[TRE] allow TRE for non-capturing calls.

[avl]

Summary

The current implementation of Tail Recursion Elimination has a very restricted
pre-requisite: AllCallsAreTailCalls. i.e. it requires that no function
call receives a pointer to local stack. Generally, function calls that
receive a pointer to local stack but do not capture it - should not
break TRE. This fix allows us to do TRE if it is proved that no pointer
to the local stack is escaped. To make sure whether pointer escaped or not,
it examines called function.

[foad]

markTails function set IsTailcall bit for functions which are not
last calls:

It's OK to set "tail" on any call that satisfies these requirements (from https://llvm.org/docs/LangRef.html#call-instruction): "Both markers [tail and musttail] imply that the callee does not access allocas from the caller. The tail marker additionally implies that the callee does not access varargs from the caller."

"tail" does not mean that the call *must* be generated as a tail call. It just means that it's safe to generate it as a tail call if it turns out to be possible (e.g. if the compiler can prove that @noarg doesn't return, or if it can prove that all the code after the call to @noarg has no effect, or so on).

So I don't think there is a bug here that needs to be fixed.

[avl]

It's OK to set "tail" on any call that satisfies these requirements (from https://llvm.org/docs/LangRef.html#call-instruction): "Both markers [tail and musttail] imply that the callee does not access allocas from the caller. The tail marker additionally implies that the callee does not access varargs from the caller."

"tail" does not mean that the call *must* be generated as a tail call. It just means that it's safe to generate it as a tail call if it turns out to be possible (e.g. if the compiler can prove that @noarg doesn't return, or if it can prove that all the code after the call to @noarg has no effect, or so on).

Yes, that is understood: ""tail" does not mean that the call *must* be generated as a tail call".
I intended to make picture consistent: when markTails sees function body, it is evident that the first call is not a tail call. I agree that a compiler "can prove that @noarg doesn't return, or if it can prove that all the code after the call to @noarg has no effect" and this first call could become tail-call. Probably the suitable strategy, in this case, would be to recalculate marking when it is broken(instead of creating false positive marks). There are other cases(compiler could add function calls), which could break existing marking and then would be necessary to recalculate marking.

Having many false positive "tail" marks could create a confusing picture.
I would describe the full problem which I am trying to solve.
(I assumed this patch would be the first one for that problem):

cat test.cpp

int count;
__attribute__((noinline)) void globalIncrement(const int* param) { count += *param; }

void test(int recurseCount)
{
    if (recurseCount == 0) return;
    {
      int temp = 10;
      globalIncrement(&temp);
    }
    test(recurseCount - 1);
}

TRE is not done for that test case. There are two reasons for that:
First is that AllocaDerivedValueTracker does not use the PointerMayBeCaptured interface, and it does not see that &temp is not escaped.
Second is that AllCallsAreTailCalls is used as a pre-requisite for making TRE.
So it requires that all calls would be marked as "tail". This looks too restricted.
Instead, it should check that "&temp" is not escaped in globalIncrement() and that "test"
is a tail recursive call not using allocas. I think the confusion happened exactly because "tail" marking was done for all calls(not for the real tailcalls).

Thus I planned to do the following fixes:

1. cleanup "tail" marking.(this patch)
2. do not use "AllCallsAreTailCalls" as a pre-requisite for TRE.(this patch).
3. use PointerMayBeCaptured inside AllocaDerivedValueTracker.

What do you think about all of this?

It seems to me that it would be good to have consistent "tail" marking.
But if it does not look like a good idea, I could continue to point 2. and 3.

[laytonio]

Relevant llvm-dev thread. Noncapture use of locals disabling TailRecursionElimination

[efriedma]

All "tail" needs to mean is "this call does not reference the current function's stack." That's all, no more or less.

The relevant documentation and APIs are a bit confusing. A "tail" marking is a prerequisite for tailcall optimization, but it's not really related to any of the other prerequisites for tailcall optimization. Dropping the "tail" marking just because the call isn't in a tail position would involve a bunch of work adding and removing "tail" markings, for no benefit.

It makes sense to teach tail recursion elimination not to depend so heavily on tail markings. But I don't think that implies we want to mess with the markings themselves.

I think the confusion happened exactly because "tail" marking was done for all calls(not for the real tailcalls).

I don't think there's any confusion here. It's just using the tail markings as a conservative estimate of capturing because it has to compute them anyway. And TRE in general is simplistic code that nobody has spent much time looking at in a long time. There are very few TRE opportunities in C/C++ code anyway.

[avl]

It makes sense to teach tail recursion elimination not to depend so heavily on tail markings. But I don't think that implies we want to mess with the markings themselves.

Ok. Thus I need to drop part of this patch related to more strict tail marking, and continue with part which changes pre-requisite for TRE.

[avl]

1. deleted code doing more strict tailcall marking.
2. left removal of "AllCallsAreTailCalls".
3. added check for non-capturing calls while tracking alloca.
4. re-titled the patch.

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

145

Please don't add code to examine the callee; if we're not deducing nocapture appropriately, we should fix that elsewhere.

332

What is the new handling for lifetime.end/assume doing?

829

Do you have to redo the AllocaDerivedValueTracker analysis? Is it not enough that the call you're trying to TRE is marked "tail"?

856

I thought we had some tests where we TRE in the presence of recursive calls, like a simple recursive fibonacci. Am I misunderstanding this?

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

145

Ok.

332

They are just skipped. In following test case:

call void @_Z5test5i(i32 %sub)
call void @llvm.lifetime.end.p0i8(i64 24, i8* nonnull %1) #5
call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #5
br label %return

they are generated in between call and ret. It is safe to ignore them while checking whether transformation is possible.

829

Do you have to redo the AllocaDerivedValueTracker analysis?

AllocaDerivedValueTracker analysis(done in markTails) could be reused here.
But marking, done in markTails(), looks like separate tasks. i.e. it is better
to make TRE not depending on markTails(). There is a review for this - https://reviews.llvm.org/D60031
Thus such separation looks useful(To not reuse result of markTails but have it computed inplace).

Is it not enough that the call you're trying to TRE is marked "tail"?

It is not enough that call which is subject to TRE is marked "Tail".
It also should be checked that other calls does not capture pointer to local stack:

// do not do TRE if any pointer to local stack has escaped.
if (!Tracker.EscapePoints.empty())
   return false;

856

right, there is a testcase for fibonacchi:

llvm/test/Transforms/TailCallElim/accum_recursion.ll:@test3_fib

areAllLastFuncCallsRecursive() checking works well for fibonacci testcase:

return fib(x-1)+fib(x-2);

Since, Last funcs call chain is : fib()->fib()->ret.
That check should prevent from such cases:

return fib(x-1)+another_call()+fib(x-2);

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

It is not enough that call which is subject to TRE is marked "Tail". It also should be checked that other calls does not capture pointer to local stack:

If there's an escaped pointer to the local stack, we wouldn't infer "tail" in the first place, would we?

856

That check should prevent from such cases: return fib(x-1)+another_call()+fib(x-2);

Why do we need to prevent this?

[laytonio]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

827

There is no need to pass the function here since its a member variable.

[hiraditya]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

805

can we use isa<> here?

820

CI->getCalledFunction() != &F seems cheaper than canMoveAboveCall

844

Do we need to visit all the instructions twice?

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

827

Ok.

