vllm.model_executor.layers.quantization.kernels.scaled_mm ¶

def choose_scaled_mm_linear_kernel(
    config: _KernelConfigT,
    possible_kernels: dict[PlatformEnum, list[type[_KernelT]]],
    compute_capability: int | None = None,
    force_kernel: type[_KernelT] | None = None,
) -> type[_KernelT]:
    """
    Choose a _KernelT that can implement the given config for the
    given compute capability. Attempts to choose the best kernel in terms of
    performance.

    Args:
        config (_KernelConfigT): Description of the linear layer
            to be implemented.
        possible_kernels (dict[PlatformEnum, list[_KernelT]]): A
            dictionary of platforms and their list of possible kernels.
        compute_capability (Optional[int], optional): The compute capability of
            the target device, if None uses `current_platform` to get the
            compute capability. Defaults to None.
        force_kernel (Optional[type[_KernelT]]): An Optional forced kernel to override
            the possible_kernels if it can be implemented. If None, it will only try the
            possible kernels.

    Raises:
        ValueError: If no kernel can implement the given config.

    Returns:
        _KernelT: Chosen kernel.
    """

    failure_reason_list = []

    if force_kernel is not None:
        can_implement, failure_reason = is_supported_and_can_implement_kernel(
            force_kernel, config, compute_capability
        )
        if can_implement:
            return force_kernel

        logger.info_once(
            "Tried to force %s, but the kernel couldn't be implemented",
            force_kernel.__name__,
            scope="global",
        )

    for kernel in possible_kernels[current_platform._enum]:
        is_supported_and_can_implement, failure_reason = (
            is_supported_and_can_implement_kernel(kernel, config, compute_capability)
        )
        if is_supported_and_can_implement:
            return kernel
        failure_reason_list.append(failure_reason)

    raise ValueError(
        "Failed to find a kernel that can implement the "
        "ScaledMM linear layer. Reasons: \n" + "\n".join(failure_reason_list)
    )