Merge branch 'add-get-cuda-driver-version' into 'main'

Add additional functions to nvml interfaces

See merge request nvidia/cloud-native/go-nvlib!18

go.mod

@@ -3,6 +3,6 @@ module gitlab.com/nvidia/cloud-native/go-nvlib
 go 1.16
 
 require (
-	github.com/NVIDIA/go-nvml v0.11.6-0.0.20220614115128-31f8b89eb740
+	github.com/NVIDIA/go-nvml v0.11.6-0.0.20220823120812-7e2082095e82
 	github.com/stretchr/testify v1.7.0
 )

go.sum

@@ -1,5 +1,5 @@
-github.com/NVIDIA/go-nvml v0.11.6-0.0.20220614115128-31f8b89eb740 h1:eOqFxx5XKIcw4U0YTIndTWWFCIJvuagS+v8IaM7XvBU=
-github.com/NVIDIA/go-nvml v0.11.6-0.0.20220614115128-31f8b89eb740/go.mod h1:hy7HYeQy335x6nEss0Ne3PYqleRa6Ct+VKD9RQ4nyFs=
+github.com/NVIDIA/go-nvml v0.11.6-0.0.20220823120812-7e2082095e82 h1:x751Xx1tdxkiA/sdkv2J769n21UbYKzVOpe9S/h1M3k=
+github.com/NVIDIA/go-nvml v0.11.6-0.0.20220823120812-7e2082095e82/go.mod h1:hy7HYeQy335x6nEss0Ne3PYqleRa6Ct+VKD9RQ4nyFs=
 github.com/davecgh/go-spew v1.1.0 h1:ZDRjVQ15GmhC3fiQ8ni8+OwkZQO4DARzQgrnXU1Liz8=
 github.com/davecgh/go-spew v1.1.0/go.mod h1:J7Y8YcW2NihsgmVo/mv3lAwl/skON4iLHjSsI+c5H38=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=

pkg/nvml/device.go

@@ -115,3 +115,21 @@ func (d nvmlDevice) GetComputeInstanceId() (int, Return) {
 	ci, r := nvml.Device(d).GetComputeInstanceId()
 	return ci, Return(r)
 }
+
+// GetCudaComputeCapability returns the compute capability major and minor versions for a device
+func (d nvmlDevice) GetCudaComputeCapability() (int, int, Return) {
+	major, minor, r := nvml.Device(d).GetCudaComputeCapability()
+	return major, minor, Return(r)
+}
+
+// GetAttributes returns the device attributes for a MIG device
+func (d nvmlDevice) GetAttributes() (DeviceAttributes, Return) {
+	a, r := nvml.Device(d).GetAttributes()
+	return DeviceAttributes(a), Return(r)
+}
+
+// GetName returns the device attributes for a MIG device
+func (d nvmlDevice) GetName() (string, Return) {
+	n, r := nvml.Device(d).GetName()
+	return n, Return(r)
+}

pkg/nvml/device_mock.go

@@ -17,9 +17,15 @@ var _ Device = &DeviceMock{}
 //
 // 		// make and configure a mocked Device
 // 		mockedDevice := &DeviceMock{
+// 			GetAttributesFunc: func() (DeviceAttributes, Return) {
+// 				panic("mock out the GetAttributes method")
+// 			},
 // 			GetComputeInstanceIdFunc: func() (int, Return) {
 // 				panic("mock out the GetComputeInstanceId method")
 // 			},
+// 			GetCudaComputeCapabilityFunc: func() (int, int, Return) {
+// 				panic("mock out the GetCudaComputeCapability method")
+// 			},
 // 			GetDeviceHandleFromMigDeviceHandleFunc: func() (Device, Return) {
 // 				panic("mock out the GetDeviceHandleFromMigDeviceHandle method")
 // 			},
@@ -50,6 +56,9 @@ var _ Device = &DeviceMock{}
 // 			GetMinorNumberFunc: func() (int, Return) {
 // 				panic("mock out the GetMinorNumber method")
 // 			},
+// 			GetNameFunc: func() (string, Return) {
+// 				panic("mock out the GetName method")
+// 			},
 // 			GetPciInfoFunc: func() (PciInfo, Return) {
 // 				panic("mock out the GetPciInfo method")
 // 			},
@@ -69,9 +78,15 @@ var _ Device = &DeviceMock{}
 //
 // 	}
 type DeviceMock struct {
+	// GetAttributesFunc mocks the GetAttributes method.
+	GetAttributesFunc func() (DeviceAttributes, Return)
+
 	// GetComputeInstanceIdFunc mocks the GetComputeInstanceId method.
 	GetComputeInstanceIdFunc func() (int, Return)
 
