This patch makes O3 CPU work along with the Ruby memory model. Ruby
overwrites the senderState pointer with another pointer. The pointer
is restored only when Ruby gets done with the packet. LSQ makes use of
senderState just after sendTiming() returns. But the dynamic_cast returns
a NULL pointer since Ruby's senderState pointer is from a different class.
Storing the senderState pointer before calling sendTiming() does away with
the problem.
Initialize flags via the Event constructor instead of calling
setFlags() in the body of the derived class's constructor. I
forget exactly why, but this made life easier when implementing
multi-queue support.
Also rename Event::getFlags() to isFlagSet() to better match
common usage, and get rid of some unused Event methods.
In FS mode the syscall function will panic, but the interface will be
consistent and code which calls syscall can be compiled in. This will allow,
for instance, instructions that use syscall to be built unconditionally but
then not returned by the decoder.
Only create a memory ordering violation when the value could have changed
between two subsequent loads, instead of just when loads go out-of-order
to the same address. While not very common in the case of Alpha, with
an architecture with a hardware table walker this can happen reasonably
frequently beacuse a translation will miss and start a table walk and
before the CPU re-schedules the faulting instruction another one will
pass it to the same address (or cache block depending on the dendency
checking).
This patch has been tested with a couple of self-checking hand crafted
programs to stress ordering between two cores.
The performance improvement on SPEC benchmarks can be substantial (2-10%).
Having two StaticInst classes, one nominally ISA dependent and the other ISA
dependent, has not been historically useful and makes the StaticInst class
more complicated that it needs to be. This change merges StaticInstBase into
StaticInst.
This change pulls the instruction decoding machinery (including caches) out of
the StaticInst class and puts it into its own class. This has a few intrinsic
benefits. First, the StaticInst code, which has gotten to be quite large, gets
simpler. Second, the code that handles decode caching is now separated out
into its own component and can be looked at in isolation, making it easier to
understand. I took the opportunity to restructure the code a bit which will
hopefully also help.
Beyond that, this change also lays some ground work for each ISA to have its
own, potentially stateful decode object. We'd be able to include less
contextualizing information in the ExtMachInst objects since that context
would be applied at the decoder. Also, the decoder could "know" ahead of time
that all the instructions it's going to see are going to be, for instance, 64
bit mode, and it will have one less thing to check when it decodes them.
Because the decode caching mechanism has been separated out, it's now possible
to have multiple caches which correspond to different types of decoding
context. Having one cache for each element of the cross product of different
configurations may become prohibitive, so it may be desirable to clear out the
cache when relatively static state changes and not to have one for each
setting.
Because the decode function is no longer universally accessible as a static
member of the StaticInst class, a new function was added to the ThreadContexts
that returns the applicable decode object.
SEV instructions were originally implemented to cause asynchronous squashes
via the generateTCSquash() function in the O3 pipeline when updating the
SEV_MAILBOX miscReg. This caused race conditions between CPUs in an MP system
that would lead to a pipeline either going inactive indefinitely or not being
able to commit squashed instructions. Fixed SEV instructions to behave like
interrupts and cause synchronous sqaushes inside the pipeline, eliminating
the race conditions. Also fixed up the semantics of the WFE instruction to
behave as documented in the ARMv7 ISA description to not sleep if SEV_MAILBOX=1
or unmasked interrupts are pending.
Two issues are fixed in this patch:
1. The load and store pc passed to the predictor are passed in reverse order.
2. The flag indicating that a barrier is inflight was never cleared when
the barrier was squashed instead of committed. This made all load insts
dependent on a non-existent barrier in-flight.
Change the way instructions are squashed on memory ordering violations
to squash the violator and younger instructions, not all instructions
that are younger than the instruction they violated (no reason to throw
away valid work).
It's possible (though until now very unlikely) for fetchAddr to get out of
sync with the actual PC of the current instruction. This change forcefull
resets fetchAddr at the end of every instruction.
Until now, the only reason a macroop would be left was because it ended at a
microop marked as the last microop. In O3 with branch prediction, it's
possible for the branch predictor to have entries which originally came from
different instructions which happened to have the same RIP. This could
theoretically happen in many ways, but it was encountered specifically when
different programs in different address spaces ran one after the other in
X86_FS.
