2005-05-27 05:30:12 +02:00
|
|
|
/*
|
2013-01-07 19:05:44 +01:00
|
|
|
* Copyright (c) 2010-2012 ARM Limited
|
2010-12-08 01:19:57 +01:00
|
|
|
* All rights reserved
|
|
|
|
*
|
|
|
|
* The license below extends only to copyright in the software and shall
|
|
|
|
* not be construed as granting a license to any other intellectual
|
|
|
|
* property including but not limited to intellectual property relating
|
|
|
|
* to a hardware implementation of the functionality of the software
|
|
|
|
* licensed hereunder. You may use the software subject to the license
|
|
|
|
* terms below provided that you ensure that this notice is replicated
|
|
|
|
* unmodified and in its entirety in all distributions of the software,
|
|
|
|
* modified or unmodified, in source code or in binary form.
|
|
|
|
*
|
2006-05-16 20:06:35 +02:00
|
|
|
* Copyright (c) 2004-2006 The Regents of The University of Michigan
|
2005-05-27 05:30:12 +02:00
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
* modification, are permitted provided that the following conditions are
|
|
|
|
* met: redistributions of source code must retain the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer;
|
|
|
|
* redistributions in binary form must reproduce the above copyright
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
* documentation and/or other materials provided with the distribution;
|
|
|
|
* neither the name of the copyright holders nor the names of its
|
|
|
|
* contributors may be used to endorse or promote products derived from
|
|
|
|
* this software without specific prior written permission.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
|
|
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
|
|
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
|
|
|
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
|
|
|
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
|
|
|
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
|
|
|
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
|
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
|
|
|
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
|
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
|
|
|
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
2006-06-01 01:26:56 +02:00
|
|
|
*
|
|
|
|
* Authors: Kevin Lim
|
2006-07-27 00:47:06 +02:00
|
|
|
* Korey Sewell
|
2005-05-27 05:30:12 +02:00
|
|
|
*/
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
#include <algorithm>
|
2013-02-15 23:40:08 +01:00
|
|
|
#include <set>
|
2006-05-19 21:53:17 +02:00
|
|
|
#include <string>
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-11-08 19:55:48 +01:00
|
|
|
#include "arch/utility.hh"
|
2006-04-23 00:26:48 +02:00
|
|
|
#include "base/loader/symtab.hh"
|
2011-04-15 19:44:06 +02:00
|
|
|
#include "base/cp_annotate.hh"
|
2009-09-23 17:34:21 +02:00
|
|
|
#include "config/the_isa.hh"
|
2012-03-09 15:59:27 +01:00
|
|
|
#include "cpu/checker/cpu.hh"
|
2006-04-23 00:26:48 +02:00
|
|
|
#include "cpu/o3/commit.hh"
|
|
|
|
#include "cpu/o3/thread_state.hh"
|
2012-01-10 01:08:20 +01:00
|
|
|
#include "cpu/base.hh"
|
2011-04-15 19:44:06 +02:00
|
|
|
#include "cpu/exetrace.hh"
|
|
|
|
#include "cpu/timebuf.hh"
|
2011-04-15 19:44:32 +02:00
|
|
|
#include "debug/Activity.hh"
|
|
|
|
#include "debug/Commit.hh"
|
|
|
|
#include "debug/CommitRate.hh"
|
2013-01-07 19:05:46 +01:00
|
|
|
#include "debug/Drain.hh"
|
2011-04-15 19:44:32 +02:00
|
|
|
#include "debug/ExecFaulting.hh"
|
2013-02-15 23:40:09 +01:00
|
|
|
#include "debug/O3PipeView.hh"
|
2009-05-26 18:23:13 +02:00
|
|
|
#include "params/DerivO3CPU.hh"
|
2011-04-15 19:44:14 +02:00
|
|
|
#include "sim/faults.hh"
|
2011-11-18 10:33:28 +01:00
|
|
|
#include "sim/full_system.hh"
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
using namespace std;
|
2008-08-11 21:22:16 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
template <class Impl>
|
|
|
|
DefaultCommit<Impl>::TrapEvent::TrapEvent(DefaultCommit<Impl> *_commit,
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID _tid)
|
2011-09-23 03:59:55 +02:00
|
|
|
: Event(CPU_Tick_Pri, AutoDelete), commit(_commit), tid(_tid)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::TrapEvent::process()
|
|
|
|
{
|
2006-05-16 20:06:35 +02:00
|
|
|
// This will get reset by commit if it was switched out at the
|
|
|
|
// time of this event processing.
|
2006-04-23 00:26:48 +02:00
|
|
|
commit->trapSquash[tid] = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
const char *
|
2008-02-06 22:32:40 +01:00
|
|
|
DefaultCommit<Impl>::TrapEvent::description() const
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
2007-07-01 02:45:58 +02:00
|
|
|
return "Trap";
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
2008-08-11 21:22:16 +02:00
|
|
|
DefaultCommit<Impl>::DefaultCommit(O3CPU *_cpu, DerivO3CPUParams *params)
|
2007-04-04 21:38:59 +02:00
|
|
|
: cpu(_cpu),
|
|
|
|
squashCounter(0),
|
2006-04-23 00:26:48 +02:00
|
|
|
iewToCommitDelay(params->iewToCommitDelay),
|
|
|
|
commitToIEWDelay(params->commitToIEWDelay),
|
|
|
|
renameToROBDelay(params->renameToROBDelay),
|
|
|
|
fetchToCommitDelay(params->commitToFetchDelay),
|
|
|
|
renameWidth(params->renameWidth),
|
|
|
|
commitWidth(params->commitWidth),
|
2008-08-11 21:22:16 +02:00
|
|
|
numThreads(params->numThreads),
|
2006-07-06 19:59:02 +02:00
|
|
|
drainPending(false),
|
2012-02-10 15:37:30 +01:00
|
|
|
trapLatency(params->trapLatency),
|
2013-02-15 23:40:07 +01:00
|
|
|
canHandleInterrupts(true),
|
|
|
|
avoidQuiesceLiveLock(false)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
_status = Active;
|
|
|
|
_nextStatus = Inactive;
|
2006-08-15 11:07:15 +02:00
|
|
|
std::string policy = params->smtCommitPolicy;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
//Convert string to lowercase
|
|
|
|
std::transform(policy.begin(), policy.end(), policy.begin(),
|
|
|
|
(int(*)(int)) tolower);
|
|
|
|
|
|
|
|
//Assign commit policy
|
|
|
|
if (policy == "aggressive"){
|
|
|
|
commitPolicy = Aggressive;
|
|
|
|
|
2011-06-11 04:15:32 +02:00
|
|
|
DPRINTF(Commit,"Commit Policy set to Aggressive.\n");
|
2006-04-23 00:26:48 +02:00
|
|
|
} else if (policy == "roundrobin"){
|
|
|
|
commitPolicy = RoundRobin;
|
|
|
|
|
|
|
|
//Set-Up Priority List
|
2009-05-26 18:23:13 +02:00
|
|
|
for (ThreadID tid = 0; tid < numThreads; tid++) {
|
2006-04-23 00:26:48 +02:00
|
|
|
priority_list.push_back(tid);
|
|
|
|
}
|
|
|
|
|
2011-06-11 04:15:32 +02:00
|
|
|
DPRINTF(Commit,"Commit Policy set to Round Robin.\n");
|
2006-04-23 00:26:48 +02:00
|
|
|
} else if (policy == "oldestready"){
|
|
|
|
commitPolicy = OldestReady;
|
|
|
|
|
2007-04-04 21:38:59 +02:00
|
|
|
DPRINTF(Commit,"Commit Policy set to Oldest Ready.");
|
2006-04-23 00:26:48 +02:00
|
|
|
} else {
|
|
|
|
assert(0 && "Invalid SMT Commit Policy. Options Are: {Aggressive,"
|
|
|
|
"RoundRobin,OldestReady}");
|
|
|
|
}
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
for (ThreadID tid = 0; tid < numThreads; tid++) {
|
|
|
|
commitStatus[tid] = Idle;
|
|
|
|
changedROBNumEntries[tid] = false;
|
|
|
|
checkEmptyROB[tid] = false;
|
|
|
|
trapInFlight[tid] = false;
|
|
|
|
committedStores[tid] = false;
|
|
|
|
trapSquash[tid] = false;
|
|
|
|
tcSquash[tid] = false;
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
pc[tid].set(0);
|
2011-01-18 23:30:05 +01:00
|
|
|
lastCommitedSeqNum[tid] = 0;
|
2013-01-07 19:05:45 +01:00
|
|
|
squashAfterInst[tid] = NULL;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
2006-11-13 08:49:03 +01:00
|
|
|
interrupt = NoFault;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2006-04-23 00:26:48 +02:00
|
|
|
std::string
|
|
|
|
DefaultCommit<Impl>::name() const
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
2006-04-23 00:26:48 +02:00
|
|
|
return cpu->name() + ".commit";
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::regStats()
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
{
|
2006-04-24 23:06:00 +02:00
|
|
|
using namespace Stats;
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
commitSquashedInsts
|
|
|
|
.name(name() + ".commitSquashedInsts")
|
|
|
|
.desc("The number of squashed insts skipped by commit")
|
|
|
|
.prereq(commitSquashedInsts);
|
|
|
|
commitSquashEvents
|
|
|
|
.name(name() + ".commitSquashEvents")
|
|
|
|
.desc("The number of times commit is told to squash")
|
|
|
|
.prereq(commitSquashEvents);
|
|
|
|
commitNonSpecStalls
|
|
|
|
.name(name() + ".commitNonSpecStalls")
|
|
|
|
.desc("The number of times commit has been forced to stall to "
|
|
|
|
"communicate backwards")
|
|
|
|
.prereq(commitNonSpecStalls);
|
|
|
|
branchMispredicts
|
|
|
|
.name(name() + ".branchMispredicts")
|
|
|
|
.desc("The number of times a branch was mispredicted")
|
|
|
|
.prereq(branchMispredicts);
|
2006-04-23 00:26:48 +02:00
|
|
|
numCommittedDist
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.init(0,commitWidth,1)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".committed_per_cycle")
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
.desc("Number of insts commited each cycle")
|
|
|
|
.flags(Stats::pdf)
|
|
|
|
;
|
2006-04-24 23:06:00 +02:00
|
|
|
|
2012-02-12 23:07:39 +01:00
|
|
|
instsCommitted
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2012-02-12 23:07:39 +01:00
|
|
|
.name(name() + ".committedInsts")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("Number of instructions committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2012-02-12 23:07:39 +01:00
|
|
|
opsCommitted
|
|
|
|
.init(cpu->numThreads)
|
|
|
|
.name(name() + ".committedOps")
|
|
|
|
.desc("Number of ops (including micro ops) committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
statComSwp
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".swp_count")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("Number of s/w prefetches committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
statComRefs
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".refs")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("Number of memory references committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
statComLoads
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".loads")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("Number of loads committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
statComMembars
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".membars")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("Number of memory barriers committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
statComBranches
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".branches")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("Number of branches committed")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2011-02-07 07:14:17 +01:00
|
|
|
statComFloating
|
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".fp_insts")
|
2011-02-07 07:14:17 +01:00
|
|
|
.desc("Number of committed floating point instructions.")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
|
|
|
statComInteger
|
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name()+".int_insts")
|
2011-02-07 07:14:17 +01:00
|
|
|
.desc("Number of committed integer instructions.")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
|
|
|
statComFunctionCalls
|
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name()+".function_calls")
|
2011-02-07 07:14:17 +01:00
|
|
|
.desc("Number of function calls committed.")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
commitEligible
|
2009-05-26 18:23:13 +02:00
|
|
|
.init(cpu->numThreads)
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".bw_limited")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("number of insts not committed due to BW limits")
|
|
|
|
.flags(total)
|
|
|
|
;
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
commitEligibleSamples
|
2011-04-20 03:45:21 +02:00
|
|
|
.name(name() + ".bw_lim_events")
|
2006-04-24 23:06:00 +02:00
|
|
|
.desc("number cycles where commit BW limit reached")
|
|
|
|
;
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
2006-08-15 11:07:15 +02:00
|
|
|
DefaultCommit<Impl>::setThreads(std::vector<Thread *> &threads)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
thread = threads;
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::setTimeBuffer(TimeBuffer<TimeStruct> *tb_ptr)
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
timeBuffer = tb_ptr;
|
|
|
|
|
|
|
|
// Setup wire to send information back to IEW.
