1 ------------------------------------------------------------------------------
3 -- GNAT COMPILER COMPONENTS --
5 -- G N A T . P E R F E C T _ H A S H _ G E N E R A T O R S --
9 -- Copyright (C) 2002-2010, AdaCore --
11 -- GNAT is free software; you can redistribute it and/or modify it under --
12 -- terms of the GNU General Public License as published by the Free Soft- --
13 -- ware Foundation; either version 2, or (at your option) any later ver- --
14 -- sion. GNAT is distributed in the hope that it will be useful, but WITH- --
15 -- OUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY --
16 -- or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License --
17 -- for more details. You should have received a copy of the GNU General --
18 -- Public License distributed with GNAT; see file COPYING. If not, write --
19 -- to the Free Software Foundation, 51 Franklin Street, Fifth Floor, --
20 -- Boston, MA 02110-1301, USA. --
22 -- As a special exception, if other files instantiate generics from this --
23 -- unit, or you link this unit with other files to produce an executable, --
24 -- this unit does not by itself cause the resulting executable to be --
25 -- covered by the GNU General Public License. This exception does not --
26 -- however invalidate any other reasons why the executable file might be --
27 -- covered by the GNU Public License. --
29 -- GNAT was originally developed by the GNAT team at New York University. --
30 -- Extensive contributions were provided by Ada Core Technologies Inc. --
32 ------------------------------------------------------------------------------
34 with Ada
.IO_Exceptions
; use Ada
.IO_Exceptions
;
35 with Ada
.Characters
.Handling
; use Ada
.Characters
.Handling
;
38 with GNAT
.Heap_Sort_G
;
39 with GNAT
.OS_Lib
; use GNAT
.OS_Lib
;
42 package body GNAT
.Perfect_Hash_Generators
is
44 -- We are using the algorithm of J. Czech as described in Zbigniew J.
45 -- Czech, George Havas, and Bohdan S. Majewski ``An Optimal Algorithm for
46 -- Generating Minimal Perfect Hash Functions'', Information Processing
47 -- Letters, 43(1992) pp.257-264, Oct.1992
49 -- This minimal perfect hash function generator is based on random graphs
50 -- and produces a hash function of the form:
52 -- h (w) = (g (f1 (w)) + g (f2 (w))) mod m
54 -- where f1 and f2 are functions that map strings into integers, and g is
55 -- a function that maps integers into [0, m-1]. h can be order preserving.
56 -- For instance, let W = {w_0, ..., w_i, ..., w_m-1}, h can be defined
57 -- such that h (w_i) = i.
59 -- This algorithm defines two possible constructions of f1 and f2. Method
60 -- b) stores the hash function in less memory space at the expense of
63 -- a) fk (w) = sum (for i in 1 .. length (w)) (Tk (i, w (i))) mod n
65 -- size (Tk) = max (for w in W) (length (w)) * size (used char set)
67 -- b) fk (w) = sum (for i in 1 .. length (w)) (Tk (i) * w (i)) mod n
69 -- size (Tk) = max (for w in W) (length (w)) but the table lookups are
70 -- replaced by multiplications.
72 -- where Tk values are randomly generated. n is defined later on but the
73 -- algorithm recommends to use a value a little bit greater than 2m. Note
74 -- that for large values of m, the main memory space requirements comes
75 -- from the memory space for storing function g (>= 2m entries).
77 -- Random graphs are frequently used to solve difficult problems that do
78 -- not have polynomial solutions. This algorithm is based on a weighted
79 -- undirected graph. It comprises two steps: mapping and assignment.
81 -- In the mapping step, a graph G = (V, E) is constructed, where = {0, 1,
82 -- ..., n-1} and E = {(for w in W) (f1 (w), f2 (w))}. In order for the
83 -- assignment step to be successful, G has to be acyclic. To have a high
84 -- probability of generating an acyclic graph, n >= 2m. If it is not
85 -- acyclic, Tk have to be regenerated.
87 -- In the assignment step, the algorithm builds function g. As G is
88 -- acyclic, there is a vertex v1 with only one neighbor v2. Let w_i be
89 -- the word such that v1 = f1 (w_i) and v2 = f2 (w_i). Let g (v1) = 0 by
90 -- construction and g (v2) = (i - g (v1)) mod n (or h (i) - g (v1) mod n).
91 -- If word w_j is such that v2 = f1 (w_j) and v3 = f2 (w_j), g (v3) = (j -
92 -- g (v2)) mod (or to be general, (h (j) - g (v2)) mod n). If w_i has no
93 -- neighbor, then another vertex is selected. The algorithm traverses G to
94 -- assign values to all the vertices. It cannot assign a value to an
95 -- already assigned vertex as G is acyclic.
97 subtype Word_Id
is Integer;
98 subtype Key_Id
is Integer;
99 subtype Vertex_Id
is Integer;
100 subtype Edge_Id
is Integer;
101 subtype Table_Id
is Integer;
103 No_Vertex
: constant Vertex_Id
:= -1;
104 No_Edge
: constant Edge_Id
:= -1;
105 No_Table
: constant Table_Id
:= -1;
107 type Word_Type
is new String_Access
;
108 procedure Free_Word
(W
: in out Word_Type
);
109 function New_Word
(S
: String) return Word_Type
;
111 procedure Resize_Word
(W
: in out Word_Type
; Len
: Natural);
112 -- Resize string W to have a length Len
114 type Key_Type
is record
117 -- A key corresponds to an edge in the algorithm graph
119 type Vertex_Type
is record
123 -- A vertex can be involved in several edges. First and Last are the bounds
124 -- of an array of edges stored in a global edge table.
126 type Edge_Type
is record
131 -- An edge is a peer of vertices. In the algorithm, a key is associated to
134 package WT
is new GNAT
.Table
(Word_Type
, Word_Id
, 0, 32, 32);
135 package IT
is new GNAT
.Table
(Integer, Integer, 0, 32, 32);
136 -- The two main tables. WT is used to store the words in their initial
137 -- version and in their reduced version (that is words reduced to their
138 -- significant characters). As an instance of GNAT.Table, WT does not
139 -- initialize string pointers to null. This initialization has to be done
140 -- manually when the table is allocated. IT is used to store several
141 -- tables of components containing only integers.
143 function Image
(Int
: Integer; W
: Natural := 0) return String;
144 function Image
(Str
: String; W
: Natural := 0) return String;
145 -- Return a string which includes string Str or integer Int preceded by
146 -- leading spaces if required by width W.
148 function Trim_Trailing_Nuls
(Str
: String) return String;
149 -- Return Str with trailing NUL characters removed
151 Output
: File_Descriptor
renames GNAT
.OS_Lib
.Standout
;
154 EOL
: constant Character := ASCII
.LF
;
156 Max
: constant := 78;
158 Line
: String (1 .. Max
);
159 -- Use this line to provide buffered IO
161 procedure Add
(C
: Character);
162 procedure Add
(S
: String);
163 -- Add a character or a string in Line and update Last
166 (F
: File_Descriptor
;
174 -- Write string S into file F as a element of an array of one or two
175 -- dimensions. Fk (resp. Lk and Ck) indicates the first (resp last and
176 -- current) index in the k-th dimension. If F1 = L1 the array is considered
177 -- as a one dimension array. This dimension is described by F2 and L2. This
178 -- routine takes care of all the parenthesis, spaces and commas needed to
179 -- format correctly the array. Moreover, the array is well indented and is
180 -- wrapped to fit in a 80 col line. When the line is full, the routine
181 -- writes it into file F. When the array is completed, the routine adds
182 -- semi-colon and writes the line into file F.
