This patch adds a new flag named -bottleneck-analysis to print out information
about throughput bottlenecks.
MCA knows how to identify and classify dynamic dispatch stalls. However, it
doesn't know how to analyze and highlight kernel bottlenecks. The goal of this
patch is to teach MCA how to correlate increases in backend pressure to backend
stalls (and therefore, the loss of throughput).
From a Scheduler point of view, backend pressure is a function of the scheduler
buffer usage (i.e. how the number of uOps in the scheduler buffers changes over
time). Backend pressure increases (or decreases) when there is a mismatch
between the number of opcodes dispatched, and the number of opcodes issued in
the same cycle. Since buffer resources are limited, continuous increases in
backend pressure would eventually leads to dispatch stalls. So, there is a
strong correlation between dispatch stalls, and how backpressure changed over
time.
This patch teaches how to identify situations where backend pressure increases
due to:
- unavailable pipeline resources.
- data dependencies.
Data dependencies may delay execution of instructions and therefore increase the
time that uOps have to spend in the scheduler buffers. That often translates to
an increase in backend pressure which may eventually lead to a bottleneck.
Contention on pipeline resources may also delay execution of instructions, and
lead to a temporary increase in backend pressure.
Internally, the Scheduler classifies instructions based on whether register /
memory operands are available or not.
An instruction is marked as "ready to execute" only if data dependencies are
fully resolved.
Every cycle, the Scheduler attempts to execute all instructions that are ready
to execute. If an instruction cannot execute because of unavailable pipeline
resources, then the Scheduler internally updates a BusyResourceUnits mask with
the ID of each unavailable resource.
ExecuteStage is responsible for tracking changes in backend pressure. If backend
pressure increases during a cycle because of contention on pipeline resources,
then ExecuteStage sends a "backend pressure" event to the listeners.
That event would contain information about instructions delayed by resource
pressure, as well as the BusyResourceUnits mask.
Note that ExecuteStage also knows how to identify situations where backpressure
increased because of delays introduced by data dependencies.
The SummaryView observes "backend pressure" events and prints out a "bottleneck
report".
Example of bottleneck report:
```
Cycles with backend pressure increase [ 99.89% ]
Throughput Bottlenecks:
Resource Pressure [ 0.00% ]
Data Dependencies: [ 99.89% ]
- Register Dependencies [ 0.00% ]
- Memory Dependencies [ 99.89% ]
```
A bottleneck report is printed out only if increases in backend pressure
eventually caused backend stalls.
About the time complexity:
Time complexity is linear in the number of instructions in the
Scheduler::PendingSet.
The average slowdown tends to be in the range of ~5-6%.
For memory intensive kernels, the slowdown can be significant if flag
-noalias=false is specified. In the worst case scenario I have observed a
slowdown of ~30% when flag -noalias=false was specified.
We can definitely recover part of that slowdown if we optimize class LSUnit (by
doing extra bookkeeping to speedup queries). For now, this new analysis is
disabled by default, and it can be enabled via flag -bottleneck-analysis. Users
of MCA as a library can enable the generation of pressure events through the
constructor of ExecuteStage.
This patch partially addresses https://bugs.llvm.org/show_bug.cgi?id=37494
Differential Revision: https://reviews.llvm.org/D58728
llvm-svn: 355308
Summary:
Before when we implemented the first EH proposal, 'catch <tag>'
instruction may not catch an exception so there were multiple EH pads an
exception can unwind to. That means a BB could have multiple EH pad
successors.
Now after we switched to the new proposal, every 'catch' instruction
catches an exception, and there is only one catchpad per catchswitch, so
we at most have one EH pad successor, making `ThrowUnwindDest` map in
`WasmEHInfo` unnecessary.
Keeping `ThrowUnwindDest` map in `WasmEHInfo` has its own problems,
because other optimization passes can split a BB that contains possibly
throwing calls (previously invokes), and we have to update the map every
time that happens, which is not easy for common CodeGen passes.
This also correctly updates successor info in LateEHPrepare when we add
a rethrow instruction.
Reviewers: dschuff
Subscribers: sbc100, jgravelle-google, sunfish, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58486
llvm-svn: 355296
Summary:
Add an SMLoc to CodeInit that records the source line it originated from.
This allows tablegen to point precisely at portions of code when reporting
errors within the CodeInit. For example, in the upcoming GlobalISel
combiner, it can report undefined expansions and point at the instance of
the expansion. This is achieved using something like:
SMLoc::getFromPointer(SMLoc::getPointer() +
(StringRef - CodeInit::getValue()))
The location is lost when producing a CodeInit by string concatenation so
a fallback SMLoc is required (e.g. the Record::getLoc()) but that's pretty
rare for CodeInits.
