manager, including both plumbing and logic to handle function pass
updates.
There are three fundamentally tied changes here:
1) Plumbing *some* mechanism for updating the CGSCC pass manager as the
CG changes while passes are running.
2) Changing the CGSCC pass manager infrastructure to have support for
the underlying graph to mutate mid-pass run.
3) Actually updating the CG after function passes run.
I can separate them if necessary, but I think its really useful to have
them together as the needs of #3 drove #2, and that in turn drove #1.
The plumbing technique is to extend the "run" method signature with
extra arguments. We provide the call graph that intrinsically is
available as it is the basis of the pass manager's IR units, and an
output parameter that records the results of updating the call graph
during an SCC passes's run. Note that "...UpdateResult" isn't a *great*
name here... suggestions very welcome.
I tried a pretty frustrating number of different data structures and such
for the innards of the update result. Every other one failed for one
reason or another. Sometimes I just couldn't keep the layers of
complexity right in my head. The thing that really worked was to just
directly provide access to the underlying structures used to walk the
call graph so that their updates could be informed by the *particular*
nature of the change to the graph.
The technique for how to make the pass management infrastructure cope
with mutating graphs was also something that took a really, really large
number of iterations to get to a place where I was happy. Here are some
of the considerations that drove the design:
- We operate at three levels within the infrastructure: RefSCC, SCC, and
Node. In each case, we are working bottom up and so we want to
continue to iterate on the "lowest" node as the graph changes. Look at
how we iterate over nodes in an SCC running function passes as those
function passes mutate the CG. We continue to iterate on the "lowest"
SCC, which is the one that continues to contain the function just
processed.
- The call graph structure re-uses SCCs (and RefSCCs) during mutation
events for the *highest* entry in the resulting new subgraph, not the
lowest. This means that it is necessary to continually update the
current SCC or RefSCC as it shifts. This is really surprising and
subtle, and took a long time for me to work out. I actually tried
changing the call graph to provide the opposite behavior, and it
breaks *EVERYTHING*. The graph update algorithms are really deeply
tied to this particualr pattern.
- When SCCs or RefSCCs are split apart and refined and we continually
re-pin our processing to the bottom one in the subgraph, we need to
enqueue the newly formed SCCs and RefSCCs for subsequent processing.
Queuing them presents a few challenges:
1) SCCs and RefSCCs use wildly different iteration strategies at
a high level. We end up needing to converge them on worklist
approaches that can be extended in order to be able to handle the
mutations.
2) The order of the enqueuing need to remain bottom-up post-order so
that we don't get surprising order of visitation for things like
the inliner.
3) We need the worklists to have set semantics so we don't duplicate
things endlessly. We don't need a *persistent* set though because
we always keep processing the bottom node!!!! This is super, super
surprising to me and took a long time to convince myself this is
correct, but I'm pretty sure it is... Once we sink down to the
bottom node, we can't re-split out the same node in any way, and
the postorder of the current queue is fixed and unchanging.
4) We need to make sure that the "current" SCC or RefSCC actually gets
enqueued here such that we re-visit it because we continue
processing a *new*, *bottom* SCC/RefSCC.
- We also need the ability to *skip* SCCs and RefSCCs that get merged
into a larger component. We even need the ability to skip *nodes* from
an SCC that are no longer part of that SCC.
This led to the design you see in the patch which uses SetVector-based
worklists. The RefSCC worklist is always empty until an update occurs
and is just used to handle those RefSCCs created by updates as the
others don't even exist yet and are formed on-demand during the
bottom-up walk. The SCC worklist is pre-populated from the RefSCC, and
we push new SCCs onto it and blacklist existing SCCs on it to get the
desired processing.
We then *directly* update these when updating the call graph as I was
never able to find a satisfactory abstraction around the update
strategy.
Finally, we need to compute the updates for function passes. This is
mostly used as an initial customer of all the update mechanisms to drive
their design to at least cover some real set of use cases. There are
a bunch of interesting things that came out of doing this:
- It is really nice to do this a function at a time because that
function is likely hot in the cache. This means we want even the
function pass adaptor to support online updates to the call graph!
- To update the call graph after arbitrary function pass mutations is
quite hard. We have to build a fairly comprehensive set of
data structures and then process them. Fortunately, some of this code
is related to the code for building the cal graph in the first place.
Unfortunately, very little of it makes any sense to share because the
nature of what we're doing is so very different. I've factored out the
one part that made sense at least.
- We need to transfer these updates into the various structures for the
CGSCC pass manager. Once those were more sanely worked out, this
became relatively easier. But some of those needs necessitated changes
to the LazyCallGraph interface to make it significantly easier to
extract the changed SCCs from an update operation.
