VPGATHERDQ Y0, 664(X14*8), Y6           // c4a2fd9034f598020000
	VPGATHERQQ X2, (BP)(X7*2), X1           // c4e2e9914c7d00
	VPGATHERQQ Y2, (BP)(Y7*2), Y1           // c4e2ed914c7d00
	VPGATHERQQ X12, (R13)(X14*2), X11       // c40299915c7500
	VPGATHERQQ Y12, (R13)(Y14*2), Y11       // c4029d915c7500
	VPGATHERQQ X2, (BP)(X7*2), X1           // c4e2e9914c7d00
	VPGATHERQQ Y2, (BP)(Y7*2), Y1           // c4e2ed914c7d00

	V4FMADDPS -17(BP)(SI*4), [Z0-Z3], K2, Z0           // 62f27f4a9a84b5efffffff
	V4FMADDPS 17(SP), [Z10-Z13], K2, Z0                // 62f22f4a9a842411000000
	V4FMADDPS -17(BP)(SI*4), [Z10-Z13], K2, Z0         // 62f22f4a9a84b5efffffff
	V4FMADDPS 17(SP), [Z20-Z23], K2, Z0                // 62f25f429a842411000000
	V4FMADDPS -17(BP)(SI*4), [Z20-Z23], K2, Z0         // 62f25f429a84b5efffffff

	VPDPBUSDS 17(SP)(BP*8), Z13, K2, Z21               // 62e2154a51acec11000000
	VPDPBUSDS 17(SP)(BP*4), Z13, K2, Z21               // 62e2154a51acac11000000
	VPDPBUSDS Z12, Z16, K2, Z5                         // 62d27d4251ec
	VPDPBUSDS Z27, Z16, K2, Z5                         // 62927d4251eb
	VPDPBUSDS 17(SP)(BP*8), Z16, K2, Z5                // 62f27d4251acec11000000

	VAESDECLAST -17(BP)(SI*8), Y11, Y24                // 62622528df84f5efffffff or 6262a528df84f5efffffff
	VAESDECLAST (R15), Y11, Y24                        // 62422528df07 or 6242a528df07
	VAESDECLAST Z8, Z23, Z23                           // 62c24540dff8 or 62c2c540dff8
	VAESDECLAST Z28, Z23, Z23                          // 62824540dffc or 6282c540dffc

	KMOVQ K2, -17(BP)(SI*4)                            // c4e1f89154b5ef
	KMOVQ K7, -17(BP)(SI*4)                            // c4e1f8917cb5ef
	KMOVQ K0, DX                                       // c4e1fb93d0
	KMOVQ K5, DX                                       // c4e1fb93d5
	KMOVQ K0, BP                                       // c4e1fb93e8
	KMOVQ K5, BP                                       // c4e1fb93ed

	VGF2P8AFFINEINVQB $64, -17(BP), X31, K3, X26       // 62638503cf95efffffff40
	VGF2P8AFFINEINVQB $64, -15(R14)(R15*8), X31, K3, X26 // 62038503cf94fef1ffffff40
	VGF2P8AFFINEINVQB $64, X8, X16, K3, X26            // 6243fd03cfd040
	VGF2P8AFFINEINVQB $64, X1, X16, K3, X26            // 6263fd03cfd140
	VGF2P8AFFINEINVQB $64, X0, X16, K3, X26            // 6263fd03cfd040
	VGF2P8AFFINEINVQB $64, -17(BP), X16, K3, X26       // 6263fd03cf95efffffff40

	{"$~15", "$-16"},
	{"(((8)&0xf)*4)(SP)", "32(SP)"},
	{"(((8-14)&0xf)*4)(SP)", "40(SP)"},
	{"(6+8)(AX)", "14(AX)"},
	{"(8*4)(BP)", "32(BP)"},
	{"(AX)", "(AX)"},
	{"(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)"},

func (bp *BytePoolCap) Populate() {
	for _, buf := range reedsolomon.AllocAligned(cap(bp.c), bp.wcap) {
		bp.Put(buf[:bp.w])
	}
}

// Get gets a []byte from the BytePool, or creates a new one if none are
// available in the pool.
func (bp *BytePoolCap) Get() (b []byte) {
	if bp == nil {
		return nil
	}
	select {
	case b = <-bp.c:
		// reuse existing buffer
	default:

	VPLZCNTD (BX), K3, Y3                              // 62f27d2b441b
	VPLZCNTD -17(BP)(SI*1), K3, Y3                     // 62f27d2b449c35efffffff
	VPLZCNTD Z21, K3, Z3                               // 62b27d4b44dd
	VPLZCNTD Z13, K3, Z3                               // 62d27d4b44dd
	VPLZCNTD 17(SP)(BP*8), K3, Z3                      // 62f27d4b449cec11000000
	VPLZCNTD 17(SP)(BP*4), K3, Z3                      // 62f27d4b449cac11000000

	VRANGEPD $11, 17(SP)(BP*4), X31, K2, X12           // 6273850250a4ac110000000b
	VRANGEPD $11, X24, X23, K2, X16                    // 6283c50250c00b
	VRANGEPD $11, X14, X23, K2, X16                    // 62c3c50250c60b
	VRANGEPD $11, X0, X23, K2, X16                     // 62e3c50250c00b
	VRANGEPD $11, 17(SP)(BP*8), X23, K2, X16           // 62e3c5025084ec110000000b