The new test llvm/test/Transforms/CodeGenPrepare/remove-assume-block.ll
breaks on non-X86 machines. Change it to look like the existing test
llvm/test/Transforms/CodeGenPrepare/X86/delete-assume-dead-code.ll
to fix it.
Reviewed By: bkramer
Differential Revision: https://reviews.llvm.org/D97952
With reference types, tables can have non-zero table numbers. This
commit adds support for element sections against these tables.
Differential Revision: https://reviews.llvm.org/D97923
Recommit bf5a5826504754788a8f1e3fec7a7dc95cda5782. Depends on
4c8fb7ddd6fa49258e0e9427e7345fb56ba522d4 which was reverted.
RegBankSelect creates zext and trunc when it selects banks for uniform i1.
Add zext_trunc_fold from generic combiner to post RegBankSelect combiner.
Differential Revision: https://reviews.llvm.org/D95432
Recommit 4112299ee761a9b6a309c8ff4a7e75f8c8d8851b. Depends on
4c8fb7ddd6fa49258e0e9427e7345fb56ba522d4 which was reverted.
Combine zext(trunc x) to x when truncated bits are known to be zero.
Differential Revision: https://reviews.llvm.org/D96031
There are certain loops like this below:
for (int i = 0; i < n; i++) {
a[i] = b[i] + 1;
*inv = a[i];
}
that can only be vectorised if we are able to extract the last lane of the
vectorised form of 'a[i]'. For fixed width vectors this already works since
we know at compile time what the final lane is, however for scalable vectors
this is a different story. This patch adds support for extracting the last
lane from a scalable vector using a runtime determined lane value. I have
added support to VPIteration for runtime-determined lanes that still permit
the caching of values. I did this by introducing a new class called VPLane,
which describes the lane we're dealing with and provides interfaces to get
both the compile-time known lane and the runtime determined value. Whilst
doing this work I couldn't find any explicit tests for extracting the last
lane values of fixed width vectors so I added tests for both scalable and
fixed width vectors.
Differential Revision: https://reviews.llvm.org/D95139
The code was using the standard isalnum function which doesn't handle
values outside the non-ascii range. Switching to using llvm::isAlnum
instead ensures we don't provoke undefined behaviour, which can in some
cases result in crashes.
Reviewed by: MaskRay
Differential Revision: https://reviews.llvm.org/D97663
The test was showing that when --strip-unneeded is specified for an
executable, all the symbols are stripped. However, the set of symbols
used in the test would be stripped by --strip-unneeded for an ET_REL
object too. Fix this by adding additional symbols that aren't normally
stripped by --strip-unneeded.
Reviewed by: MaskRay
Differential Revision: https://reviews.llvm.org/D97664
This pass runs in any situations but we skip it when it is not O0 and the
function doesn't have optnone attribute. With -O0, the def of shape to amx
intrinsics is near the amx intrinsics code. We are not able to find a
point which post-dominate all the shape and dominate all amx intrinsics.
To decouple the dependency of the shape, we transform amx intrinsics
to scalar operation, so that compiling doesn't fail. In long term, we
should improve fast register allocation to allocate amx register.
Reviewed By: pengfei
Differential Revision: https://reviews.llvm.org/D93594
GCC warning:
```
In file included from /usr/include/c++/9/cassert:44,
from /home/vsts/work/1/llvm-project/llvm/include/llvm/ADT/BitVector.h:21,
from /home/vsts/work/1/llvm-project/llvm/include/llvm/Support/Program.h:17,
from /home/vsts/work/1/llvm-project/llvm/include/llvm/Support/Process.h:32,
from /home/vsts/work/1/llvm-project/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp:11:
/home/vsts/work/1/llvm-project/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp: In member function ‘virtual llvm::Expected<std::unique_ptr<llvm::jitlink::JITLinkMemoryManager::Allocation> > llvm::jitlink::InProcessMemoryManager::allocate(const llvm::jitlink::JITLinkDylib*, const SegmentsRequestMap&)’:
/home/vsts/work/1/llvm-project/llvm/lib/ExecutionEngine/JITLink/JITLinkMemoryManager.cpp:129:40: warning: comparison of unsigned expression >= 0 is always true [-Wtype-limits]
129 | assert(SlabRemaining.allocatedSize() >= 0 && "Mapping exceeds allocation");
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~
```
The return type of `allocatedSize()` is `size_t`, thus the expression
`SlabRemaining.allocatedSize() >= 0` always evaluate to `true`.