184 procedure New_Line
(File
: File_Descriptor
);
185 -- Simulate Ada.Text_IO.New_Line with GNAT.OS_Lib
187 procedure Put
(File
: File_Descriptor
; Str
: String);
188 -- Simulate Ada.Text_IO.Put with GNAT.OS_Lib
190 procedure Put_Used_Char_Set
(File
: File_Descriptor
; Title
: String);
191 -- Output a title and a used character set
193 procedure Put_Int_Vector
194 (File
: File_Descriptor
;
198 -- Output a title and a vector
200 procedure Put_Int_Matrix
201 (File
: File_Descriptor
;
206 -- Output a title and a matrix. When the matrix has only one non-empty
207 -- dimension (Len_2 = 0), output a vector.
209 procedure Put_Edges
(File
: File_Descriptor
; Title
: String);
210 -- Output a title and an edge table
212 procedure Put_Initial_Keys
(File
: File_Descriptor
; Title
: String);
213 -- Output a title and a key table
215 procedure Put_Reduced_Keys
(File
: File_Descriptor
; Title
: String);
216 -- Output a title and a key table
218 procedure Put_Vertex_Table
(File
: File_Descriptor
; Title
: String);
219 -- Output a title and a vertex table
221 function Ada_File_Base_Name
(Pkg_Name
: String) return String;
222 -- Return the base file name (i.e. without .ads/.adb extension) for an
223 -- Ada source file containing the named package, using the standard GNAT
224 -- file-naming convention. For example, if Pkg_Name is "Parent.Child", we
225 -- return "parent-child".
227 ----------------------------------
228 -- Character Position Selection --
229 ----------------------------------
231 -- We reduce the maximum key size by selecting representative positions
232 -- in these keys. We build a matrix with one word per line. We fill the
233 -- remaining space of a line with ASCII.NUL. The heuristic selects the
234 -- position that induces the minimum number of collisions. If there are
235 -- collisions, select another position on the reduced key set responsible
236 -- of the collisions. Apply the heuristic until there is no more collision.
238 procedure Apply_Position_Selection
;
239 -- Apply Position selection and build the reduced key table
241 procedure Parse_Position_Selection
(Argument
: String);
242 -- Parse Argument and compute the position set. Argument is list of
243 -- substrings separated by commas. Each substring represents a position
244 -- or a range of positions (like x-y).
246 procedure Select_Character_Set
;
247 -- Define an optimized used character set like Character'Pos in order not
248 -- to allocate tables of 256 entries.
250 procedure Select_Char_Position
;
251 -- Find a min char position set in order to reduce the max key length. The
252 -- heuristic selects the position that induces the minimum number of
253 -- collisions. If there are collisions, select another position on the
254 -- reduced key set responsible of the collisions. Apply the heuristic until
255 -- there is no collision.
257 -----------------------------
258 -- Random Graph Generation --
259 -----------------------------
261 procedure Random
(Seed
: in out Natural);
262 -- Simulate Ada.Discrete_Numerics.Random
264 procedure Generate_Mapping_Table
268 Seed
: in out Natural);
269 -- Random generation of the tables below. T is already allocated
271 procedure Generate_Mapping_Tables
273 Seed
: in out Natural);
274 -- Generate the mapping tables T1 and T2. They are used to define fk (w) =
275 -- sum (for i in 1 .. length (w)) (Tk (i, w (i))) mod n. Keys, NK and Chars
276 -- are used to compute the matrix size.
278 ---------------------------
279 -- Algorithm Computation --
280 ---------------------------
282 procedure Compute_Edges_And_Vertices
(Opt
: Optimization
);
283 -- Compute the edge and vertex tables. These are empty when a self loop is
284 -- detected (f1 (w) = f2 (w)). The edge table is sorted by X value and then
285 -- Y value. Keys is the key table and NK the number of keys. Chars is the
286 -- set of characters really used in Keys. NV is the number of vertices
287 -- recommended by the algorithm. T1 and T2 are the mapping tables needed to
288 -- compute f1 (w) and f2 (w).
290 function Acyclic
return Boolean;
291 -- Return True when the graph is acyclic. Vertices is the current vertex
292 -- table and Edges the current edge table.
294 procedure Assign_Values_To_Vertices
;
295 -- Execute the assignment step of the algorithm. Keys is the current key
296 -- table. Vertices and Edges represent the random graph. G is the result of
297 -- the assignment step such that:
298 -- h (w) = (g (f1 (w)) + g (f2 (w))) mod m
303 Opt
: Optimization
) return Natural;
304 -- For an optimization of CPU_Time return
305 -- fk (w) = sum (for i in 1 .. length (w)) (Tk (i, w (i))) mod n
306 -- For an optimization of Memory_Space return
307 -- fk (w) = sum (for i in 1 .. length (w)) (Tk (i) * w (i)) mod n
310 -------------------------------
311 -- Internal Table Management --
312 -------------------------------
314 function Allocate
(N
: Natural; S
: Natural := 1) return Table_Id
;
315 -- Allocate N * S ints from IT table
321 Keys
: Table_Id
:= No_Table
;
323 -- NK : Number of Keys
325 function Initial
(K
: Key_Id
) return Word_Id
;
326 pragma Inline
(Initial
);
328 function Reduced
(K
: Key_Id
) return Word_Id
;
329 pragma Inline
(Reduced
);
331 function Get_Key
(N
: Key_Id
) return Key_Type
;
332 procedure Set_Key
(N
: Key_Id
; Item
: Key_Type
);
333 -- Get or Set Nth element of Keys table
339 Char_Pos_Set
: Table_Id
:= No_Table
;
340 Char_Pos_Set_Len
: Natural;
341 -- Character Selected Position Set
343 function Get_Char_Pos
(P
: Natural) return Natural;
344 procedure Set_Char_Pos
(P
: Natural; Item
: Natural);
345 -- Get or Set the string position of the Pth selected character
351 Used_Char_Set
: Table_Id
:= No_Table
;
352 Used_Char_Set_Len
: Natural;
353 -- Used Character Set : Define a new character mapping. When all the
354 -- characters are not present in the keys, in order to reduce the size
355 -- of some tables, we redefine the character mapping.
357 function Get_Used_Char
(C
: Character) return Natural;
358 procedure Set_Used_Char
(C
: Character; Item
: Natural);
364 T1
: Table_Id
:= No_Table
;
365 T2
: Table_Id
:= No_Table
;
368 -- T1 : Values table to compute F1
369 -- T2 : Values table to compute F2
371 function Get_Table
(T
: Integer; X
, Y
: Natural) return Natural;
372 procedure Set_Table
(T
: Integer; X
, Y
: Natural; Item
: Natural);
378 G
: Table_Id
:= No_Table
;
380 -- Values table to compute G
382 NT
: Natural := Default_Tries
;
383 -- Number of tries running the algorithm before raising an error
385 function Get_Graph
(N
: Natural) return Integer;
386 procedure Set_Graph
(N
: Natural; Item
: Integer);
387 -- Get or Set Nth element of graph
393 Edge_Size
: constant := 3;
394 Edges
: Table_Id
:= No_Table
;
396 -- Edges : Edge table of the random graph G
398 function Get_Edges
(F
: Natural) return Edge_Type
;
399 procedure Set_Edges
(F
: Natural; Item
: Edge_Type
);
405 Vertex_Size
: constant := 2;
407 Vertices
: Table_Id
:= No_Table
;
408 -- Vertex table of the random graph G
411 -- Number of Vertices
413 function Get_Vertices
(F
: Natural) return Vertex_Type
;
414 procedure Set_Vertices
(F
: Natural; Item
: Vertex_Type
);
415 -- Comments needed ???
