This reverts commit 6951b723.
Breaks several bots, see comments on https://reviews.llvm.org/D99409

Extend D94856 to handle 'and', 'or' and 'xor' instructions as well

We still fail on many i8/i16 cases as the test and the logic-op are performed on different widths

No need to lookup through and/or try to vectorize operands of the
CmpInst instructions during attempts to find/vectorize min/max
reductions. Compiler implements postanalysis of the CmpInsts so we can
skip extra attempts in tryToVectorizeHorReductionOrInstOperands and save
compile time.

Differential Revision: https://reviews.llvm.org/D99950

The -fdebug-dump-provenance flag is meant to be used with
needProvenanceRangeToCharBlockMappings set to true. This way, extra
mapping is generated that allows e.g. IDEs to retrieve symbol's scope
(offset into cooked character stream) based on symbol's source code
location. This patch makes sure that this option is set when using
-fdebug-dump-provenance.

With this patch, the implementation of  -fdebug-dump-provenance in
`flang-new -fc1` becomes consistent with `f18`. The corresponding LIT
test is updated so that it can be shared with `f18`. I refined it a bit
so that:
  * it becomes a frontend-only test
  * it's stricter about the expected output

Differential Revision: https://reviews.llvm.org/D98847

This patch adds support for the `-cpp` and `-nocpp` flags. The
implemented semantics match f18 (i.e. the "throwaway" driver), but are
different to gfortran. In Flang the preprocessor is always run. Instead,
`-cpp/-nocpp` are used to control whether predefined and command-line
preprocessor macro definitions are enabled or not. In practice this is
sufficient to model gfortran`s `-cpp/-nocpp`.

In the absence of `-cpp/-nocpp`, the driver will use the extension of
the input file to decide whether to include the standard macro
predefinitions. gfortran's documentation [1] was used to decide which
file extension to use for this.

The logic mentioned above was added in FrontendAction::BeginSourceFile.
That's relatively late in the driver set-up, but this roughly where the
name of the input file becomes available. The logic for deciding between
fixed and free form works in a similar way and was also moved to
FrontendAction::BeginSourceFile for consistency (and to reduce
code-duplication).

The `-cpp/-nocpp` flags are respected also when the input is read from
stdin. This is different to:
   * gfortran (behaves as if `-cpp` was used)
   * f18 (behaves as if `-nocpp` was used)

Starting with this patch, file extensions are significant and some test
files had to be renamed to reflect that. Where possible, preprocessor
tests were updated so that they can be shared between `f18` and
`flang-new`. This was implemented on top of adding new test for
`-cpp/-nocpp`.

[1] https://gcc.gnu.org/onlinedocs/gcc/Overall-Options.html

Reviewed By: kiranchandramohan

Differential Revision: https://reviews.llvm.org/D99292

D97493 separate target creation out to a single function
`CompilerInstance::createTarget`. However, it would overwrite AuxTarget
even if it has been set.
As @kadircet recommended in D98128, this patch check the existence of
AuxTarget and not overwrite it when it has been set.

Reviewed By: kadircet

Differential Revision: https://reviews.llvm.org/D100024

The swap of the operands can affect later transforms that
are expecting a constant as operand 1. I don't think we
can trigger a bug with the current code, but I hit that
problem while drafting a new transform for min/max intrinsics.

Also factors out out-of-bounds mask generation from vector.transfer_read/write into a new MaterializeTransferMask pattern.

Differential Revision: https://reviews.llvm.org/D100001

D94856 covered the BMI cases where we had existing tests, this adds missing AND/OR/XOR test cases

Extend test coverage by checking various standard versions with
-f[no-]finite-loops. Suggested as part of D96418.

Clang spends a decent amount of time in the LineOffsetMapping::get(...)
function. This function used to be vectorized (through SSE2) then the
optimization got dropped because the sequential version was on-par performance
wise.

This provides an optimization of the sequential version that works on a word at
a time, using (documented) bithacks to provide a portable vectorization.

When preprocessing the sqlite amalgamation, this yields a sweet 3% speedup.

Differential Revision: https://reviews.llvm.org/D99409

If we allocate memory, the extent of the MemRegion will be the symbolic
value of the size parameter. This way, if that symbol gets constrained,
the extent will be also constrained.

This test demonstrates that the extent is indeed the same symbol.

Reviewed By: NoQ

Differential Revision: https://reviews.llvm.org/D99959

Extend D94856 to handle 'bzhi' instructions as well

rdar://75704162

Differential Revision: https://reviews.llvm.org/D99601

I do not see any bit-width restriction from the point of the
LLVM Lang Ref - Operand Bundles on the types of the deopt bundle
operands. Statepoint Lowering seems to be able to work with any
types.
This patch relaxes the two related assertions and adds a new test
for this change.

Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D100006

When property is declared in a superclass (or in a protocol),
it still can be of CXXRecord type and Sema could've already
generated a body for us.  This patch joins two branches and
two ways of acquiring IVar in order to reuse the existing code.
And prevent us from generating l-value to r-value casts for
C++ types.

rdar://67416721

Differential Revision: https://reviews.llvm.org/D99194

D94856 covered the BMI cases where we had existing tests, this adds a missing BZHI test case

When you pass in a payload with an invalid URI in a build with assertions enabled, it will crash.
Consuming the error from the failed URI parse prevents the error.

The crash is caused by the [llvm::expected](https://llvm.org/doxygen/classllvm_1_1Expected.html) having protection around trying to deconstruct without consuming the error first.

