Module memory 

Device and host memory management.

Implements the CUDA Runtime memory API:

• cudaMalloc / cudaFree
• cudaMallocHost / cudaFreeHost (pinned host memory)
• cudaMallocManaged (unified memory)
• cudaMallocPitch (pitched 2-D allocation)
• cudaMemcpy / cudaMemcpyAsync
• cudaMemset / cudaMemsetAsync
• cudaMemGetInfo

All memory addresses returned for device allocations are represented as DevicePtr, a newtype around u64 that matches the driver API’s CUdeviceptr.

Structs

DevicePtr

Opaque CUDA device-memory address (mirrors CUdeviceptr).

Enums

MemAttachFlags

Flags for cudaMallocManaged.

MemcpyKind

Direction of a cudaMemcpy transfer.

Functions

free

Free device memory previously allocated with malloc.

free_host^⚠

Free page-locked host memory previously allocated with malloc_host.

malloc

Allocate size bytes of device memory.

malloc_host

Allocate size bytes of pinned (page-locked) host memory.

malloc_managed

Allocate unified managed memory accessible from both CPU and GPU.

malloc_pitch

Allocate pitched device memory for 2-D arrays.

mem_get_info

Returns (free_bytes, total_bytes) for the current device’s global memory.

memcpy^⚠

Synchronously copy count bytes between memory regions.

memcpy_async^⚠

Asynchronously copy count bytes on stream.

memcpy_d2d

Copy between two device allocations.

memcpy_d2h

Copy device memory to a host slice.

memcpy_h2d

Copy a slice of host data to a device allocation.

memset

Set count bytes of device memory starting at ptr to value.

memset32

Set device memory to 32-bit value pattern.