+	// GetCudaComputeCapabilityFunc mocks the GetCudaComputeCapability method.
+	GetCudaComputeCapabilityFunc func() (int, int, Return)
+
 	// GetDeviceHandleFromMigDeviceHandleFunc mocks the GetDeviceHandleFromMigDeviceHandle method.
 	GetDeviceHandleFromMigDeviceHandleFunc func() (Device, Return)
 
@@ -102,6 +117,9 @@ type DeviceMock struct {
 	// GetMinorNumberFunc mocks the GetMinorNumber method.
 	GetMinorNumberFunc func() (int, Return)
 
+	// GetNameFunc mocks the GetName method.
+	GetNameFunc func() (string, Return)
+
 	// GetPciInfoFunc mocks the GetPciInfo method.
 	GetPciInfoFunc func() (PciInfo, Return)
 
@@ -116,9 +134,15 @@ type DeviceMock struct {
 
 	// calls tracks calls to the methods.
 	calls struct {
+		// GetAttributes holds details about calls to the GetAttributes method.
+		GetAttributes []struct {
+		}
 		// GetComputeInstanceId holds details about calls to the GetComputeInstanceId method.
 		GetComputeInstanceId []struct {
 		}
+		// GetCudaComputeCapability holds details about calls to the GetCudaComputeCapability method.
+		GetCudaComputeCapability []struct {
+		}
 		// GetDeviceHandleFromMigDeviceHandle holds details about calls to the GetDeviceHandleFromMigDeviceHandle method.
 		GetDeviceHandleFromMigDeviceHandle []struct {
 		}
@@ -155,6 +179,9 @@ type DeviceMock struct {
 		// GetMinorNumber holds details about calls to the GetMinorNumber method.
 		GetMinorNumber []struct {
 		}
+		// GetName holds details about calls to the GetName method.
+		GetName []struct {
+		}
 		// GetPciInfo holds details about calls to the GetPciInfo method.
 		GetPciInfo []struct {
 		}
@@ -170,7 +197,9 @@ type DeviceMock struct {
 			Mode int
 		}
 	}
+	lockGetAttributes                      sync.RWMutex
 	lockGetComputeInstanceId               sync.RWMutex
+	lockGetCudaComputeCapability           sync.RWMutex
 	lockGetDeviceHandleFromMigDeviceHandle sync.RWMutex
 	lockGetGpuInstanceId                   sync.RWMutex
 	lockGetGpuInstanceProfileInfo          sync.RWMutex
@@ -181,12 +210,39 @@ type DeviceMock struct {
 	lockGetMigDeviceHandleByIndex          sync.RWMutex
 	lockGetMigMode                         sync.RWMutex
 	lockGetMinorNumber                     sync.RWMutex
+	lockGetName                            sync.RWMutex
 	lockGetPciInfo                         sync.RWMutex
 	lockGetUUID                            sync.RWMutex
 	lockIsMigDeviceHandle                  sync.RWMutex
 	lockSetMigMode                         sync.RWMutex
 }
 
+// GetAttributes calls GetAttributesFunc.
+func (mock *DeviceMock) GetAttributes() (DeviceAttributes, Return) {
+	if mock.GetAttributesFunc == nil {
+		panic("DeviceMock.GetAttributesFunc: method is nil but Device.GetAttributes was just called")
+	}
+	callInfo := struct {
+	}{}
+	mock.lockGetAttributes.Lock()
+	mock.calls.GetAttributes = append(mock.calls.GetAttributes, callInfo)
+	mock.lockGetAttributes.Unlock()
+	return mock.GetAttributesFunc()
+}
+
+// GetAttributesCalls gets all the calls that were made to GetAttributes.
+// Check the length with:
+//     len(mockedDevice.GetAttributesCalls())
+func (mock *DeviceMock) GetAttributesCalls() []struct {
+} {
+	var calls []struct {
+	}
+	mock.lockGetAttributes.RLock()
+	calls = mock.calls.GetAttributes
+	mock.lockGetAttributes.RUnlock()
+	return calls
+}
+
 // GetComputeInstanceId calls GetComputeInstanceIdFunc.
 func (mock *DeviceMock) GetComputeInstanceId() (int, Return) {
 	if mock.GetComputeInstanceIdFunc == nil {
@@ -213,6 +269,32 @@ func (mock *DeviceMock) GetComputeInstanceIdCalls() []struct {
 	return calls
 }
 