What would happen in that case was that the macroop would continue to be
looped over and microops fetched from it until it reached the last microop
even though the macropc had moved out from under it. If things lined up
properly, this could mean that the end bytes of an instruction actually fell
into the instruction sized block of memory after the one in the predecoder.
The fetch loop implicitly assumes that the last instruction sized chunk of
memory processed was the last one needed for the instruction it just finished
executing. It would then tell the predecoder to move to an offset within the
bytes it was given that is larger than those bytes, and that would trip an
assert in the x86 predecoder.
This change fixes this problem by making fetch stop processing the current
macroop if the address it should be fetching from changed when the PC is
updated. That happens when the last microop was reached because the instruction
handled it properly, and it also catches the case where the branch predictor
makes fetch do a macro level branch when it shouldn't.
The check of isLastMicroop is retained because otherwise, a macroop that
branches back to itself would act like a single, long macroop instead of
multiple instances of the same microop. There may be situations (which may
turn out to be purely hypothetical) where that matters.
This also fixes a relatively minor issue where the curMacroop variable would
be set to NULL immediately after seeing that a microop was the last one before
curMacroop was used to build the dyninst. The traceData structure would have a
NULL pointer to the macroop for that microop.
Before this change, the commit stage would wait until the ROB and store queue
were empty before recognizing an interrupt. The fetch stage would stop
generating instructions at an appropriate point, so commit would then wait
until a valid time to interrupt the instruction stream. Instructions might be
in flight after fetch but not the in the ROB or store queue (in rename, for
instance), so this change makes commit wait until all in flight instructions
are finished.
This patch replaces RUBY with PROTOCOL in all the SConscript files as
the environment variable that decides whether or not certain components
of the simulator are compiled.
This constructor assumes that the ExtMachInst can be decoded directly into a
StaticInst that's useful to execute. With the advent of microcoded
instructions that's no longer true.
When fetching from the microcode ROM, if the PC is set so that it isn't in the
cache block that's been fetched the CPU will get stuck. The fetch stage
notices that it's in the ROM so it doesn't try to fetch from the current PC.
It then later notices that it's outside of the current cache block so it skips
generating instructions expecting to continue once the right bytes have been
fetched. This change lets the fetch stage attempt to generate instructions,
and only checks if the bytes it's going to use are valid if it's really going
to use them.
Implemented a pipeline activity viewer as a python script (util/o3-pipeview.py)
and modified O3 code base to support an extra trace flag (O3PipeView) for
generating traces to be used as inputs by the tool.
Branch predictor could not predict a branch in a nested loop because:
1. The global history was not updated after a mispredict squash.
2. The global history was updated in the fetch stage. The choice predictors
that were updated used the changed global history. This is incorrect, as
it incorporates the state of global history after the branch in
encountered. Fixed update to choice predictor using the global history
state before the branch happened.
3. The global predictor table was also updated using the global history state
before the branch happened as above.
Additionally, parameters to initialize ctr and history size were reversed.
Fixed up the patch from Yasuko Watanabe that enabled pipelining of fetch accessess to
icache to work with recent changes to main repository.
Also added in ability for fetch stage to delay issuing the fault carrying
nop when a pipeline fetch causes a fault and no fetch bandwidth is available
until the next cycle.
readBytes and writeBytes had the word "bytes" in their names because they
accessed blobs of bytes. This distinguished them from the read and write
functions which handled higher level data types. Because those functions don't
exist any more, this change renames readBytes and writeBytes to more general
names, readMem and writeMem, which reflect the fact that they are how you read
and write memory. This also makes their names more consistent with the
register reading/writing functions, although those are still read and set for
some reason.
This patch rpovides functional access support in Ruby. Currently only
the M5Port of RubyPort supports functional accesses. The support for
functional through the PioPort will be added as a separate patch.
The DTB expects the correct PC in the ThreadContext
but how if the memory accesses are speculative? Shouldn't
we send along the requestor's PC to the translate functions?
if a faulting instruction reaches an execution unit,
then ignore it and pass it through the pipeline.
Once we recognize the fault in the graduation unit,
dont allow a second fault to creep in on the same cycle.