418 -- Ratio between Keys and Vertices (parameter of Czech's algorithm)
421 -- Optimization mode (memory vs CPU)
423 Max_Key_Len
: Natural := 0;
424 Min_Key_Len
: Natural := 0;
425 -- Maximum and minimum of all the word length
430 function Type_Size
(L
: Natural) return Natural;
431 -- Given the last L of an unsigned integer type T, return its size
437 function Acyclic
return Boolean is
438 Marks
: array (0 .. NV
- 1) of Vertex_Id
:= (others => No_Vertex
);
440 function Traverse
(Edge
: Edge_Id
; Mark
: Vertex_Id
) return Boolean;
441 -- Propagate Mark from X to Y. X is already marked. Mark Y and propagate
442 -- it to the edges of Y except the one representing the same key. Return
443 -- False when Y is marked with Mark.
449 function Traverse
(Edge
: Edge_Id
; Mark
: Vertex_Id
) return Boolean is
450 E
: constant Edge_Type
:= Get_Edges
(Edge
);
451 K
: constant Key_Id
:= E
.Key
;
452 Y
: constant Vertex_Id
:= E
.Y
;
453 M
: constant Vertex_Id
:= Marks
(E
.Y
);
460 elsif M
= No_Vertex
then
462 V
:= Get_Vertices
(Y
);
464 for J
in V
.First
.. V
.Last
loop
466 -- Do not propagate to the edge representing the same key
468 if Get_Edges
(J
).Key
/= K
469 and then not Traverse
(J
, Mark
)
481 -- Start of processing for Acyclic
484 -- Edges valid range is
486 for J
in 1 .. Edges_Len
- 1 loop
488 Edge
:= Get_Edges
(J
);
490 -- Mark X of E when it has not been already done
492 if Marks
(Edge
.X
) = No_Vertex
then
493 Marks
(Edge
.X
) := Edge
.X
;
496 -- Traverse E when this has not already been done
498 if Marks
(Edge
.Y
) = No_Vertex
499 and then not Traverse
(J
, Edge
.X
)
508 ------------------------
509 -- Ada_File_Base_Name --
510 ------------------------
512 function Ada_File_Base_Name
(Pkg_Name
: String) return String is
514 -- Convert to lower case, then replace '.' with '-'
516 return Result
: String := To_Lower
(Pkg_Name
) do
517 for J
in Result
'Range loop
518 if Result
(J
) = '.' then
523 end Ada_File_Base_Name
;
529 procedure Add
(C
: Character) is
530 pragma Assert
(C
/= ASCII
.NUL
);
532 Line
(Last
+ 1) := C
;
540 procedure Add
(S
: String) is
541 Len
: constant Natural := S
'Length;
543 for J
in S
'Range loop
544 pragma Assert
(S
(J
) /= ASCII
.NUL
);
548 Line
(Last
+ 1 .. Last
+ Len
) := S
;
556 function Allocate
(N
: Natural; S
: Natural := 1) return Table_Id
is
557 L
: constant Integer := IT
.Last
;
559 IT
.Set_Last
(L
+ N
* S
);
561 -- Initialize, so debugging printouts don't trip over uninitialized
564 for J
in L
+ 1 .. IT
.Last
loop
571 ------------------------------
572 -- Apply_Position_Selection --
573 ------------------------------
575 procedure Apply_Position_Selection
is
577 for J
in 0 .. NK
- 1 loop
579 IW
: constant String := WT
.Table
(Initial
(J
)).all;
580 RW
: String (1 .. IW
'Length) := (others => ASCII
.NUL
);
581 N
: Natural := IW
'First - 1;
584 -- Select the characters of Word included in the position
587 for C
in 0 .. Char_Pos_Set_Len
- 1 loop
588 exit when IW
(Get_Char_Pos
(C
)) = ASCII
.NUL
;
590 RW
(N
) := IW
(Get_Char_Pos
(C
));
593 -- Build the new table with the reduced word. Be careful
594 -- to deallocate the old version to avoid memory leaks.
596 Free_Word
(WT
.Table
(Reduced
(J
)));
597 WT
.Table
(Reduced
(J
)) := New_Word
(RW
);
598 Set_Key
(J
, (Edge
=> No_Edge
));
601 end Apply_Position_Selection
;
603 -------------------------------
604 -- Assign_Values_To_Vertices --
605 -------------------------------
607 procedure Assign_Values_To_Vertices
is
610 procedure Assign
(X
: Vertex_Id
);
611 -- Execute assignment on X's neighbors except the vertex that we are
612 -- coming from which is already assigned.
618 procedure Assign
(X
: Vertex_Id
) is
620 V
: constant Vertex_Type
:= Get_Vertices
(X
);
623 for J
in V
.First
.. V
.Last
loop
626 if Get_Graph
(E
.Y
) = -1 then
627 Set_Graph
(E
.Y
, (E
.Key
- Get_Graph
(X
)) mod NK
);
633 -- Start of processing for Assign_Values_To_Vertices
636 -- Value -1 denotes an uninitialized value as it is supposed to
637 -- be in the range 0 .. NK.
