           break;
@@ -1325,6 +1324,17 @@ class GraphKernelNode : public GraphNode {
     }
     return hipSuccess;
   }
+
+  virtual bool GraphCaptureEnabled() override {
+    bool isGraphCapture = false;
+    if (DEBUG_CLR_GRAPH_PACKET_CAPTURE) {
+      // Disable capture for cooperative kernels
+      if (!coopKernel_) {
+        isGraphCapture = true;
+      }
+    }
+    return isGraphCapture;
+  }
 };
 
 class GraphMemcpyNode : public GraphNode {

projects/clr/hipamd/src/hip_module.cpp

@@ -572,6 +572,9 @@ hipError_t hipModuleLaunchCooperativeKernel(hipFunction_t f, unsigned int gridDi
     HIP_RETURN(hipErrorInvalidValue);
   }
 
+  STREAM_CAPTURE(hipModuleLaunchCooperativeKernel, stream, f, gridDimX, gridDimY, gridDimZ,
+                 blockDimX, blockDimY, blockDimZ, sharedMemBytes, kernelParams);
+
   size_t globalWorkSizeX = static_cast<size_t>(gridDimX) * blockDimX;
   size_t globalWorkSizeY = static_cast<size_t>(gridDimY) * blockDimY;
   size_t globalWorkSizeZ = static_cast<size_t>(gridDimZ) * blockDimZ;