Before graduating an instruction, explicitly check fault
by making the fault check it's own separate command
that can be put on an instruction schedule.
remove events in the resource pool that can be called from the CPU event, since the CPU
event is scheduled at the same time at the resource pool event.
----
Also, match the resPool event function names to the cpu event function names
----
only update BTB on a taken branch and update branch predictor w/pcstate from instruction
---
only pay attention to branch predictor updates if the the inst. is in fact a branch
formerly, this was implicit when you accessed the execution unit
or the use-def unit but it's better that this just be something
that a user can specify.
Architectures like SPARC need to read the window pointer
in order to figure out it's register dependence. However,
this may not get updated until after an instruction gets
executed, so now we lazily detect the register dependence
in the EXE stage (execution unit or use_def). This
makes sure we get the mapping after the most current change.
Add a few constants and functions that the InOrder model wants for SPARC.
* * *
sparc: add eaComp function
InOrder separates the address generation from the actual access so give
Sparc that functionality
* * *
sparc: add control flags for branches
branch predictors and other cpu model functions need to know specific information
about branches, so add the necessary flags here
Calculation of offset to copy from storeQueue[idx].data structure for load to
store forwarding fixed to be difference in bytes between store and load virtual
addresses. Previous method would induce bug where a load would index into
buffer at the wrong location.
If a split load fails on a blocked cache wbOutstanding can be decremented
twice if the first part of the split load succeeds and the second part fails.
Condition the decrementing on not having completed the first part of the load.
This patch fixes two problems with the O3 cpu model. The first is an issue
with an instruction fetch causing a fault on the next address while the
current macro-op is being issued. This happens when the micro-ops exceed
the fetch bandwdith and then on the next cycle the fetch stage attempts
to issue a request to the next line while it still has micro-ops to issue
if the next line faults a fault is attached to a micro-op in the currently
executing macro-op rather than a "nop" from the next instruction block.
This leads to an instruction incorrectly faulting when on fetch when
it had no reason to fault.
A similar problem occurs with interrupts. When an interrupt occurs the
fetch stage nominally stops issuing instructions immediately. This is incorrect
in the case of a macro-op as the current location might not be interruptable.
Debug flags are ExecUser, ExecKernel, and ExecAsid. ExecUser and
ExecKernel are set by default when Exec is specified. Use minus
sign with ExecUser or ExecKernel to remove user or kernel tracing
respectively.
Instructions that load an address and are control instructions can
execute down the wrong path if they were predicted correctly and then
instructions following them are squashed. If an instruction is a
memory and control op use the predicted address for the next PC instead
of just advancing the PC. Without this change NPC is used for the next
instruction, but predPC is used to verify that the branch was successful
so the wrong path is silently executed.
The network tester terminates after injecting for sim_cycles
(default=1000), instead of having to explicitly pass --maxticks from the
command line as before. If fixed_pkts is enabled, the tester only
injects maxpackets number of packets, else it keeps injecting till sim_cycles.
The tester also works with zero command line arguments now.
At the same time, rename the trace flags to debug flags since they
have broader usage than simply tracing. This means that
--trace-flags is now --debug-flags and --trace-help is now --debug-help
This change fixes a small bug in the arm copyRegs() code where some registers
wouldn't be copied if the processor was in a mode other than MODE_USER.
Additionally, this change simplifies the way the O3 switchCpu code works by
utilizing TheISA::copyRegs() to copy the required context information
rather than the adhoc copying that goes on in the CPU model. The current code
makes assumptions about the visibility of int and float registers that aren't
true for all architectures in FS mode.
The comment in the code suggests that the checking granularity should be 16
bytes, however in reality the shift by 8 is 256 bytes which seems much
larger than required.
***
(1): get rid of expandForMT function
MIPS is the only ISA that cares about having a piece of ISA state integrate
multiple threads so add constants for MIPS and relieve the other ISAs from having
to define this. Also, InOrder was the only core that was actively calling
this function
* * *
(2): get rid of corespecific type
The CoreSpecific type was used as a proxy to pass in HW specific params to
a MIPS CPU, but since MIPS FS hasnt been touched for awhile, it makes sense
to not force every other ISA to use CoreSpecific as well use a special
reset function to set it. That probably should go in a PowerOn reset fault
anyway.
