The extload on X87 is free.

[X86] Swap the 0 and the fudge factor in the constant pool for the 32-bit mode i64->f32/f64/f80 uint_to_fp algorithm.

This allows us to generate better code for selecting the fixup
to load.

Previously when the sign was set we had to load offset 0. And
when it was clear we had to load offset 4. This required a testl,
setns, zero extend, and finally a mul by 4. By switching the offsets
we can just shift the sign bit into the lsb and multiply it by 4.

Summary: Fix the ViewOpShapeFolder in case of no affine mapping associated with a Memref construct identity mapping.

Reviewers: nicolasvasilache

Subscribers: mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, arpith-jacob, mgester, lucyrfox, liufengdb, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72735

On Fuchsia, pthread API is emulated on top of C11 thread API. Using C11
thread API directly is more efficient.

While this implementation is only used by Fuchsia at the moment, it's
not Fuchsia specific, and could be used by other platforms that use C11
threads rather than pthreads in the future.

Differential Revision: https://reviews.llvm.org/D64378

(clang diagnostic handler for IR input files)

The reference counter for global objects marked with declare target is INF. This patch prevents the runtime from incrementing /decrementing INF refcounts. Without it, the map(delete: global_object) directive actually deallocates the global on the device. With this patch, such a directive becomes a no-op.

Differential Revision: https://reviews.llvm.org/D72525

Summary:
- `dead-mi-elimination` assumes MIR in the SSA form and cannot be
  arranged after phi elimination or DeSSA. It's enhanced to handle the
  dead register definition by skipping use check on it. Once a register
  def is `dead`, all its uses, if any, should be `undef`.
- Re-arrange the DIE in RA phase for AMDGPU by placing it directly after
  `detect-dead-lanes`.
- Many relevant tests are refined due to different register assignment.

Reviewers: rampitec, qcolombet, sunfish

Subscribers: arsenm, kzhuravl, jvesely, wdng, nhaehnle, yaxunl, dstuttard, tpr, t-tye, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72709

This commit defines a new SPIR-V dialect attribute for specifying
a SPIR-V target environment. It is a dictionary attribute containing
the SPIR-V version, supported extension list, and allowed capability
list. A SPIRVConversionTarget subclass is created to take in the
target environment and sets proper dynmaically legal ops by querying
the op availability interface of SPIR-V ops to make sure they are
available in the specified target environment. All existing conversions
targeting SPIR-V is changed to use this SPIRVConversionTarget. It
probes whether the input IR has a `spv.target_env` attribute,
otherwise, it uses the default target environment: SPIR-V 1.0 with
Shader capability and no extra extensions.

Differential Revision: https://reviews.llvm.org/D72256

For IR input files, we currently use LLVM diagnostic handler even the
compilation is from clang. As a result, we are not able to use -Rpass
to get the transformation reports. Some warnings are not handled
properly either: We found many mysterious warnings in our ThinLTO backend
compilations in SamplePGO and CSPGO. An example of the warning:
"warning: net/proto2/public/metadata_lite.h:51:21: 0.02% (1 / 4999)"

This turns out to be a warning by Wmisexpect, which is supposed to be
filtered out by default. But since the filter is in clang's
diagnostic hander, we emit these incomplete warnings from LLVM's
diagnostic handler.

This patch uses clang diagnostic handler for IR input files. We create
a fake backendconsumer just to install the diagnostic handler.

With this change, we will have proper handling of all the warnings and we can
use -Rpass* options in IR input files compilation.
Also note that with is patch, LLVM's diagnostic options, like
"-mllvm -pass-remarks=*", are no longer be able to get optimization remarks.

Differential Revision: https://reviews.llvm.org/D72523

Summary:
This allows for users to cache printer state, which can be costly to recompute. Each of the IR print methods gain a new overload taking this new state class.

Depends On D72293

Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D72294

NVPTX does not support RTTI, so disable it by default.

Summary:
This was previously disabled as FunctionType TypeAttrs could not be roundtripped in the IR. This has been fixed, so we can now generically print FuncOp.

Depends On D72429

Reviewed By: jpienaar, mehdi_amini

Differential Revision: https://reviews.llvm.org/D72642

Allow to build PCH's (with -building-pch-with-obj and the extra .o file)
with -fmodules-codegen -fmodules-debuginfo to allow emitting shared code
into the extra .o file, similarly to how it works with modules. A bit of
a misnomer, but the underlying functionality is the same. This saves up
to 20% of build time here.

