AESKEYGENASSIST $0x10, X2, X1 CALL _expand_key_256a<>(SB) AESKEYGENASSIST $0x10, X0, X1 CALL _expand_key_256b<>(SB) AESKEYGENASSIST $0x20, X2, X1 CALL _expand_key_256a<>(SB) AESKEYGENASSIST $0x20, X0, X1 CALL _expand_key_256b<>(SB) AESKEYGENASSIST $0x40, X2, X1 CALL _expand_key_256a<>(SB) JMP Lexp_dec Lexp_enc192: MOVQ 16(AX), X2 AESKEYGENASSIST $0x01, X2, X1 CALL _expand_key_192a<>(SB) AESKEYGENASSIST $0x02, X2, X1 CALL _expand_key_192b<>(SB) AESKEYGENASSIST $0x04, X2, X1 CALL _expand_key_192a<>(SB) AESKEYGENASSIST...

        requires all optimizer variables to be created at the first
        `apply_gradients()` or `minimize()` call. If your workflow calls
        optimizer to update different parts of model in multiple stages,
        please call `optimizer.build(model.trainable_variables)` before the
        training loop.
    *   **Performance regression on `ParameterServerStrategy`.** This could be

	//TODO: LCALLQ* (R11)                   // 49ff1b
	CBW                                     // 6698
	CDQ                                     // 99
	CDQE                                    // 4898
	CLAC                                    // 0f01ca
	CLC                                     // f8
	CLD                                     // fc
	CLFLUSH (BX)                            // 0fae3b
	CLFLUSH (R11)                           // 410fae3b

This has **several benefits**, one of the main ones is that now the parameters of your functions with `Annotated` would **not be affected** at all.