829

If function receives pointer to alloca then it would not be marked with "Tail". Then we do not have a possibility to understand whether this function receives pointer to alloca but does not capture it:

void test(int recurseCount)
{
    if (recurseCount == 0) return;
    int temp = 10;
    globalIncrement(&temp);
    test(recurseCount - 1);
}

test - marked with Tail.
globalIncrement - not marked with Tail. But TRE could be done since it does not capture pointer. But if it will capture the pointer then we could not do TRE. So we need to check !Tracker.EscapePoints.empty().

856

We do not. I misunderstood the canTransformAccumulatorRecursion(). That check could be removed.

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

test - marked with Tail.

For the given code, TRE won't mark the recursive call "tail". That transform isn't legal: the recursive call could access the caller's version of "temp".

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

For the given code, TRE won't mark the recursive call "tail". That transform isn't legal: the recursive call could access the caller's version of "temp".

it looks like recursive call could NOT access the caller's version of "temp":

test(recurseCount - 1);

Caller`s version of temp is accessed by non-recursive call:

globalIncrement(&temp);

If globalIncrement does not capture the "&temp" then TRE looks to be legal for that case.

globalIncrement() would not be marked with "Tail". test() would be marked with Tail.

Thus the pre-requisite for TRE would be: tail-recursive call must not receive pointer to local stack(Tail) and non-recursive calls must not capture the pointer to local stack.

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

Can you give a complete IR example where we infer "tail", but TRE is illegal?

Can you give a complete IR example, we we don't infer "tail", but we still do the TRE transform here?

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

Can you give a complete IR example where we infer "tail", but TRE is illegal?

there is no such example. Currently all cases where we infer "tail" would be valid for TRE.

Can you give a complete IR example, we we don't infer "tail", but we still do the TRE transform here?

For the following example current code base would not infer "tail" for _Z15globalIncrementPKi and as the result would not do TRE for _Z4testi.
This patch changes this behavior: so that if _Z15globalIncrementPKi is not marked with "tail" and does not capture its pointer argument - TRE would be allowed for _Z4testi.

@count = dso_local local_unnamed_addr global i32 0, align 4

; Function Attrs: nofree noinline norecurse nounwind uwtable
define dso_local void @_Z15globalIncrementPKi(i32* nocapture readonly %param) local_unnamed_addr #0 {
entry:
  %0 = load i32, i32* %param, align 4
  %1 = load i32, i32* @count, align 4
  %add = add nsw i32 %1, %0
  store i32 %add, i32* @count, align 4
  ret void
}

; Function Attrs: nounwind uwtable
define dso_local void @_Z4testi(i32 %recurseCount) local_unnamed_addr #1 {
entry:
  %temp = alloca i32, align 4
  %cmp = icmp eq i32 %recurseCount, 0
  br i1 %cmp, label %return, label %if.end

if.end:                                           ; preds = %entry
  %0 = bitcast i32* %temp to i8*
  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #6
  store i32 10, i32* %temp, align 4
  call void @_Z15globalIncrementPKi(i32* nonnull %temp)
  %sub = add nsw i32 %recurseCount, -1
  call void @_Z4testi(i32 %sub)
  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #6
  br label %return

return:                                           ; preds = %entry, %if.end
  ret void
}

; Function Attrs: argmemonly nounwind willreturn
declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #2

; Function Attrs: argmemonly nounwind willreturn
declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #2

attributes #0 = { nofree noinline norecurse nounwind uwtable }
attributes #1 = { nounwind uwtable }
attributes #2 = { argmemonly nounwind willreturn }

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

In your example, we don't infer "tail" for globalIncrement... but we do infer it for the recursive call to test(). I'm suggesting you could just check for "tail" on test(), instead of using AllocaDerivedValueTracker.

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

Checking only test() is not enough. There additionally should be checked globalIncrement().

It is correct that while checking test() there could be checked "Tail" flag. Which does not require using AllocaDerivedValueTracker.

But while checking globalIncrement() there should be checked whether some alloca value escaped or not. "Tail" flag could not be used for that. AllocaDerivedValueTracker allow to do such check:

Tracker.EscapePoints.empty()

If we would not do check for globalIncrement then it is not valid to do TRE. Thus it seems we need to check globalIncrement for escaping pointer and we need to use AllocaDerivedValueTracker for that.

[avl]

addressed comments:

1. removed PointerMayBeCaptured() used for CalledFunction.
2. rewrote CanTRE() to visiting instructions only once.
3. replaced areAllLastFuncCallsRecursive() with isInTREPosition().

I did not address request for not using AllocaDerivedValueTracker yet.
Since there is open question on it. I would address it as soon as the question would be resolved.

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

Checking only test() is not enough. There additionally should be checked globalIncrement().

Can you give a complete IR example where we infer "tail", but TRE is illegal?

there is no such example. Currently all cases where we infer "tail" would be valid for TRE.

I'm not sure how to reconcile this. Are you saying we could infer "tail" in some case where TRE is illegal, but don't right now? Or are you saying that you plan to extend TRE to handle cases where we can't infer "tail" on the recursive call?

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

Or are you saying that you plan to extend TRE to handle cases where we can't infer "tail" on the recursive call?

I think not exactly. more precise would probably be : "I am saying that plan to extend TRE to handle cases where we can't infer "tail" on the NON-recursive NON-last call"

globalIncrement() is non-recursive non-last call in above example. But we need to check whether it captures argument pointer or not to decide whether it is OK to do TRE.

To make things clear - I am suggesting instead of current pre-requisite for TRE :

"All call sites are marked with Tail"

to make following:

"Recursive last calls are marked with "Tail", non-recursive non-last calls are proved to not capture alloca".

For the above example it means : the requirement for test() should stay the same(should be marked with Tail). The requirement for globalIncrement() should be "does not capture alloca".

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

"Recursive last calls are marked with 'tail'" implies "non-recursive non-last calls are proved to not capture alloca".

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

829

I see, thank you for explanations.
There is a test which makes me think that above rule is not always correct:

Transforms/TailCallElim/basic.ll:@test1

; PR615. Make sure that we do not move the alloca so that it interferes with the tail call.
define i32 @test1() {
; CHECK: i32 @test1()
; CHECK-NEXT: alloca
        %A = alloca i32         ; <i32*> [#uses=2]
        store i32 5, i32* %A
        call void @use(i32* %A)
; CHECK: tail call i32 @test1
        %X = tail call i32 @test1()             ; <i32> [#uses=1]
        ret i32 %X
}

I removed usages of AllocaDerivedValueTracker and corrected the test1 from Transforms/TailCallElim/basic.ll.

[avl]

removed usages of AllocaDerivedValueTracker from canTRE().

[laytonio]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

830

You can use for (Instruction &I : instructions(F)) here.

844

Is this correct? I think we want to check these per TRE candidate in findTRECandidate, not just disable TRE in general if one is found.

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

844

I tried to minimize changes and keep old logic here - but yes, it is better to move that check into findTRECandidate(). Will do.

[avl]

check valid TRE candidate into findTRECandidate()().

[laytonio]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

842

Is there any reason to find and validate candidates now only to have to redo it when we actually perform the eliminations? If so, is there any reason this needs to follow a different code path than findTRECandidate? findTRECandidate is doing the same checks, except for canMoveAboveCall and canTransformAccumulatorRecursion, which should probably be refactored into findTRECandidate from eliminateCall anyway. If not then all of this code goes away and we're left with the same canTRE as in trunk.