641 G
:= Allocate
(G_Len
, 1);
644 for J
in 0 .. G_Len
- 1 loop
648 for K
in 0 .. NK
- 1 loop
649 X
:= Get_Edges
(Get_Key
(K
).Edge
).X
;
651 if Get_Graph
(X
) = -1 then
657 for J
in 0 .. G_Len
- 1 loop
658 if Get_Graph
(J
) = -1 then
664 Put_Int_Vector
(Output
, "Assign Values To Vertices", G
, G_Len
);
666 end Assign_Values_To_Vertices
;
672 procedure Compute
(Position
: String := Default_Position
) is
673 Success
: Boolean := False;
677 raise Program_Error
with "keywords set cannot be empty";
681 Put_Initial_Keys
(Output
, "Initial Key Table");
684 if Position
'Length /= 0 then
685 Parse_Position_Selection
(Position
);
687 Select_Char_Position
;
692 (Output
, "Char Position Set", Char_Pos_Set
, Char_Pos_Set_Len
);
695 Apply_Position_Selection
;
698 Put_Reduced_Keys
(Output
, "Reduced Keys Table");
701 Select_Character_Set
;
704 Put_Used_Char_Set
(Output
, "Character Position Table");
707 -- Perform Czech's algorithm
709 for J
in 1 .. NT
loop
710 Generate_Mapping_Tables
(Opt
, S
);
711 Compute_Edges_And_Vertices
(Opt
);
713 -- When graph is not empty (no self-loop from previous operation) and
716 if 0 < Edges_Len
and then Acyclic
then
723 raise Too_Many_Tries
;
726 Assign_Values_To_Vertices
;
729 --------------------------------
730 -- Compute_Edges_And_Vertices --
731 --------------------------------
733 procedure Compute_Edges_And_Vertices
(Opt
: Optimization
) is
738 Vertex
: Vertex_Type
;
739 Not_Acyclic
: Boolean := False;
741 procedure Move
(From
: Natural; To
: Natural);
742 function Lt
(L
, R
: Natural) return Boolean;
743 -- Subprograms needed for GNAT.Heap_Sort_G
749 function Lt
(L
, R
: Natural) return Boolean is
750 EL
: constant Edge_Type
:= Get_Edges
(L
);
751 ER
: constant Edge_Type
:= Get_Edges
(R
);
753 return EL
.X
< ER
.X
or else (EL
.X
= ER
.X
and then EL
.Y
< ER
.Y
);
760 procedure Move
(From
: Natural; To
: Natural) is
762 Set_Edges
(To
, Get_Edges
(From
));
765 package Sorting
is new GNAT
.Heap_Sort_G
(Move
, Lt
);
767 -- Start of processing for Compute_Edges_And_Vertices
770 -- We store edges from 1 to 2 * NK and leave zero alone in order to use
773 Edges_Len
:= 2 * NK
+ 1;
775 if Edges
= No_Table
then
776 Edges
:= Allocate
(Edges_Len
, Edge_Size
);
779 if Vertices
= No_Table
then
780 Vertices
:= Allocate
(NV
, Vertex_Size
);
783 for J
in 0 .. NV
- 1 loop
784 Set_Vertices
(J
, (No_Vertex
, No_Vertex
- 1));
787 -- For each w, X = f1 (w) and Y = f2 (w)
789 for J
in 0 .. NK
- 1 loop
794 X
:= Sum
(WT
.Table
(Reduced
(J
)), T1
, Opt
);
795 Y
:= Sum
(WT
.Table
(Reduced
(J
)), T2
, Opt
);
797 -- Discard T1 and T2 as soon as we discover a self loop
804 -- We store (X, Y) and (Y, X) to ease assignment step
806 Set_Edges
(2 * J
+ 1, (X
, Y
, J
));
807 Set_Edges
(2 * J
+ 2, (Y
, X
, J
));
810 -- Return an empty graph when self loop detected
817 Put_Edges
(Output
, "Unsorted Edge Table");
818 Put_Int_Matrix
(Output
, "Function Table 1", T1
,
820 Put_Int_Matrix
(Output
, "Function Table 2", T2
,
824 -- Enforce consistency between edges and keys. Construct Vertices and
825 -- compute the list of neighbors of a vertex First .. Last as Edges
826 -- is sorted by X and then Y. To compute the neighbor list, sort the
829 Sorting
.Sort
(Edges_Len
- 1);
832 Put_Edges
(Output
, "Sorted Edge Table");
833 Put_Int_Matrix
(Output
, "Function Table 1", T1
,
835 Put_Int_Matrix
(Output
, "Function Table 2", T2
,
839 -- Edges valid range is 1 .. 2 * NK
841 for E
in 1 .. Edges_Len
- 1 loop
842 Edge
:= Get_Edges
(E
);
843 Key
:= Get_Key
(Edge
.Key
);
845 if Key
.Edge
= No_Edge
then
847 Set_Key
(Edge
.Key
, Key
);
850 Vertex
:= Get_Vertices
(Edge
.X
);
852 if Vertex
.First
= No_Edge
then
857 Set_Vertices
(Edge
.X
, Vertex
);
861 Put_Reduced_Keys
(Output
, "Key Table");
862 Put_Edges
(Output
, "Edge Table");
863 Put_Vertex_Table
(Output
, "Vertex Table");
866 end Compute_Edges_And_Vertices
;
874 Item_Size
: out Natural;
875 Length_1
: out Natural;
876 Length_2
: out Natural)
880 when Character_Position
=>
882 Length_1
:= Char_Pos_Set_Len
;
885 when Used_Character_Set
=>
890 when Function_Table_1
891 | Function_Table_2
=>
892 Item_Size
:= Type_Size
(NV
);
897 Item_Size
:= Type_Size
(NK
);
907 procedure Finalize
is
910 Put
(Output
, "Finalize");
914 -- Deallocate all the WT components (both initial and reduced
915 -- ones) to avoid memory leaks.
917 for W
in 0 .. WT
.Last
loop
918 Free_Word
(WT
.Table
(W
));
923 -- Reset all variables for next usage
927 Char_Pos_Set
:= No_Table
;
928 Char_Pos_Set_Len
:= 0;
930 Used_Char_Set
:= No_Table
;
931 Used_Char_Set_Len
:= 0;
945 Vertices
:= No_Table
;
957 procedure Free_Word
(W
: in out Word_Type
) is
964 ----------------------------
965 -- Generate_Mapping_Table --
966 ----------------------------
968 procedure Generate_Mapping_Table
972 Seed
: in out Natural)
975 for J
in 0 .. L1
- 1 loop
976 for K
in 0 .. L2
- 1 loop
978 Set_Table
(Tab
, J
, K
, Seed
mod NV
);
981 end Generate_Mapping_Table
;
983 -----------------------------
984 -- Generate_Mapping_Tables --
985 -----------------------------
987 procedure Generate_Mapping_Tables
989 Seed
: in out Natural)
992 -- If T1 and T2 are already allocated no need to do it twice. Reuse them
993 -- as their size has not changed.
