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_gpu

def gemv_gpu[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, pdl_level: PDLLevel = PDLLevel.ON](c: TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type], a: TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type], b: TileTensor[Storage=b.Storage, address_space=b.address_space, linear_idx_type=b.linear_idx_type], ctx: DeviceContext)

Selects and dispatches the appropriate GPU GEMV kernel based on shape and hardware.

Examines runtime dimensions M, N, K and static shape information to choose among GEMV_KERNEL, GEMV_KERNEL_VECTOR, GEMV_SPLIT_K, GEVM_KERNEL, and MATMUL_NAIVE variants.

Parameters:

transpose_b (Bool): When True, B is treated as transposed (N, K) row-major.
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 element-wise to each output.
pdl_level (PDLLevel): Programmatic dependent launch level for PDL barriers.

Args:

c (TileTensor[Storage=c.Storage, address_space=c.address_space, linear_idx_type=c.linear_idx_type]): Rank-2 output TileTensor.
a (TileTensor[Storage=a.Storage, address_space=a.address_space, linear_idx_type=a.linear_idx_type]): Rank-2 input matrix TileTensor.
b (TileTensor[Storage=b.Storage, address_space=b.address_space, linear_idx_type=b.linear_idx_type]): Rank-2 input vector or matrix TileTensor.
ctx (DeviceContext): Device context for kernel launch.