"(*mspan).base",
			"(*mspan).markBitsForBase",
			"(*mspan).markBitsForIndex",
			"(*mspan).writeUserArenaHeapBits",
			"(*muintptr).set",
			"(*puintptr).set",
			"(*wbBuf).get1",
			"(*wbBuf).get2",

			// Trace-related ones.
			"traceLocker.ok",
			"traceEnabled",
		},
		"runtime/internal/sys": {},
		"runtime/internal/math": {
			"MulUintptr",
		},
		"bytes": {

//
//go:nosplit
//go:linkname rand
func rand() uint64 {
	// Note: We avoid acquirem here so that in the fast path
	// there is just a getg, an inlined c.Next, and a return.
	// The performance difference on a 16-core AMD is
	// 3.7ns/call this way versus 4.3ns/call with acquirem (+16%).
	mp := getg().m
	c := &mp.chacha8
	for {
		// Note: c.Next is marked nosplit,
		// so we don't need to use mp.locks

// and then calls coroexit to remove the extra concurrency.
func corostart() {
	gp := getg()
	c := gp.coroarg
	gp.coroarg = nil

	defer coroexit(c)
	c.f(c)
}

// coroexit is like coroswitch but closes the coro
// and exits the current goroutine
func coroexit(c *coro) {
	gp := getg()
	gp.coroarg = c
	gp.coroexit = true
	mcall(coroswitch_m)
}

//go:linkname coroswitch

//
// Ensure mayMoreStackPreempt can be called for all ABIs.
//
//go:nosplit
//go:linkname mayMoreStackPreempt
func mayMoreStackPreempt() {
	// Don't do anything on the g0 or gsignal stack.
	gp := getg()
	if gp == gp.m.g0 || gp == gp.m.gsignal {
		return
	}
	// Force a preemption, unless the stack is already poisoned.
	if gp.stackguard0 < stackPoisonMin {
		gp.stackguard0 = stackPreempt
	}
}

//   - fortio.org/log
//
// Do not remove or change the type signature.
// See go.dev/issue/67401.
//
//go:linkname getm
func getm() uintptr {
	return uintptr(unsafe.Pointer(getg().m))
}

var (
	// Locking linked list of extra M's, via mp.schedlink. Must be accessed
	// only via lockextra/unlockextra.
	//
	// Can't be atomic.Pointer[m] because we use an invalid pointer as a

		// symbolizer.
		pcBuf[0] = logicalStackSentinel
		if getg() == gp {
			nstk += callers(skip+1, pcBuf[1:])
		} else if gp != nil {
			nstk += gcallers(gp, skip, pcBuf[1:])
		}
	} else {
		// Fast path: Unwind using frame pointers.
		pcBuf[0] = uintptr(skip)
		if getg() == gp {
			nstk += fpTracebackPCs(unsafe.Pointer(getfp()), pcBuf[1:])
		} else if gp != nil {
			// Three cases:

    auto l = mlir::dyn_cast<TypedValue<RankedTensorType>>(op.getX());
    auto r = mlir::dyn_cast<TypedValue<RankedTensorType>>(op.getY());
    if (!l || !r) return failure();

    auto element_type = getElementTypeOrSelf(l.getType());
    if (!element_type.isBF16()) return failure();

    auto out_type = op.getZ().getType();

    l = rewriter.create<ConvertOp>(op.getLoc(), l, rewriter.getF32Type());

// sigpanic calls can be injected.
//
// The signal handler must not inject a call to sigpanic if
// getg().throwsplit, since sigpanic may need to grow the stack.
//
// This is exported via linkname to assembly in runtime/cgo.
//
//go:linkname sigpanic
func sigpanic() {
	gp := getg()
	if !canpanic() {
		throw("unexpected signal during runtime execution")
	}

	switch gp.sig {
	case _SIGBUS:

	if goarch.IsWasm == 0 && stringslite.HasPrefix(funcname(findfunc(pc)), "runtime.") {
		// Note: wasm can't tail call, so we can't get the original caller's pc.
		throw(msg)
	}
	// TODO: is this redundant? How could we be in malloc
	// but not in the runtime? runtime/internal/*, maybe?
	gp := getg()
	if gp != nil && gp.m != nil && gp.m.mallocing != 0 {
		throw(msg)
	}
}

			lock(&rw.rLock)
			if rw.readerPass > 0 {
				// Writer finished.
				rw.readerPass -= 1
				unlock(&rw.rLock)
			} else {
				// Queue this reader to be woken by
				// the writer.
				m := getg().m
				m.schedlink = rw.readers
				rw.readers.set(m)
				unlock(&rw.rLock)
				notesleep(&m.park)
				noteclear(&m.park)
			}
		})
	}
}

// runlock undoes a single rlock call on rw.