995 if T1
= No_Table
and then T2
= No_Table
then
997 Used_Char_Last
: Natural := 0;
1001 if Opt
= CPU_Time
then
1002 for P
in reverse Character'Range loop
1003 Used_Char
:= Get_Used_Char
(P
);
1004 if Used_Char
/= 0 then
1005 Used_Char_Last
:= Used_Char
;
1011 T1_Len
:= Char_Pos_Set_Len
;
1012 T2_Len
:= Used_Char_Last
+ 1;
1013 T1
:= Allocate
(T1_Len
* T2_Len
);
1014 T2
:= Allocate
(T1_Len
* T2_Len
);
1018 Generate_Mapping_Table
(T1
, T1_Len
, T2_Len
, Seed
);
1019 Generate_Mapping_Table
(T2
, T1_Len
, T2_Len
, Seed
);
1022 Put_Used_Char_Set
(Output
, "Used Character Set");
1023 Put_Int_Matrix
(Output
, "Function Table 1", T1
,
1025 Put_Int_Matrix
(Output
, "Function Table 2", T2
,
1028 end Generate_Mapping_Tables
;
1034 function Get_Char_Pos
(P
: Natural) return Natural is
1035 N
: constant Natural := Char_Pos_Set
+ P
;
1037 return IT
.Table
(N
);
1044 function Get_Edges
(F
: Natural) return Edge_Type
is
1045 N
: constant Natural := Edges
+ (F
* Edge_Size
);
1048 E
.X
:= IT
.Table
(N
);
1049 E
.Y
:= IT
.Table
(N
+ 1);
1050 E
.Key
:= IT
.Table
(N
+ 2);
1058 function Get_Graph
(N
: Natural) return Integer is
1060 return IT
.Table
(G
+ N
);
1067 function Get_Key
(N
: Key_Id
) return Key_Type
is
1070 K
.Edge
:= IT
.Table
(Keys
+ N
);
1078 function Get_Table
(T
: Integer; X
, Y
: Natural) return Natural is
1079 N
: constant Natural := T
+ (Y
* T1_Len
) + X
;
1081 return IT
.Table
(N
);
1088 function Get_Used_Char
(C
: Character) return Natural is
1089 N
: constant Natural := Used_Char_Set
+ Character'Pos (C
);
1091 return IT
.Table
(N
);
1098 function Get_Vertices
(F
: Natural) return Vertex_Type
is
1099 N
: constant Natural := Vertices
+ (F
* Vertex_Size
);
1102 V
.First
:= IT
.Table
(N
);
1103 V
.Last
:= IT
.Table
(N
+ 1);
1111 function Image
(Int
: Integer; W
: Natural := 0) return String is
1112 B
: String (1 .. 32);
1115 procedure Img
(V
: Natural);
1116 -- Compute image of V into B, starting at B (L), incrementing L
1122 procedure Img
(V
: Natural) is
1129 B
(L
) := Character'Val ((V
mod 10) + Character'Pos ('0'));
1132 -- Start of processing for Image
1143 return Image
(B
(1 .. L
), W
);
1150 function Image
(Str
: String; W
: Natural := 0) return String is
1151 Len
: constant Natural := Str
'Length;
1152 Max
: Natural := Len
;
1160 Buf
: String (1 .. Max
) := (1 .. Max
=> ' ');
1163 for J
in 0 .. Len
- 1 loop
1164 Buf
(Max
- Len
+ 1 + J
) := Str
(Str
'First + J
);
1175 function Initial
(K
: Key_Id
) return Word_Id
is
1184 procedure Initialize
1186 K_To_V
: Float := Default_K_To_V
;
1187 Optim
: Optimization
:= Memory_Space
;
1188 Tries
: Positive := Default_Tries
)
1192 Put
(Output
, "Initialize");
1196 -- Deallocate the part of the table concerning the reduced words.
1197 -- Initial words are already present in the table. We may have reduced
1198 -- words already there because a previous computation failed. We are
1199 -- currently retrying and the reduced words have to be deallocated.
1201 for W
in Reduced
(0) .. WT
.Last
loop
1202 Free_Word
(WT
.Table
(W
));
1207 -- Initialize of computation variables
1211 Char_Pos_Set
:= No_Table
;
1212 Char_Pos_Set_Len
:= 0;
1214 Used_Char_Set
:= No_Table
;
1215 Used_Char_Set_Len
:= 0;
1229 Vertices
:= No_Table
;
1238 raise Program_Error
with "K to V ratio cannot be lower than 2.0";
1241 -- Do not accept a value of K2V too close to 2.0 such that once
1242 -- rounded up, NV = 2 * NK because the algorithm would not converge.
1244 NV
:= Natural (Float (NK
) * K2V
);
1245 if NV
<= 2 * NK
then
1249 Keys
:= Allocate
(NK
);
1251 -- Resize initial words to have all of them at the same size
1252 -- (so the size of the largest one).
1254 for K
in 0 .. NK
- 1 loop
1255 Resize_Word
(WT
.Table
(Initial
(K
)), Max_Key_Len
);
1258 -- Allocated the table to store the reduced words. As WT is a
1259 -- GNAT.Table (using C memory management), pointers have to be
1260 -- explicitly initialized to null.
1262 WT
.Set_Last
(Reduced
(NK
- 1));
1263 for W
in 0 .. NK
- 1 loop
1264 WT
.Table
(Reduced
(W
)) := null;
1272 procedure Insert
(Value
: String) is
1273 Len
: constant Natural := Value
'Length;
1277 Put
(Output
, "Inserting """ & Value
& """");
1281 for J
in Value
'Range loop
1282 pragma Assert
(Value
(J
) /= ASCII
.NUL
);
1287 WT
.Table
(NK
) := New_Word
(Value
);
1290 if Max_Key_Len
< Len
then
1294 if Min_Key_Len
= 0 or else Len
< Min_Key_Len
then
1303 procedure New_Line
(File
: File_Descriptor
) is
1305 if Write
(File
, EOL
'Address, 1) /= 1 then
1306 raise Program_Error
;
1314 function New_Word
(S
: String) return Word_Type
is
1316 return new String'(S);
1319 ------------------------------
1320 -- Parse_Position_Selection --
1321 ------------------------------
1323 procedure Parse_Position_Selection (Argument : String) is
1324 N : Natural := Argument'First;
1325 L : constant Natural := Argument'Last;
1326 M : constant Natural := Max_Key_Len;
1328 T : array (1 .. M) of Boolean := (others => False);
1330 function Parse_Index return Natural;
1331 -- Parse argument starting at index N to find an index
1337 function Parse_Index return Natural is
1338 C : Character := Argument (N);
1347 if C not in '0' .. '9' then
1348 raise Program_Error with "cannot read position argument";
1351 while C in '0' .. '9' loop
1352 V := V * 10 + (Character'Pos (C) - Character'Pos ('0'));
1361 -- Start of processing for Parse_Position_Selection
1364 -- Empty specification means all the positions
1367 Char_Pos_Set_Len := M;
1368 Char_Pos_Set := Allocate (Char_Pos_Set_Len);
1370 for C in 0 .. Char_Pos_Set_Len - 1 loop
1371 Set_Char_Pos (C, C + 1);
1377 First, Last : Natural;
1380 First := Parse_Index;
1385 if N <= L and then Argument (N) = '-' then
1387 Last := Parse_Index;
1390 -- Include the positions in the selection
1392 for J in First .. Last loop
1399 if Argument (N) /= ',' then
1400 raise Program_Error with "cannot read position argument";
1406 -- Compute position selection length
1409 for J in T'Range loop
1415 -- Fill position selection
1417 Char_Pos_Set_Len := N;
1418 Char_Pos_Set := Allocate (Char_Pos_Set_Len);
1421 for J in T'Range loop
1423 Set_Char_Pos (N, J);
1428 end Parse_Position_Selection;
1435 (Pkg_Name : String := Default_Pkg_Name;
1436 Use_Stdout : Boolean := False)
1438 File : File_Descriptor := Standout;
1441 -- For call to Close
1443 function Array_Img (N, T, R1 : String; R2 : String := "") return String;
1444 -- Return string "N : constant array (R1[, R2]) of T;"
1446 function Range_Img (F, L : Natural; T : String := "") return String;
1447 -- Return string "[T range ]F .. L"
1449 function Type_Img (L : Natural) return String;
1450 -- Return the larger unsigned type T such that T'Last < L
1458 R2 : String := "") return String
1464 Add (" : constant array (");
1475 return Line (1 .. Last);
1482 function Range_Img (F, L : Natural; T : String := "") return String is
1483 FI : constant String := Image (F);
1484 FL : constant Natural := FI'Length;
1485 LI : constant String := Image (L);
1486 LL : constant Natural := LI'Length;
1487 TL : constant Natural := T'Length;
1488 RI : String (1 .. TL + 7 + FL + 4 + LL);
1493 RI (Len + 1 .. Len + TL) := T;
1495 RI (Len + 1 .. Len + 7) := " range ";
1499 RI (Len + 1 .. Len + FL) := FI;
1501 RI (Len + 1 .. Len + 4) := " .. ";
1503 RI (Len + 1 .. Len + LL) := LI;
1505 return RI (1 .. Len);
1512 function Type_Img (L : Natural) return String is
1513 S : constant String := Image (Type_Size (L));
1514 U : String := "Unsigned_ ";
1518 for J in S'Range loop
1530 FName : String := Ada_File_Base_Name (Pkg_Name) & ".ads";
1531 -- Initially, the name of the spec file, then modified to be the name of
1532 -- the body file. Not used if Use_Stdout is True.
