SWDEV-475244 - Memory Usage and Bandwidth: max mem and current mem

Implemented max memory bandwith and current memory bandwidth. Added two
new field ids: RDC_FI_GPU_MEMORY_MAX_BANDWIDTH, RDC_FI_GPU_MEMORY_CUR_BANDWIDTH

Signed-off-by: Li Ma <li.ma@amd.com>
Change-Id: I453e49937a84777146575f4f5bdd69fd4fe53bfc


[ROCm/rdc commit: 30f9b2ac2f]

projects/rdc/common/rdc_field.data

@@ -52,6 +52,8 @@ FLD_DESC_ENT(RDC_FI_GPU_MEMORY_TOTAL,    "Total memory of the GPU instance",
 FLD_DESC_ENT(RDC_FI_GPU_MM_ENC_UTIL,     "Mutilmedia encoder busy percentage",          "GPU_MM_ENC_UTIL",  true)
 FLD_DESC_ENT(RDC_FI_GPU_MM_DEC_UTIL,     "Mutilmedia decoder busy percentage",          "GPU_MM_DEC_UTIL",  true)
 FLD_DESC_ENT(RDC_FI_GPU_MEMORY_ACTIVITY, "Memory busy percentage",                      "GPU_MEM_UTIL",     true)
+FLD_DESC_ENT(RDC_FI_GPU_MEMORY_MAX_BANDWIDTH,       "Memory max bandwidth",             "GPU_MEM_MAX_BANDWIDTH",   true)
+FLD_DESC_ENT(RDC_FI_GPU_MEMORY_CUR_BANDWIDTH,       "Memory current bandwidth",         "GPU_MEM_CUR_BANDWIDTH",   true)
 
 FLD_DESC_ENT(RDC_FI_GPU_PAGE_RETRIED,    "Retried page of the GPU instance",            "GPU_PAGE_RETRIED", true)
 

projects/rdc/include/rdc/rdc.h

@@ -188,6 +188,8 @@ typedef enum {
   RDC_FI_GPU_MM_ENC_UTIL,  //!< Multimedia encoder busy percentage
   RDC_FI_GPU_MM_DEC_UTIL,  //!< Multimedia decoder busy percentage
   RDC_FI_GPU_MEMORY_ACTIVITY,  //!< Memory busy percentage
+  RDC_FI_GPU_MEMORY_MAX_BANDWIDTH,  //<! The Memory max bandwidth at current memory clock in Mb/Second
+  RDC_FI_GPU_MEMORY_CUR_BANDWIDTH,  //<! The Memory current bandwidth in Mb/Second
 
   /**
    * @brief GPU page related fields

projects/rdc/python_binding/RdcReader.py

@@ -8,6 +8,8 @@ default_field_ids = [
         rdc_field_t.RDC_FI_GPU_MM_ENC_UTIL,
         rdc_field_t.RDC_FI_GPU_MM_DEC_UTIL,
         rdc_field_t.RDC_FI_GPU_MEMORY_ACTIVITY,
+        rdc_field_t.RDC_FI_GPU_MEMORY_MAX_BANDWIDTH,
+        rdc_field_t.RDC_FI_GPU_MEMORY_CUR_BANDWIDTH,
         rdc_field_t.RDC_FI_OAM_ID,
         rdc_field_t.RDC_FI_POWER_USAGE,
         rdc_field_t.RDC_FI_GPU_CLOCK,

projects/rdc/python_binding/rdc_bootstrap.py

@@ -90,6 +90,8 @@ class rdc_field_t(c_int):
      RDC_FI_GPU_MM_ENC_UTIL = 503
      RDC_FI_GPU_MM_DEC_UTIL = 504
      RDC_FI_GPU_MEMORY_ACTIVITY = 505
+     RDC_FI_GPU_MEMORY_MAX_BANDWIDTH = 506
+     RDC_FI_GPU_MEMORY_CUR_BANDWIDTH = 507
      RDC_FI_ECC_CORRECT_TOTAL = 600
      RDC_FI_ECC_UNCORRECT_TOTAL = 601
      RDC_FI_ECC_SDMA_SEC = 602

projects/rdc/rdc_libs/rdc/src/RdcMetricFetcherImpl.cc

@@ -573,6 +573,32 @@ rdc_status_t RdcMetricFetcherImpl::fetch_smi_field(uint32_t gpu_index, rdc_field
       }
       break;
     }
+    case RDC_FI_GPU_MEMORY_MAX_BANDWIDTH: {
+      amdsmi_vram_info_t vram_info;
+
+      value->status = amdsmi_get_gpu_vram_info(processor_handle, &vram_info);
+      value->type = INTEGER;
+      if (value->status == AMDSMI_STATUS_SUCCESS) {
+        value->value.l_int = vram_info.vram_max_bandwidth;
+      }
+      break;
+    }
+    case RDC_FI_GPU_MEMORY_CUR_BANDWIDTH: {
+      amdsmi_engine_usage_t engine_usage;
+      amdsmi_vram_info_t vram_info;
+
+      value->status = amdsmi_get_gpu_activity(processor_handle, &engine_usage);
+      value->type = INTEGER;
+      if (value->status == AMDSMI_STATUS_SUCCESS) {
+        value->value.l_int = static_cast<int64_t>(engine_usage.umc_activity);
+      }
+
+      value->status = amdsmi_get_gpu_vram_info(processor_handle, &vram_info);
+      if (value->status == AMDSMI_STATUS_SUCCESS) {
+        value->value.l_int = value->value.l_int / 100 * vram_info.vram_max_bandwidth;
+      }
+      break;
+    }
     case RDC_FI_GPU_COUNT: {
       uint32_t processor_count = 0;
       // amdsmi is initialized in AMDSMI_INIT_AMD_GPUS mode -> returned sockets are GPUs

projects/rdc/rdc_libs/rdc/src/RdcSmiLib.cc

@@ -183,6 +183,7 @@ rdc_status_t RdcSmiLib::rdc_telemetry_fields_query(uint32_t field_ids[MAX_NUM_FI
       RDC_HEALTH_XGMI_ERROR,          RDC_HEALTH_PCIE_REPLAY_COUNT,   RDC_HEALTH_RETIRED_PAGE_NUM,
       RDC_HEALTH_PENDING_PAGE_NUM,    RDC_HEALTH_RETIRED_PAGE_LIMIT,  RDC_HEALTH_UNCORRECTABLE_PAGE_LIMIT,
       RDC_HEALTH_POWER_THROTTLE_TIME, RDC_HEALTH_THERMAL_THROTTLE_TIME,
+      RDC_FI_GPU_MEMORY_MAX_BANDWIDTH,  RDC_FI_GPU_MEMORY_CUR_BANDWIDTH,
   };
   std::copy(fields.begin(), fields.end(), field_ids);
   *field_count = fields.size();