isSmallestNormalized() only considers the magnitude, not the sign.

We can only use fli for the positive normalized value.

https://reviews.llvm.org/D146358 was assuming that all subobjects have
their own name (`SubobjectDecl`), but it was not true for array
elements.

Fixes https://github.com/llvm/llvm-project/issues/67317

Leak could be real, as the code terminates before freeing the memory.

This patch fixes the various crashlog test failures following
ec456ba9

, which renamed the process member variable in the Scripted
Thread python base class.

This patch updates the crashlog scripted process implementation to
reflect that change.

Signed-off-by: Med Ismail Bennani <ismail@bennani.ma>

* Consider the darker/clang-format command to have failed if the exit
   code is non-zero, regardless of the stdout/stderr output.
* Propagate stderr from the formatter command to the script's
   caller (and into the GitHub log).
* On success, dump stdout to the caller, so it ends up in GitHub's log.
   I'm not sure what this would ever be, but if it exists, it should be
   preserved.
* Just before the script exits, if any formatter failed, print a line
   showing which formatters failed.

Renaming cds-> cdsort for consistency. This is NFC unless somebody uses
older names

The interesting stuff is of course still coming ;-)

These tests all need an x86 backend for the correct data layout.

This reverts commit ddbaa11e.

Reapply the original commit, the broken test was repaired in 5e51363f in the meantime.

This reverts commit c2a1249a.

The MLIR bots are broken with an omp test failure.

This also renames DoubleXLenLLT to sDoubleXLen.

Changes:

1. For both dimToLvl and lvlToDim, always returns the actual map instead
of AffineMap() for identity map.
2. Updated custom builder for encoding to have default values.
3. Non-inferable lvlToDim will still return AffineMap() during
inference, so it will be caught by verifier.

pthread_rwlock is faster than std::shared_mutex, especially in heavily
threaded code such as the Swift compiler.

rdar://117445844

Make the cache have the fully control on how many blocks to be popped
(At before, it depended the number of blocks stored in the
TransferBatch)

This patch introduces an `OperatingSystem` base implementation in the
`lldb` python module to make it easier for lldb users to write their own
implementation.

The `OperatingSystem` base implementation is derived itself from the
`ScriptedThread` base implementation since they share some common grounds.

To achieve that, this patch makes changes to the `ScriptedThread`
initializer since it gets called by the `OperatingSystem` initializer.

I also took the opportunity to document the `OperatingSystem` base
class and methods.

Differential Revision: https://reviews.llvm.org/D159315



Signed-off-by: Med Ismail Bennani <ismail@bennani.ma>

This patch aims to consolidate the OperatingSystem scripting affordance
by introducing a stable interface that conforms to the
Scripted{,Python}Interface.

This unify the way we call into python methods from lldb while
also improving its capabilities by allowing us to pass lldb_private
objects are arguments.

Differential Revision: https://reviews.llvm.org/D159314



Signed-off-by: Med Ismail Bennani <ismail@bennani.ma>

This reverts commit 85f6b2fa.

[NFC] In InstrProf, generalize helper functions to take 'GlobalObject'. They currently take 'Functions' as function parameters or have 'Func' in the name. (#70287)

- For instance, `collectPGOFuncNameStrings` is reused a lot in https://github.com/llvm/llvm-project/pull/66825 to get the compressed vtable names; and in some added callsites it's just confusing to see 'func' since context clearly shows it's not. This function currently just takes a list of strings as input so name it to `collectGlobalObjectNameStrings`
- Do the rename in a standalone patch since the method is used in non-llvm codebase. It's easier to rollback this NFC in case rename in that codebase takes longer.

The runtime needs to know about the acceptable launch bounds, especially
if the compiler (middle- or backend) assumed those bounds. While this
patch does not yet inform the runtime, it stores the bounds in a place
that can/will be accessed and is associated with the kernel.

This test depends on the datalayout.

These are copies of tests in flang/test/Lower/OpenMP/FIR
copyin.f90 is autogenerated.

We now provide the information about the min/max thread and team count
from to the OMPIRBuilder, no matter what the source was. That means we
unify `thread_limit`, `num_teams`, `num_threads` handling with the
target specific attriutes (`__launch_bounds__` and
`amdgpu_flat_work_group_size`). This is in preparation to pass the
values to the runtime, and to allow the middle-end (OpenMP-opt) to
tighten the values if it seems appropriate. There is no "real" change
after this commit.

Globalization and SPMD are different things that used to be conflated.
Some leftover crossover interactions remain, trying to remove them now.

The workaround code ensure we always call __kmpc_kernel_parallel, but it
did so in a racy manner as the initialization might not have been
completed yet. To avoid introducing a sync, we move the workaround into
the deinit function for now.

On Windows the warning/error messages print opt.exe instead of opt, so
just drop this part of the check.

Since e39f6c18 these tests require
a valid target in order to compute the data layout.