I'm not sure this would catch all such issues, but it would catch some.
The problem for PR49393 was that we were holding a reference to a node that
wasn't connect edto the DAG across a function that could delete unused nodes. In
this particular case we managed to try to use the deleted node while it was in
the deleted state before its memory got recycled.
It could also happen that we delete the node, something allocates a new node
which recycles the memory. Then we try to use the reference we were holding and
it is now a completely different node with different valid opcode. This patch
would not catch that.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D97969
For binary or ternary ops we call getNegatedExpression multiple
times and then compare costs. While we're doing this we need to
hold a node from the first call across the second call, but its
not yet attached to the DAG. Its possible the second call creates
an identical node and then decides it didn't need it so will try
to delete it if it has no uses. This can cause a reference to the
node we're holding further up the call stack to become invalidated.
To prevent this, we can use a HandleSDNode to artifically give
the node a use without connecting it to the DAG.
I've used a std::list of HandleSDNodes so we can create handles
only when we have a node to hold. HandleSDNode does not have
default constructor and cannot be copied or moved.
Fixes PR49393.
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D97914
Initial support for using the OpenMPIRBuilder by clang to generate loops using the OpenMPIRBuilder. This initial support is intentionally limited to:
* Only the worksharing-loop directive.
* Recognizes only the nowait clause.
* No loop nests with more than one loop.
* Untested with templates, exceptions.
* Semantic checking left to the existing infrastructure.
This patch introduces a new AST node, OMPCanonicalLoop, which becomes parent of any loop that has to adheres to the restrictions as specified by the OpenMP standard. These restrictions allow OMPCanonicalLoop to provide the following additional information that depends on base language semantics:
* The distance function: How many loop iterations there will be before entering the loop nest.
* The loop variable function: Conversion from a logical iteration number to the loop variable.
These allow the OpenMPIRBuilder to act solely using logical iteration numbers without needing to be concerned with iterator semantics between calling the distance function and determining what the value of the loop variable ought to be. Any OpenMP logical should be done by the OpenMPIRBuilder such that it can be reused MLIR OpenMP dialect and thus by flang.
The distance and loop variable function are implemented using lambdas (or more exactly: CapturedStmt because lambda implementation is more interviewed with the parser). It is up to the OpenMPIRBuilder how they are called which depends on what is done with the loop. By default, these are emitted as outlined functions but we might think about emitting them inline as the OpenMPRuntime does.
For compatibility with the current OpenMP implementation, even though not necessary for the OpenMPIRBuilder, OMPCanonicalLoop can still be nested within OMPLoopDirectives' CapturedStmt. Although OMPCanonicalLoop's are not currently generated when the OpenMPIRBuilder is not enabled, these can just be skipped when not using the OpenMPIRBuilder in case we don't want to make the AST dependent on the EnableOMPBuilder setting.
Loop nests with more than one loop require support by the OpenMPIRBuilder (D93268). A simple implementation of non-rectangular loop nests would add another lambda function that returns whether a loop iteration of the rectangular overapproximation is also within its non-rectangular subset.
Reviewed By: jdenny
Differential Revision: https://reviews.llvm.org/D94973
By implementing the method "unsigned RISCVTTIImpl::getRegisterBitWidth(bool Vector)",
fixed-length vectorization is enabled when possible. Without this method, the
"#pragma clang loop" directive is needed to enable vectorization(or the cost model
may inform LLVM that "Vectorization is possible but not beneficial").
Reviewed By: frasercrmck
Differential Revision: https://reviews.llvm.org/D97549
sample loader pass.
In https://reviews.llvm.org/rG5fb65c02ca5e91e7e1a00e0efdb8edc899f3e4b9,
to prevent repeated indirect call promotion for the same indirect call
and the same target, we used zero-count value profile to indicate an
indirect call has been promoted for a certain target. We removed
PromotedInsns cache in the same patch. However, there was a problem in
that patch described below, and that problem led me to add PromotedInsns
back as a mitigation in
https://reviews.llvm.org/rG4ffad1fb489f691825d6c7d78e1626de142f26cf.
