1
0
mirror of https://github.com/golang/go synced 2024-11-19 06:24:41 -07:00
go/ssa/dom.go

310 lines
7.9 KiB
Go
Raw Normal View History

// Copyright 2013 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
// This file defines algorithms related to dominance.
// Dominator tree construction ----------------------------------------
//
// We use the algorithm described in Lengauer & Tarjan. 1979. A fast
// algorithm for finding dominators in a flowgraph.
// http://doi.acm.org/10.1145/357062.357071
//
// We also apply the optimizations to SLT described in Georgiadis et
// al, Finding Dominators in Practice, JGAA 2006,
// http://jgaa.info/accepted/2006/GeorgiadisTarjanWerneck2006.10.1.pdf
// to avoid the need for buckets of size > 1.
import (
"fmt"
"io"
"math/big"
"os"
)
// domNode represents a node in the dominator tree.
//
// TODO(adonovan): export this, when ready.
type domNode struct {
Block *BasicBlock // the basic block; n.Block.dom == n
Idom *domNode // immediate dominator (parent in dominator tree)
Children []*domNode // nodes dominated by this one
Level int // level number of node within tree; zero for root
Pre, Post int // pre- and post-order numbering within dominator tree
// Working state for Lengauer-Tarjan algorithm
// (during which Pre is repurposed for CFG DFS preorder number).
// TODO(adonovan): opt: measure allocating these as temps.
semi *domNode // semidominator
parent *domNode // parent in DFS traversal of CFG
ancestor *domNode // ancestor with least sdom
}
// ltDfs implements the depth-first search part of the LT algorithm.
func ltDfs(v *domNode, i int, preorder []*domNode) int {
preorder[i] = v
v.Pre = i // For now: DFS preorder of spanning tree of CFG
i++
v.semi = v
v.ancestor = nil
for _, succ := range v.Block.Succs {
if w := succ.dom; w.semi == nil {
w.parent = v
i = ltDfs(w, i, preorder)
}
}
return i
}
// ltEval implements the EVAL part of the LT algorithm.
func ltEval(v *domNode) *domNode {
// TODO(adonovan): opt: do path compression per simple LT.
u := v
for ; v.ancestor != nil; v = v.ancestor {
if v.semi.Pre < u.semi.Pre {
u = v
}
}
return u
}
// ltLink implements the LINK part of the LT algorithm.
func ltLink(v, w *domNode) {
w.ancestor = v
}
// buildDomTree computes the dominator tree of f using the LT algorithm.
// Precondition: all blocks are reachable (e.g. optimizeBlocks has been run).
//
func buildDomTree(f *Function) {
// The step numbers refer to the original LT paper; the
// reodering is due to Georgiadis.
// Initialize domNode nodes.
for _, b := range f.Blocks {
dom := b.dom
if dom == nil {
dom = &domNode{Block: b}
b.dom = dom
} else {
dom.Block = b // reuse
}
}
// Step 1. Number vertices by depth-first preorder.
n := len(f.Blocks)
preorder := make([]*domNode, n)
root := f.Blocks[0].dom
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
prenum := ltDfs(root, 0, preorder)
var recover *domNode
if f.Recover != nil {
recover = f.Recover.dom
ltDfs(recover, prenum, preorder)
}
buckets := make([]*domNode, n)
copy(buckets, preorder)
// In reverse preorder...
for i := n - 1; i > 0; i-- {
w := preorder[i]
// Step 3. Implicitly define the immediate dominator of each node.
for v := buckets[i]; v != w; v = buckets[v.Pre] {
u := ltEval(v)
if u.semi.Pre < i {
v.Idom = u
} else {
v.Idom = w
}
}
// Step 2. Compute the semidominators of all nodes.
w.semi = w.parent
for _, pred := range w.Block.Preds {
v := pred.dom
u := ltEval(v)
if u.semi.Pre < w.semi.Pre {
w.semi = u.semi
}
}
ltLink(w.parent, w)
if w.parent == w.semi {
w.Idom = w.parent
} else {
buckets[i] = buckets[w.semi.Pre]
buckets[w.semi.Pre] = w
}
}
// The final 'Step 3' is now outside the loop.
for v := buckets[0]; v != root; v = buckets[v.Pre] {
v.Idom = root
}
// Step 4. Explicitly define the immediate dominator of each
// node, in preorder.
for _, w := range preorder[1:] {
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
if w == root || w == recover {
w.Idom = nil
} else {
if w.Idom != w.semi {
w.Idom = w.Idom.Idom
}
// Calculate Children relation as inverse of Idom.
w.Idom.Children = append(w.Idom.Children, w)
}
// Clear working state.
w.semi = nil
w.parent = nil
w.ancestor = nil
}
numberDomTree(root, 0, 0, 0)
// printDomTreeDot(os.Stderr, f) // debugging
// printDomTreeText(os.Stderr, root, 0) // debugging
if f.Prog.mode&SanityCheckFunctions != 0 {
sanityCheckDomTree(f)
}
}
// numberDomTree sets the pre- and post-order numbers of a depth-first
// traversal of the dominator tree rooted at v. These are used to
// answer dominance queries in constant time. Also, it sets the level
// numbers (zero for the root) used for frontier computation.
//
func numberDomTree(v *domNode, pre, post, level int) (int, int) {
v.Level = level
level++
v.Pre = pre
pre++
for _, child := range v.Children {
pre, post = numberDomTree(child, pre, post, level)
}
v.Post = post
post++
return pre, post
}
// dominates returns true if b dominates c.
// Requires that dominance information is up-to-date.