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

842

We are enumerating all instructions here, so we could understand if there are not TRE candidates and stop earlier. That is the reason for doing it here.

I agree that findTRECandidate should be refactored to have the same checks as here.

What do you think is better to do:

• leave early check for TRE candidates in canTRE or remove it
• refactor findTRECandidate or leave it as is

?

[laytonio]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

842

Yes we are iterating all the instructions here but, unless I am missing something, we would literally just be doing the checks twice for no reason. Look at it this way, best case scenario we have to check all possible candidates once, find none and we're done. Worst case, we check all possible candidates once, find one and have to check all possible candidates a second time. Where as if we remove the early checks we only ever have to check the candidates once. So we wouldn't really be stopping any earlier.

As for refactoring findTRECandidate, I do think that should be done and we should strive to move all the failure conditions out of eliminateCall in order to avoid having to fold a return only to find out we didn't need to. But, I think that is out of the scope of this change, and if we do decide to keep the early checks here then we should say that findTRECandidate does a good enough job to consider this function as having valid candidates.

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

842

Yes we are iterating all the instructions here but, unless I am missing something, we would literally just be doing the checks twice for no reason. Look at it this way, best case scenario we have to check all possible candidates once, find none and we're done. Worst case, we check all possible candidates once, find one and have to check all possible candidates a second time. Where as if we remove the early checks we only ever have to check the candidates once. So we wouldn't really be stopping any earlier.

yes. we would do check twice if there are TRE candidates. my idea was that number of cases when TRE is applicable less then when TRE is not applicable. Thus we would do double instruction navigation more often than double check for candidates. But, I did not measure real impact. Thus, let`s return old logic here as you suggested.

[avl]

removed early check for TRE candidates from canTRE().

[avl]

ping.

[avl]

rebased.

[avl]

@efriedma What do you think about current state of this patch? Is it OK?

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

95

If we're not going to try to do TRE at all on calls not marked "tail", we can probably drop this check.

332

It makes sense we can ignore lifetime.end on an alloca: we know the call doesn't refer to the alloca. (Maybe we should check that the pointer argument is pointing at an alloca? That should usually be true anyway, but better to be on the safe side, I guess.)

I don't think it's safe to hoist assume without additional checks; I think we'd need to check that the call is marked "willreturn"?

Since this is sort of tricky, I'd prefer to split this off into a followup.

810–811

Can you move this FIXME into a more appropriate spot?

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

95

It looks to me that original idea(PR962) was to avoid inefficient code which is generated for dynamic alloca.

Currently there would still be generated inefficient code:

Doing TRE for dynamic alloca requires correct stack adjustment to avoid extra stack usage.
i.e. dynamic stack reservation done for alloca should be restored
in the end of the current iteration. Current TRE implementation does not do this.

Please, consider the test case:

#include <alloca.h>

int count;
__attribute__((noinline)) void globalIncrement(const int* param) { 
	count += *param; 
}

void test(int recurseCount)
{
    if (recurseCount == 0) return;
    {
    int *temp = (int*)alloca(100);
    globalIncrement(temp);
    }
    test(recurseCount - 1);
}

Following is the x86 asm generated for the above test case in assumption that dynamic allocas are possible:

.LBB1_2:
        movq    %rsp, %rdi
        addq    $-112, %rdi   <<<<<<<<<<<<<< dynamic stack reservation, need to be restored before "jne .LBB1_2"
        movq    %rdi, %rsp
        callq   _Z15globalIncrementPKi
        addl    $-1, %ebx
        jne     .LBB1_2

So, it looks like we still have inefficient code here and it was a reason for avoiding TRE.

332

It makes sense we can ignore lifetime.end on an alloca: we know the call doesn't refer to the alloca. (Maybe we should check that the pointer argument is pointing at an alloca? That should usually be true anyway, but better to be on the safe side, I guess.)

OK, I would add checking that the pointer argument of lifetime.end is pointing to an alloca.

I don't think it's safe to hoist assume without additional checks; I think we'd need to check that the call is marked "willreturn"?

Since this is sort of tricky, I'd prefer to split this off into a followup.

Ok, I would split Intrinsic::assume into another review.

810–811

OK.

[efriedma]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

95

I guess we can leave this for a later patch.

This isn't really any worse than the stack usage before TRE, assuming we can't emit a sibling call in the backend. And we could avoid this by making TRE insert stacksave/stackrestore intrinsics. But better to do one thing at a time.

[avl]

addressed comments.

[efriedma]

I think I'd like to see a testcase where there are multiple recursive calls, but only one is a tail call that can be eliminated.

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

474

The hasOperandBundles() check looks completely new; is there some test for it?

The isNoTailCall() check is currently redundant; it isn't legal to write "tail notail". I guess it makes sense to guard against that, though.

llvm/test/Transforms/TailCallElim/basic.ll

24

I'm not sure this is testing what it was originally supposed to. I guess that's okay, but please fix the comment at least.

llvm/test/Transforms/TailCallElim/tre-noncapturing-alloca-calls.ll

21

For the purpose of this testcase, we don't need the definition of _Z15globalIncrementPKi.

38

I think I'd prefer to just generate this with update_test_checks.py

[avl]

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

474

The hasOperandBundles() check looks completely new; is there some test for it?

it is not new. it is copied from 245 line. Now, when patch changed from its original state all above conditions could be changed just to :

if (!CI->isTailCall())

the test is Transforms/TailCallElim/deopt-bundle.ll

The isNoTailCall() check is currently redundant; it isn't legal to write "tail notail". I guess it makes sense to guard against that, though.

would add checking for that.

[avl]

addressed comments:

added test for multiple recursive calls,
removed duplicated check for operand bundles,
simplified and commented tests.

[efriedma]

LGTM

[avl]

Thank you, for the review.

[davezarzycki]

Hello. I have an auto-bisecting multi-stage bot that is failing on two after this change. Can we please revert this or commit a quick fix?

FAIL: Clang :: CXX/class/class.compare/class.spaceship/p1.cpp (6232 of 64222)
******************** TEST 'Clang :: CXX/class/class.compare/class.spaceship/p1.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';   /tmp/_update_lc/t/bin/clang -cc1 -internal-isystem /tmp/_update_lc/t/lib/clang/11.0.0/include -nostdsysteminc -std=c++2a -verify /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp -fcxx-exceptions
--
Exit Code: 134

