Changes

[LoopVectorize][X86] Add operands to make it more obvious what line the CHECK concerns

As we're checking the cost debug analysis these should match the original IR line - so we shouldn't have any variable naming issues.

I'm investigating v4i32 mul -> PMADDDW costs handling (for PR47437) and these CHECK lines were proving tricky to keep track of

[SelectionDAG] Make WidenVecRes_Convert work for scalable vectors.

Most of the code wasn't yet scalable safe, although most of the
code conceptually just works for scalable vectors. This change
makes the algorithm work on ElementCount, where appropriate,
and leaves the fixed-width only code to use `getFixedNumElements`.

Reviewed By: david-arm

Differential Revision: https://reviews.llvm.org/D110058

[hwasan] also omit safe mem[cpy|mov|set].

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D109816

Don't fold (select C, (gep Ptr, Idx), Ptr) if C is vector but Idx is scalar

The folding rule (select C, (gep Ptr, Idx), Ptr) -> (gep Ptr, (select C,
Idx, 0)) creates a malformed SELECT IR if C is a vector while Idx is scalar.

  SELECT VecC, ScalarIdx, 0

We could splat Idx to a vector but it defeats the purpose of
optimisation. Don't apply the folding rule in this case.

This fixes a regression from commit d561b6fbdbe6d1da05fd92003a4ac1e37bf4b8bc.

Unbreak module builds by making InstructionWorklist.h non-modular

This regressed in D110181 and apparently the header intentionally requires
DEBUG_TYPE to be defined by the including file. Just exclude the header from
the module to unbreak the build.

[ARM] Allow smaller VMOVL in tail predicated loops

This allows VMOVL in tail predicated loops so long as the the vector
size the VMOVL is extending into is less than or equal to the size of
the VCTP in the tail predicated loop. These cases represent a
sign-extend-inreg (or zero-extend-inreg), which needn't block tail
predication as in https://godbolt.org/z/hdTsEbx8Y.

For this a vecsize has been added to the TSFlag bits of MVE
instructions, which stores the size of the elements that the MVE
instruction operates on. In the case of multiple size (such as a
MVE_VMOVLs8bh that extends from i8 to i16, the largest size was be
chosen). The sizes are encoded as 00 = i8, 01 = i16, 10 = i32 and 11 =
i64, which often (but not always) comes from the instruction encoding
directly. A unit test was added, and although only a subset of the
vecsizes are currently used, the rest should be useful for other cases.

Differential Revision: https://reviews.llvm.org/D109706

[lldb] [Windows] Fix continuing from breakpoints and singlestepping on ARM/AArch64

Based on suggestions by Eric Youngdale.

This fixes https://llvm.org/PR51673.

Differential Revision: https://reviews.llvm.org/D109777

[Matrix] Emit assumption that matrix indices are valid.

The matrix extension requires the indices for matrix subscript
expression to be valid and it is UB otherwise.

extract/insertelement produce poison if the index is invalid, which
limits the optimizer to not be bale to scalarize load/extract pairs for
example, which causes very suboptimal code to be generated when using
matrix subscript expressions with variable indices for large matrixes.

This patch updates IRGen to emit assumes to for index expression to
convey the information that the index must be valid.

This also adjusts the order in which operations are emitted slightly, so
indices & assumes are added before the load of the matrix value.

Reviewed By: erichkeane

Differential Revision: https://reviews.llvm.org/D102478

Revert "[CodeGen] regenerate test checks; NFC"

This reverts commit 52832cd917af00e2b9c6a9d1476ba79754dcabff.
The motivating commit 2f6b07316f5 caused several bots to hit
an infinite loop at stage 2, so that needs to be reverted too
while figuring out how to fix that.

Revert "[InstCombine] fold cast of right-shift if high bits are not demanded"

This reverts commit 2f6b07316f560a1f6d225919019dff2e5d6346e5.

This caused several bots to hit an infinite loop at stage 2,
so it needs to be reverted while figuring out how to fix that.

[Passes] Run vector-combine early with -fenable-matrix.

IR with matrix intrinsics is likely to also contain large vector
operations, which can benefit from early simplifications.

