[Cuda] Add mem type query in device caps

Author: EgorOrachyov

cubool/include/cubool/cubool.h

@@ -118,6 +118,7 @@ typedef struct cuBool_Vector_t* cuBool_Vector;
 typedef struct cuBool_DeviceCaps {
     char name[256];
     bool cudaSupported;
+    bool managedMem;
     int major;
     int minor;
     int warp;

cubool/sources/core/library.cpp

@@ -249,6 +249,7 @@ namespace cubool {
     void Library::queryCapabilities(cuBool_DeviceCaps &caps) {
         caps.name[0] = '\0';
         caps.cudaSupported = false;
+        caps.managedMem = false;
         caps.major = 0;
         caps.minor = 0;
         caps.warp = 0;
@@ -272,6 +273,7 @@ namespace cubool {
                    << " name: " << caps.name << ","
                    << " major: " << caps.major << ","
                    << " minor: " << caps.minor << ","
+                   << " mem type: " << (caps.managedMem? "managed": "default") << ","
                    << " warp size: " << caps.warp << ","
                    << " globalMemoryKiBs: " << caps.globalMemoryKiBs << ","
                    << " sharedMemoryPerMultiProcKiBs: " << caps.sharedMemoryPerMultiProcKiBs << ","

cubool/sources/cuda/cuda_backend.cu

@@ -92,7 +92,7 @@ namespace cubool {
     }
 
     void CudaBackend::queryCapabilities(cuBool_DeviceCaps &caps) {
-        CudaInstance::queryDeviceCapabilities(caps);
+        mInstance->queryDeviceCapabilities(caps);
     }
 
     CudaInstance & CudaBackend::getInstance() {

cubool/sources/cuda/cuda_instance.cu

@@ -86,17 +86,7 @@ namespace cubool {
         }
     }
 
-    CudaInstance::MemType CudaInstance::getMemoryType() const {
-        return mMemoryType;
-    }
-
-    bool CudaInstance::isCudaDeviceSupported() {
-        int device;
-        cudaError error = cudaGetDevice(&device);
-        return error == cudaSuccess;
-    }
-
-    void CudaInstance::queryDeviceCapabilities(cuBool_DeviceCaps &deviceCaps) {
+    void CudaInstance::queryDeviceCapabilities(cuBool_DeviceCaps &deviceCaps) const {
         const unsigned long long KiB = 1024;
 
         int device;
@@ -109,6 +99,7 @@ namespace cubool {
             if (error == cudaSuccess) {
                 strcpy(deviceCaps.name, deviceProp.name);
                 deviceCaps.cudaSupported = true;
+                deviceCaps.managedMem = mMemoryType == MemType::Managed;
                 deviceCaps.minor = deviceProp.minor;
                 deviceCaps.major = deviceProp.major;
                 deviceCaps.warp = deviceProp.warpSize;
@@ -119,6 +110,16 @@ namespace cubool {
         }
     }
 
+    CudaInstance::MemType CudaInstance::getMemoryType() const {
+        return mMemoryType;
+    }
+
+    bool CudaInstance::isCudaDeviceSupported() {
+        int device;
+        cudaError error = cudaGetDevice(&device);
+        return error == cudaSuccess;
+    }
+
     void CudaInstance::allocate(void* &ptr, size_t size) const {
         ptr = malloc(size);
         CHECK_RAISE_ERROR(ptr != nullptr, MemOpFailed, "Failed to allocate memory on the CPU");

cubool/sources/cuda/cuda_instance.hpp

@@ -51,12 +51,12 @@ namespace cubool {
         void allocateOnGpu(void* &ptr, size_t s) const;
         void deallocate(void* ptr) const;
         void deallocateOnGpu(void* ptr) const;
+        void queryDeviceCapabilities(cuBool_DeviceCaps& deviceCaps) const;
 
         void syncHostDevice() const;
         MemType getMemoryType() const;
 
         static bool isCudaDeviceSupported();
-        static void queryDeviceCapabilities(cuBool_DeviceCaps& deviceCaps);
         static CudaInstance& getInstanceRef();
         static CudaInstance* getInstancePtr();
         static bool isInstancePresent();