Currently, CopyPropagation only canonicalizes defs that are "canonical",
that is, the root of the copy_value tree. When that canonical
def is lexical, however, the canonicalization respects deinit barriers.
But copies of lexical values are not themselves lexical, so their
lifetimes can be shortened without respect to deinit barriers.
Here, immediate copies of lexical values are canonicalized before the
lexical values themselves are.
rdar://107197935
Previously, the utility bailed out on lexical lifetimes because it
didn't respect deinit barriers. Here, deinit barriers are found and
added to liveness if the value is lexical. This enables copies to be
propagated without hoisting destroys over deinit barriers.
rdar://104630103
If a move_value is determined to be redundant and removed, take the
opportunity its removal makes available to canonicalize the moved-from
value without the obstruction of the move_value.
Add a run of ComputeSideEffects before the first run of CopyPropagation.
Allow hoisting over applies of functions that are able to be analyzed
not to be deinit barriers at this early point.
SemanticARCOpts already eliminates move values that are redundant that
block its optimizations. But it's always run after CopyPropagation.
Because move_values divide copy-extended lifetimes, move_values obstruct
lifetime canonicalization. If a move_value isn't separating lifetimes
with different characteristics (specifically: lexicallity, escaping),
then it is only obstructing lifetime canonicalization. Remove it
before canonicalizing the lifetime of the moved-from value.
Pass a BasicCalleeAnalysis instance to isDeinitBarrier. This enables
LexicalDestroyHoisting to hoist destroys over applies of functions which
are not deinit barriers.
Pass a BasicCalleeAnalysis instance to isDeinitBarrier. This will allow
ShrinkBorrowScope to hoist end_borrows over applies of functions which
are not deinit barriers.
To improve the debugging experience of values whose lifetimes are
canonicalized without compromising the semantics expressed in the source
language, when canonicalizing OSSA lifetimes at Onone, lengthen
lifetimes as much as possible without incurring copies that would be
eliminated at O.
rdar://99618502
Andy some time ago already created the new API but didn't go through and update
the old occurences. I did that in this PR and then deprecated the old API. The
tree is clean, so I could just remove it, but I decided to be nicer to
downstream people by deprecating it first.
Previously, the lifetimes of only values which were copied were
canonicalized during the non-mandatory CopyPropagation pass. The
mandatory pass (i.e. MandatoryCopyPropagation) has been disabled,
meaning that non-copied values lifetimes were never canonicalized. So
enable canonicalization of all lifetimes during regular CopyPropagation.
rdar://54335055
When canonicalizing an owned value's lifetime, also check whether the
value is dead. If it is, track it for deletion. In particular, this
eliminates dead copy_values.
Assertion failed: (succeed && "should be filtered by
FindBorrowScopeUses"), function canonicalizeFunctionArgument, file
CanonicalizeBorrowScope.cpp, line 798
Canonicalization for guaranteed function arguments is triggered by
SILCombine without any up-front analysis. Because the canonicalization
rewrites the function argument's copies in place, it must always
succeed.
Fix the visitBorrowScopeUses utility to be aware that it is being
invoked on a function argument and avoid bailing out.
Mandatory copy propagation was primarily a stop-gap until lexcial
lifetimes were implemented. It supposedly made variables lifetimes
more consistent between -O and -Onone builds. Now that lexical
lifetimes are enabled, it is no longer needed for that purpose (and
will never satisfactorily meet that goal anyway).
Mandatory copy propagation may be enabled again later as a -Onone "
optimization. But that requires a more careful audit of the effect on
debug information.
For now, it should be disabled.
Replaced ShrinkBorrowScope's own data flow with the general
BackwardReachability.
Took this opportunity to refactor and document the utility.
Taken together these changes make ShrinkBorrowScope serve as a template
for a future LexicalDestroyHoisting which will operate on owned lexical
values (rather than guaranteed as here) and hoist destroy_values (rather
than end_borrows as here) but should otherwise be quite similar.
So that CopyPropagation and other clients can react accordingly, pass
back a list of copy_value instructions that were rewritten by
ShrinkBorrowScope. In CopyPropagation, add each modified copy to the
copy worklist.
During copy propagation (for which -enable-copy-propagation must still
be passed), also try to shrink borrow scopes by hoisting end_borrows
using the newly added ShrinkBorrowScope utility.
Allow end_borrow instructions to be hoisted over instructions that are
not deinit barriers for the value which is borrowed. Deinit barriers
include uses of the value, loads of memory, loads of weak references
that may be zeroed during deinit, and "synchronization points".
rdar://79149830
OSSA rauw cleans up end of scope markers before rauw'ing.
This can lead to inadvertant deleting of end_lifetime, later
resulting in an ownership verifier error indicating a leak.
This PR stops treating end_lifetime scope ending like end_borrow/end_access.
This rewrites functionality that was mostly disabled but is now ready
to be enabled.
Allow lifetime canonicalization of owned values and function arguments
as a simple stand-alone utility. This is now being called from within
SILCombine, so we should only do the kind of canonicalization that
makes sense in that context.
