All relevant operations have been switched to primarily use the strided
layout, but still support the affine map layout. Update the relevant
tests to use the strided format instead for compatibility with how ops
now print by default.
Reviewed By: nicolasvasilache
Differential Revision: https://reviews.llvm.org/D134045
This reverts commit 01ffe31cbb.
A build breakage with GCC 7.3 has been reported:
https://reviews.llvm.org/D132311#3797053
FWIW, GCC 7.5 is OK according to Pavel Chupin. I also personally
tested GCC 8.4.0.
Use salvageDebugInfo for instructions erased as trivially dead in
GlobalISel.
It would be helpful to implement support of G_PTR_ADD and G_FRAME_INDEX
in salvageDebugInfo in future in order to preserve more variable
location.
Reviewed by: arsenm
Differential Revision: https://reviews.llvm.org/D133986
Now libc headers can be installed separately from installing the rest of
the libc.
Reviewed By: sivachandra
Differential Revision: https://reviews.llvm.org/D133960
DefaultInlineOrder was largely an exercise in generalizing the
traversal order of call sites within the inliner.
Now that the module inliner is starting to form its shape, there is no
point in sharing DefaultInlineOrder between the module inliner and the
CGSCC inliner. DefaultInlineOrder and all the other inline orders are
mutually exclusive in the following sense:
- The use of DefaultInlineOrder doesn't make sense in the module
inliner because there is no priority inherent in the order in which
call sites are added to the list of call sites -- SmallVector.
- The use of any other inline order doesn't make sense in the CGSCC
inliner because little prioritization can be done within one CGSCC.
This patch essentially reverts the addition of DefaultInlineOrder so
that the loop structure of Inliner.cpp looks like the state just
before we started working on the module inliner (circa June 2021).
At the same time, ww remove the choice of DefaultInlineOrder from
UseInlinePriority.
Differential Revision: https://reviews.llvm.org/D134080
Added the mask and the analysis of the buildvector sequence in the
isUndefVector function, improves codegen and cost estimation.
Metric: SLP.NumVectorInstructions
Program SLP.NumVectorInstructions
results results0 diff
test-suite :: External/SPEC/CFP2017rate/526.blender_r/526.blender_r.test 27362.00 27360.00 -0.0%
Metric: size..text
Program size..text
results results0 diff
test-suite :: External/SPEC/CFP2017rate/508.namd_r/508.namd_r.test 805299.00 806035.00 0.1%
526.blender_r - some extra code is vectorized.
508.namd_r - some extra code is optimized out.
Differential Revision: https://reviews.llvm.org/D133891
Some HLSL functionality is gated on the target shader model version.
Enabling the use of availability markup allows us to diagnose
availability issues easily in the frontend.
Reviewed By: erichkeane
Differential Revision: https://reviews.llvm.org/D134067
This is a follow-up to D133777, which resolved a use-after-free case but
did not cover all possible memory bugs due to misplacement of loads.
In short, the overall problem was that sinked loads could be moved after
state-modifying instructions leading to memory bugs.
The solution is to restrict load sinking unless it is found to be sound.
i) Within a basic block (to-be-sinked load and select-user are in the same BB),
loads can be sinked only if there is no intervening state-modifying instruction.
This is a conservative approach to avoid resorting to alias analysis to detect
potential memory overlap.
ii) Across basic blocks, sinking of loads is avoided. This is because going over
multiple basic blocks looking for memory conflicts could be computationally
expensive and also unlikely to allow loads to sink. Further, experiments showed
that not sinking these loads has a slight positive performance effect.
Maybe for some of these loads, having some separation allows enough time
for the load to be executed in time for its user. This is not the case for
floating point operations that benefit more from sinking.
The solution in D133777 was essentially undone in this patch,
since the latter is a complete solution to the observed problem.
Overall, the performance impact of this patch is minimal.
Tested on two internal Google workloads with instrPGO.
Search application showed <0.05% perf difference,
while the database one showed a slight improvement,
but not statistically significant.
Reviewed By: davidxl
Differential Revision: https://reviews.llvm.org/D133999
This fixes https://github.com/llvm/llvm-project/issues/57616.
Type test lowering in ThinLTO modules relies on having type id
summaries set up for the referenced types, which provide the type
test resolution. If there is no summary, the type tests are lowered
to false. At the very least, a default type id summary gives the
type tests a resolution of Unknown, which is handled correctly (ignored
by the first invocation of LTT, and lowered to true by the second).
