Treat unknown address spaces as global

We don't need FrameInfo's concept of the stack alignment. We just
need to tell it the desired alignment. Which in this case is 2.

AArch64 handles this pattern in their lowering code. By emitting
CMN. ARM handles it as an isel pattern.

This reverts commit https://reviews.llvm.org/rG449a13509190b1c57e5fcf5cd7e8f0f647f564b4,
due to buildbot failures such as
http://lab.llvm.org:8011/builders/clang-cmake-x86_64-avx2-linux/builds/13251.

This required legal and regbankselected MIR for seemingly no
reason. For AMDGPU this wouldn't see legalized G_GLOBAL_VALUEs.

I readily admit that I don't know why this fixes the modules build, but
it seems to get things building again. Previously I saw the error
message:

http://lab.llvm.org:8080/green/view/LLDB/job/lldb-cmake/9404/consoleFull#-361314398a1ca8a51-895e-46c6-af87-ce24fa4cd561

```
/Users/buildslave/jenkins/workspace/lldb-cmake/llvm-project/llvm/include/llvm/IR/IRBuilderFolder.h:18:10: fatal error: cyclic dependency in module 'LLVM_intrinsic_gen': LLVM_intrinsic_gen -> LLVM_IR -> LLVM_intrinsic_gen

         ^
While building module 'LLVM_intrinsic_gen' imported from /Users/buildslave/jenkins/workspace/lldb-cmake/llvm-project/llvm/lib/IR/IRBuilder.cpp:14:
In file included from <module-includes>:1:
/Users/buildslave/jenkins/workspace/lldb-cmake/llvm-project/llvm/include/llvm/IR/Argument.h:19:10: fatal error: could not build module 'LLVM_IR'
 ~~~~~~~~^~~~~~~~~~~~~~~~~
/Users/buildslave/jenkins/workspace/lldb-cmake/llvm-project/llvm/lib/IR/IRBuilder.cpp:14:10: fatal error: could not build module 'LLVM_intrinsic_gen'
```

And reproduced with:

cmake -G Ninja /Users/vsk/src/llvm-backup-master/llvm -DCLANG_ENABLE_ARCMT=Off -DCLANG_ENABLE_STATIC_ANALYZER=Off -DLLVM_ENABLE_PROJECTS='clang;clang-tools-extra;lld;libcxx;libcxxabi;compiler-rt;libunwind;lldb' -DLLDB_USE_SYSTEM_DEBUGSERVER=On -DCMAKE_BUILD_TYPE=RelWithDebInfo -DLLVM_ENABLE_ASSERTIONS=On -DLLVM_ENABLE_MODULES=On

ProcessPOSIX and ProcessLinux are not real plugins and should not be
marked as such. This should fix the Linux bots.

AMDGPUCodeGenPrepare expands this most of the time, but not always. We
will always at least need a fallback option here. This is the 3rd
implementation of the same expansion in the backend. Eventually I
would like to eliminate the IR expansion (and the DAG version
obviously).

Currently the new legalizer path produces a better result, since the
IR expansion results in extra operations which need to be combined
out. Notably, the IR expansion results in multiplies by 0.

The check for 'HAVE_CXX_ATOMICS_WITHOUT_LIB' may create false
positives in RISC-V. This is reproducible when compiling LLVM natively
using GCC on a rv64gc (rv64imafdgc) host. Due to the 'A' (atomic)
extension, g++ replaces calls to libatomic operations on the
std::atomic<int> type with the native hardware instructions. As a
result, the compilation succeeds and the build system thinks it
doesn't need to pass '-latomic'.

Improve the reliability of the 'HAVE_CXX_ATOMICS_WITHOUT_LIB' test in
two steps:

1. Force a pre-increment on x (++x), which should force a call to a
libatomic function;

2. Because step 1 would resolve the increment to 'amoadd.w.aq' under
the 'A' extension, force the same operation on sub-word types, for
which there is no hardware support.

Reviewers: jfb, hintonda, smeenai, mgorny, JDevlieghere, jyknight
Reviewed By: jfb
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D68964

Summary:
The first in a series of patches that ports the LLVM coroutines passes
to the new pass manager infrastructure. This patch implements
'coro-early'.