|
|
|
|
toIEW = timeBuffer->getWire(0);
|
|
|
|
|
|
|
|
// Setup wire to read data from IEW (for the ROB).
|
|
|
|
robInfoFromIEW = timeBuffer->getWire(-iewToCommitDelay);
|
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::setFetchQueue(TimeBuffer<FetchStruct> *fq_ptr)
|
|
|
|
{
|
|
|
|
fetchQueue = fq_ptr;
|
|
|
|
|
|
|
|
// Setup wire to get instructions from rename (for the ROB).
|
|
|
|
fromFetch = fetchQueue->getWire(-fetchToCommitDelay);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::setRenameQueue(TimeBuffer<RenameStruct> *rq_ptr)
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
renameQueue = rq_ptr;
|
|
|
|
|
|
|
|
// Setup wire to get instructions from rename (for the ROB).
|
|
|
|
fromRename = renameQueue->getWire(-renameToROBDelay);
|
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::setIEWQueue(TimeBuffer<IEWStruct> *iq_ptr)
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
iewQueue = iq_ptr;
|
|
|
|
|
|
|
|
// Setup wire to get instructions from IEW.
|
|
|
|
fromIEW = iewQueue->getWire(-iewToCommitDelay);
|
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::setIEWStage(IEW *iew_stage)
|
|
|
|
{
|
|
|
|
iewStage = iew_stage;
|
|
|
|
}
|
|
|
|
|
|
|
|
template<class Impl>
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
DefaultCommit<Impl>::setActiveThreads(list<ThreadID> *at_ptr)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
activeThreads = at_ptr;
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::setRenameMap(RenameMap rm_ptr[])
|
|
|
|
{
|
2009-05-26 18:23:13 +02:00
|
|
|
for (ThreadID tid = 0; tid < numThreads; tid++)
|
|
|
|
renameMap[tid] = &rm_ptr[tid];
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::setROB(ROB *rob_ptr)
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
rob = rob_ptr;
|
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2013-01-07 19:05:44 +01:00
|
|
|
DefaultCommit<Impl>::startupStage()
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
rob->setActiveThreads(activeThreads);
|
|
|
|
rob->resetEntries();
|
|
|
|
|
|
|
|
// Broadcast the number of free entries.
|
2009-05-26 18:23:13 +02:00
|
|
|
for (ThreadID tid = 0; tid < numThreads; tid++) {
|
|
|
|
toIEW->commitInfo[tid].usedROB = true;
|
|
|
|
toIEW->commitInfo[tid].freeROBEntries = rob->numFreeEntries(tid);
|
|
|
|
toIEW->commitInfo[tid].emptyROB = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
2007-04-04 21:38:59 +02:00
|
|
|
// Commit must broadcast the number of free entries it has at the
|
|
|
|
// start of the simulation, so it starts as active.
|
|
|
|
cpu->activateStage(O3CPU::CommitIdx);
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
cpu->activityThisCycle();
|
|
|
|
}
|
|
|
|
|
2006-05-04 17:36:20 +02:00
|
|
|
template <class Impl>
|
2013-01-07 19:05:46 +01:00
|
|
|
void
|
2006-07-06 19:59:02 +02:00
|
|
|
DefaultCommit<Impl>::drain()
|
2006-05-04 17:36:20 +02:00
|
|
|
{
|
2006-07-06 19:59:02 +02:00
|
|
|
drainPending = true;
|
2006-05-16 20:06:35 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
2013-01-07 19:05:46 +01:00
|
|
|
DefaultCommit<Impl>::drainResume()
|
2006-05-16 20:06:35 +02:00
|
|
|
{
|
2006-07-06 19:59:02 +02:00
|
|
|
drainPending = false;
|
2006-05-04 17:36:20 +02:00
|
|
|
}
|
|
|
|
|
2006-07-06 19:59:02 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
2013-01-07 19:05:46 +01:00
|
|
|
DefaultCommit<Impl>::drainSanityCheck() const
|
2006-07-06 19:59:02 +02:00
|
|
|
{
|
2013-01-07 19:05:46 +01:00
|
|
|
assert(isDrained());
|
|
|
|
rob->drainSanityCheck();
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
bool
|
|
|
|
DefaultCommit<Impl>::isDrained() const
|
|
|
|
{
|
|
|
|
/* Make sure no one is executing microcode. There are two reasons
|
|
|
|
* for this:
|
|
|
|
* - Hardware virtualized CPUs can't switch into the middle of a
|
|
|
|
* microcode sequence.
|
|
|
|
* - The current fetch implementation will most likely get very
|
|
|
|
* confused if it tries to start fetching an instruction that
|
|
|
|
* is executing in the middle of a ucode sequence that changes
|
|
|
|
* address mappings. This can happen on for example x86.
|
|
|
|
*/
|
|
|
|
for (ThreadID tid = 0; tid < numThreads; tid++) {
|
|
|
|
if (pc[tid].microPC() != 0)
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Make sure that all instructions have finished committing before
|
|
|
|
* declaring the system as drained. We want the pipeline to be
|
|
|
|
* completely empty when we declare the CPU to be drained. This
|
|
|
|
* makes debugging easier since CPU handover and restoring from a
|
|
|
|
* checkpoint with a different CPU should have the same timing.
|
|
|
|
*/
|
|
|
|
return rob->isEmpty() &&
|
|
|
|
interrupt == NoFault;
|
2006-07-06 19:59:02 +02:00
|
|
|
}
|
|
|
|
|
2006-05-04 17:36:20 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::takeOverFrom()
|
|
|
|
{
|
|
|
|
_status = Active;
|
|
|
|
_nextStatus = Inactive;
|
2009-05-26 18:23:13 +02:00
|
|
|
for (ThreadID tid = 0; tid < numThreads; tid++) {
|
|
|
|
commitStatus[tid] = Idle;
|
|
|
|
changedROBNumEntries[tid] = false;
|
|
|
|
trapSquash[tid] = false;
|
|
|
|
tcSquash[tid] = false;
|
2013-01-07 19:05:45 +01:00
|
|
|
squashAfterInst[tid] = NULL;
|
2006-05-04 17:36:20 +02:00
|
|
|
}
|
|
|
|
squashCounter = 0;
|
|
|
|
rob->takeOverFrom();
|
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::updateStatus()
|
|
|
|
{
|
2006-05-16 20:06:35 +02:00
|
|
|
// reset ROB changed variable
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-12-21 07:20:11 +01:00
|
|
|
|
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-12-21 07:20:11 +01:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
changedROBNumEntries[tid] = false;
|
|
|
|
|
|
|
|
// Also check if any of the threads has a trap pending
|
|
|
|
if (commitStatus[tid] == TrapPending ||
|
|
|
|
commitStatus[tid] == FetchTrapPending) {
|
|
|
|
_nextStatus = Active;
|
|
|
|
}
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (_nextStatus == Inactive && _status == Active) {
|
|
|
|
DPRINTF(Activity, "Deactivating stage.\n");
|
2006-06-16 23:08:47 +02:00
|
|
|
cpu->deactivateStage(O3CPU::CommitIdx);
|
2006-04-23 00:26:48 +02:00
|
|
|
} else if (_nextStatus == Active && _status == Inactive) {
|
|
|
|
DPRINTF(Activity, "Activating stage.\n");
|
2006-06-16 23:08:47 +02:00
|
|
|
cpu->activateStage(O3CPU::CommitIdx);
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
_status = _nextStatus;
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::setNextStatus()
|
|
|
|
{
|
|
|
|
int squashes = 0;
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (commitStatus[tid] == ROBSquashing) {
|
|
|
|
squashes++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-06-13 01:04:42 +02:00
|
|
|
squashCounter = squashes;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
// If commit is currently squashing, then it will have activity for the
|
|
|
|
// next cycle. Set its next status as active.