WPD sets up the type id summaries (with a default type test resolution)
as it is processing the type tests, but only does this for the patterns
handled by WPD, which is a type test directly feeding an assume. In the
case of type tests feeding an assume via a phi, the type id summary was
not being set up, leading to the type tests being lowered to false
incorrectly.
Fix this by adding the default type id summary entries for all type ids
used on globals during index-only WPD.
This is not an issue for hybrid (split-lto-unit) LTO, as in that case
the type test resolution is determined and set up during LTT, since the
type definitions are in the regular LTO split module, and exported via
the summary to the ThinLTO split module.
Differential Revision: https://reviews.llvm.org/D134012
If we want to be able to close and then re-open a library then we need to reset
the data section states when the library is closed. This commit updates
MachOPlatform and the ORC runtime to track __data and __common sections, and
reset the state in MachOPlatformRuntimeState::dlcloseDeinitialize.
This is only a first step to full support -- there are other data sections that
we're not capturing, and we'll probably want a more efficient representation
for the sections (rather than passing their string name over IPC), but this is
a reasonable first step.
This commit also contains a fix to MapperJITLinkMemoryManager that prevents it
from calling OnDeallocated twice in the case of an error.
When we derive EFMM from SectionMemoryManager, it brings into EFMM extra
functionality, such as the registry of exception handling sections,
page permission management, etc. Such functionality is of no use to
llvm-bolt and can even be detrimental (see
https://github.com/llvm/llvm-project/issues/56726).
Change the base class of ExecutableFileMemoryManager to MemoryManager,
avoid registering EH sections, and skip memory finalization.
Fixes#56726
Reviewed By: yota9
Differential Revision: https://reviews.llvm.org/D133994
In non-relocation mode, every function is emitted in its own section. If
a function is empty, RuntimeDyld will still allocate 1-byte section
for the function and initialize it with zero. As a result, we will
overwrite the first byte of the original function contents with zero.
Such scenario can happen when the input function had only NOP
instructions which BOLT removes by default. Even though such functions
likely cause undefined behavior, it's better to preserve their contents.
Reviewed By: yota9
Differential Revision: https://reviews.llvm.org/D133978
This is a partial port of the code used by the SelectionDAGBuilder to
translate selects.
In particular, see matchSelectPattern in ValueTracking.cpp. This is a
GISel-equivalent of the portion which handles fminnum/fmaxnum/fminimum/fmaximum.
I tried to set it up so it'd be easy to add the non-FP cases. Those are simpler.
On the AArch64-end, it seems like the FP cases are more important for perf
right now, so I bit the bullet and went at the more complicated problem. :)
I elected to do this as a post-legalize combine rather than in the
IRTranslator because
Deciding which fmax/fmin to use can depend on legalization rules
Philosophically-speaking (TM), putting it in a combine just feels cleaner
Being able to enable/disable the combine is handy
Another option would be to use the ValueTracking code in the IRTranslator and
match what SelectionDAGBuilder::visitSelect does. I think that may be somewhat
annoying since we'd need to write lowerings back into the selects in the
legalizer. I'm not strongly opposed to the approach.
We'd also want to be careful with vector selects once that's implemented,
which explicitly check if a vector select is legal on the target. That'd
probably need a hook.
From what I can tell, doing this as a combine is probably a cleaner option
long-term.
Differential Revision: https://reviews.llvm.org/D116702
Currently, it can become extremely costly to compute MayIncreasePressure if the
size of CSUses turns out to be very large. In that case, it's no longer cost
effective to keep computing MayIncreasePressure. Therefore, to limit the amount
of time spent in isProfitableToCSE, we simply conservatively assume
MayIncreasePressure if the size of CSUses is too large. This can reduce overall
compile time by 30% for some benchmarks.
Reviewed By: arsenm
Differential Revision: https://reviews.llvm.org/D134003
I want to default all VP operations to Expand. These 2 were blocking
because VE doesn't support them and the tests were expecting them
to fail a specific way. Using Expand caused them to fail differently.
Seemed better to emulate them using operations that are supported.
@simoll mentioned on Discord that VE has some expansion downstream. Not
sure if its done like this or in the VE target.
Reviewed By: frasercrmck, efocht
Differential Revision: https://reviews.llvm.org/D133514
Summary:
This patch enables libc++ LIT test case last_write_time.pass.cpp for AIX. Because system call utimensat() of AIX which is used in the libc++ implementation of last_write_time() does not accept the times parameter with a negative tv_sec or tv_nsec field, testing of setting file time to before epoch time is excluded for AIX.
