VRCP28PS 15(DX)(BX*1), K3, Z28                     // 62627d4bcaa41a0f000000
	VRCP28PS -7(CX)(DX*2), K3, Z28                     // 62627d4bcaa451f9ffffff
	VRCP28PS Z13, K3, Z6                               // 62d27d4bcaf5
	VRCP28PS Z21, K3, Z6                               // 62b27d4bcaf5
	VRCP28PS 15(DX)(BX*1), K3, Z6                      // 62f27d4bcab41a0f000000

	VPDPWSSD -7(CX)(DX*4), Y2, K3, Y14                 // 62726d2b52b491f9ffffff
	VPDPWSSD Y8, Y22, K3, Y14                          // 62524d2352f0
	VPDPWSSD Y9, Y22, K3, Y14                          // 62524d2352f1
	VPDPWSSD Y22, Y22, K3, Y14                         // 62324d2352f6
	VPDPWSSD 15(R8)(R14*4), Y22, K3, Y14               // 62124d2352b4b00f000000

	VPLZCNTD 15(DX)(BX*1), K3, X17                     // 62e27d0b448c1a0f000000
	VPLZCNTD -7(CX)(DX*2), K3, X17                     // 62e27d0b448c51f9ffffff
	VPLZCNTD X3, K3, X15                               // 62727d0b44fb
	VPLZCNTD X26, K3, X15                              // 62127d0b44fa
	VPLZCNTD X23, K3, X15                              // 62327d0b44ff

	VGF2P8AFFINEINVQB $47, (DX), Z14, K2, Z28          // 62638d4acf222f
	VGF2P8AFFINEINVQB $47, Z3, Z28, K2, Z28            // 62639d42cfe32f
	VGF2P8AFFINEINVQB $47, Z12, Z28, K2, Z28           // 62439d42cfe42f
	VGF2P8AFFINEINVQB $47, 99(R15)(R15*1), Z28, K2, Z28 // 62039d42cfa43f630000002f
	VGF2P8AFFINEINVQB $47, (DX), Z28, K2, Z28          // 62639d42cf222f

	VPMADD52HUQ 15(DX)(BX*2), Y31, K7, Y17             // 62e28527b58c5a0f000000
	VPMADD52HUQ Y28, Y8, K7, Y17                       // 6282bd2fb5cc
	VPMADD52HUQ Y13, Y8, K7, Y17                       // 62c2bd2fb5cd
	VPMADD52HUQ Y7, Y8, K7, Y17                        // 62e2bd2fb5cf
	VPMADD52HUQ (R8), Y8, K7, Y17                      // 62c2bd2fb508
	VPMADD52HUQ 15(DX)(BX*2), Y8, K7, Y17              // 62e2bd2fb58c5a0f000000

	VAESDEC (DX), Y0, Y23                              // 62e27d28de3a or 62e2fd28de3a
	VAESDEC Z27, Z3, Z11                               // 62126548dedb or 6212e548dedb
	VAESDEC Z15, Z3, Z11                               // 62526548dedf or 6252e548dedf
	VAESDEC 99(R15)(R15*1), Z3, Z11                    // 62126548de9c3f63000000 or 6212e548de9c3f63000000

	VPSHUFBITQMB 15(DX)(BX*2), X7, K6, K0              // 62f2450e8f845a0f000000
	VPSHUFBITQMB X24, X13, K6, K0                      // 6292150e8fc0
	VPSHUFBITQMB X7, X13, K6, K0                       // 62f2150e8fc7
	VPSHUFBITQMB X0, X13, K6, K0                       // 62f2150e8fc0
	VPSHUFBITQMB (R8), X13, K6, K0                     // 62d2150e8f00
	VPSHUFBITQMB 15(DX)(BX*2), X13, K6, K0             // 62f2150e8f845a0f000000

	VPERMB -7(CX)(DX*8), Y31, K7, Y22                  // 62e205278db4d1f9ffffff
	VPERMB Y5, Y5, K7, Y22                             // 62e2552f8df5
	VPERMB Y19, Y5, K7, Y22                            // 62a2552f8df3
	VPERMB Y31, Y5, K7, Y22                            // 6282552f8df7
	VPERMB 17(SP)(BP*1), Y5, K7, Y22                   // 62e2552f8db42c11000000

	{"(AX)(CX*8)", "(AX)(CX*8)"},
	{"(BP)(CX*4)", "(BP)(CX*4)"},
	{"(BP)(DX*4)", "(BP)(DX*4)"},
	{"(BP)(R8*4)", "(BP)(R8*4)"},
	{"(BX)", "(BX)"},
	{"(DI)", "(DI)"},
	{"(DI)(BX*1)", "(DI)(BX*1)"},
	{"(DX)", "(DX)"},
	{"(R9)", "(R9)"},
	{"(R9)(BX*8)", "(R9)(BX*8)"},
	{"(SI)", "(SI)"},
	{"(SI)(BX*1)", "(SI)(BX*1)"},
	{"(SI)(DX*1)", "(SI)(DX*1)"},
	{"(SP)", "(SP)"},
	{"(SP)(AX*4)", "(SP)(AX*4)"},

     *
     *    dX = U / Y
     *    dY = -U*X / Y^2 = (X/Y) * -U / Y = -U*Z / Y
     *
     */

    AbstractTensorHandle* upstream_grad = grad_outputs[0];
    AbstractTensorHandle* Y = forward_inputs_[1];
    AbstractTensorHandle* Z = forward_outputs_[0];

    // Calculate dX =  U / Y
    std::string name = "Div_Grad_X";
    TF_RETURN_IF_ERROR(