| Code Generation Conventions |
The
following sections describe
register conventions and patch writing.
By
default, the compiler generates a patch area in each compiled module. This
patch area provides space for you to apply maintenance to your modules in
object code or load module form (zaps).
The patch area is generated in the first 4096 bytes
of a CSECT known as the constants CSECT for the compilation. The constants
CSECT is permanently addressed by general register 4, so the patch area is
always addressable. The default patch area is 1/64 the size of the generated
code for the compilation, rounded up to a multiple of 8 bytes, with a minimum
size of 24 bytes and a maximum size of 256 bytes. The patch area is generated
as a series of S-cons (address constants in base-displacement form). Each
SCON contains its own address in base-displacement form, using register 4
as a base register. This minimizes errors in patch-writing, both for branches
to the patch area and for branches within it.
To find the patch area for a module, look near the end
of the OMD listing for the constants CSECT. (See Compiler-generated Names.) The patch area is found at the end of the
CSECT or just before 4096 bytes if the CSECT exceeds 4096 bytes in size.
It is easily recognized by its distinctive S-con format.
It is possible, for very large compilations, or when
the zapspace option is specified, for more patch space to be required than
the compiler can generate in the first 4096 bytes of the patch area. In this
case, the compiler will generate one or more secondary patch areas later in
the constants CSECT. These areas have the form of S-cons, using register
0 as the base register. Because these areas are not in the first 4K of the
constants section, they are not directly addressable but can be branched to
from the previous patch area.
The
zapspace
compiler option can be used to alter the size of the compiler-generated
patch area. The size of the patch area can be increased or its generation
suppressed.
The
zapspace
option accepts an integer
value between 0 and 22, inclusive, that specifies the factor by which the
default patch area size is to be multiplied. If the factor is 0, then no
patch area is generated. For example, if the default patch area is 48 bytes
and the
zapspace
option specifies a factor of 3, then the
patch area actually generated is 144 bytes long. In no case does the compiler
generate more than 512 bytes of patch area.
The
zapmin
compiler option can be used to specify the minimum size of the
compiler-generated patch area.
The
zapmin
option accepts an integer value
that specifies the number of bytes in the patch area. The default is 24.
For example,
zapmin(64)
ensures that the patch area is at least 64 bytes.
In no case does the compiler generate more than 512
bytes of patch area.
The
following list summarizes register conventions. You need this information
if you are writing patches.
The best way to write a zap is to examine other code
in the function that does something similar to what is required and to model
the zap on that code.
General register 1 (R1) can be used as a scratch register
for zaps unless it is already directly used in the zapped code or a function
call sequence (which sets R1) is being zapped. The compiler never remembers
the value in R1 across statements. If it is not possible to use R1, you should
be aware that values can be kept in registers for a large number of statements,
including conditional branches and function calls. The best register use for
a zap (other than R1) is a register that is set soon after the zap without
being used between the zap and the code that sets the register to a new value
(that is, one in which the existing value is discarded).
Copyright © 2001
by SAS Institute Inc., Cary, NC, USA. All rights reserved.