NB: All coroutines passes begin by checking that coroutine intrinsics are
declared within the LLVM IR module they're operating on. To do so, they call
`coro::declaresIntrinsics`. The next 3 patches in this series, which add new
pass manager implementations of the 'coro-split', 'coro-elide', and
'coro-cleanup' passes, use a similar pattern as the one used here: a static
function is shared across both old and new passes to detect if relevant
coroutine intrinsics are delcared. To make this pattern easier to read, this
patch adds `const` keywords to the parameters of `coro::declaresIntrinsics`.

Reviewers: GorNishanov, lewissbaker, junparser, chandlerc, deadalnix, wenlei

Reviewed By: wenlei

Subscribers: ychen, wenlei, EricWF, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71898

This should fix the unresolved external symbol error.

mutateStrictFPToFP can delete the node and replace it with another with the same
value which can later cause problems, and returning the result of
mutateStrictFPToFP doesn't work because SelectionDAGLegalize expects that the
returned value has the same number of results as the original. Instead handle
things by doing the mutation manually.

Differential Revision: https://reviews.llvm.org/D74726

Relative to the original commit, this fixes some warnings,
and is based on the deletion of the IRBuilder copy constructor
in D74693. The automatic copy constructor would no longer be
safe.

-----

Related llvm-dev thread:
http://lists.llvm.org/pipermail/llvm-dev/2020-February/138951.html

This patch moves the IRBuilder from templating over the constant
folder and inserter towards making both of these virtual.
There are a couple of motivations for this:

1. It's not possible to share code between use-sites that use
different IRBuilder folders/inserters (short of templating the code
and moving it into headers).
2. Methods currently defined on IRBuilderBase (which is not templated)
do not use the custom inserter, resulting in subtle bugs (e.g.
incorrect InstCombine worklist management). It would be possible to
move those into the templated IRBuilder, but...
3. The vast majority of the IRBuilder implementation has to live
in the header, because it depends on the template arguments.
4. We have many unnecessary dependencies on IRBuilder.h,
because it is not easy to forward-declare. (Significant parts of
the backend depend on it via TargetLowering.h, for example.)

This patch addresses the issue by making the following changes:

* IRBuilderDefaultInserter::InsertHelper becomes virtual.
  IRBuilderBase accepts a reference to it.
* IRBuilderFolder is introduced as a virtual base class. It is
 implemented by ConstantFolder (default), NoFolder and TargetFolder.
  IRBuilderBase has a reference to this as well.
* All the logic is moved from IRBuilder to IRBuilderBase. This means
  that methods can in the future replace their IRBuilder<> & uses
  (or other specific IRBuilder types) with IRBuilderBase & and thus
  be usable with different IRBuilders.
* The IRBuilder class is now a thin wrapper around IRBuilderBase.
  Essentially it only stores the folder and inserter and takes care
  of constructing the base builder.

What this patch doesn't do, but should be simple followups after this change:

* Fixing use of the inserter for creation methods originally defined
  on IRBuilderBase.
* Replacing IRBuilder<> uses in arguments with IRBuilderBase, where useful.
* Moving code from the IRBuilder header to the source file.

From the user perspective, these changes should be mostly transparent:
The only thing that consumers using a custom inserted may need to do is
inherit from IRBuilderDefaultInserter publicly and mark their InsertHelper
as public.

Differential Revision: https://reviews.llvm.org/D73835

Simply dropping the createPollyIRBuilder() function here, because
it doesn't do much. Also directly initialize Expander in
ScopExpander instead of going through the copy-constructor.

Summary:
Depends on https://reviews.llvm.org/D70927.

`LazyCallGraph::addNewFunctionIntoSCC` allows users to insert a new
function node into a call graph, into a specific, existing SCC.

Extend this interface such that functions can be added even when they do
not belong in any existing SCC, but instead in a new SCC within an
existing RefSCC.

The ability to insert new functions as part of a RefSCC is necessary for
outlined functions that do not form a strongly connected cycle with the
function they are outlined from. An example of such a function would be the
coroutine funclets 'f.resume', etc., which are outlined from a coroutine 'f'.