This is the last step in a series of changes to improve code-gen for
code using matrix subscript operators with the C/C++ matrix extension in
CLang, like

    using matrix_t = double __attribute__((matrix_type(15, 15)));

    void foo(unsigned i, matrix_t &A, matrix_t &B) {
      for (unsigned j = 0; j < 4; ++j)
        for (unsigned k = 0; k < i; k++)
          B[k][j] -= A[k][j] * B[i][j];
    }

https://clang.godbolt.org/z/6dKxK1Ed7

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D102496

[gn build] (manually) port f8b1cc365786

[gn build] Port 7a320b279d07

[SelectionDAG] Add PromoteIntOp_INSERT_SUBVECTOR.

This is required to codegen something like:
  <vscale x 8 x i16> @llvm.experimental.vector.insert(<vscale x 8 x i16> %vec,
                                                      <vscale x 2 x i16> %subvec,
                                                      i64 %idx)
where the output vector is legal, but the input vector needs promoting.

It implements this by performing the whole operation on the promoted type,
and then truncating the result.

Reviewed By: david-arm, craig.topper

Differential Revision: https://reviews.llvm.org/D110059

[lldb] JITLoaderGDB tests can use lli in ORC greedy mode

At first, lli only supported lazy mode for ORC. Greedy mode was added with e1579894d205 and is the default settings now. JITLoaderGDB tests don't rely on laziness, so we can switch them to greedy and remove some complexity.

[SLP][NFC]Rename function in the test for better matching of the
transformation.

[ELF][test] Restore important part of ICF alignment test

Restore the checking of addresses in ICF test which was testing the
behaviour of ICF with regards to different alignments of otherwise
identical sections. Also make the test more robust to layout changes.

Differential Revision: https://reviews.llvm.org/D110090

AArch64: use indivisible cmpxchg for 128-bit atomic loads at O0

Like normal atomicrmw operations, at -O0 the simple register-allocator can
insert spills into the LL/SC loop if it's expanded and visible when regalloc
runs. This can cause the operation to never succeed by repeatedly clearing the
monitor. Instead expand to a cmpxchg, which has a pseudo-instruction for -O0.

[SelectionDAG] Remove PromoteIntOp_EXTRACT_SUBVECTOR.

This code seems untested and is likely obsolete, because this case
should already be handled by the code that legalizes the result type
of EXTRACT_SUBVECTOR.

Reviewed By: craig.topper

Differential Revision: https://reviews.llvm.org/D110061

[AArch64][SVE] Add extract_subvector patterns for unpacked fp16 and bfloat types.

Reviewed By: david-arm

Differential Revision: https://reviews.llvm.org/D110163

[mailmap] Add entry for myself

[OpenMP] Make sure the Thread ID function is not removed

Summary:
The thread ID function was reintroduced in D110195, but could
potentially be removed by the optimizer. Make the function noinline to
preserve the call sites and add it to the externalization RAII so its
definition is not removed by the attributor.

[AArch64][SVE] NFC: Move extract_subvector tests around.

This patch splits up sve-extract-vector.ll into
  * sve-extract-fixed-vector.ll
  * sve-extract-scalable-vector.ll

For testing extracts of a fixed-width or scalable sub-vector from a
scalable source vector, respectively.

[InstCombine] Update InstCombine to use poison instead of undef for shufflevector's placeholder (1/3)

This patch is for fixing potential shufflevector-related bugs like D93818.
As D93818, this patch change shufflevector's default placeholder to poison.
To reduce risk, it was divided into several patches, and this patch is for InstCombineCasts.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D110226

[Target][CodeGen] Remove default CostKind arguments on inner/impl TTI overrides

Based off a discussion on D110100, we should be avoiding default CostKinds whenever possible.

This initial patch removes them from the 'inner' target implementation callbacks - these should only be used by the main TTI calls, so this should guarantee that we don't cause changes in CostKind by missing it in an inner call. This exposed a few missing arguments in getGEPCost and reduction cost calls that I've cleaned up.