|
|
|
|
if (squashCounter) {
|
|
|
|
_nextStatus = Active;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
bool
|
|
|
|
DefaultCommit<Impl>::changedROBEntries()
|
|
|
|
{
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (changedROBNumEntries[tid]) {
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
2009-05-26 18:23:13 +02:00
|
|
|
size_t
|
|
|
|
DefaultCommit<Impl>::numROBFreeEntries(ThreadID tid)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
return rob->numFreeEntries(tid);
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
DefaultCommit<Impl>::generateTrapEvent(ThreadID tid)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
DPRINTF(Commit, "Generating trap event for [tid:%i]\n", tid);
|
|
|
|
|
|
|
|
TrapEvent *trap = new TrapEvent(this, tid);
|
|
|
|
|
2012-08-28 20:30:31 +02:00
|
|
|
cpu->schedule(trap, cpu->clockEdge(trapLatency));
|
2007-03-23 18:13:10 +01:00
|
|
|
trapInFlight[tid] = true;
|
2011-08-19 22:08:07 +02:00
|
|
|
thread[tid]->trapPending = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
DefaultCommit<Impl>::generateTCEvent(ThreadID tid)
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
2007-03-23 18:13:10 +01:00
|
|
|
assert(!trapInFlight[tid]);
|
2006-06-06 23:32:21 +02:00
|
|
|
DPRINTF(Commit, "Generating TC squash event for [tid:%i]\n", tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-06-06 23:32:21 +02:00
|
|
|
tcSquash[tid] = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
DefaultCommit<Impl>::squashAll(ThreadID tid)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
|
|
|
// If we want to include the squashing instruction in the squash,
|
|
|
|
// then use one older sequence number.
|
|
|
|
// Hopefully this doesn't mess things up. Basically I want to squash
|
|
|
|
// all instructions of this thread.
|
|
|
|
InstSeqNum squashed_inst = rob->isEmpty() ?
|
2011-01-18 23:30:05 +01:00
|
|
|
lastCommitedSeqNum[tid] : rob->readHeadInst(tid)->seqNum - 1;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
// All younger instructions will be squashed. Set the sequence
|
|
|
|
// number as the youngest instruction in the ROB (0 in this case.
|
|
|
|
// Hopefully nothing breaks.)
|
2011-01-18 23:30:05 +01:00
|
|
|
youngestSeqNum[tid] = lastCommitedSeqNum[tid];
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
rob->squash(squashed_inst, tid);
|
|
|
|
changedROBNumEntries[tid] = true;
|
|
|
|
|
|
|
|
// Send back the sequence number of the squashed instruction.
|
|
|
|
toIEW->commitInfo[tid].doneSeqNum = squashed_inst;
|
|
|
|
|
|
|
|
// Send back the squash signal to tell stages that they should
|
|
|
|
// squash.
|
|
|
|
toIEW->commitInfo[tid].squash = true;
|
|
|
|
|
|
|
|
// Send back the rob squashing signal so other stages know that
|
|
|
|
// the ROB is in the process of squashing.
|
|
|
|
toIEW->commitInfo[tid].robSquashing = true;
|
|
|
|
|
2011-01-18 23:30:05 +01:00
|
|
|
toIEW->commitInfo[tid].mispredictInst = NULL;
|
2011-03-18 01:20:19 +01:00
|
|
|
toIEW->commitInfo[tid].squashInst = NULL;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
toIEW->commitInfo[tid].pc = pc[tid];
|
2006-05-16 20:06:35 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
DefaultCommit<Impl>::squashFromTrap(ThreadID tid)
|
2006-05-16 20:06:35 +02:00
|
|
|
{
|
|
|
|
squashAll(tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Squashing from trap, restarting at PC %s\n", pc[tid]);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
thread[tid]->trapPending = false;
|
2013-01-07 19:05:33 +01:00
|
|
|
thread[tid]->noSquashFromTC = false;
|
2007-03-23 18:13:10 +01:00
|
|
|
trapInFlight[tid] = false;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
trapSquash[tid] = false;
|
|
|
|
|
|
|
|
commitStatus[tid] = ROBSquashing;
|
|
|
|
cpu->activityThisCycle();
|
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
2009-05-26 18:23:13 +02:00
|
|
|
DefaultCommit<Impl>::squashFromTC(ThreadID tid)
|
2006-04-23 00:26:48 +02:00
|
|
|
{
|
2006-05-16 20:06:35 +02:00
|
|
|
squashAll(tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Squashing from TC, restarting at PC %s\n", pc[tid]);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2013-01-07 19:05:33 +01:00
|
|
|
thread[tid]->noSquashFromTC = false;
|
2006-04-23 00:26:48 +02:00
|
|
|
assert(!thread[tid]->trapPending);
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
commitStatus[tid] = ROBSquashing;
|
|
|
|
cpu->activityThisCycle();
|
|
|
|
|
2006-06-06 23:32:21 +02:00
|
|
|
tcSquash[tid] = false;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
2010-12-08 01:19:57 +01:00
|
|
|
template <class Impl>
|
|
|
|
void
|
2013-01-07 19:05:45 +01:00
|
|
|
DefaultCommit<Impl>::squashFromSquashAfter(ThreadID tid)
|
2010-12-08 01:19:57 +01:00
|
|
|
{
|
2013-01-07 19:05:45 +01:00
|
|
|
DPRINTF(Commit, "Squashing after squash after request, "
|
|
|
|
"restarting at PC %s\n", pc[tid]);
|
2010-12-08 01:19:57 +01:00
|
|
|
|
2013-01-07 19:05:45 +01:00
|
|
|
squashAll(tid);
|
|
|
|
// Make sure to inform the fetch stage of which instruction caused
|
|
|
|
// the squash. It'll try to re-fetch an instruction executing in
|
|
|
|
// microcode unless this is set.
|
|
|
|
toIEW->commitInfo[tid].squashInst = squashAfterInst[tid];
|
|
|
|
squashAfterInst[tid] = NULL;
|
2010-12-08 01:19:57 +01:00
|
|
|
|
2013-01-07 19:05:45 +01:00
|
|
|
commitStatus[tid] = ROBSquashing;
|
|
|
|
cpu->activityThisCycle();
|
|
|
|
}
|
2010-12-08 01:19:57 +01:00
|
|
|
|
2013-01-07 19:05:45 +01:00
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::squashAfter(ThreadID tid, DynInstPtr &head_inst)
|
|
|
|
{
|
2010-12-08 01:19:57 +01:00
|
|
|
DPRINTF(Commit, "Executing squash after for [tid:%i] inst [sn:%lli]\n",
|
2013-01-07 19:05:45 +01:00
|
|
|
tid, head_inst->seqNum);
|
|
|
|
|
|
|
|
assert(!squashAfterInst[tid] || squashAfterInst[tid] == head_inst);
|
|
|
|
commitStatus[tid] = SquashAfterPending;
|
|
|
|
squashAfterInst[tid] = head_inst;
|
2010-12-08 01:19:57 +01:00
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::tick()
|
|
|
|
{
|
|
|
|
wroteToTimeBuffer = false;
|
|
|
|
_nextStatus = Inactive;
|
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
if (activeThreads->empty())
|
2006-07-07 10:06:26 +02:00
|
|
|
return;
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Check if any of the threads are done squashing. Change the
|
|
|
|
// status if they are done.
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
// Clear the bit saying if the thread has committed stores
|
|
|
|
// this cycle.
|
|
|
|
committedStores[tid] = false;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
if (commitStatus[tid] == ROBSquashing) {
|
|
|
|
|
|
|
|
if (rob->isDoneSquashing(tid)) {
|
|
|
|
commitStatus[tid] = Running;
|
|
|
|
} else {
|
|
|
|
DPRINTF(Commit,"[tid:%u]: Still Squashing, cannot commit any"
|
2006-07-07 21:58:03 +02:00
|
|
|
" insts this cycle.\n", tid);
|
2006-06-13 01:04:42 +02:00
|
|
|
rob->doSquash(tid);
|
|
|
|
toIEW->commitInfo[tid].robSquashing = true;
|
|
|
|
wroteToTimeBuffer = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
commit();
|
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
markCompletedInsts();
|
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
threads = activeThreads->begin();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (!rob->isEmpty(tid) && rob->readHeadInst(tid)->readyToCommit()) {
|
|
|
|
// The ROB has more instructions it can commit. Its next status
|
|
|
|
// will be active.
|
|
|
|
_nextStatus = Active;
|
|
|
|
|
|
|
|
DynInstPtr inst = rob->readHeadInst(tid);
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit,"[tid:%i]: Instruction [sn:%lli] PC %s is head of"
|
2006-04-23 00:26:48 +02:00
|
|
|
" ROB and ready to commit\n",
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
tid, inst->seqNum, inst->pcState());
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
} else if (!rob->isEmpty(tid)) {
|
|
|
|
DynInstPtr inst = rob->readHeadInst(tid);
|
|
|
|
|
|
|
|
DPRINTF(Commit,"[tid:%i]: Can't commit, Instruction [sn:%lli] PC "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"%s is head of ROB and not ready\n",
|
|
|
|
tid, inst->seqNum, inst->pcState());
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
DPRINTF(Commit, "[tid:%i]: ROB has %d insts & %d free entries.\n",
|
|
|
|
tid, rob->countInsts(tid), rob->numFreeEntries(tid));
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
if (wroteToTimeBuffer) {
|
2006-05-16 20:06:35 +02:00
|
|
|
DPRINTF(Activity, "Activity This Cycle.\n");
|
2006-04-23 00:26:48 +02:00
|
|
|
cpu->activityThisCycle();
|
|
|
|
}
|
|
|
|
|
|
|
|
updateStatus();
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2007-03-23 18:13:10 +01:00
|
|
|
DefaultCommit<Impl>::handleInterrupt()
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
2011-01-18 23:30:01 +01:00
|
|
|
// Verify that we still have an interrupt to handle
|
|
|
|
if (!cpu->checkInterrupts(cpu->tcBase(0))) {
|
|
|
|
DPRINTF(Commit, "Pending interrupt is cleared by master before "
|
|
|
|
"it got handled. Restart fetching from the orig path.\n");
|
|
|
|
toIEW->commitInfo[0].clearInterrupt = true;
|
|
|
|
interrupt = NoFault;
|
2013-02-15 23:40:07 +01:00
|
|
|
avoidQuiesceLiveLock = true;
|
2011-01-18 23:30:01 +01:00
|
|
|
return;
|
|
|
|
}
|
2006-11-13 02:15:30 +01:00
|
|
|
|
2011-08-09 12:37:43 +02:00
|
|
|
// Wait until all in flight instructions are finished before enterring
|
|
|
|
// the interrupt.