1534 -- Start of processing for Produce
1538 if Verbose and then not Use_Stdout then
1540 "Producing " & Ada.Directories.Current_Directory & "/" & FName);
1544 if not Use_Stdout then
1545 File := Create_File (FName, Binary);
1547 if File = Invalid_FD then
1548 raise Program_Error with "cannot create: " & FName;
1552 Put (File, "package ");
1553 Put (File, Pkg_Name);
1556 Put (File, " function Hash (S : String) return Natural;");
1559 Put (File, Pkg_Name);
1563 if not Use_Stdout then
1564 Close (File, Status);
1571 if not Use_Stdout then
1573 -- Set to body file name
1575 FName (FName'Last) := 'b
';
1577 File := Create_File (FName, Binary);
1579 if File = Invalid_FD then
1580 raise Program_Error with "cannot create: " & FName;
1584 Put (File, "with Interfaces; use Interfaces;");
1587 Put (File, "package body ");
1588 Put (File, Pkg_Name);
1593 if Opt = CPU_Time then
1594 Put (File, Array_Img ("C", Type_Img (256), "Character"));
1597 F := Character'Pos (Character'First);
1598 L := Character'Pos (Character'Last);
1600 for J in Character'Range loop
1601 P := Get_Used_Char (J);
1602 Put (File, Image (P), 1, 0, 1, F, L, Character'Pos (J));
1609 L := Char_Pos_Set_Len - 1;
1611 Put (File, Array_Img ("P", "Natural", Range_Img (F, L)));
1614 for J in F .. L loop
1615 Put (File, Image (Get_Char_Pos (J)), 1, 0, 1, F, L, J);
1624 Array_Img ("T1", Type_Img (NV),
1625 Range_Img (0, T1_Len - 1),
1626 Range_Img (0, T2_Len - 1, Type_Img (256))),
1627 T1, T1_Len, T2_Len);
1629 when Memory_Space =>
1632 Array_Img ("T1", Type_Img (NV),
1633 Range_Img (0, T1_Len - 1)),
1643 Array_Img ("T2", Type_Img (NV),
1644 Range_Img (0, T1_Len - 1),
1645 Range_Img (0, T2_Len - 1, Type_Img (256))),
1646 T2, T1_Len, T2_Len);
1648 when Memory_Space =>
1651 Array_Img ("T2", Type_Img (NV),
1652 Range_Img (0, T1_Len - 1)),
1660 Array_Img ("G", Type_Img (NK),
1661 Range_Img (0, G_Len - 1)),
1665 Put (File, " function Hash (S : String) return Natural is");
1667 Put (File, " F : constant Natural := S'First - 1;");
1669 Put (File, " L : constant Natural := S'Length;");
1671 Put (File, " F1, F2 : Natural := 0;");
1674 Put (File, " J : ");
1678 Put (File, Type_Img (256));
1679 when Memory_Space =>
1680 Put (File, "Natural");
1686 Put (File, " begin");
1688 Put (File, " for K in P'Range loop");
1690 Put (File, " exit when L < P (K);");
1692 Put (File, " J := ");
1697 when Memory_Space =>
1698 Put (File, "Character'Pos");
1701 Put (File, " (S (P (K) + F));");
1704 Put (File, " F1 := (F1 + Natural (T1 (K");
1706 if Opt = CPU_Time then
1712 if Opt = Memory_Space then
1716 Put (File, ") mod ");
1717 Put (File, Image (NV));
1721 Put (File, " F2 := (F2 + Natural (T2 (K");
1723 if Opt = CPU_Time then
1729 if Opt = Memory_Space then
1733 Put (File, ") mod ");
1734 Put (File, Image (NV));
1738 Put (File, " end loop;");
1742 " return (Natural (G (F1)) + Natural (G (F2))) mod ");
1744 Put (File, Image (NK));
1747 Put (File, " end Hash;");
1751 Put (File, Pkg_Name);
1755 if not Use_Stdout then
1756 Close (File, Status);
1768 procedure Put (File : File_Descriptor; Str : String) is
1769 Len : constant Natural := Str'Length;
1771 for J in Str'Range loop
1772 pragma Assert (Str (J) /= ASCII.NUL);
1776 if Write (File, Str'Address, Len) /= Len then
1777 raise Program_Error;
1786 (F : File_Descriptor;
1795 Len : constant Natural := S'Length;
1798 -- Write current line, followed by LF
1806 Put (F, Line (1 .. Last));
1811 -- Start of processing for Put
1814 if C1 = F1 and then C2 = F2 then
1818 if Last + Len + 3 >= Max then
1826 if C1 = F1 and then C2 = F2 then
1878 procedure Put_Edges (File : File_Descriptor; Title : String) is
1880 F1 : constant Natural := 1;
1881 L1 : constant Natural := Edges_Len - 1;
1882 M : constant Natural := Max / 5;
1888 -- Edges valid range is 1 .. Edge_Len - 1
1890 for J in F1 .. L1 loop
1892 Put (File, Image (J, M), F1, L1, J, 1, 4, 1);
1893 Put (File, Image (E.X, M), F1, L1, J, 1, 4, 2);
1894 Put (File, Image (E.Y, M), F1, L1, J, 1, 4, 3);
1895 Put (File, Image (E.Key, M), F1, L1, J, 1, 4, 4);
1899 ----------------------
1900 -- Put_Initial_Keys --
1901 ----------------------
1903 procedure Put_Initial_Keys (File : File_Descriptor; Title : String) is
1904 F1 : constant Natural := 0;
1905 L1 : constant Natural := NK - 1;
1906 M : constant Natural := Max / 5;
1913 for J in F1 .. L1 loop
1915 Put (File, Image (J, M), F1, L1, J, 1, 3, 1);
1916 Put (File, Image (K.Edge, M), F1, L1, J, 1, 3, 2);
1917 Put (File, Trim_Trailing_Nuls (WT.Table (Initial (J)).all),
1918 F1, L1, J, 1, 3, 3);
1920 end Put_Initial_Keys;
1922 --------------------
1923 -- Put_Int_Matrix --
1924 --------------------
1926 procedure Put_Int_Matrix
1927 (File : File_Descriptor;
1933 F1 : constant Integer := 0;
1934 L1 : constant Integer := Len_1 - 1;
1935 F2 : constant Integer := 0;
1936 L2 : constant Integer := Len_2 - 1;
1944 for J in F1 .. L1 loop
1945 Ix := IT.Table (Table + J);
1946 Put (File, Image (Ix), 1, 0, 1, F1, L1, J);
1950 for J in F1 .. L1 loop
1951 for K in F2 .. L2 loop
1952 Ix := IT.Table (Table + J + K * Len_1);
1953 Put (File, Image (Ix), F1, L1, J, F2, L2, K);
1959 --------------------
1960 -- Put_Int_Vector --
1961 --------------------
1963 procedure Put_Int_Vector
1964 (File : File_Descriptor;
1969 F2 : constant Natural := 0;
1970 L2 : constant Natural := Length - 1;
1976 for J in F2 .. L2 loop
1977 Put (File, Image (IT.Table (Vector + J)), 1, 0, 1, F2, L2, J);
1981 ----------------------
1982 -- Put_Reduced_Keys --
1983 ----------------------
1985 procedure Put_Reduced_Keys (File : File_Descriptor; Title : String) is
