1------------------------------------------------------------------------------ 2-- -- 3-- GNAT COMPILER COMPONENTS -- 4-- -- 5-- E X P _ P A K D -- 6-- -- 7-- S p e c -- 8-- -- 9-- Copyright (C) 1992-2001 Free Software Foundation, Inc. -- 10-- -- 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, 59 Temple Place - Suite 330, Boston, -- 20-- MA 02111-1307, USA. -- 21-- -- 22-- GNAT was originally developed by the GNAT team at New York University. -- 23-- Extensive contributions were provided by Ada Core Technologies Inc. -- 24-- -- 25------------------------------------------------------------------------------ 26 27-- Expand routines for manipulation of packed arrays 28 29with Types; use Types; 30 31package Exp_Pakd is 32 33 ------------------------------------- 34 -- Implementation of Packed Arrays -- 35 ------------------------------------- 36 37 -- When a packed array (sub)type is frozen, we create a corresponding 38 -- type that will be used to hold the bits of the packed value, and 39 -- store the entity for this type in the Packed_Array_Type field of the 40 -- E_Array_Type or E_Array_Subtype entity for the packed array. 41 42 -- This packed array type has the name xxxPn, where xxx is the name 43 -- of the packed type, and n is the component size. The expanded 44 -- declaration declares a type that is one of the following: 45 46 -- For an unconstrained array with component size 1,2,4 or any other 47 -- odd component size. These are the cases in which we do not need 48 -- to align the underlying array. 49 50 -- type xxxPn is new Packed_Bytes1; 51 52 -- For an unconstrained array with component size that is divisible 53 -- by 2, but not divisible by 4 (other than 2 itself). These are the 54 -- cases in which we can generate better code if the underlying array 55 -- is 2-byte aligned (see System.Pack_14 in file s-pack14 for example). 56 57 -- type xxxPn is new Packed_Bytes2; 58 59 -- For an unconstrained array with component size that is divisible 60 -- by 4, other than powers of 2 (which either come under the 1,2,4 61 -- exception above, or are not packed at all). These are cases where 62 -- we can generate better code if the underlying array is 4-byte 63 -- aligned (see System.Pack_20 in file s-pack20 for example). 64 65 -- type xxxPn is new Packed_Bytes4; 66 67 -- For a constrained array with a static index type where the number 68 -- of bits does not exceed the size of Unsigned: 69 70 -- type xxxPn is new Unsigned range 0 .. 2 ** nbits - 1; 71 72 -- For a constrained array with a static index type where the number 73 -- of bits is greater than the size of Unsigned, but does not exceed 74 -- the size of Long_Long_Unsigned: 75 76 -- type xxxPn is new Long_Long_Unsigned range 0 .. 2 ** nbits - 1; 77 78 -- For all other constrained arrays, we use one of 79 80 -- type xxxPn is new Packed_Bytes1 (0 .. m); 81 -- type xxxPn is new Packed_Bytes2 (0 .. m); 82 -- type xxxPn is new Packed_Bytes4 (0 .. m); 83 84 -- where m is calculated (from the length of the original packed array) 85 -- to hold the required number of bits, and the choice of the particular 86 -- Packed_Bytes{1,2,4} type is made on the basis of alignment needs as 87 -- described above for the unconstrained case. 88 89 -- When a variable of packed array type is allocated, gigi will allocate 90 -- the amount of space indicated by the corresponding packed array type. 91 -- However, we do NOT attempt to rewrite the types of any references or 92 -- to retype the variable itself, since this would cause all kinds of 93 -- semantic problems in the front end (remember that expansion proceeds 94 -- at the same time as analysis). 95 96 -- For an indexed reference to a packed array, we simply convert the 97 -- reference to the appropriate equivalent reference to the object 98 -- of the packed array type (using unchecked conversion). 99 100 -- In some cases (for internally generated types, and for the subtypes 101 -- for record fields that depend on a discriminant), the corresponding 102 -- packed type cannot be easily generated in advance. In these cases, 103 -- we generate the required subtype on the fly at the reference point. 