The tester code is in testers/networktest.
The tester can be invoked by configs/example/ruby_network_test.py.
A dummy coherence protocol called Network_test is also addded for network-only simulations and testing. The protocol takes in messages from the tester and just pushes them into the network in the appropriate vnet, without storing any state.
This change speeds up booting, especially in MP cases, by not executing
udelay() on the core but instead skipping ahead tha amount of time that is being
delayed.
This change fixes the problem for all the cases we actively use. If you want to try
more creative I/O device attachments (E.g. sharing an L2), this won't work. You
would need another level of caching between the I/O device and the cache
(which you actually need anyway with our current code to make sure writes
propagate). This is required so that you can mark the cache in between as
top level and it won't try to send ownership of a block to the I/O device.
Asserts have been added that should catch any issues.
Without this change the a store can be issued to the cache multiple times.
If this case occurs when the l1 cache is out of mshrs (and thus blocked)
the processor will never make forward progress because each cycle it will
send a single request using the recently freed mshr and not completing the
multipart store. This will continue forever.
There may not be a formally correct spelling for the past tense of mmap, but
mmapped is the spelling Google doesn't try to autocorrect. This makes sense
because it mirrors the past tense of map->mapped and not the past tense of
cape->caped.
--HG--
rename : src/arch/alpha/mmaped_ipr.hh => src/arch/alpha/mmapped_ipr.hh
rename : src/arch/arm/mmaped_ipr.hh => src/arch/arm/mmapped_ipr.hh
rename : src/arch/mips/mmaped_ipr.hh => src/arch/mips/mmapped_ipr.hh
rename : src/arch/power/mmaped_ipr.hh => src/arch/power/mmapped_ipr.hh
rename : src/arch/sparc/mmaped_ipr.hh => src/arch/sparc/mmapped_ipr.hh
rename : src/arch/x86/mmaped_ipr.hh => src/arch/x86/mmapped_ipr.hh
This patch changes DataBlock.hh so that it is not dependent on RubySystem.
This dependence seems unecessary. All those functions that depende on
RubySystem have been moved to DataBlock.cc file.
Because int and not InstSeqNum was used in a couple of places, you can
overflow the int type and thus get wierd bugs when the sequence number
is negative (or some wierd value)
remove constructors that werent being used (it just gets confusing)
use initialization list for all the variables instead of relying on initVars()
function
-use a pointer to CacheReqPacket instead of PacketPtr so correct destructors
get called on packet deletion
- make sure to delete the packet if the cache blocks the sendTiming request
or for some reason we dont use the packet
- dont overwrite memory requests since in the worst case an instruction will
be replaying a request so no need to keep allocating a new request
- we dont use retryPkt so delete it
- fetch code was split out already, so just assert that this is a memory
reference inst. and that the staticInst is available
If there is an outstanding table walk and no other activity in the CPU
it can go to sleep and never wake up. This change makes the instruction
queue always active if the CPU is waiting for a store to translate.
If Gabe changes the way this code works then the below should be removed
as indicated by the todo.
keep track of when an instruction needs the execution
behind it to be serialized. Without this, in SE Mode
instructions can execute behind a system call exit().
resources don't need to call getLatency because the latency is already a member
in the class. If there is some type of special case where different instructions
impose a different latency inside a resource then we can revisit this and
add getLatency() back in
each resource has a certain # of requests it can take per cycle. update the #s here
to be more realistic based off of the pipeline width and if the resource needs to
be accessed on multiple cycles
---
need to delete the cache request's data on clearRequest() now that we are recycling
requests
---
fetch unit needs to deallocate the fetch buffer blocks when they are replaced or
squashed.
formerly, to free up bandwidth in a resource, we could just change the pointer in that resource
but at the same time the pipeline stages had visibility to see what happened to a resource request.
