Differential Revision: https://reviews.llvm.org/D107920

[RISCV] Teach vsetvli insertion pass that it doesn't need to insert vsetvli for unit-stride or strided loads/stores in some cases.

For unit-stride and strided load/stores we set the SEW operand of
the pseudo instruction equal the EEW in the opcode. The LMUL
of the pseudo instruction is the LMUL we want.

These instructions calculate EMUL=(EEW/SEW) * LMUL. We can use
this to avoid changing vtype if the SEW/LMUL of the previous
vtype matches the EEW/EMUL ratio we need for the instruction.

Due to how the global analysis works, we can only do this
optimization when the previous vsetvli was produced in the block
containing the store. We need to know in the first phase if the
vsetvli will be inserted so we can propagate information to
the successors in the second phase correctly. This means we can't
depend on predecessors.

Reviewed By: rogfer01

Differential Revision: https://reviews.llvm.org/D106601

Instead of using scalar size divided by 8 for segment loads, get
the alignment from clang's type system.

Make vleff match for consistency.

Also replace uses of getPointerElementType() which will be removed as part of the OpaquePtr changes.

Reviewed By: HsiangKai

Differential Revision: https://reviews.llvm.org/D106738

We really shouldn't deal with a conditional branch that can be trivially
constant-folded into an unconditional branch.

Indeed, barring failure to trigger BB reprocessing, that should be true,
so let's assert as much, and hope the assertion never fires.
If it does, we have a bug to fix.

Mainly, i want to add an assertion that `SimplifyCFGOpt::simplifyCondBranch()`
doesn't get asked to deal with non-unconditional branches,
and if i do that, then said assertion fires on existing tests,
and this is what prevents it from firing.

The function was introduced without a use on Sep 16, 2011 in
commit 5acab501.

This reverts commit b6113548.

This is a direct translation of the select folds added with
D53033 / D53036 and another step towards canonicalization
using the intrinsics (see D98152).

Recent work in runtime assignments failed an assertion in fir-dev
while running tests (flang/test/Semantics/defined-ops.f90).  This
test didn't fail in llvm-project/main because only the "new" Arm
driver is used now, and that only builds runtime derived type information
tables when some debug dumping options are enabled.

So add a reproducing test case to another test that is run with
-fdebug-dump-symbols, and fix the crash by emitting special procedure
binding information only for type-bound generic ASSIGNMENT(=) bindings
that are relevant to the runtime support library for use in intrinsic
assignment of derived types.

Differential Revision: https://reviews.llvm.org/D107918

New ports in glibc typically don't define ELF_INITFINI, so
DT_INIT/DT_FINI support is disabled.
(rhel ppc64le likely patches their glibc this way as well.)
musl can disable DT_INIT/DT_FINI via -DNO_LEGACY_INITFINI.

So we cannot guarantee ctor()/dtor() will be printed.

Fixes miscompile of calls into ocml. Bug 51445.

The stack variable `double __tmp` is moved to dynamically allocated shared
memory by CGOpenMPRuntimeGPU. This is usually fine, but when the variable
is passed to a function that is explicitly annotated address_space(5) then
allocating the variable off-stack leads to a miscompile in the back end,
which cannot decide to move the variable back to the stack from shared.

This could be fixed by removing the AS(5) annotation from the math library
or by explicitly marking the variables as thread_mem_alloc. The cast to
AS(5) is still a no-op once IR is reached.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D107971

Currently, LNICM pass does not support sinking instructions out of loop nest.
This patch enables LNICM to sink down as many instructions to the exit block of outermost loop as possible.

Reviewed By: Whitney

Differential Revision: https://reviews.llvm.org/D107219

This is a fix for commit d2c5cbc3

Previoulsy debug-info-for-profiling and pseudo-probe-for-profiling are mutual exclusive because they compete the dwarf discrimnator for callsites on the IR. This changes allows to use the two switches together. The side effect is that callsite discriminators will be taken by pseudo probe, while discriminators for other instructions are still available for AutoFDO use. This is less than ideal, however, it still allows us a chance to smoothly transition from AutoFDO to CSSPGO, by collecting both profiles from a CSSPGO binary.

Reviewed By: wenlei, wmi

Differential Revision: https://reviews.llvm.org/D107876

This was just added with 6de1dbbd , and I missed
pulling the extra arg from the final revision.

If we cast a function at the call site it is hard(er) to get the rewrite
correct, let's not attempt it for now.

Fixes PR51448.

See D106309.

Differential Revision: https://reviews.llvm.org/D107906

Given a constant operand, the MVE and DAGCombine combines could fight,
each redistributing in the opposite order. Add a guard to the MVE
vecreduce distribution to prevent that.

