|
Operation |
Semantics and Operands |
| negate |
The negate operation may be applied to a source operand with any
numeric type to get a result of any other numeric type. The result
is the source value with the sign reversed. |
|
invert |
The invert operation returns the multiplicative inverse of the
input. That is, if the input is x, the result is 1/x. The invert
operation may be applied to a source of any numeric type to
return any other numeric type, though it's practically useless
for integer types no matter how the rounding is defined.
|
|
absolute_value |
The absolute_value operation may be applied to an operand of any
numeric type to give a result of any other numeric type. |
|
bitwise_not |
The bitwise_not operation may be applied to any unsigned integer
type operand to give back a value of any other unsigned integer
type. If the result is larger, it is zero-filled while if the
result it smaller than the source type the high-order bits are
dropped. |
|
logical_not |
The logical_not operation is allowed for a source operand of any
boolean type and may have any other boolean type as result type.
|
|
convert |
The convert operation is used to convert a value in one type to the
corresponding value in a different type. For example, converting a
32-bit integer with value 1 into a 32-bit floating-point value
should give the floating-point value 1.0. The bit patterns will
often be totally different, but the value will be the same (or
the closest appropriate value, if applicable). Conversions
are allowed between any numeric, enumerated, or pointer type and any
other numeric, enumerated, or pointer type. For conversions between
integer types where the value is representable in both types, the
value must stay the same. Other than that, Standard SUIF does
not specify anything about the range of values that are legal for
conversion or what the effects of converting various values are.
|
|
treat_as |
The treat_as operation takes a source value of any data type at all
and returns a result of any other data type at all that has the same
bit pattern. That is, it treats the bits as a different type.
So applying the treat_as operation to a 32-bit integer with value 1
to get a 32-bit floating-point value is likely not to give the
floating-point representation for 1.0. Rather it will give
whatever floating-point number is represented by a one in the
low-order bit and all other bits zero. Note that using this
operation can cause very machine-dependent effects, since it all
depends on what bit patterns are used for various types. If the
result type is smaller, the high-order bits are dropped. If the
result is larger, the high-order bits are filled with zeros.
|
