...

Source file src/pkg/cmd/compile/internal/ssa/sparsetreemap.go

     1	// Copyright 2016 The Go Authors. All rights reserved.
     2	// Use of this source code is governed by a BSD-style
     3	// license that can be found in the LICENSE file.
     4	
     5	package ssa
     6	
     7	import "fmt"
     8	
     9	// A SparseTreeMap encodes a subset of nodes within a tree
    10	// used for sparse-ancestor queries.
    11	//
    12	// Combined with a SparseTreeHelper, this supports an Insert
    13	// to add a tree node to the set and a Find operation to locate
    14	// the nearest tree ancestor of a given node such that the
    15	// ancestor is also in the set.
    16	//
    17	// Given a set of blocks {B1, B2, B3} within the dominator tree, established
    18	// by stm.Insert()ing B1, B2, B3, etc, a query at block B
    19	// (performed with stm.Find(stm, B, adjust, helper))
    20	// will return the member of the set that is the nearest strict
    21	// ancestor of B within the dominator tree, or nil if none exists.
    22	// The expected complexity of this operation is the log of the size
    23	// the set, given certain assumptions about sparsity (the log complexity
    24	// could be guaranteed with additional data structures whose constant-
    25	// factor overhead has not yet been justified.)
    26	//
    27	// The adjust parameter allows positioning of the insertion
    28	// and lookup points within a block -- one of
    29	// AdjustBefore, AdjustWithin, AdjustAfter,
    30	// where lookups at AdjustWithin can find insertions at
    31	// AdjustBefore in the same block, and lookups at AdjustAfter
    32	// can find insertions at either AdjustBefore or AdjustWithin
    33	// in the same block.  (Note that this assumes a gappy numbering
    34	// such that exit number or exit number is separated from its
    35	// nearest neighbor by at least 3).
    36	//
    37	// The Sparse Tree lookup algorithm is described by
    38	// Paul F. Dietz. Maintaining order in a linked list. In
    39	// Proceedings of the Fourteenth Annual ACM Symposium on
    40	// Theory of Computing, pages 122–127, May 1982.
    41	// and by
    42	// Ben Wegbreit. Faster retrieval from context trees.
    43	// Communications of the ACM, 19(9):526–529, September 1976.
    44	type SparseTreeMap RBTint32
    45	
    46	// A SparseTreeHelper contains indexing and allocation data
    47	// structures common to a collection of SparseTreeMaps, as well
    48	// as exposing some useful control-flow-related data to other
    49	// packages, such as gc.
    50	type SparseTreeHelper struct {
    51		Sdom   []SparseTreeNode // indexed by block.ID
    52		Po     []*Block         // exported data; the blocks, in a post-order
    53		Dom    []*Block         // exported data; the dominator of this block.
    54		Ponums []int32          // exported data; Po[Ponums[b.ID]] == b; the index of b in Po
    55	}
    56	
    57	// NewSparseTreeHelper returns a SparseTreeHelper for use
    58	// in the gc package, for example in phi-function placement.
    59	func NewSparseTreeHelper(f *Func) *SparseTreeHelper {
    60		dom := f.Idom()
    61		ponums := make([]int32, f.NumBlocks())
    62		po := postorderWithNumbering(f, ponums)
    63		return makeSparseTreeHelper(newSparseTree(f, dom), dom, po, ponums)
    64	}
    65	
    66	func (h *SparseTreeHelper) NewTree() *SparseTreeMap {
    67		return &SparseTreeMap{}
    68	}
    69	
    70	func makeSparseTreeHelper(sdom SparseTree, dom, po []*Block, ponums []int32) *SparseTreeHelper {
    71		helper := &SparseTreeHelper{Sdom: []SparseTreeNode(sdom),
    72			Dom:    dom,
    73			Po:     po,
    74			Ponums: ponums,
    75		}
    76		return helper
    77	}
    78	
    79	// A sparseTreeMapEntry contains the data stored in a binary search
    80	// data structure indexed by (dominator tree walk) entry and exit numbers.
    81	// Each entry is added twice, once keyed by entry-1/entry/entry+1 and
    82	// once keyed by exit+1/exit/exit-1.
    83	//
    84	// Within a sparse tree, the two entries added bracket all their descendant
    85	// entries within the tree; the first insertion is keyed by entry number,
    86	// which comes before all the entry and exit numbers of descendants, and
    87	// the second insertion is keyed by exit number, which comes after all the
    88	// entry and exit numbers of the descendants.
    89	type sparseTreeMapEntry struct {
    90		index        *SparseTreeNode // references the entry and exit numbers for a block in the sparse tree
    91		block        *Block          // TODO: store this in a separate index.