Now that we are recycling these requests (to avoid too much dynamic allocation), we can't throw
away the request too early or the pipeline stage gets bad information. Instead, mark when a request
is done with the resource all together and then let the pipeline stage call back to the resource
that it's time to free up the bandwidth for more instructions
*** inteface notes ***
- When an instruction completes and is done in a resource for that cycle, call done()
- When an instruction fails and is done with a resource for that cycle, call done(false)
- When an instruction completes, but isnt finished with a resource, call completed()
- When an instruction fails, but isnt finished with a resource, call completed(false)
* * *
inorder: tlbmiss wakeup bug fix
take away all instances of reqMap in the code and make all references use the built-in
request vectors inside of each resource. The request map was dynamically allocating
a request per instruction. The request vector just allocates N number of requests
during instantiation and then the surrounding code is fixed up to reuse those N requests
***
setRequest() and clearRequest() are the new accessors needed to define a new
request in a resource
we are going to be getting away from creating new resource requests for every
instruction so no more need to keep track of a reqRemoveList and clean it up
every tick
first change in an optimization that will stop InOrder from allocating new memory for every instruction's
request to a resource. This gets expensive since every instruction needs to access ~10 requests before
graduation. Instead, the plan is to allocate just enough resource request objects to satisfy each resource's
bandwidth (e.g. the execution unit would need to allocate 3 resource request objects for a 1-issue pipeline
since on any given cycle it could have 2 read requests and 1 write request) and then let the instructions
contend and reuse those allocated requests. The end result is a smaller memory footprint for the InOrder model
and increased simulation performance
resource skeds are divided into two parts: front end (all insts) and back end (inst. specific)
each of those are implemented as separate lists, so this iterator wraps around
the traditional list iterator so that an instruction can walk it's schedule but seamlessly
transfer from front end to back end when necessary
add a stage scheduler class to replace InstStage in pipeline_traits.cc
use that class to define a default front-end, resource schedule that all
instructions will follow. This will also replace the back end schedule in
pipeline_traits.cc. The reason for adding this is so that we can cache
instruction schedules in the future instead of calling the same function
over/over again as well as constantly dynamically alllocating memory on
every instruction to try to figure out it's schedule
When a table walk is initiated by the fetch stage, the CPU can
potentially move to the idle state and never wake up.
The fetch stage must call cpu->wakeCPU() when a translation completes
(in finishTranslation()).
This change fixes an issue where a DTLB fault occurs and redirects fetch to
handle the fault and the ITLB requires a walk which delays translation. In this
case the status of the cpu isn't updated appropriately, and an additional
instruction fetch occurs. Eventually this hits an assert as multiple instruction
fetches are occuring in the system and when the second one returns the
processor is in the wrong state.
Some asserts below are removed because it was always true (typo) and the state
after the initiateAcc() the processor could be in any valid state when a
d-side fault occurs.
Some ISAs (like ARM) relies on hardware page table walkers. For those ISAs,
when a TLB miss occurs, initiateTranslation() can return with NoFault but with
the translation unfinished.
Instructions experiencing a delayed translation due to a hardware page table
walk are deferred until the translation completes and kept into the IQ. In
order to keep track of them, the IQ has been augmented with a queue of the
outstanding delayed memory instructions. When their translation completes,
instructions are re-executed (only their initiateAccess() was already
executed; their DTB translation is now skipped). The IEW stage has been
modified to support such a 2-pass execution.
In sendSplitData, keep a pointer to the senderState that may be updated after
the call to handle*Packet. This way, if the receiver updates the packet
senderState, it can still be accessed in sendSplitData.
Maintain all information about an instruction's fault in the DynInst object rather
than any cpu-request object. Also, if there is a fault during the execution stage
then just save the fault inside the instruction and trap once the instruction
tries to graduate
Give fetch unit it's own parameterizable fetch buffer to read from. Very inefficient
(architecturally and in simulation) to continually fetch at the granularity of the
wordsize. As expected, the number of fetch memory requests drops dramatically
instead of having one cache-unit class be responsible for both data and code
accesses, separate code that is just for fetch in it's own derived class off the
original base class. This makes the code easier to manage as well as handle
future cases of special fetch handling
allow the user to specify how many instructions a pipeline stage can process
on any given cycle (stageWidth...i.e.bandwidth) by setting the parameter through
the python interface rather than compile the code after changing the *.cc file.
