https://discourse.llvm.org/t/opening-up-prs-experimentally-for-a-subset-of-the-llvm-project/70375/11

Reviewed By: ldionne

Differential Revision: https://reviews.llvm.org/D150314

My newly added backtrace test (https://reviews.llvm.org/D150491)
broke the Windows buildbot (https://lab.llvm.org/buildbot/#/builders/127/builds/48103)
because they do not have execinfo.h.
I expect the same will happen with the backtrace_symbols test (https://reviews.llvm.org/D150498) as well.

This patch does a quick fix by restricting the test scope to glibc-2.27.
(A tricker fix would take into account SANITIZER_INTERCEPT_BACKTRACE,
which is defined as (SI_FREEBSD || SI_NETBSD || SI_GLIBC || SI_SOLARIS))

This is another patch for https://github.com/google/sanitizers/issues/321
(sanitizer interceptors can write to freed memory, causing corruption),
in this case for backtrace_symbols.

backtrace_symbols is already correct, hence this patch removes the
TODO note. Additionally, this patch adds a test case for it.

Differential Revision: https://reviews.llvm.org/D150498

Currenlty there is a mismatch between LoongArch gcc and clang about
handling register name in inlineasm, i.e. gcc allows both `$`-prefixed
and non-prefiexed names for GPRs while clang only allows `$`-prefixed
one. This patch fixes this mismatch by adding non-prefixed GPR names
in clang.

Take `$r4` for example. With this patch, clang accepts `$r4`, `r4`,
`$a0` and `a0` like what gcc does.

Reviewed By: xen0n

Differential Revision: https://reviews.llvm.org/D136436

Add 'REQUIRES: asserts' to test added in D150002 (53a4adc0) because it tests for a crash that is caused by an assertion failure.

This is part of an effort to migrate from llvm::Optional to
std::optional:

https://discourse.llvm.org/t/deprecating-llvm-optional-x-hasvalue-getvalue-getvalueor/63716

This was reverted in d4994d0e b/c a bolt test failed after the
encoding changed.

Relanded the patch with the updated test.

This is to improve a performance bottleneck of printf for long double.

Reviewed By: michaelrj

Differential Revision: https://reviews.llvm.org/D150475

Main value of this is simplifying code, and making a few of the tests easier to read.

Differential Revision: https://reviews.llvm.org/D150474

This was reverted in 4e84149c due to a problem report which has been confirmed to be misattributed.

Original commit message:

Use a switch to avoid repeat tests on the opcode, and factor out some common code out of another switch.  (Not all branches had both common bits, but the one which didn't left the values unchanged and their starting value is the respective operand - so storing it back is a nop.)

This reverts commit d3a6a05b.

Some bots don't like it.

Boo.

Sometimes you want to see how much is being allocated in your data structure
in general.

Add statistics that show how many internal and leaf nodes have been allocated
in the suffix tree over the course of its construction.

Also add a testcase that shows that we actually get these stats out when we're
outlining stuff.

The test shows that we get the expected O(n) leaf nodes, a split, and so on.

Add if condition removal pattern for acc.host_data in a same way as
acc.enter_data, acc.exit_data and acc.update.

The condition is removed from the op if it is a true constant. If
it is a false constant then the region is inlined before the op
and the op is removed.

Reviewed By: vzakhari

Differential Revision: https://reviews.llvm.org/D150480

This makes a (possible) change to the internal representation easier in the future, and makes the code easier to read now.

This reverts commit c13ed1cc.

Reason: causes bolt tests to fail: https://crbug.com/1445243

This reverts commit 13984608.  A problem was reported - the attribution seems slightly questionable, but this is low enough value after 297e06cf that reverting is easy.

It is a known, longstanding issue that some ASan interceptors
may write to freed memory, causing corruption
(https://github.com/google/sanitizers/issues/321). This patch
adds a testcase for the backtrace interceptor (one of the
known cases).

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D150491

Predefine __AMDGCN_CUMODE__ as 1 or 0 when compilation assumes CU or WGP modes.

If WGP mode is not supported, ignore -mno-cumode and emit a warning.

This is needed for implementing device functions like __smid
(https://github.com/ROCm-Developer-Tools/hipamd/blob/312dff7b794337aa040be0691acc78e9f968a8d2/include/hip/amd_detail/amd_device_functions.h#L957)

Reviewed by: Matt Arsenault, Artem Belevich, Brian Sumner

Differential Revision: https://reviews.llvm.org/D145343

Adds support to set the hot/cold new hint values with an option. Change
the defaults slightly to make it easier to distinguish between compiler
synthesized vs manually inserted calls to the interface.

Differential Revision: https://reviews.llvm.org/D150488

The 'loc' argument is optional, and some compilers (e.g. MSVC) do no supply it.

Differential Revision: https://reviews.llvm.org/D148393

This patch migrates uses of StringRef::{starts,ends}with_insensitive
to StringRef::{starts,ends}_with_insensitive so that we can use names
similar to those used in std::string_view.  I'm planning to deprecate
StringRef::{starts,ends}with_insensitive once the migration is
complete across the code base.

Differential Revision: https://reviews.llvm.org/D150426

[OpenMP] Fix GCC build issues and restore "Additional APIs used by the MSVC compiler for loop collapse (rectangular and non-rectangular loops)"

Fixes a GCC build issue (an instance of unallowed typename keyword) and reworks memory allocation
to avoid the use of C++ library based primitives ) in and restores the earlier commit https://reviews.llvm.org/D148393

Differential Revision: https://reviews.llvm.org/D149010

The short granule logic made this test flaky because with low
probability there would be no tag mismatch by coincidence.