Differential Revision: https://reviews.llvm.org/D69778

If a header contains 'extern template', then the template should be provided
somewhere by an explicit instantiation, so it is not necessary to generate
a copy. Worse, this can lead to an unresolved symbol, because the codegen's
object file will not actually contain functions from such a template
because of the GVA_AvailableExternally, but the object file for the explicit
instantiation will not contain them either because it will be blocked
by the information provided by the module.

Differential Revision: https://reviews.llvm.org/D69779

Summary:
This diff fixes issues with the semantics of linalg.generic on tensors that appeared when converting directly from HLO to linalg.generic.
The changes are self-contained within MLIR and can be captured and tested independently of XLA.

The linalg.generic and indexed_generic are updated to:

To allow progressive lowering from the value world (a.k.a tensor values) to
the buffer world (a.k.a memref values), a linalg.generic op accepts
mixing input and output ranked tensor values with input and output memrefs.

```
%1 = linalg.generic #trait_attribute %A, %B {other-attributes} :
  tensor<?x?xf32>,
  memref<?x?xf32, stride_specification>
  -> (tensor<?x?xf32>)
```

In this case, the number of outputs (args_out) must match the sum of (1) the
number of output buffer operands and (2) the number of tensor return values.
The semantics is that the linalg.indexed_generic op produces (i.e.
allocates and fills) its return values.

Tensor values must be legalized by a buffer allocation pass before most
transformations can be applied. Such legalization moves tensor return values
into output buffer operands and updates the region argument accordingly.

Transformations that create control-flow around linalg.indexed_generic
operations are not expected to mix with tensors because SSA values do not
escape naturally. Still, transformations and rewrites that take advantage of
tensor SSA values are expected to be useful and will be added in the near
future.

Subscribers: bmahjour, mehdi_amini, rriddle, jpienaar, burmako, shauheen, antiagainst, arpith-jacob, mgester, lucyrfox, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72555

- Prefer `getVectorIdxTy()` as the index operand type for
  `EXTRACT_SUBVECTOR` as targets expect different types by overloading
  `getVectorIdxTy()`.

There are no special virtual function handlers (like __cxa_pure_virtual)
defined for NVPTX target, so just emit such functions as null pointers
to prevent issues with linking and unresolved references.

Summary: bfloat16 doesn't have a valid APFloat format, so we have to use double semantics when storing it. This change makes sure that hexadecimal values can be round-tripped properly given this fact.

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D72667

These intrinsics expand to a variable number of instructions so just like in
ISelLowering.cpp we use custom code to deal with them.

Committing Tim's original patch.

Differential Revision: https://reviews.llvm.org/D65656

Reviewed By: nathanchance, sidneym

Differential Revision: https://reviews.llvm.org/D72668

This code is untested in tree because the "APFloat::semanticsPrecision(sem) >= SrcVT.getSizeInBits() - 1" check is false for most combinations for int and fp types except maybe i32 and f64. For that you would need i32 to be an illegal type, but f64 to be legal and have custom handling for legalizing the split sint_to_fp. The precision check itself was added in 2010 to fix a double rounding issue in the algorithm that would occur if the sint_to_fp was not able to do the conversion without rounding.

Differential Revision: https://reviews.llvm.org/D72728

The tests aren't concerned at all by the actual sanitizer - only by blacklist being reported as a dependency.
We're unfortunately limited by platform support for any particular sanitizer but we can at least use one that is widely supported.

Post-commit review:
https://reviews.llvm.org/D72729

When multiple guard intrinsics are merged into one, currently the
result of eraseInstFromFunction() is returned -- however, this
should only be done if the current instruction is being removed.
In this case we're removing a different instruction and should
instead report that the current one has been modified by returning it.

For this test case, this reduces the number of instcombine iterations
from 5 to 2 (the minimum possible).

Differential Revision: https://reviews.llvm.org/D72558

Summary:
This patch adds an option to limit debug info by only emitting complete class
type information when its constructor is emitted. This applies to classes
that have nontrivial user defined constructors.

I implemented the option by adding another level to `DebugInfoKind`, and
a flag `-flimit-debug-info-constructor`.

Total object file size on Windows, compiling with RelWithDebInfo:
  before: 4,257,448 kb
  after:  2,104,963 kb

And on Linux
  before: 9,225,140 kb
  after:  4,387,464 kb

According to the Windows clang.pdb files, here is a list of types that are no
longer complete with this option enabled: https://reviews.llvm.org/P8182

Reviewers: rnk, dblaikie

Subscribers: aprantl, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D72427

Differential revision: https://reviews.llvm.org/D70689

The primary motivation for this is to add another dimension to the
Swift LLDB test matrix, but this seems generally useful.

