diff --git a/projects/clr/hipamd/src/hip_graph_internal.cpp b/projects/clr/hipamd/src/hip_graph_internal.cpp index f3a5be0880..472c94af9a 100644 --- a/projects/clr/hipamd/src/hip_graph_internal.cpp +++ b/projects/clr/hipamd/src/hip_graph_internal.cpp @@ -585,8 +585,9 @@ hipError_t GraphExec::Run(hipStream_t stream) { amd::AccumulateCommand* accumulate = nullptr; bool isLastPacketKernel = false; if (DEBUG_CLR_GRAPH_PACKET_CAPTURE) { - uint8_t* lastCapturedPacket = (topoOrder_.back()->GetType() == hipGraphNodeTypeKernel) ? - topoOrder_.back()->GetAqlPacket() : nullptr; + uint8_t* lastCapturedPacket = (topoOrder_.back()->GetType() == hipGraphNodeTypeKernel) + ? topoOrder_.back()->GetAqlPacket() + : nullptr; accumulate = new amd::AccumulateCommand(*hip_stream, {}, nullptr, lastCapturedPacket); } diff --git a/projects/clr/rocclr/device/pal/palsettings.cpp b/projects/clr/rocclr/device/pal/palsettings.cpp index a5d08c50be..c68eaff6f8 100644 --- a/projects/clr/rocclr/device/pal/palsettings.cpp +++ b/projects/clr/rocclr/device/pal/palsettings.cpp @@ -146,6 +146,7 @@ Settings::Settings() { useDeviceKernelArg_ = flagIsDefault(HIP_FORCE_DEV_KERNARG) ? true : HIP_FORCE_DEV_KERNARG; limit_blit_wg_ = 16; + DEBUG_CLR_GRAPH_PACKET_CAPTURE = false; // disable graph performance optimizations for PAL } bool Settings::create(const Pal::DeviceProperties& palProp, diff --git a/projects/clr/rocclr/utils/flags.hpp b/projects/clr/rocclr/utils/flags.hpp index cd1f40c42d..3bdc66a735 100644 --- a/projects/clr/rocclr/utils/flags.hpp +++ b/projects/clr/rocclr/utils/flags.hpp @@ -235,7 +235,7 @@ release(size_t, HIP_INITIAL_DM_SIZE, 8 * Mi, \ "Set initial heap size for device malloc.") \ release(bool, HIP_FORCE_DEV_KERNARG, 0, \ "Force device mem for kernel args.") \ -release(bool, DEBUG_CLR_GRAPH_PACKET_CAPTURE, false, \ +release(bool, DEBUG_CLR_GRAPH_PACKET_CAPTURE, true, \ "Enable/Disable graph packet capturing") \ release(bool, GPU_DEBUG_ENABLE, false, \ "Enables collection of extra info for debugger at some perf cost") \