Reviewed By: kadircet

Differential Revision: https://reviews.llvm.org/D99872

//-E create_symlink//  is available on windows since CMake 3.13 (LLVM now uses 3.13.4)
It may needs administrator privileges or enabled developer mode (Windows 10)
See https://cmake.org/cmake/help/latest/release/3.13.html

Reviewed By: kbobyrev

Differential Revision: https://reviews.llvm.org/D99170

Differential Revision: https://reviews.llvm.org/D99934

size_t and friends are built-in scalar data types and s6.4.4.2 of the
OpenCL C Specification says the as_type() operator must be available
for these data types.

Differential Revision: https://reviews.llvm.org/D98959

This reverts commit a547b4e2,
relanding commit 31d219d2,
which was reverted because there was a conflicting inverse transform,
which was causing an endless combine loop, which has now been adjusted.

Original commit message:

https://alive2.llvm.org/ce/z/67w-wQ

We prefer `add`s over `sub`, and this particular xform
allows further folds to happen:

Fixes https://bugs.llvm.org/show_bug.cgi?id=49858

I.e., if any/all of the consants is an expression, don't do it.
Since those constants won't reduce into an immediate,
but would be left as an constant expression, they could cause
endless combine loops after 31d219d2
added an inverse transformation.

These would cause endless combine loop after 31d219d2.

LLVM test CodeGen/PowerPC/ppc-disable-non-volatile-cr.ll tries to check
for the absence of a sequence of instructions with several CHECK-NOT
with one of those directives using a variable defined in another.
However CHECK-NOT are checked independently so that is using a variable
defined in a pattern that should not occur in the input.

This commit changes occurence of the variable for the regex used in its
definition, thereby making each CHECK-NOT independent.

Reviewed By: NeHuang, nemanjai

Differential Revision: https://reviews.llvm.org/D99880

Clang test CodeGen/libcalls.c contains CHECK-NOT directives using a
variable defined in a CHECK directive with a different prefix never
enabled together, therefore causing the variable to be undefined in that
CHECK-NOT.

The intent of the test is to check that some declaration do not have the
same attribute as when compiling the test without -fmath-errno. This
commits instead changes all CHECK-NOT to CHECK directive, checking that
they all use the same attribute. It also adds an extra CHECK for that
prefix to check the expected attributes these functions should have when
compiling with -fmath-errno.

Reviewed By: rjmccall

Differential Revision: https://reviews.llvm.org/D99898

LLVM test Transforms/Coroutine/coro-split-sink-lifetime-O2.ll tries to
check for the absence of a sequence of instructions with several
CHECK-NOT with one of those directives using a variable defined in
another. However CHECK-NOT are checked independently so that is using a
variable defined in a pattern that should not occur in the input.

This commit simplifies the CHECK-NOT block to only check for the
presence of any lifetime start marker since that is effectively what
the test was testing at the moment.

Reviewed By: junparser

Differential Revision: https://reviews.llvm.org/D99856

As it is being noted in D99249, lack of alignment information on `this`
has been preventing LICM from happening.

For some time now, lack of alignment attribute does *not* imply
natural alignment, but an alignment of `1`.
Also, we used to treat dereferenceable as implying alignment,
but we no longer do, so it's a bugfix.

Differential Revision: https://reviews.llvm.org/D99790

This will fix swap-reduction in DAGISel for cases where COPY_TO_REGCLASS
has multiple uses.

The test case added in 258f055e was lacking two important details for the test infrastructure. ae217bf1 added the executable to LLVM_TEST_DEPENDS in CMake to make sure the exectubale gets built before we run the test suite. This patch adds a ToolSubst for the executable in LIT, which replaces the tool invokation in the RUN line with an absolute path. It makes sure we don't run accidentally run some other tool from the user's PATH. The test works without it in case LLVM's main binary directory happens to be the working directory (which is default apparently). Configurations that don't build the examples ignore failures for this ToolSubst (and won't run the test).

Reviewed By: echristo

Differential Revision: https://reviews.llvm.org/D99931

D88631 introduced a set of knobs to tweak how the stack protector is codegen'd for x86 targets, including the offset from the base register where the stack cookie is located. The `StackProtectorGuardOffset` field in `TargetOptions` was left uninitialized instead of being reset to its neutral value -1, making it possible to emit nonsensical code if the frontend doesn't change the field value at all before feeding the `TargetOptions` to the target machine initializer.

Reviewed By: nickdesaulniers

Differential Revision: https://reviews.llvm.org/D99952

A value from reachable block may come to a Phi node as its input from
unreachable block. This may confuse matchSimpleRecurrence  which
has no access to DomTree and can falsely recognize something as a recurrency
because of this effect, as the attached test shows.

Patch `ae7b1e` deals with half of this problem, but it only accounts from
the case when an unreachable instruction comes to Phi as an input.

This patch provides a generalization by checking that no Phi block's
predecessor is unreachable (no matter what the input is).

Differential Revision: https://reviews.llvm.org/D99929
Reviewed By: reames

This reverts commit 849d3729 which
depends on 31d219d2 that was reverted.

LWG3175 identifies that the `common_reference` requirement for
`swappable_with` is over-constraining and doesn't need to concern itself
with cv- or reference qualifiers.

Differential Revision: https://reviews.llvm.org/D99817

This reverts commit 31d219d2 which
causes an infinite loop when compiling the XRay runtime.

When both arguments are the same for these ops, propagate this argument.

Stop emulating dsymutil-classic which only cached the last used CIE for
reuse.