Add a check for enforcing minimum length for variable names. A default
minimum length of three characters is applied to regular variables
(including function parameters). Loop counters and exception variables
have a minimum of two characters. Additionally, the 'i', 'j' and 'k'
are accepted as legacy values.

All three sizes, as well as the list of accepted legacy loop counter
names are configurable.

This diff was accidentally committed with:
1b5a1958

This is an adaptation of D41603 and another step on the way
to canonicalizing to the intrinsic forms of min/max.

See D98152 for status.

When depth > 0, callee frame address is used to compute the return address of
callee producing improper return address. This patch adds the fix to use caller
frame address to compute the return address of callee.

Reviewed By: nemanjai, #powerpc

Differential revision: https://reviews.llvm.org/D107646

Move InlineOrder to separated file.

Reviewed By: kazu

Differential Revision: https://reviews.llvm.org/D107831

Add CHECK-LABELs and fix string substitution to actually match the
previous definition.

The compilation of the file
526.blender_r/src/blender/source/blender/editors/space_logic/logic_ops.c
from the SPEC CPU 2017 benchmarks took excessive time to compute

    InvalidDomain.gist_params(Ctx)

Simplifying beforehand, specifically using isl_set_detect_equalities,
reduces the computation time to a negible level again.

The default compiler-generated MutexSet::Desc::operator=()
now contains memcpy() call since Desc become bigger.
This fails in debug mode since we call interceptor from within the runtime.
Define own operator=() using internal_memcpy().
This also makes copy ctor necessary, otherwise:
tsan_mutexset.h:33:11: warning: definition of implicit copy constructor for
'Desc' is deprecated because it has a user-declared copy assignment operator
And if we add copy ctor, we also need the default ctor
since it's called by MutexSet ctor.

Depends on D107911.

Reviewed By: melver

Differential Revision: https://reviews.llvm.org/D107959

Dependent commit a1ef81de has been reverted in a1ef81de.

This reverts commit a1ef81de.

Broke the MLIR buildbot.

We currently memorize u64 id + epoch for each mutex.
The new tsan runtime will memorize address + stack_id instead.
But switching to address + stack_id requires new trace,
which in turn requires new MutexSet and some other changes.
Extend MutexSet to support both new and old info to break
the dependency cycles. The plan is to remove the old
info/methods after switching to the new runtime.

Reviewed By: vitalybuka, melver

Differential Revision: https://reviews.llvm.org/D107910

Introducing a plugin API and a simple HelloWorld Plugin example.
This patch adds the `-load` and `-plugin` flags to frontend driver and
the code around using custom frontend actions from within a plugin
shared library object.

It also adds to the Driver-help test to check the help option with the
updated driver flags.

Additionally, the patch creates a plugin-example test to check the
HelloWorld plugin example runs correctly. As part of this, a new CMake
flag (`FLANG_BUILD_EXAMPLES`) is added to allow the example to be built
and for the test to run.

This Plugin API has only been tested on Linux.

Reviewed By: awarzynski

Differential Revision: https://reviews.llvm.org/D106137

The existing logic for per-target libc++ include directories only
seem to exist for the Gnu and Fuchsia drivers, added in
ea12d779 / D89013.

This is less generic than the corresponding case in the Gnu driver,
but matches the existing level of genericity in the MinGW driver
(and others too).

Differential Revision: https://reviews.llvm.org/D107893

Consistently use 'w64' as vendor string; it was 'pc' for the original
i686 triple added in 91bd6c92, but the later x86_64 triple
used 'w64' as vendor, added in d6c1f37f.

When the arm triples were added in c84ad73a, the differing
vendors were copied over accidentally to the arm targets too.

When using per-target runtime directories, having inconsistent vendor
parts of the target triples is fatal.

Differential Revision: https://reviews.llvm.org/D107894

In e72403f9, we added the flag
"--no-dynamicbase" for disabling the dynamicbase flag which we set
by default. At the time, ld.bfd didn't have any corresponding
option (as ld.bfd defaulted to not setting the flag). Almost at
the same time, corresponding options were added to ld.bfd for
disabling it (while it was being enabled by default), with a
different name, "--disable-dynamicbase".

Thus add the "--disable-dynamicbase" option. Make this default
one advertised in the help listing, but keep the "--no-dynamicbase"
form as an alias. Also improve checking for the last option set
if there are multiple ones on the same command line.

Also add corresponding disable options for a lot of other flags
that we set by default, also added in ld.bfd in the same commit:
https://sourceware.org/git/?p=binutils-gdb.git;a=commitdiff;h=514b4e191d5f46de8e142fe216e677a35fa9c4bb

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