vector.fragment.load

← vector dialect

Load a target-shaped matrix fragment payload from a typed view at a full-rank logical origin. Unlike vector.load, the result vector shape is the physical fragment payload selected by role, logical matrix shape, view layout, and target legality; it is not an ordinary trailing-axis footprint of the view. The result carries fragment facts directly so vector.mma can consume it without a separate vector.fragment wrapper. When the view and payload element types differ, the operation represents a fragment-shaped numeric conversion at the load boundary and target lowering must either select that conversion explicitly or reject it with target diagnostics. When the view storage schema requires runtime auxiliary values such as sparse metadata, scale values, or codebooks, the optional keyed using operands provide those SSA values while the view type remains the source of truth for the storage schema.

Operation contract

Property	Value
Semantic phase	executable
Traits	RefinableResultTypeRefs
Interfaces	MemoryAccess
Memory effects	read view

Signature

Kind	Name	Type	Cardinality	Description
Operand	view	view	required	Typed source view holding logical matrix data.
Operand	indices	index	variadic	Dynamic logical origin indices.
Operand	blocks	index	optional	Optional independent matrix block count.
Operand	rows	index	required	Logical matrix row count for this fragment role.
Operand	columns	index	required	Logical matrix column count for this fragment role.
Operand	auxiliary	vector	variadic	Optional keyed runtime auxiliary operands required by the view storage schema.
Result	result	vector	required	Loaded physical matrix fragment payload.
Attribute	role	enum VectorFragmentRole	required	—
Attribute	auxiliary_names	dict	optional	Sorted auxiliary operand keys mapped to auxiliary operand ordinals.
Attribute	cache_scope	enum CacheScope	optional	Optional cache/coherency scope required by target lowering.
Attribute	cache_temporal	enum CacheTemporal	optional	Optional temporal cache policy required by target lowering.
Attribute	static_indices	i64_array	required	Static logical origin indices with INT64_MIN sentinels for dynamics.

Examples

%lhs = vector.fragment.load<lhs> %a[%row, %k0] shape [%m, %k] : view<[%M]x[%K]xf16, %layout> -> vector<16xf16>

%rhs = vector.fragment.load<rhs> %b[%k0, %col] shape [%k, %n] using {sparsity = %metadata : vector<1xi32>} : view<[%K]x[%N]xf8E4M3, %storage> -> vector<8xi32>