[CMake] optimized_kernels and quantized_kernel will depend on portable_kernels.

[kirklandsign]

🚀 The feature, motivation and pitch

For example,

executorch/kernels/optimized/cpu/op_native_layer_norm.cpp

Lines 128 to 133 in 94f7b10

	ET_KERNEL_CHECK(
	ctx,
	check_layer_norm_args(
	input, normalized_shape, weight, bias, out, mean_out, rstd_out),
	InvalidArgument,
	ret_val);

here it uses check_layer_norm_args, and it's defined in

executorch/kernels/portable/cpu/util/normalization_ops_util.cpp

Lines 76 to 85 in 94f7b10

	bool check_layer_norm_args(
	const Tensor& in,
	IntArrayRef normalized_shape,
	const executorch::aten::optional<Tensor>& weight,
	const executorch::aten::optional<Tensor>& bias,
	Tensor& out,
	Tensor& mean_out,
	Tensor& rstd_out) {
	size_t ndim = normalized_shape.size();
	ET_CHECK_OR_RETURN_FALSE(

So it means
Either optimized_kernels/quantized_kernel depends on portable_kernels privately, or we split these helpers into its own lib.

Right now, it depends on user of optimized_kernels to pull in portable_kernels as well, so it's "it happens to work"

Alternatives

No response

Additional context

No response

RFC (Optional)

No response

cc @larryliu0820 @jathu

[swolchok]

@manuelcandales

[kirklandsign]

Close for now. At least now the dependency is correct. Ideally we should factor out the common code