EasyManua.ls Logo

AMD AMD5K86 - Retire

AMD AMD5K86
416 pages
Print Icon
To Next Page IconTo Next Page
To Next Page IconTo Next Page
To Previous Page IconTo Previous Page
To Previous Page IconTo Previous Page
Loading...
AMD~
AMD5J!36
Processor
Technical
Reference
Manual
18524B/O-Marl996
2.2.5
Retire
2-12
The
processor
implements
a
real.
state
(non-speculative)
regis-
ter file
that
contains
the
x86-architecture
registers
and
a
real-
state
8-Kbyte
data
cache.
While
ROPs
complete
out
of
order
and
their
results
are
forwarded
to
other
execution
units
and
to
the
ROB
out
of
order,
their
results
are
always
written
at
retire-
ment
time
to
the
real-state
x86
registers
in
program
order.
Likewise,
as
results
are
written
from
the
load/store
units
to
the
store
buffer
out
of
order,
they
are
always
written
at
retirement
time
to
the
data
cache
and/or
memory
in
program
order.
An
x86
instruction
is
said
to
retire
when
the
ROB
or
store
buffer
writes
the
operands
for
all
of
its
ROPs,
in
program
order,
to
the
x86
real-state
registers
or
the
data
cache.
At
the
point
of
retirement,
the
register
file
and
data
cache
fully
reflect
the
execution
of
an
instruction.
Any
associated
excep-
tions
are
recognized
(the
ROB
facilitates
precise
exception
handling),
any
external
interrupts
that
were
latched
or
are
cur-
rently
held
asserted
are
recognized,
and
the
instruction
pointer
is
updated.
For
instructions
that
store
an
operand
to
memory,
retirement
is
the
time
at
which
the
store
is
guaran-
teed
to
be
written
externally.
When
a pipeline invalidation
(flush)
occurs,
it
does
so
at
the
retirement
stage,
causing
all
instructions
in
the
pipeline
that
have
not
reached
the
retire-
ment
stage
to
be
invalidated.
The
retirement
stage
is also
called
the
instruction-retirement
boundary,
or
simply
instruction boundary.
The
processor
can
retire
up
to
four
instructions
per
processor
clock.
Thus,
the
next
set
of
up
to
four
instructions
that
are
candidates
to
retire
determines
the
next
instruction
boundary
at
which
an
external
interrupt
can
be
recognized.
Only
one
store
(from
the
store
buffer
or
from
either
of
the
pro-
cessor's
two
writeback
buffers)
can
be
among
the
set
of
up
to
four
instructions
that
retire
simultaneously.
Thus,
for
example,
the
processor
will
only
finish
one
in-progress
store
cycle
on
the
bus
before
recognizing
an
asserted
HOLD, sm,
or
STpCLK.
If
the
set
of
retirement
candidates
in
any
clock
includes
more
than
one
store,
only
those
instructions
up
to
(but
not
including)
the
second
store
will
retire.
The
remaining
stores
occur
one
at
a
time,
in
their
queued
order,
during
subsequent
retirements.
Internal
Architecture

Table of Contents

Related product manuals