+// GetCudaComputeCapability calls GetCudaComputeCapabilityFunc.
+func (mock *DeviceMock) GetCudaComputeCapability() (int, int, Return) {
+	if mock.GetCudaComputeCapabilityFunc == nil {
+		panic("DeviceMock.GetCudaComputeCapabilityFunc: method is nil but Device.GetCudaComputeCapability was just called")
+	}
+	callInfo := struct {
+	}{}
+	mock.lockGetCudaComputeCapability.Lock()
+	mock.calls.GetCudaComputeCapability = append(mock.calls.GetCudaComputeCapability, callInfo)
+	mock.lockGetCudaComputeCapability.Unlock()
+	return mock.GetCudaComputeCapabilityFunc()
+}
+
+// GetCudaComputeCapabilityCalls gets all the calls that were made to GetCudaComputeCapability.
+// Check the length with:
+//     len(mockedDevice.GetCudaComputeCapabilityCalls())
+func (mock *DeviceMock) GetCudaComputeCapabilityCalls() []struct {
+} {
+	var calls []struct {
+	}
+	mock.lockGetCudaComputeCapability.RLock()
+	calls = mock.calls.GetCudaComputeCapability
+	mock.lockGetCudaComputeCapability.RUnlock()
+	return calls
+}
+
 // GetDeviceHandleFromMigDeviceHandle calls GetDeviceHandleFromMigDeviceHandleFunc.
 func (mock *DeviceMock) GetDeviceHandleFromMigDeviceHandle() (Device, Return) {
 	if mock.GetDeviceHandleFromMigDeviceHandleFunc == nil {
@@ -488,6 +570,32 @@ func (mock *DeviceMock) GetMinorNumberCalls() []struct {
 	return calls
 }
 
+// GetName calls GetNameFunc.
+func (mock *DeviceMock) GetName() (string, Return) {
+	if mock.GetNameFunc == nil {
+		panic("DeviceMock.GetNameFunc: method is nil but Device.GetName was just called")
+	}
+	callInfo := struct {
+	}{}
+	mock.lockGetName.Lock()
+	mock.calls.GetName = append(mock.calls.GetName, callInfo)
+	mock.lockGetName.Unlock()
+	return mock.GetNameFunc()
+}
+
+// GetNameCalls gets all the calls that were made to GetName.
+// Check the length with:
+//     len(mockedDevice.GetNameCalls())
+func (mock *DeviceMock) GetNameCalls() []struct {
+} {
+	var calls []struct {
+	}
+	mock.lockGetName.RLock()
+	calls = mock.calls.GetName
+	mock.lockGetName.RUnlock()
+	return calls
+}
+
 // GetPciInfo calls GetPciInfoFunc.
 func (mock *DeviceMock) GetPciInfo() (PciInfo, Return) {
 	if mock.GetPciInfoFunc == nil {

pkg/nvml/nvml.go

@@ -46,7 +46,7 @@ func (n *nvmlLib) Init() Return {
 	if n.refcount == 0 {
 		errorStringFunc = nvml.ErrorString
 	}
-	n.refcount += 1
+	n.refcount++
 
 	return SUCCESS
 }
@@ -60,7 +60,7 @@ func (n *nvmlLib) Shutdown() Return {
 
 	n.Lock()
 	defer n.Unlock()
-	n.refcount -= 1
+	n.refcount--
 	if n.refcount == 0 {
 		errorStringFunc = defaultErrorStringFunc
 	}
@@ -92,6 +92,12 @@ func (n *nvmlLib) SystemGetDriverVersion() (string, Return) {
 	return v, Return(r)
 }
 
