Fix the other typo

arc messed up

Nah, it is a typo in both places...
I don't understand what's going on.

Remove unnecessary -Werror.

In D146920#4230541, @ahatanak wrote:

I added -Werror because the changes I made in https://reviews.llvm.org/D131714 broke someone's internal build (see @wlei's comment). Apparently, gcc emits a warning if you try to use -Wformat-security without passing -Wformat too.

-Werror is probably only needed for -Wformat-security, but I think passing it in other places doesn't do any harm.

It appears there is a typo in the use of COMPILER_RT_HAS_BUILTIN_FORMAT_SECURITY_FLAG:
append_list_if(COMPILER_RT_HAS_BUILTIN_FORMAL_SECURITY_FLAG -Werror=format-security flags)
FORMAL vs FORMAT.

This now breaks M68k build https://lab.llvm.org/buildbot#builders/192/builds/1022

cc1: error: -Wformat-security ignored without -Wformat [-Werror=format-security]

ping

Mention "Reland" in the commit title

Drop changes to -ftrivial-auto-var-init causing build failures

I briefly glanced at other patches and can't review / accept them without good test coverage.

I like the idea, but the implementation looks more like a PoC than a complete solution.
There is clearly an overlap between RangeList and SliceElements and I don't see why they should be distinct.
I am not a TableGen expert so take my words with a grain of salt. I won't be able to review it well.
Maybe @nhaehnle could take a look?

In D146920#4231272, @smeenai wrote:

Feel free to include the changes from D147164 as part of this patch, which should fix the errors.

Try to refine assertion message.

There seem to be an issue with tsan?

In D146920#4231154, @hctim wrote:

Looks like @smeenai was working on a fix-forward in https://reviews.llvm.org/D147164. Those should be either merged and submitted together or landed together. Thanks :)

Abandoned in favor of D147093.

It crashes on the test from D146930.

Related: D147085
__USER_LABEL_PREFIX__ is always defined. It is usually defined to nothing though, i.e. just
#define __USER_LABEL_PREFIX__
In this case the file compiles just fine. It does not compile if the prefix is non-empty, e.g.:
#define __USER_LABEL_PREFIX__ _
because it is just a token and not a string.
I don't know if it is valid for the label prefix to be non-empty on ELF platforms (and couldn't find an example),
but I don't see why it couldn't be.

Wait for D147093 resolution.

The prefix has only caused troubles; it will probably go away once I am able to build newlib using clang.

In D147085#4228710, @MaskRay wrote:

The code only makes sense for some ELF platforms, not Mach-O platforms where __USER_LABEL_PREFIX__ is (usually) _.
What platforms are you building the code with and seeing a problem?

Fix another occurance

The commit message does not make it clear why this change is necessary.
(And the title does not tell anything at all.)

Does it try to solve the same problem as D146930?

IMHO .gitignore is for files that are strictly project-related. Everything else should go into local .git/info/exclude.

Half of the includes and forward declarations are unused.
I'm not familiar with the ABI to review this part.
However, manual handling of i1, i8, i16 and f32 looks suspicious, because they are all illegal.
Looks like a dead code to me, but I may be wrong.

Drop 'renamable' on the registers that are not really renamable.

Strip changes that are not necessary for fixing the bug.
They will be in a separate commit.

Rebase

The change effectively disables BackwardCopyPropagateBlock that folded the first copy into
the preceding load, eliminating definition of s[60-67], which is necessary to reproduce the bug.

Add a positive test

In D146906#4222742, @chapuni wrote:

I noticed we could deduplicate MVT when I read the article.
https://discourse.llvm.org/t/issues-in-llvm-tblgen-high-parallelized-build/68037/3

The start point of this series is here.
https://discourse.llvm.org/t/rfc-let-intrinsics-td-emit-iit-info-typesig/69186

In D146906#4222713, @chapuni wrote:

@barannikov88 Could you see D146352 and the series of D145937?
I will create a small tablegen (including this) w/o MVT.h.

This wont work as llvm-tblgen depends on MachineValueTypes.h.

I'm not sure this is the right thing to do. The regression was introduced in https://github.com/llvm/llvm-project/commit/f051c1ded40970169cda84b0966331ae7ad424ed

Update commit message

Still LGTM

@thevinster
Thanks for the confirmation.

This still does not look right to me.
Shouldn't PASSTHROUGH_PREFIXES have just been removed in this invocation?

clang's unwind.h also gets installed and is found first when included from a source file (and it includes_next the libunwind's version, but only for apple platforms).
Not sure I understand the logic, may it be somehow related to the fact that clang driver supports different unwind libraries?
Either way, installing libunwind's header by default does not update clang's version, so the issue persists.

LGTM, great job!

Update the comment

I'm afraid this change will go unnoticed by build bots. They must be updated first, and a deprecation warning could help (before removing the support).

LGTM with one note

LGTM
I can confirm that MTVRSAVEv change is the only non-comment change in the generated files.
Thorough work!

I intent to make make InstAlias with "complex" operands (non-empty MIOperandInfo) more robust (stricter syntax). There is also a couple of bugs I'd like to fix.
The change will clobber most of the code here, and I thought it is a good time to move the class to a dedicated file.
Another weak point is that this class is only used in AsmMatcherEmitter and AsmWriterEmitter, while CodeGenInstruction.h is included everywhere.
The class does not have long history; there were only a few functional changes since initial commit.

Keep both spellings