Saves a bunch of compile time when compiling the stdlib. Specifically, when
compiling the iOS 32 bit stdlib + sil-verify-all, this shaves off ~20% of the
compile time.
Enable most simplify-cfg optimizations as long as the block arguments
have trivial types. Enable most simplify CFG unit tests cases.
This massively reduces the size of the CFG during OSSA passes.
Test cases that aren't supported in OSSA yet have been moved to a
separate test file for disabled OSSA tests,
Full simplify-cfg support is currently blocked on OSSA utilities which
I haven't checked in yet.
This is kind of complicated, because an enum can be trivial for one case and not trivial for another case. We need to check at which parts of the function we can prove that the enum does (or could) have a trivial case. In such a branch, it's not required in SIL to destroy the enum location.
Also, document the rules and requirements for enum memory locations in SIL.rst.
rdar://73770085
Specifically, given code like the following:
```
(%1, %2) = multi_result_inst %0 (a)
inst_to_delete %1 (b)
inst_to_delete %2 (c)
```
We would sometimes visit (c) before (b). So if nextII was (b) at that point, we
would think that we were safe. Then we process (b), move nextII from (b) ->
(c). Then we delete both (b) and (c). =><=.
The solution to this is each iteration of the delete loop, we track /all/
instructions that eliminateDeadInstruction is going to eliminate and ensure that
after each callback is called we have moved past all visited instructions (even
ones we visited previously). This is done using a small set that I clear each
iteration.
I also re-enabled the dictionary test that exposed this issue when testing
-O/-Osize.
rdar://71933996
- Properly clone and use debug scopes for all instructions in pullback functions.
- Emit `debug_value` instructions for adjoint values.
- Add debug locations and variable info to adjoint buffer allocations.
- Add `TangentBuilder` (a `SILBuilder` subclass) to unify and simplify special emitter utilities for tangent vector code generation. More simplifications to come.
Pullback variable inspection example:
```console
(lldb) n
Process 50984 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = step over
frame #0: 0x0000000100003497 main`pullback of foo(x=0) at main.swift:12:11
9 import _Differentiation
10
11 func foo(_ x: Float) -> Float {
-> 12 let y = sin(x)
13 let z = cos(y)
14 let k = tanh(z) + cos(y)
15 return k
Target 0: (main) stopped.
(lldb) fr v
(Float) x = 0
(Float) k = 1
(Float) z = 0.495846391
(Float) y = -0.689988375
```
Resolves rdar://68616528 / SR-13535.
Deinitializer side effects severely handicap the connection-graph
based EscapeAnalysis. Because it's not flow-sensitive, this approach
falls apart whenever an object is released in the current function,
which makes it seem to have escaped everywhere (it generally doesn't
matter if it doesn't escape until the release point, but the analysis
can't discern that).
This can be slightly mitigated by realizing that releasing an object
can only cause things it points to to escape if the object itself has
references or pointers.
Fix: Don't create a escaping content node when releasing an object
that can't contain any references or pointers. The previous commit,
"Fix EscapeAnalysis::mayReleaseContent" would defeat release-hoisting
in important cases without doing anything else. Adding this extra
precision to the connection graph avoids some regressions.
The previous implementation made extremely subtle and specific
assumptions about how the API is used which doesn't apply
everywhere. It was trying very hard to avoid regressing performance
relative to an even older implementation that didn't even try to consider deinitializer side effects.
The aggressive logic was based on the idea that a release must have a
corresponding retain somewhere in the same function and we don't care
if the last release happens early if there are no more aliasing
uses. All the unit tests I wrote previously were based on release
hoisting, which happens to work given the way the API is used.
But this logic is incorrect for retain sinking. In that case sinking
past an "unrelated" release could cause the object to be freed
early. See test/SILOptimizer/arc_crash.swift.
With SemanticARC and other SIL improvements being made, I'm afraid
bugs like this will begin to surface.
To fix it, just remove the subtle logic to leave behind a simple and
sound EscapeAnalysis API. To do better, we will need to rewrite the
AliasAnalysis logic for release side effects, which is currently
a tangled web. In the meantime, SemanticARC can handle many cases without EscapeAnalysis.
Fixes rdar://74469299 (ARC miscompile:
EscapeAnalysis::mayReleaseContent; potential use-after-free)
While fixing this, add support for address-type queries too:
Fixes rdar://74360041 (Assertion failed:
(!releasedReference->getType().isAddress() && "an address is never a
reference"), function mayReleaseContent
The DiagnoseLifetimeIssuesPass pass prints a warning if an object is stored to a weak property (or is weakly captured) and destroyed before the property (or captured reference) is ever used again.
This can happen if the programmer relies on the lexical scope to keep an object alive, but copy-propagation can shrink the object's lifetime to its last use.
For example:
func test() {
let k = Klass()
// k is deallocated immediately after the closure capture (a store_weak).
functionWithClosure({ [weak k] in
// crash!
k!.foo()
})
}
Unfortunately this pass can only catch simple cases, but it's better than nothing.
rdar://73910632
A store_borrow is a manner to temporarily "borrow" a guaranteed value into
memory for purposes like reabstraction. To make this safer in OSSA, we treat a
store_borrow's result as an interior pointer into the stored guaranteed value,
causing all uses of that result to be validated as being within the lifetime of
the guaranteed value.
