The MC68040 is user-object-code compatible with the MC68030 and MC68881/MC68882.
The MC68040 floating-point unit is optimized to directly execute the most commonly used
subset of the extensive MC68881/MC68882 instruction set through hardware. Special
traps and stack frames for the unimplemented instructions and data types provide support
for the remaining instructions. These functions coupled with Motorola’s floating-point
software package (M68040FPSP) ensure complete user-object-code compatibility.
There are two versions of the M68040FPSP, one for applications compiled for the
MC68881/MC68882 (kernel version) and the other for applications compiled for the
MC68040 (library version). System integrators can install the kernel version as part of an
MC68040-based operating system. The kernel version is used to execute preexisting user
object code written for the MC68881/MC68882 as part of the operating system. User
applications need not be recompiled or modified in any way once the kernel version is
The MC68040 compiler writer and system integrator use the library version which provides
less overhead than the kernel version. Overhead is reduced because the appropriate
floating-point exception routine is called directly rather than taking an unimplemented
instruction trap. The library is M68000 application binary interface (ABI) and IEEE
exception-reporting compliant; it is not UNIX® exception-reporting compliant.
The M68040FPSP provides the following features:
Arithmetic and Transcendental Instructions
IEEE-Compliant Exception Handlers
MC68040 Unimplemented Data Type and Data Format Handlers
Can Reside in a 64-Kbyte ROM
Code Is Reentrant
The M68040FPSP satisfies the IEEE
Standard 754 for Binary Floating-Point Arithmetic
The average 25-MHz performance of the transcendental function subroutines is equivalent
to that of the 33-MHz MC68881/MC68882. The error bound is equivalent to that of the
®UNIX is a registered trademark of AT&T Bell Laboratories.
Terms of Use | Privacy Policy | DMCA Policy
2006-2020 Rsmanuals.com