Motorola M68060 User Manual page 375

MC68060 Software Package
other rounding modes. Transcendental instructions have an error bound of less than 0.6 unit
in the last place of double precision. The error bound for decimal conversions is 0.97 unit in
the destination precision for the round-to-nearest mode and 1.47 units in the last digit of the
destination precision for the other rounding modes.
C.3.2 Module 3: Full Floating-Point Kernel
The full floating-point kernel includes the following exception handlers:
1. Floating-point unimplemented instruction handler
2. Floating-point unimplemented data type handler
3. Unimplemented effective address handler
4. Floating-point arithmetic exception handlers
When the full floating-point kernel is integrated into the system, the entire MC68881 floating-
point coprocessor instruction set object-code compatibility is attained, and IEEE-754 trap
reporting compliance is achieved. This module stands on its own and is ideal for systems
whose applications were written with the full MC68881 instruction set in mind.
C.3.2.1 FULL FLOATING-POINT KERNEL MODULE ENTRY POINTS. The _fpsp_fline,
_fpsp_unsupp and _fpsp_effadd are entry points supplied for the floating-point unimple-
mented instruction, floating-point unimplemented data type and unimplemented effective
address handlers respectively. The _fpsp_snan, _fpsp_operr, _fpsp_ovfl, _fpsp_unfl,
_fpsp_dz, _fpsp_inex entry points are supplied as the floating-point arithmetic exception
handlers. These entry points are implemented such that the appropriate vector table entries
typically point directly to these functions. If the system chooses to perform certain system
functions prior to entering these entry points, the system can do so with the condition that
the system stack pointer must point to the exception stack frame at the time of the function
entry. Figure C-12 illustrates the relationship of the module to the vector table and system
software envelope.
C.3.2.2 FULL FLOATING-POINT KERNEL MODULE CALL-OUTS. The full floating-point
kernel requires the following call-outs: _real_fline, _real_fpu_disabled, _real_trace,
_real_trap, _real_bsun, _real_snan, _real_operr, _real_ovfl, _real_unfl, _real_dz,
_real_inex, _fpsp_done. In addition, C.4 Operating System Dependencies discusses the
_real_access call-out and other call-outs that are common to the unimplemented integer
instruction exception module.
C.3.2.2.1 The F-Line Exception Call-Outs. When the _fpsp_fline function is entered, it
checks the stack frame format and determines whether this is an unimplemented floating-
point instruction, FPU disabled or F-line illegal exception. If it is determined that the FPU is
disabled, the call-out _real_fpu_disabled is taken. It is up to the system software to either
emulate the instruction using integer instructions or simply turn on the FPU before returning
to restart the instruction. If the instruction is not recognized as an MC68881 instruction, the
call-out _real_fline is taken. The system software is responsible for taking the appropriate
action. If neither the FPU disabled or F-line illegal exception cases is true, then the
M68060SP emulates the instruction.
C-14 M68060 USER'S MANUAL MOTOROLA