src/cmd/compile/internal/ssa/deadcode.go

// Copyright 2015 The Go Authors. All rights reserved.
// Use of this source code is governed by a BSD-style
// license that can be found in the LICENSE file.

package ssa

import (
	"cmd/internal/src"
)

// findlive returns the reachable blocks and live values in f.
func findlive(f *Func) (reachable []bool, live []bool) {
	reachable = ReachableBlocks(f)
	live, _ = liveValues(f, reachable)
	return
}

// ReachableBlocks returns the reachable blocks in f.
func ReachableBlocks(f *Func) []bool {
	reachable := make([]bool, f.NumBlocks())
	reachable[f.Entry.ID] = true
	p := make([]*Block, 0, 64) // stack-like worklist
	p = append(p, f.Entry)
	for len(p) > 0 {
		// Pop a reachable block
		b := p[len(p)-1]
		p = p[:len(p)-1]
		// Mark successors as reachable
		s := b.Succs
		if b.Kind == BlockFirst {
			s = s[:1]
		}
		for _, e := range s {
			c := e.b
			if int(c.ID) >= len(reachable) {
				f.Fatalf("block %s >= f.NumBlocks()=%d?", c, len(reachable))
			}
			if !reachable[c.ID] {
				reachable[c.ID] = true
				p = append(p, c) // push
			}
		}
	}
	return reachable
}

// liveValues returns the live values in f and a list of values that are eligible
// to be statements in reversed data flow order.
// The second result is used to help conserve statement boundaries for debugging.
// reachable is a map from block ID to whether the block is reachable.
func liveValues(f *Func, reachable []bool) (live []bool, liveOrderStmts []*Value) {
	live = make([]bool, f.NumValues())

	// After regalloc, consider all values to be live.
	// See the comment at the top of regalloc.go and in deadcode for details.
	if f.RegAlloc != nil {
		for i := range live {
			live[i] = true
		}
		return
	}

	// Find all live values
	q := make([]*Value, 0, 64) // stack-like worklist of unscanned values

	// Starting set: all control values of reachable blocks are live.
	// Calls are live (because callee can observe the memory state).
	for _, b := range f.Blocks {
		if !reachable[b.ID] {
			continue
		}
		if v := b.Control; v != nil && !live[v.ID] {
			live[v.ID] = true
			q = append(q, v)
			if v.Pos.IsStmt() != src.PosNotStmt {
				liveOrderStmts = append(liveOrderStmts, v)
			}
		}
		for _, v := range b.Values {
			if (opcodeTable[v.Op].call || opcodeTable[v.Op].hasSideEffects) && !live[v.ID] {
				live[v.ID] = true
				q = append(q, v)
				if v.Pos.IsStmt() != src.PosNotStmt {
					liveOrderStmts = append(liveOrderStmts, v)
				}
			}
			if v.Type.IsVoid() && !live[v.ID] {
				// The only Void ops are nil checks and inline marks.  We must keep these.
				live[v.ID] = true
				q = append(q, v)
				if v.Pos.IsStmt() != src.PosNotStmt {
					liveOrderStmts = append(liveOrderStmts, v)
				}
			}
		}
	}

	// Compute transitive closure of live values.
	for len(q) > 0 {
		// pop a reachable value
		v := q[len(q)-1]
		q = q[:len(q)-1]
		for i, x := range v.Args {
			if v.Op == OpPhi && !reachable[v.Block.Preds[i].b.ID] {
				continue
			}
			if !live[x.ID] {
				live[x.ID] = true
				q = append(q, x) // push
				if x.Pos.IsStmt() != src.PosNotStmt {
					liveOrderStmts = append(liveOrderStmts, x)
				}
			}
		}
	}

	return
}

// deadcode removes dead code from f.
func deadcode(f *Func) {
	// deadcode after regalloc is forbidden for now. Regalloc
	// doesn't quite generate legal SSA which will lead to some
	// required moves being eliminated. See the comment at the
	// top of regalloc.go for details.
	if f.RegAlloc != nil {
		f.Fatalf("deadcode after regalloc")
	}

	// Find reachable blocks.
	reachable := ReachableBlocks(f)

	// Get rid of edges from dead to live code.
	for _, b := range f.Blocks {
		if reachable[b.ID] {
			continue
		}
		for i := 0; i < len(b.Succs); {
			e := b.Succs[i]
			if reachable[e.b.ID] {
				b.removeEdge(i)
			} else {
				i++
			}
		}
	}

	// Get rid of dead edges from live code.
	for _, b := range f.Blocks {
		if !reachable[b.ID] {
			continue
		}
		if b.Kind != BlockFirst {
			continue
		}
		b.removeEdge(1)
		b.Kind = BlockPlain
		b.Likely = BranchUnknown
	}

	// Splice out any copies introduced during dead block removal.
	copyelim(f)

	// Find live values.
	live, order := liveValues(f, reachable)