There's a reasonable case for extending tracking of a couple other Init
objects, for example StringInit's are often parsed and it would be good to
point inside the string when reporting errors about that. However, location
tracking also harms de-duplication. This is fine for CodeInit where there's
only a few hundred of them (~160 for X86) and it may be worth it for
StringInit (~86k up to ~1.9M for roughly 15MB increase for X86).
However the origin tracking would be a _terrible_ idea for IntInit, BitInit,
and UnsetInit. I haven't measured either of those three but BitInit would
most likely be on the order of increasing the current 2 BitInit values up
to billions.
Reviewers: volkan, aditya_nandakumar, bogner, paquette, aemerson
Reviewed By: paquette
Subscribers: javed.absar, kristof.beyls, dexonsmith, llvm-commits, kristina
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58141
llvm-svn: 355245
IntrArgMemOnly implies that only memory pointed to by pointer typed arguments will be accessed. But these intrinsics allow you to pass null to the pointer argument and put the full address into the index argument. Other passes won't be able to understand this.
A colleague found that ISPC was creating gathers like this and then dead store elimination removed some stores because it didn't understand what the gather was doing since the pointer argument was null.
Differential Revision: https://reviews.llvm.org/D58805
llvm-svn: 355228
This was sometimes causing clang or llvm-mc to crash, and in other
cases could emit a bogus DWARF line-table header. I did an interim
patch in r352541; this patch should be a cleaner and more complete
fix, and retains the test.
Addresses PR40538.
Differential Revision: https://reviews.llvm.org/D58750
llvm-svn: 355226
Previously we had build_vector PatFrags that called ISD::isBuildVectorAllZeros/Ones. Internally the ISD::isBuildVectorAllZeros/Ones look through bitcasts, but we aren't able to take advantage of that in isel. Instead of we have to canonicalize the types of the all zeros/ones build_vectors and insert bitcasts. Then we have to pattern match those exact bitcasts.
By emitting specific matchers for these 2 nodes, we can make isel look through any bitcasts without needing to explicitly match them. We should also be able to remove the canonicalization to vXi32 from lowering, but I've left that for a follow up.
This removes something like 40,000 bytes from the X86 isel table.
Differential Revision: https://reviews.llvm.org/D58595
llvm-svn: 355224
This is for tweaking SHT_SYMTAB sections.
Their sh_info contains the (number of symbols + 1) usually.
But for creating invalid inputs for test cases it would be convenient
to allow explicitly override this field from YAML.
Differential revision: https://reviews.llvm.org/D58779
llvm-svn: 355193
This is a small addition to arithmetic operations that improves
expressiveness of the language.
Differential Revision: https://reviews.llvm.org/D58775
llvm-svn: 355187
This patch allows all forms of values for options to be used at the end
of a group. With the fix, it is possible to follow the way GNU binutils
tools handle grouping options better. For example, the -j option can be
used with objdump in any of the following ways:
$ objdump -d -j .text a.o
$ objdump -d -j.text a.o
$ objdump -dj .text a.o
$ objdump -dj.text a.o
Differential Revision: https://reviews.llvm.org/D58711
llvm-svn: 355185
Summary:
ConstIntInfoVec contains elements extracted from the previous function.
In new PM, releaseMemory() is not called and the dangling elements can
cause segfault in findConstantInsertionPoint.
Rename releaseMemory() to cleanup() to deliver the idea that it is
mandatory and call cleanup() in ConstantHoistingPass::runImpl to fix
this.
Reviewers: ormris, zzheng, dmgreen, wmi
Reviewed By: ormris, wmi
Subscribers: llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58589
llvm-svn: 355174
Subtarget features are stored in a std::bitset that has been subclassed. There is a special constructor to allow the tablegen files to provide a list of bits to initialize the std::bitset to. This constructor isn't constexpr and std::bitset doesn't support many constexpr operations either. This results in a static global constructor being used to initialize the feature bitsets in these files at startup.
To fix this I've introduced a new FeatureBitArray class that holds three 64-bit values representing the initial bit values and taught tablegen to emit hex constants for them based on the feature enum values. This makes the tablegen files less readable than they were before. I can add the list of features back as a comment if we think that's important.
I've added a method to convert from this class into the std::bitset subclass we had before. I considered making the new FeatureBitArray class just implement the std::bitset interface we need instead, but thought I'd see how others felts about that first.
I've simplified the interfaces to SetImpliedBits and ClearImpliedBits a little minimize the number of times we need to convert to the bitset.
This removes about 27K from my local release+asserts build of llc.
Differential Revision: https://reviews.llvm.org/D58520
llvm-svn: 355167
Add support for cloning DWARF expressions that contain base type DIE
references in dsymutil.