Command Output (stderr):
--
clang: /home/dave/s/lp/clang/lib/Basic/SourceManager.cpp:917: clang::FileID clang::SourceManager::getFileIDLoaded(unsigned int) const: Assertion `0 && "Invalid SLocOffset or bad function choice"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.  Program arguments: /tmp/_update_lc/t/bin/clang -cc1 -internal-isystem /tmp/_update_lc/t/lib/clang/11.0.0/include -nostdsysteminc -std=c++2a -verify /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp -fcxx-exceptions
1.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp:127:38: current parser token ','
2.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp:39:1: parsing namespace 'Deletedness'
3.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp:123:12: parsing function body 'Deletedness::g'
4.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp:123:12: in compound statement ('{}')
 #0 0x000000000359273f llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/tmp/_update_lc/t/bin/clang+0x359273f)
 #1 0x0000000003590912 llvm::sys::RunSignalHandlers() (/tmp/_update_lc/t/bin/clang+0x3590912)
 #2 0x0000000003592bb5 SignalHandler(int) (/tmp/_update_lc/t/bin/clang+0x3592bb5)
 #3 0x00007ffff7fa6a90 __restore_rt (/lib64/libpthread.so.0+0x14a90)
 #4 0x00007ffff7b3da25 raise (/lib64/libc.so.6+0x3ca25)
 #5 0x00007ffff7b26895 abort (/lib64/libc.so.6+0x25895)
 #6 0x00007ffff7b26769 _nl_load_domain.cold (/lib64/libc.so.6+0x25769)
 #7 0x00007ffff7b35e86 (/lib64/libc.so.6+0x34e86)
 #8 0x000000000375636c clang::SourceManager::getFileIDLoaded(unsigned int) const (/tmp/_update_lc/t/bin/clang+0x375636c)
 #9 0x0000000003ee0bbb clang::VerifyDiagnosticConsumer::HandleDiagnostic(clang::DiagnosticsEngine::Level, clang::Diagnostic const&) (/tmp/_update_lc/t/bin/clang+0x3ee0bbb)
#10 0x00000000037501ab clang::DiagnosticIDs::ProcessDiag(clang::DiagnosticsEngine&) const (/tmp/_update_lc/t/bin/clang+0x37501ab)
#11 0x0000000003749fca clang::DiagnosticsEngine::EmitCurrentDiagnostic(bool) (/tmp/_update_lc/t/bin/clang+0x3749fca)
#12 0x0000000004df0c60 clang::Sema::EmitCurrentDiagnostic(unsigned int) (/tmp/_update_lc/t/bin/clang+0x4df0c60)
#13 0x0000000005092783 (anonymous namespace)::DefaultedComparisonAnalyzer::visitBinaryOperator(clang::OverloadedOperatorKind, llvm::ArrayRef<clang::Expr*>, (anonymous namespace)::DefaultedComparisonSubobject, clang::OverloadCandidateSet*) (/tmp/_update_lc/t/bin/clang+0x5092783)
#14 0x0000000005091dba (anonymous namespace)::DefaultedComparisonAnalyzer::visitExpandedSubobject(clang::QualType, (anonymous namespace)::DefaultedComparisonSubobject) (/tmp/_update_lc/t/bin/clang+0x5091dba)
#15 0x0000000005091b86 (anonymous namespace)::DefaultedComparisonVisitor<(anonymous namespace)::DefaultedComparisonAnalyzer, (anonymous namespace)::DefaultedComparisonInfo, (anonymous namespace)::DefaultedComparisonInfo, (anonymous namespace)::DefaultedComparisonSubobject>::visitSubobjects((anonymous namespace)::DefaultedComparisonInfo&, clang::CXXRecordDecl*, clang::Qualifiers) (/tmp/_update_lc/t/bin/clang+0x5091b86)
#16 0x0000000005058c8c (anonymous namespace)::DefaultedComparisonAnalyzer::visit() (/tmp/_update_lc/t/bin/clang+0x5058c8c)
#17 0x000000000505ab22 clang::Sema::DiagnoseDeletedDefaultedFunction(clang::FunctionDecl*) (/tmp/_update_lc/t/bin/clang+0x505ab22)
#18 0x00000000053e60ed clang::Sema::CreateOverloadedBinOp(clang::SourceLocation, clang::BinaryOperatorKind, clang::UnresolvedSetImpl const&, clang::Expr*, clang::Expr*, bool, bool, clang::FunctionDecl*) (/tmp/_update_lc/t/bin/clang+0x53e60ed)
#19 0x000000000514270a BuildOverloadedBinOp(clang::Sema&, clang::Scope*, clang::SourceLocation, clang::BinaryOperatorKind, clang::Expr*, clang::Expr*) (/tmp/_update_lc/t/bin/clang+0x514270a)
#20 0x00000000050fbf49 clang::Sema::ActOnBinOp(clang::Scope*, clang::SourceLocation, clang::tok::TokenKind, clang::Expr*, clang::Expr*) (/tmp/_update_lc/t/bin/clang+0x50fbf49)
#21 0x0000000004d52ccc clang::Parser::ParseRHSOfBinaryExpression(clang::ActionResult<clang::Expr*, true>, clang::prec::Level) (/tmp/_update_lc/t/bin/clang+0x4d52ccc)
#22 0x0000000004d51be9 clang::Parser::ParseAssignmentExpression(clang::Parser::TypeCastState) (/tmp/_update_lc/t/bin/clang+0x4d51be9)
#23 0x0000000004d60dba clang::Parser::ParseExpressionList(llvm::SmallVectorImpl<clang::Expr*>&, llvm::SmallVectorImpl<clang::SourceLocation>&, llvm::function_ref<void ()>) (/tmp/_update_lc/t/bin/clang+0x4d60dba)
#24 0x0000000004d542d9 clang::Parser::ParsePostfixExpressionSuffix(clang::ActionResult<clang::Expr*, true>) (/tmp/_update_lc/t/bin/clang+0x4d542d9)
#25 0x0000000004d55b95 clang::Parser::ParseCastExpression(clang::Parser::CastParseKind, bool, bool&, clang::Parser::TypeCastState, bool, bool*) (/tmp/_update_lc/t/bin/clang+0x4d55b95)
#26 0x0000000004d51b89 clang::Parser::ParseAssignmentExpression(clang::Parser::TypeCastState) (/tmp/_update_lc/t/bin/clang+0x4d51b89)
#27 0x0000000004d51ac9 clang::Parser::ParseExpression(clang::Parser::TypeCastState) (/tmp/_update_lc/t/bin/clang+0x4d51ac9)
#28 0x0000000004d78368 clang::Parser::ParseExprStatement(clang::Parser::ParsedStmtContext) (/tmp/_update_lc/t/bin/clang+0x4d78368)
#29 0x0000000004d76ba0 clang::Parser::ParseStatementOrDeclarationAfterAttributes(llvm::SmallVector<clang::Stmt*, 32u>&, clang::Parser::ParsedStmtContext, clang::SourceLocation*, clang::Parser::ParsedAttributesWithRange&) (/tmp/_update_lc/t/bin/clang+0x4d76ba0)
#30 0x0000000004d76614 clang::Parser::ParseStatementOrDeclaration(llvm::SmallVector<clang::Stmt*, 32u>&, clang::Parser::ParsedStmtContext, clang::SourceLocation*) (/tmp/_update_lc/t/bin/clang+0x4d76614)
#31 0x0000000004d7ecd2 clang::Parser::ParseCompoundStatementBody(bool) (/tmp/_update_lc/t/bin/clang+0x4d7ecd2)
#32 0x0000000004d7fcd0 clang::Parser::ParseFunctionStatementBody(clang::Decl*, clang::Parser::ParseScope&) (/tmp/_update_lc/t/bin/clang+0x4d7fcd0)
#33 0x0000000004cfacc0 clang::Parser::ParseFunctionDefinition(clang::ParsingDeclarator&, clang::Parser::ParsedTemplateInfo const&, clang::Parser::LateParsedAttrList*) (/tmp/_update_lc/t/bin/clang+0x4cfacc0)
#34 0x0000000004d28f2d clang::Parser::ParseDeclGroup(clang::ParsingDeclSpec&, clang::DeclaratorContext, clang::SourceLocation*, clang::Parser::ForRangeInit*) (/tmp/_update_lc/t/bin/clang+0x4d28f2d)
#35 0x0000000004cf9f32 clang::Parser::ParseDeclOrFunctionDefInternal(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec&, clang::AccessSpecifier) (/tmp/_update_lc/t/bin/clang+0x4cf9f32)
#36 0x0000000004cf9938 clang::Parser::ParseDeclarationOrFunctionDefinition(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec*, clang::AccessSpecifier) (/tmp/_update_lc/t/bin/clang+0x4cf9938)
#37 0x0000000004cf86fc clang::Parser::ParseExternalDeclaration(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec*) (/tmp/_update_lc/t/bin/clang+0x4cf86fc)
#38 0x0000000004d02c15 clang::Parser::ParseInnerNamespace(llvm::SmallVector<clang::Parser::InnerNamespaceInfo, 4u> const&, unsigned int, clang::SourceLocation&, clang::ParsedAttributes&, clang::BalancedDelimiterTracker&) (/tmp/_update_lc/t/bin/clang+0x4d02c15)
#39 0x0000000004d0251a clang::Parser::ParseNamespace(clang::DeclaratorContext, clang::SourceLocation&, clang::SourceLocation) (/tmp/_update_lc/t/bin/clang+0x4d0251a)
#40 0x0000000004d22f0a clang::Parser::ParseDeclaration(clang::DeclaratorContext, clang::SourceLocation&, clang::Parser::ParsedAttributesWithRange&, clang::SourceLocation*) (/tmp/_update_lc/t/bin/clang+0x4d22f0a)
#41 0x0000000004cf7e39 clang::Parser::ParseExternalDeclaration(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec*) (/tmp/_update_lc/t/bin/clang+0x4cf7e39)
#42 0x0000000004cf6858 clang::Parser::ParseTopLevelDecl(clang::OpaquePtr<clang::DeclGroupRef>&, bool) (/tmp/_update_lc/t/bin/clang+0x4cf6858)
#43 0x0000000004cf16ed clang::ParseAST(clang::Sema&, bool, bool) (/tmp/_update_lc/t/bin/clang+0x4cf16ed)
#44 0x0000000003e3eb21 clang::FrontendAction::Execute() (/tmp/_update_lc/t/bin/clang+0x3e3eb21)
#45 0x0000000003dba0e3 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/tmp/_update_lc/t/bin/clang+0x3dba0e3)
#46 0x0000000003ee796b clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/tmp/_update_lc/t/bin/clang+0x3ee796b)
#47 0x0000000002244636 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/tmp/_update_lc/t/bin/clang+0x2244636)
#48 0x000000000224297d ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) (/tmp/_update_lc/t/bin/clang+0x224297d)
#49 0x0000000002242619 main (/tmp/_update_lc/t/bin/clang+0x2242619)
#50 0x00007ffff7b28042 __libc_start_main (/lib64/libc.so.6+0x27042)
#51 0x000000000223f8ce _start (/tmp/_update_lc/t/bin/clang+0x223f8ce)
/tmp/_update_lc/t/tools/clang/test/CXX/class/class.compare/class.spaceship/Output/p1.cpp.script: line 1: 4146089 Aborted                 /tmp/_update_lc/t/bin/clang -cc1 -internal-isystem /tmp/_update_lc/t/lib/clang/11.0.0/include -nostdsysteminc -std=c++2a -verify /home/dave/s/lp/clang/test/CXX/class/class.compare/class.spaceship/p1.cpp -fcxx-exceptions