+// SystemGetCudaDriverVersion returns the version of CUDA associated with the NVIDIA driver
+func (n *nvmlLib) SystemGetCudaDriverVersion() (int, Return) {
+	v, r := nvml.SystemGetCudaDriverVersion()
+	return v, Return(r)
+}
+
 // ErrorString returns the error string associated with a given return value
 func (n *nvmlLib) ErrorString(ret Return) string {
 	return nvml.ErrorString(nvml.Return(ret))

pkg/nvml/nvml_mock.go

@@ -35,6 +35,9 @@ var _ Interface = &InterfaceMock{}
 // 			ShutdownFunc: func() Return {
 // 				panic("mock out the Shutdown method")
 // 			},
+// 			SystemGetCudaDriverVersionFunc: func() (int, Return) {
+// 				panic("mock out the SystemGetCudaDriverVersion method")
+// 			},
 // 			SystemGetDriverVersionFunc: func() (string, Return) {
 // 				panic("mock out the SystemGetDriverVersion method")
 // 			},
@@ -63,6 +66,9 @@ type InterfaceMock struct {
 	// ShutdownFunc mocks the Shutdown method.
 	ShutdownFunc func() Return
 
+	// SystemGetCudaDriverVersionFunc mocks the SystemGetCudaDriverVersion method.
+	SystemGetCudaDriverVersionFunc func() (int, Return)
+
 	// SystemGetDriverVersionFunc mocks the SystemGetDriverVersion method.
 	SystemGetDriverVersionFunc func() (string, Return)
 
@@ -92,17 +98,21 @@ type InterfaceMock struct {
 		// Shutdown holds details about calls to the Shutdown method.
 		Shutdown []struct {
 		}
+		// SystemGetCudaDriverVersion holds details about calls to the SystemGetCudaDriverVersion method.
+		SystemGetCudaDriverVersion []struct {
+		}
 		// SystemGetDriverVersion holds details about calls to the SystemGetDriverVersion method.
 		SystemGetDriverVersion []struct {
 		}
 	}
-	lockDeviceGetCount         sync.RWMutex
-	lockDeviceGetHandleByIndex sync.RWMutex
-	lockDeviceGetHandleByUUID  sync.RWMutex
-	lockErrorString            sync.RWMutex
-	lockInit                   sync.RWMutex
-	lockShutdown               sync.RWMutex
-	lockSystemGetDriverVersion sync.RWMutex
+	lockDeviceGetCount             sync.RWMutex
+	lockDeviceGetHandleByIndex     sync.RWMutex
+	lockDeviceGetHandleByUUID      sync.RWMutex
+	lockErrorString                sync.RWMutex
+	lockInit                       sync.RWMutex
+	lockShutdown                   sync.RWMutex
+	lockSystemGetCudaDriverVersion sync.RWMutex
+	lockSystemGetDriverVersion     sync.RWMutex
 }
 
 // DeviceGetCount calls DeviceGetCountFunc.
@@ -276,6 +286,32 @@ func (mock *InterfaceMock) ShutdownCalls() []struct {
 	return calls
 }
 
+// SystemGetCudaDriverVersion calls SystemGetCudaDriverVersionFunc.
+func (mock *InterfaceMock) SystemGetCudaDriverVersion() (int, Return) {
+	if mock.SystemGetCudaDriverVersionFunc == nil {
+		panic("InterfaceMock.SystemGetCudaDriverVersionFunc: method is nil but Interface.SystemGetCudaDriverVersion was just called")
+	}
+	callInfo := struct {
+	}{}
+	mock.lockSystemGetCudaDriverVersion.Lock()
+	mock.calls.SystemGetCudaDriverVersion = append(mock.calls.SystemGetCudaDriverVersion, callInfo)
+	mock.lockSystemGetCudaDriverVersion.Unlock()
+	return mock.SystemGetCudaDriverVersionFunc()
+}
+
+// SystemGetCudaDriverVersionCalls gets all the calls that were made to SystemGetCudaDriverVersion.
+// Check the length with:
+//     len(mockedInterface.SystemGetCudaDriverVersionCalls())
+func (mock *InterfaceMock) SystemGetCudaDriverVersionCalls() []struct {
+} {
+	var calls []struct {
+	}
+	mock.lockSystemGetCudaDriverVersion.RLock()
+	calls = mock.calls.SystemGetCudaDriverVersion
+	mock.lockSystemGetCudaDriverVersion.RUnlock()
+	return calls
+}
+
 // SystemGetDriverVersion calls SystemGetDriverVersionFunc.
 func (mock *InterfaceMock) SystemGetDriverVersion() (string, Return) {
 	if mock.SystemGetDriverVersionFunc == nil {