104 105 -- For the modular case, any unused bits are initialized to zero, and 106 -- all operations maintain these bits as zero (where necessary all 107 -- unchecked conversions from corresponding array values require 108 -- these bits to be clear, which is done automatically by gigi). 109 110 -- For the array cases, there can be unused bits in the last byte, and 111 -- these are neither initialized, nor treated specially in operations 112 -- (i.e. it is allowable for these bits to be clobbered, e.g. by not). 113 114 --------------------------- 115 -- Endian Considerations -- 116 --------------------------- 117 118 -- The standard does not specify the way in which bits are numbered in 119 -- a packed array. There are two reasonable rules for deciding this: 120 121 -- Store the first bit at right end (low order) word. This means 122 -- that the scaled subscript can be used directly as a right shift 123 -- count (if we put bit 0 at the left end, then we need an extra 124 -- subtract to compute the shift count. 125 126 -- Layout the bits so that if the packed boolean array is overlaid on 127 -- a record, using unchecked conversion, then bit 0 of the array is 128 -- the same as the bit numbered bit 0 in a record representation 129 -- clause applying to the record. For example: 130 131 -- type Rec is record 132 -- C : Bits4; 133 -- D : Bits7; 134 -- E : Bits5; 135 -- end record; 136 137 -- for Rec use record 138 -- C at 0 range 0 .. 3; 139 -- D at 0 range 4 .. 10; 140 -- E at 0 range 11 .. 15; 141 -- end record; 142 143 -- type P16 is array (0 .. 15) of Boolean; 144 -- pragma Pack (P16); 145 146 -- Now if we use unchecked conversion to convert a value of the record 147 -- type to the packed array type, according to this second criterion, 148 -- we would expect field D to occupy bits 4..10 of the Boolean array. 149 150 -- Although not required, this correspondence seems a highly desirable 151 -- property, and is one that GNAT decides to guarantee. For a little 152 -- endian machine, we can also meet the first requirement, but for a 153 -- big endian machine, it will be necessary to store the first bit of 154 -- a Boolean array in the left end (most significant) bit of the word. 155 -- This may cost an extra instruction on some machines, but we consider 156 -- that a worthwhile price to pay for the consistency. 157 158 -- One more important point arises in the case where we have a constrained 159 -- subtype of an unconstrained array. Take the case of 20-bits. For the 160 -- unconstrained representation, we would use an array of bytes: 161 162 -- Little-endian case 163 -- 8-7-6-5-4-3-2-1 16-15-14-13-12-11-10-9 x-x-x-x-20-19-18-17 164 165 -- Big-endian case 166 -- 1-2-3-4-5-6-7-8 9-10-11-12-13-14-15-16 17-18-19-20-x-x-x-x 167 168 -- For the constrained case, we use a 20-bit modular value, but in 169 -- general this value may well be stored in 32 bits. Let's look at 170 -- what it looks like: 171 172 -- Little-endian case 173 174 -- x-x-x-x-x-x-x-x-x-x-x-x-20-19-18-17-...-10-9-8-7-6-5-4-3-2-1 175 176 -- which stored in memory looks like 177 178 -- 8-7-...-2-1 16-15-...-10-9 x-x-x-x-20-19-18-17 x-x-x-x-x-x-x 179 180 -- An important rule is that the constrained and unconstrained cases 181 -- must have the same bit representation in memory, since we will often 182 -- convert from one to the other (e.g. when calling a procedure whose 183 -- formal is unconstrained). As we see, that criterion is met for the 184 -- little-endian case above. Now let's look at the big-endian case: 185 186 -- Big-endian case 187 188 -- x-x-x-x-x-x-x-x-x-x-x-x-1-2-3-4-5-6-7-8-9-10-...-17-18-19-20 189 190 -- which stored in memory looks like 191 192 -- x-x-x-x-x-x-x-x x-x-x-x-1-2-3-4 5-6-...11-12 13-14-...-19-20 193 194 -- That won't do, the representation value in memory is NOT the same in 195 -- the constrained and unconstrained case. The solution is to store the 196 -- modular value left-justified: 197 198 -- 1-2-3-4-5-6-7-8-9-10-...-17-18-19-20-x-x-x-x-x-x-x-x-x-x-x 199 200 -- which stored in memory looks like 201 202 -- 1-2-...-7-8 9-10-...15-16 17-18-19-20-x-x-x-x x-x-x-x-x-x-x-x 203 204 -- and now, we do indeed have the same representation. The special flag 205 -- Is_Left_Justified_Modular is set in the modular type used as the 206 -- packed array type in the big-endian case to ensure that this required 207 -- left justification occurs. 