By default, only consider AMD GPU's in RSMI device list

With newly added initialization parameters that can be
passed to rsmi_init(), you can tell RSMI to consider other
devices.

Also:
-fixed incorrect header file name that would break in C
builds
-modified rsmi_init() and rsmi_shut_down() to reinitialize and
clear static structures

src/rocm_smi.cc

@@ -94,7 +94,7 @@ static rsmi_status_t handleException() {
 #define TRY try {
 #define CATCH } catch (...) {return handleException();}
 #define GET_DEV_FROM_INDX  \
-    amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance(); \
+    amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance(); \
     if (dv_ind >= smi.monitor_devices().size()) { \
       return RSMI_STATUS_INVALID_ARGS; \
     } \
@@ -106,7 +106,8 @@ static rsmi_status_t handleException() {
     amd::smi::ScopedPthread _lock(_pw);
 
 static pthread_mutex_t *get_mutex(uint32_t dv_ind) {
-  amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance();
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
+
   if (dv_ind >= smi.monitor_devices().size()) {
     return nullptr;
   }
@@ -382,7 +383,7 @@ static rsmi_status_t set_dev_mon_value(amd::smi::MonitorTypes type,
 
 static rsmi_status_t get_power_mon_value(amd::smi::PowerMonTypes type,
                                       uint32_t dv_ind, uint64_t *val) {
-  amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance();
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
 
   if (dv_ind >= smi.monitor_devices().size() || val == nullptr) {
     return RSMI_STATUS_INVALID_ARGS;
@@ -416,11 +417,12 @@ static bool is_power_of_2(uint64_t n) {
 }
 
 rsmi_status_t
-rsmi_init(uint64_t init_flags) {
+rsmi_init(uint64_t flags) {
   TRY
-  (void)init_flags;  // unused for now; for future use
 
-  amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance();
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
+  smi.Initialize(flags);
+
   return RSMI_STATUS_SUCCESS;
   CATCH
 }
@@ -430,6 +432,11 @@ rsmi_init(uint64_t init_flags) {
 rsmi_status_t
 rsmi_shut_down(void) {
   TRY
+
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
+
+  smi.Cleanup();
+
   return RSMI_STATUS_SUCCESS;
   CATCH
 }
@@ -441,7 +448,7 @@ rsmi_num_monitor_devices(uint32_t *num_devices) {
     return RSMI_STATUS_INVALID_ARGS;
   }
 
-  amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance();
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
 
   *num_devices = smi.monitor_devices().size();
   return RSMI_STATUS_SUCCESS;
@@ -1086,7 +1093,7 @@ rsmi_dev_gpu_clk_freq_set(uint32_t dv_ind,
 
   assert(freqs.num_supported <= RSMI_MAX_NUM_FREQUENCIES);
 
-  amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance();
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
 
   // Above call to rsmi_dev_get_gpu_clk_freq should have emitted an error if
   // assert below is not true
@@ -1366,7 +1373,7 @@ rsmi_dev_pci_bandwidth_set(uint32_t dv_ind, uint64_t bw_bitmask) {
 
   assert(bws.transfer_rate.num_supported <= RSMI_MAX_NUM_FREQUENCIES);
 
-  amd::smi::RocmSMI smi = amd::smi::RocmSMI::getInstance();
+  amd::smi::RocmSMI& smi = amd::smi::RocmSMI::getInstance();
 
   // Above call to rsmi_dev_pci_bandwidth_get() should have emitted an error
   // if assert below is not true

src/rocm_smi_main.cc

@@ -234,10 +234,13 @@ static uint32_t GetMonitorDevices(const std::shared_ptr<amd::smi::Device> &d,
 
 std::vector<std::shared_ptr<amd::smi::Device>> RocmSMI::s_monitor_devices;
 
-RocmSMI::RocmSMI(void) {
+void
+RocmSMI::Initialize(uint64_t flags) {
   auto i = 0;
   uint32_t ret;
 
+  init_options_ = flags;
+
   GetEnvVariables();
 
   while (std::string(kAMDMonitorTypes[i]) != "") {
@@ -260,15 +263,23 @@ RocmSMI::RocmSMI(void) {
   }
 }
 
-RocmSMI::~RocmSMI() {
+void
+RocmSMI::Cleanup() {
+  s_monitor_devices.clear();
   devices_.clear();
   monitors_.clear();
 }
 
-RocmSMI& RocmSMI::getInstance(void) {
+RocmSMI::RocmSMI(uint64_t flags) : init_options_(flags) {
+}
+
+RocmSMI::~RocmSMI() {
+}
+
+RocmSMI& RocmSMI::getInstance(uint64_t flags) {
   // Assume c++11 or greater. static objects will be created by only 1 thread
   // and creation will be thread-safe.
-  static RocmSMI singleton;
+  static RocmSMI singleton(flags);
   return singleton;
 }
 
@@ -324,6 +335,33 @@ RocmSMI::AddToDeviceList(std::string dev_name) {
   return;
 }
 
+static const uint32_t kAmdGpuId=0x1002;
+
+static bool isAMDGPU(std::string dev_path) {
+
+  std::string vend_path = dev_path + "/device/vendor";
+  if (!FileExists(vend_path.c_str())) {
+    return false;
+  }
+
+  std::ifstream fs;
+  fs.open(vend_path);
+
+  if (!fs.is_open()) {
+      return errno;
+  }
+
+  uint32_t vendor_id;
+
+  fs >> std::hex >> vendor_id;
+
+  fs.close();
+
+  if (vendor_id == kAmdGpuId) {
+    return true;
+  }
+  return false;
+}
 
 uint32_t RocmSMI::DiscoverDevices(void) {
   auto ret = 0;
@@ -346,7 +384,14 @@ uint32_t RocmSMI::DiscoverDevices(void) {
   while (dentry != nullptr) {
     if (memcmp(dentry->d_name, kDeviceNamePrefix, strlen(kDeviceNamePrefix))
                                                                        == 0) {
-      AddToDeviceList(dentry->d_name);
+      std::string vend_str_path = kPathDRMRoot;
+      vend_str_path += "/";
+      vend_str_path += dentry->d_name;
+
+      if (isAMDGPU(vend_str_path) ||
+          (init_options_ & RSMI_INIT_FLAG_ALL_GPUS)) {
+        AddToDeviceList(dentry->d_name);
+      }
     }
     dentry = readdir(drm_dir);
   }