VBROADCASTF64X2 15(DX)(BX*1), K1, Z14              // 6272fd491ab41a0f000000
	VBROADCASTF64X2 -7(CX)(DX*2), K1, Z14              // 6272fd491ab451f9ffffff
	VBROADCASTF64X2 15(DX)(BX*1), K1, Z28              // 6262fd491aa41a0f000000
	VBROADCASTF64X2 -7(CX)(DX*2), K1, Z28              // 6262fd491aa451f9ffffff
	VBROADCASTI32X2 X14, K1, X19                       // 62c27d0959de

	VMOVDQU8 X11, K5, -7(CX)(DX*1)                     // 62717f0d7f9c11f9ffffff
	VMOVDQU8 X11, K5, -15(R14)(R15*4)                  // 62117f0d7f9cbef1ffffff
	VMOVDQU8 X24, K3, X31                              // 62017f0b7fc7
	VMOVDQU8 15(DX)(BX*1), K3, X31                     // 62617f0b6fbc1a0f000000
	VMOVDQU8 -7(CX)(DX*2), K3, X31                     // 62617f0b6fbc51f9ffffff

	ANDNL (R11), R9, DX                     // c4c230f213
	ANDNL DX, R9, DX                        // c4e230f2d2
	ANDNL R11, R9, DX                       // c4c230f2d3
	ANDNL (BX), R9, R11                     // c46230f21b
	ANDNL (R11), R9, R11                    // c44230f21b
	ANDNL DX, R9, R11                       // c46230f2da
	ANDNL R11, R9, R11                      // c44230f2db

	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

	VFIXUPIMMPS $82, 15(DX)(BX*1), Z5, K1, Z13         // 6273554954ac1a0f00000052
	VFIXUPIMMPS $82, -7(CX)(DX*2), Z5, K1, Z13         // 6273554954ac51f9ffffff52
	VFIXUPIMMPS $82, Z2, Z23, K1, Z13                  // 6273454154ea52
	VFIXUPIMMPS $82, 15(DX)(BX*1), Z23, K1, Z13        // 6273454154ac1a0f00000052
	VFIXUPIMMPS $82, -7(CX)(DX*2), Z23, K1, Z13        // 6273454154ac51f9ffffff52

	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

	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

	V4FNMADDPS 99(R15)(R15*1), [Z1-Z4], K3, Z15        // 6212774baabc3f63000000
	V4FNMADDPS (DX), [Z1-Z4], K3, Z15                  // 6272774baa3a
	V4FNMADDPS 99(R15)(R15*1), [Z11-Z14], K3, Z15      // 6212274baabc3f63000000
	V4FNMADDPS (DX), [Z11-Z14], K3, Z15                // 6272274baa3a
	V4FNMADDPS 99(R15)(R15*1), [Z21-Z24], K3, Z15      // 62125743aabc3f63000000
	V4FNMADDPS (DX), [Z21-Z24], K3, Z15                // 62725743aa3a

	VSCATTERPF0DPD K5, (SP)(Y4*2)                      // 62f2fd4dc62c64
	VSCATTERPF0DPD K5, (DX)(Y10*4)                     // 62b2fd4dc62c92
	VSCATTERPF0DPS K3, (DX)(Z10*4)                     // 62b27d4bc62c92
	VSCATTERPF0DPS K3, (AX)(Z4*1)                      // 62f27d4bc62c20
	VSCATTERPF0DPS K3, (SP)(Z4*2)                      // 62f27d4bc62c64
	VSCATTERPF0QPD K4, (DX)(Z10*4)                     // 62b2fd4cc72c92

	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