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.
Code in the CPUs that need a nop to carry a fault can't easily deal with a
microcoded nop. This instruction format provides for one that isn't.
--HG--
rename : src/arch/x86/isa/formats/syscall.isa => src/arch/x86/isa/formats/nop.isa
In the process make add skipFuction() to handle isa specific function skipping
instead of ifdefs and other ugliness. For almost all ABIs, 64 bit arguments can
only start in even registers. Size is now passed to getArgument() so that 32
bit systems can make decisions about register selection for 64 bit arguments.
The number argument is now passed by reference because getArgument() will need
to change it based on the size of the argument and the current argument number.
For ARM, if the argument number is odd and a 64-bit register is requested the
number must first be incremented to because all 64 bit arguments are passed
in an even argument register. Then the number will be incremented again to
access both halves of the argument.
This reduces the scope of those includes and makes it less likely for there to
be a dependency loop. This also moves the hashing functions associated with
ExtMachInst objects to be with the ExtMachInst definitions and out of
utility.hh.
Also move the "Fault" reference counted pointer type into a separate file,
sim/fault.hh. It would be better to name this less similarly to sim/faults.hh
to reduce confusion, but fault.hh matches the name of the type. We could change
Fault to FaultPtr to match other pointer types, and then changing the name of
the file would make more sense.
When decoding a srs instruction, invalid mode encoding returns invalid instruction.
This can happen when garbage instructions are fetched from mispredicted path
This is to help tidy up arch/x86. These files should not be used external to
the ISA.
--HG--
rename : src/arch/x86/apicregs.hh => src/arch/x86/regs/apic.hh
rename : src/arch/x86/floatregs.hh => src/arch/x86/regs/float.hh
rename : src/arch/x86/intregs.hh => src/arch/x86/regs/int.hh
rename : src/arch/x86/miscregs.hh => src/arch/x86/regs/misc.hh
rename : src/arch/x86/segmentregs.hh => src/arch/x86/regs/segment.hh
This single parameter replaces the collection of bools that set up various
flavors of microops. A flag parameter also allows other flags to be set like
the serialize before/after flags, etc., without having to change the
constructor.
This allows one two different OS requirements for the same ISA to be handled.
Some OSes are compiled for a virtual address and need to be loaded into physical
memory that starts at address 0, while other bare metal tools generate
images that start at address 0.
Replace direct call to unserialize() on each SimObject with a pair of
calls for better control over initialization in both ckpt and non-ckpt
cases.
If restoring from a checkpoint, loadState(ckpt) is called on each
SimObject. The default implementation simply calls unserialize() if
there is a corresponding checkpoint section, so we get backward
compatibility for existing objects. However, objects can override
loadState() to get other behaviors, e.g., doing other programmed
initializations after unserialize(), or complaining if no checkpoint
section is found. (Note that the default warning for a missing
checkpoint section is now gone.)
If not restoring from a checkpoint, we call the new initState() method
on each SimObject instead. This provides a hook for state
initializations that are only required when *not* restoring from a
checkpoint.
Given this new framework, do some cleanup of LiveProcess subclasses
and X86System, which were (in some cases) emulating initState()
behavior in startup via a local flag or (in other cases) erroneously
doing initializations in startup() that clobbered state loaded earlier
by unserialize().
When doing an unsigned 64 bit division with a divisor that has its most
significant bit set, the division code would spill a bit off of the end of a
uint64_t trying to shift the dividend into position. This change adds code
that handles that case specially by purposefully letting it spill and then
going ahead assuming there was a 65th one bit.
When each load or store is sent to the LSQ, we check whether it will cross a
cache line boundary and, if so, split it in two. This creates two TLB
translations and two memory requests. Care has to be taken if the first
packet of a split load is sent but the second blocks the cache. Similarly,
for a store, if the first packet cannot be sent, we must store the second
one somewhere to retry later.
This modifies the LSQSenderState class to record both packets in a split
load or store.
Finally, a new const variable, HasUnalignedMemAcc, is added to each ISA
to indicate whether unaligned memory accesses are allowed. This is used
throughout the changed code so that compiler can optimise away code dealing
with split requests for ISAs that don't need them.
Some of the micro-ops weren't casting 1 to ULL before shifting,
which can cause problems. On the perl makerand input this
caused some values to be negative that shouldn't have been.
The casts are done as ULL(1) instead of 1ULL to match others
in the m5 code base.
