rocm_smi_lib: add gpu_metrics_v1_3 support

Signed-off-by: Elena Sakhnovitch
Change-Id: I4a9dedc80b8fce60e12c5baf8651d54d16a6a41c

include/rocm_smi/rocm_smi.h

@@ -835,6 +835,7 @@ struct metrics_table_header_t {
 // existing field sizes are changed.
 #define RSMI_GPU_METRICS_API_CONTENT_VER_1 1
 #define RSMI_GPU_METRICS_API_CONTENT_VER_2 2
+#define RSMI_GPU_METRICS_API_CONTENT_VER_3 3
 
 // This should match NUM_HBM_INSTANCES
 #define RSMI_NUM_HBM_INSTANCES 4
@@ -900,6 +901,19 @@ typedef struct {
   uint32_t       gfx_activity_acc;   // new in v1
   uint32_t       mem_actvity_acc;     // new in v1
   uint16_t       temperature_hbm[RSMI_NUM_HBM_INSTANCES];  // new in v1
+
+  /* PMFW attached timestamp (10ns resolution) */
+  uint64_t       firmware_timestamp;      // added in v1_2
+
+  /* Voltage (mV) */
+  uint16_t       voltage_soc;            // added in v1_3
+  uint16_t       voltage_gfx;            // added in v1_3
+  uint16_t       voltage_mem;            // added in v1_3
+
+  uint16_t       padding1;
+
+  /* Throttle status (ASIC independent) */
+  uint64_t       indep_throttle_status;   // added in v1_3
   /// \endcond
 } rsmi_gpu_metrics_t;
 
@@ -2155,7 +2169,7 @@ rsmi_dev_busy_percent_get(uint32_t dv_ind, uint32_t *busy_percent);
  *
  *  @param[inout] utilization_counters Multiple utilization counters can be retreived with a single
  *  call. The caller must allocate enough space to the utilization_counters array. The caller also
- *  needs to set valid RSMI_UTILIZATION_COUNTER_TYPE type for each element of the array. 
+ *  needs to set valid RSMI_UTILIZATION_COUNTER_TYPE type for each element of the array.
  *  ::RSMI_STATUS_NOT_SUPPORTED if it is not supported with the provided arguments.
  *
  *  If the function reutrns RSMI_STATUS_SUCCESS, the counter will be set in the value field of

src/rocm_smi_gpu_metrics.cc

@@ -121,14 +121,32 @@ typedef struct {
   uint64_t           firmware_timestamp;
 } rsmi_gpu_metrics_v_1_2;
 
+typedef struct {
+  rsmi_gpu_metrics_t base;
+	/* PMFW attached timestamp (10ns resolution) */
+	uint64_t			firmware_timestamp;
+
+	/* Voltage (mV) */
+	uint16_t			voltage_soc;
+	uint16_t			voltage_gfx;
+	uint16_t			voltage_mem;
+
+	/* Throttle status (ASIC independent) */
+	uint64_t			indep_throttle_status;
+
+} rsmi_gpu_metrics_v_1_3;
+
 static rsmi_status_t GetGPUMetricsFormat1(uint32_t dv_ind,
                                 rsmi_gpu_metrics_t *data, uint8_t content_v) {
   assert(content_v != RSMI_GPU_METRICS_API_CONTENT_VER_1 &&
-         content_v != RSMI_GPU_METRICS_API_CONTENT_VER_2 );
+         content_v != RSMI_GPU_METRICS_API_CONTENT_VER_2 &&
+         content_v != RSMI_GPU_METRICS_API_CONTENT_VER_3 );
   if (content_v == RSMI_GPU_METRICS_API_CONTENT_VER_1 ||
-      content_v == RSMI_GPU_METRICS_API_CONTENT_VER_2 ) {
+      content_v == RSMI_GPU_METRICS_API_CONTENT_VER_2 ||
+      content_v == RSMI_GPU_METRICS_API_CONTENT_VER_3 ) {
     // This function shouldn't be called if content version is
     // RSMI_GPU_METRICS_API_CONTENT_VER_1 or RSMI_GPU_METRICS_API_CONTENT_VER_2
+    // or RSMI_GPU_METRICS_API_CONTENT_VER_3
     return RSMI_STATUS_INVALID_ARGS;
   }
   void *metric_data = nullptr;
@@ -226,6 +244,17 @@ static void map_gpu_metrics_1_2_to_rsmi_gpu_metrics_t(
           gpu_metrics_v_1_2->firmware_timestamp * 10;
 }
 
+static void map_gpu_metrics_1_3_to_rsmi_gpu_metrics_t(
+        const rsmi_gpu_metrics_v_1_3 *gpu_metrics_v_1_3,
+        rsmi_gpu_metrics_t *rsmi_gpu_metrics)
+{
+    memcpy(rsmi_gpu_metrics, &gpu_metrics_v_1_3->base,
+           sizeof(rsmi_gpu_metrics_t));
+    // firmware_timestamp is at 10ns resolution
+    rsmi_gpu_metrics->system_clock_counter =
+          gpu_metrics_v_1_3->firmware_timestamp * 10;
+
+}
 
 rsmi_status_t
 rsmi_dev_gpu_metrics_info_get(uint32_t dv_ind, rsmi_gpu_metrics_t *smu) {
@@ -233,6 +262,7 @@ rsmi_dev_gpu_metrics_info_get(uint32_t dv_ind, rsmi_gpu_metrics_t *smu) {
   DEVICE_MUTEX
   CHK_SUPPORT_NAME_ONLY(smu)
   rsmi_gpu_metrics_v_1_2 smu_v_1_2;
+  rsmi_gpu_metrics_v_1_3 smu_v_1_3;
   rsmi_status_t ret;
 
   if (!dev->gpu_metrics_ver().structure_size) {
@@ -248,6 +278,10 @@ rsmi_dev_gpu_metrics_info_get(uint32_t dv_ind, rsmi_gpu_metrics_t *smu) {
     return RSMI_STATUS_NOT_SUPPORTED;
   }
 
+  // Initialize the smu fiedls to zero as some of them only valid in
+  // a specific version.
+  *smu = {};
+
   if (dev->gpu_metrics_ver().content_revision ==
                                         RSMI_GPU_METRICS_API_CONTENT_VER_1) {
         ret = GetDevBinaryBlob(amd::smi::kDevGpuMetrics, dv_ind,
@@ -257,6 +291,11 @@ rsmi_dev_gpu_metrics_info_get(uint32_t dv_ind, rsmi_gpu_metrics_t *smu) {
         ret = GetDevBinaryBlob(amd::smi::kDevGpuMetrics, dv_ind,
                                 sizeof(rsmi_gpu_metrics_v_1_2), &smu_v_1_2);
         map_gpu_metrics_1_2_to_rsmi_gpu_metrics_t(&smu_v_1_2, smu);
+  } else if (dev->gpu_metrics_ver().content_revision ==
+                                        RSMI_GPU_METRICS_API_CONTENT_VER_3) {
+        ret = GetDevBinaryBlob(amd::smi::kDevGpuMetrics, dv_ind,
+                                sizeof(rsmi_gpu_metrics_v_1_3), &smu_v_1_3);
+        map_gpu_metrics_1_3_to_rsmi_gpu_metrics_t(&smu_v_1_3, smu);
   } else {
             ret = GetGPUMetricsFormat1(dv_ind, smu,
                                      dev->gpu_metrics_ver().content_revision);