For systems with a tightly coupled L2, a stride-based prefetcher may observe
access requests from both instruction and data L1 caches. However, the PC
address of an instruction miss gives no relevant training information to the
stride based prefetcher(there is no stride to train). In theses cases, its
better if the L2 stride prefetcher simply reverted back to a simple N-block
ahead prefetcher. This patch enables this option.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This patch extends the classic prefetcher to work on non-block aligned
addresses. Because the existing prefetchers in gem5 mask off the lower
address bits of cache accesses, many predictable strides fail to be
detected. For example, if a load were to stride by 48 bytes, with 64 byte
cachelines, the current stride based prefetcher would see an access pattern
of 0, 64, 64, 128, 192.... Thus not detecting a constant stride pattern. This
patch fixes this, by training the prefetcher on access and not masking off the
lower address bits.
It also adds the following configuration options:
1) Training/prefetching only on cache misses,
2) Training/prefetching only on data acceses,
3) Optionally tagging prefetches with a PC address.
#3 allows prefetchers to train off of prefetch requests in systems with
multiple cache levels and PC-based prefetchers present at multiple levels.
It also effectively allows a pipelining of prefetch requests (like in POWER4)
across multiple levels of cache hierarchy.
Improves performance on my gem5 configuration by 4.3% for SPECINT and 4.7% for SPECFP (geomean).
The patch
(1) removes the redundant writeback argument from findVictim()
(2) fixes the description of access() function
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
Adds very basic statistics on the number of tag and data accesses within the
cache, which is important for power modelling. For the tags, simply count
the associativity of the cache each time. For the data, this depends on
whether tags and data are accessed sequentially, which is given by a new
parameter. In the parallel case, all data blocks are accessed each time, but
with sequential accesses, a single data block is accessed only on a hit.
This patch enables tracking of cache occupancy per thread along with
ages (in buckets) per cache blocks. Cache occupancy stats are
recalculated on each stat dump.
Add some values and methods to the request object to track the translation
and access latency for a request and which level of the cache hierarchy responded
to the request.
This patch makes it possible to once again build gem5 without any
ISA. The main purpose is to enable work around the interconnect and
memory system without having to build any CPU models or device models.
The regress script is updated to include the NULL ISA target. Currently
no regressions make use of it, but all the testers could (and perhaps
should) transition to it.
--HG--
rename : build_opts/NOISA => build_opts/NULL
rename : src/arch/noisa/SConsopts => src/arch/null/SConsopts
rename : src/arch/noisa/cpu_dummy.hh => src/arch/null/cpu_dummy.hh
rename : src/cpu/intr_control.cc => src/cpu/intr_control_noisa.cc
This patch removes the notion of a peer block size and instead sets
the cache line size on the system level.
Previously the size was set per cache, and communicated through the
interconnect. There were plenty checks to ensure that everyone had the
same size specified, and these checks are now removed. Another benefit
that is not yet harnessed is that the cache line size is now known at
construction time, rather than after the port binding. Hence, the
block size can be locally stored and does not have to be queried every
time it is used.
A follow-on patch updates the configuration scripts accordingly.
This patch reorganizes the cache tags to allow more flexibility to
implement new replacement policies. The base tags class is now a
clocked object so that derived classes can use a clock if they need
one. Also having deriving from SimObject allows specialized Tag
classes to be swapped in/out in .py files.
The cache set is now templatized to allow it to contain customized
cache blocks with additional informaiton. This involved moving code to
the .hh file and removing cacheset.cc.
The statistics belonging to the cache tags are now including ".tags"
in their name. Hence, the stats need an update to reflect the change
in naming.
This patch fixes an outstanding issue in the cache timing calculations
where an atomic access returned a time in Cycles, but the port
forwarded it on as if it was in Ticks.
A separate patch will update the regression stats.
This patch changes the updards snoop packet to avoid allocating and
later deleting it. As the code executes in 0 time and the lifetime of
the packet does not extend beyond the block there is no reason to heap
allocate it.
This patch does some minor tidying up of the MSHR and MSHRQueue. The
clean up started as part of some ad-hoc tracing and debugging, but
seems worthwhile enough to go in as a separate patch.
The highlights of the changes are reduced scoping (private) members
where possible, avoiding redundant new/delete, and constructor
initialisation to please static code analyzers.
This patch provides useful printouts throughut the memory system. This
includes pretty-printed cache tags and function call messages
(call-stack like).
Fixes a latency calculation bug for accesses during a cache line fill.
Under a cache miss, before the line is filled, accesses to the cache are
associated with a MSHR and marked as targets. Once the line fill completes,
MSHR target packets pay an additional latency of
"responseLatency + busSerializationLatency". However, the "whenReady"
field of the cache line is only set to an additional delay of
"busSerializationLatency". This lacks the responseLatency component of
the fill. It is possible for accesses that occur on the cycle of
(or briefly after) the line fill to respond without properly paying the
responseLatency. This also creates the situation where two accesses to the
same address may be serviced in an order opposite of how they were received
by the cache. For stores to the same address, this means that although the
cache performs the stores in the order they were received, acknowledgements
may be sent in a different order.
