Add getTotalMem GpuMgr method

Author: kurapov-peter

omniscidb/CudaMgr/CudaMgr.h

@@ -151,6 +151,10 @@ class CudaMgr : public GpuMgr {
   bool isArchMaxwellOrLaterForAll() const;
   bool isArchVoltaOrGreaterForAll() const;
 
+  size_t getTotalMem(const int device_num) const {
+    return getDeviceProperties(device_num)->globalMem;
+  }
+
   uint32_t getMaxBlockSize() const override {
     return getAllDeviceProperties().front().maxThreadsPerBlock;
   }

omniscidb/DataMgr/DataMgr.cpp

@@ -282,19 +282,9 @@ void DataMgr::populateMgrs(const Config& config,
       int num_gpus = mgr->getDeviceCount();
       device_context->gpu_count = num_gpus;
       for (int gpu_num = 0; gpu_num < num_gpus; ++gpu_num) {
-        size_t device_mem_size = 0;
-        // TODO: get rid of manager-specific branches by introducing some kind of device
-        // properties in GpuMgr
-        switch (p) {
-          case GpuMgrPlatform::CUDA:
-            device_mem_size = getCudaMgr()->getDeviceProperties(gpu_num)->globalMem;
-            break;
-          case GpuMgrPlatform::L0:
-            device_mem_size = getL0Mgr()->getMaxAllocationSize(gpu_num);
-            page_size = getL0Mgr()->getPageSize(gpu_num);
-            break;
-          default:
-            CHECK(false);
+        size_t device_mem_size = mgr->getTotalMem(gpu_num);
+        if (p == GpuMgrPlatform::L0) {
+          page_size = getL0Mgr()->getPageSize(gpu_num);
         }
 
         size_t gpu_max_mem_size = config.mem.gpu.max_size;

omniscidb/DataMgr/GpuMgr.h

@@ -57,6 +57,7 @@ struct GpuMgr {
   virtual void synchronizeDevices() const = 0;
   virtual int getDeviceCount() const = 0;
   virtual GpuMgrPlatform getPlatform() const = 0;
+  virtual size_t getTotalMem(const int device_num) const = 0;
   virtual uint32_t getMaxBlockSize() const = 0;
   virtual int8_t getSubGroupSize() const = 0;
   virtual uint32_t getGridSize() const = 0;

omniscidb/L0Mgr/L0Mgr.cpp

@@ -419,6 +419,20 @@ size_t L0Manager::getMaxAllocationSize(const int device_num) const {
   return device_properties.maxMemAllocSize;
 }
 
+size_t L0Manager::getTotalMem(const int device_num) const {
+  CHECK_GE(device_num, 0);
+  CHECK_LT(device_num, drivers_[0]->devices().size());
+  uint32_t p_count{0};
+  L0_SAFE_CALL(zeDeviceGetMemoryProperties(
+      drivers_[0]->devices()[device_num]->device(), &p_count, nullptr));
+  CHECK_GT(p_count, 0u);
+  std::vector<ze_device_memory_properties_t> mem_properties;
+  mem_properties.resize(p_count);
+  L0_SAFE_CALL(zeDeviceGetMemoryProperties(
+      drivers_[0]->devices()[device_num]->device(), &p_count, mem_properties.data()));
+  return static_cast<size_t>(mem_properties[0].totalSize);
+}
+
 uint32_t L0Manager::getMaxBlockSize() const {
   CHECK_GT(drivers_[0]->devices().size(), size_t(0));
   unsigned sz = drivers_[0]->devices()[0]->maxGroupSize();

omniscidb/L0Mgr/L0Mgr.h

@@ -237,6 +237,7 @@ class L0Manager : public GpuMgr {
   };
 
   size_t getMaxAllocationSize(const int device_num) const;
+  size_t getTotalMem(const int device_num) const override;
   size_t getPageSize(const int device_num) const { return 4096u; }
 
   uint32_t getMaxBlockSize() const override;

omniscidb/L0Mgr/L0MgrNoL0.cpp

@@ -118,6 +118,11 @@ size_t L0Manager::getMaxAllocationSize(const int device_num) const {
   return 0u;
 }
 
+size_t L0Manager::getTotalMem(const int device_num) const {
+  CHECK(false);
+  return 0u;
+}
+
 unsigned L0Manager::getMaxBlockSize() const {
   CHECK(false);
   return 0u;