pkg/nvml/types.go

@@ -20,53 +20,61 @@ import (
 	"github.com/NVIDIA/go-nvml/pkg/nvml"
 )
 
-//go:generate moq -out nvml_mock.go . Interface
 // Interface defines the functions implemented by an NVML library
+//
+//go:generate moq -out nvml_mock.go . Interface
 type Interface interface {
-	Init() Return
-	Shutdown() Return
 	DeviceGetCount() (int, Return)
 	DeviceGetHandleByIndex(Index int) (Device, Return)
 	DeviceGetHandleByUUID(UUID string) (Device, Return)
-	SystemGetDriverVersion() (string, Return)
 	ErrorString(r Return) string
+	Init() Return
+	Shutdown() Return
+	SystemGetCudaDriverVersion() (int, Return)
+	SystemGetDriverVersion() (string, Return)
 }
 
-//go:generate moq -out device_mock.go . Device
 // Device defines the functions implemented by an NVML device
+//
+//go:generate moq -out device_mock.go . Device
 type Device interface {
-	GetIndex() (int, Return)
-	GetPciInfo() (PciInfo, Return)
-	GetMemoryInfo() (Memory, Return)
-	GetUUID() (string, Return)
-	GetMinorNumber() (int, Return)
-	IsMigDeviceHandle() (bool, Return)
+	GetAttributes() (DeviceAttributes, Return)
+	GetComputeInstanceId() (int, Return)
+	GetCudaComputeCapability() (int, int, Return)
 	GetDeviceHandleFromMigDeviceHandle() (Device, Return)
-	SetMigMode(Mode int) (Return, Return)
-	GetMigMode() (int, int, Return)
+	GetGpuInstanceId() (int, Return)
 	GetGpuInstanceProfileInfo(Profile int) (GpuInstanceProfileInfo, Return)
 	GetGpuInstances(Info *GpuInstanceProfileInfo) ([]GpuInstance, Return)
+	GetIndex() (int, Return)
 	GetMaxMigDeviceCount() (int, Return)
+	GetMemoryInfo() (Memory, Return)
 	GetMigDeviceHandleByIndex(Index int) (Device, Return)
-	GetGpuInstanceId() (int, Return)
-	GetComputeInstanceId() (int, Return)
+	GetMigMode() (int, int, Return)
+	GetMinorNumber() (int, Return)
+	GetName() (string, Return)
+	GetPciInfo() (PciInfo, Return)
+	GetUUID() (string, Return)
+	IsMigDeviceHandle() (bool, Return)
+	SetMigMode(Mode int) (Return, Return)
 }
 
-//go:generate moq -out gi_mock.go . GpuInstance
 // GpuInstance defines the functions implemented by a GpuInstance
+//
+//go:generate moq -out gi_mock.go . GpuInstance
 type GpuInstance interface {
-	GetInfo() (GpuInstanceInfo, Return)
-	GetComputeInstanceProfileInfo(Profile int, EngProfile int) (ComputeInstanceProfileInfo, Return)
 	CreateComputeInstance(Info *ComputeInstanceProfileInfo) (ComputeInstance, Return)
-	GetComputeInstances(Info *ComputeInstanceProfileInfo) ([]ComputeInstance, Return)
 	Destroy() Return
+	GetComputeInstanceProfileInfo(Profile int, EngProfile int) (ComputeInstanceProfileInfo, Return)
+	GetComputeInstances(Info *ComputeInstanceProfileInfo) ([]ComputeInstance, Return)
+	GetInfo() (GpuInstanceInfo, Return)
 }
 
