| Age | Commit message (Collapse) | Author |
|---|
|
The previous implementation did a pair of nested RMW operations,
which isn't compatible with the way that locked RMW operations are
implemented in the cache models. It was convenient though in that
it didn't require any new micro-ops, and supported cmpxchg16b using
64-bit memory ops. It also worked in AtomicSimpleCPU where
atomicity was guaranteed by the core and not by the memory system.
It did not work with timing CPU models though.
This new implementation defines new 'split' load and store micro-ops
which allow a single memory operation to use a pair of registers as
the source or destination, then uses a single ldsplit/stsplit RMW
pair to implement cmpxchg. This patch requires support for 128-bit
memory accesses in the ISA (added via a separate patch) to support
cmpxchg16b.
|
|
Although the cache models support wider accesses, the ISA descriptions
assume that (for the most part) memory operands are integer types,
which makes it difficult to define instructions that do memory accesses
larger than 64 bits.
This patch adds some generic support for memory operands that are arrays
of uint64_t, and specifically a 'u2qw' operand type for x86 that is an
array of 2 uint64_ts (128 bits). This support is unused at this point,
but will be needed shortly for cmpxchg16b. Ideally the 128-bit SSE
memory accesses will also be rewritten to use this support.
Support for 128-bit accesses could also have been added using the gcc
__int128_t extension, which would have been less disruptive. However,
although clang also supports __int128_t, it's still non-standard.
Also, more importantly, this approach creates a path to defining
256- and 512-byte operands as well, which will be useful for eventual
AVX support.
|
|
Result of running 'hg m5style --skip-all --fix-white -a'.
|
|
For historical reasons, the ExecContext interface had a single
function, readMem(), that did two different things depending on
whether the ExecContext supported atomic memory mode (i.e.,
AtomicSimpleCPU) or timing memory mode (all the other models).
In the former case, it actually performed a memory read; in the
latter case, it merely initiated a read access, and the read
completion did not happen until later when a response packet
arrived from the memory system.
This led to some confusing things, including timing accesses
being required to provide a pointer for the return data even
though that pointer was only used in atomic mode.
This patch splits this interface, adding a new initiateMemRead()
function to the ExecContext interface to replace the timing-mode
use of readMem().
For consistency and clarity, the readMemTiming() helper function
in the ISA definitions is renamed to initiateMemRead() as well.
For x86, where the access size is passed in explicitly, we can
also get rid of the data parameter at this level. For other ISAs,
where the access size is determined from the type of the data
parameter, we have to keep the parameter for that purpose.
|
|
These are packed single-precision approximate reciprocal operations,
vector and scalar versions, respectively.
This code was basically developed by copying the code for
sqrtps and sqrtss. The mrcp micro-op was simplified relative to
msqrt since there are no double-precision versions of this operation.
|
|
fild loads an integer value into the x87 top of stack register.
fucomi/fucomip compare two x87 register values (the latter
also doing a stack pop).
These instructions are used by some versions of GNU libstdc++.
|
|
Same exception is raised whether division with zero is performed or the
quotient is greater than the maximum value that the provided space can hold.
Divide-by-Zero is the AMD terminology, while Divide-Error is Intel's.
|
|
Makes x86-style locked operations even more distinct from
LLSC operations. Using "locked" by itself should be
obviously ambiguous now.
|
|
This patch takes quite a large step in transitioning from the ad-hoc
RefCountingPtr to the c++11 shared_ptr by adopting its use for all
Faults. There are no changes in behaviour, and the code modifications
are mostly just replacing "new" with "make_shared".
|
|
The o3 cpu relies upon instructions that suspend a thread context being
flagged as "IsQuiesce". If they are not, unpredictable behavior can occur.
This patch fixes that for the x86 ISA.
|
|
This patch sets op class of two fp instructions: movfp and pop x87 stack
as IntAluOp since these instructions do not make use of the fp alu.
|
|
This patch encompasses several interrelated and interdependent changes
to the ISA generation step. The end goal is to reduce the size of the
generated compilation units for instruction execution and decoding so
that batch compilation can proceed with all CPUs active without
exhausting physical memory.
The ISA parser (src/arch/isa_parser.py) has been improved so that it can
accept 'split [output_type];' directives at the top level of the grammar
and 'split(output_type)' python calls within 'exec {{ ... }}' blocks.
This has the effect of "splitting" the files into smaller compilation
units. I use air-quotes around "splitting" because the files themselves
are not split, but preprocessing directives are inserted to have the same
effect.
Architecturally, the ISA parser has had some changes in how it works.
In general, it emits code sooner. It doesn't generate per-CPU files,
and instead defers to the C preprocessor to create the duplicate copies
for each CPU type. Likewise there are more files emitted and the C
preprocessor does more substitution that used to be done by the ISA parser.
