
CONTENTS
--------
* Const data support
* Pointer Models
* Storing 14 bit data
* Data of size 16 bit or more
* Code Pages
* Merging data
* Examples



CONST DATA SUPPORT
------------------

CC5X supports constant data stored in program memory. The C keyword
'const' tells the compiler that these data do not change.

Examples:
    const char *ps = "Hello world!";
    const float ftx[] = { 1.0, 33.34, 1.3e-10 };
    ..
    t = *ps;
    ps = "";
    fx = ftx[i];

The implementation of constant data supports the following features:

    * both 8 and 16 bit pointers in the same application

    * the size of single pointers is normally automatically chosen

    * 'const' pointers can access both RAM and program memory

    * the compiler will not put all constant data in a single table,
      but rather make smaller tables if this saves code space

    * some devices supports 14 bits data (PIC16F87X). The compiler
      will automatically chose this format if space can be saved.
      This allows very compact storage of 7 bit ASCII strings.

    * duplicate strings and other data are automatically merged to
      save space

Command line options:

    -mc1 : default 'const' pointer size is 1 byte (8 bits)
    -mc2 : default 'const' pointer size is 2 bytes (16 bits)
    -mr1 : default RAM pointer size is 1 byte
    -mr2 : default RAM pointer size is 2 bytes
    -mm1 : default pointer size is 1 byte (all pointer types)
    -mm2 : default pointer size is 2 bytes (all pointer types)

That default pointer sizes are used only when the pointer size is
not chosen dynamically. The priority when deciding the pointer size
is:

    1) Pointer size type modifiers
    2) Automatic chosen pointer size (single pointers)
    3) Pointer size chosen according to the default model

Pointer size type modifiers:

    'size1': pointer size is 1 byte (8 bits)
    'size2': pointer size is 2 byte (16 bits)

The size type modifer must be put in front of the data type:

    bank1 size2 float *pf;


Recommendations:

   It is recommended to use small data tables and structures. This
   allows the compiler to merge equal data items and build optimal
   blocks of constant data.


Limitations:
   1) The compiler will not initialize RAM variables on startup
   2) Data items of 16 bit or more in structures with more than 256
      byte data must be aligned



POINTER MODELS
--------------

Using 8 bit pointers when possible saves both code and RAM space.
CC5X allows the size of all single pointers to be decided
automatically. However, pointers in structures and arrays have to be
decided in advance, by using the memory model command line options
or a size type modifier.

Note that the operator sizeof(pointer) will fix the size according
to the chosen default model. Using sizeof(pointer) is normally not
required and should be avoided.

Several different pointer types are supported. This is done
automatically and is transparent to the application program.

    a) 8 bit pointer to RAM. The compiler will automatically
       update the IRP bit if required (when RAM space exceeds 256
       bytes).

    b) 16 bit pointer to RAM. This format is required only
       when the same pointer have to access locations in different
       256 byte RAM segments.

    c) 8 bit pointer to program memory. This pointer can access
       up to 256 byte data.

    d) 16 bit pointer to program memory. This pointer can access
       more than 256 byte data.

    e) 8 bit pointer to RAM or program memory. This pointer can
       access up to 128 byte data and 128 byte RAM. Bit 7 is used to
       detect RAM or program memory access. The compiler will only
       chose this format if all RAM addresses loaded to the pointer
       is in the same bank (14 bit core).

    f) 16 bit pointer to RAM or program memory. Bit 15 is used to
       detect RAM or program memory access.



STORING 14 BIT DATA
-------------------

Some devices supports 14 bits data stored in program memory. This
allows compact storage of 7 bit ASCII strings and 14 bits data. The
code sequence required for accessing these bits is longer than the
code for a return table. This means that code is saved when the
amount of data exceeds 40-50 bytes. The compiler will automatically
chose the optimal storage format.

The 14 bit data format allows storing up to 16000 bytes of
7 bit ASCII text on a 8k device.

The compiler will use the 14 bit format when a pragma statement
is used. This is normally found in the header file for the device.
The device must also contain certain special purpose registers:

  1)
    #pragma wideConstData
    /*char*/ EEADRH, EEADR, EEDATH, EEDATA, /*bit*/ EEPGD, RD
  2)
    #pragma wideConstData r
    /*char*/ PMADRH, PMADR, PMDATH, PMDATA, /*bit*/ RD
  3)
    #pragma wideConstData p
    /*char*/ PMADRH, PMADRL, PMDATH, PMDATL, /*bit*/ RD

When a constant table contains less than 256 byte of data, there
will be a tradeoff between speed and size. Using a return table
executes faster but requires more code when the table contains more
than 40-50 bytes of data. If speed is required, the following
pragma statement defines a new limit for size optimization.

    #pragma wideConstData 200  // return table up to 200 byte data



DATA OF SIZE 16 BIT OR MORE
---------------------------

The compiler allows access of 8,16,24 and 32 bits data, including
fixed and floating point formats. When using arrays or structures
with more than 256 byte data, single data items have to be aligned.
Alignment means that there should not be any remainder when dividing
the offset with the size of the data item. This is only a problem
when defining structures containing data of different sizes.

  const long tl[5] = { 10000, -10000, 0, 30000, -1 };
  const uns24 th[] = { 1000000, 0xFFFFFF, 9000000 };
  const int32 ti[] = { 1000000000, 0x7FFFFFFF, -900000000 };
  const fixed8_8 tf[] = { -1.1, 200.25, -100.25 };
  const float tp[] = { -1.1, 200.25, 23e20 };
  const double td[] = { -1.1, 200.25, 23e-30};
  const float16 ts[] = { -1.1, 200.25, 23e-30};
  ..
  l = tl[i];  // reading a long integer
  d = td[x];  // reading a double float constant



CODE PAGES
----------

When using devices with more than one codepage, the compiler will
automatically calculate the optimal codepage for the data. The
disadvantage is that the compiler will not know when a codepage is
full, so the functions still have to be moved manually between
codepages to find allowed and optimal combinations. Also, the
data can be located on a specific codepage by using a page type
modifier.

    const page1 char tx[] = "Hello!";



MERGING DATA
------------

The compiler will automatically merge equal strings and sub-strings,
and also other data items. Using small tables will increase the
chance of finding data items that can be merged.

Note that data containing initialized addresses are not merged.

Examples:

1. The string "world!" is identical to the last part of the
   string "Hello world!". It is therefore not required to use
   additional storage for the first string. The compiler handles
   the address calculations so that merged (or overlapping)
   strings are handled fully automatically. Note that the string
   termination '\0' also have to be equal, otherwise merging is
   not possible. For example, the string "world" can not be
   merged with the above strings.

2. Merging applies to all kinds of data. Data is compared byte by
   byte. This allows the first two of the following tables to be
   merged with the last one.

   const char a1[] = { 10, 20, 30 };
   const char a2[] = "ab";
   const char a3[] = { 5, 10, 20, 30, 'a', 'b', 0 };



EXAMPLES
--------

A table of pointer to strings:

    const struct {
        const char *s;
    } tb[] = {
        "Hello world",
        "Monday",
        "",
        "world"  // automatically merged with first string
    };

Note that 'const struct' is required to put the pointers in program
memory. Using 'const char *tx[];' means that the data resides in
program memory, but the table itself 'tx[]' resides in RAM.

    p = tb[i].s;  // const char *p; char i;

    t = *p++;     // char t;
    t = p[x];     // char x;

String parameters:

    myfunc("Hello");  // void myfunc(const char *str);
    myfunc(&tab[i]);  // char tab[20]; // string in RAM
    myfunc(ctab);  // const char ctab[] = "A string";