Adding the responseLatency component to the whenReady field preserves the
penalty that should be paid and prevents these ordering issues.
Committed by: Nilay Vaish <nilay@cs.wisc.edu>
This patch solves the corner case scenario where the sendRetryEvent could be
scheduled twice, when an io device stresses the IOcache in the system. This
should not be possible in the cache system.
This patch fixes a newly introduced bug where the sender state was
popped before checking that it should be. Amazingly all regressions
pass, but Linux fails to boot on the detailed CPU with caches enabled.
This patch address the most important name shadowing warnings (as
produced when using gcc/clang with -Wshadow). There are many
locations where constructor parameters and function parameters shadow
local variables, but these are left unchanged.
This patch adds a check to ensure that the delay incurred by
the bus is not simply disregarded, but accounted for by someone. At
this point, all the modules do is to zero it out, and no additional
time is spent. This highlights where the bus timing is simply dropped
instead of being paid for.
As a follow up, the locations identified in this patch should add this
additional time to the packets in one way or another. For now it
simply acts as a sanity check and highlights where the delay is simply
ignored.
Since no time is added, all regressions remain the same.
This patch changes the names of the cache accessor functions to be in
line with those used by the ports. This is done to avoid confusion and
get closer to a one-to-one correspondence between the interface of the
memory object (the cache in this case) and the port itself.
The member function timingAccess has been split into a snoop/non-snoop
part to avoid branching on the isResponse() of the packet.
This patch changes the bus-related time accounting done in the packet
to be relative. Besides making it easier to align the cache timing to
cache clock cycles, it also makes it possible to create a Last-Level
Cache (LLC) directly to a memory controller without a bus inbetween.
The bus is unique in that it does not ever make the packets wait to
reflect the time spent forwarding them. Instead, the cache is
currently responsible for making the packets wait. Thus, the bus
annotates the packets with the time needed for the first word to
appear, and also the last word. The cache then delays the packets in
its queues before passing them on. It is worth noting that every
object attached to a bus (devices, memories, bridges, etc) should be
doing this if we opt for keeping this way of accounting for the bus
timing.
This patch removes the time field from the packet as it was only used
by the preftecher. Similar to the packet queue, the prefetcher now
wraps the packet in a deferred packet, which also has a tick
representing the absolute time when the packet should be sent.
This patch makes the clock member private to the ClockedObject and
forces all children to access it using clockPeriod(). This makes it
impossible to inadvertently change the clock, and also makes it easier
to transition to a situation where the clock is derived from e.g. a
clock domain, or through a multiplier.
This patch fixes a potential deadlock in the caches. This deadlock
could occur when more than one cache is used in a system, and
pkt->senderState is modified in between the two caches. This happened
as the caches relied on the senderState remaining unchanged, and used
it for instantaneous upstream communication with other caches.
This issue has been addressed by iterating over the linked list of
senderStates until we are either able to cast to a MSHR* or
senderState is NULL. If the cast is successful, we know that the
packet has previously passed through another cache, and therefore
update the downstreamPending flag accordingly. Otherwise, we do
nothing.
This patch adds a predecessor field to the SenderState base class to
make the process of linking them up more uniform, and enable a
traversal of the stack without knowing the specific type of the
subclasses.
There are a number of simplifications done as part of changing the
SenderState, particularly in the RubyTest.
This patch merely adopts a more strict use of const for the cache
member functions and variables, and also moves a large portion of the
member functions from public to protected.
Virtualized CPUs and the fastmem mode of the atomic CPU require direct
access to physical memory. We currently require caches to be disabled
when using them to prevent chaos. This is not ideal when switching
between hardware virutalized CPUs and other CPU models as it would
require a configuration change on each switch. This changeset
introduces a new version of the atomic memory mode,
'atomic_noncaching', where memory accesses are inserted into the
memory system as atomic accesses, but bypass caches.
To make memory mode tests cleaner, the following methods are added to
the System class:
* isAtomicMode() -- True if the memory mode is 'atomic' or 'direct'.
* isTimingMode() -- True if the memory mode is 'timing'.
* bypassCaches() -- True if caches should be bypassed.
The old getMemoryMode() and setMemoryMode() methods should never be
used from the C++ world anymore.
the cache drainManager is set but never cleared, this is because
the cache itself does not need to be drained and thus never
triggers a signalDrainDone(). because the drainManager variable
is not used properly and does not appear to be necessary it has
been removed with this patch.
The current implementation in gem5 just keeps a list of locks per cacheline.
Due to this, a store to a non-overlapping portion of the cacheline can cause an
LL/SC pair to fail. This patch simply adds an address range to the lock
structure, so that the lock is only invalidated if the store overlaps the lock
range.
When the classic gem5 cache sees an uncacheable memory access, it used
to ignore it or silently drop the cache line in case of a
write. Normally, there shouldn't be any data in the cache belonging to
an uncacheable address range. However, since some architecture models
don't implement cache maintenance instructions, there might be some
dirty data in the cache that is discarded when this happens. The
reason it has mostly worked before is because such cache lines were
most likely evicted by normal memory activity before a TLB flush was
requested by the OS.
