VADDPD 2032(DX), X29, X0        // 62f1950058427f
	VADDPD 2032(DX), X1, X29        // 6261f508586a7f
	VADDPD 2032(DX), X29, X28       // 6261950058627f
	VADDPD 2032(DX)(AX*2), X29, X0  // 62f195005844427f
	VADDPD 2032(DX)(AX*2), X1, X29  // 6261f508586c427f
	VADDPD 2032(DX)(AX*2), X29, X28 // 626195005864427f
	VADDPD 4064(DX), Y0, Y29        // 6261fd28586a7f
	VADDPD 4064(DX), Y29, Y1        // 62f19520584a7f

	VPCOMPRESSB X13, K1, X28                           // 62127d0963ec
	VPCOMPRESSB X8, K1, X28                            // 62127d0963c4
	VPCOMPRESSB X7, K1, -7(CX)(DX*1)                   // 62f27d09637c11f9
	VPCOMPRESSB X13, K1, -7(CX)(DX*1)                  // 62727d09636c11f9
	VPCOMPRESSB X8, K1, -7(CX)(DX*1)                   // 62727d09634411f9
	VPCOMPRESSB X7, K1, -15(R14)(R15*4)                // 62927d09637cbef1

// i.e., d(y_1 + y_2 + ...)/dx_1, d(y_1 + y_2 + ...)/dx_2...
//
// `dx` are used as initial gradients (which represent the symbolic partial
// derivatives of some loss function `L` w.r.t. `y`).
// `dx` must be nullptr or have size `ny`.
// If `dx` is nullptr, the implementation will use dx of `OnesLike` for all
// shapes in `y`.
// The partial derivatives are returned in `dy`. `dy` should be allocated to

				"application/csta+xml",
				"application/cstadata+xml",
				"application/cu-seeme",
				"application/cybercash",
				"application/davmount+xml",
				"application/dca-rft",
				"application/dec-dx",
				"application/dialog-info+xml",
				"application/dicom",
				"application/dita+xml",
				"application/dita+xml;format=map",
				"application/dita+xml;format=topic",
				"application/dita+xml;format=task",

    // The expected graph looks like this if grad_inputs_provided.
    // If grad_inputs_provided is false, Const_0 will be a OnesLike op.
    //      ^             ^
    //    dy|           dx|        // MatMul Gradient Graph
    //      |             |
    //   MatMul_2      MatMul_1
    //   ^   ^          ^    ^
    //   |   |----------|    |
    //   |        ^          |