Finally, the build system (SCons) needs to be able to cope with a
dynamic list of source files coming out of the ISA parser. The changes
to the SCons{cript,truct} files support this. In broad strokes, the
targets requested on the command line are hidden from SCons until all
the build dependencies are determined, otherwise it would try, realize
it can't reach the goal, and terminate in failure. Since build steps
(i.e. running the ISA parser) must be taken to determine the file list,
several new build stages have been inserted at the very start of the
build. First, the build dependencies from the ISA parser will be emitted
to arch/$ISA/generated/inc.d, which is then read by a new SCons builder
to finalize the dependencies. (Once inc.d exists, the ISA parser will not
need to be run to complete this step.) Once the dependencies are known,
the 'Environments' are made by the makeEnv() function. This function used
to be called before the build began but now happens during the build.
It is easy to see that this step is quite slow; this is a known issue
and it's important to realize that it was already slow, but there was
no obvious cause to attribute it to since nothing was displayed to the
terminal. Since new steps that used to be performed serially are now in a
potentially-parallel build phase, the pathname handling in the SCons scripts
has been tightened up to deal with chdir() race conditions. In general,
pathnames are computed earlier and more likely to be stored, passed around,
and processed as absolute paths rather than relative paths. In the end,
some of these issues had to be fixed by inserting serializing dependencies
in the build.
Minor note:
For the null ISA, we just provide a dummy inc.d so SCons is never
compelled to try to generate it. While it seems slightly wrong to have
anything in src/arch/*/generated (i.e. a non-generated 'generated' file),
it's by far the simplest solution.
|
|
With (upcoming) separate compilation, they are useless. Only
link-time optimization could re-inline them, but ideally
feedback-directed optimization would choose to do so only for
profitable (i.e. common) instructions.
|
|
|
|
|
|
The x87 FPU supports three floating point formats: 32-bit, 64-bit, and
80-bit floats. The current gem5 implementation supports 32-bit and
64-bit floats, but only works correctly for 64-bit floats. This
changeset fixes the 32-bit float handling by correctly loading and
rounding (using truncation) 32-bit floats instead of simply truncating
the bit pattern.
80-bit floats are loaded by first loading the 80-bits of the float to
two temporary integer registers. A micro-op (cvtint_fp80) then
converts the contents of the two integer registers to the internal FP
representation (double). Similarly, when storing an 80-bit float,
there are two conversion routines (ctvfp80h_int and cvtfp80l_int) that
convert an internal FP register to 80-bit and stores the upper 64-bits
or lower 32-bits to an integer register, which is the written to
memory using normal integer stores.
|
|
X87 store instructions typically loads and pops the top value of the
stack and stores it in memory. The current implementation pops the
stack at the same time as the floating point value is loaded to a
temporary register. This will corrupt the state of the x87 stack if
the store fails. This changeset introduces a pop87 micro-instruction
that pops the stack and uses this instruction in the affected
macro-instructions to pop the stack after storing the value to memory.
|
|
The current implementation of the x87 never updates the x87 tag
word. This is currently not a big issue since the simulated x87 never
checks for stack overflows, however this becomes an issue when
switching between a virtualized CPU and a simulated CPU. This
changeset adds support, which is enabled by default, for updating the
tag register to every floating point microop that updates the stack
top using the spm mechanism.
The new tag words is generated by the helper function
X86ISA::genX87Tags(). This function is currently limited to flagging a
stack position as valid or invalid and does not try to distinguish
between the valid, zero, and special states.
|
|
This changeset actually fixes two issues:
* The lfpimm instruction didn't work correctly when applied to a
floating point constant (it did work for integers containing the
bit string representation of a constant) since it used
reinterpret_cast to convert a double to a uint64_t. This caused a
compilation error, at least, in gcc 4.6.3.
* The instructions loading floating point constants in the x87
processor didn't work correctly since they just stored a truncated
integer instead of a double in the floating point register. This
changeset fixes the old microcode by using lfpimm instruction
instead of the limm instructions.
|
|
The current implementation of fprem simply does an fmod and doesn't
simulate any of the iterative behavior in a real fprem. This isn't
normally a problem, however, it can lead to problems when switching
between CPU models. If switching from a real CPU in the middle of an
fprem loop to a simulated CPU, the output of the fprem loop becomes
correupted. This changeset changes the fprem implementation to work
like the one on real hardware.
|
|
This changeset fixes two problems in the FABS and FCHS
implementation. First, the ISA parser expects the assignment in
flag_code to be a pure assignment and not an and-assignment, which
leads to the isa_parser omitting the misc reg update. Second, the FCHS
and FABS macro-ops don't set the SetStatus flag, which means that the
default micro-op version, which doesn't update FSW, is executed.
|
|
Currently all the integer microops are marked as IntAluOp and the floating
point microops are marked as FloatAddOp. This patch adds support for marking
different microops differently. Now IntMultOp, IntDivOp, FloatDivOp,
FloatMultOp, FloatCvtOp, FloatSqrtOp classes will be used as well. This will
help in providing different latencies for different op class.
|
|
This patch implements ftan, fprem, fyl2x, fld* floating-point instructions.
|
|
|
|
|
|
This patch implements the fsincos instruction. The code was originally written
by Vince Weaver. Gabe had made some comments about the code, but those were
never addressed. This patch addresses those comments.