(we always had the parameter there, but still used the static 'ThePipeline::StageWidth'
instead)
-
Since StageWidth is now dynamically defined, change the interstage communication
structure to use a vector and get rid of array and array handling index (toNextStageIndex)
since we can just make calls to the list for the same information
use skidbuffer as only location for instructions between stages. before,
we had the insts queue from the prior stage and the skidbuffer for the
current stage, but that gets confusing and this consolidation helps
when handling squash cases
manage insertion and deletion like a queue but will need
access to internal elements for future changes
Currently, skidbuffer manages any instruction that was
in a stage but could not complete processing, however
we will want to manage all blocked instructions (from prev stage
and from cur. stage) in just one buffer.
Previous code was marking CPU activity on almost every cycle due to a bug in
tracking the status of pipeline stages. This disables the CPU from sleeping
on long latency stalls and increases simulation time
This makes sure that the address ranges requested for caches and uncached ports
don't conflict with each other, and that accesses which are always uncached
(message signaled interrupts for instance) don't waste time passing through
caches.
Without this change 0 is always used for the youngest sequence number if
a squash occured and the ROB was empty (E.g. an instruction is marked
serializeAfter or a fetch stall prevents other instructions from issuing).
Using 0 there is a race to rename where an instruction that committed the
same cycle as the squashing instruction can have it's renamed state undone
by the squash using sequence number 0.
I'm not positive this is the correct fix, but it's working right now.
Either we need to do something like this, prevent the misc reg from being renamed at all,
or there something else going on. We need to find the root cause as to why
this is only a problem sometimes.
The squash inside the fetch unit should not attempt to remove them from the
branch predictor as non-control instructions are not pushed into the predictor.
When this condition occurs the cpu should restart the fetch stage to fetch from
the original execution path. Fault handling in the commit stage is cleaned up a
little bit so the control flow is simplier. Finally, if an instruction is being
used to carry a fault it isn't executed, so the fault propagates appropriately.
There were several copies of similar functions that looked
like they all replicated reschedule(), so I replaced them
with direct calls. Keeping this separate from the previous
cset since there may be some subtle functional differences
if the code ever reschedules an event that is scheduled but
not squashed (though none were detected in the regressions).
Events need to be scheduled on the queue assigned
to the SimObject, not on the global queue (which
should be going away).
Also cleaned up a number of redundant expressions
that made the code unnecessarily verbose.
These files really aren't general enough to belong in src/base.
This patch doesn't reorder include lines, leaving them unsorted
in many cases, but Nate's magic script will fix that up shortly.
--HG--
rename : src/base/sched_list.hh => src/cpu/sched_list.hh
rename : src/base/timebuf.hh => src/cpu/timebuf.hh
Ran all the source files through 'perl -pi' with this script:
s|\s*(};?\s*)?/\*\s*(end\s*)?namespace\s*(\S+)\s*\*/(\s*})?|} // namespace $3|;
s|\s*};?\s*//\s*(end\s*)?namespace\s*(\S+)\s*|} // namespace $2\n|;
s|\s*};?\s*//\s*(\S+)\s*namespace\s*|} // namespace $1\n|;
Also did a little manual editing on some of the arch/*/isa_traits.hh files
and src/SConscript.
The store queue doesn't need to be ISA specific and architectures can
frequently store more than an int registers worth of data. A 128 bits seems
more common, but even 256 bits may be appropriate. Pretty much anything less
than a cache line size is buildable.
For SPARC ASIs are added to the ExtMachInst. If the ASI is changed simply
marking the instruction as Serializing isn't enough beacuse that only
stops rename. This provides a mechanism to squash all the instructions
and refetch them
ARM instructions updating cumulative flags (ARM FP exceptions and saturation
flags) are not serialized.
Added aliases for ARM FP exceptions and saturation flags in FPSCR. Removed
write accesses to the FP condition codes for most ARM VFP instructions: only
VCMP and VCMPE instructions update the FP condition codes. Removed a potential
cause of seg. faults in the O3 model for NEON memory macro-ops (ARM).
This change makes O3 flatten floating point destination registers, and also
fixes misc register flattening so that it's correctly repositioned relative to
the resized regions for integer and floating point indices.