|
2012-02-10 15:37:30 +01:00
|
|
|
if (canHandleInterrupts && cpu->instList.empty()) {
|
2011-01-18 23:30:01 +01:00
|
|
|
// Squash or record that I need to squash this cycle if
|
|
|
|
// an interrupt needed to be handled.
|
|
|
|
DPRINTF(Commit, "Interrupt detected.\n");
|
2007-03-23 18:13:10 +01:00
|
|
|
|
2011-01-18 23:30:01 +01:00
|
|
|
// Clear the interrupt now that it's going to be handled
|
|
|
|
toIEW->commitInfo[0].clearInterrupt = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2013-01-07 19:05:33 +01:00
|
|
|
assert(!thread[0]->noSquashFromTC);
|
|
|
|
thread[0]->noSquashFromTC = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2012-01-31 16:46:03 +01:00
|
|
|
if (cpu->checker) {
|
|
|
|
cpu->checker->handlePendingInt();
|
|
|
|
}
|
|
|
|
|
2013-03-29 20:05:26 +01:00
|
|
|
// CPU will handle interrupt. Note that we ignore the local copy of
|
|
|
|
// interrupt. This is because the local copy may no longer be the
|
|
|
|
// interrupt that the interrupt controller thinks is being handled.
|
|
|
|
cpu->processInterrupts(cpu->getInterrupts());
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2013-01-07 19:05:33 +01:00
|
|
|
thread[0]->noSquashFromTC = false;
|
2006-11-13 02:15:30 +01:00
|
|
|
|
2011-01-18 23:30:01 +01:00
|
|
|
commitStatus[0] = TrapPending;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2011-01-18 23:30:01 +01:00
|
|
|
// Generate trap squash event.
|
|
|
|
generateTrapEvent(0);
|
|
|
|
|
|
|
|
interrupt = NoFault;
|
2013-02-15 23:40:07 +01:00
|
|
|
avoidQuiesceLiveLock = false;
|
2011-01-18 23:30:01 +01:00
|
|
|
} else {
|
2012-02-10 15:37:30 +01:00
|
|
|
DPRINTF(Commit, "Interrupt pending: instruction is %sin "
|
|
|
|
"flight, ROB is %sempty\n",
|
|
|
|
canHandleInterrupts ? "not " : "",
|
|
|
|
cpu->instList.empty() ? "" : "not " );
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
2007-03-23 18:13:10 +01:00
|
|
|
}
|
2011-01-18 23:30:01 +01:00
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::propagateInterrupt()
|
|
|
|
{
|
|
|
|
if (commitStatus[0] == TrapPending || interrupt || trapSquash[0] ||
|
|
|
|
tcSquash[0])
|
|
|
|
return;
|
|
|
|
|
|
|
|
// Process interrupts if interrupts are enabled, not in PAL
|
|
|
|
// mode, and no other traps or external squashes are currently
|
|
|
|
// pending.
|
|
|
|
// @todo: Allow other threads to handle interrupts.
|
|
|
|
|
|
|
|
// Get any interrupt that happened
|
|
|
|
interrupt = cpu->getInterrupts();
|
|
|
|
|
|
|
|
// Tell fetch that there is an interrupt pending. This
|
|
|
|
// will make fetch wait until it sees a non PAL-mode PC,
|
|
|
|
// at which point it stops fetching instructions.
|
|
|
|
if (interrupt != NoFault)
|
|
|
|
toIEW->commitInfo[0].interruptPending = true;
|
|
|
|
}
|
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::commit()
|
|
|
|
{
|
2011-11-18 10:33:28 +01:00
|
|
|
if (FullSystem) {
|
|
|
|
// Check if we have a interrupt and get read to handle it
|
|
|
|
if (cpu->checkInterrupts(cpu->tcBase(0)))
|
|
|
|
propagateInterrupt();
|
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
|
|
|
|
////////////////////////////////////
|
2006-05-16 20:06:35 +02:00
|
|
|
// Check for any possible squashes, handle them first
|
2004-08-20 20:54:07 +02:00
|
|
|
////////////////////////////////////
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// Not sure which one takes priority. I think if we have
|
|
|
|
// both, that's a bad sign.
|
|
|
|
if (trapSquash[tid] == true) {
|
2006-06-06 23:32:21 +02:00
|
|
|
assert(!tcSquash[tid]);
|
2006-04-23 00:26:48 +02:00
|
|
|
squashFromTrap(tid);
|
2006-06-06 23:32:21 +02:00
|
|
|
} else if (tcSquash[tid] == true) {
|
2007-03-23 18:13:10 +01:00
|
|
|
assert(commitStatus[tid] != TrapPending);
|
2006-06-06 23:32:21 +02:00
|
|
|
squashFromTC(tid);
|
2013-01-07 19:05:45 +01:00
|
|
|
} else if (commitStatus[tid] == SquashAfterPending) {
|
|
|
|
// A squash from the previous cycle of the commit stage (i.e.,
|
|
|
|
// commitInsts() called squashAfter) is pending. Squash the
|
|
|
|
// thread now.
|
|
|
|
squashFromSquashAfter(tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
// Squashed sequence number must be older than youngest valid
|
|
|
|
// instruction in the ROB. This prevents squashes from younger
|
|
|
|
// instructions overriding squashes from older instructions.
|
|
|
|
if (fromIEW->squash[tid] &&
|
|
|
|
commitStatus[tid] != TrapPending &&
|
|
|
|
fromIEW->squashedSeqNum[tid] <= youngestSeqNum[tid]) {
|
|
|
|
|
2011-03-18 01:20:19 +01:00
|
|
|
if (fromIEW->mispredictInst[tid]) {
|
|
|
|
DPRINTF(Commit,
|
|
|
|
"[tid:%i]: Squashing due to branch mispred PC:%#x [sn:%i]\n",
|
2006-04-23 00:26:48 +02:00
|
|
|
tid,
|
2011-03-18 01:20:19 +01:00
|
|
|
fromIEW->mispredictInst[tid]->instAddr(),
|
2006-04-23 00:26:48 +02:00
|
|
|
fromIEW->squashedSeqNum[tid]);
|
2011-03-18 01:20:19 +01:00
|
|
|
} else {
|
|
|
|
DPRINTF(Commit,
|
|
|
|
"[tid:%i]: Squashing due to order violation [sn:%i]\n",
|
|
|
|
tid, fromIEW->squashedSeqNum[tid]);
|
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Commit, "[tid:%i]: Redirecting to PC %#x\n",
|
|
|
|
tid,
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
fromIEW->pc[tid].nextInstAddr());
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
commitStatus[tid] = ROBSquashing;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// If we want to include the squashing instruction in the squash,
|
|
|
|
// then use one older sequence number.
|
|
|
|
InstSeqNum squashed_inst = fromIEW->squashedSeqNum[tid];
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
|
2006-07-23 19:39:42 +02:00
|
|
|
if (fromIEW->includeSquashInst[tid] == true) {
|
|
|
|
squashed_inst--;
|
|
|
|
}
|
2007-03-23 18:13:10 +01:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// All younger instructions will be squashed. Set the sequence
|
|
|
|
// number as the youngest instruction in the ROB.
|
|
|
|
youngestSeqNum[tid] = squashed_inst;
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-09-01 02:51:30 +02:00
|
|
|
rob->squash(squashed_inst, tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
changedROBNumEntries[tid] = true;
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
toIEW->commitInfo[tid].doneSeqNum = squashed_inst;
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
toIEW->commitInfo[tid].squash = true;
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// Send back the rob squashing signal so other stages know that
|
|
|
|
// the ROB is in the process of squashing.
|
|
|
|
toIEW->commitInfo[tid].robSquashing = true;
|
|
|
|
|
2011-01-18 23:30:05 +01:00
|
|
|
toIEW->commitInfo[tid].mispredictInst =
|
|
|
|
fromIEW->mispredictInst[tid];
|
2006-04-23 00:26:48 +02:00
|
|
|
toIEW->commitInfo[tid].branchTaken =
|
|
|
|
fromIEW->branchTaken[tid];
|
2012-02-10 15:37:28 +01:00
|
|
|
toIEW->commitInfo[tid].squashInst =
|
|
|
|
rob->findInst(tid, squashed_inst);
|
2012-02-13 19:26:24 +01:00
|
|
|
if (toIEW->commitInfo[tid].mispredictInst) {
|
|
|
|
if (toIEW->commitInfo[tid].mispredictInst->isUncondCtrl()) {
|
|
|
|
toIEW->commitInfo[tid].branchTaken = true;
|
|
|
|
}
|
|
|
|
}
|
2006-04-23 00:26:48 +02:00
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
toIEW->commitInfo[tid].pc = fromIEW->pc[tid];
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2011-03-18 01:20:19 +01:00
|
|
|
if (toIEW->commitInfo[tid].mispredictInst) {
|
2006-04-23 00:26:48 +02:00
|
|
|
++branchMispredicts;
|
|
|
|
}
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
}
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
setNextStatus();
|
|
|
|
|
|
|
|
if (squashCounter != numThreads) {
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
// If we're not currently squashing, then get instructions.
|
2004-08-20 20:54:07 +02:00
|
|
|
getInsts();
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
// Try to commit any instructions.
|
2004-08-20 20:54:07 +02:00
|
|
|
commitInsts();
|
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
//Check for any activity
|
2006-12-21 07:20:11 +01:00
|
|
|
threads = activeThreads->begin();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (changedROBNumEntries[tid]) {
|
|
|
|
toIEW->commitInfo[tid].usedROB = true;
|
|
|
|
toIEW->commitInfo[tid].freeROBEntries = rob->numFreeEntries(tid);
|
|
|
|
|
|
|
|
wroteToTimeBuffer = true;
|
|
|
|
changedROBNumEntries[tid] = false;
|
2007-03-23 18:13:10 +01:00
|
|
|
if (rob->isEmpty(tid))
|
|
|
|
checkEmptyROB[tid] = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
2007-03-23 18:13:10 +01:00
|
|
|
|
|
|
|
// ROB is only considered "empty" for previous stages if: a)
|
|
|
|
// ROB is empty, b) there are no outstanding stores, c) IEW
|
|
|
|
// stage has received any information regarding stores that
|
|
|
|
// committed.
