For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo function
load
def load[dtype: DType, //, width: Int = Int(1), *, read_only: Bool = False, prefetch_size: Optional[Int] = None, cache_policy: CacheOperation = CacheOperation.ALWAYS, eviction_policy: CacheEviction = CacheEviction.EVICT_NORMAL, alignment: Int = Int((get_alignof Scalar[dtype], _current_target())) if is_nvidia_gpu() else Int(1)](ptr: Pointer[Scalar[dtype]]) -> SIMD[dtype, width]
Loads data from global memory into a SIMD vector.
Provides a high-level interface for vectorized memory loads with configurable cache behavior and memory access patterns.
Parameters:
- dtype (
DType): The data type to load. - width (
Int): Vector width (number of elements to load). - read_only (
Bool): If True, marks the load as read-only for cache optimization. - prefetch_size (
Optional[Int]): Optional L2 cache prefetch size (64, 128, or 256 bytes). - cache_policy (
CacheOperation): Cache operation policy for the load. - eviction_policy (
CacheEviction): Cache eviction policy. - alignment (
Int): Memory alignment in bytes.
Args:
- ptr (
Pointer[Scalar[dtype]]): Pointer to global memory to load from.
Returns:
SIMD[dtype, width]: SIMD vector containing the loaded data.
def load[OffsetType: Indexer, dtype: DType, //, width: Int = Int(1), *, read_only: Bool = False, prefetch_size: Optional[Int] = None, cache_policy: CacheOperation = CacheOperation.ALWAYS, eviction_policy: CacheEviction = CacheEviction.EVICT_NORMAL, alignment: Int = Int((get_alignof Scalar[dtype], _current_target())) if is_nvidia_gpu() else Int(1)](ptr: Pointer[Scalar[dtype]], offset: OffsetType) -> SIMD[dtype, width]
Loads data from global memory with an offset into a SIMD vector.
Provides a high-level interface for vectorized memory loads with configurable cache behavior and memory access patterns, supporting offset-based addressing.
Parameters:
- OffsetType (
Indexer): Type of the offset value. - dtype (
DType): The data type to load. - width (
Int): Vector width (number of elements to load). - read_only (
Bool): If True, marks the load as read-only for cache optimization. - prefetch_size (
Optional[Int]): Optional L2 cache prefetch size (64, 128, or 256 bytes). - cache_policy (
CacheOperation): Cache operation policy for the load. - eviction_policy (
CacheEviction): Cache eviction policy. - alignment (
Int): Memory alignment in bytes.
Args:
- ptr (
Pointer[Scalar[dtype]]): Base pointer to global memory. - offset (
OffsetType): Offset from base pointer in elements.
Returns:
SIMD[dtype, width]: SIMD vector containing the loaded data.