Struct GpuUtils 

pub struct GpuUtils { /* private fields */ }

GPU computing utilities

Implementations

impl GpuUtils

pub fn new() -> Self

Create new GPU utilities

pub fn init_devices(&mut self) -> Result<(), GpuError>

Initialize GPU devices.

Behind the gpu feature, this walks the real oxicuda-driver device list. On a host with no CUDA driver installed, or with the gpu feature disabled, this always yields an empty device list – never a fabricated device. See enumerate_real_devices.

pub fn get_devices(&self) -> &[GpuDevice]

Get available GPU devices

pub fn get_device(&self, id: u32) -> Option<&GpuDevice>

Get device by ID

pub fn get_best_device(&self) -> Option<&GpuDevice>

Get best device for ML workloads: the one with the highest multiprocessor_count * clock_rate_mhz (a real-hardware-derived throughput proxy), falling back to the first enumerated device.

Unlike the pre-migration version, this no longer filters out “integrated” devices: the driver API this crate queries (oxicuda-driver’s DeviceInfo) has no integrated/discrete flag, so pretending to know that distinction would itself be a fabrication.

pub fn allocate_memory( &self, size: u64, device_id: u32, name: &str, ) -> Result<u64, GpuError>

Allocate GPU memory

pub fn free_memory(&self, ptr: u64) -> Result<(), GpuError>

Free GPU memory

pub fn get_memory_stats(&self) -> HashMap<u32, MemoryStats>

Get memory usage statistics

pub fn execute_kernel( &self, kernel: &GpuKernelInfo, ) -> Result<GpuKernelExecution, GpuError>

Execute GPU kernel

pub fn get_kernel_history(&self) -> Vec<GpuKernelExecution>

Get kernel execution history

pub fn get_performance_counters(&self) -> HashMap<String, f64>

Get performance counters

pub fn update_counter(&self, name: &str, value: f64)

Update performance counter

pub fn estimate_throughput( &self, device_id: u32, array_size: usize, operation: &str, ) -> f64

Get throughput estimate for array operations

pub fn should_use_gpu(&self, array_size: usize, operation: &str) -> bool

Check if operation should use GPU

pub fn get_utilization(&self) -> HashMap<u32, f64>

Get GPU utilization

pub fn cleanup(&self) -> Result<(), GpuError>

Cleanup all resources

Trait Implementations

impl Debug for GpuUtils

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

Formats the value using the given formatter.

impl Default for GpuUtils

fn default() -> Self

Returns the “default value” for a type.