Differential Revision: https://reviews.llvm.org/D72662

This also makes this function consistent with the rest of the
libc++ provided fallbacks.

The locale support in msvcrt.dll is very limited anyway; it can
only be configured processwide, not per thread, and it only seems
to support the locales "C" and "" (the user set locale), so it's
hard to make any meaningful automatic test for it. But manually tested,
this change does make time formatting locale code in libc++ output
times in the user requested format, when using locale "".

Differential Revision: https://reviews.llvm.org/D69554

Summary: Add the missing MUL pattern for integer immediate instructions.

Reviewers: sdesmalen, huntergr, efriedma, c-rhodes, kmclaughlin

Subscribers: tschuett, hiraditya, rkruppe, psnobl, llvm-commits, amehsan

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72654

Fedora wants to build projects with -fno-semantic-interposition (e.g.
https://fedoraproject.org/wiki/Changes/PythonNoSemanticInterpositionSpeedup),
which is supported by GCC>=5.

Clang's current behavior is similar to -fno-semantic-interposition and
the end goal is to make it more so
(https://lists.llvm.org/pipermail/llvm-dev/2016-November/107625.html).
Ignore this option.

We should let users know -fsemantic-interposition is not currently
supported, so it should remain a hard error.

Reviewed By: serge-sans-paille

Differential Revision: https://reviews.llvm.org/D72724

TSan spuriously reports for any OpenMP application a race on the initialization
of a runtime internal mutex:

```
Atomic read of size 1 at 0x7b6800005940 by thread T4:
  #0 pthread_mutex_lock <null> (a.out+0x43f39e)
  #1 __kmp_resume_64 <null> (libomp.so.5+0x84db4)

Previous write of size 1 at 0x7b6800005940 by thread T7:
  #0 pthread_mutex_init <null> (a.out+0x424793)
  #1 __kmp_suspend_initialize_thread <null> (libomp.so.5+0x8422e)
```

According to @AndreyChurbanov this is a false positive report, as the control
flow of the runtime guarantees the ordering of the mutex initialization and
the lock:
https://software.intel.com/en-us/forums/intel-open-source-openmp-runtime-library/topic/530363

To suppress this report, I suggest the use of
TSAN_OPTIONS='ignore_uninstrumented_modules=1'.
With this patch, a runtime warning is provided in case an OpenMP application
is built with Tsan and executed without this Tsan-option.

Reviewed By: jdoerfert

Different...

This ports the MergeFunctions pass to the NewPM. This was rather
straightforward, as no analyses are used.

Additionally MergeFunctions needs to be conditionally enabled in
the PassBuilder, but I left that part out of this patch.

Differential Revision: https://reviews.llvm.org/D72537

Need to analayze the type of the expression for mapping, not the type of
the declaration.

Fix for https://bugs.llvm.org/show_bug.cgi?id=44245.

The optimizeBitCastFromPhi() and FoldPHIArgOpIntoPHI() end up
fighting against each other, because optimizeBitCastFromPhi()
assumes that bitcasts of loads will get folded. This doesn't
happen here, because a dangling phi node prevents the one-use
fold in https://github.com/llvm/llvm-project/blob/master/llvm/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp#L620-L628 from triggering.

This patch fixes the issue by explicitly performing the load
combine as part of the bitcast of phi transform. Other attempts
to force the load to be combined first were ultimately too
unreliable.

Differential Revision: https://reviews.llvm.org/D71164

So it can be reused as part of other combines.
In particular for D71164.

This fixes the issue encountered in D71164. Instead of using a
range-based for, manually iterate over the users and advance the
iterator beforehand, so we do not skip any users due to iterator
invalidation.

Differential Revision: https://reviews.llvm.org/D72657

Summary:
[nfc][libomptarget] Refactor nxptx/target_impl.cu

Use __kmpc_impl_atomic_add instead of atomicAdd to match the rest of the file.
Alternatively, target_impl.cu could use the cuda functions directly. Using a mixture in this
file was an oversight, happy to resolve in either direction.

Removed some comments that look outdated.

Call __kmpc_impl_unset_lock directly to avoid a redundant diagnostic and remove an implict
dependency on interface.h.

Reviewers: ABataev, grokos, jdoerfert

Reviewed By: jdoerfert

Subscribers: jfb, openmp-commits

Tags: #openmp

Differential Revision: https://reviews.llvm.org/D72719