// Recompose any decomposed variables, and establish the canonical
	// IDs for each var and slot by filling out state.vars and state.slots.

	state.slots = state.slots[:0]
	state.vars = state.vars[:0]
	for i, slot := range f.Names {
		state.slots = append(state.slots, *slot)
		if ir.IsSynthetic(slot.N) {
			continue
		}

		topSlot := slot
		for topSlot.SplitOf != nil {
			topSlot = topSlot.SplitOf
		}

	slots := f.Cache.allocIntSlice(f.NumValues())
	defer f.Cache.freeIntSlice(slots)
	for i := range slots {
		slots[i] = -1
	}

	// Pick a stack slot for each value needing one.
	used := f.Cache.allocBoolSlice(f.NumValues())
	defer f.Cache.freeBoolSlice(used)
	for _, b := range f.Blocks {
		for _, v := range b.Values {
			if !s.values[v.ID].needSlot {
				s.nNotNeed++
				continue
			}

	for i := 0; i < n; i++ {
		if name.Type.FieldType(i).IsStruct() {
			slots = decomposeUserStructInto(f, fnames[i], slots)
			delete(f.NamedValues, *fnames[i])
		} else if name.Type.FieldType(i).IsArray() {
			slots = decomposeUserArrayInto(f, fnames[i], slots)
			delete(f.NamedValues, *fnames[i])
		}
	}
	return slots
}
func decomposeUserPhi(v *Value) {
	switch {
	case v.Type.IsStruct():

#include <windows.h>

int main(int argc, char **argv) {
    if (argc < 3) {
        return 1;
    }
    // Allocate more than 64 TLS indices
    // so the Go runtime doesn't find
    // enough space in the TEB TLS slots.
    for (int i = 0; i < 65; i++) {
        TlsAlloc();
    }
    HMODULE hlib = LoadLibrary(argv[1]);
    if (hlib == NULL) {
        return 2;
    }
    FARPROC proc = GetProcAddress(hlib, argv[2]);

   * number of CPUS. Table slots remain empty (null) until they are
   * needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these

		// reading the value to take back ownership of this
		// slot.
		head--
		ptrs2 := d.pack(head, tail)
		if d.headTail.CompareAndSwap(ptrs, ptrs2) {
			// We successfully took back slot.
			slot = &d.vals[head&uint32(len(d.vals)-1)]
			break
		}
	}

	val := *(*any)(unsafe.Pointer(slot))
	if val == dequeueNil(nil) {
		val = nil
	}
	// Zero the slot. Unlike popTail, this isn't racing with

   * number of CPUS. Table slots remain empty (null) until they are
   * needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these

   * number of CPUS. Table slots remain empty (null) until they are
   * needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these

   * number of CPUS. Table slots remain empty (null) until they are
   * needed.
   *
   * A single spinlock ("busy") is used for initializing and
   * resizing the table, as well as populating slots with new Cells.
   * There is no need for a blocking lock; when the lock is not
   * available, threads try other slots (or the base).  During these

In Go 1.23, overhead should be in the single digit percentages.

<!-- https://go.dev/issue/62737 , https://golang.org/cl/576681,  https://golang.org/cl/577615 -->
The compiler in Go 1.23 can now overlap the stack frame slots of local variables
accessed in disjoint regions of a function, which reduces stack usage
for Go applications.

<!-- https://go.dev/cl/577935 -->
For 386 and amd64, the compiler will use information from PGO to align certain