    92		data         interface{}
    93		sparseParent *sparseTreeMapEntry // references the nearest ancestor of this block in the sparse tree.
    94		adjust       int32               // at what adjustment was this node entered into the sparse tree? The same block may be entered more than once, but at different adjustments.
    95	}
    96	
    97	// Insert creates a definition within b with data x.
    98	// adjust indicates where in the block should be inserted:
    99	// AdjustBefore means defined at a phi function (visible Within or After in the same block)
   100	// AdjustWithin means defined within the block (visible After in the same block)
   101	// AdjustAfter means after the block (visible within child blocks)
   102	func (m *SparseTreeMap) Insert(b *Block, adjust int32, x interface{}, helper *SparseTreeHelper) {
   103		rbtree := (*RBTint32)(m)
   104		blockIndex := &helper.Sdom[b.ID]
   105		if blockIndex.entry == 0 {
   106			// assert unreachable
   107			return
   108		}
   109		// sp will be the sparse parent in this sparse tree (nearest ancestor in the larger tree that is also in this sparse tree)
   110		sp := m.findEntry(b, adjust, helper)
   111		entry := &sparseTreeMapEntry{index: blockIndex, block: b, data: x, sparseParent: sp, adjust: adjust}
   112	
   113		right := blockIndex.exit - adjust
   114		_ = rbtree.Insert(right, entry)
   115	
   116		left := blockIndex.entry + adjust
   117		_ = rbtree.Insert(left, entry)
   118	
   119		// This newly inserted block may now be the sparse parent of some existing nodes (the new sparse children of this block)
   120		// Iterate over nodes bracketed by this new node to correct their parent, but not over the proper sparse descendants of those nodes.
   121		_, d := rbtree.Lub(left) // Lub (not EQ) of left is either right or a sparse child
   122		for tme := d.(*sparseTreeMapEntry); tme != entry; tme = d.(*sparseTreeMapEntry) {
   123			tme.sparseParent = entry
   124			// all descendants of tme are unchanged;
   125			// next sparse sibling (or right-bracketing sparse parent == entry) is first node after tme.index.exit - tme.adjust
   126			_, d = rbtree.Lub(tme.index.exit - tme.adjust)
   127		}
   128	}
   129	
   130	// Find returns the definition visible from block b, or nil if none can be found.
   131	// Adjust indicates where the block should be searched.
   132	// AdjustBefore searches before the phi functions of b.
   133	// AdjustWithin searches starting at the phi functions of b.
   134	// AdjustAfter searches starting at the exit from the block, including normal within-block definitions.
   135	//
   136	// Note that Finds are properly nested with Inserts:
   137	// m.Insert(b, a) followed by m.Find(b, a) will not return the result of the insert,
   138	// but m.Insert(b, AdjustBefore) followed by m.Find(b, AdjustWithin) will.
   139	//
   140	// Another way to think of this is that Find searches for inputs, Insert defines outputs.
   141	func (m *SparseTreeMap) Find(b *Block, adjust int32, helper *SparseTreeHelper) interface{} {
   142		v := m.findEntry(b, adjust, helper)
   143		if v == nil {
   144			return nil
   145		}
   146		return v.data
   147	}
   148	
   149	func (m *SparseTreeMap) findEntry(b *Block, adjust int32, helper *SparseTreeHelper) *sparseTreeMapEntry {
   150		rbtree := (*RBTint32)(m)
   151		if rbtree == nil {
   152			return nil
   153		}
   154		blockIndex := &helper.Sdom[b.ID]
   155	
   156		// The Glb (not EQ) of this probe is either the entry-indexed end of a sparse parent
   157		// or the exit-indexed end of a sparse sibling
   158		_, v := rbtree.Glb(blockIndex.entry + adjust)
   159	
   160		if v == nil {
   161			return nil
   162		}
   163	
   164		otherEntry := v.(*sparseTreeMapEntry)
   165		if otherEntry.index.exit >= blockIndex.exit { // otherEntry exit after blockIndex exit; therefore, brackets
   166			return otherEntry
   167		}
   168		// otherEntry is a sparse Sibling, and shares the same sparse parent (nearest ancestor within larger tree)
   169		sp := otherEntry.sparseParent
   170		if sp != nil {
   171			if sp.index.exit < blockIndex.exit { // no ancestor found
   172				return nil
   173			}
   174			return sp
   175		}
   176		return nil
   177	}
   178	
   179	func (m *SparseTreeMap) String() string {
   180		tree := (*RBTint32)(m)
   181		return tree.String()
   182	}
   183	
   184	func (e *sparseTreeMapEntry) String() string {
   185		if e == nil {
   186			return "nil"
   187		}
   188		return fmt.Sprintf("(index=%v, block=%v, data=%v)->%v", e.index, e.block, e.data, e.sparseParent)
   189	}
   190	

View as plain text