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D150484

Signed-off-by: Jerry Ge <jerry.ge@arm.com>
Change-Id: I922eeda95223da2c7da0653451e604651b7594bd

Reviewed By: mehdi_amini

Differential Revision: https://reviews.llvm.org/D150487

A follow-up change for 6d861d49:
remove an unnecessary const-qualifier so that the code doesn't have to
remove the qualifier explicitly using `std::remove_const_t`, which
triggers a warning at some bots (e.g.,
https://lab.llvm.org/buildbot/#/builders/247/builds/4442).

Add support for --dump-section on COFF files. This is helpful for
extracting specific content from an object file on Windows.

Differential Revision: https://reviews.llvm.org/D150305
Reviewed By: @alexander-shaposhnikov, @jhenderson, @hjyamauchi

Cleanup profile reader classes to prepare for complex refactoring as propsed in D147740, continuing D148872
This is patch 3/n. This patch changes the behavior of function offset table.

Previously when reading ExtBinary profile, the funcOffsetTable (map) is always populated, and in addition if the profile is CS, the orderedFuncOffsets (list) is also populated. However when reading the function samples, only one of the container is being used, never both, so it's a huge waste of time to populate both. Added logic to select which one to use, and completely skip reading function offset table if we are in tool mode (all function samples are to be read sequentially regardless)

Reviewed By: davidxl, wenlei

Differential Revision: https://reviews.llvm.org/D149124

The global names were created using a hash based on the address
of std::vector::data address. Since the memory may be reused
by different std::vector's, this may cause non-equivalent
constant expressions to map to the same name. This is what is happening
in the modified flang/test/Lower/constant-literal-mangling.f90 test.

I changed the name creation to use a map between the constant expressions
and corresponding unique names. The uniquing is done using a name counter
in FirConverter. The effect of this change is that the equivalent
constant expressions are now mapped to the same global, and the naming
is "stable" (i.e. it does not change from compilation to compilation).

Though, the issue is not HLFIR specific it was affecting several tests
when using HLFIR lowering.

Differential Revision: https://reviews.llvm.org/D150380

Tosa supports fp16 and fp32 accumulator type for fp16 input, but
no way to tell for computational operators whether accumulator
should be fp16 or fp32 from input type. Add this new attribute
to specify the type.

Set to fp32 by default for now. When the time fp16 is supported,
the accumulator type can be selected based on trade-off between
performance and accuracy.

Reviewed By: jpienaar

Differential Revision: https://reviews.llvm.org/D146317

DWARFAttribute is used in 2 classes: DWARFAbbreviationDecl and
DWARFAttributes. The former stores a std::vector of them and the latter
has a small structure called AttributeValue that contains a
DWARFAttribute. DWARFAttributes maintains a llvm::SmallVector of
AttributeValues.

My end goal is to have `DWARFAttributes` have a llvm::SmallVector
specialized on DWARFAttribute. In order to do that, we'll have to move
the other elements of AttributeValue into DWARFAttribute itself. But we
don't want to do this while DWARFAbbreviationDecl is using
DWARFAttribute because it will needlessly increase the size of
DWARFAbbreviationDecl. So instead I will create a small type containing
only what DWARFAbbreviationDecl needs and call it `AttributeSpec`. This
is the exact same thing that LLVM does today.

I've elected to swap std::vector for llvm::SmallVector here with a pre-allocated
size of 8. I've collected time and memory measurements before this change and
after it as well. Using a c++ project with 10,000 object files and no dSYM, I
place a breakpoint by file + lineno and see how long it takes to resolve.

Before this patch:
  Time (mean ± σ):     13.577 s ±  0.024 s    [User: 12.418 s, System: 1.247 s]
Total number of bytes allocated: 1.38 GiB
Total number of allocations: 6.47 million allocations

After this patch:
  Time (mean ± σ):     13.287 s ±  0.020 s    [User: 12.128 s, System: 1.250 s]
Total number of bytes allocated: 1.59 GiB
Total number of allocations: 4.61 million allocations

So we consume more memory than before, but we actually make less allocations on
average.

I also measured with an llvm::SmallVector with a pre-allocated size of 4 instead
of 8 to measure how well it performs:

  Time (mean ± σ):     13.246 s ±  0.048 s    [User: 12.074 s, System: 1.268 s]
Total memory consumption: 1.50 GiB
Total number of allocations: 5.74 million

Of course this data may look very different depending on the actual program
being debugged, but each of the object files had 100+ AbbreviationDeclarations
each with between 0 and 10 Attributes, so I feel this was a fair example to
consider.

Differential Revision: https://reviews.llvm.org/D150418

Reviewed By: ldionne, #libc

Spies: arichardson, pcwang-thead, libcxx-commits, miyuki

Differential Revision: https://reviews.llvm.org/D150277

IndirectCopyToAGPR should be reworked as to avoid optimizing during copy lowering. However, as it stands, the code is buggy. This patch replaces the call to definesRegister with modifiesRegister, and confirms that the dest reg of the found accvgpr_write is in fact the src reg of our copy.

Differential Revision: https://reviews.llvm.org/D149873

Change-Id: Id8a61659ac15565dcb970069d0624f0925a46e6d

This is preparation for an additional combine.