|
|
|
|
// c) is checked by making sure to not consider the ROB empty
|
|
|
|
// on the same cycle as when stores have been committed.
|
|
|
|
// @todo: Make this handle multi-cycle communication between
|
|
|
|
// commit and IEW.
|
|
|
|
if (checkEmptyROB[tid] && rob->isEmpty(tid) &&
|
2008-09-26 16:44:07 +02:00
|
|
|
!iewStage->hasStoresToWB(tid) && !committedStores[tid]) {
|
2007-03-23 18:13:10 +01:00
|
|
|
checkEmptyROB[tid] = false;
|
|
|
|
toIEW->commitInfo[tid].usedROB = true;
|
|
|
|
toIEW->commitInfo[tid].emptyROB = true;
|
|
|
|
toIEW->commitInfo[tid].freeROBEntries = rob->numFreeEntries(tid);
|
|
|
|
wroteToTimeBuffer = true;
|
|
|
|
}
|
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::commitInsts()
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
////////////////////////////////////
|
|
|
|
// Handle commit
|
2006-05-16 20:06:35 +02:00
|
|
|
// Note that commit will be handled prior to putting new
|
|
|
|
// instructions in the ROB so that the ROB only tries to commit
|
|
|
|
// instructions it has in this current cycle, and not instructions
|
|
|
|
// it is writing in during this cycle. Can't commit and squash
|
|
|
|
// things at the same time...
|
2004-08-20 20:54:07 +02:00
|
|
|
////////////////////////////////////
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Commit, "Trying to commit instructions in the ROB.\n");
|
2004-08-20 20:54:07 +02:00
|
|
|
|
|
|
|
unsigned num_committed = 0;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
DynInstPtr head_inst;
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
// Commit as many instructions as possible until the commit bandwidth
|
|
|
|
// limit is reached, or it becomes impossible to commit any more.
|
2006-04-23 00:26:48 +02:00
|
|
|
while (num_committed < commitWidth) {
|
2012-02-10 15:37:30 +01:00
|
|
|
// Check for any interrupt that we've already squashed for
|
|
|
|
// and start processing it.
|
|
|
|
if (interrupt != NoFault)
|
|
|
|
handleInterrupt();
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
int commit_thread = getCommittingThread();
|
|
|
|
|
|
|
|
if (commit_thread == -1 || !rob->isHeadReady(commit_thread))
|
|
|
|
break;
|
|
|
|
|
|
|
|
head_inst = rob->readHeadInst(commit_thread);
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = head_inst->threadNumber;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
assert(tid == commit_thread);
|
|
|
|
|
|
|
|
DPRINTF(Commit, "Trying to commit head instruction, [sn:%i] [tid:%i]\n",
|
|
|
|
head_inst->seqNum, tid);
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// If the head instruction is squashed, it is ready to retire
|
|
|
|
// (be removed from the ROB) at any time.
|
2004-08-20 20:54:07 +02:00
|
|
|
if (head_inst->isSquashed()) {
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Commit, "Retiring squashed instruction from "
|
2004-08-20 20:54:07 +02:00
|
|
|
"ROB.\n");
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
rob->retireHead(commit_thread);
|
2004-08-20 20:54:07 +02:00
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
++commitSquashedInsts;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// Record that the number of ROB entries has changed.
|
|
|
|
changedROBNumEntries[tid] = true;
|
2004-08-20 20:54:07 +02:00
|
|
|
} else {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
pc[tid] = head_inst->pcState();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
// Increment the total number of non-speculative instructions
|
|
|
|
// executed.
|
|
|
|
// Hack for now: it really shouldn't happen until after the
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
// commit is deemed to be successful, but this count is needed
|
|
|
|
// for syscalls.
|
2006-04-23 00:26:48 +02:00
|
|
|
thread[tid]->funcExeInst++;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
|
|
|
// Try to commit the head instruction.
|
|
|
|
bool commit_success = commitHead(head_inst, num_committed);
|
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
if (commit_success) {
|
2004-08-20 20:54:07 +02:00
|
|
|
++num_committed;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
changedROBNumEntries[tid] = true;
|
|
|
|
|
|
|
|
// Set the doneSeqNum to the youngest committed instruction.
|
|
|
|
toIEW->commitInfo[tid].doneSeqNum = head_inst->seqNum;
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2012-02-10 15:37:30 +01:00
|
|
|
if (tid == 0) {
|
|
|
|
canHandleInterrupts = (!head_inst->isDelayedCommit()) &&
|
|
|
|
((THE_ISA != ALPHA_ISA) ||
|
|
|
|
(!(pc[0].instAddr() & 0x3)));
|
|
|
|
}
|
|
|
|
|
2010-12-08 01:19:57 +01:00
|
|
|
// Updates misc. registers.
|
|
|
|
head_inst->updateMiscRegs();
|
|
|
|
|
2012-01-10 01:08:20 +01:00
|
|
|
cpu->traceFunctions(pc[tid].instAddr());
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
TheISA::advancePC(pc[tid], head_inst->staticInst);
|
2006-07-23 19:39:42 +02:00
|
|
|
|
2011-01-18 23:30:05 +01:00
|
|
|
// Keep track of the last sequence number commited
|
|
|
|
lastCommitedSeqNum[tid] = head_inst->seqNum;
|
|
|
|
|
2010-12-08 01:19:57 +01:00
|
|
|
// If this is an instruction that doesn't play nicely with
|
|
|
|
// others squash everything and restart fetch
|
|
|
|
if (head_inst->isSquashAfter())
|
2013-01-07 19:05:45 +01:00
|
|
|
squashAfter(tid, head_inst);
|
2010-12-08 01:19:57 +01:00
|
|
|
|
2013-01-07 19:05:46 +01:00
|
|
|
if (drainPending) {
|
|
|
|
DPRINTF(Drain, "Draining: %i:%s\n", tid, pc[tid]);
|
|
|
|
if (pc[tid].microPC() == 0 && interrupt == NoFault) {
|
|
|
|
squashAfter(tid, head_inst);
|
|
|
|
cpu->commitDrained(tid);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
int count = 0;
|
|
|
|
Addr oldpc;
|
2007-10-03 03:25:37 +02:00
|
|
|
// Debug statement. Checks to make sure we're not
|
|
|
|
// currently updating state while handling PC events.
|
2013-01-07 19:05:33 +01:00
|
|
|
assert(!thread[tid]->noSquashFromTC && !thread[tid]->trapPending);
|
2006-04-23 00:26:48 +02:00
|
|
|
do {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
oldpc = pc[tid].instAddr();
|
2007-10-03 03:25:37 +02:00
|
|
|
cpu->system->pcEventQueue.service(thread[tid]->getTC());
|
2006-04-23 00:26:48 +02:00
|
|
|
count++;
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
} while (oldpc != pc[tid].instAddr());
|
2006-04-23 00:26:48 +02:00
|
|
|
if (count > 1) {
|
2007-10-03 03:25:37 +02:00
|
|
|
DPRINTF(Commit,
|
|
|
|
"PC skip function event, stopping commit\n");
|
2006-04-23 00:26:48 +02:00
|
|
|
break;
|
|
|
|
}
|
2013-02-15 23:40:07 +01:00
|
|
|
|
|
|
|
// Check if an instruction just enabled interrupts and we've
|
|
|
|
// previously had an interrupt pending that was not handled
|
|
|
|
// because interrupts were subsequently disabled before the
|
|
|
|
// pipeline reached a place to handle the interrupt. In that
|
|
|
|
// case squash now to make sure the interrupt is handled.
|
|
|
|
//
|
|
|
|
// If we don't do this, we might end up in a live lock situation
|
|
|
|
if (!interrupt && avoidQuiesceLiveLock &&
|
|
|
|
(!head_inst->isMicroop() || head_inst->isLastMicroop()) &&
|
|
|
|
cpu->checkInterrupts(cpu->tcBase(0)))
|
|
|
|
squashAfter(tid, head_inst);
|
2004-08-20 20:54:07 +02:00
|
|
|
} else {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Unable to commit head instruction PC:%s "
|
2006-04-23 00:26:48 +02:00
|
|
|
"[tid:%i] [sn:%i].\n",
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
head_inst->pcState(), tid ,head_inst->seqNum);
|
2004-08-20 20:54:07 +02:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2005-01-12 00:52:29 +01:00
|
|
|
DPRINTF(CommitRate, "%i\n", num_committed);
|
2006-04-23 00:26:48 +02:00
|
|
|
numCommittedDist.sample(num_committed);
|
2006-05-04 17:36:20 +02:00
|
|
|
|
|
|
|
if (num_committed == commitWidth) {
|
2006-06-01 21:39:45 +02:00
|
|
|
commitEligibleSamples++;
|
2006-05-04 17:36:20 +02:00
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
bool
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::commitHead(DynInstPtr &head_inst, unsigned inst_num)
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
assert(head_inst);
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = head_inst->threadNumber;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// If the instruction is not executed yet, then it will need extra
|
|
|
|
// handling. Signal backwards that it should be executed.
