Struct Device 

pub struct Device(/* private fields */);

A CUDA device (Runtime API view — a bare ordinal).

Implementations

impl Device

pub fn count() -> Result<u32>

Number of CUDA devices visible to the process.

pub const fn from_ordinal(ordinal: u32) -> Self

Construct a Device for the given ordinal. Does not validate — use Device::all if you want a checked enumeration.

pub fn all() -> Result<Vec<Self>>

All visible devices, in ordinal order.

Examples found in repository

examples/device_query.rs (line 22)

fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("CUDA driver version: {}", driver_version()?);
    println!("CUDA runtime version: {}", runtime_version()?);

    let devices = Device::all()?;
    println!("found {} CUDA device(s)", devices.len());

    for dev in &devices {
        // Switch current-thread device before querying properties:
        // `device_properties` reads via `cudaGetDeviceProperties`, which
        // is keyed off the ordinal we pass — but other Runtime calls
        // (memory queries, kernel launches) operate on the current
        // device, so it's good practice to bind first.
        dev.set_current()?;
        let props = device_properties(dev)?;
        let (cc_major, cc_minor) = dev.compute_capability()?;
        let sm_count = dev.multiprocessor_count()?;
        let warp = dev.warp_size()?;

        let gib = (props.total_global_memory_bytes as f64) / (1024.0 * 1024.0 * 1024.0);
        println!(
            "  [{}] {:30}  cc {}.{}  {} SMs  warp {}  {:.2} GiB",
            dev.ordinal(),
            props.name,
            cc_major,
            cc_minor,
            sm_count,
            warp,
            gib,
        );
    }

    // Make sure all pending work on every device (none here) is done.
    device_synchronize()?;
    println!("OK");
    Ok(())
}

pub fn set_current(&self) -> Result<()>

Set this device as current on the calling thread. Subsequent Runtime API calls (allocations, launches, …) operate on this device.

Examples found in repository

examples/device_query.rs (line 31)

fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("CUDA driver version: {}", driver_version()?);
    println!("CUDA runtime version: {}", runtime_version()?);

    let devices = Device::all()?;
    println!("found {} CUDA device(s)", devices.len());

    for dev in &devices {
        // Switch current-thread device before querying properties:
        // `device_properties` reads via `cudaGetDeviceProperties`, which
        // is keyed off the ordinal we pass — but other Runtime calls
        // (memory queries, kernel launches) operate on the current
        // device, so it's good practice to bind first.
        dev.set_current()?;
        let props = device_properties(dev)?;
        let (cc_major, cc_minor) = dev.compute_capability()?;
        let sm_count = dev.multiprocessor_count()?;
        let warp = dev.warp_size()?;

        let gib = (props.total_global_memory_bytes as f64) / (1024.0 * 1024.0 * 1024.0);
        println!(
            "  [{}] {:30}  cc {}.{}  {} SMs  warp {}  {:.2} GiB",
            dev.ordinal(),
            props.name,
            cc_major,
            cc_minor,
            sm_count,
            warp,
            gib,
        );
    }

    // Make sure all pending work on every device (none here) is done.
    device_synchronize()?;
    println!("OK");
    Ok(())
}

pub fn current() -> Result<Self>

Retrieve the device currently selected on the calling thread.

pub fn ordinal(&self) -> i32

Ordinal of this device (0, 1, …).

Examples found in repository

examples/device_query.rs (line 40)

fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("CUDA driver version: {}", driver_version()?);
    println!("CUDA runtime version: {}", runtime_version()?);

    let devices = Device::all()?;
    println!("found {} CUDA device(s)", devices.len());

    for dev in &devices {
        // Switch current-thread device before querying properties:
        // `device_properties` reads via `cudaGetDeviceProperties`, which
        // is keyed off the ordinal we pass — but other Runtime calls
        // (memory queries, kernel launches) operate on the current
        // device, so it's good practice to bind first.
        dev.set_current()?;
        let props = device_properties(dev)?;
        let (cc_major, cc_minor) = dev.compute_capability()?;
        let sm_count = dev.multiprocessor_count()?;
        let warp = dev.warp_size()?;

        let gib = (props.total_global_memory_bytes as f64) / (1024.0 * 1024.0 * 1024.0);
        println!(
            "  [{}] {:30}  cc {}.{}  {} SMs  warp {}  {:.2} GiB",
            dev.ordinal(),
            props.name,
            cc_major,
            cc_minor,
            sm_count,
            warp,
            gib,
        );
    }

    // Make sure all pending work on every device (none here) is done.
    device_synchronize()?;
    println!("OK");
    Ok(())
}

pub fn compute_capability(&self) -> Result<(u32, u32)>

Compute capability as (major, minor).

Examples found in repository

examples/device_query.rs (line 33)

fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("CUDA driver version: {}", driver_version()?);
    println!("CUDA runtime version: {}", runtime_version()?);

    let devices = Device::all()?;
    println!("found {} CUDA device(s)", devices.len());

    for dev in &devices {
        // Switch current-thread device before querying properties:
        // `device_properties` reads via `cudaGetDeviceProperties`, which
        // is keyed off the ordinal we pass — but other Runtime calls
        // (memory queries, kernel launches) operate on the current
        // device, so it's good practice to bind first.
        dev.set_current()?;
        let props = device_properties(dev)?;
        let (cc_major, cc_minor) = dev.compute_capability()?;
        let sm_count = dev.multiprocessor_count()?;
        let warp = dev.warp_size()?;

        let gib = (props.total_global_memory_bytes as f64) / (1024.0 * 1024.0 * 1024.0);
        println!(
            "  [{}] {:30}  cc {}.{}  {} SMs  warp {}  {:.2} GiB",
            dev.ordinal(),
            props.name,
            cc_major,
            cc_minor,
            sm_count,
            warp,
            gib,
        );
    }

    // Make sure all pending work on every device (none here) is done.
    device_synchronize()?;
    println!("OK");
    Ok(())
}

pub fn multiprocessor_count(&self) -> Result<u32>

Multiprocessor count.