1986 F1 : constant Natural := 0;
1987 L1 : constant Natural := NK - 1;
1988 M : constant Natural := Max / 5;
1995 for J in F1 .. L1 loop
1997 Put (File, Image (J, M), F1, L1, J, 1, 3, 1);
1998 Put (File, Image (K.Edge, M), F1, L1, J, 1, 3, 2);
1999 Put (File, Trim_Trailing_Nuls (WT.Table (Reduced (J)).all),
2000 F1, L1, J, 1, 3, 3);
2002 end Put_Reduced_Keys;
2004 -----------------------
2005 -- Put_Used_Char_Set --
2006 -----------------------
2008 procedure Put_Used_Char_Set (File : File_Descriptor; Title : String) is
2009 F : constant Natural := Character'Pos (Character'First);
2010 L : constant Natural := Character'Pos (Character'Last);
2016 for J in Character'Range loop
2018 (File, Image (Get_Used_Char (J)), 1, 0, 1, F, L, Character'Pos (J));
2020 end Put_Used_Char_Set;
2022 ----------------------
2023 -- Put_Vertex_Table --
2024 ----------------------
2026 procedure Put_Vertex_Table (File : File_Descriptor; Title : String) is
2027 F1 : constant Natural := 0;
2028 L1 : constant Natural := NV - 1;
2029 M : constant Natural := Max / 4;
2036 for J in F1 .. L1 loop
2037 V := Get_Vertices (J);
2038 Put (File, Image (J, M), F1, L1, J, 1, 3, 1);
2039 Put (File, Image (V.First, M), F1, L1, J, 1, 3, 2);
2040 Put (File, Image (V.Last, M), F1, L1, J, 1, 3, 3);
2042 end Put_Vertex_Table;
2048 procedure Random (Seed : in out Natural) is
2050 -- Park & Miller Standard Minimal using Schrage's algorithm to avoid
2051 -- overflow: Xn+1 = 16807 * Xn mod (2 ** 31 - 1)
2058 R := Seed mod 127773;
2060 X := 16807 * R - 2836 * Q;
2062 Seed := (if X < 0 then X + 2147483647 else X);
2069 function Reduced (K : Key_Id) return Word_Id is
2078 procedure Resize_Word (W : in out Word_Type; Len : Natural) is
2079 S1 : constant String := W.all;
2080 S2 : String (1 .. Len) := (others => ASCII.NUL);
2081 L : constant Natural := S1'Length;
2090 --------------------------
2091 -- Select_Char_Position --
2092 --------------------------
2094 procedure Select_Char_Position is
2096 type Vertex_Table_Type is array (Natural range <>) of Vertex_Type;
2098 procedure Build_Identical_Keys_Sets
2099 (Table : in out Vertex_Table_Type;
2100 Last : in out Natural;
2102 -- Build a list of keys subsets that are identical with the current
2103 -- position selection plus Pos. Once this routine is called, reduced
2104 -- words are sorted by subsets and each item (First, Last) in Sets
2105 -- defines the range of identical keys.
2106 -- Need comment saying exactly what Last is ???
2108 function Count_Different_Keys
2109 (Table : Vertex_Table_Type;
2111 Pos : Natural) return Natural;
2112 -- For each subset in Sets, count the number of different keys if we add
2113 -- Pos to the current position selection.
2115 Sel_Position : IT.Table_Type (1 .. Max_Key_Len);
2116 Last_Sel_Pos : Natural := 0;
2117 Max_Sel_Pos : Natural := 0;
2119 -------------------------------
2120 -- Build_Identical_Keys_Sets --
2121 -------------------------------
2123 procedure Build_Identical_Keys_Sets
2124 (Table : in out Vertex_Table_Type;
2125 Last : in out Natural;
2128 S : constant Vertex_Table_Type := Table (Table'First .. Last);
2129 C : constant Natural := Pos;
2130 -- Shortcuts (why are these not renames ???)
2134 -- First and last words of a subset
2137 -- GNAT.Heap_Sort assumes that the first array index is 1. Offset
2138 -- defines the translation to operate.
2140 function Lt (L, R : Natural) return Boolean;
2141 procedure Move (From : Natural; To : Natural);
2142 -- Subprograms needed by GNAT.Heap_Sort_G
2148 function Lt (L, R : Natural) return Boolean is
2149 C : constant Natural := Pos;
2156 Right := Offset + R;
2162 Right := Offset + R;
2165 return WT.Table (Left)(C) < WT.Table (Right)(C);
2172 procedure Move (From : Natural; To : Natural) is
2173 Target, Source : Natural;
2178 Target := Offset + To;
2180 Source := Offset + From;
2183 Source := Offset + From;
2184 Target := Offset + To;
2187 WT.Table (Target) := WT.Table (Source);
2188 WT.Table (Source) := null;
2191 package Sorting is new GNAT.Heap_Sort_G (Move, Lt);
2193 -- Start of processing for Build_Identical_Key_Sets
2198 -- For each subset in S, extract the new subsets we have by adding C
2199 -- in the position selection.
2201 for J in S'Range loop
2202 if S (J).First = S (J).Last then
2206 Table (Last) := (F, L);
2209 Offset := Reduced (S (J).First) - 1;
2210 Sorting.Sort (S (J).Last - S (J).First + 1);
2214 for N in S (J).First .. S (J).Last loop
2216 -- For the last item, close the last subset
2218 if N = S (J).Last then
2220 Table (Last) := (F, N);
2222 -- Two contiguous words are identical when they have the
2223 -- same Cth character.
2225 elsif WT.Table (Reduced (N))(C) =
2226 WT.Table (Reduced (N + 1))(C)
2230 -- Find a new subset of identical keys. Store the current
2231 -- one and create a new subset.
2235 Table (Last) := (F, L);
2242 end Build_Identical_Keys_Sets;
2244 --------------------------
2245 -- Count_Different_Keys --
2246 --------------------------
2248 function Count_Different_Keys
2249 (Table : Vertex_Table_Type;
2251 Pos : Natural) return Natural
2253 N : array (Character) of Natural;
2258 -- For each subset, count the number of words that are still
2259 -- different when we include Pos in the position selection. Only
2260 -- focus on this position as the other positions already produce
2263 for S in 1 .. Last loop
2265 -- Count the occurrences of the different characters
2268 for K in Table (S).First .. Table (S).Last loop
2269 C := WT.Table (Reduced (K))(Pos);
2273 -- Update the number of different keys. Each character used
2274 -- denotes a different key.