--

********************
Testing:  0..
FAIL: Clang :: CXX/class/class.compare/class.eq/p2.cpp (6242 of 64222)
******************** TEST 'Clang :: CXX/class/class.compare/class.eq/p2.cpp' FAILED ********************
Script:
--
: 'RUN: at line 1';   /tmp/_update_lc/t/bin/clang -cc1 -internal-isystem /tmp/_update_lc/t/lib/clang/11.0.0/include -nostdsysteminc -std=c++2a -verify /home/dave/s/lp/clang/test/CXX/class/class.compare/class.eq/p2.cpp
--
Exit Code: 134

Command Output (stderr):
--
clang: /home/dave/s/lp/clang/lib/Basic/SourceManager.cpp:917: clang::FileID clang::SourceManager::getFileIDLoaded(unsigned int) const: Assertion `0 && "Invalid SLocOffset or bad function choice"' failed.
PLEASE submit a bug report to https://bugs.llvm.org/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.  Program arguments: /tmp/_update_lc/t/bin/clang -cc1 -internal-isystem /tmp/_update_lc/t/lib/clang/11.0.0/include -nostdsysteminc -std=c++2a -verify /home/dave/s/lp/clang/test/CXX/class/class.compare/class.eq/p2.cpp
1.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.eq/p2.cpp:47:30: current parser token ')'
2.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.eq/p2.cpp:30:13: parsing function body 'test'
3.  /home/dave/s/lp/clang/test/CXX/class/class.compare/class.eq/p2.cpp:30:13: in compound statement ('{}')
 #0 0x000000000359273f llvm::sys::PrintStackTrace(llvm::raw_ostream&) (/tmp/_update_lc/t/bin/clang+0x359273f)
 #1 0x0000000003590912 llvm::sys::RunSignalHandlers() (/tmp/_update_lc/t/bin/clang+0x3590912)
 #2 0x0000000003592bb5 SignalHandler(int) (/tmp/_update_lc/t/bin/clang+0x3592bb5)
 #3 0x00007ffff7fa6a90 __restore_rt (/lib64/libpthread.so.0+0x14a90)
 #4 0x00007ffff7b3da25 raise (/lib64/libc.so.6+0x3ca25)
 #5 0x00007ffff7b26895 abort (/lib64/libc.so.6+0x25895)
 #6 0x00007ffff7b26769 _nl_load_domain.cold (/lib64/libc.so.6+0x25769)
 #7 0x00007ffff7b35e86 (/lib64/libc.so.6+0x34e86)
 #8 0x000000000375636c clang::SourceManager::getFileIDLoaded(unsigned int) const (/tmp/_update_lc/t/bin/clang+0x375636c)
 #9 0x0000000003ee0bbb clang::VerifyDiagnosticConsumer::HandleDiagnostic(clang::DiagnosticsEngine::Level, clang::Diagnostic const&) (/tmp/_update_lc/t/bin/clang+0x3ee0bbb)
#10 0x00000000037501ab clang::DiagnosticIDs::ProcessDiag(clang::DiagnosticsEngine&) const (/tmp/_update_lc/t/bin/clang+0x37501ab)
#11 0x0000000003749fca clang::DiagnosticsEngine::EmitCurrentDiagnostic(bool) (/tmp/_update_lc/t/bin/clang+0x3749fca)
#12 0x0000000004df0c60 clang::Sema::EmitCurrentDiagnostic(unsigned int) (/tmp/_update_lc/t/bin/clang+0x4df0c60)
#13 0x00000000050928b7 (anonymous namespace)::DefaultedComparisonAnalyzer::visitBinaryOperator(clang::OverloadedOperatorKind, llvm::ArrayRef<clang::Expr*>, (anonymous namespace)::DefaultedComparisonSubobject, clang::OverloadCandidateSet*) (/tmp/_update_lc/t/bin/clang+0x50928b7)
#14 0x0000000005091dba (anonymous namespace)::DefaultedComparisonAnalyzer::visitExpandedSubobject(clang::QualType, (anonymous namespace)::DefaultedComparisonSubobject) (/tmp/_update_lc/t/bin/clang+0x5091dba)
#15 0x0000000005091b86 (anonymous namespace)::DefaultedComparisonVisitor<(anonymous namespace)::DefaultedComparisonAnalyzer, (anonymous namespace)::DefaultedComparisonInfo, (anonymous namespace)::DefaultedComparisonInfo, (anonymous namespace)::DefaultedComparisonSubobject>::visitSubobjects((anonymous namespace)::DefaultedComparisonInfo&, clang::CXXRecordDecl*, clang::Qualifiers) (/tmp/_update_lc/t/bin/clang+0x5091b86)
#16 0x0000000005058c8c (anonymous namespace)::DefaultedComparisonAnalyzer::visit() (/tmp/_update_lc/t/bin/clang+0x5058c8c)
#17 0x000000000505ab22 clang::Sema::DiagnoseDeletedDefaultedFunction(clang::FunctionDecl*) (/tmp/_update_lc/t/bin/clang+0x505ab22)
#18 0x00000000053e60ed clang::Sema::CreateOverloadedBinOp(clang::SourceLocation, clang::BinaryOperatorKind, clang::UnresolvedSetImpl const&, clang::Expr*, clang::Expr*, bool, bool, clang::FunctionDecl*) (/tmp/_update_lc/t/bin/clang+0x53e60ed)
#19 0x000000000514270a BuildOverloadedBinOp(clang::Sema&, clang::Scope*, clang::SourceLocation, clang::BinaryOperatorKind, clang::Expr*, clang::Expr*) (/tmp/_update_lc/t/bin/clang+0x514270a)
#20 0x00000000050fbf49 clang::Sema::ActOnBinOp(clang::Scope*, clang::SourceLocation, clang::tok::TokenKind, clang::Expr*, clang::Expr*) (/tmp/_update_lc/t/bin/clang+0x50fbf49)