	// Remove dead & duplicate entries from namedValues map.
	s := f.newSparseSet(f.NumValues())
	defer f.retSparseSet(s)
	i := 0
	for _, name := range f.Names {
		j := 0
		s.clear()
		values := f.NamedValues[name]
		for _, v := range values {
			if live[v.ID] && !s.contains(v.ID) {
				values[j] = v
				j++
				s.add(v.ID)
			}
		}
		if j == 0 {
			delete(f.NamedValues, name)
		} else {
			f.Names[i] = name
			i++
			for k := len(values) - 1; k >= j; k-- {
				values[k] = nil
			}
			f.NamedValues[name] = values[:j]
		}
	}
	for k := len(f.Names) - 1; k >= i; k-- {
		f.Names[k] = LocalSlot{}
	}
	f.Names = f.Names[:i]

	pendingLines := f.cachedLineStarts // Holds statement boundaries that need to be moved to a new value/block
	pendingLines.clear()

	// Unlink values and conserve statement boundaries
	for i, b := range f.Blocks {
		if !reachable[b.ID] {
			// TODO what if control is statement boundary? Too late here.
			b.SetControl(nil)
		}
		for _, v := range b.Values {
			if !live[v.ID] {
				v.resetArgs()
				if v.Pos.IsStmt() == src.PosIsStmt && reachable[b.ID] {
					pendingLines.set(v.Pos.Line(), int32(i)) // TODO could be more than one pos for a line
				}
			}
		}
	}

	// Find new homes for lost lines -- require earliest in data flow with same line that is also in same block
	for i := len(order) - 1; i >= 0; i-- {
		w := order[i]
		if j := pendingLines.get(w.Pos.Line()); j > -1 && f.Blocks[j] == w.Block {
			w.Pos = w.Pos.WithIsStmt()
			pendingLines.remove(w.Pos.Line())
		}
	}

	// Any boundary that failed to match a live value can move to a block end
	for i := 0; i < pendingLines.size(); i++ {
		l, bi := pendingLines.getEntry(i)
		b := f.Blocks[bi]
		if b.Pos.Line() == l {
			b.Pos = b.Pos.WithIsStmt()
		}
	}

	// Remove dead values from blocks' value list. Return dead
	// values to the allocator.
	for _, b := range f.Blocks {
		i := 0
		for _, v := range b.Values {
			if live[v.ID] {
				b.Values[i] = v
				i++
			} else {
				f.freeValue(v)
			}
		}
		// aid GC
		tail := b.Values[i:]
		for j := range tail {
			tail[j] = nil
		}
		b.Values = b.Values[:i]
	}

	// Remove dead blocks from WBLoads list.
	i = 0
	for _, b := range f.WBLoads {
		if reachable[b.ID] {
			f.WBLoads[i] = b
			i++
		}
	}
	for j := i; j < len(f.WBLoads); j++ {
		f.WBLoads[j] = nil
	}
	f.WBLoads = f.WBLoads[:i]

	// Remove unreachable blocks. Return dead blocks to allocator.
	i = 0
	for _, b := range f.Blocks {
		if reachable[b.ID] {
			f.Blocks[i] = b
			i++
		} else {
			if len(b.Values) > 0 {
				b.Fatalf("live values in unreachable block %v: %v", b, b.Values)
			}
			f.freeBlock(b)
		}
	}
	// zero remainder to help GC
	tail := f.Blocks[i:]
	for j := range tail {
		tail[j] = nil
	}
	f.Blocks = f.Blocks[:i]
}

// removeEdge removes the i'th outgoing edge from b (and
// the corresponding incoming edge from b.Succs[i].b).
func (b *Block) removeEdge(i int) {
	e := b.Succs[i]
	c := e.b
	j := e.i

	// Adjust b.Succs
	b.removeSucc(i)

	// Adjust c.Preds
	c.removePred(j)

	// Remove phi args from c's phis.
	n := len(c.Preds)
	for _, v := range c.Values {
		if v.Op != OpPhi {
			continue
		}
		v.Args[j].Uses--
		v.Args[j] = v.Args[n]
		v.Args[n] = nil
		v.Args = v.Args[:n]
		phielimValue(v)
		// Note: this is trickier than it looks. Replacing
		// a Phi with a Copy can in general cause problems because
		// Phi and Copy don't have exactly the same semantics.
		// Phi arguments always come from a predecessor block,
		// whereas copies don't. This matters in loops like:
		// 1: x = (Phi y)
		//    y = (Add x 1)
		//    goto 1
		// If we replace Phi->Copy, we get
		// 1: x = (Copy y)
		//    y = (Add x 1)
		//    goto 1
		// (Phi y) refers to the *previous* value of y, whereas
		// (Copy y) refers to the *current* value of y.
		// The modified code has a cycle and the scheduler
		// will barf on it.
		//
		// Fortunately, this situation can only happen for dead
		// code loops. We know the code we're working with is
		// not dead, so we're ok.
		// Proof: If we have a potential bad cycle, we have a
		// situation like this:
		//   x = (Phi z)
		//   y = (op1 x ...)
		//   z = (op2 y ...)
		// Where opX are not Phi ops. But such a situation
		// implies a cycle in the dominator graph. In the
		// example, x.Block dominates y.Block, y.Block dominates
		// z.Block, and z.Block dominates x.Block (treating
		// "dominates" as reflexive).  Cycles in the dominator
		// graph can only happen in an unreachable cycle.
	}
}