2276 for J in N'Range loop
2284 end Count_Different_Keys;
2286 -- Start of processing for Select_Char_Position
2289 -- Initialize the reduced words set
2291 for K in 0 .. NK - 1 loop
2292 WT.Table (Reduced (K)) := New_Word (WT.Table (Initial (K)).all);
2296 Differences : Natural;
2297 Max_Differences : Natural := 0;
2298 Old_Differences : Natural;
2299 Max_Diff_Sel_Pos : Natural := 0; -- init to kill warning
2300 Max_Diff_Sel_Pos_Idx : Natural := 0; -- init to kill warning
2301 Same_Keys_Sets_Table : Vertex_Table_Type (1 .. NK);
2302 Same_Keys_Sets_Last : Natural := 1;
2305 for C in Sel_Position'Range loop
2306 Sel_Position (C) := C;
2309 Same_Keys_Sets_Table (1) := (0, NK - 1);
2312 -- Preserve maximum number of different keys and check later on
2313 -- that this value is strictly incrementing. Otherwise, it means
2314 -- that two keys are strictly identical.
2316 Old_Differences := Max_Differences;
2318 -- The first position should not exceed the minimum key length.
2319 -- Otherwise, we may end up with an empty word once reduced.
2322 (if Last_Sel_Pos = 0 then Min_Key_Len else Max_Key_Len);
2324 -- Find which position increases more the number of differences
2326 for J in Last_Sel_Pos + 1 .. Max_Sel_Pos loop
2327 Differences := Count_Different_Keys
2328 (Same_Keys_Sets_Table,
2329 Same_Keys_Sets_Last,
2334 "Selecting position" & Sel_Position (J)'Img &
2335 " results in" & Differences'Img &
2340 if Differences > Max_Differences then
2341 Max_Differences := Differences;
2342 Max_Diff_Sel_Pos := Sel_Position (J);
2343 Max_Diff_Sel_Pos_Idx := J;
2347 if Old_Differences = Max_Differences then
2348 raise Program_Error with "some keys are identical";
2351 -- Insert selected position and sort Sel_Position table
2353 Last_Sel_Pos := Last_Sel_Pos + 1;
2354 Sel_Position (Last_Sel_Pos + 1 .. Max_Diff_Sel_Pos_Idx) :=
2355 Sel_Position (Last_Sel_Pos .. Max_Diff_Sel_Pos_Idx - 1);
2356 Sel_Position (Last_Sel_Pos) := Max_Diff_Sel_Pos;
2358 for P in 1 .. Last_Sel_Pos - 1 loop
2359 if Max_Diff_Sel_Pos < Sel_Position (P) then
2360 Sel_Position (P + 1 .. Last_Sel_Pos) :=
2361 Sel_Position (P .. Last_Sel_Pos - 1);
2362 Sel_Position (P) := Max_Diff_Sel_Pos;
2367 exit when Max_Differences = NK;
2369 Build_Identical_Keys_Sets
2370 (Same_Keys_Sets_Table,
2371 Same_Keys_Sets_Last,
2376 "Selecting position" & Max_Diff_Sel_Pos'Img &
2377 " results in" & Max_Differences'Img &
2382 for J in 1 .. Same_Keys_Sets_Last loop
2384 Same_Keys_Sets_Table (J).First ..
2385 Same_Keys_Sets_Table (J).Last
2388 Trim_Trailing_Nuls (WT.Table (Reduced (K)).all));
2398 Char_Pos_Set_Len := Last_Sel_Pos;
2399 Char_Pos_Set := Allocate (Char_Pos_Set_Len);
2401 for C in 1 .. Last_Sel_Pos loop
2402 Set_Char_Pos (C - 1, Sel_Position (C));
2404 end Select_Char_Position;
2406 --------------------------
2407 -- Select_Character_Set --
2408 --------------------------
2410 procedure Select_Character_Set is
2411 Last : Natural := 0;
2412 Used : array (Character) of Boolean := (others => False);
2416 for J in 0 .. NK - 1 loop
2417 for K in 0 .. Char_Pos_Set_Len - 1 loop
2418 Char := WT.Table (Initial (J))(Get_Char_Pos (K));
2419 exit when Char = ASCII.NUL;
2420 Used (Char) := True;
2424 Used_Char_Set_Len := 256;
2425 Used_Char_Set := Allocate (Used_Char_Set_Len);
2427 for J in Used'Range loop
2429 Set_Used_Char (J, Last);
2432 Set_Used_Char (J, 0);
2435 end Select_Character_Set;
2441 procedure Set_Char_Pos (P : Natural; Item : Natural) is
2442 N : constant Natural := Char_Pos_Set + P;
2444 IT.Table (N) := Item;
2451 procedure Set_Edges (F : Natural; Item : Edge_Type) is
2452 N : constant Natural := Edges + (F * Edge_Size);
2454 IT.Table (N) := Item.X;
2455 IT.Table (N + 1) := Item.Y;
2456 IT.Table (N + 2) := Item.Key;
2463 procedure Set_Graph (N : Natural; Item : Integer) is
2465 IT.Table (G + N) := Item;
2472 procedure Set_Key (N : Key_Id; Item : Key_Type) is
2474 IT.Table (Keys + N) := Item.Edge;
2481 procedure Set_Table (T : Integer; X, Y : Natural; Item : Natural) is
2482 N : constant Natural := T + ((Y * T1_Len) + X);
2484 IT.Table (N) := Item;
2491 procedure Set_Used_Char (C : Character; Item : Natural) is
2492 N : constant Natural := Used_Char_Set + Character'Pos (C);
2494 IT.Table (N) := Item;
2501 procedure Set_Vertices (F : Natural; Item : Vertex_Type) is
2502 N : constant Natural := Vertices + (F * Vertex_Size);
2504 IT.Table (N) := Item.First;
2505 IT.Table (N + 1) := Item.Last;
2515 Opt : Optimization) return Natural
2523 for J in 0 .. T1_Len - 1 loop
2524 exit when Word (J + 1) = ASCII.NUL;
2525 R := Get_Table (Table, J, Get_Used_Char (Word (J + 1)));
2526 S := (S + R) mod NV;
2529 when Memory_Space =>
2530 for J in 0 .. T1_Len - 1 loop
2531 exit when Word (J + 1) = ASCII.NUL;
2532 R := Get_Table (Table, J, 0);
2533 S := (S + R * Character'Pos (Word (J + 1))) mod NV;
2540 ------------------------
2541 -- Trim_Trailing_Nuls --
2542 ------------------------
2544 function Trim_Trailing_Nuls (Str : String) return String is
2546 for J in reverse Str'Range loop
2547 if Str (J) /= ASCII.NUL then
2548 return Str (Str'First .. J);
2553 end Trim_Trailing_Nuls;
2559 function Type_Size (L : Natural) return Natural is
2563 elsif L <= 2 ** 16 then
2577 K : Natural := 0) return Natural
2581 when Character_Position =>
2582 return Get_Char_Pos (J);
2584 when Used_Character_Set =>
2585 return Get_Used_Char (Character'Val (J));
2587 when Function_Table_1 =>
2588 return Get_Table (T1, J, K);
2590 when Function_Table_2 =>
2591 return Get_Table (T2, J, K);
2594 return Get_Graph (J);
2599 end GNAT.Perfect_Hash_Generators;