Coroutine 'f' only references the funclets' addresses, it does not call
them directly.

Reviewers: jdoerfert, chandlerc, wenlei, hfinkel

Reviewed By: jdoerfert

Subscribers: hfinkel, JonChesterfield, mehdi_amini, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D72226

Summary:
This patch adds vector-scalar variants to the following families of
MVE intrinsics:
* vaddq
* vsubq
* vmulq
* vqaddq
* vqsubq
* vhaddq
* vhsubq
* vqdmulhq
* vqrdmulhq

The vector-scalar variants perform a splat operation on the scalar
operand and then perform the same operations as their vector-vector
counterparts. Code generation is done accordingly (using LLVM IR 'insert'
and 'shuffle' operations which are later converted into an ARMvdup
SDNode).

Reviewers: simon_tatham, dmgreen, MarkMurrayARM, ostannard

Reviewed By: dmgreen

Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D74620

D73835 will make IRBuilder no longer trivially copyable. This patch
deletes the copy constructor in advance, to separate out the breakage.

Currently, the IRBuilder copy constructor is usually used by accident,
not by intention.  In rG7c362b25 I've fixed a number of cases where
functions accepted IRBuilder rather than IRBuilder &, thus performing
an unnecessary copy. In rG5f7b92b1 I've fixed cases where an
IRBuilder was copied, while an InsertPointGuard should have been used
instead.

The only non-trivial use of the copy constructor is the
getIRBForDbgInsertion() helper, for which I separated construction and
setting of the insertion point in this patch.

Differential Revision: https://reviews.llvm.org/D74693

I've fixed most of these before, but missed some occurrences
in targets I don't usually build.

This patch changes the way we initialize and terminate the plugins in
the system initializer. It uses an approach similar to LLVM's
TARGETS_TO_BUILD with a def file that enumerates the plugins.

Differential revision: https://reviews.llvm.org/D73067

Should fix error C2039: 'toupper': is not a member of 'std'.

This reverts commit 82b47b29.

This broke Clang and LLDB module builds without -fmodules-local-submodule-visbility.
I'll revert this for now until we have a fix and reland once Clang
can properly handle this code.

See also the discussion in https://reviews.llvm.org/rG82b47b2978405f802a33b00d046e6f18ef6a47be

The buildbot failures on MSVC should have been fixed by f128f442.

The way fallback to SelectionDAG works is somewhat surprising to
me. When the fallback path is enabled, the entire set of SelectionDAG
selector passes is added to the pass pipeline, and each one needs to
check if the function was selected. This results in the surprising
behavior of running SIFixSGPRCopies for example, but only if
-global-isel-abort=2 is used.

SIAddIMGInitPass is also added in addInstSelector, but I'm not sure
why we have this pass or if it should be added somewhere else for
GlobalISel.

These are going to be useful in TargetLowering::SimplifyDemandedBits, so expose these helpers outside of SelectionDAG.cpp

Also add an getValidShiftAmountConstant early-out to getValidMinimumShiftAmountConstant/getValidMaximumShiftAmountConstant so we can use them for scalar cases as well.

Produce an unmerge to a narrower type and introduce a narrower shift
if needed. I wasn't sure if there was a better way to parameterize the
target's preferred shift type for the GICombineRule, so manually call
the combine helper.

Doesn't try to fail on the dlc bit pre-gfx10 like the DAG lowering
does.

Summary:
We are missing the default parmeter arguments when IndexFunctionLocals
is true.

Fixes https://github.com/clangd/clangd/issues/285.

Reviewers: kadircet

Subscribers: kristof.beyls, ilya-biryukov, jkorous, arphaman, usaxena95, cfe-commits

Tags: #clang

Differential Revision: https://reviews.llvm.org/D74610

Summary:
This patch adds assembly-level support for a new Arm M-profile
architecture extension, Custom Datapath Extension (CDE).