When we get value profile from metadata by calling getValueProfDataFromInst,
we need to specify the maximum possible number of values we expect to read.
We uses MaxNumPromotions in the last patch so the maximum number of value
information extracted from metadata is MaxNumPromotions. If we have many
values including zero-count values when we write the metadata, some of them
will be dropped when we read them because we only read MaxNumPromotions
values. It will allow repeated indirect call promotion again. We need to
make sure if there are values indicating promoted targets, those values need
to be saved in metadata with higher priority than other values.
The patch fixed that problem. We change to use -1 to represent the count
of a promoted target instead of 0 so it is easier to sort the values.
When we prepare to update the metadata in updateIDTMetaData, we will sort
the values in the descending count order and extract only MaxNumPromotions
values to write into metadata. Since -1 is the max uint64_t number, if we
have equal to or less than MaxNumPromotions of -1 count values, they will
all be kept in metadata. If we have more than MaxNumPromotions of -1 count
values, we will only save MaxNumPromotions such values maximally. In such
case, we have logic in place in doesHistoryAllowICP to guarantee no more
promotion in sample loader pass will happen for the indirect call, because
it has been promoted enough.
With this change, now we can remove PromotedInsns without problem.
Differential Revision: https://reviews.llvm.org/D97350
As a preparation step for fast8 support, we need to update the tests
to pass in both modes. That requires generalizing the shadow width
and remove any hard coded references that assume it's always 2 bytes.
Reviewed By: stephan.yichao.zhao
Differential Revision: https://reviews.llvm.org/D97988
Lorenz Bauer from Cloudflare tried to use "const struct <name>"
as the type for __builtin_btf_type_id(*(const struct <name>)0, 1)
relocation and hit a llvm BPF fatal error.
https://lore.kernel.org/bpf/a3782f71-3f6b-1e75-17a9-1827822c2030@fb.com/
...
fatal error: error in backend: Empty type name for BTF_TYPE_ID_REMOTE reloc
Currently, we require the debuginfo type itself must have a name.
In this case, the debuginfo type is "const" which points to "struct <name>".
The "const" type does not have a name, hence the above fatal error
will be triggered.
Let us permit "const" and "volatile" type modifiers. We skip modifiers
in some other cases as well like structure member type tracing.
This can aviod the above fatal error.
Differential Revision: https://reviews.llvm.org/D97986
This is included from IR files, and IR doesn't/can't depend on Analysis
(because Analysis depends on IR).
Also fix the implementation - don't use non-member static in headers, as
it leads to ODR violations, inaccurate "unused function" warnings, etc.
And fix the header protection macro name (we don't generally include
"LIB" in the names, so far as I can tell).
This is actually two changes. One is to avoid copies when fp values are fed into
a build_vector, without being able to tell from the opcode.
The other is that build_vectors are also marked as only defining FP, since they
produce vector results.
Differential Revision: https://reviews.llvm.org/D97968
This is a case D97677 missed. When taking out remaining BBs that are
reachable from already-taken-out exceptions (because they are not
subexcptions but unwind destinations), I assumed the remaining BBs are
not EH pads, but they can be. For example,
```
try {
try {
throw 0;
} catch (int) { // (a)
}
} catch (int) { // (b)
}
try {
foo();
} catch (int) { // (c)
}
```
In this code, (b) is the unwind destination of (a) so its exception is
taken out of (a)'s exception, But even though the next try-catch is not
inside the first two-level try-catches, because the first try always
throws, its continuation BB is unreachable and the whole rest of the
function is dominated by EH pad (a), including EH pad (c). So after we
take out of (b)'s exception out of (a)'s, we also need to take out (c)'s
exception out of (a)'s, because (c) is reachable from (b).
This adds one more step before what we did for remaining BBs in D97677;
it traverses EH pads first to take subexceptions out of their incorrect
parent exception. It's the same thing as D97677, but because we can do
this before we add BBs to exceptions' sets, we don't need to fix sets
and only need to fix parent exception pointers.
Other changes are variable name changes (I changed `WE` -> `SrcWE`,
`UnwindWE` -> `DstWE` for clarity), some comment changes, and a drive-by
fix in a bug in a `LLVM_DEBUG` print statement.