NOTE: This is not the complete store_borrow verification story. We also will
verify that the memory that is being store_borrowed into is not written to while
the store_borrow's result is live.
This only runs on OSSA now, so I am deleting the non-OSSA tests and put in an
early bail out since there are some tests that have not been updated yet for
OSSA, that run the diagnostic passes, but for which this pass never kicked in
on.
NOTE: I also while updating the tests noticed a bunch were assuming load_borrow
like semantics. So I updated them with that in mind and implemented support in
the pass for that.
Add the following new mangling rules.
```
global ::= from-type to-type 'TJO' AUTODIFF-FUNCTION-KIND // autodiff self-reordering reabstraction thunk
global ::= from-type 'TJS' AUTODIFF-FUNCTION-KIND INDEX-SUBSET 'p' INDEX-SUBSET 'r' INDEX-SUBSET 'P' // autodiff linear map subset parameters thunk
global ::= global to-type 'TJS' AUTODIFF-FUNCTION-KIND INDEX-SUBSET 'p' INDEX-SUBSET 'r' INDEX-SUBSET 'P' // autodiff derivative function subset parameters thunk
```
Example:
```console
$s13TangentVector16_Differentiation14DifferentiablePQzAaDQy_SdAFIegnnnr_TJSdSSSpSrSUSP ---> autodiff subset parameters thunk for differential from @escaping @callee_guaranteed (@in_guaranteed A._Differentiation.Differentiable.TangentVector, @in_guaranteed B._Differentiation.Differentiable.TangentVector, @in_guaranteed Swift.Double) -> (@out B._Differentiation.Differentiable.TangentVector) with respect to parameters {0, 1, 2} and results {0} to parameters {0, 2}
$sS2f8mangling3FooV13TangentVectorVIegydd_SfAESfIegydd_TJOp ---> autodiff self-reordering reabstraction thunk for pullback from @escaping @callee_guaranteed (@unowned Swift.Float) -> (@unowned Swift.Float, @unowned mangling.Foo.TangentVector) to @escaping @callee_guaranteed (@unowned Swift.Float) -> (@unowned mangling.Foo.TangentVector, @unowned Swift.Float)
```
Resolves rdar://72666310 / SR-13508.
Also fix a bug in `AutoDiffFunction` mangling where the original may be a global that contains more than 1 node (rdar://74151229 / SR-14106).
There is some sort of ASAN issue that this exposes on Linux, so I am going to do
this on Darwin and then debug the Linux issue using ASAN over the weekend/next
week.
This eliminates some regressions by eliminating phase ordering in between
ARCSequenceOpts/inlining with read only functions whose read onlyness is lost
after inlining.
In OSSA, we do not allow for address phis, but in certain cases the logic of
LoopRotate really wants them. To work around this issue, I added some code in
this PR to loop rotate that as a post-pass fixes up any address phis by
inserting address <-> raw pointer adapters and changing the address phi to
instead be of raw pointer type.
This commit is fixing two things:
1. In certain cases, we are seeing cases where either SILGen or the optimizer
are eliminating destroy_addr along paths where we know that an enum is
dynamically trivial. This can not be expressed in OSSA, so I added code to
pred-deadalloc-elim so that I check if any of our available values after we
finish promoting away an allocation now need to have their consuming use set
completed.
2. That led me to discover that in certain cases load [take] that we were
promoting were available values of other load [take]. This means that we have a
memory safety issue if we promote one load before the other. Consider the
following SIL:
```
%mem = alloc_stack
store %arg to [init] %mem
%0 = load [take] %mem
store %0 to [init] %mem
%1 = load [take] %mem
destroy_value %1
dealloc_stack %mem
```
In this case, if we eliminate %0 before we eliminate %1, we will have a stale
pointer to %0.
I also took this as an opportunity to turn off predictable mem access opt on SIL
that was deserialized canonicalized and non-OSSA SIL. We evidently need to still
do this for pred mem opts for perf reasons (not sure why). But I am pretty sure
this isn't needed and allows me to avoid some nasty code.
In SILCombine, we do not want to add or delete edges. We are ok with swapping
edges or replacing edges when the CFG structure is preserved. This becomes an
issue since by performing this optimization, we are going to get rid of the
error parameter but leave a try_apply, breaking SIL invariants. So to do perform
this optimization, we would need to convert to an apply and eliminate the error
edge, breaking the aforementioned SILCombine invariant. So, just do not perform
this for now and leave it to other passes like SimplifyCFG.
strong_retain/strong_release with an optional reference as operand are rejected by the verifier and are not supported by IRGen.
SILCombine created such retains/releases when optimizing ref_cast instructions.
This fixes a compiler crash.
rdar://74146617
This PR just removes an unnecessary error raised in
OwnershipLifetimeExtender::createPlusOneCopy.
We can ownership rauw a value inside the loop with a value outside the
loop. findJointPostDominatingSet correctly helps create control
equivalent copies inside the loop for replacement.