208 209 ----------------- 210 -- Subprograms -- 211 ----------------- 212 213 procedure Create_Packed_Array_Type (Typ : Entity_Id); 214 -- Typ is a array type or subtype to which pragma Pack applies. If the 215 -- Packed_Array_Type field of Typ is already set, then the call has no 216 -- effect, otherwise a suitable type or subtype is created and stored 217 -- in the Packed_Array_Type field of Typ. This created type is an Itype 218 -- so that Gigi will simply elaborate and freeze the type on first use 219 -- (which is typically the definition of the corresponding array type). 220 -- 221 -- Note: although this routine is included in the expander package for 222 -- packed types, it is actually called unconditionally from Freeze, 223 -- whether or not expansion (and code generation) is enabled. We do this 224 -- since we want gigi to be able to properly compute type charactersitics 225 -- (for the Data Decomposition Annex of ASIS, and possible other future 226 -- uses) even if code generation is not active. Strictly this means that 227 -- this procedure is not part of the expander, but it seems appropriate 228 -- to keep it together with the other expansion routines that have to do 229 -- with packed array types. 230 231 procedure Expand_Packed_Boolean_Operator (N : Node_Id); 232 -- N is an N_Op_And, N_Op_Or or N_Op_Xor node whose operand type is a 233 -- packed boolean array. This routine expands the appropriate operations 234 -- to carry out the logical operation on the packed arrays. It handles 235 -- both the modular and array representation cases. 236 237 procedure Expand_Packed_Element_Reference (N : Node_Id); 238 -- N is an N_Indexed_Component node whose prefix is a packed array. In 239 -- the bit packed case, this routine can only be used for the expression 240 -- evaluation case not the assignment case, since the result is not a 241 -- variable. See Expand_Bit_Packed_Element_Set for how he assignment case 242 -- is handled in the bit packed case. For the enumeration case, the result 243 -- of this call is always a variable, so the call can be used for both the 244 -- expression evaluation and assignment cases. 245 246 procedure Expand_Bit_Packed_Element_Set (N : Node_Id); 247 -- N is an N_Assignment_Statement node whose name is an indexed 248 -- component of a bit-packed array. This procedure rewrites the entire 249 -- assignment statement with appropriate code to set the referenced 250 -- bits of the packed array type object. Note that this procedure is 251 -- used only for the bit-packed case, not for the enumeration case. 252 253 procedure Expand_Packed_Eq (N : Node_Id); 254 -- N is an N_Op_Eq node where the operands are packed arrays whose 255 -- representation is an array-of-bytes type (the case where a modular 256 -- type is used for the representation does not require any special 257 -- handling, because in the modular case, unused bits are zeroes. 258 259 procedure Expand_Packed_Not (N : Node_Id); 260 -- N is an N_Op_Not node where the operand is packed array of Boolean 261 -- in standard representation (i.e. component size is one bit). This 262 -- procedure expands the corresponding not operation. Note that the 263 -- non-standard representation case is handled by using a loop through 264 -- elements generated by the normal non-packed circuitry. 265 266 function Involves_Packed_Array_Reference (N : Node_Id) return Boolean; 267 -- N is the node for a name. This function returns true if the name 268 -- involves a packed array reference. A node involves a packed array 269 -- reference if it is itself an indexed compoment referring to a bit- 270 -- packed array, or it is a selected component whose prefix involves 271 -- a packed array reference. 272 273 procedure Expand_Packed_Address_Reference (N : Node_Id); 274 -- The node N is an attribute reference for the 'Address reference, where 275 -- the prefix involves a packed array reference. This routine expands the 276 -- necessary code for performing the address reference in this case. 277 278end Exp_Pakd; 279