-//go:generate moq -out ci_mock.go . ComputeInstance
 // ComputeInstance defines the functions implemented by a ComputeInstance
+//
+//go:generate moq -out ci_mock.go . ComputeInstance
 type ComputeInstance interface {
-	GetInfo() (ComputeInstanceInfo, Return)
 	Destroy() Return
+	GetInfo() (ComputeInstanceInfo, Return)
 }
 
 // GpuInstanceInfo holds info about a GPU Instance
@@ -92,7 +100,7 @@ type Return nvml.Return
 // Memory holds info about GPU device memory
 type Memory nvml.Memory
 
-//PciInfo holds info about the PCI connections of a GPU dvice
+// PciInfo holds info about the PCI connections of a GPU dvice
 type PciInfo nvml.PciInfo
 
 // GpuInstanceProfileInfo holds info about a GPU Instance Profile
@@ -106,3 +114,6 @@ type ComputeInstanceProfileInfo nvml.ComputeInstanceProfileInfo
 
 // ComputeInstancePlacement holds placement info about a Compute Instance
 type ComputeInstancePlacement nvml.ComputeInstancePlacement
+
+// DeviceAttributes stores information about MIG devices
+type DeviceAttributes nvml.DeviceAttributes

vendor/github.com/NVIDIA/go-nvml/pkg/nvml/device.go

@@ -38,21 +38,21 @@ func DeviceGetHandleByIndex(Index int) (Device, Return) {
 // nvml.DeviceGetHandleBySerial()
 func DeviceGetHandleBySerial(Serial string) (Device, Return) {
 	var Device Device
-	ret := nvmlDeviceGetHandleBySerial(Serial, &Device)
+	ret := nvmlDeviceGetHandleBySerial(Serial + string(rune(0)), &Device)
 	return Device, ret
 }
 
 // nvml.DeviceGetHandleByUUID()
 func DeviceGetHandleByUUID(Uuid string) (Device, Return) {
 	var Device Device
-	ret := nvmlDeviceGetHandleByUUID(Uuid, &Device)
+	ret := nvmlDeviceGetHandleByUUID(Uuid + string(rune(0)), &Device)
 	return Device, ret
 }
 
 // nvml.DeviceGetHandleByPciBusId()
 func DeviceGetHandleByPciBusId(PciBusId string) (Device, Return) {
 	var Device Device
-	ret := nvmlDeviceGetHandleByPciBusId(PciBusId, &Device)
+	ret := nvmlDeviceGetHandleByPciBusId(PciBusId + string(rune(0)), &Device)
 	return Device, ret
 }
 

vendor/github.com/NVIDIA/go-nvml/pkg/nvml/init.go

@@ -32,13 +32,9 @@ var nvml *dl.DynamicLibrary
 // nvml.Init()
 func Init() Return {
 	lib := dl.New(nvmlLibraryName, nvmlLibraryLoadFlags)
-	if lib == nil {
-		panic(fmt.Sprintf("error instantiating DynamicLibrary for %s", nvmlLibraryName))
-	}
-
 	err := lib.Open()
 	if err != nil {
-		panic(fmt.Sprintf("error opening %s: %v", nvmlLibraryName, err))
+		return ERROR_LIBRARY_NOT_FOUND
 	}
 
 	nvml = lib
@@ -50,13 +46,9 @@ func Init() Return {
 // nvml.InitWithFlags()
 func InitWithFlags(Flags uint32) Return {
 	lib := dl.New(nvmlLibraryName, nvmlLibraryLoadFlags)
-	if lib == nil {
-		panic(fmt.Sprintf("error instantiating DynamicLibrary for %s", nvmlLibraryName))
-	}
-
 	err := lib.Open()
 	if err != nil {
-		panic(fmt.Sprintf("error opening %s: %v", nvmlLibraryName, err))
+		return ERROR_LIBRARY_NOT_FOUND
 	}
 
 	nvml = lib

vendor/modules.txt

@@ -1,4 +1,4 @@
-# github.com/NVIDIA/go-nvml v0.11.6-0.0.20220614115128-31f8b89eb740
+# github.com/NVIDIA/go-nvml v0.11.6-0.0.20220823120812-7e2082095e82
 ## explicit
 github.com/NVIDIA/go-nvml/pkg/dl
 github.com/NVIDIA/go-nvml/pkg/nvml