Canonicalizing other borrow scopes should *not* be invoked as a
single-value cleanup because it affects other lifetimes outside the
borrow scope boundary. It is a somewhat complicated process that
hoists and sinks forwarding instructions and can generate surrounding
compensation code. The copy propagation pass knows how to post-process
the related lifetimes in just the right order. So borrow scope
rewriting should only be done in the copy propagation pass.
Similarly, only do simple canonicalization of owned values and
function arguments at -Onone.
The feature to canoncalize borrow scopes is now ready to be
enabled (-canonical-ossa-rewrite-borrows), but flipping the switch
should be a separate commit. So most of the functionality that was
affected is not exposed by this PR.
Changes:
Split canonicalization of owned lifetimes vs. borrowed lifetimes into
separate utilities. The owned lifetime utility is now back to being
the simple utility that I originally envisioned. So not much happened
to it other than removing complexity.
We now have a separate entry point for finding the starting point for
rewriting borrow scopes:
CanonicalizeBorrowScope::getCanonicalBorrowedDef.
We now have a utility that defines forwarding instructions that we can
treat consistently as part of a guaranteed lifetime,
CanonicalizeBorrowScope::isRewritableOSSAForward.
We now have a utility that defines the uses of a borrowed value that
are considered part of its lifetime,
CanonicalizeBorrowScope::visitBorrowScopeUses. This single utility is
used to implement three different parts of the alogrithm:
1. Find any uses of the borrowed value that need to be propagated
outside the borrow scope
2. RewriteInnerBorrowUses for SILFunction arguments and borrow scopes
with no outer uses.
3. RewriteOuterBorrowUses for borrow scopes with outer uses. Handling
these involves creating new copies outside the borrow scope and
hoisting forwarding instructions.
The end result is that a lot of borrow scopes can be eliminated and
owned values can be forwarded to destructures, reducing copies and
destroys.
If we stop generating borrow scopes for all interior pointers, then
we'll need to design a comparable optimization that works on
"implicit" borrow scopes:
%ownedDef = ...
%element struct_extract %ownedDef
%copy = copy_value %element
apply(@guaranteed %element)
apply(@owned %copy)
destroy %ownedDef
Should be:
%ownedDef = ...
%borrowedElement = destructure_struct @guaranteed %ownedDef
apply(@guaranteed %borrowedElement)
%ownedElement = destructure_struct %ownedDef
apply(@owned %copy)
It now handles looking through forwarding consumes such as
destructures.
Expected to be NFC since borrow consolidation is still disabled by
default.
TODO: Write unit tests for various forwarding consumes in addition to
destructure.
struct_extract and tuple_extract do not belong in OSSA (except to
workaround certain extreme cases). They completely defeat
simplification that OSSA provides for optimizing owned lifetimes.
Copy propagation uses this utility to canonicalize owned values before
canonicalizing their lifetime.
Ensure that any object lifetime that will be shortened ends in
destroy_value [poison], indicating that the reference should not be
dereferenced (e.g. by the debugger) beyond this point.
This has no way of knowing whether the object will actually be
deallocated at this point. It conservatively avoids showing garbage to
debuggers.
Part 1/2: rdar://75012368 (-Onone compiler support for early object
deinitialization with sentinel dead references)
-enable-copy-propagation: enables whatever form of copy propagation
the current pipeline runs (mandatory-copy-propagation at -Onone,
regular copy-propation at -O).
-disable-copy-propagation: similarly disables any form of copy
propagation in the current pipelien.
This bleeds into the implementation where "guaranteed" is used
everywhere to talk about optimization of guaranteed values. We need to
use mandatory to indicate we're talking about the pass pipeline.
It is currently disabled so this commit is NFC.
MandatoryCopyPropagation canonicalizes all all OSSA lifetimes with
either CopyValue or DestroyValue operations. While regular
CopyPropagation only canonicalizes lifetimes with copies. This ensures
that more lifetime program bugs are found in debug builds. Eventually,
regular CopyPropagation will also canonicalize all lifetimes, but for
now, we don't want to expose optimized code to more behavior change
than necessary.
Add frontend flags for developers to easily control copy propagation:
-enable-copy-propagation: enables whatever form of copy propagation
the current pipeline runs (mandatory-copy-propagation at -Onone,
regular copy-propation at -O).
-disable-copy-propagation: similarly disables any form of copy
propagation in the current pipelien.
To control a specific variant of the passes, use
-Xllvm -disable-pass=mandatory-copy-propagation
or -Xllvm -disable-pass=copy-propagation instead.
The meaning of these flags will stay the same as we adjust the
defaults. Soon mandatory-copy-propagation will be enabled by
default. There are two reasons to do this, both related to predictable
behavior across Debug and Release builds.
1. Shortening object lifetimes can cause observable changes in program
behavior in the presense of weak/unowned reference and
deinitializer side effects.
2. Programmers need to know reliably whether a given code pattern will
copy the storage for copy-on-write types (Array, Set). Eliminating
the "unexpected" copies the same way at -Onone and -O both makes
debugging tractable and provides assurance that the code isn't
relying on the luck of the optimizer in a particular compiler
release.