#21 0x0000000004d52ccc clang::Parser::ParseRHSOfBinaryExpression(clang::ActionResult<clang::Expr*, true>, clang::prec::Level) (/tmp/_update_lc/t/bin/clang+0x4d52ccc)
#22 0x0000000004d51be9 clang::Parser::ParseAssignmentExpression(clang::Parser::TypeCastState) (/tmp/_update_lc/t/bin/clang+0x4d51be9)
#23 0x0000000004d60dba clang::Parser::ParseExpressionList(llvm::SmallVectorImpl<clang::Expr*>&, llvm::SmallVectorImpl<clang::SourceLocation>&, llvm::function_ref<void ()>) (/tmp/_update_lc/t/bin/clang+0x4d60dba)
#24 0x0000000004d4b29c clang::Parser::ParseCXXTypeConstructExpression(clang::DeclSpec const&) (/tmp/_update_lc/t/bin/clang+0x4d4b29c)
#25 0x0000000004d57617 clang::Parser::ParseCastExpression(clang::Parser::CastParseKind, bool, bool&, clang::Parser::TypeCastState, bool, bool*) (/tmp/_update_lc/t/bin/clang+0x4d57617)
#26 0x0000000004d51b89 clang::Parser::ParseAssignmentExpression(clang::Parser::TypeCastState) (/tmp/_update_lc/t/bin/clang+0x4d51b89)
#27 0x0000000004d51ac9 clang::Parser::ParseExpression(clang::Parser::TypeCastState) (/tmp/_update_lc/t/bin/clang+0x4d51ac9)
#28 0x0000000004d78368 clang::Parser::ParseExprStatement(clang::Parser::ParsedStmtContext) (/tmp/_update_lc/t/bin/clang+0x4d78368)
#29 0x0000000004d76ba0 clang::Parser::ParseStatementOrDeclarationAfterAttributes(llvm::SmallVector<clang::Stmt*, 32u>&, clang::Parser::ParsedStmtContext, clang::SourceLocation*, clang::Parser::ParsedAttributesWithRange&) (/tmp/_update_lc/t/bin/clang+0x4d76ba0)
#30 0x0000000004d76614 clang::Parser::ParseStatementOrDeclaration(llvm::SmallVector<clang::Stmt*, 32u>&, clang::Parser::ParsedStmtContext, clang::SourceLocation*) (/tmp/_update_lc/t/bin/clang+0x4d76614)
#31 0x0000000004d7ecd2 clang::Parser::ParseCompoundStatementBody(bool) (/tmp/_update_lc/t/bin/clang+0x4d7ecd2)
#32 0x0000000004d7fcd0 clang::Parser::ParseFunctionStatementBody(clang::Decl*, clang::Parser::ParseScope&) (/tmp/_update_lc/t/bin/clang+0x4d7fcd0)
#33 0x0000000004cfacc0 clang::Parser::ParseFunctionDefinition(clang::ParsingDeclarator&, clang::Parser::ParsedTemplateInfo const&, clang::Parser::LateParsedAttrList*) (/tmp/_update_lc/t/bin/clang+0x4cfacc0)
#34 0x0000000004d28f2d clang::Parser::ParseDeclGroup(clang::ParsingDeclSpec&, clang::DeclaratorContext, clang::SourceLocation*, clang::Parser::ForRangeInit*) (/tmp/_update_lc/t/bin/clang+0x4d28f2d)
#35 0x0000000004cf9f32 clang::Parser::ParseDeclOrFunctionDefInternal(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec&, clang::AccessSpecifier) (/tmp/_update_lc/t/bin/clang+0x4cf9f32)
#36 0x0000000004cf9938 clang::Parser::ParseDeclarationOrFunctionDefinition(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec*, clang::AccessSpecifier) (/tmp/_update_lc/t/bin/clang+0x4cf9938)
#37 0x0000000004cf86fc clang::Parser::ParseExternalDeclaration(clang::Parser::ParsedAttributesWithRange&, clang::ParsingDeclSpec*) (/tmp/_update_lc/t/bin/clang+0x4cf86fc)
#38 0x0000000004cf6858 clang::Parser::ParseTopLevelDecl(clang::OpaquePtr<clang::DeclGroupRef>&, bool) (/tmp/_update_lc/t/bin/clang+0x4cf6858)
#39 0x0000000004cf16ed clang::ParseAST(clang::Sema&, bool, bool) (/tmp/_update_lc/t/bin/clang+0x4cf16ed)
#40 0x0000000003e3eb21 clang::FrontendAction::Execute() (/tmp/_update_lc/t/bin/clang+0x3e3eb21)
#41 0x0000000003dba0e3 clang::CompilerInstance::ExecuteAction(clang::FrontendAction&) (/tmp/_update_lc/t/bin/clang+0x3dba0e3)
#42 0x0000000003ee796b clang::ExecuteCompilerInvocation(clang::CompilerInstance*) (/tmp/_update_lc/t/bin/clang+0x3ee796b)
#43 0x0000000002244636 cc1_main(llvm::ArrayRef<char const*>, char const*, void*) (/tmp/_update_lc/t/bin/clang+0x2244636)
#44 0x000000000224297d ExecuteCC1Tool(llvm::SmallVectorImpl<char const*>&) (/tmp/_update_lc/t/bin/clang+0x224297d)
#45 0x0000000002242619 main (/tmp/_update_lc/t/bin/clang+0x2242619)
#46 0x00007ffff7b28042 __libc_start_main (/lib64/libc.so.6+0x27042)
#47 0x000000000223f8ce _start (/tmp/_update_lc/t/bin/clang+0x223f8ce)
/tmp/_update_lc/t/tools/clang/test/CXX/class/class.compare/class.eq/Output/p2.cpp.script: line 1: 4146047 Aborted                 /tmp/_update_lc/t/bin/clang -cc1 -internal-isystem /tmp/_update_lc/t/lib/clang/11.0.0/include -nostdsysteminc -std=c++2a -verify /home/dave/s/lp/clang/test/CXX/class/class.compare/class.eq/p2.cpp