<rdar://problem/48167812>
Differential Revision: https://reviews.llvm.org/D58534
llvm-svn: 355148
The basic idea of the pass is to use a circular buffer to log the execution ordering of the functions. We only log the function when it is first executed. We use a 8-byte hash to log the function symbol name.
In this pass, we add three global variables:
(1) an order file buffer: a circular buffer at its own llvm section.
(2) a bitmap for each module: one byte for each function to say if the function is already executed.
(3) a global index to the order file buffer.
At the function prologue, if the function has not been executed (by checking the bitmap), log the function hash, then atomically increase the index.
Differential Revision: https://reviews.llvm.org/D57463
llvm-svn: 355133
Part 2 of CSPGO changes (mostly related to ProfileSummary).
Note that I use a default parameter in setProfileSummary() and getSummary().
This is to break the dependency in clang. I will make the parameter explicit
after changing clang in a separated patch.
Differential Revision: https://reviews.llvm.org/D54175
llvm-svn: 355131
Summary:
The description of KnownBits::zext() and
KnownBits::zextOrTrunc() has confusingly been telling
that the operation is equivalent to zero extending the
value we're tracking. That has not been true, instead
the user has been forced to explicitly set the extended
bits as known zero afterwards.
This patch adds a second argument to KnownBits::zext()
and KnownBits::zextOrTrunc() to control if the extended
bits should be considered as known zero or as unknown.
Reviewers: craig.topper, RKSimon
Reviewed By: RKSimon
Subscribers: javed.absar, hiraditya, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58650
llvm-svn: 355099
Summary:
If the two sequences are not equal, std::is_permutation may be O(N^2)
and indeed the case in libstdc++ and libc++. Use SmallPtrSet to prevent
pessimizing cases. On my machine, SmallPtrSet starts to outperform
std::is_permutation when there are 16 elements.
Reviewers: kuhar
Reviewed By: kuhar
Subscribers: kristina, jdoerfert, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58373
llvm-svn: 355070
In D57601, I described the expectations around usage of AtomicSDNode and LoadSDNode/StoreSDNode w.r.t. atomic and volatiles. This patch simply embeds those expectations in assertions so that they can't be accidentally weakened.
Note: The reason only AtomicSDNodes of ATOMIC_LOAD and ATOMIC_STORE opcode are currently checked is that AMDGPU has an intrinsic which gets lowered to an ATOMIC_LOAD_FADD w/o a corresponding atomic MMO. This is suspicious, and I've brought it to the attention of the relevant developers. Once resolved, I'll strengthen that assertion.
Note 2: If this breaks your out-of-tree backend, go read the update instructions associated w/the previously mentioned patch. This will assert on (a subset of) things you need to update per those instructions.
Differential Revision: https://reviews.llvm.org/D58738
llvm-svn: 355069
OptBisect is in IR due to LLVMContext using it. However, it uses IR units from
Analysis as well. This change moves getDescription functions from OptBisect
to their respective IR units. Generating names for IR units will now be up
to the callers, keeping the Analysis IR units in Analysis. To prevent
unnecessary string generation, isEnabled function is added so that callers know
when the description needs to be generated.
Differential Revision: https://reviews.llvm.org/D58406
llvm-svn: 355068
This patch introduces Memory::MF_HUGE_HINT which indicates that allocateMappedMemory() shall return a pointer to a large memory page.
However the flag is a hint because we're not guaranteed in any way that we will get back a large memory page. There are several restrictions:
- Large/huge memory pages aren't enabled by default on modern OSes (Windows 10 and Linux at least), and should be manually enabled/reserved.
- Once enabled, it should be kept in mind that large pages are physical only, they can't be swapped.
- Memory fragmentation can affect the availability of large pages, especially after running the OS for a long time and/or running along many other applications.
Memory::allocateMappedMemory() will fallback to 4KB pages if it can't allocate 2MB large pages (if Memory::MF_HUGE_HINT is provided)
Currently, Memory::MF_HUGE_HINT only works on Windows. The hint will be ignored on Linux, 4KB pages will always be returned.
Differential Revision: https://reviews.llvm.org/D58718
llvm-svn: 355065
When using full LTO it is possible that template function definition DIE
is bound to one compilation unit and it's declaration to another. We should
add function declaration attributes on behalf of its owner CU otherwise
we may end up with malformed file identifier in function declaration
DW_AT_decl_file attribute.
Differential revision: https://reviews.llvm.org/D58538
llvm-svn: 354978
That patch is the fix for https://bugs.llvm.org/show_bug.cgi?id=40703
"wrong line number info for obj file compiled with -ffunction-sections"
bug. The problem happened with only .o files. If object file contains
several .text sections then line number information showed incorrectly.
