func bitSetTest(x int) bool {
	// amd64:"ANDL\t[$]9, AX"
	// amd64:"CMPQ\tAX, [$]9"
	return x&9 == 9
}

// mask contiguous one bits
func cont1Mask64U(x uint64) uint64 {
	// s390x:"RISBGZ\t[$]16, [$]47, [$]0,"
	return x & 0x0000ffffffff0000
}

// mask contiguous zero bits
func cont0Mask64U(x uint64) uint64 {
	// s390x:"RISBGZ\t[$]48, [$]15, [$]0,"
	return x & 0xffff00000000ffff
}

  %dq_weight = "quantfork.dcast"(%q_weight) : (tensor<2x3x3x2x!quant.uniform<i8:f32, 0.074855112561992565:-1>>) -> tensor<2x3x3x2xf32>
  %q_bias = "quantfork.qcast"(%bias) : (tensor<2xf32>) -> tensor<2x!quant.uniform<i32:f32, 0.044022349891595126>>
  %dq_bias = "quantfork.dcast"(%q_bias) : (tensor<2x!quant.uniform<i32:f32, 0.044022349891595126>>) -> tensor<2xf32>

}

// writeLSB writes the code c for "Least Significant Bits first" data.
func (w *Writer) writeLSB(c uint32) error {
	w.bits |= c << w.nBits
	w.nBits += w.width
	for w.nBits >= 8 {
		if err := w.w.WriteByte(uint8(w.bits)); err != nil {
			return err
		}
		w.bits >>= 8
		w.nBits -= 8
	}
	return nil
}

// writeMSB writes the code c for "Most Significant Bits first" data.
func (w *Writer) writeMSB(c uint32) error {

		if s.state.get() != mSpanInUse {
			continue
		}

		// Find all allocated objects.
		abits := s.allocBitsForIndex(0)
		for i := uintptr(0); i < uintptr(s.nelems); i++ {
			if abits.index < uintptr(s.freeindex) || abits.isMarked() {
				x := s.base() + i*s.elemsize
				trace.HeapObjectExists(x, s.typePointersOfUnchecked(x).typ)
			}
			abits.advance()
		}
	}

	// Write out all the goroutine stacks.
	forEachGRace(func(gp *g) {