For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /max/get-started.md).

Mojo function

gemv_kernel_vector

def gemv_kernel_vector[c_type: DType, a_type: DType, b_type: DType, c_layout: TensorLayout, a_layout: TensorLayout, b_layout: TensorLayout, c_storage: TensorStorage, a_storage: TensorStorage, b_storage: TensorStorage, *, simd_width: Int, transpose_b: Bool = False, elementwise_lambda_fn: Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None] = None, accum_type: DType = get_accum_type[c_type](), check_bounds: Bool = True, pdl_level: PDLLevel = PDLLevel()](c: TileTensor[c_type, c_layout, MutAnyOrigin, Storage=c_storage], a: TileTensor[a_type, a_layout, ImmutAnyOrigin, Storage=a_storage], b: TileTensor[b_type, b_layout, ImmutAnyOrigin, Storage=b_storage], m: Int, n: Int, k: Int)

GPU kernel for matrix-vector multiplication using vectorized warp-level loads.

Each warp processes one output row. Threads collaborate to load simd_width-wide vectors from A and B, accumulate dot products locally, then reduce across the warp.

Parameters:

c_type (DType): Output element type.
a_type (DType): A (matrix) element type.
b_type (DType): B (vector) element type.
c_layout (TensorLayout): Layout descriptor for the output tensor.
a_layout (TensorLayout): Layout descriptor for the A matrix.
b_layout (TensorLayout): Layout descriptor for the B vector.
c_storage (TensorStorage): Storage kind for the output tensor.
a_storage (TensorStorage): Storage kind for A.
b_storage (TensorStorage): Storage kind for B.
simd_width (Int): Number of elements loaded per vectorized access.
transpose_b (Bool): When True, writes the result transposed.
elementwise_lambda_fn (Optional[def[dtype: DType, width: SIMDLength, *, alignment: Int = Int(1)](IndexList[Int(2)], SIMD[dtype, width]) capturing thin -> None]): Optional epilogue applied to each output element.
accum_type (DType): Accumulation precision type.
check_bounds (Bool): When True, bounds-checks the last K iteration.
pdl_level (PDLLevel): Programmatic dependent launch level.

Args:

c (TileTensor[c_type, c_layout, MutAnyOrigin, Storage=c_storage]): Rank-2 output TileTensor.
a (TileTensor[a_type, a_layout, ImmutAnyOrigin, Storage=a_storage]): Rank-2 input matrix TileTensor, shape (m, k).
b (TileTensor[b_type, b_layout, ImmutAnyOrigin, Storage=b_storage]): Rank-2 input vector TileTensor, shape (1, k).
m (Int): Number of output rows.
n (Int): Unused; retained for interface consistency.
k (Int): Shared reduction dimension.