Download MIPSproTM Assembly Language Programmer`s Guide

6: Coprocessor Instruction Set
Software may simulate this exception for other operations that produce a signed
infinity, such as ln(0), sec(p/2), csc(0) or 0-1.
Overflow Exception
The overflow exception is signaled when what would have been the magnitude of the
rounded floating-point result, were the exponent range unbounded, is larger than the
destination format’s largest finite number. The result, when no trap occurs, is
determined by the rounding mode and the sign of the intermediate result.
If overflow traps are enabled, the result register is not modified, and the source
registers are preserved.
Underflow Exception
Two related events contribute to underflow. One is the creation of a tiny non-zero
result between 2Emin (minimum expressible exponent) which, because it is tiny, may
cause some other exception later. The other is extraordinary loss of accuracy during
the approximation of such tiny numbers by denormalized numbers.
The IEEE standard permits a choice in how these events are detected, but requires
that they must be detected the same way for all operations.
The IEEE standard specifies that “tininess” may be detected either: “after rounding”
(when a nonzero result computed as though the exponent range were unbounded
would lie strictly between 2Emin), or “before rounding” (when a nonzero result
computed as though the exponent range and the precision were unbounded would lie
strictly between 2Emin). The architecture requires that tininess be detected after
rounding.
Loss of accuracy may be detected as either “denormalization loss” (when the delivered
result differs from what would have been computed if the exponent range were
unbounded), or “inexact result” (when the delivered result differs from what would
have been computed if the exponent range and precision were both unbounded). The
architecture requires that loss of accuracy be detected as inexact result.
When an underflow trap is not enabled, underflow is signaled (via the underflow
flag) only when both tininess and loss of accuracy have been detected. The delivered
result might be zero, denormalized, or 2Emin. When an underflow trap is enabled,
underflow is signaled when tininess is detected regardless of loss of accuracy.
If underflow traps are enabled, the result register is not modified, and the source
registers are preserved.
84
007–2418–006