rocr/driver: add GetClockCounters API to driver interface

This commit introduces a new GetClockCounters API to the driver interface.

- Implemented GetClockCounters in KfdDriver to fetch clock counters
  using hsaKmtGetClockCounters.
- Added a stub implementation of GetClockCounters in XdnaDriver that
  returns HSA_STATUS_ERROR.
- Modified GpuAgent to use driver().GetClockCounters instead of
  directly calling hsaKmtGetClockCounters.

Signed-off-by: Honglei Huang <Honglei1.Huang@amd.com>


[ROCm/ROCR-Runtime commit: 8d077dba3b]

projects/rocr-runtime/runtime/hsa-runtime/core/driver/kfd/amd_kfd_driver.cpp

@@ -583,6 +583,14 @@ hsa_status_t KfdDriver::GetDeviceHandle(uint32_t node_id, void** device_handle)
   return HSA_STATUS_SUCCESS;
 }
 
+hsa_status_t KfdDriver::GetClockCounters(uint32_t node_id, HsaClockCounters* clock_counter) const {
+  assert(clock_counter);
+
+  if (HSAKMT_CALL(hsaKmtGetClockCounters(node_id, clock_counter)) != HSAKMT_STATUS_SUCCESS)
+    return HSA_STATUS_ERROR;
+  return HSA_STATUS_SUCCESS;
+}
+
 hsa_status_t KfdDriver::IsModelEnabled(bool* enable) const {
   // AIE does not support streaming performance monitor.
   HSAKMT_STATUS status = HSAKMT_STATUS_ERROR;

projects/rocr-runtime/runtime/hsa-runtime/core/driver/xdna/amd_xdna_driver.cpp

@@ -881,5 +881,9 @@ hsa_status_t XdnaDriver::GetDeviceHandle(uint32_t node_id, void** device_handle)
   return HSA_STATUS_ERROR;
 }
 
+hsa_status_t XdnaDriver::GetClockCounters(uint32_t node_id, HsaClockCounters* clock_counter) const {
+  return HSA_STATUS_ERROR;
+}
+
 } // namespace AMD
 } // namespace rocr

projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_kfd_driver.h

@@ -125,6 +125,7 @@ public:
   hsa_status_t SetTrapHandler(uint32_t node_id, const void* base, uint64_t base_size,
                               const void* buffer_base, uint64_t buffer_base_size) const override;
   hsa_status_t GetDeviceHandle(uint32_t node_id, void** device_handle) const override;
+  hsa_status_t GetClockCounters(uint32_t node_id, HsaClockCounters* clock_counter) const override;
 
   hsa_status_t OpenSMI(uint32_t node_id, int* fd) const override;
 

projects/rocr-runtime/runtime/hsa-runtime/core/inc/amd_xdna_driver.h

@@ -239,6 +239,7 @@ public:
   hsa_status_t SetTrapHandler(uint32_t node_id, const void* base, uint64_t base_size,
                               const void* buffer_base, uint64_t buffer_base_size) const override;
   hsa_status_t GetDeviceHandle(uint32_t node_id, void** device_handle) const override;
+  hsa_status_t GetClockCounters(uint32_t node_id, HsaClockCounters* clock_counter) const override;
 
   hsa_status_t IsModelEnabled(bool* enable) const override;
 

projects/rocr-runtime/runtime/hsa-runtime/core/inc/driver.h

@@ -276,6 +276,13 @@ public:
   virtual hsa_status_t GetDeviceHandle(uint32_t node_id, void** device_handle) const = 0;
 
 
+  /// @brief Gets clock counters for particular Node
+  /// @param[in] node_id Node ID of the agent
+  /// @param[out] clock_counter Clock counter
+  /// @return HSA_STATUS_SUCCESS if the driver successfully returns the clock
+  virtual hsa_status_t GetClockCounters(uint32_t node_id,
+                                        HsaClockCounters* clock_counter) const = 0;
+
   /// @brief Check if the HSA KMT Model is enabled
   /// @param[out] enable True if the model is enabled, false otherwise
   virtual hsa_status_t IsModelEnabled(bool* enable) const = 0;

projects/rocr-runtime/runtime/hsa-runtime/core/runtime/amd_gpu_agent.cpp

@@ -125,10 +125,10 @@ GpuAgent::GpuAgent(HSAuint32 node, const HsaNodeProperties& node_props, bool xna
   if (node_props.Capability.ui32.DoorbellType != 2)
     throw AMD::hsa_exception(HSA_STATUS_ERROR, "Agent creation failed.\nThe GPU node uses a deprecated doorbell type\n");
 
-  HSAKMT_STATUS err = HSAKMT_CALL(hsaKmtGetClockCounters(node_id(), &t0_));
+  hsa_status_t err = driver().GetClockCounters(node_id(), &t0_);
   t1_ = t0_;
   historical_clock_ratio_ = 0.0;
-  assert(err == HSAKMT_STATUS_SUCCESS && "hsaGetClockCounters error");
+  assert(err == HSA_STATUS_SUCCESS && "hsaGetClockCounters error");
 
   const core::Isa *isa_base;
 
@@ -1678,7 +1678,8 @@ hsa_status_t GpuAgent::GetInfo(hsa_agent_info_t attribute, void* value) const {
       HsaClockCounters hsakmt_counters = {};
       hsa_amd_clock_counters_t* counters = static_cast<hsa_amd_clock_counters_t*>(value);
 
-      if (hsaKmtGetClockCounters(node_id(), &hsakmt_counters) == HSAKMT_STATUS_SUCCESS ) {
+      hsa_status_t err = driver().GetClockCounters(node_id(), &hsakmt_counters);
+      if (err == HSA_STATUS_SUCCESS) {
         counters->cpu_clock_counter = hsakmt_counters.CPUClockCounter;
         counters->gpu_clock_counter = hsakmt_counters.GPUClockCounter;
         counters->system_clock_counter = hsakmt_counters.SystemClockCounter;
@@ -2199,8 +2200,8 @@ uint16_t GpuAgent::GetSdmaMicrocodeVersion() const {
 }
 
 void GpuAgent::SyncClocks() {
-  HSAKMT_STATUS err = HSAKMT_CALL(hsaKmtGetClockCounters(node_id(), &t1_));
-  assert(err == HSAKMT_STATUS_SUCCESS && "hsaGetClockCounters error");
+  hsa_status_t err = driver().GetClockCounters(node_id(), &t1_);
+  assert(err == HSA_STATUS_SUCCESS && "hsaGetClockCounters error");
 }
 
 hsa_status_t GpuAgent::UpdateTrapHandlerWithPCS(pcs_sampling_data_t* pcs_hosttrap_buffers, pcs_sampling_data_t* pcs_stochastic_buffers) {