Examples found in repository

examples/device_query.rs (line 34)

fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("CUDA driver version: {}", driver_version()?);
    println!("CUDA runtime version: {}", runtime_version()?);

    let devices = Device::all()?;
    println!("found {} CUDA device(s)", devices.len());

    for dev in &devices {
        // Switch current-thread device before querying properties:
        // `device_properties` reads via `cudaGetDeviceProperties`, which
        // is keyed off the ordinal we pass — but other Runtime calls
        // (memory queries, kernel launches) operate on the current
        // device, so it's good practice to bind first.
        dev.set_current()?;
        let props = device_properties(dev)?;
        let (cc_major, cc_minor) = dev.compute_capability()?;
        let sm_count = dev.multiprocessor_count()?;
        let warp = dev.warp_size()?;

        let gib = (props.total_global_memory_bytes as f64) / (1024.0 * 1024.0 * 1024.0);
        println!(
            "  [{}] {:30}  cc {}.{}  {} SMs  warp {}  {:.2} GiB",
            dev.ordinal(),
            props.name,
            cc_major,
            cc_minor,
            sm_count,
            warp,
            gib,
        );
    }

    // Make sure all pending work on every device (none here) is done.
    device_synchronize()?;
    println!("OK");
    Ok(())
}

pub fn warp_size(&self) -> Result<u32>

Warp size in threads.

Examples found in repository

examples/device_query.rs (line 35)

fn main() -> Result<(), Box<dyn std::error::Error>> {
    println!("CUDA driver version: {}", driver_version()?);
    println!("CUDA runtime version: {}", runtime_version()?);

    let devices = Device::all()?;
    println!("found {} CUDA device(s)", devices.len());

    for dev in &devices {
        // Switch current-thread device before querying properties:
        // `device_properties` reads via `cudaGetDeviceProperties`, which
        // is keyed off the ordinal we pass — but other Runtime calls
        // (memory queries, kernel launches) operate on the current
        // device, so it's good practice to bind first.
        dev.set_current()?;
        let props = device_properties(dev)?;
        let (cc_major, cc_minor) = dev.compute_capability()?;
        let sm_count = dev.multiprocessor_count()?;
        let warp = dev.warp_size()?;

        let gib = (props.total_global_memory_bytes as f64) / (1024.0 * 1024.0 * 1024.0);
        println!(
            "  [{}] {:30}  cc {}.{}  {} SMs  warp {}  {:.2} GiB",
            dev.ordinal(),
            props.name,
            cc_major,
            cc_minor,
            sm_count,
            warp,
            gib,
        );
    }

    // Make sure all pending work on every device (none here) is done.
    device_synchronize()?;
    println!("OK");
    Ok(())
}

pub fn attribute(&self, attr: i32) -> Result<i32>

Raw device-attribute query. See baracuda_cuda_sys::runtime::types::cudaDeviceAttr.

pub fn can_access_peer(&self, peer: &Device) -> Result<bool>

true if this device can peer-access peer’s allocations (P2P). Call Device::enable_peer_access before actually using peer pointers in kernels.

pub fn enable_peer_access(peer: &Device) -> Result<()>

Enable peer access from the current device to peer’s allocations. Call Device::set_current() on the accessing device first.

pub fn disable_peer_access(peer: &Device) -> Result<()>

Disable peer access previously enabled via Device::enable_peer_access.

Trait Implementations

impl Clone for Device

fn clone(&self) -> Device

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for Device

impl Debug for Device

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Eq for Device

impl Hash for Device

fn hash<__H: Hasher>(&self, state: &mut __H)

Feeds this value into the given Hasher.

fn hash_slice<H>(data: &[Self], state: &mut H)

where H: Hasher, Self: Sized,

Feeds a slice of this type into the given Hasher.

impl Ord for Device

fn cmp(&self, other: &Device) -> Ordering

This method returns an Ordering between self and other.

fn max(self, other: Self) -> Self

where Self: Sized,

Compares and returns the maximum of two values.

fn min(self, other: Self) -> Self

where Self: Sized,

Compares and returns the minimum of two values.

fn clamp(self, min: Self, max: Self) -> Self

where Self: Sized,

Restrict a value to a certain interval.

impl PartialEq for Device

fn eq(&self, other: &Device) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl PartialOrd for Device

fn partial_cmp(&self, other: &Device) -> Option<Ordering>

This method returns an ordering between self and other values if one exists.

fn lt(&self, other: &Rhs) -> bool

Tests less than (for self and other) and is used by the < operator.

fn le(&self, other: &Rhs) -> bool

Tests less than or equal to (for self and other) and is used by the <= operator.

fn gt(&self, other: &Rhs) -> bool

Tests greater than (for self and other) and is used by the > operator.

fn ge(&self, other: &Rhs) -> bool

Tests greater than or equal to (for self and other) and is used by the >= operator.

impl StructuralPartialEq for Device