Unfortunately my implementation of the movd instruction had two bugs.
In one case, when moving a 32-bit value into an xmm register, the
lower half of the xmm register was not zero extended.
The other case is that xmm was used instead of xmmlm as the source
for a register move. My test case didn't notice this at first
as it moved xmm0 to eax, which both have the same register
number.
This double cast led to rounding errors which caused
some benchmarks to get the wrong values, most notably lucas
which failed spectacularly due to CVTTSD2SI returning an
off-by-one value. equake was also broken.
This problem is like the one fixed with movhpd a few weeks ago.
A +8 displacement is used to access memory when there should
be none.
This fix is needed for the perlbmk spec2k benchmark to run.
64-bit vsyscall is different than 32-bit.
There are only two syscalls, time and gettimeofday.
On a real system, there is complicated code that implements these
without entering the kernel. That would be complicated to implement in m5.
Instead we just place code that calls the regular syscalls (this is how
tools such as valgrind handle this case).
This is needed for the perlbmk spec2k benchmark.
These are complicated instructions and the micro-code might be suboptimal.
This has been tested with some small sample programs (attached)
The psrldq instruction is needed by various spec2k programs.
This patch implements the movd_Vo_Edp series of instructions.
It addresses various concerns by Gabe Black about which file the
instruction belonged in, as well as supporting REX prefixed
instructions properly.
This instruction is needed for some of the spec2k benchmarks, most
notably bzip2.
This patch implements the haddpd instruction.
It fixes the problem in the previous version (pointed out by Gabe Black)
where an incorrect result would happen if you issue the instruction
with the same argument twice, i.e. "haddpd %xmm0,%xmm0"
This instruction is used by many spec2k benchmarks.
This patch hooks up the truncate, ftruncate, truncate64 and ftruncate64
system calls on 32-bit and 64-bit X86.
These have been tested on both architectures.
ftruncate/ftruncate64 is needed for the f90 spec2k benchmarks.
When accessing arguments for a syscall, the position of an argument depends on
the policies of the ISA, how much space preceding arguments took up, and the
"alignment" of the index for this particular argument into the number of
possible storate locations. This change adjusts getSyscallArg to take its
index parameter by reference instead of value and to adjust it to point to the
possible location of the next argument on the stack, basically just after the
current one. This way, the rules for the new argument can be applied locally
without knowing about other arguments since those have already been taken into
account implicitly.
All system calls have also been changed to reflect the new interface. In a
number of cases this made the implementation clearer since it encourages
arguments to be collected in one place in order and then used as necessary
later, as opposed to scattering them throughout the function or using them in
place in long expressions. It also discourages using getSyscallArg over and
over to retrieve the same value when a temporary would do the job.
The movdqa instruction should enforce 16-byte alignment.
This implementation does not do that.
These instructions are needed for most of x86_64 spec2k to run.
The st_size entry was in the wrong place
(see linux-2.6.29/arch/x86/include/asm/stat.h )
Also, the packed attribute is needed when compiling on a
64-bit machine, otherwise gcc adds extra padding that
break the layout of the structure.
I've tested these on x86 and they work as expected.
In theory for 32-bit x86 we should have some sort of special
handling for the legacy 16-bit uid/gid syscalls, but in practice
modern toolchains don't use the 16-bit versions, and m5 sets the uid
and gid values to be less than 16-bits anyway.
This fix is needed for the perl spec2k benchmarks to run.
Get rid of misc.py and just stick misc things in __init__.py
Move utility functions out of SCons files and into m5.util
Move utility type stuff from m5/__init__.py to m5/util/__init__.py
Remove buildEnv from m5 and allow access only from m5.defines
Rename AddToPath to addToPath while we're moving it to m5.util
Rename read_command to readCommand while we're moving it
Rename compare_versions to compareVersions while we're moving it.
--HG--
rename : src/python/m5/convert.py => src/python/m5/util/convert.py
rename : src/python/m5/smartdict.py => src/python/m5/util/smartdict.py
The manuals from both AMD and Intel say that when writing to a 32 bit
destination in 64 bit mode, the upper 32 bits of the register are filled with
zeros. They also both say that the CMOV instructions leave their destination
alone when their condition fails. Unfortunately, it seems that CMOV will zero
extend its destination register whether or not it was supposed to actually do
a move on both platforms. This seems to be the only case where this happens,
but it would be hard to say for sure.
