SWDEV-422207 - Enable DEBUG_CLR_GRAPH_PACKET_CAPTURE environiment variable

Change-Id: I9bf72b9c1a56980352109bd4d42b54ecb2d1b8f9


[ROCm/clr commit: ae0368d12d]

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);
     }
 

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,

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")    \