|
|
The patch introduces two predicates for condition code registers -- one
tests if a register needs to be read, the other tests whether a register
needs to be written to. These predicates are evaluated twice -- during
construction of the microop and during its execution. Register reads
and writes are elided depending on how the predicates evaluate.
|
|
The D flag bit is part of the cc flag bit register currently. But since it
is not being used any where in the implementation, it creates an unnecessary
dependency. Hence, it is being moved to a separate register.
|
|
This patch moves the ECF and EZF bits to individual registers (ecfBit and
ezfBit) and the CF and OF bits to cfofFlag registers. This is being done
so as to lower the read after write dependencies on the the condition code
register. Ultimately we will have the following registers [ZAPS], [OF],
[CF], [ECF], [EZF] and [DF]. Note that this is only one part of the
solution for lowering the dependencies. The other part will check whether
or not the condition code register needs to be actually read. This would
be done through a separate patch.
|
|
This patch addresses a number of minor issues that cause problems when
compiling with clang >= 3.0 and gcc >= 4.6. Most importantly, it
avoids using the deprecated ext/hash_map and instead uses
unordered_map (and similarly so for the hash_set). To make use of the
new STL containers, g++ and clang has to be invoked with "-std=c++0x",
and this is now added for all gcc versions >= 4.6, and for clang >=
3.0. For gcc >= 4.3 and <= 4.5 and clang <= 3.0 we use the tr1
unordered_map to avoid the deprecation warning.
The addition of c++0x in turn causes a few problems, as the
compiler is more stringent and adds a number of new warnings. Below,
the most important issues are enumerated:
1) the use of namespaces is more strict, e.g. for isnan, and all
headers opening the entire namespace std are now fixed.
2) another other issue caused by the more stringent compiler is the
narrowing of the embedded python, which used to be a char array,
and is now unsigned char since there were values larger than 128.
3) a particularly odd issue that arose with the new c++0x behaviour is
found in range.hh, where the operator< causes gcc to complain about
the template type parsing (the "<" is interpreted as the beginning
of a template argument), and the problem seems to be related to the
begin/end members introduced for the range-type iteration, which is
a new feature in c++11.
As a minor update, this patch also fixes the build flags for the clang
debug target that used to be shared with gcc and incorrectly use
"-ggdb".
|
|
Virtual (pre-segmentation) addresses are truncated based on address size, and
any non-64 bit linear address is truncated to 32 bits. This means that real
mode addresses aren't truncated down to 16 bits after their segment bases are
added in.
|
|
If an instruction is executed speculatively and hits a situation where it
wants to panic, it should return a fault instead. If the instruction was
misspeculated, the fault can be thrown away. If the instruction wasn't
misspeculated, the fault will be invoked and the panic will still happen.
|
|
--HG--
extra : rebase_source : 755f4f6eae52f88ed516a1f1ac9e2565725d89c1
|
|
This patch adds a new microop for memory barrier. The microop itself does
nothing, but since it is marked as a memory barrier, the O3 CPU should flush
all the pending loads and stores before the fence to the memory system.
|
|
And by "everything" I mean all the quick regressions.
|
|
By using an underscore, the "." is still available and can unambiguously be
used to refer to members of a structure if an operand is a structure, class,
etc. This change mostly just replaces the appropriate "."s with "_"s, but
there were also a few places where the ISA descriptions where handling the
extensions themselves and had their own regular expressions to update. The
regular expressions in the isa parser were updated as well. It also now
looks for one of the defined type extensions specifically after connecting "_"
where before it would look for any sequence of characters after a "."
following an operand name and try to use it as the extension. This helps to
disambiguate cases where a "_" may legitimately be part of an operand name but
not separate the name from the type suffix.
Because leaving the "_" and suffix on the variable name still leaves a valid
C++ identifier and all extensions need to be consistent in a given context, I
considered leaving them on as a breadcrumb that would show what the intended
type was for that operand. Unfortunately the operands can be referred to in
code templates, the Mem operand in particular, and since the exact type of Mem
can be different for different uses of the same template, that broke things.
|
|
|
|
If a fault was returned by the CPU when a store initiated it's write, the
store instruction would ignore the fault. This change fixes that.
|
|
|
|
|
|
|
|
This was making certain versions of gcc omit the function from the object file
which would break the build.
|
|
Using the destination register directly causes the ISA parser to treat it as a
source even if none of the original bits are used.
|
|
In x86, 32 and 64 bit writes to registers in which registers appear to be 32 or
64 bits wide overwrite all bits of the destination register. This change
removes false dependencies in these cases where the previous value of a
register doesn't need to be read to write a new value. New versions of most
microops are created that have a "Big" suffix which simply overwrite their
destination, and the right version to use is selected during microop
allocation based on the selected data size.
This does not change the performance of the O3 CPU model significantly, I
assume because there are other false dependencies from the condition code bits
in the flags register.
|
|
These faults can panic/warn/warn_once, etc., instead of instructions doing
that themselves directly. That way, instructions can be speculatively
executed, and only if they're actually going to commit will their fault be
invoked and the panic, etc., happen.
|
|
|
|
|
|
|
|
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.
|
|
|
