MOVL $4, DX
loop:
	QR(X0, X4, X8, X12, X15)
	MOVOU X4, (4*16)(BX) // save X4
	QR(X1, X5, X9, X13, X15)
	MOVOU (15*16)(BX), X15 // reload X15; temp now X4
	QR(X2, X6, X10, X14, X4)
	QR(X3, X7, X11, X15, X4)

	QR(X0, X5, X10, X15, X4)
	MOVOU X15, (15*16)(BX) // save X15
	QR(X1, X6, X11, X12, X4)
	MOVOU (4*16)(BX), X4  // reload X4; temp now X15
	QR(X2, X7, X8, X13, X15)
	QR(X3, X4, X9, X14, X15)

	MOVD F15, 336(R3)
	MOVD F16, 344(R3)
	MOVD F17, 352(R3)
	MOVD F18, 360(R3)
	MOVD F19, 368(R3)
	MOVD F20, 376(R3)
	MOVD F21, 384(R3)
	MOVD F22, 392(R3)
	MOVD F23, 400(R3)
	MOVD F24, 408(R3)
	MOVD F25, 416(R3)
	MOVD F26, 424(R3)
	MOVD F27, 432(R3)
	MOVD F28, 440(R3)
	MOVD F29, 448(R3)
	MOVD F30, 456(R3)
	MOVD F31, 464(R3)
	MOVV FCC0, R4
	MOVV R4, 472(R3)
	CALL ·asyncPreempt2(SB)
	MOVV 472(R3), R4

	MOVD F22, 384(X2)
	MOVD F23, 392(X2)
	MOVD F24, 400(X2)
	MOVD F25, 408(X2)
	MOVD F26, 416(X2)
	MOVD F27, 424(X2)
	MOVD F28, 432(X2)
	MOVD F29, 440(X2)
	MOVD F30, 448(X2)
	MOVD F31, 456(X2)
	CALL ·asyncPreempt2(SB)
	MOVD 456(X2), F31
	MOVD 448(X2), F30
	MOVD 440(X2), F29
	MOVD 432(X2), F28
	MOVD 424(X2), F27
	MOVD 416(X2), F26
	MOVD 408(X2), F25
	MOVD 400(X2), F24
	MOVD 392(X2), F23
	MOVD 384(X2), F22

	out |= 4
	// amd64:-`.*SP.*`
	out |= in[0] << 4
	// amd64:-`.*SP.*`
	out |= in[1] << 7
	// amd64:-`.*SP.*`
	out |= in[2] << 10
	// amd64:-`.*SP.*`
	out |= in[3] << 13
	// amd64:-`.*SP.*`
	out |= in[4] << 16
	// amd64:-`.*SP.*`
	out |= in[5] << 19
	// amd64:-`.*SP.*`
	out |= in[6] << 22
	// amd64:-`.*SP.*`
	out |= in[7] << 25
	// amd64:-`.*SP.*`
	out |= in[8] << 28
	// amd64:-`.*SP.*`
	out |= in[9] << 31

    $('.card').each(function (index, row) {
        var currentTags = $(row).attr('tag');
        if (currentTags === undefined) {
            currentTags = ''
        }
        // tag="FAILURE-THRESHOLD(4.16%),FLAKY(9.30%)"
        // tag="UNTAGGED"
        currentTags = currentTags.split(',').map(tag => tag.split('(')[0])
        if (selectedTags.some(tag => currentTags.indexOf(tag) != -1)) {
            $(row).show()

			valSize := ef.ByteOrder.Uint32(note[4:])
			tag := ef.ByteOrder.Uint32(note[8:])
			nname := note[12:16]
			if nameSize == 4 && 16+valSize <= uint32(len(note)) && tag == elfGoBuildIDTag && bytes.Equal(nname, elfGoNote) {
				return string(note[16 : 16+valSize]), nil
			}

			if nameSize == 4 && 16+valSize <= uint32(len(note)) && tag == gnuBuildIDTag && bytes.Equal(nname, elfGNUNote) {
				gnu = string(note[16 : 16+valSize])
			}

      lhs_contracting_dimensions = [4, 1, 6],
      rhs_contracting_dimensions = [6, 0, 4]
  >} : (tensor<1x9x2x3x8x4x10xf32>, tensor<9x2x1x5x10x5x8x7xf32>) -> tensor<1x2x3x4x5x5x7xf32>
  %1 = "mhlo.dot_general"(%arg1, %arg2) {
    dot_dimension_numbers = #mhlo.dot<
      lhs_batching_dimensions = [0, 2],
      rhs_batching_dimensions = [2, 1],
      lhs_contracting_dimensions = [4, 1, 6],
      rhs_contracting_dimensions = [6, 0, 4]

	LDP	3*16(a_ptr), (R4, R5)
	LDP	4*16(a_ptr), (R6, R7)
	LDP	5*16(a_ptr), (R8, R9)
	LDP	3*16(b_ptr), (R16, R17)
	LDP	4*16(b_ptr), (R19, R20)
	LDP	5*16(b_ptr), (R21, R22)
	CSEL	EQ, R16, R4, R4
	CSEL	EQ, R17, R5, R5
	CSEL	EQ, R19, R6, R6
	CSEL	EQ, R20, R7, R7
	CSEL	EQ, R21, R8, R8
	CSEL	EQ, R22, R9, R9
	STP	(R4, R5), 3*16(res_ptr)
	STP	(R6, R7), 4*16(res_ptr)
	STP	(R8, R9), 5*16(res_ptr)

	RET

	VADD C.S4, D.S4, C.S4; VEOR B.B16, C.B16, V30.B16; VSHL  $7, V30.S4, B.S4; VSRI $25, V30.S4, B.S4

// block runs 4 ChaCha8 block transformations in the four stripes of the V registers.

// func block(seed *[8]uint32, blocks *[4][16]uint32, counter uint32)
TEXT ·block<ABIInternal>(SB), NOSPLIT, $16
	// seed in R0
	// blocks in R1
	// counter in R2

	// Load initial constants into top row.
	MOVD $·chachaConst(SB), R10

	SYS_WRITEV_NOCANCEL                = 412
	SYS_SENDTO_NOCANCEL                = 413
	SYS_PREAD_NOCANCEL                 = 414
	SYS_PWRITE_NOCANCEL                = 415
	SYS_WAITID_NOCANCEL                = 416
	SYS_POLL_NOCANCEL                  = 417
	SYS_MSGSND_NOCANCEL                = 418
	SYS_MSGRCV_NOCANCEL                = 419
	SYS_SEM_WAIT_NOCANCEL              = 420
	SYS_AIO_SUSPEND_NOCANCEL           = 421