Fixes https://github.com/emscripten-core/emscripten/issues/13588.
Reviewed By: dschuff
Differential Revision: https://reviews.llvm.org/D97929
Background:
Wasm EH, while using Windows EH (catchpad/cleanuppad based) IR, uses
Itanium-based libraries and ABIs with some modifications.
`__clang_call_terminate` is a wrapper generated in Clang's Itanium C++
ABI implementation. It contains this code, in C-style pseudocode:
```
void __clang_call_terminate(void *exn) {
__cxa_begin_catch(exn);
std::terminate();
}
```
So this function is a wrapper to call `__cxa_begin_catch` on the
exception pointer before termination.
In Itanium ABI, this function is called when another exception is thrown
while processing an exception. The pointer for this second, violating
exception is passed as the argument of this `__clang_call_terminate`,
which calls `__cxa_begin_catch` with that pointer and calls
`std::terminate` to terminate the program.
The spec (https://libcxxabi.llvm.org/spec.html) for `__cxa_begin_catch`
says,
```
When the personality routine encounters a termination condition, it
will call __cxa_begin_catch() to mark the exception as handled and then
call terminate(), which shall not return to its caller.
```
In wasm EH's Clang implementation, this function is called from
cleanuppads that terminates the program, which we also call terminate
pads. Cleanuppads normally don't access the thrown exception and the
wasm backend converts them to `catch_all` blocks. But because we need
the exception pointer in this cleanuppad, we generate
`wasm.get.exception` intrinsic (which will eventually be lowered to
`catch` instruction) as we do in the catchpads. But because terminate
pads are cleanup pads and should run even when a foreign exception is
thrown, so what we have been doing is:
1. In `WebAssemblyLateEHPrepare::ensureSingleBBTermPads()`, we make sure
terminate pads are in this simple shape:
```
%exn = catch
call @__clang_call_terminate(%exn)
unreachable
```
2. In `WebAssemblyHandleEHTerminatePads` pass at the end of the
pipeline, we attach a `catch_all` to terminate pads, so they will be in
this form:
```
%exn = catch
call @__clang_call_terminate(%exn)
unreachable
catch_all
call @std::terminate()
unreachable
```
In `catch_all` part, we don't have the exception pointer, so we call
`std::terminate()` directly. The reason we ran HandleEHTerminatePads at
the end of the pipeline, separate from LateEHPrepare, was it was
convenient to assume there was only a single `catch` part per `try`
during CFGSort and CFGStackify.
---
Problem:
While it thinks terminate pads could have been possibly split or calls
to `__clang_call_terminate` could have been duplicated,
`WebAssemblyLateEHPrepare::ensureSingleBBTermPads()` assumes terminate
pads contain no more than calls to `__clang_call_terminate` and
`unreachable` instruction. I assumed that because in LLVM very limited
forms of transformations are done to catchpads and cleanuppads to
maintain the scoping structure. But it turned out to be incorrect;
passes can merge cleanuppads into one, including terminate pads, as long
as the new code has a correct scoping structure. One pass that does this
I observed was `SimplifyCFG`, but there can be more. After this
transformation, a single cleanuppad can contain any number of other
instructions with the call to `__clang_call_terminate` and can span many
BBs. It wouldn't be practical to duplicate all these BBs within the
cleanuppad to generate the equivalent `catch_all` blocks, only with
calls to `__clang_call_terminate` replaced by calls to `std::terminate`.
Unless we do more complicated transformation to split those calls to
`__clang_call_terminate` into a separate cleanuppad, it is tricky to
solve.
---
Solution (?):
This CL just disables the generation and use of `__clang_call_terminate`
and calls `std::terminate()` directly in its place.
The possible downside of this approach can be, because the Itanium ABI
intended to "mark" the violating exception handled, we don't do that
anymore. What `__cxa_begin_catch` actually does is increment the
exception's handler count and decrement the uncaught exception count,
which in my opinion do not matter much given that we are about to
terminate the program anyway. Also it does not affect info like stack
traces that can be possibly shown to developers.
And while we use a variant of Itanium EH ABI, we can make some
deviations if we choose to; we are already different in that in the
current version of the EH spec we don't support two-phase unwinding. We
can possibly consider a more complicated transformation later to
reenable this, but I don't think that has high priority.