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
if (!head_inst->isExecuted()) {
|
|
|
|
// Keep this number correct. We have not yet actually executed
|
|
|
|
// and committed this instruction.
|
2006-04-23 00:26:48 +02:00
|
|
|
thread[tid]->funcExeInst--;
|
|
|
|
|
|
|
|
if (head_inst->isNonSpeculative() ||
|
2006-05-31 17:45:02 +02:00
|
|
|
head_inst->isStoreConditional() ||
|
2006-04-23 00:26:48 +02:00
|
|
|
head_inst->isMemBarrier() ||
|
|
|
|
head_inst->isWriteBarrier()) {
|
2006-05-16 20:06:35 +02:00
|
|
|
|
|
|
|
DPRINTF(Commit, "Encountered a barrier or non-speculative "
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
"instruction [sn:%lli] at the head of the ROB, PC %s.\n",
|
|
|
|
head_inst->seqNum, head_inst->pcState());
|
2006-05-16 20:06:35 +02:00
|
|
|
|
2008-09-26 16:44:07 +02:00
|
|
|
if (inst_num > 0 || iewStage->hasStoresToWB(tid)) {
|
2006-04-23 00:26:48 +02:00
|
|
|
DPRINTF(Commit, "Waiting for all stores to writeback.\n");
|
|
|
|
return false;
|
|
|
|
}
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
toIEW->commitInfo[tid].nonSpecSeqNum = head_inst->seqNum;
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
|
|
|
|
// Change the instruction so it won't try to commit again until
|
|
|
|
// it is executed.
|
|
|
|
head_inst->clearCanCommit();
|
|
|
|
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
++commitNonSpecStalls;
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
return false;
|
|
|
|
} else if (head_inst->isLoad()) {
|
2008-09-26 16:44:07 +02:00
|
|
|
if (inst_num > 0 || iewStage->hasStoresToWB(tid)) {
|
2007-03-23 18:13:10 +01:00
|
|
|
DPRINTF(Commit, "Waiting for all stores to writeback.\n");
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(head_inst->uncacheable());
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "[sn:%lli]: Uncached load, PC %s.\n",
|
|
|
|
head_inst->seqNum, head_inst->pcState());
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
// Send back the non-speculative instruction's sequence
|
2006-05-16 20:06:35 +02:00
|
|
|
// number. Tell the lsq to re-execute the load.
|
2006-04-23 00:26:48 +02:00
|
|
|
toIEW->commitInfo[tid].nonSpecSeqNum = head_inst->seqNum;
|
|
|
|
toIEW->commitInfo[tid].uncached = true;
|
|
|
|
toIEW->commitInfo[tid].uncachedLoad = head_inst;
|
|
|
|
|
|
|
|
head_inst->clearCanCommit();
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
return false;
|
|
|
|
} else {
|
2006-04-23 00:26:48 +02:00
|
|
|
panic("Trying to commit un-executed instruction "
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
"of unknown type!\n");
|
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
if (head_inst->isThreadSync()) {
|
2006-04-23 00:26:48 +02:00
|
|
|
// Not handled for now.
|
2006-05-16 20:06:35 +02:00
|
|
|
panic("Thread sync instructions are not handled yet.\n");
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
// Check if the instruction caused a fault. If so, trap.
|
|
|
|
Fault inst_fault = head_inst->getFault();
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Stores mark themselves as completed.
|
2007-03-23 18:13:10 +01:00
|
|
|
if (!head_inst->isStore() && inst_fault == NoFault) {
|
2006-05-11 21:39:02 +02:00
|
|
|
head_inst->setCompleted();
|
|
|
|
}
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Use checker prior to updating anything due to traps or PC
|
|
|
|
// based events.
|
|
|
|
if (cpu->checker) {
|
2006-06-16 19:10:47 +02:00
|
|
|
cpu->checker->verify(head_inst);
|
2006-05-16 20:06:35 +02:00
|
|
|
}
|
|
|
|
|
2006-02-16 20:55:15 +01:00
|
|
|
if (inst_fault != NoFault) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Inst [sn:%lli] PC %s has a fault\n",
|
|
|
|
head_inst->seqNum, head_inst->pcState());
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2008-09-26 16:44:07 +02:00
|
|
|
if (iewStage->hasStoresToWB(tid) || inst_num > 0) {
|
2006-05-16 20:06:35 +02:00
|
|
|
DPRINTF(Commit, "Stores outstanding, fault must wait.\n");
|
|
|
|
return false;
|
|
|
|
}
|
2006-05-11 21:39:02 +02:00
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
head_inst->setCompleted();
|
|
|
|
|
2012-01-31 16:46:03 +01:00
|
|
|
if (cpu->checker) {
|
|
|
|
// Need to check the instruction before its fault is processed
|
2006-06-16 19:10:47 +02:00
|
|
|
cpu->checker->verify(head_inst);
|
2006-05-16 20:06:35 +02:00
|
|
|
}
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2013-01-07 19:05:33 +01:00
|
|
|
assert(!thread[tid]->noSquashFromTC);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Mark that we're in state update mode so that the trap's
|
|
|
|
// execution doesn't generate extra squashes.
|
2013-01-07 19:05:33 +01:00
|
|
|
thread[tid]->noSquashFromTC = true;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Execute the trap. Although it's slightly unrealistic in
|
|
|
|
// terms of timing (as it doesn't wait for the full timing of
|
|
|
|
// the trap event to complete before updating state), it's
|
|
|
|
// needed to update the state as soon as possible. This
|
|
|
|
// prevents external agents from changing any specific state
|
|
|
|
// that the trap need.
|
2010-09-20 11:46:42 +02:00
|
|
|
cpu->trap(inst_fault, tid, head_inst->staticInst);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Exit state update mode to avoid accidental updating.
|
2013-01-07 19:05:33 +01:00
|
|
|
thread[tid]->noSquashFromTC = false;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
commitStatus[tid] = TrapPending;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2011-02-23 22:10:49 +01:00
|
|
|
DPRINTF(Commit, "Committing instruction with fault [sn:%lli]\n",
|
|
|
|
head_inst->seqNum);
|
2007-03-23 18:13:10 +01:00
|
|
|
if (head_inst->traceData) {
|
2009-09-26 19:50:50 +02:00
|
|
|
if (DTRACE(ExecFaulting)) {
|
|
|
|
head_inst->traceData->setFetchSeq(head_inst->seqNum);
|
2012-02-12 23:07:39 +01:00
|
|
|
head_inst->traceData->setCPSeq(thread[tid]->numOp);
|
2009-09-26 19:50:50 +02:00
|
|
|
head_inst->traceData->dump();
|
|
|
|
}
|
2007-03-25 06:47:14 +02:00
|
|
|
delete head_inst->traceData;
|
2007-03-23 18:13:10 +01:00
|
|
|
head_inst->traceData = NULL;
|
|
|
|
}
|
|
|
|
|
2006-05-16 20:06:35 +02:00
|
|
|
// Generate trap squash event.
|
|
|
|
generateTrapEvent(tid);
|
2004-08-20 20:54:07 +02:00
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
2006-04-24 23:06:00 +02:00
|
|
|
updateComInstStats(head_inst);
|
Check in of various updates to the CPU. Mainly adds in stats, improves
branch prediction, and makes memory dependence work properly.
SConscript:
Added return address stack, tournament predictor.
cpu/base_cpu.cc:
Added debug break and print statements.
cpu/base_dyn_inst.cc:
cpu/base_dyn_inst.hh:
Comment out possibly unneeded variables.
cpu/beta_cpu/2bit_local_pred.cc:
2bit predictor no longer speculatively updates itself.
cpu/beta_cpu/alpha_dyn_inst.hh:
Comment formatting.
cpu/beta_cpu/alpha_full_cpu.hh:
Formatting
cpu/beta_cpu/alpha_full_cpu_builder.cc:
Added new parameters for branch predictors, and IQ parameters.
cpu/beta_cpu/alpha_full_cpu_impl.hh:
Register stats.
cpu/beta_cpu/alpha_params.hh:
Added parameters for IQ, branch predictors, and store sets.
cpu/beta_cpu/bpred_unit.cc:
Removed one class.
cpu/beta_cpu/bpred_unit.hh:
Add in RAS, stats. Changed branch predictor unit functionality
so that it holds a history of past branches so it can update, and also
hold a proper history of the RAS so it can be restored on branch
mispredicts.
cpu/beta_cpu/bpred_unit_impl.hh:
Added in stats, history of branches, RAS. Now bpred unit actually
modifies the instruction's predicted next PC.
cpu/beta_cpu/btb.cc:
Add in sanity checks.
cpu/beta_cpu/comm.hh:
Add in communication where needed, remove it where it's not.
cpu/beta_cpu/commit.hh:
cpu/beta_cpu/rename.hh:
cpu/beta_cpu/rename_impl.hh:
Add in stats.
cpu/beta_cpu/commit_impl.hh:
Stats, update what is sent back on branch mispredict.
cpu/beta_cpu/cpu_policy.hh:
Change the bpred unit being used.
cpu/beta_cpu/decode.hh:
cpu/beta_cpu/decode_impl.hh:
Stats.
cpu/beta_cpu/fetch.hh:
Stats, change squash so it can handle squashes from decode differently
than squashes from commit.
cpu/beta_cpu/fetch_impl.hh:
Add in stats. Change how a cache line is fetched. Update to work with
caches. Also have separate functions for different behavior if squash
is coming from decode vs commit.
cpu/beta_cpu/free_list.hh:
Remove some old comments.
cpu/beta_cpu/full_cpu.cc:
cpu/beta_cpu/full_cpu.hh:
Added function to remove instructions from back of instruction list
until a certain sequence number.
cpu/beta_cpu/iew.hh:
Stats, separate squashing behavior due to branches vs memory.
cpu/beta_cpu/iew_impl.hh:
Stats, separate squashing behavior for branches vs memory.
cpu/beta_cpu/inst_queue.cc:
Debug stuff
cpu/beta_cpu/inst_queue.hh:
Stats, change how mem dep unit works, debug stuff
cpu/beta_cpu/inst_queue_impl.hh:
Stats, change how mem dep unit works, debug stuff. Also add in
parameters that used to be hardcoded.
cpu/beta_cpu/mem_dep_unit.hh:
cpu/beta_cpu/mem_dep_unit_impl.hh:
Add in stats, change how memory dependence unit works. It now holds
the memory instructions that are waiting for their memory dependences
to resolve. It provides which instructions are ready directly to the
IQ.
cpu/beta_cpu/regfile.hh:
Fix up sanity checks.
cpu/beta_cpu/rename_map.cc:
Fix loop variable type.
cpu/beta_cpu/rob_impl.hh:
Remove intermediate DynInstPtr
cpu/beta_cpu/store_set.cc:
Add in debugging statements.
cpu/beta_cpu/store_set.hh:
Reorder function arguments to match the rest of the calls.