It also fixes some overly long lines.
In the case of a split transaction and a cache that is faster than a CPU we
could get two responses before next_tick expires. Add an event that is
scheduled in this case and return false rather than asserting.
This change modifies the way prefetches work. They are now like normal loads
that don't writeback a register. Previously prefetches were supposed to call
prefetch() on the exection context, so they executed with execute() methods
instead of initiateAcc() completeAcc(). The prefetch() methods for all the CPUs
are blank, meaning that they get executed, but don't actually do anything.
On Alpha dead cache copy code was removed and prefetches are now normal ops.
They count as executed operations, but still don't do anything and IsMemRef is
not longer set on them.
On ARM IsDataPrefetch or IsInstructionPreftech is now set on all prefetch
instructions. The timing simple CPU doesn't try to do anything special for
prefetches now and they execute with the normal memory code path.
This change is a low level and pervasive reorganization of how PCs are managed
in M5. Back when Alpha was the only ISA, there were only 2 PCs to worry about,
the PC and the NPC, and the lsb of the PC signaled whether or not you were in
PAL mode. As other ISAs were added, we had to add an NNPC, micro PC and next
micropc, x86 and ARM introduced variable length instruction sets, and ARM
started to keep track of mode bits in the PC. Each CPU model handled PCs in
its own custom way that needed to be updated individually to handle the new
dimensions of variability, or, in the case of ARMs mode-bit-in-the-pc hack,
the complexity could be hidden in the ISA at the ISA implementation's expense.
Areas like the branch predictor hadn't been updated to handle branch delay
slots or micropcs, and it turns out that had introduced a significant (10s of
percent) performance bug in SPARC and to a lesser extend MIPS. Rather than
perpetuate the problem by reworking O3 again to handle the PC features needed
by x86, this change was introduced to rework PC handling in a more modular,
transparent, and hopefully efficient way.
PC type:
Rather than having the superset of all possible elements of PC state declared
in each of the CPU models, each ISA defines its own PCState type which has
exactly the elements it needs. A cross product of canned PCState classes are
defined in the new "generic" ISA directory for ISAs with/without delay slots
and microcode. These are either typedef-ed or subclassed by each ISA. To read
or write this structure through a *Context, you use the new pcState() accessor
which reads or writes depending on whether it has an argument. If you just
want the address of the current or next instruction or the current micro PC,
you can get those through read-only accessors on either the PCState type or
the *Contexts. These are instAddr(), nextInstAddr(), and microPC(). Note the
move away from readPC. That name is ambiguous since it's not clear whether or
not it should be the actual address to fetch from, or if it should have extra
bits in it like the PAL mode bit. Each class is free to define its own
functions to get at whatever values it needs however it needs to to be used in
ISA specific code. Eventually Alpha's PAL mode bit could be moved out of the
PC and into a separate field like ARM.
These types can be reset to a particular pc (where npc = pc +
sizeof(MachInst), nnpc = npc + sizeof(MachInst), upc = 0, nupc = 1 as
appropriate), printed, serialized, and compared. There is a branching()
function which encapsulates code in the CPU models that checked if an
instruction branched or not. Exactly what that means in the context of branch
delay slots which can skip an instruction when not taken is ambiguous, and
ideally this function and its uses can be eliminated. PCStates also generally
know how to advance themselves in various ways depending on if they point at
an instruction, a microop, or the last microop of a macroop. More on that
later.
Ideally, accessing all the PCs at once when setting them will improve
performance of M5 even though more data needs to be moved around. This is
because often all the PCs need to be manipulated together, and by getting them
all at once you avoid multiple function calls. Also, the PCs of a particular
thread will have spatial locality in the cache. Previously they were grouped
by element in arrays which spread out accesses.
Advancing the PC:
The PCs were previously managed entirely by the CPU which had to know about PC
semantics, try to figure out which dimension to increment the PC in, what to
set NPC/NNPC, etc. These decisions are best left to the ISA in conjunction
with the PC type itself. Because most of the information about how to
increment the PC (mainly what type of instruction it refers to) is contained
in the instruction object, a new advancePC virtual function was added to the
StaticInst class. Subclasses provide an implementation that moves around the
right element of the PC with a minimal amount of decision making. In ISAs like
Alpha, the instructions always simply assign NPC to PC without having to worry
about micropcs, nnpcs, etc. The added cost of a virtual function call should
be outweighed by not having to figure out as much about what to do with the
PCs and mucking around with the extra elements.