The reason for this is that DwarfLineTable could not detect section which
corresponds to specified address(because address is the local to the
section). And as the result it could not select proper sequence in the
line table. The fix is to pass SectionIndex with the address. So that it
would be possible to differentiate addresses from various sections. With
this fix llvm-objdump shows correct line numbers for disassembled code.
Differential review: https://reviews.llvm.org/D58194
llvm-svn: 354972
DWARFFormValues can be created from a data extractor or by passing its
value directly. Until now this was done by member functions that
modified an existing object's internal state. This patch replaces a
subset of these methods with static method that return a new
DWARFFormValue.
llvm-svn: 354941
Current PGO profile counts are not context sensitive. The branch probabilities
for the inlined functions are kept the same for all call-sites, and they might
be very different from the actual branch probabilities. These suboptimal
profiles can greatly affect some downstream optimizations, in particular for
the machine basic block placement optimization.
In this patch, we propose to have a post-inline PGO instrumentation/use pass,
which we called Context Sensitive PGO (CSPGO). For the users who want the best
possible performance, they can perform a second round of PGO instrument/use on
the top of the regular PGO. They will have two sets of profile counts. The
first pass profile will be manly for inline, indirect-call promotion, and
CGSCC simplification pass optimizations. The second pass profile is for
post-inline optimizations and code-gen optimizations.
A typical usage:
// Regular PGO instrumentation and generate pass1 profile.
> clang -O2 -fprofile-generate source.c -o gen
> ./gen
> llvm-profdata merge default.*profraw -o pass1.profdata
// CSPGO instrumentation.
> clang -O2 -fprofile-use=pass1.profdata -fcs-profile-generate -o gen2
> ./gen2
// Merge two sets of profiles
> llvm-profdata merge default.*profraw pass1.profdata -o profile.profdata
// Use the combined profile. Pass manager will invoke two PGO use passes.
> clang -O2 -fprofile-use=profile.profdata -o use
This change touches many components in the compiler. The reviewed patch
(D54175) will committed in phrases.
Differential Revision: https://reviews.llvm.org/D54175
llvm-svn: 354930
This patch enables the following
1) AMD family 17h "znver2" tune flag (-march, -mcpu).
2) ISAs that are enabled for "znver2" architecture.
3) For the time being, it uses the znver1 scheduler model.
4) Tests are updated.
5) Scheduler descriptions are yet to be put in place.
Reviewers: craig.topper
Differential Revision: https://reviews.llvm.org/D58343
llvm-svn: 354897
Dispatch stall cycles may be associated to multiple dispatch stall events.
Before this patch, each stall cycle was associated with a single stall event.
This patch also improves a couple of code comments, and adds a helper method to
query the Scheduler for dispatch stalls.
llvm-svn: 354877
This allows tools to parse/dump the architecture specific tags
like DT_MIPS_*, DT_PPC64_* and DT_HEXAGON_*
Also fixes a bug in DynamicTags.def which was revealed in this patch.
Differential revision: https://reviews.llvm.org/D58667
llvm-svn: 354876
The --disassembler-options, or -M, are used to customize
the disassembler and affect its output.
The two implemented options allow selecting register names on ARM:
* With -Mreg-names-raw, the disassembler uses rNN for all registers.
* With -Mreg-names-std it prints sp, lr and pc for r13, r14 and r15,
which is the default behavior of llvm-objdump.
Differential Revision: https://reviews.llvm.org/D57680
llvm-svn: 354870
Summary:
In D58580 i have noted that `llvm::to_string()` is a memory hog.
It uses `raw_string_ostream`, and since it was buffered,
every `raw_string_ostream` had a cost of `BUFSIZ` bytes
(which is `8192` at least here). So every `llvm::to_string()`
call, even to just print an `int`, costed `8192` bytes.
In D58580, getting rid of that buffering //had// significant
performance and memory consumption improvements for `llvm-xray convert`.
Similarly, in D58580 @rnk pointed out that the `raw_svector_ostream`
is already unbuffered, and `write_unsigned_impl` and friends
do internal buffering. So it should be ok performance-wise to just
make the `raw_string_ostream` itself unbuffered.
Here, i don't have any perf measurements.
Another letdown is that i'm leaving a loose-end - not deleting the
`flush()` method. I don't expect that cleanup to be anything more
than just fixing every new compiler error, but i'm presently unable
to do that. Will look into that later.
Reviewers: rnk, zturner
Reviewed By: rnk
Subscribers: kristina, jdoerfert, llvm-commits, rnk
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D58643
llvm-svn: 354819
Andrea Di Biagio