Struct CudaBuffer 

pub struct CudaBuffer {
    pub columns: Vec<CudaColumn>,
    pub row_cap: u64,
    pub d_num_rows: TrackedCudaSlice<u32>,
    pub schema: Schema,
    /* private fields */
}

Column-oriented GPU buffer

Holds columnar data on the GPU with an associated schema. Each column is stored as a separate CudaSlice<u8>.

Fields

columns: Vec<CudaColumn>

Column data stored as raw bytes

row_cap: u64

Row capacity for allocated columns

d_num_rows: TrackedCudaSlice<u32>

Device-resident row count (len = 1)

schema: Schema

Schema describing the column types

Implementations

impl CudaBuffer

pub fn from_columns( columns: Vec<CudaColumn>, row_cap: u64, d_num_rows: TrackedCudaSlice<u32>, schema: Schema, ) -> Self

Create a buffer from existing columns

Arguments

• columns - Pre-allocated column data
• row_cap - Row capacity for the buffer
• d_num_rows - Device-resident row count
• schema - Schema describing the columns

Panics

Panics if the number of columns doesn’t match the schema arity

pub fn from_columns_with_host_count( columns: Vec<CudaColumn>, row_cap: u64, d_num_rows: TrackedCudaSlice<u32>, schema: Schema, host_row_count: u32, ) -> Self

Like from_columns, but eagerly populates the row-count cache. Use when the host already knows the exact row count (e.g., buffer_from_columns).

pub fn cached_row_count(&self) -> Option<u32>

Returns the cached row count if available (not sentinel u32::MAX).

pub fn set_cached_row_count_if_unset(&self, count: u32)

Sets the cached row count if not already set (CAS from sentinel). No-op if already cached.

pub fn num_rows(&self) -> u64

Get the row capacity

pub fn num_rows_device(&self) -> &TrackedCudaSlice<u32>

Get the device-resident row count

pub fn is_empty(&self) -> bool

Check if the buffer has zero row capacity

pub fn schema(&self) -> &Schema

Get the schema

pub fn arity(&self) -> usize

Get the number of columns (arity)

pub fn estimated_bytes(&self) -> u64

Estimated memory usage in bytes

pub fn column(&self, index: usize) -> Option<&CudaColumn>

Get a reference to a specific column by index