This mandatory pass inserts struct_extract operations before earlier
stores to the aggregate but didn't set the debug location of those new
instructions to the location of the store, which caused unexpected
stepping behavior in the debugger.
Fixes a regression from e74367f2b3.
<rdar://problem/35459092>
This simplifies the code and enables me to extract out the code for deleting
dead allocations into a separate struct from AllocOptimize. Otherwise, I would
have to pass in a reference to AllocOptimize creating back references, a
complexity that isn't necessary and will force me to break encapsulation
properties.
Just chopping this off of a larger commit.
rdar://31521023
Previously, we just always promoted destroy_addr. With ownership, this does not
work as well since we need to be able to reason about the take operation that we
are performing. In the general case, this would require adding code to
pred-memopts for tracking reads and for compensation store code since we would
need to eliminate the store of the taken value to prevent a double use.
I am going to loop back around later in the year and add back this code once the
time is available. I filed SR-6341. In case any third party contributor is
interested in looking at re-enabling this optimization before I get back to it.
rdar://31521023
Previously in PredMemOpts, we would insert any extracts at the load site, i.e.:
store %foo to %0
...
%1 = struct_element_addr %0, $Type, $Type.field
%2 = load %1
...
apply %use(%2)
would transform to:
store %foo to %0
...
%2 = struct_extract %foo
apply %use(%2)
This framework will not work with Ownership enabled since the value stored is
considered consumed by the store. This change fixes the issue by moving such
non-destructive extracts to occur while %foo is considered live, i.e. before the
store:
%2 = struct_extract %foo
store %foo to %0
...
apply %use(%2)
This means that we have to store insertion points for each store that provides
us with available values and insert the extracts at those points. This creates
some complications in the case where we have multiple stores since we need to
deal with phi nodes. Rather than dealing with it by hand, we just insert the
extracts at each point and then use the SSA updater to insert the relevant phi
nodes.
rdar://31521023
For pred-memopt to work with ownership, we need to insert instructions in two
different places:
1. When loading not-available values, we must insert the load at the site of the
load we are trying to promote.
2. When propagating an available value, we must destructure and or copy before
each one of the stores that provide our value.
By changing AvailableValue to be a struct, I am providing a cleaner API, but
more importantly the ability to start tracking that information.
rdar://31521023
In non-ownership SIL, PMOpt did not need to distinguish in between destructive
and non-destructive sub element extraction. This allowed PMOpt to just use the
"non-destructive" sub element (i.e. extractSubElement) everywhere. In
preparation for introducing this distinction (and a destructive sub element
extractor), rename extractSubElement to be nonDestructivelyExtractSubElement for
clarity.
NFC.
rdar://31521023
AssignInst is eliminated by DI so we should /never/ see any AssignInst once DI
runs. So this code is dead.
Just shaved off of a larger commit to make it easier to review the larger
commit.
rdar://31521023
introduce a common superclass, SILNode.
This is in preparation for allowing instructions to have multiple
results. It is also a somewhat more elegant representation for
instructions that have zero results. Instructions that are known
to have exactly one result inherit from a class, SingleValueInstruction,
that subclasses both ValueBase and SILInstruction. Some care must be
taken when working with SILNode pointers and testing for equality;
please see the comment on SILNode for more information.
A number of SIL passes needed to be updated in order to handle this
new distinction between SIL values and SIL instructions.
Note that the SIL parser is now stricter about not trying to assign
a result value from an instruction (like 'return' or 'strong_retain')
that does not produce any.
At some point, pass definitions were heavily macro-ized. Pass
descriptive names were added in two places. This is not only redundant
but a source of confusion. You could waste a lot of time grepping for
the wrong string. I removed all the getName() overrides which, at
around 90 passes, was a fairly significant amount of code bloat.
Any pass that we want to be able to invoke by name from a tool
(sil-opt) or pipeline plan *should* have unique type name, enum value,
commend-line string, and name string. I removed a comment about the
various inliner passes that contradicted that.
Side note: We should be consistent with the policy that a pass is
identified by its type. We have a couple passes, LICM and CSE, which
currently violate that convention.
Handling address_to_pointer as a plain inout missed some mutations and lead to miscompiles.
We now treat address_to_pointer as escaping address.
Fixes SR-3554
Applying nontrivial generic arguments to a nontrivial SIL layout requires lowered SILType substitution, which requires a SILModule. NFC yet, just an API change.
There's no longer a single element type to speak of. Update uses to either iterate all box fields or to assert that they're working with a single-field box.
Today, loads and stores are treated as having @unowned(unsafe) ownership
semantics. This leaves the user to specify ownership changes on the loaded or
stored value independently of the load/store by inserting ARC operations. With
the change to Semantic SIL, this will no longer be true. Instead loads, stores
have ownership semantics that one must reason about such as copy, take, and
trivial.
This change moves us closer to that world by eliminating the default
OwnershipQualification argument from create{Load,Store}. This means that the
compiler developer cannot ignore reasoning about the ownership semantics of the
memory operation that they are creating.
Operationally, this is a NFC change since I have just gone through the compiler
and updated all places where we create loads, stores to pass in the former
default argument ({Load,Store}OwnershipQualifier::Unqualified), to
SILBuilder::create{Load,Store}(...). For now, one can just do that in situations
where one needs to create loads/stores, but over time, I am going to tighten the
semantics up via the verifier.
rdar://28685236
This instruction creates a "virtual" address to represent a property with a behavior that supports definite initialization. The instruction holds references to functions that perform the initialization and 'set' logic for the property. It will be DI's job to rewrite assignments into this virtual address into calls to the initializer or setter based on the initialization state of the property at the time of assignment.
As there are no instructions left which produce multiple result values, this is a NFC regarding the generated SIL and generated code.
Although this commit is large, most changes are straightforward adoptions to the changes in the ValueBase and SILValue classes.
And use project_box to get to the address value.
SILGen now generates a project_box for each alloc_box.
And IRGen re-uses the address value from the alloc_box if the operand of project_box is an alloc_box.
This lets the generated code be the same as before.
Other than that most changes of this (quite large) commit are straightforward.
(libraries now)
It has been generally agreed that we need to do this reorg, and now
seems like the perfect time. Some major pass reorganization is in the
works.
This does not have to be the final word on the matter. The consensus
among those working on the code is that it's much better than what we
had and a better starting point for future bike shedding.
Note that the previous organization was designed to allow separate
analysis and optimization libraries. It turns out this is an
artificial distinction and not an important goal.