Differential Revision: https://reviews.llvm.org/D110242

[Sanitizer] Add Windows header for _mkdir

This will hopefully fix the sanitizer_windows bot failure after D109794:
https://lab.llvm.org/buildbot/#/builders/127/builds/17222

[libc++][NFC] Add link to Discord channel from documentation

[InstCombine] Update InstCombine to use poison instead of undef for shufflevector's placeholder (2/3)

This patch is for fixing potential shufflevector-related bugs like D93818.
As D93818, this patch change shufflevector's default placeholder to poison.
To reduce risk, it was divided into several patches, and this patch is for InstCombineCompares and InstructionCombining.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D110227

[OpenMP][Offloading] Use bitset to indicate execution mode instead of value

The execution mode of a kernel is stored in a global variable, whose value means:
- 0 - SPMD mode
- 1 - indicates generic mode
- 2 - SPMD mode execution with generic mode semantics

We are going to add support for SIMD execution mode. It will be come with another
execution mode, such as SIMD-generic mode. As a result, this value-based indicator
is not flexible.

This patch changes to bitset based solution to encode execution mode. Each
position is:
[0] - generic mode
[1] - SPMD mode
[2] - SIMD mode (will be added later)

In this way, `0x1` is generic mode, `0x2` is SPMD mode, and `0x3` is SPMD mode
execution with generic mode semantics. In the future after we add the support for
SIMD mode, `0b1xx` will be in SIMD mode.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D110029

[libc++] Disallow volatile types in std::allocator

LWG 2447 is marked as `Complete`, but there is no `static_assert` to
reject volatile types in `std::allocator`. See the discussion at
https://reviews.llvm.org/D108856.

Add `static_assert` in `std::allocator` to disallow volatile types. Since this
is an implementation choice, mark the binding test as `libc++` only.

Remove tests that use containers backed by `std::allocator` that test
the container when used with a volatile type.

Reviewed By: ldionne, #libc

Differential Revision: https://reviews.llvm.org/D109056

[InstCombine] Update InstCombine to use poison instead of undef for shufflevector's placeholder (3/3)

This patch is for fixing potential shufflevector-related bugs like D93818.
As D93818, this patch change shufflevector's default placeholder to poison.
To reduce risk, it was divided into several patches, and this patch is for InstCombineVectorOps.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D110230

[SLP] getReductionCost - use explicit TTI::TCK_RecipThroughput CostKind. NFCI.

Avoid relying on the default cost kinds in TTI calls (we already do this in other places in SLP) - noticed while trying to see how much work it'd be to extend D110242 and remove all remaining uses of default CostKind arguments.

[ORC] Re-enable ELF DebugObjectManagerPlugin tests

These tests were disabled by accident after D107640. Actually, REQUIRES lines don't support `x86_64` and so these tests stopped running on all targets.
`native && target-x86_64` should be the correct term to express "x86_64 host targeting native arch".

[CodeGen] Remove redundant declaration MIRCanonicalizerID (NFC)

Note that MIRCanonicalizerID is declared in
llvm/include/llvm/CodeGen/Passes.h, which MIRCanonicalizerPass.cpp
includes.

Identified with readability-redundant-declaration.

[SLP][NFC]Add a test to show an issue with incorrectly extracted
pointers.

[CSSPGO] Set PseudoProbeInserter as a default pass.

Currenlty PseudoProbeInserter is a pass conditioned on a target switch. It works well with a single clang invocation. It doesn't work so well when the backend is called separately (i.e, through the linker or llc), where user has always to pass -pseudo-probe-for-profiling explictly. I'm making the pass a default pass that requires no command line arg to trigger, but will be actually run depending on whether the CU comes with `llvm.pseudo_probe_desc` metadata.

Reviewed By: wenlei

Differential Revision: https://reviews.llvm.org/D110209

[NFC] clang-format -i llvm/lib/Frontend/OpenMP/OMPIRBuilder.cpp

[RISCV] Optimize vp.store with an all ones mask to avoid a vmset.

We can use riscv_vse intrinsic instead of riscv_vse_mask. The code here
is based on similar code for handling masked.scatter and vp.scatter.

Reviewed By: frasercrmck

Differential Revision: https://reviews.llvm.org/D110206

[mlir][sparse] replace ad-hoc MemRef struct with CRunnerUtils definition

This revision removes the ad-hoc MemRefs that were needed using the old
ABI (when we still passed by value) and replaces them with the shared
StridedMemRef definitions of CRunnerUtils (possible now that we pass by
pointer). This avoids code duplication and makes sure we have a consistent
view of strided memory references in all our support libraries.

Reviewed By: jsetoain

Differential Revision: https://reviews.llvm.org/D110221

[OpenMP] Add function tracing debugging to device RTL

This patch adds support for an RAII struct that will print function
traces when placed inside of a function declaration. Each successive
call will increase the indentation to make it easier to visually
inspect.

