Fix: use cuda_device_handle in cu_memory_provider_initialize

cu_provider->device is not set (equals 0) in this place yet,
so we have to use cu_params->cuda_device_handle instead of
cu_provider->device here.

Signed-off-by: Lukasz Dorau <[email protected]>

Author: ldorau

src/provider/provider_cuda.c

@@ -313,7 +313,7 @@ static umf_result_t cu_memory_provider_initialize(void *params,
     CUmemAllocationProp allocProps = {0};
     allocProps.location.type = CU_MEM_LOCATION_TYPE_DEVICE;
     allocProps.type = CU_MEM_ALLOCATION_TYPE_PINNED;
-    allocProps.location.id = cu_provider->device;
+    allocProps.location.id = cu_params->cuda_device_handle;
     CUresult cu_result = g_cu_ops.cuMemGetAllocationGranularity(
         &min_alignment, &allocProps, CU_MEM_ALLOC_GRANULARITY_MINIMUM);
     if (cu_result != CUDA_SUCCESS) {