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

	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

	VPCLMULQDQ $97, 15(DX)(BX*8), Z26, Z24             // 62632d404484da0f00000061 or 6263ad404484da0f00000061
	VPCLMULQDQ $97, Z9, Z0, Z12                        // 62537d4844e161 or 6253fd4844e161
	VPCLMULQDQ $97, Z3, Z0, Z12                        // 62737d4844e361 or 6273fd4844e361
	VPCLMULQDQ $97, 7(SI)(DI*1), Z0, Z12               // 62737d4844a43e0700000061 or 6273fd4844a43e0700000061

	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

	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

	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

		lines(
			"#define LOAD(off, reg) \\",
			"\tMOVBLZX	(off*4)(R12),	reg \\",
			"\tADDB	reg,		DX",
			"",
			"LOAD(8, AX)",
		),
		"\n.\n.MOVBLZX.(.8.*.4.).(.R12.).,.AX.\n.ADDB.AX.,.DX.\n",
	},
	{
		"nested multiline macro",
		lines(
			"#define KEYROUND(xmm, load, off, r1, r2, index) \\",
			"\tMOVBLZX	(BP)(DX*4),	R8 \\",
			"\tload((off+1), r2) \\",
			"\tMOVB	R8,		(off*4)(R12) \\",

	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

	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

	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