"VPSHLDVD",
	"VPSHLDVQ",
	"VPSHLDVW",
	"VPSHLDW",
	"VPSHRDD",
	"VPSHRDQ",
	"VPSHRDVD",
	"VPSHRDVQ",
	"VPSHRDVW",
	"VPSHRDW",
	"VPSHUFB",
	"VPSHUFBITQMB",
	"VPSHUFD",
	"VPSHUFHW",
	"VPSHUFLW",
	"VPSIGNB",
	"VPSIGND",
	"VPSIGNW",
	"VPSLLD",
	"VPSLLDQ",
	"VPSLLQ",
	"VPSLLVD",
	"VPSLLVQ",
	"VPSLLVW",
	"VPSLLW",
	"VPSRAD",

	VPSHUFD $121, Z3, K7, Z8                           // 62717d4f70c379
	VPSHUFD $121, Z27, K7, Z8                          // 62117d4f70c379
	VPSHUFD $121, 7(AX)(CX*4), K7, Z8                  // 62717d4f7084880700000079
	VPSHUFD $121, 7(AX)(CX*1), K7, Z8                  // 62717d4f7084080700000079
	VPSHUFD $121, Z3, K7, Z2                           // 62f17d4f70d379

	// but previous assembler permitted negative arguments.
	// This behavior is preserved for compatibility reasons.
	VPSHUFD $-79, X7, X7         // c5f970ffb1
	RORXL $-1, (AX), DX          // c4e37bf010ff
	RORXQ $-1, (AX), DX          // c4e3fbf010ff
	VPSHUFD $-1, X1, X2          // c5f970d1ff
	VPSHUFD $-1, Y1, Y2          // c5fd70d1ff
	VPSHUFHW $-1, X1, X2         // c5fa70d1ff
	VPSHUFHW $-1, Y1, Y2         // c5fe70d1ff

	ANDL    c, T1;                       \ // T1 = a&c						// MAJB
	ADDL    y0, y2;                      \ // y2 = S1 + CH					// --
	;                                    \
	VPXOR   XTMP4, XTMP3, XTMP1;         \ // XTMP1 = s0
	VPSHUFD $0xFA, XDWORD3, XTMP2;       \ // XTMP2 = W[-2] {BBAA}
	ORL     T1, y3;                      \ // y3 = MAJ = (a|c)&b)|(a&c)             // MAJ
	ADDL    y1, h;                       \ // h = k + w + h + S0                    // --

	VPSHUFD $7, X11, X11            // c4417970db07
	VPSHUFD $7, (BX), Y2            // c4e17d701307 or c5fd701307
	VPSHUFD $7, (R11), Y2           // c4c17d701307
	VPSHUFD $7, Y2, Y2              // c4e17d70d207 or c5fd70d207
	VPSHUFD $7, Y11, Y2             // c4c17d70d307
	VPSHUFD $7, (BX), Y11           // c4617d701b07 or c57d701b07
	VPSHUFD $7, (R11), Y11          // c4417d701b07
	VPSHUFD $7, Y2, Y11             // c4617d70da07 or c57d70da07