dsPIC30F1010/202X
DS70178C-page 26
Preliminary
2006 Microchip Technology Inc.
The SA and SB bits are modified each time data passes
through the adder/subtracter, but can only be cleared by
the user. When set, they indicate that the accumulator
has overflowed its maximum range (bit 31 for 32-bit sat-
uration, or bit 39 for 40-bit saturation) and will be satu-
rated (if saturation is enabled). When saturation is not
enabled, SA and SB default to bit 39 overflow and thus
indicate that a catastrophic overflow has occurred. If the
COVTE bit in the INTCON1 register is set, SA and SB
bits will generate an arithmetic warning trap when
saturation is disabled.
The overflow and saturation Status bits can optionally
be viewed in the STATUS Register (SR) as the logical
OR of OA and OB (in bit OAB) and the logical OR of SA
and SB (in bit SAB). This allows programmers to check
one bit in the STATUS Register to determine if either
accumulator has overflowed, or one bit to determine if
either accumulator has saturated. This is useful for
complex number arithmetic, which typically uses both
the accumulators.
The device supports three Saturation and Overflow
modes.
1.
Bit 39 Overflow and Saturation:
When bit 39 overflow and saturation occurs, the
saturation logic loads the maximally positive 9.31
(0x7FFFFFFFFF) or maximally negative 9.31
value (0x8000000000) into the target accumula-
tor. The SA or SB bit is set and remains set until
cleared by the user. This is referred to as ‘super
saturation’ and provides protection against erro-
neous data or unexpected algorithm problems
(e.g., gain calculations).
2.
Bit 31 Overflow and Saturation:
When bit 31 overflow and saturation occurs, the
saturation logic then loads the maximally positive
1.31 value (0x007FFFFFFF) or maximally nega-
tive 1.31 value (0x0080000000) into the target
accumulator. The SA or SB bit is set and remains
set until cleared by the user. When this Saturation
mode is in effect, the guard bits are not used (so
the OA, OB or OAB bits are never set).
3.
Bit 39 Catastrophic Overflow
The bit 39 overflow Status bit from the adder is
used to set the SA or SB bit, which remain set
until cleared by the user. No saturation operation
is performed and the accumulator is allowed to
overflow (destroying its sign). If the COVTE bit in
the INTCON1 register is set, a catastrophic
overflow can initiate a trap exception.
2.4.2.2
Accumulator ‘Write Back’
The MAC class of instructions (with the exception of
MPY, MPY.N, ED
and EDAC) can optionally write a
rounded version of the high word (bits 31 through 16)
of the accumulator that is not targeted by the instruction
into data space memory. The write is performed across
the X bus into combined X and Y address space. The
following addressing modes are supported:
1.
W13, Register Direct:
The rounded contents of the non-target
accumulator are written into W13 as a 1.15
fraction.
2.
[W13] + = 2, Register Indirect with Post-Incre-
ment: The rounded contents of the non-target
accumulator are written into the address pointed
to by W13 as a 1.15 fraction. W13 is then
incremented by 2 (for a word write).
2.4.2.3
Round Logic
The round logic is a combinational block, which per-
forms a conventional (biased) or convergent (unbiased)
round function during an accumulator write (store). The
Round mode is determined by the state of the RND bit
in the CORCON register. It generates a 16-bit, 1.15 data
value which is passed to the data space write saturation
logic. If rounding is not indicated by the instruction, a
truncated 1.15 data value is stored and the least
significant word (lsw) is simply discarded.
Conventional rounding takes bit 15 of the accumulator,
zero-extends it and adds it to the ACCxH word (bits 16
through 31 of the accumulator). If the ACCxL word (bits
0 through 15 of the accumulator) is between 0x8000
and 0xFFFF (0x8000 included), ACCxH is incre-
mented. If ACCxL is between 0x0000 and 0x7FFF,
ACCxH is left unchanged. A consequence of this
algorithm is that over a succession of random rounding
operations, the value will tend to be biased slightly
positive.
Convergent (or unbiased) rounding operates in the
same manner as conventional rounding, except when
ACCxL equals 0x8000. If this is the case, the LSb (bit
16 of the accumulator) of ACCxH is examined. If it is ‘1’,
ACCxH is incremented. If it is ‘0’, ACCxH is not modi-
fied. Assuming that bit 16 is effectively random in
nature, this scheme will remove any rounding bias that
may accumulate.
The SAC and SAC.R instructions store either a trun-
cated (SAC) or rounded (SAC.R) version of the contents
of the target accumulator to data memory, via the X bus
of instructions, the accumulator write back operation
will function in the same manner, addressing combined
MCU (X and Y) data space though the X bus. For this
class of instructions, the data is always subject to
rounding.