Reviewed By: jdoerfert

Differential Revision: https://reviews.llvm.org/D110202

[ORC] DebugObjectManagerPlugin tests can use lli in ORC greedy mode

Initially, lli only supported lazy mode for ORC. Greedy mode was added with e1579894d205 and it's the default setting now. DebugObjectManagerPlugin tests don't rely on laziness, so we can switch them to greedy in order to avoid some unnecessary complexity.

[HWASan] Use a single .weak binding in asm.

Specifying .global and .weak causes a compiler warning:

  warning: __sigsetjmp changed binding to STB_WEAK

Specifying only .weak should have the same effect without causing a
warning.

Reviewed By: eugenis

Differential Revision: https://reviews.llvm.org/D110178

[libc] Add an implementation of bsearch.

Reviewed By: michaelrj

Differential Revision: https://reviews.llvm.org/D110222

[SimplifyCFG] Ignore free instructions when computing cost for folding branch to common dest

When determining whether to fold branches to a common destination by
merging two blocks, SimplifyCFG will count the number of instructions to
be moved into the first basic block. However, there's no reason to count
free instructions like bitcasts and other similar instructions.

This resolves missed branch foldings with -fstrict-vtable-pointers in
llvm-test-suite's lambda benchmark.

Reviewed By: spatel

Differential Revision: https://reviews.llvm.org/D108837

Add document numbers for the C99 status page.

This doesn't add all of the document numbers, but it adds a bunch of
them. Not all of the documents are available on the committee page
(they're old enough that they come from a time when the mailing was
comprised of physical pieces of paper), so some of the documents listed
are assumed to be correct based on my reading of editor's reports.

[NFCI][CodeGen, AArch64] Fix inconsistent TargetCostKind types.

The pass uses different cost kinds to estimate "old" and "interleaved" costs:
default cost kind for all targets override `getInterleavedMemoryOpCost()` is
`TCK_SizeAndLatency`. Although at the moment estimated `TCK_Latency` costs are
equal to `TCK_SizeAndLatency`, (so the change is NFC) it may change in future.

Reviewed By: RKSimon

Differential Revision: https://reviews.llvm.org/D110100

[llvm-profgen] An option to dump disasm of specified symbols

For large app, dumping disasm of the whole program can be slow and result in gianant output. Adding a switch to dump specific symbols only.

Reviewed By: wlei

Differential Revision: https://reviews.llvm.org/D110079

[mlir][linalg] Fix interchange initialization in fusion on tensors.

If no interchange vector is given initialize it with the identity permutation from 0 to number of loops.

Reviewed By: mravishankar

Differential Revision: https://reviews.llvm.org/D110249

Don't diagnose unused but set when the Cleanup attribute is used.

This applies to -Wunused-but-set-variable and
-Wunused-but-set-parameter.

This addresses bug 51865.

Differential Revision: https://reviews.llvm.org/D109862

[Analysis] improve function matching for strlen libcall

The return type of strlen is size_t, not just any integer.

This is a partial fix for an example based on:
https://llvm.org/PR50836

There's another bug here because we can still crash
processing a real strlen or something that looks like it.

Change error for storage-class to mean linkage, fix lang-linkage diag

Allow multiversioning declarations to match when the actual formal
linkage matches, not just when the storage class is identical.
Additionally, change the ambiguous 'linkage' mismatch to be more
specific and say 'language linkage'.

Mark CFG as preserved in TypePromotion and InterleaveAccess passes

Neither of these passes modify the CFG, allowing us to preserve DomTree
and LoopInfo across them by using setPreservesCFG.

Differential Revision: https://reviews.llvm.org/D110161

[lldb] Remove IRExecutionUnit::CollectFallbackNames

The work that IRExecutionUnit::CollectFallbackNames is basically the
work that `CPlusPlusLanguage::GetDemangledFunctionNameWithoutArguments`
does already. It's also (at time or writing) specific to C++, so it can
be folded into `IRExecutionUnit::CollectCandidateCPlusPlusNames`.

Differential Revision: https://reviews.llvm.org/D109928

DebugInfo: Add (initially no-op) -gsimple-template-names={simple,mangled}

This is to build the foundation of a new debug info feature to use only
the base name of template as its debug info name (eg: "t1" instead of
the full "t1<int>"). The intent being that a consumer can still retrieve
all that information from the DW_TAG_template_*_parameters.

