JE dataTLS
	CMPQ autLen, $128
	JB startSinglesLoop
	JMP dataOctaLoop

dataTLS:
	MOVOU (16*14)(pTbl), T1
	MOVOU (16*15)(pTbl), T2
	PXOR B0, B0
	MOVQ (aut), B0
	PINSRD $2, 8(aut), B0
	PINSRB $12, 12(aut), B0
	XORQ autLen, autLen
	JMP dataMul

dataOctaLoop:
		CMPQ autLen, $128
		JB startSinglesLoop
		SUBQ $128, autLen

		MOVOU (16*0)(aut), X0
		MOVOU (16*1)(aut), X1

	"PDEPL",
	"PDEPQ",
	"PEXTL",
	"PEXTQ",
	"PEXTRB",
	"PEXTRD",
	"PEXTRQ",
	"PEXTRW",
	"PHADDD",
	"PHADDSW",
	"PHADDW",
	"PHMINPOSUW",
	"PHSUBD",
	"PHSUBSW",
	"PHSUBW",
	"PINSRB",
	"PINSRD",
	"PINSRQ",
	"PINSRW",
	"PMADDUBSW",
	"PMADDWL",
	"PMAXSB",
	"PMAXSD",
	"PMAXSW",
	"PMAXUB",
	"PMAXUD",
	"PMAXUW",
	"PMINSB",
	"PMINSD",
	"PMINSW",
	"PMINUB",

		"tzcntq", "tzcntl", "tzcnt",
	},
	"bmi2": {
		"sarxq", "sarxl", "sarx",
		"shlxq", "shlxl", "shlx",
		"shrxq", "shrxl", "shrx",
	},
	"sse41": {
		"roundsd",
		"pinsrq", "pinsrl", "pinsrd", "pinsrb", "pinsr",
		"pextrq", "pextrl", "pextrd", "pextrb", "pextr",
		"pminsb", "pminsd", "pminuw", "pminud", // Note: ub and sw are ok.
		"pmaxsb", "pmaxsd", "pmaxuw", "pmaxud",

	PINSRB $7, (BX), X2                     // 660f3a201307
	PINSRB $7, (R11), X2                    // 66410f3a201307
	PINSRB $7, DX, X2                       // 660f3a20d207
	PINSRB $7, R11, X2                      // 66410f3a20d307
	PINSRB $7, (BX), X11                    // 66440f3a201b07
	PINSRB $7, (R11), X11                   // 66450f3a201b07
	PINSRB $7, DX, X11                      // 66440f3a20da07

	PHADDSW:         "PHADDSW",
	PHADDW:          "PHADDW",
	PHMINPOSUW:      "PHMINPOSUW",
	PHSUBD:          "PHSUBD",
	PHSUBSW:         "PHSUBSW",
	PHSUBW:          "PHSUBW",
	PINSRB:          "PINSRB",
	PINSRD:          "PINSRD",
	PINSRQ:          "PINSRQ",
	PINSRW:          "PINSRW",
	PMADDUBSW:       "PMADDUBSW",
	PMADDWD:         "PMADDWD",
	PMAXSB:          "PMAXSB",