A brief description of the extension is available at
https://developer.arm.com/architectures/instruction-sets/custom-instructions

The latest specification for CDE is currently a beta release and is
available at
https://static.docs.arm.com/ddi0607/aa/DDI0607A_a_armv8m_arm_supplement_cde.pdf

CDE allows chip vendors to add custom CPU instructions.  The CDE
instructions re-use the same encoding space as existing coprocessor
instructions (such as MRC, MCR, CDP etc.). Each coprocessor in range
cp0-cp7 can be configured as either general purpose (GCP) or custom
datapath (CDEv1).  This configuration is defined by the CPU vendor and
is provided to LLVM using 8 subtarget features: cdecp0 ... cdecp7.

The semantics of CDE instructions are implementation-defined, but the
instructions are guaranteed to be pure (that is, they are stateless,
they do not access memory or any registers except their explicit
inputs/outputs).

CDE requires the CPU to support at least Armv8.0-M mainline
architecture. CDE includes 3 sets of instructions:
* Instructions that operate on general purpose registers and NZCV
  flags
* Instructions that operate on the S or D register file (require
  either FP or MVE extension)
* Instructions that operate on the Q register file, require MVE

The user-facing names that can be specified on the command line are
the same as the 8 subtarget feature names. For example:

    $ clang -target arm-none-none-eabi -march=armv8m.main+cdecp0+cdecp3

tells the compiler that the coprocessors 0 and 3 are configured as
CDEv1 and the remaining coprocessors are configured as GCP (which is
the default).

Reviewers: simon_tatham, ostannard, dmgreen, eli.friedman

Reviewed By: simon_tatham

Subscribers: kristof.beyls, hiraditya, cfe-commits, llvm-commits

Tags: #clang, #llvm

Differential Revision: https://reviews.llvm.org/D74044

While looking at the output on real sized programs, there is a lot of
extra SGPR spilling compared to the DAG path. This seems to largely be
from all constants being SGPRs in the entry block.

As discussed in https://reviews.llvm.org/D73206#1871895 there is both
`DIERef` and `user_id_t` and sometimes (for DWZ) we need to encode Main
CU into them and sometimes we cannot as it is unavailable at that point
and at the same time not even needed.

I have also noticed `DIERef` and `user_id_t` in fact contain the same
information which can be seen in SymbolFileDWARF::GetUID.

SB* API/ABI is already using `user_id_t` and it needs to encode Main CU
for DWZ. Therefore what about making `DIERef` the identifier not
containing Main CU and `user_id_t` the identifier containing Main CU?

It is sort of a revert of D63322.

I find this patch as a NFC cleanup to the codebase - to satisfy a new
premise `user_id_t` is used as little as possible and thus only for
external interfaces which must not deal with MainCU in any way.

Its larger goal is to satisfy a plan to implement DWZ support.

Differential Revision: https://reviews.llvm.org/D74637

We were skipping the libatomic requirement check for MSVC, but not setting
the corresponding variable, HAVE_CXX_ATOMICS_WITHOUT_LIB. D69869 seems to
have to failed to build on ARM MSVC because of that, and was reverted. This
should probably fix the issue. The plan is to check the result of the build
bots and then submit a more thoroughly refactored version for review.

Summary:
This patch implements the part of the calling convention
where SVE Vectors are passed by reference. This means the
caller must allocate stack space for these objects and
pass the address to the callee.

Reviewers: efriedma, rovka, cameron.mcinally, c-rhodes, rengolin

Reviewed By: efriedma

Subscribers: tschuett, kristof.beyls, hiraditya, llvm-commits

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D71216

Summary:
Linalg's promotion pass was only supporting f32 buffers due to how the
zero value was build for the `fill` operation.

Moreover, `promoteSubViewOperands` was returning a vector with one entry
per float subview while omitting integer subviews. For a program
with only integer subviews the return vector would be of size 0.
However, `promoteSubViewsOperands` would try to access a non zero
number of entries of this vector, resulting in a sefgault.

Reviewers: nicolasvasilache, ftynse

Reviewed By: ftynse

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

Tags: #llvm

Differential Revision: https://reviews.llvm.org/D74532

This adds another pattern to the combiner for a case that we were not handling
to generate the REV16 instruction for ARM/Thumb2 and a bswap+ror on X86.

Differential Revision: https://reviews.llvm.org/D74032