- We also need to update the CGSCC analysis manager as the shape of the
graph changes. When an SCC is merged away we need to clear analyses
associated with it from the analysis manager which we didn't have
support for in the analysis manager infrsatructure. New SCCs are easy!
But then we have the case that the original SCC has its shape changed
but remains in the call graph. There we need to *invalidate* the
analyses associated with it.
- We also need to invalidate analyses after we *finish* processing an
SCC. But the analyses we need to invalidate here are *only those for
the newly updated SCC*!!! Because we only continue processing the
bottom SCC, if we split SCCs apart the original one gets invalidated
once when its shape changes and is not processed farther so its
analyses will be correct. It is the bottom SCC which continues being
processed and needs to have the "normal" invalidation done based on
the preserved analyses set.
All of this is mostly background and context for the changes here.
Many thanks to all the reviewers who helped here. Especially Sanjoy who
caught several interesting bugs in the graph algorithms, David, Sean,
and others who all helped with feedback.
Differential Revision: http://reviews.llvm.org/D21464
llvm-svn: 279618
Re-apply this patch, hopefully I will get away without any warnings
in the constructor now.
This patch removes the MachineFunctionAnalysis. Instead we keep a
map from IR Function to MachineFunction in the MachineModuleInfo.
This allows the insertion of ModulePasses into the codegen pipeline
without breaking it because the MachineFunctionAnalysis gets dropped
before a module pass.
Peak memory should stay unchanged without a ModulePass in the codegen
pipeline: Previously the MachineFunction was freed at the end of a codegen
function pipeline because the MachineFunctionAnalysis was dropped; With
this patch the MachineFunction is freed after the AsmPrinter has
finished.
Differential Revision: http://reviews.llvm.org/D23736
llvm-svn: 279602
Specifying isSSA is an extra line at best and results in invalid MI at
worst. Compute the value instead.
Differential Revision: http://reviews.llvm.org/D22722
llvm-svn: 279600
Change this pass constructor to just accept a const TargetMachine * and
use INITIALIZE_TM_PASS, that way we can get rid of the dummy
constructor. The pass will still fail when calling the default
constructor leading to TM == nullptr, this is no different than before
but is more in line what other codegen passes are doing and avoids the
dummy constructor.
llvm-svn: 279598
Summary:
In clang commit r268509 we started to invoke loop-unroll pass from the
driver even under -Os. However, we happen to not initialize optsize
thresholds properly, which si fixed with this change.
r268509 led to some big compile time regressions, because we started to
unroll some loops that we didn't unroll before. With this change I hope
to recover most of the regressions. We still are slightly slower than
before, because we do some checks here and there in loop-unrolling
before we bail out, but at least the slowdown is not that huge now.
Reviewers: hfinkel, chandlerc
Subscribers: mzolotukhin, llvm-commits
Differential Revision: https://reviews.llvm.org/D23388
llvm-svn: 279585
Add the ability to plug a cache on the LTO API.
I tried to write such that a linker implementation can
control the cache backend. This is intrusive and I'm
not totally happy with it, but I can't figure out a
better design right now.
Differential Revision: https://reviews.llvm.org/D23599
llvm-svn: 279576
I want to compute the SSA property of .mir files automatically in
upcoming patches. The problem with this is that some inputs will be
reported as static single assignment with some passes claiming not to
support SSA form. In reality though those passes do not support PHI
instructions => Track the presence of PHI instructions separate from the
SSA property.
Differential Revision: https://reviews.llvm.org/D22719
llvm-svn: 279573
They really should have both types represented, but early variants were created
before MachineInstrs could have multiple types so they're rather ambiguous.
llvm-svn: 279567
Re-apply this commit with the deletion of a MachineFunction delegated to
a separate pass to avoid use after free when doing this directly in
AsmPrinter.
This patch removes the MachineFunctionAnalysis. Instead we keep a
map from IR Function to MachineFunction in the MachineModuleInfo.
This allows the insertion of ModulePasses into the codegen pipeline
without breaking it because the MachineFunctionAnalysis gets dropped
before a module pass.
Peak memory should stay unchanged without a ModulePass in the codegen
pipeline: Previously the MachineFunction was freed at the end of a codegen
function pipeline because the MachineFunctionAnalysis was dropped; With
this patch the MachineFunction is freed after the AsmPrinter has
finished.
Differential Revision: http://reviews.llvm.org/D23736
llvm-svn: 279564
Instructions like G_ICMP have multiple types that may need to be legalized (the
boolean output and nearly arbitrary inputs in this case). So the legalizer must
be capable of deciding what to do for each of them separately.
llvm-svn: 279554
Summary:
I assume there was a use case, so maybe this strawman patch will help
clarifying if it is legit.
In any case the current situation is not legit: a ThinLTO compilation
should not trigger an unexpected full LTO compilation.
Right now, adding a --save-temps option triggers this and makes the
number of output differs.
