SDKZSPF LSI, SDKZSPF Datasheet - Page 57

• Manufacturer Part Number: SDKZSPF
• Manufacturer: LSI

Specifications of SDKZSPF

• Lead Free Status / Rohs Status: Supplier Unconfirmed

performance and code size can be obtained by selectively using the
-mlong_call option, but this may require repetitive trial and error to
determine which files can safely be converted to use call-immediates.
One important exception is a file which does not call a function external
to the file; in this case, the necessity of -mlong_call does not depend
on other files or on the link order—this kind of file either always requires
-mlong_call or never requires it. Note that with sdcc, the use of -
mlong_call does not guarantee that all calls will be long calls; that is,
the assembly optimizer, sdopt, converts long calls to call immediates if
it can determine that such a conversion is safe. The assembly optimizer
can be disabled by specifying the -mno_sdopt option; otherwise, it is
automatically invoked when optimization levels greater than -O1 are
selected. Note that for debugging optimized code, the -mno_sdopt
option should be used, because the assembly optimizers perform
optimizations that make debugging extremely difficult.
sdopt takes in assembly generated by the compiler proper and optimizes
it to produce improved assembly. The optimizations that sdopt performs
include simplification of constant generation, conversion of loops to use
loop registers, dead code elimination, loop invariant code motion,
conversion of long calls to call-immediate, instruction scheduling, and
improved register utilization.
zdopt takes in assembly generated by zdcc and optimizes it to produce
improved assembly. The optimizations that zdopt performs include dead
code elimination, loop invariant code motion, instruction scheduling, and
improved register utilization.
zdcc supports two models of memory. The default small memory model
requires that data and bss sections be placed in the first 64K words of
data memory. The large data model has no requirements on the size or
placement of the data and bss sections. The large data model is selected
with the “-mlarge_data” option. For both models, the pointer size is
32 bits. Both models allow stack and heap space to use all addressable
memory. Code generated with the small data model is more compact and
has better performance than code generated with the large data model.
The small data model allows a shorter instruction sequence to be used
to access memory in the data or bss sections.
Compiler Options
3-3
Copyright © 1999-2003 by LSI Logic Corporation. All rights reserved.