So gno-simple-template-names is business as usual/previously ("t1<int>")
   =simple is the simplified name ("t1")
   =mangled is a special mode to communicate the full information, but
   also indicate that the name should be able to be simplified. The data
   is encoded as "_STNt1|<int>" which will be matched with an
   llvm-dwarfdump --verify feature to deconstruct this name, rebuild the
   original name, and then try to rebuild the simple name via the DWARF
   tags - then compare the latter and the former to ensure that all the
   data necessary to fully rebuild the name is present.

[OpenMP] Fix KeepAlive usage

Summary:
Functions were called the wrong way around, this didn't keep the symbol
alive.

[ELF][AArch64] Refine and fix the condition when BTI/PAC PLT needs bti c

(As I mentioned in https://reviews.llvm.org/D62609#1534158 ,
the condition for using bti c for executable can be loosened.)

In two cases the address of a PLT may escape:

* canonical PLT entry for a STT_FUNC
* non-preemptible STT_GNU_IFUNC which is converted to STT_FUNC

The first case can be detected with `needsPltAddr`.

The second case is not straightforward to detect because for the Relocations.cpp
created `directSym`, it's difficult to know whether the associated `sym` has
exercised the `!needsPlt(expr)` code path. Just use the conservative `isInIplt`
condition. A non-preemptible ifunc not referenced by non-GOT-generating
non-PLT-generating relocations will have an unneeded `bti c`, but the cost is acceptable.

The second case fixes a bug as well: a -shared link may have non-preemptible ifunc.
Before the patch we did not emit `bti c` and could be wrong if the PLT address escaped.
GNU ld doesn't handle the case: `relocation R_AARCH64_ADR_PREL_PG_HI21 against STT_GNU_IFUNC symbol 'ifunc2' isn't handled by elf64_aarch64_final_link_relocate` (https://sourceware.org/bugzilla/show_bug.cgi?id=28370)

For -shared, if BTI is enabled but PAC is disabled, the PLT entry size increases
from 16 to 24 because we have to select the PLT scheme early, but the cost is
acceptable.

Reviewed By: peter.smith

Differential Revision: https://reviews.llvm.org/D110217

[Analysis] reduce code for isOnlyUsedInZeroEqualityComparison; NFC

There's a bug here noted by the FIXME and visible in variations of PR50836.

[ValueTracking] fix isOnlyUsedInZeroEqualityComparison with no users

This is another problem exposed by:
https://bugs.llvm.org/PR50836

[mlir][sparse] generalize reduction support in sparse compiler

Now not just SUM, but also PRODUCT, AND, OR, XOR. The reductions
MIN and MAX are still to be done (also depends on recognizing
these operations in cmp-select constructs).

Reviewed By: bixia

Differential Revision: https://reviews.llvm.org/D110203

[libc++][test] Remove disable_missing_braces_warning.h from tests

Several tests include `disable_missing_braces_warning.h` but do not need
to. Remove the include.

Inspired from discussion at https://reviews.llvm.org/D109668

Reviewed By: ldionne, #libc, Mordante

Differential Revision: https://reviews.llvm.org/D109711

[libc++][NFCI] Remove uses of _LIBCPP_INLINE_VAR

All supported compilers provide support for inline variables in C++17 now.
Also, as a fly-by fix, replace some uses of _LIBCPP_CONSTEXPR by just
constexpr.

The only exception in this patch is `std::ignore`, which is provided
prior to C++17. Since it is defined in an anonymous namespace, it always
has internal linkage anyway, so using an inline variable there doesn't
provide any benefit. Instead, `inline` was removed entirely on `std::ignore`.

Differential Revision: https://reviews.llvm.org/D110243

Reset operation when canceling root update transaction

Should reset the operation to original state when canceling the updates.

Reviewed By: rriddle, ftynse

Differential Revision: https://reviews.llvm.org/D110176

[mlir:DataFlowAnalysis] Reprocess the arguments of already executable edges

This fixes a bug where we discover new information about the arguments of an
already executable edge, but don't visit the arguments. We only visit the arguments, and not the block itself, so this commit shouldn't really affect performance at all.

Fixes PR#51871

Differential Revision: https://reviews.llvm.org/D110197