Reviewed by: ldionne, libc++
Differential Revision: https://reviews.llvm.org/D133124
Before this patch, when compiling an IR file (eg the .llvmbc section
from an object file compiled with -Xclang -fembed-bitcode=all) and
profile data was passed in using the -fprofile-instrument-use-path
flag, there would be no error printed (as the previous implementation
relied on the error getting caught again in the constructor of
CodeGenModule which isn't called when -x ir is set). This patch
moves the error checking directly to where the error is caught
originally rather than failing silently in setPGOUseInstrumentor and
waiting to catch it in CodeGenModule to print diagnostic information to
the user.
Regression test added.
Reviewed By: xur, mtrofin
Differential Revision: https://reviews.llvm.org/D132991
These classes are referred to only from getInlineOrder in
InlineOrder.cpp. This patch hides the entire class declarations and
definitions in InlineOrder.cpp.
Differential Revision: https://reviews.llvm.org/D134056
This patch modifies the testcase to use error substitution so it will pass on all platforms.
Reviewed By: fanbo-meng, zibi
Differential Revision: https://reviews.llvm.org/D134034
For functions with references to internal offsets from data, verify externally
referenced blocks against the set of jump table targets. Mark the function
as non-simple if there are any unclaimed data to code references.
Reviewed By: #bolt, maksfb
Differential Revision: https://reviews.llvm.org/D132495
In a separate patch I want to modify ValueTracking.h. When I touch the
header, arc wants to clang-format the lines I touch (reasonable!). But
then these whitespace changes get mixed into my patch.
Adds support for default arguments in the internal representation and reads these values from the source. Implements writing these values to YAML but does not implement this for the HTML or markdown outputs.
Reviewed By: paulkirth
Differential Revision: https://reviews.llvm.org/D133732
Before, it returns the outermost blocks if nested blocks have the same
address range. That casuses lldb unable to find variables that are inside
inner blocks.
Reviewed By: labath
Differential Revision: https://reviews.llvm.org/D133601
The `ScanInstance` is a local variable in `DependencyScanningAction::runInvocation()` that is referenced by `ModuleDepCollector`. Since D132405, `ModuleDepCollector` can escape the function and can outlive its `ScanInstance`. This patch fixes that.
Reviewed By: benlangmuir
Differential Revision: https://reviews.llvm.org/D133988
The uncached lookup is mainly used in the ASTImporter/LLDB code-path
where we're not allowed to load from external storage. When importing
a FieldDecl with a DeclContext that had no external visible storage
(but came from a Clang module or PCH) the above call to `lookup(Name)`
the regular `DeclContext::lookup` fails because:
1. `DeclContext::buildLookup` doesn't set `LookupPtr` for decls
that came from a module
2. LLDB doesn't use the `SharedImporterState`
In such a case we would never continue with the "slow" path of iterating
through the decl chain on the DeclContext. In some cases this means that
ASTNodeImporter::VisitFieldDecl ends up importing a decl into the
DeclContext a second time.
The patch removes the short-circuit in the case where we don't find
any decls via the regular lookup.
**Tests**
* Un-skip the failing LLDB API tests
Differential Revision: https://reviews.llvm.org/D133945
The problem here is that the ASTImporter adds
the template class member FieldDecl to
the DeclContext twice. This happens because
we don't construct a `LookupPtr` for decls
that originate from modules and thus the
ASTImporter never realizes that the FieldDecl
has already been imported. These duplicate
decls then break the assumption of the LayoutBuilder
which expects only a single member decl to
exist.
The test will be fixed by a follow-up revision
and is thus skipped for now.
Differential Revision: https://reviews.llvm.org/D133944
These comments refer to the nested loop in the module inliner where
the inner loop grouped call sites from the same caller. We don't
group call sites anymore, so the comment has become stale.
In the CGSCC inliner, DidInline was used as an indicator to update the call graph.
In the module inliner, DidInline is always true at the end of the
"while" loop, so can just drop it.
This revision revisits the implementation of `transform.fuse_into_containing_op` so that it iterates on
producers one use at a time.
Support is added to fuse a producer through a foreach_thread shared tensor argument, in which case we
tile and fuse the op inside the containing op and update the shared tensor argument to the unique destination operand.
If one cannot find such a unique destination operand the transform fails.
Differential Revision: https://reviews.llvm.org/D134051
Alex Zinenko