align.Target alignment considerations

Product : ISaGRAF 3.3

Date : 18-August-1998

File : align.Target alignment considerations .htm

Subject : Target alignment considerations

Keywords: Alignment - ISA_ALIGN - tasy0def.h - SYSTEM

____________________________________________________________________

The alignment is defined by rules applied to the address location

where data is to be allocated according to its type. For example

with an alignment at 4, it generaly means that:

- 16 bits variable is to be allocated at an address multiple of 2

- 32 bits variable is to be allocated at an address multiple of 4

This is also true within structures. Therefore it may leads to

gaps between different fields of a structure.

These alignment constraints are closely linked to the hardware, the

CPU and, in the end, the compiler. For instance some CPUs cannot

access a long (32bits) variable at an address non multiple of 4.

As a consequence you get a bus error or something similar.

In some other cases (Intel x86, pentium CPUs or some Motorola

Power Pc) it is just for execution speed concerns.

On previous target versions, the alignment value was fixed to 2 (1

is obviously ok)

The result of an alignment mismatch can then either be a system

error or some data mismatches between target structures and

workbench generated application code. Any way it does not work.

From target 3.3 version, some modifications have been made to try

to make it work with an alignment different from 2 (generaly 4).

In that way, the IDK customer needs to define ISA_ALIGN macro in the

definition file (tasy0def.h) with its system alignment. When not

defined in this file, the default value is 2 (this keeps

compatibility).

Any way, some parts are still not taken into account by this

alignment. They are:

- IOs OEM data. This can be worked arround in defining some

constraints: The IO integrator must declare homogeneous

oem data such as there is no alignment mismatches.

- Call to the ISaGRAF SYSTEM function for saving ALL. This can be

worked around by calling the function several times for each

type.

- All points not discovered yet...Please tell us

____________________________________________________________________