.\"/*
.\" * Copyright (c) 2005-2018, NVIDIA CORPORATION.  All rights reserved.
.\" *
.\" * Licensed under the Apache License, Version 2.0 (the "License");
.\" * you may not use this file except in compliance with the License.
.\" * You may obtain a copy of the License at
.\" *
.\" *     http://www.apache.org/licenses/LICENSE-2.0
.\" *
.\" * Unless required by applicable law or agreed to in writing, software
.\" * distributed under the License is distributed on an "AS IS" BASIS,
.\" * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
.\" * See the License for the specific language governing permissions and
.\" * limitations under the License.
.\" *
.\" */
.NS 14 "Data Initialization File"
.lp
The Data Initialization File is an external binary file defining the 
initialization of external and static variables.  It is written by 
the Semantic Analyzer as it processes initializers for external 
and static variables, and is read by the Assembler in order to
write the necessary \*(cfTXT\*(rf blocks and/or assembler directives
to initialize the data.
.lp
The file consists of one or more
.ul
records.
Each record consists of 2 fields, a \*(cfshort int\*(rf (dtype)
and a 32-bit \*(cfINT\*(rf (conval).
Records are written to, and read from the file via the routines
\*(mfdinit_put\*(rf and \*(mfdinit_read\*(rf in the module 
\*(ffdinitutil.c\*(rf.  These routines are also responsible 
for opening and possibly buffering the file.
.lp
There are several types of record in the file,
distinguished by the value of the first field (dtype):
.IP "dtype = DINIT_LOC" CW
.br
\*(cfDINIT_LOC\*(rf
is a special value chosen so as to not conflict
with any of the other possible values of this field.
\*(cfconval\*(rf, the second field of the record, will be a
symbol table pointer to a variable, array, pointer, or structure.
This record indicates that a new symbol is being initialized.
It results in a \*(cf$LOC\*(rf directive being written to 
the Object Code Listing.  Ensuing records, up to the next 
\*(cfDINIT_LOC\*(rf, will initialize consecutive locations 
within this variable.
.IP "dtype = DINIT_OFFSET" CW
.cw "(FORTRAN)" .
The value in \*(cfconval\*(rf is added to the base address given by the
\*(cfDINIT_LOC\*(rf record to determine the effective address for data
initialization.  This is useful for Fortran style structures where fields
in a structure can be skipped and individual array elements can be
initialized.
.IP "dtype = DINIT_PAD" CW
.br
This record indicates that the next \*(cfconval\*(rf bytes are to 
be zerofilled for padding.
.IP "dtype = DINIT_ZEROES" CW
.br
This record indicates that the next \*(cfconval\*(rf bytes are to 
be zerofilled.
.IP "dtype = DINIT_REPEAT" CW
.br
.cw "(FORTRAN)" .
The value of \*(cfconval\*(rf specifies the repeat count for the constant
that is specified in the following dinit record.
.IP "dtype = DINIT_LEN" CW
.br
.cw "(C)" .
This record follows the record initializing an array of char with a string
and is used to provide extra information for the initialization.
Its conval field indicates the number of characters to use from the string.
This value will either be the length of the string with the null terminating
character or the length of the string without the null terminating character.
.IP "dtype = DT_DBLE" CW
.br
Indicates that 2 32-bit words are to be initialized with
a 64-bit double precision floating point value.
\*(cfconval\*(rf is a symbol table pointer to a double precision constant.
.IP "dtype = DT_CMPLX" CW
.br
.cw "(FORTRAN)" .
Indicates that 2 32-bit words are to be initialized with
two 32-bit floating point values.
\*(cfconval\*(rf is a symbol table pointer to a
.cw complex
constant.
.IP "dtype = DT_DCMPLX" CW
.br
.cw "(FORTRAN)" .
Indicates that 2 64-bit words are to be initialized with
two 64-bit double precision floating point values.
\*(cfconval\*(rf is a symbol table pointer to a
.cw doublecomplex
constant.
.IP "dtype = pointer to x, conval = address constant" CW
.br
Indicates that the next 32-bit word is to be initialized
to a variable address plus an optional byte offset.
\*(cfconval\*(rf is a symbol table pointer to an address constant.
.IP "dtype = pointer to x, conval = string" CW
.br
.cw "(C)" .
Indicates that a sequence of bytes is to be initialized
to the characters in the string, possibly including the terminating
null character.  This record immediately precedes a record of type 
.cw DINIT_LEN
which indicates the number of characters to use from string
for the initialization.
\*(cfconval\*(rf is a symbol table pointer to a string.
.IP "dtype = other C types" CW
.br
Includes the
.i integral
and single precision data types.
Indicates that the current byte (e.g. \*(cfDT_CHAR\*(rf),
halfword (e.g. \*(cfDT_SINT\*(rf),
or word (e.g. \*(cfDT_FLOAT\*(rf),
is to be initialized
to the value in conval.
The Assembler must perform word or halfword 
alignment if necessary.
.IP "dtype = Fortran Hollerith or CHARACTER" CW
.br
.cw dtype
is actually \*(cfDT_CHAR\*(rf which indicates that
a sequence of bytes is to be initialized to the
characters in the
.cw Hollerith
or
.cw CHARACTER
string.
.cw conval
is a symbol table pointer to the
.i string
constant.
The length of the initialization is derived from the
.i string "'s"
.cw DTYPE ,
not
.cw dtype .
.IP "dtype = other Fortran types" CW
.br
Includes the integer, logical, and single precision data
types.
Indicates that the current byte (e.g. \*(cfDT_BINT\*(rf),
halfword (e.g. \*(cfDT_SINT\*(rf),
or word (e.g. \*(cfDT_FLOAT\*(rf),
is to be initialized
to the value in conval.
The Assembler must perform word or halfword 
alignment if necessary.
.IP "dtype = DINIT_LABEL" CW
.br
Indicates that the next 64-bit word is to be initialized
to the address of a label.  \*(cfconval\*(rf is a symbol 
table pointer to a label.
.IP "dtype = 0 - alignment record" CW
.br
Indicates that byte, halfword, or word alignment is required.
The value of conval will be 0, 1, 3, or 7 respectively.
Note that in the usual case (e.g. a char followed by an int),
alignment is automatic and this record is not written.
This record is only required for certain cases where a struct
is a member of another struct or is an array element.
For example, word alignment is required between the initialization
record for c and the one for d in the following code:
.CS

    struct {
        char c;
        struct {
            char d;
            int e;
        } f;
    } x = { 'c', {'d', 77}};
.CE
.IP "dtype = DINIT_FIELD" CW
.br
.cw "(C)" .
This record is only used for certain implementations where bit fields
are initialized in a special way.
Normally, bit fields are initialized using "dtype" dinit records
whose \*(cfconval\*(rf fields have been filled by the data initalizing
process during semantic analysis.
If used, this record always precedes the records which initialize a
sequence of fields.