VPMADD52HUQ 15(DX)(BX*2), Y8, K7, Y17              // 62e2bd2fb58c5a0f000000
	VPMADD52HUQ Y28, Y1, K7, Y17                       // 6282f52fb5cc
	VPMADD52HUQ Y13, Y1, K7, Y17                       // 62c2f52fb5cd
	VPMADD52HUQ Y7, Y1, K7, Y17                        // 62e2f52fb5cf
	VPMADD52HUQ (R8), Y1, K7, Y17                      // 62c2f52fb508
	VPMADD52HUQ 15(DX)(BX*2), Y1, K7, Y17              // 62e2f52fb58c5a0f000000

	VGF2P8MULB Y12, Y13, K1, Y1                        // 62d21529cfcc
	VGF2P8MULB (R8), Y13, K1, Y1                       // 62d21529cf08
	VGF2P8MULB 15(DX)(BX*2), Y13, K1, Y1               // 62f21529cf8c5a0f000000
	VGF2P8MULB Y2, Y7, K1, Y1                          // 62f24529cfca
	VGF2P8MULB Y21, Y7, K1, Y1                         // 62b24529cfcd
	VGF2P8MULB Y12, Y7, K1, Y1                         // 62d24529cfcc

	VPDPWSSDS Y31, Y1, K1, Y1                          // 6292752953cf
	VPDPWSSDS (R8), Y1, K1, Y1                         // 62d275295308
	VPDPWSSDS 15(DX)(BX*2), Y1, K1, Y1                 // 62f27529538c5a0f000000
	VPDPWSSDS Y0, Y9, K1, Y1                           // 62f2352953c8
	VPDPWSSDS Y19, Y9, K1, Y1                          // 62b2352953cb

	VPCONFLICTQ Y5, K4, Y1                             // 62f2fd2cc4cd
	VPCONFLICTQ Y18, K4, Y1                            // 62b2fd2cc4ca
	VPCONFLICTQ Y20, K4, Y1                            // 62b2fd2cc4cc
	VPCONFLICTQ 99(R15)(R15*8), K4, Y1                 // 6292fd2cc48cff63000000
	VPCONFLICTQ 7(AX)(CX*8), K4, Y1                    // 62f2fd2cc48cc807000000

	VPMULTISHIFTQB Y1, Y30, K7, Y12                    // 62728d2783e1
	VPMULTISHIFTQB Y30, Y30, K7, Y12                   // 62128d2783e6
	VPMULTISHIFTQB 17(SP)(BP*2), Y30, K7, Y12          // 62728d2783a46c11000000
	VPMULTISHIFTQB -7(DI)(R8*4), Y30, K7, Y12          // 62328d2783a487f9ffffff
	VPMULTISHIFTQB Y16, Y26, K7, Y12                   // 6232ad2783e0
	VPMULTISHIFTQB Y1, Y26, K7, Y12                    // 6272ad2783e1

	VPOPCNTB Y1, K4, Y24                               // 62627d2c54c1
	VPOPCNTB 15(R8)(R14*8), K4, Y24                    // 62027d2c5484f00f000000
	VPOPCNTB -15(R14)(R15*2), K4, Y24                  // 62027d2c54847ef1ffffff
	VPOPCNTB Y14, K4, Y13                              // 62527d2c54ee
	VPOPCNTB Y21, K4, Y13                              // 62327d2c54ed
	VPOPCNTB Y1, K4, Y13                               // 62727d2c54e9

    /*
     * For an explanation of the basic principle behind this check, see
     * https://stackoverflow.com/a/35754308/28465
     *
     * In that explanation's notation, our `overlap` check would be `x1 < y2 && y1 < x2`. We've
     * flipped one part of the check so that we're using "less than" in both cases (rather than a
     * mix of "less than" and "greater than"). We've also switched to "strictly less than" rather

    /*
     * For an explanation of the basic principle behind this check, see
     * https://stackoverflow.com/a/35754308/28465
     *
     * In that explanation's notation, our `overlap` check would be `x1 < y2 && y1 < x2`. We've
     * flipped one part of the check so that we're using "less than" in both cases (rather than a
     * mix of "less than" and "greater than"). We've also switched to "strictly less than" rather