Add support for interleaved borrow scopes:
%b1 = begin_borrow %a
%c = copy
%b2 = begin_borrow %b1
end_borrow %b1
use %c
end_borrow %b2
Will be transformed to:
%c = copy %a
%b1 = begin_borrow %a
%b2 = begin_borrow %b1
end_borrow %b1
use %c
end_borrow %b2
This was the original intention but the implementation was incomplete.
This option can be enabled as soon as we need it for performance.
The intention is also to handle multi-block borrows, but that hasn't
been implemented.
Access scopes for enforcing exclusivity are currently the only
exception to our ability to canonicalize OSSA lifetime purely based on
the SSA value's known uses. This is because access scopes have
semantics relative to object deinitializers.
In general, deinitializers are asynchronous with respect to code that
is unrelated to the object's uses. Ignoring exclusivity, the optimizer
may always destroy objects as early as it wants, as long as the object
won't be used again. The optimizer may also extend the lifetime
(although in the future this lifetime extension should be limited by
"synchronization points").
The optimizer's freedom is however limited by exclusivity
enforcement. Optimization may never introduce new exclusivity
violations. Destroying an object within an access scope is an
exclusivity violation if the deinitializer accesses the same variable.
To handle this, OSSA canonicalization must detect access scopes that
overlap with the end of the pruned extended lifetime. Essentially:
%def
begin_access // access scope unrelated to def
use %def // pruned liveness ends here
end_access
destroy %def
Support for access scopes composes cleanly with the existing algorithm
without adding significant cost in the usual case. Overlapping access
scopes are unusual. A single CFG walk within the original extended
lifetime is normally sufficient. Only the blocks that are not already
LiveOut in the pruned liveness need to be visited. During this walk,
local overlapping access are detected by scanning for end_access
instructions after the last use point. Global overlapping accesses are
detected by checking NonLocalAccessBlockAnalysis. This avoids scanning
instructions in the common case. NonLocalAccessBlockAnalysis is a
trivial analysis that caches the rare occurence of nonlocal access
scopes. The analysis itself is a single linear scan over the
instruction stream. This analysis can be preserved across most
transformations and I expect it to be used to speed up other
optimizations related to access marker.
When an overlapping access is detected, pruned liveness is simply
extended to include the end_access as a new use point. Extending the
lifetime is iterative, but with each iteration, blocks that are now
marked LiveOut no longer need to be visited. Furthermore, interleaved
accessed scopes are not expected to happen in practice.
MandatoryCopyPropagation must be a separate pass in order to preserve
all debug_value instructions. CopyPropagation cannot preserve
debug_value because, as a rule, debug information cannot affect -O
behavior.
Canonicalizing OSSA provably minimizes the number of retains and
releases within the boundaries of that lifetime. This eliminates the
need for ad-hoc optimization of OSSA copies.
This initial implementation only canonicalizes owned values, but
canonicalizing guaranteed values is a simple extension.
This was originally part of the CopyPropagation prototype years
ago. Now OSSA is specified completely enough that it can be turned
into a simple utility instead.
CanonicalOSSALifetime uses PrunedLiveness to find the extended live
range and identify the consumes on the boundary. All other consumes
need their own copy. No other copies are needed.
By running this after other transformations that affect OSSA
lifetimes, we can avoid the need to run pattern-matching optimization
to SemanticARC to recover from suboptimal patterns, which is not
robust, maintainable, or efficient.
This makes it easier to understand conceptually why a ValueOwnershipKind with
Any ownership is invalid and also allowed me to explicitly document the lattice
that relates ownership constraints/value ownership kinds.
This makes it clearer that isConsumingUse() is not an owned oriented API and
returns also for instructions that end the lifetime of guaranteed values like
end_borrow.
This is in preparation for adding a run of this around ownership lowering in
order to eliminate extra copies that passes may introduce as they transform IR.
The tests for the pass all still pass in the exact same way so no updates were
needed.
...and avoid reallocation.
This is immediately necessary for LICM, in addition to its current
uses. I suspect this could be used by many passes that work with
addresses. RLE/DSE should absolutely migrate to it.
The XXOptUtils.h convention is already established and parallels
the SIL/XXUtils convention.
New:
- InstOptUtils.h
- CFGOptUtils.h
- BasicBlockOptUtils.h
- ValueLifetime.h
Removed:
- Local.h
- Two conflicting CFG.h files
This reorganization is helpful before I introduce more
utilities for block cloning similar to SinkAddressProjections.
Move the control flow utilies out of Local.h, which was an
unreadable, unprincipled mess. Rename it to InstOptUtils.h, and
confine it to small APIs for working with individual instructions.
These are the optimizer's additions to /SIL/InstUtils.h.
Rename CFG.h to CFGOptUtils.h and remove the one in /Analysis. Now
there is only SIL/CFG.h, resolving the naming conflict within the
swift project (this has always been a problem for source tools). Limit
this header to low-level APIs for working with branches and CFG edges.
Add BasicBlockOptUtils.h for block level transforms (it makes me sad
that I can't use BBOptUtils.h, but SIL already has
BasicBlockUtils.h). These are larger APIs for cloning or removing
whole blocks.