--

********************
Testing:  0.. 10.. 20.. 30.. 40.. 50.. 60.. 70.. 80.. 90..
********************
Failed Tests (2):
  Clang :: CXX/class/class.compare/class.eq/p2.cpp
  Clang :: CXX/class/class.compare/class.spaceship/p1.cpp


Testing Time: 117.51s
  Unsupported      : 12906
  Passed           : 51214
  Expectedly Failed:   100
  Failed           :     2
FAILED: CMakeFiles/check-all
cd /tmp/_update_lc/t && /usr/bin/python3.8 /tmp/_update_lc/t/./bin/llvm-lit -sv --param USE_Z3_SOLVER=0 /tmp/_update_lc/t/tools/clang/test /tmp/_update_lc/t/tools/lld/test /tmp/_update_lc/t/tools/lldb/test /tmp/_update_lc/t/utils/lit /tmp/_update_lc/t/test
ninja: build stopped: subcommand failed.
+ do_error 'FAILURE -- STAGE TWO BUILD of LLVM' 12
+ echo FAILURE -- STAGE TWO BUILD of LLVM
FAILURE -- STAGE TWO BUILD of LLVM
+ exit 12

llvm/lib/Transforms/Scalar/TailRecursionElimination.cpp

 #include "llvm/IR/Module.h"
 #include "llvm/IR/ValueHandle.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 
 #define DEBUG_TYPE "tailcallelim"
 
 STATISTIC(NumEliminated, "Number of tail calls removed");
 STATISTIC(NumRetDuped,   "Number of return duplicated");
 STATISTIC(NumAccumAdded, "Number of accumulators introduced");
 
 /// Scan the specified function for alloca instructions.
 /// If it contains any dynamic allocas, returns false.
 static bool canTRE(Function &F) {
-  // Because of PR962, we don't TRE dynamic allocas.
+  // TODO: We don't do TRE if dynamic allocas are used.
+  // Dynamic allocas allocate stack space which should be
+  // deallocated before new iteration started. That is
+  // currently not implemented.
   return llvm::all_of(instructions(F), [](Instruction &I) {
     auto *AI = dyn_cast<AllocaInst>(&I);
     return !AI || AI->isStaticAlloca();
   });
 }
 
 namespace {
 struct AllocaDerivedValueTracker {
while (!Worklist.empty()) {
         if (CB.isArgOperand(U) && CB.isByValArgument(CB.getArgOperandNo(U)))
           continue;
         bool IsNocapture =
             CB.isDataOperand(U) && CB.doesNotCapture(CB.getDataOperandNo(U));
         callUsesLocalStack(CB, IsNocapture);
         if (IsNocapture) {
           // If the alloca-derived argument is passed in as nocapture, then it
           // can't propagate to the call's return. That would be capturing.
           continue;
         }
         break;
       }
       case Instruction::Load: {
         // The result of a load is not alloca-derived (unless an alloca has
         // otherwise escaped, but this is a local analysis).
         continue;
       }
if (!CB.onlyReadsMemory())
       EscapePoints.insert(&CB);
   }
 
   SmallPtrSet<Instruction *, 32> AllocaUsers;
   SmallPtrSet<Instruction *, 32> EscapePoints;
 };
 }
 
-static bool markTails(Function &F, bool &AllCallsAreTailCalls,
-                      OptimizationRemarkEmitter *ORE) {
+static bool markTails(Function &F, OptimizationRemarkEmitter *ORE) {
   if (F.callsFunctionThatReturnsTwice())
     return false;
-  AllCallsAreTailCalls = true;
 
   // The local stack holds all alloca instructions and all byval arguments.
   AllocaDerivedValueTracker Tracker;
   for (Argument &Arg : F.args()) {
     if (Arg.hasByValAttr())
       Tracker.walk(&Arg);
   }
   for (auto &BB : F) {
for (auto &I : *BB) {
                    << "marked as tail call candidate (readnone)";
           });
           CI->setTailCall();
           Modified = true;
           continue;
         }
       }
 
-      if (!IsNoTail && Escaped == UNESCAPED && !Tracker.AllocaUsers.count(CI)) {
+      if (!IsNoTail && Escaped == UNESCAPED && !Tracker.AllocaUsers.count(CI))
         DeferredTails.push_back(CI);
-      } else {
-        AllCallsAreTailCalls = false;
-      }
     }
 
     for (auto *SuccBB : make_range(succ_begin(BB), succ_end(BB))) {
       auto &State = Visited[SuccBB];
       if (State < Escaped) {
         State = Escaped;
         if (State == ESCAPED)
           WorklistEscaped.push_back(SuccBB);
static bool markTails(Function &F, OptimizationRemarkEmitter *ORE) {
 
   for (CallInst *CI : DeferredTails) {
     if (Visited[CI->getParent()] != ESCAPED) {
       // If the escape point was part way through the block, calls after the
       // escape point wouldn't have been put into DeferredTails.
       LLVM_DEBUG(dbgs() << "Marked as tail call candidate: " << *CI << "\n");
       CI->setTailCall();
       Modified = true;
-    } else {
-      AllCallsAreTailCalls = false;
     }
   }
 
   return Modified;
 }
 
 /// Return true if it is safe to move the specified
 /// instruction from after the call to before the call, assuming that all
 /// instructions between the call and this instruction are movable.
 ///
-static bool canMoveAboveCall(Instruction *I, CallInst *CI, AliasAnalysis *AA) {
+static bool canMoveAboveCall(Instruction *I, CallInst *CI, AliasAnalysis *AA,
+                             DenseMap<Value *, AllocaInst *> &AllocaForValue) {
+  if (isa<DbgInfoIntrinsic>(I))
+    return true;
+
+  if (const IntrinsicInst *II = dyn_cast<IntrinsicInst>(I))
+    if (II->getIntrinsicID() == Intrinsic::lifetime_end &&
+        llvm::findAllocaForValue(II->getArgOperand(1), AllocaForValue))
+      return true;
+
   // FIXME: We can move load/store/call/free instructions above the call if the
   // call does not mod/ref the memory location being processed.
   if (I->mayHaveSideEffects())  // This also handles volatile loads.
     return false;
 
   if (LoadInst *L = dyn_cast<LoadInst>(I)) {
     // Loads may always be moved above calls without side effects.
     if (CI->mayHaveSideEffects()) {
class TailRecursionEliminator {
   OptimizationRemarkEmitter *ORE;
   DomTreeUpdater &DTU;
 
   // The below are shared state we want to have available when eliminating any
   // calls in the function. There values should be populated by
   // createTailRecurseLoopHeader the first time we find a call we can eliminate.
   BasicBlock *HeaderBB = nullptr;
   SmallVector<PHINode *, 8> ArgumentPHIs;
-  bool RemovableCallsMustBeMarkedTail = false;
 
   // PHI node to store our return value.
   PHINode *RetPN = nullptr;
 
   // i1 PHI node to track if we have a valid return value stored in RetPN.
   PHINode *RetKnownPN = nullptr;
 
   // Vector of select instructions we insereted. These selects use RetKnownPN
   // to either propagate RetPN or select a new return value.
   SmallVector<SelectInst *, 8> RetSelects;
 
   // The below are shared state needed when performing accumulator recursion.
   // There values should be populated by insertAccumulator the first time we
   // find an elimination that requires an accumulator.
 