//
func dominates(b, c *BasicBlock) bool {
return b.dom.Pre <= c.dom.Pre && c.dom.Post <= b.dom.Post
}
// Testing utilities ----------------------------------------
// sanityCheckDomTree checks the correctness of the dominator tree
// computed by the LT algorithm by comparing against the dominance
// relation computed by a naive Kildall-style forward dataflow
// analysis (Algorithm 10.16 from the "Dragon" book).
//
func sanityCheckDomTree(f *Function) {
n := len(f.Blocks)
// D[i] is the set of blocks that dominate f.Blocks[i],
// represented as a bit-set of block indices.
D := make([]big.Int, n)
one := big.NewInt(1)
// all is the set of all blocks; constant.
var all big.Int
all.Set(one).Lsh(&all, uint(n)).Sub(&all, one)
// Initialization.
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
for i, b := range f.Blocks {
if i == 0 || b == f.Recover {
// The root is dominated only by itself.
D[i].SetBit(&D[0], 0, 1)
} else {
// All other blocks are (initially) dominated
// by every block.
D[i].Set(&all)
}
}
// Iteration until fixed point.
for changed := true; changed; {
changed = false
for i, b := range f.Blocks {
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
if i == 0 || b == f.Recover {
continue
}
// Compute intersection across predecessors.
var x big.Int
x.Set(&all)
for _, pred := range b.Preds {
x.And(&x, &D[pred.Index])
}
x.SetBit(&x, i, 1) // a block always dominates itself.
if D[i].Cmp(&x) != 0 {
D[i].Set(&x)
changed = true
}
}
}
// Check the entire relation. O(n^2).
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
// The Recover block (if any) must be treated specially so we skip it.
ok := true
for i := 0; i < n; i++ {
for j := 0; j < n; j++ {
b, c := f.Blocks[i], f.Blocks[j]
go.tools/ssa: implement correct control flow for recovered panic. A function such as this: func one() (x int) { defer func() { recover() }() x = 1 panic("return") } that combines named return parameters (NRPs) with deferred calls that call recover, may return non-zero values despite the fact it doesn't even contain a return statement. (!) This requires a change to the SSA API: all functions' control-flow graphs now have a second entry point, called Recover, which is the block at which control flow resumes after a recovered panic. The Recover block simply loads the NRPs and returns them. As an optimization, most functions don't need a Recover block, so it is omitted. In fact it is only needed for functions that have NRPs and defer a call to another function that _may_ call recover. Dataflow analysis of SSA now requires extra work, since every may-panic instruction has an implicit control-flow edge to the Recover block. The only dataflow analysis so far implemented is SSA renaming, for which we make the following simplifying assumption: the Recover block only loads the NRPs and returns. This means we don't really need to analyze it, we can just skip the "lifting" of such NRPs. We also special-case the Recover block in the dominance computation. Rejected alternative approaches: - Specifying a Recover block for every defer instruction (like a traditional exception handler). This seemed like excessive generality, since Go programs only need the same degenerate form of Recover block. - Adding an instruction to set the Recover block immediately after the named return values are set up, so that dominance can be computed without special-casing. This didn't seem worth the effort. Interpreter: - This CL completely reimplements the panic/recover/ defer logic in the interpreter. It's clearer and simpler and closer to the model in the spec. - Some runtime panic messages have been changed to be closer to gc's, since tests depend on it. - The interpreter now requires that the runtime.runtimeError type be part of the SSA program. This requires that clients import this package prior to invoking the interpreter. This in turn requires (Importer).ImportPackage(path string), which this CL adds. - All $GOROOT/test/recover{,1,2,3}.go tests are now passing. NB, the bug described in coverage.go (defer/recover in a concatenated init function) remains. Will be fixed in a follow-up. Fixes golang/go#6381 R=gri CC=crawshaw, golang-dev https://golang.org/cl/13844043
2013-10-14 13:38:56 -06:00
if c == f.Recover {
continue
}
actual := dominates(b, c)
expected := D[j].Bit(i) == 1
if actual != expected {
fmt.Fprintf(os.Stderr, "dominates(%s, %s)==%t, want %t\n", b, c, actual, expected)
ok = false
}
}
}
if !ok {
panic("sanityCheckDomTree failed for " + f.String())
}
}
// Printing functions ----------------------------------------
// printDomTree prints the dominator tree as text, using indentation.
func printDomTreeText(w io.Writer, v *domNode, indent int) {
fmt.Fprintf(w, "%*s%s\n", 4*indent, "", v.Block)
for _, child := range v.Children {
printDomTreeText(w, child, indent+1)
}
}
// printDomTreeDot prints the dominator tree of f in AT&T GraphViz
// (.dot) format.
func printDomTreeDot(w io.Writer, f *Function) {
fmt.Fprintln(w, "//", f)
fmt.Fprintln(w, "digraph domtree {")
for i, b := range f.Blocks {
v := b.dom
fmt.Fprintf(w, "\tn%d [label=\"%s (%d, %d)\",shape=\"rectangle\"];\n", v.Pre, b, v.Pre, v.Post)
// TODO(adonovan): improve appearance of edges
// belonging to both dominator tree and CFG.
// Dominator tree edge.
if i != 0 {
fmt.Fprintf(w, "\tn%d -> n%d [style=\"solid\",weight=100];\n", v.Idom.Pre, v.Pre)
}
// CFG edges.
for _, pred := range b.Preds {
fmt.Fprintf(w, "\tn%d -> n%d [style=\"dotted\",weight=0];\n", pred.dom.Pre, v.Pre)
}
}
fmt.Fprintln(w, "}")
}