One drawback of making the StaticInsts advance the PC is that you have to
actually have one to advance the PC. This would, superficially, seem to
require decoding an instruction before fetch could advance. This is, as far as
I can tell, realistic. fetch would advance through memory addresses, not PCs,
perhaps predicting new memory addresses using existing ones. More
sophisticated decisions about control flow would be made later on, after the
instruction was decoded, and handed back to fetch. If branching needs to
happen, some amount of decoding needs to happen to see that it's a branch,
what the target is, etc. This could get a little more complicated if that gets
done by the predecoder, but I'm choosing to ignore that for now.
Variable length instructions:
To handle variable length instructions in x86 and ARM, the predecoder now
takes in the current PC by reference to the getExtMachInst function. It can
modify the PC however it needs to (by setting NPC to be the PC + instruction
length, for instance). This could be improved since the CPU doesn't know if
the PC was modified and always has to write it back.
ISA parser:
To support the new API, all PC related operand types were removed from the
parser and replaced with a PCState type. There are two warts on this
implementation. First, as with all the other operand types, the PCState still
has to have a valid operand type even though it doesn't use it. Second, using
syntax like PCS.npc(target) doesn't work for two reasons, this looks like the
syntax for operand type overriding, and the parser can't figure out if you're
reading or writing. Instructions that use the PCS operand (which I've
consistently called it) need to first read it into a local variable,
manipulate it, and then write it back out.
Return address stack:
The return address stack needed a little extra help because, in the presence
of branch delay slots, it has to merge together elements of the return PC and
the call PC. To handle that, a buildRetPC utility function was added. There
are basically only two versions in all the ISAs, but it didn't seem short
enough to put into the generic ISA directory. Also, the branch predictor code
in O3 and InOrder were adjusted so that they always store the PC of the actual
call instruction in the RAS, not the next PC. If the call instruction is a
microop, the next PC refers to the next microop in the same macroop which is
probably not desirable. The buildRetPC function advances the PC intelligently
to the next macroop (in an ISA specific way) so that that case works.
Change in stats:
There were no change in stats except in MIPS and SPARC in the O3 model. MIPS
runs in about 9% fewer ticks. SPARC runs with 30%-50% fewer ticks, which could
likely be improved further by setting call/return instruction flags and taking
advantage of the RAS.
TODO:
Add != operators to the PCState classes, defined trivially to be !(a==b).
Smooth out places where PCs are split apart, passed around, and put back
together later. I think this might happen in SPARC's fault code. Add ISA
specific constructors that allow setting PC elements without calling a bunch
of accessors. Try to eliminate the need for the branching() function. Factor
out Alpha's PAL mode pc bit into a separate flag field, and eliminate places
where it's blindly masked out or tested in the PC.
This reduces the scope of those includes and makes it less likely for there to
be a dependency loop. This also moves the hashing functions associated with
ExtMachInst objects to be with the ExtMachInst definitions and out of
utility.hh.
This code is no longer needed because of the preceeding change which adds a
StaticInstPtr parameter to the fault's invoke method, obviating the only use
for this pair of functions.
Also move the "Fault" reference counted pointer type into a separate file,
sim/fault.hh. It would be better to name this less similarly to sim/faults.hh
to reduce confusion, but fault.hh matches the name of the type. We could change
Fault to FaultPtr to match other pointer types, and then changing the name of
the file would make more sense.
Don't assert that the response packet is marked as a response
since it won't always be so for functional accesses.
Also cleanup code to refer to functional accesses rather
than "probes" (old terminology), and mention in the
DPRINTF which type of access we're doing.
When decoding a srs instruction, invalid mode encoding returns invalid instruction.
This can happen when garbage instructions are fetched from mispredicted path
Since miscellaneous registers bypass wakeup logic, force serialization
to resolve data dependencies through them
* * *
ARM: adding non-speculative/serialize flags for instructions change CPSR