Previously, the cache model would invalidate cache lines when they
were accessed by an uncacheable write. This changeset alters this
behavior so all uncacheable memory accesses cause a cache flush with
an associated writeback if necessary. This is implemented by reusing
the cache flushing machinery used when draining the cache, which
implies that writebacks are performed using functional accesses.
The IIC replacement policy seems to be unused and has probably
gathered too much bit rot to be useful. This patch removes the IIC and
its associated cache parameters.
This patch adds support for the following optional drain methods in
the classical memory system's cache model:
memWriteback() - Write back all dirty cache lines to memory using
functional accesses.
memInvalidate() - Invalidate all cache lines. Dirty cache lines
are lost unless a writeback is requested.
Since memWriteback() is called when checkpointing systems, this patch
adds support for checkpointing systems with caches. The serialization
code now checks whether there are any dirty lines in the cache. If
there are dirty lines in the cache, the checkpoint is flagged as bad
and a warning is printed.
This patch moves the draining interface from SimObject to a separate
class that can be used by any object needing draining. However,
objects not visible to the Python code (i.e., objects not deriving
from SimObject) still depend on their parents informing them when to
drain. This patch also gets rid of the CountedDrainEvent (which isn't
really an event) and replaces it with a DrainManager.
When casting objects in the generated SWIG interfaces, SWIG uses
classical C-style casts ( (Foo *)bar; ). In some cases, this can
degenerate into the equivalent of a reinterpret_cast (mainly if only a
forward declaration of the type is available). This usually works for
most compilers, but it is known to break if multiple inheritance is
used anywhere in the object hierarchy.
This patch introduces the cxx_header attribute to Python SimObject
definitions, which should be used to specify a header to include in
the SWIG interface. The header should include the declaration of the
wrapped object. We currently don't enforce header the use of the
header attribute, but a warning will be generated for objects that do
not use it.
This patch adds an additional level of ports in the inheritance
hierarchy, separating out the protocol-specific and protocl-agnostic
parts. All the functionality related to the binding of ports is now
confined to use BaseMaster/BaseSlavePorts, and all the
protocol-specific parts stay in the Master/SlavePort. In the future it
will be possible to add other protocol-specific implementations.
The functions used in the binding of ports, i.e. getMaster/SlavePort
now use the base classes, and the index parameter is updated to use
the PortID typedef with the symbolic InvalidPortID as the default.
This patch addresses a number of smaller issues identified by the code
inspection utility cppcheck. There are a number of identified leaks in
the arm/linux/system.cc (although the function only get's called once
so it is not a major problem), a few deletes in dev/x86/i8042.cc that
were not array deletes, and sprintfs where the character array had one
element less than needed. In the IIC tags there was a function
allocating an array of longs which is in fact never used.
This patch changes the cache-related latencies from an absolute time
expressed in Ticks, to a number of cycles that can be scaled with the
clock period of the caches. Ultimately this patch serves to enable
future work that involves dynamic frequency scaling. As an immediate
benefit it also makes it more convenient to specify cache performance
without implicitly assuming a specific CPU core operating frequency.
The stat blocked_cycles that actually counter in ticks is now updated
to count in cycles.
As the timing is now rounded to the clock edges of the cache, there
are some regressions that change. Plenty of them have very minor
changes, whereas some regressions with a short run-time are perturbed
quite significantly. A follow-on patch updates all the statistics for
the regressions.
In the current caches the hit latency is paid twice on a miss. This patch lets
a configurable response latency be set of the cache for the backward path.
This patch takes the final plunge and transitions from the templated
Range class to the more specific AddrRange. In doing so it changes the
obvious Range<Addr> to AddrRange, and also bumps the range_map to be
AddrRangeMap.
In addition to the obvious changes, including the removal of redundant
includes, this patch also does some house keeping in preparing for the
introduction of address interleaving support in the ranges. The Range
class is also stripped of all the functionality that is never used.
--HG--
rename : src/base/range.hh => src/base/addr_range.hh
rename : src/base/range_map.hh => src/base/addr_range_map.hh
This patch addresses a few minor issues reported by the clang static
analyzer.
The analysis was run with:
scan-build -disable-checker deadcode \
-enable-checker experimental.core \
-disable-checker experimental.core.CastToStruct \
-enable-checker experimental.cpluscplus
This seperates the functionality to clear the state in a block into
blk.hh and the functionality to udpate the tag information into the
tags. This gets rid of the case where calling invalidateBlk on an
already-invalid block does something different than calling it on a
valid block, which was confusing.
This patch is a first step to using Cycles as a parameter type. The
main affected modules are the CPUs and the Ruby caches. There are
definitely plenty more places that are affected, but this patch serves
as a starting point to making the transition.
An important part of this patch is to actually enable parameters to be
specified as Param.Cycles which involves some changes to params.py.