--HG--
extra : convert_revision : aabf9b1fecd1d743265dfc3b174d6159937c6f44
2004-10-22 00:02:36 +02:00
|
|
|
|
2011-11-18 10:33:28 +01:00
|
|
|
if (FullSystem) {
|
|
|
|
if (thread[tid]->profile) {
|
|
|
|
thread[tid]->profilePC = head_inst->instAddr();
|
|
|
|
ProfileNode *node = thread[tid]->profile->consume(
|
|
|
|
thread[tid]->getTC(), head_inst->staticInst);
|
2006-08-24 23:43:08 +02:00
|
|
|
|
2011-11-18 10:33:28 +01:00
|
|
|
if (node)
|
|
|
|
thread[tid]->profileNode = node;
|
|
|
|
}
|
|
|
|
if (CPA::available()) {
|
|
|
|
if (head_inst->isControl()) {
|
|
|
|
ThreadContext *tc = thread[tid]->getTC();
|
|
|
|
CPA::cpa()->swAutoBegin(tc, head_inst->nextInstAddr());
|
|
|
|
}
|
2009-02-27 01:29:17 +01:00
|
|
|
}
|
|
|
|
}
|
2011-08-19 22:08:05 +02:00
|
|
|
DPRINTF(Commit, "Committing instruction with [sn:%lli] PC %s\n",
|
|
|
|
head_inst->seqNum, head_inst->pcState());
|
2004-08-20 20:54:07 +02:00
|
|
|
if (head_inst->traceData) {
|
2006-04-23 00:26:48 +02:00
|
|
|
head_inst->traceData->setFetchSeq(head_inst->seqNum);
|
2012-02-12 23:07:39 +01:00
|
|
|
head_inst->traceData->setCPSeq(thread[tid]->numOp);
|
2007-02-11 00:14:50 +01:00
|
|
|
head_inst->traceData->dump();
|
|
|
|
delete head_inst->traceData;
|
2006-04-23 00:26:48 +02:00
|
|
|
head_inst->traceData = NULL;
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
2012-02-13 19:26:25 +01:00
|
|
|
if (head_inst->isReturn()) {
|
|
|
|
DPRINTF(Commit,"Return Instruction Committed [sn:%lli] PC %s \n",
|
|
|
|
head_inst->seqNum, head_inst->pcState());
|
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
// Update the commit rename map
|
|
|
|
for (int i = 0; i < head_inst->numDestRegs(); i++) {
|
2006-12-06 11:51:18 +01:00
|
|
|
renameMap[tid]->setEntry(head_inst->flattenedDestRegIdx(i),
|
2006-04-23 00:26:48 +02:00
|
|
|
head_inst->renamedDestRegIdx(i));
|
|
|
|
}
|
|
|
|
|
|
|
|
// Finally clear the head ROB entry.
|
|
|
|
rob->retireHead(tid);
|
2004-08-20 20:54:07 +02:00
|
|
|
|
2011-07-15 18:53:35 +02:00
|
|
|
#if TRACING_ON
|
2013-02-15 23:40:09 +01:00
|
|
|
if (DTRACE(O3PipeView)) {
|
|
|
|
head_inst->commitTick = curTick() - head_inst->fetchTick;
|
|
|
|
}
|
2011-07-15 18:53:35 +02:00
|
|
|
#endif
|
|
|
|
|
2007-03-23 18:13:10 +01:00
|
|
|
// If this was a store, record it for this cycle.
|
|
|
|
if (head_inst->isStore())
|
|
|
|
committedStores[tid] = true;
|
|
|
|
|
2004-08-20 20:54:07 +02:00
|
|
|
// Return true to indicate that we have committed an instruction.
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::getInsts()
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
2006-07-23 19:39:42 +02:00
|
|
|
DPRINTF(Commit, "Getting instructions from Rename stage.\n");
|
|
|
|
|
2006-09-01 02:51:30 +02:00
|
|
|
// Read any renamed instructions and place them into the ROB.
|
|
|
|
int insts_to_process = std::min((int)renameWidth, fromRename->size);
|
2006-07-27 00:47:06 +02:00
|
|
|
|
|
|
|
for (int inst_num = 0; inst_num < insts_to_process; ++inst_num) {
|
|
|
|
DynInstPtr inst;
|
|
|
|
|
2006-09-01 02:51:30 +02:00
|
|
|
inst = fromRename->insts[inst_num];
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = inst->threadNumber;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (!inst->isSquashed() &&
|
2007-03-23 18:13:10 +01:00
|
|
|
commitStatus[tid] != ROBSquashing &&
|
|
|
|
commitStatus[tid] != TrapPending) {
|
2006-04-23 00:26:48 +02:00
|
|
|
changedROBNumEntries[tid] = true;
|
|
|
|
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Inserting PC %s [sn:%i] [tid:%i] into ROB.\n",
|
|
|
|
inst->pcState(), inst->seqNum, tid);
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
rob->insertInst(inst);
|
|
|
|
|
|
|
|
assert(rob->getThreadEntries(tid) <= rob->getMaxEntries(tid));
|
|
|
|
|
|
|
|
youngestSeqNum[tid] = inst->seqNum;
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
} else {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Instruction PC %s [sn:%i] [tid:%i] was "
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
"squashed, skipping.\n",
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
inst->pcState(), inst->seqNum, tid);
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
2006-07-27 00:47:06 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::skidInsert()
|
|
|
|
{
|
|
|
|
DPRINTF(Commit, "Attempting to any instructions from rename into "
|
|
|
|
"skidBuffer.\n");
|
|
|
|
|
|
|
|
for (int inst_num = 0; inst_num < fromRename->size; ++inst_num) {
|
|
|
|
DynInstPtr inst = fromRename->insts[inst_num];
|
|
|
|
|
|
|
|
if (!inst->isSquashed()) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Inserting PC %s [sn:%i] [tid:%i] into ",
|
|
|
|
"skidBuffer.\n", inst->pcState(), inst->seqNum,
|
2006-10-08 06:53:41 +02:00
|
|
|
inst->threadNumber);
|
2006-07-27 00:47:06 +02:00
|
|
|
skidBuffer.push(inst);
|
|
|
|
} else {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "Instruction PC %s [sn:%i] [tid:%i] was "
|
2006-07-27 00:47:06 +02:00
|
|
|
"squashed, skipping.\n",
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
inst->pcState(), inst->seqNum, inst->threadNumber);
|
2006-07-27 00:47:06 +02:00
|
|
|
}
|
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
|
Update to make multiple instruction issue and different latencies work.
Also change to ref counted DynInst.
SConscript:
Add branch predictor, BTB, load store queue, and storesets.
arch/isa_parser.py:
Specify the template parameter for AlphaDynInst
base/traceflags.py:
Add load store queue, store set, and mem dependence unit to the
list of trace flags.
cpu/base_dyn_inst.cc:
Change formating, add in debug statement.
cpu/base_dyn_inst.hh:
Change DynInst to be RefCounted, add flag to clear whether or not this
instruction can commit. This is likely to be removed in the future.
cpu/beta_cpu/alpha_dyn_inst.cc:
AlphaDynInst has been changed to be templated, so now this CC file
is just used to force instantiations of AlphaDynInst.
cpu/beta_cpu/alpha_dyn_inst.hh:
Changed AlphaDynInst to be templated on Impl. Removed some unnecessary
functions.
cpu/beta_cpu/alpha_full_cpu.cc:
AlphaFullCPU has been changed to be templated, so this CC file is now
just used to force instantation of AlphaFullCPU.
cpu/beta_cpu/alpha_full_cpu.hh:
Change AlphaFullCPU to be templated on Impl.
cpu/beta_cpu/alpha_impl.hh:
Update it to reflect AlphaDynInst and AlphaFullCPU being templated
on Impl. Also removed time buffers from here, as they are really
a part of the CPU and are thus in the CPU policy now.
cpu/beta_cpu/alpha_params.hh:
Make AlphaSimpleParams inherit from the BaseFullCPU so that it doesn't
need to specifically declare any parameters that are already in the
BaseFullCPU.
cpu/beta_cpu/comm.hh:
Changed the structure of the time buffer communication structs. Now
they include the size of the packet of instructions it is sending.
Added some parameters to the backwards communication struct, mainly
for squashing.
cpu/beta_cpu/commit.hh:
Update typenames to reflect change in location of time buffer structs.
Update DynInst to DynInstPtr (it is refcounted now).
cpu/beta_cpu/commit_impl.hh:
Formatting changes mainly. Also sends back proper information
on branch mispredicts so that the bpred unit can update itself.
Updated behavior for non-speculative instructions (stores, any
other non-spec instructions): once they reach the head of the ROB,
the ROB signals back to the IQ that it can go ahead and issue the
non-speculative instruction. The instruction itself is updated so that
commit won't try to commit it again until it is done executing.
cpu/beta_cpu/cpu_policy.hh:
Added branch prediction unit, mem dependence prediction unit, load
store queue. Moved time buffer structs from AlphaSimpleImpl to here.
cpu/beta_cpu/decode.hh:
Changed typedefs to reflect change in location of time buffer structs
and also the change from DynInst to ref counted DynInstPtr.
cpu/beta_cpu/decode_impl.hh:
Continues to buffer instructions even while unblocking now. Changed
how it loops through groups of instructions so it can properly block
during the middle of a group of instructions.
cpu/beta_cpu/fetch.hh:
Changed typedefs to reflect change in location of time buffer structs
and the change to ref counted DynInsts. Also added in branch
brediction unit.
cpu/beta_cpu/fetch_impl.hh:
Add in branch prediction. Changed how fetch checks inputs and its
current state to make for easier logic.
cpu/beta_cpu/free_list.cc:
Changed int regs and float regs to logically use one flat namespace.