Reviewers: tejohnson
Subscribers: pcc, llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D23600
llvm-svn: 279550
Summary:
This greatly simplifies our handling of SDNode::SubclassData.
NFC, hopefully. :)
See discussion in D23035 for discussion about the design API of these
bitfields.
Reviewers: chandlerc
Subscribers: llvm-commits, rnk
Differential Revision: https://reviews.llvm.org/D23036
llvm-svn: 279537
This patch removes the MachineFunctionAnalysis. Instead we keep a
map from IR Function to MachineFunction in the MachineModuleInfo.
This allows the insertion of ModulePasses into the codegen pipeline
without breaking it because the MachineFunctionAnalysis gets dropped
before a module pass.
Peak memory should stay unchanged without a ModulePass in the codegen
pipeline: Previously the MachineFunction was freed at the end of a codegen
function pipeline because the MachineFunctionAnalysis was dropped; With
this patch the MachineFunction is freed after the AsmPrinter has
finished.
Differential Revision: http://reviews.llvm.org/D23736
llvm-svn: 279502
Separate algorithms in iplist<T> that don't depend on T into ilist_base,
and unit test them.
While I was adding unit tests for these algorithms anyway, I also added
unit tests for ilist_node_base and ilist_sentinel<T>.
To make the algorithms and unit tests easier to write, I also did the
following minor changes as a drive-by:
- encapsulate Prev/Next in ilist_node_base to so that algorithms are
easier to read, and
- update ilist_node_access API to take nodes by reference.
There should be no real functionality change here.
llvm-svn: 279484
Summary: Before the change, *Opt never actually gets updated by the end
of toNext(), so for every next time the loop has to start over from
child_begin(). This bug doesn't affect the correctness, since Visited prevents
it from re-entering the same node again; but it's slow.
Reviewers: dberris, dblaikie, dannyb
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D23649
llvm-svn: 279482
Remove all the dead code around ilist_*sentinel_traits. This is a
follow-up to gutting them as part of r279314 (originally r278974),
staged to prevent broken builds in sub-projects.
Uses were removed from clang in r279457 and lld in r279458.
llvm-svn: 279473
Philip commented on r279113 to ask for better comments as to
when to use the different versions of getName. Its also possible
to assert in the simple case that we aren't an overloaded intrinsic
as those have to use the more capable version of getName.
Thanks for the comments Philip.
llvm-svn: 279466
Assembler directives .dtprelword, .dtpreldword, .tprelword, and
.tpreldword generates relocations R_MIPS_TLS_DTPREL32, R_MIPS_TLS_DTPREL64,
R_MIPS_TLS_TPREL32, and R_MIPS_TLS_TPREL64 respectively.
The main motivation for this patch is to be able to write test cases
for checking correctness of the LLD linker's behaviour.
Differential Revision: https://reviews.llvm.org/D23669
llvm-svn: 279439
It use to be non-const for the sole purpose of custom handling of
commons symbol. This is moved now in the regular LTO handling now
and such we can constify the callback.
llvm-svn: 279438
The gold-plugin was doing this internally, now the API is handling
commons correctly based on the given resolution.
Differential Revision: https://reviews.llvm.org/D23739
llvm-svn: 279417
Summary: Reduce store size to avoid leading and trailing zeros.
Reviewers: kcc, eugenis
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D23648
llvm-svn: 279379
Summary:
We are going to combine poisoning of red zones and scope poisoning.
PR27453
Reviewers: kcc, eugenis
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D23623
llvm-svn: 279373
unit for use in the PreservedAnalyses set.
This doesn't have any important functional change yet but it cleans
things up and makes the analysis substantially more efficient by
avoiding querying through the type erasure for every analysis.
I also think it makes it much easier to reason about how analyses are
preserved when walking across pass managers and across IR unit
abstractions.
Thanks to Sean and Mehdi both for the comments and suggestions.
Differential Revision: https://reviews.llvm.org/D23691
llvm-svn: 279360
- Always compile print() regardless of LLVM_ENABLE_DUMP. (We usually
only gard dump() functions with that).
- Only show the set properties to reduce output clutter.
- Remove the unused variant that even shows the unset properties.
- Fix comments
llvm-svn: 279338
Currently nodes_iterator may dereference to a NodeType* or a NodeType&. Make them all dereference to NodeType*, which is NodeRef later.
Differential Revision: https://reviews.llvm.org/D23704
Differential Revision: https://reviews.llvm.org/D23705
llvm-svn: 279326
This reverts commit r279053, reapplying r278974 after fixing PR29035
with r279104.
Note that r279312 has been committed in the meantime, and this has been
rebased on top of that. Otherwise it's identical to r278974.