Changes in this CL contains:
- In Clang, we don't generate a call to `wasm.get.exception()` intrinsic
and `__clang_call_terminate` function in terminate pads anymore; we
simply generate calls to `std::terminate()`, which is the default
implementation of `CGCXXABI::emitTerminateForUnexpectedException`.
- Remove `WebAssembly::ensureSingleBBTermPads() function and
`WebAssemblyHandleEHTerminatePads` pass, because terminate pads are
already `catch_all` now (because they don't need the exception
pointer) and we don't need these transformations anymore.
- Change tests to use `std::terminate` directly. Also removes tests that
tested `LateEHPrepare::ensureSingleBBTermPads` and
`HandleEHTerminatePads` pass.
- Drive-by fix: Add some function attributes to EH intrinsic
declarations
Fixes https://github.com/emscripten-core/emscripten/issues/13582.
Reviewed By: dschuff, tlively
Differential Revision: https://reviews.llvm.org/D97834
This code assumed that FP math was only permissable if it was
fully "fast", so it hard-coded "fast" when creating new instructions.
The underlying code already allows matching recurrences/reductions
that are only "reassoc", so this change should prevent the potential
miscompile seen in the test diffs (we created "fast" ops even though
none existed in the original code).
I don't know if we need to create the temporary IRBuilder objects
used here, so that could be follow-up clean-up.
There's an open question about whether we should require "nsz" in
addition to "reassoc" here. InstCombine uses that combo for its
reassociative folds, but I think codegen is not as strict.
The hazard where a VMEM reads an SGPR written by a VALU counts as a data
dependency hazard, so no nops are required on GFX10. Tested with Vulkan
CTS on GFX10.1 and GFX10.3.
Differential Revision: https://reviews.llvm.org/D97926
BasicAA knows how to analyze phis, but to control compile time, we're fairly limited in doing so. This patch loosens that restriction just slightly when there is exactly one phi input (after discounting induction variable increments). The result of this is that we can handle more cases around nested and sibling loops with pointer induction variables.
A few points to note.
* This is deliberately extremely restrictive about recursing through at most one input of the phi. There's a known general problem with BasicAA sometimes hitting exponential compile time already, and this patch makes every effort not to compound the problem. Once the root issue is fixed, we can probably loosen the restrictions here a bit.
* As seen in the test file, we're still missing cases which aren't *directly* based on phis (e.g. using the indvar increment). I believe this to be a separate problem and am going to explore this in another patch once this one lands.
* As seen in the test file, this results in the unfortunate fact that using phivalues sometimes results in worse quality results. I believe this comes down to an oversight in how recursive phi detection was implemented for phivalues. I'm happy to tackle this in a follow up change.
Differential Revision: https://reviews.llvm.org/D97401
Add more expectations in vector.ll and select.ll based on command-line option combinations.
Also, remove hard-coded shadow width references to enable fast8 transition.
Reviewed By: stephan.yichao.zhao
Differential Revision: https://reviews.llvm.org/D97903
This enhances the auto-init remark with information about the variable
that is auto-initialized.
This is based of debug info if available, or alloca names (mostly for
development purposes).
```
auto-init.c:4:7: remark: Call to memset inserted by -ftrivial-auto-var-init. Memory operation size: 4096 bytes.Variables: var (4096 bytes). [-Rpass-missed=annotation-remarks]
int var[1024];
^
```
This allows to see things like partial initialization of a variable that
the optimizer won't be able to completely remove.
Differential Revision: https://reviews.llvm.org/D97734
This is recommit of 4c8fb7ddd6fa49258e0e9427e7345fb56ba522d4.
MIR in one unit test had mismatched types.
For vectors we consider a bit as known if it is the same for all demanded
vector elements (all elements by default). KnownBits BitWidth for vector
type is size of vector element. Add support for G_BUILD_VECTOR.
This allows combines of urem_pow2_to_mask in pre-legalizer combiner.
Differential Revision: https://reviews.llvm.org/D96122
This reverts commit 7479a2e00bc41f399942e5106fbdf9b4b0c11506.
This commit causes compile errors on clang-x64-windows-msvc, so I'm
reverting the patch for now.