Future change will be moving them to a single scoreboard to conserve
space.
cpu/beta_cpu/free_list.hh:
Mostly debugging statements. Might be removed for performance in future.
cpu/beta_cpu/full_cpu.cc:
Added in some debugging statements. Updated BaseFullCPU to take
a params object.
cpu/beta_cpu/full_cpu.hh:
Added params class within BaseCPU that other param classes will be
able to inherit from. Updated typedefs to reflect change in location
of time buffer structs and ref counted DynInst.
cpu/beta_cpu/iew.hh:
Updated typedefs to reflect change in location of time buffer structs
and use of ref counted DynInsts.
cpu/beta_cpu/iew_impl.hh:
Added in load store queue, updated iew to be able to execute non-
speculative instructions, instead of having them execute in commit.
cpu/beta_cpu/inst_queue.hh:
Updated change to ref counted DynInsts. Changed inst queue to hold
non-speculative instructions as well, which are issued only when
commit signals backwards that a nonspeculative instruction is at
the head of the ROB.
cpu/beta_cpu/inst_queue_impl.hh:
Updated to allow for non-speculative instructions to be in the inst
queue. Also added some debug functions.
cpu/beta_cpu/regfile.hh:
Added debugging statements, changed formatting.
cpu/beta_cpu/rename.hh:
Updated typedefs, added some functions to clean up code.
cpu/beta_cpu/rename_impl.hh:
Moved some code into functions to make it easier to read.
cpu/beta_cpu/rename_map.cc:
Changed int and float reg behavior to use a single flat namespace. In
the future, the rename maps can be combined to a single rename map to
save space.
cpu/beta_cpu/rename_map.hh:
Added destructor.
cpu/beta_cpu/rob.hh:
Updated it with change from DynInst to ref counted DynInst.
cpu/beta_cpu/rob_impl.hh:
Formatting, updated to use ref counted DynInst.
cpu/static_inst.hh:
Updated forward declaration for AlphaDynInst now that it is templated.
--HG--
extra : convert_revision : 1045f240ee9b6a4bd368e1806aca029ebbdc6dd3
2004-09-23 20:06:03 +02:00
|
|
|
template <class Impl>
|
2004-08-20 20:54:07 +02:00
|
|
|
void
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::markCompletedInsts()
|
2004-08-20 20:54:07 +02:00
|
|
|
{
|
|
|
|
// Grab completed insts out of the IEW instruction queue, and mark
|
|
|
|
// instructions completed within the ROB.
|
|
|
|
for (int inst_num = 0;
|
2005-05-03 16:56:47 +02:00
|
|
|
inst_num < fromIEW->size && fromIEW->insts[inst_num];
|
2004-08-20 20:54:07 +02:00
|
|
|
++inst_num)
|
|
|
|
{
|
2006-04-23 00:26:48 +02:00
|
|
|
if (!fromIEW->insts[inst_num]->isSquashed()) {
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
DPRINTF(Commit, "[tid:%i]: Marking PC %s, [sn:%lli] ready "
|
2006-05-16 20:06:35 +02:00
|
|
|
"within ROB.\n",
|
2006-04-23 00:26:48 +02:00
|
|
|
fromIEW->insts[inst_num]->threadNumber,
|
ISA,CPU,etc: Create an ISA defined PC type that abstracts out ISA behaviors.
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.
2010-10-31 08:07:20 +01:00
|
|
|
fromIEW->insts[inst_num]->pcState(),
|
2006-04-23 00:26:48 +02:00
|
|
|
fromIEW->insts[inst_num]->seqNum);
|
|
|
|
|
|
|
|
// Mark the instruction as ready to commit.
|
|
|
|
fromIEW->insts[inst_num]->setCanCommit();
|
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
template <class Impl>
|
|
|
|
bool
|
|
|
|
DefaultCommit<Impl>::robDoneSquashing()
|
|
|
|
{
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (!rob->isDoneSquashing(tid))
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2006-04-24 23:06:00 +02:00
|
|
|
template <class Impl>
|
|
|
|
void
|
|
|
|
DefaultCommit<Impl>::updateComInstStats(DynInstPtr &inst)
|
|
|
|
{
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = inst->threadNumber;
|
2006-04-24 23:06:00 +02:00
|
|
|
|
2012-02-12 23:07:39 +01:00
|
|
|
if (!inst->isMicroop() || inst->isLastMicroop())
|
|
|
|
instsCommitted[tid]++;
|
|
|
|
opsCommitted[tid]++;
|
2006-04-24 23:06:00 +02:00
|
|
|
|
2012-09-12 17:35:52 +02:00
|
|
|
// To match the old model, don't count nops and instruction
|
|
|
|
// prefetches towards the total commit count.
|
|
|
|
if (!inst->isNop() && !inst->isInstPrefetch()) {
|
|
|
|
cpu->instDone(tid, inst);
|
|
|
|
}
|
|
|
|
|
2006-04-24 23:06:00 +02:00
|
|
|
//
|
|
|
|
// Control Instructions
|
|
|
|
//
|
|
|
|
if (inst->isControl())
|
2009-05-26 18:23:13 +02:00
|
|
|
statComBranches[tid]++;
|
2006-04-24 23:06:00 +02:00
|
|
|
|
|
|
|
//
|
|
|
|
// Memory references
|
|
|
|
//
|
|
|
|
if (inst->isMemRef()) {
|
2009-05-26 18:23:13 +02:00
|
|
|
statComRefs[tid]++;
|
2006-04-24 23:06:00 +02:00
|
|
|
|
|
|
|
if (inst->isLoad()) {
|
2009-05-26 18:23:13 +02:00
|
|
|
statComLoads[tid]++;
|
2006-04-24 23:06:00 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (inst->isMemBarrier()) {
|
2009-05-26 18:23:13 +02:00
|
|
|
statComMembars[tid]++;
|
2006-04-24 23:06:00 +02:00
|
|
|
}
|
2011-02-07 07:14:17 +01:00
|
|
|
|
|
|
|
// Integer Instruction
|
|
|
|
if (inst->isInteger())
|
|
|
|
statComInteger[tid]++;
|
|
|
|
|
|
|
|
// Floating Point Instruction
|
|
|
|
if (inst->isFloating())
|
|
|
|
statComFloating[tid]++;
|
|
|
|
|
|
|
|
// Function Calls
|
|
|
|
if (inst->isCall())
|
|
|
|
statComFunctionCalls[tid]++;
|
|
|
|
|
2006-04-24 23:06:00 +02:00
|
|
|
}
|
|
|
|
|
2006-04-23 00:26:48 +02:00
|
|
|
////////////////////////////////////////
|
|
|
|
// //
|
2006-05-16 20:06:35 +02:00
|
|
|
// SMT COMMIT POLICY MAINTAINED HERE //
|
2006-04-23 00:26:48 +02:00
|
|
|
// //
|
|
|
|
////////////////////////////////////////
|
|
|
|
template <class Impl>
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::getCommittingThread()
|
|
|
|
{
|
|
|
|
if (numThreads > 1) {
|
|
|
|
switch (commitPolicy) {
|
|
|
|
|
|
|
|
case Aggressive:
|
|
|
|
//If Policy is Aggressive, commit will call
|
|
|
|
//this function multiple times per
|
|
|
|
//cycle
|
|
|
|
return oldestReady();
|
|
|
|
|
|
|
|
case RoundRobin:
|
|
|
|
return roundRobin();
|
|
|
|
|
|
|
|
case OldestReady:
|
|
|
|
return oldestReady();
|
|
|
|
|
|
|
|
default:
|
2009-05-26 18:23:13 +02:00
|
|
|
return InvalidThreadID;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
} else {
|
2006-12-21 07:20:11 +01:00
|
|
|
assert(!activeThreads->empty());
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = activeThreads->front();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (commitStatus[tid] == Running ||
|
|
|
|
commitStatus[tid] == Idle ||
|
|
|
|
commitStatus[tid] == FetchTrapPending) {
|
|
|
|
return tid;
|
|
|
|
} else {
|
2009-05-26 18:23:13 +02:00
|
|
|
return InvalidThreadID;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
template<class Impl>
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::roundRobin()
|
|
|
|
{
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator pri_iter = priority_list.begin();
|
|
|
|
list<ThreadID>::iterator end = priority_list.end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
while (pri_iter != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *pri_iter;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (commitStatus[tid] == Running ||
|
2006-07-03 05:11:24 +02:00
|
|
|
commitStatus[tid] == Idle ||
|
|
|
|
commitStatus[tid] == FetchTrapPending) {
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (rob->isHeadReady(tid)) {
|
|
|
|
priority_list.erase(pri_iter);
|
|
|
|
priority_list.push_back(tid);
|
|
|
|
|
|
|
|
return tid;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
pri_iter++;
|
|
|
|
}
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
return InvalidThreadID;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
template<class Impl>
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID
|
2006-04-23 00:26:48 +02:00
|
|
|
DefaultCommit<Impl>::oldestReady()
|
|
|
|
{
|
|
|
|
unsigned oldest = 0;
|
|
|
|
bool first = true;
|
|
|
|
|
2009-05-26 18:23:13 +02:00
|
|
|
list<ThreadID>::iterator threads = activeThreads->begin();
|
|
|
|
list<ThreadID>::iterator end = activeThreads->end();
|
2006-04-23 00:26:48 +02:00
|
|
|
|
2006-12-21 07:20:11 +01:00
|
|
|
while (threads != end) {
|
2009-05-26 18:23:13 +02:00
|
|
|
ThreadID tid = *threads++;
|
2006-04-23 00:26:48 +02:00
|
|
|
|
|
|
|
if (!rob->isEmpty(tid) &&
|
|
|
|
(commitStatus[tid] == Running ||
|
|
|
|
commitStatus[tid] == Idle ||
|
|
|
|
commitStatus[tid] == FetchTrapPending)) {
|
|
|
|
|
|
|
|
if (rob->isHeadReady(tid)) {
|
|
|
|
|
|
|
|
DynInstPtr head_inst = rob->readHeadInst(tid);
|
|
|
|
|
|
|
|
if (first) {
|
|
|
|
oldest = tid;
|
|
|
|
first = false;
|
|
|
|
} else if (head_inst->seqNum < oldest) {
|
|
|
|
oldest = tid;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!first) {
|
|
|
|
return oldest;
|
|
|
|
} else {
|
2009-05-26 18:23:13 +02:00
|
|
|
return InvalidThreadID;
|
2006-04-23 00:26:48 +02:00
|
|
|
}
|
2004-08-20 20:54:07 +02:00
|
|
|
}
|