Note for maintainers of out-of-tree code (that I missed in the original
message): if the new isKnownSentinel() assertion is firing from
ilist_iterator<>::operator*(), this patch has identified a bug in your
code. There are a few common patterns:
- Some IR-related APIs htake an IRUnit* that might be nullptr, and pass
in an incremented iterator as an insertion point. Some old code was
using "&*++I", which in the case of end() only worked by fluke. If
the IRUnit in question inherits from ilist_node_with_parent<>, you can
use "I->getNextNode()". Otherwise, use "List.getNextNode(*I)".
- In most other cases, crashes on &*I just need to check for I==end()
before dereferencing.
- There's also occasional code that sends iterators into a function, and
then starts calling I->getOperand() (or other API). Either check for
end() before the entering the function, or early exit.
Note for if the static_assert with HasObsoleteCustomization is firing
for you:
- r278513 has examples of how to stop using custom sentinel traits.
- r278532 removed ilist_nextprev_traits since no one was using it. See
lld's r278469 for the only migration I needed to do.
Original commit message follows.
----
This removes the undefined behaviour (UB) in ilist/ilist_node/etc.,
mainly by removing (gutting) the ilist_sentinel_traits customization
point and canonicalizing on a single, efficient memory layout. This
fixes PR26753.
The new ilist is a doubly-linked circular list.
- ilist_node_base has two ilist_node_base*: Next and Prev. Size-of: two
pointers.
- ilist_node<T> (size-of: two pointers) is a type-safe wrapper around
ilist_node_base.
- ilist_iterator<T> (size-of: two pointers) operates on an
ilist_node<T>*, and downcasts to T* on dereference.
- ilist_sentinel<T> (size-of: two pointers) is a wrapper around
ilist_node<T> that has some extra API for list management.
- ilist<T> (size-of: two pointers) has an ilist_sentinel<T>, whose
address is returned for end().
The new memory layout matches ilist_half_embedded_sentinel_traits<T>
exactly. The Head pointer that previously lived in ilist<T> is
effectively glued to the ilist_half_node<T> that lived in
ilist_half_embedded_sentinel_traits<T>, becoming the Next and Prev in
the ilist_sentinel_node<T>, respectively. sizeof(ilist<T>) is now the
size of two pointers, and there is never any additional storage for a
sentinel.
This is a much simpler design for a doubly-linked list, removing most of
the corner cases of list manipulation (add, remove, etc.). In follow-up
commits, I intend to move as many algorithms as possible into a
non-templated base class (ilist_base) to reduce code size.
Moreover, this fixes the UB in ilist_iterator/getNext/getPrev
operations. Previously, ilist_iterator<T> operated on a T*, even when
the sentinel was not of type T (i.e., ilist_embedded_sentinel_traits and
ilist_half_embedded_sentinel_traits). This added UB to all operations
involving end(). Now, ilist_iterator<T> operates on an ilist_node<T>*,
and only downcasts when the full type is guaranteed to be T*.
What did we lose? There used to be a crash (in some configurations) on
++end(). Curiously (via UB), ++end() would return begin() for users of
ilist_half_embedded_sentinel_traits<T>, but otherwise ++end() would
cause a nice dependable nullptr dereference, crashing instead of a
possible infinite loop. Options:
1. Lose that behaviour.
2. Keep it, by stealing a bit from Prev in asserts builds.
3. Crash on dereference instead, using the same technique.
Hans convinced me (because of the number of problems this and r278532
exposed on Windows) that we really need some assertion here, at least in
the short term. I've opted for #3 since I think it catches more bugs.
I added only a couple of unit tests to root out specific bugs I hit
during bring-up, but otherwise this is tested implicitly via the
extensive usage throughout LLVM.
Planned follow-ups:
- Remove ilist_*sentinel_traits<T>. Here I've just gutted them to
prevent build failures in sub-projects. Once I stop referring to them
in sub-projects, I'll come back and delete them.
- Add ilist_base and move algorithms there.
- Check and fix move construction and assignment.
Eventually, there are other interesting directions:
- Rewrite reverse iterators, so that rbegin().getNodePtr()==&*rbegin().
This allows much simpler logic when erasing elements during a reverse
traversal.
- Remove ilist_traits::createNode, by deleting the remaining API that
creates nodes. Intrusive lists shouldn't be creating nodes
themselves.
- Remove ilist_traits::deleteNode, by (1) asserting that lists are empty
on destruction and (2) changing API that calls it to take a Deleter
functor (intrusive lists shouldn't be in the memory management
business).
- Reconfigure the remaining callback traits (addNodeToList, etc.) to be
higher-level, pulling out a simple_ilist<T> that is much easier to
read and understand.
- Allow tags (e.g., ilist_node<T,tag1> and ilist_node<T,tag2>) so that T
can be a member of multiple intrusive lists.
llvm-svn: 279314
Chandler Carruth