For reference, the error in question is:
```
error C2280: 'llvm::raw_ostream_iterator<char,char>
&llvm::raw_ostream_iterator<char,char>::operator =(const
llvm::raw_ostream_iterator<char,char> &)': attempting to reference a deleted
function
note: compiler has generated 'llvm::raw_ostream_iterator<char,char>::operator ='
here
note: 'llvm::raw_ostream_iterator<char,char>
&llvm::raw_ostream_iterator<char,char>::operator =(const
llvm::raw_ostream_iterator<char,char> &)': function was implicitly deleted
because 'llvm::raw_ostream_iterator<char,char>' has a data member
'llvm::raw_ostream_iterator<char,char>::OutStream' of reference type
```
This change adds '-use-interfacestub' option to allow llvm-ifs
to use InterfaceStub lib when generating ELF binary.
Differential Revision: https://reviews.llvm.org/D94461
explicitly emitting retainRV or claimRV calls in the IR
This reapplies ed4718eccb12bd42214ca4fb17d196d49561c0c7, which was reverted
because it was causing a miscompile. The bug that was causing the miscompile
has been fixed in 75805dce5ff874676f3559c069fcd6737838f5c0.
Original commit message:
Background:
This fixes a longstanding problem where llvm breaks ARC's autorelease
optimization (see the link below) by separating calls from the marker
instructions or retainRV/claimRV calls. The backend changes are in
https://reviews.llvm.org/D92569.
https://clang.llvm.org/docs/AutomaticReferenceCounting.html#arc-runtime-objc-autoreleasereturnvalue
What this patch does to fix the problem:
- The front-end adds operand bundle "clang.arc.attachedcall" to calls,
which indicates the call is implicitly followed by a marker
instruction and an implicit retainRV/claimRV call that consumes the
call result. In addition, it emits a call to
@llvm.objc.clang.arc.noop.use, which consumes the call result, to
prevent the middle-end passes from changing the return type of the
called function. This is currently done only when the target is arm64
and the optimization level is higher than -O0.
- ARC optimizer temporarily emits retainRV/claimRV calls after the calls
with the operand bundle in the IR and removes the inserted calls after
processing the function.
- ARC contract pass emits retainRV/claimRV calls after the call with the
operand bundle. It doesn't remove the operand bundle on the call since
the backend needs it to emit the marker instruction. The retainRV and
claimRV calls are emitted late in the pipeline to prevent optimization
passes from transforming the IR in a way that makes it harder for the
ARC middle-end passes to figure out the def-use relationship between
the call and the retainRV/claimRV calls (which is the cause of
PR31925).
- The function inliner removes an autoreleaseRV call in the callee if
nothing in the callee prevents it from being paired up with the
retainRV/claimRV call in the caller. It then inserts a release call if
claimRV is attached to the call since autoreleaseRV+claimRV is
equivalent to a release. If it cannot find an autoreleaseRV call, it
tries to transfer the operand bundle to a function call in the callee.
This is important since the ARC optimizer can remove the autoreleaseRV
returning the callee result, which makes it impossible to pair it up
with the retainRV/claimRV call in the caller. If that fails, it simply
emits a retain call in the IR if retainRV is attached to the call and
does nothing if claimRV is attached to it.
- SCCP refrains from replacing the return value of a call with a
constant value if the call has the operand bundle. This ensures the
call always has at least one user (the call to
@llvm.objc.clang.arc.noop.use).
- This patch also fixes a bug in replaceUsesOfNonProtoConstant where
multiple operand bundles of the same kind were being added to a call.
Future work:
- Use the operand bundle on x86-64.
- Fix the auto upgrader to convert call+retainRV/claimRV pairs into
calls with the operand bundles.
rdar://71443534
Differential Revision: https://reviews.llvm.org/D92808
Adds a class `raw_ostream_iterator` that behaves like
std::ostream_iterator, but can be used with raw_ostream.
This is useful for using raw_ostream with std algorithms.
For example, it can be used to output std containers as follows:
```
std::vector<int> V = { 1, 2, 3 };
std::copy(V.begin(), V.end(), raw_ostream_iterator<int>(outs(), ", "));
// Output: "1, 2, 3, "
```
The API tries to follow std::ostream_iterator as closely as is
practically possible.
Reviewed By: dblaikie, mkitzan
Differential Revision: https://reviews.llvm.org/D78795