   // PHI node to store our current accumulated value.
   PHINode *AccPN = nullptr;
 
   // The instruction doing the accumulating.
   Instruction *AccumulatorRecursionInstr = nullptr;
 
+  // The cache for <value, alloca instruction> pairs.
+  DenseMap<Value *, AllocaInst *> AllocaForValue;
+
   TailRecursionEliminator(Function &F, const TargetTransformInfo *TTI,
                           AliasAnalysis *AA, OptimizationRemarkEmitter *ORE,
                           DomTreeUpdater &DTU)
       : F(F), TTI(TTI), AA(AA), ORE(ORE), DTU(DTU) {}
 
-  CallInst *findTRECandidate(Instruction *TI,
-                             bool CannotTailCallElimCallsMarkedTail);
+  CallInst *findTRECandidate(Instruction *TI);
 
   void createTailRecurseLoopHeader(CallInst *CI);
 
   void insertAccumulator(Instruction *AccRecInstr);
 
   bool eliminateCall(CallInst *CI);
 
-  bool foldReturnAndProcessPred(ReturnInst *Ret,
-                                bool CannotTailCallElimCallsMarkedTail);
+  bool foldReturnAndProcessPred(ReturnInst *Ret);
 
-  bool processReturningBlock(ReturnInst *Ret,
-                             bool CannotTailCallElimCallsMarkedTail);
+  bool processReturningBlock(ReturnInst *Ret);
 
   void cleanupAndFinalize();
 
 public:
   static bool eliminate(Function &F, const TargetTransformInfo *TTI,
                         AliasAnalysis *AA, OptimizationRemarkEmitter *ORE,
                         DomTreeUpdater &DTU);
 };
 } // namespace
 
-CallInst *TailRecursionEliminator::findTRECandidate(
-    Instruction *TI, bool CannotTailCallElimCallsMarkedTail) {
+CallInst *TailRecursionEliminator::findTRECandidate(Instruction *TI) {
   BasicBlock *BB = TI->getParent();
 
   if (&BB->front() == TI) // Make sure there is something before the terminator.
     return nullptr;
 
   // Scan backwards from the return, checking to see if there is a tail call in
   // this block.  If so, set CI to it.
   CallInst *CI = nullptr;
   BasicBlock::iterator BBI(TI);
   while (true) {
     CI = dyn_cast<CallInst>(BBI);
     if (CI && CI->getCalledFunction() == &F)
       break;
 
     if (BBI == BB->begin())
       return nullptr;          // Didn't find a potential tail call.
     --BBI;
   }
 
-  // If this call is marked as a tail call, and if there are dynamic allocas in
-  // the function, we cannot perform this optimization.
-  if (CI->isTailCall() && CannotTailCallElimCallsMarkedTail)
+  assert((!CI->isTailCall() || !CI->isNoTailCall()) &&
+         "Incompatible call site attributes(Tail,NoTail)");
+  if (!CI->isTailCall())
     return nullptr;
 
   // As a special case, detect code like this:
   //   double fabs(double f) { return __builtin_fabs(f); } // a 'fabs' call
   // and disable this xform in this case, because the code generator will
   // lower the call to fabs into inline code.
   if (BB == &F.getEntryBlock() &&
       firstNonDbg(BB->front().getIterator()) == CI &&
       firstNonDbg(std::next(BB->begin())) == TI && CI->getCalledFunction() &&

 void TailRecursionEliminator::createTailRecurseLoopHeader(CallInst *CI) {
   HeaderBB = &F.getEntryBlock();
   BasicBlock *NewEntry = BasicBlock::Create(F.getContext(), "", &F, HeaderBB);
   NewEntry->takeName(HeaderBB);
   HeaderBB->setName("tailrecurse");
   BranchInst *BI = BranchInst::Create(HeaderBB, NewEntry);
   BI->setDebugLoc(CI->getDebugLoc());
 
-  // If this function has self recursive calls in the tail position where some
-  // are marked tail and some are not, only transform one flavor or another.
-  // We have to choose whether we move allocas in the entry block to the new
-  // entry block or not, so we can't make a good choice for both. We make this
-  // decision here based on whether the first call we found to remove is
-  // marked tail.
-  // NOTE: We could do slightly better here in the case that the function has
-  // no entry block allocas.
-  RemovableCallsMustBeMarkedTail = CI->isTailCall();
-
-  // If this tail call is marked 'tail' and if there are any allocas in the
-  // entry block, move them up to the new entry block.
-  if (RemovableCallsMustBeMarkedTail)
-    // Move all fixed sized allocas from HeaderBB to NewEntry.
-    for (BasicBlock::iterator OEBI = HeaderBB->begin(), E = HeaderBB->end(),
-                              NEBI = NewEntry->begin();
-         OEBI != E;)
-      if (AllocaInst *AI = dyn_cast<AllocaInst>(OEBI++))
-        if (isa<ConstantInt>(AI->getArraySize()))
-          AI->moveBefore(&*NEBI);
 
   // Now that we have created a new block, which jumps to the entry
   // block, insert a PHI node for each argument of the function.
   // For now, we initialize each PHI to only have the real arguments
   // which are passed in.
   Instruction *InsertPos = &HeaderBB->front();
   for (Function::arg_iterator I = F.arg_begin(), E = F.arg_end(); I != E; ++I) {
     PHINode *PN =
bool TailRecursionEliminator::eliminateCall(CallInst *CI) {
 
   // Ok, we found a potential tail call.  We can currently only transform the
   // tail call if all of the instructions between the call and the return are
   // movable to above the call itself, leaving the call next to the return.
   // Check that this is the case now.
   Instruction *AccRecInstr = nullptr;
   BasicBlock::iterator BBI(CI);
   for (++BBI; &*BBI != Ret; ++BBI) {
-    if (canMoveAboveCall(&*BBI, CI, AA))
+    if (canMoveAboveCall(&*BBI, CI, AA, AllocaForValue))
       continue;
 
     // If we can't move the instruction above the call, it might be because it
     // is an associative and commutative operation that could be transformed
     // using accumulator recursion elimination.  Check to see if this is the
     // case, and if so, remember which instruction accumulates for later.
     if (AccPN || !canTransformAccumulatorRecursion(&*BBI, CI))
       return false; // We cannot eliminate the tail recursion!
return OptimizationRemark(DEBUG_TYPE, "tailcall-recursion", CI)
            << "transforming tail recursion into loop";
   });
 
   // OK! We can transform this tail call.  If this is the first one found,
   // create the new entry block, allowing us to branch back to the old entry.
   if (!HeaderBB)
     createTailRecurseLoopHeader(CI);
 
-  if (RemovableCallsMustBeMarkedTail && !CI->isTailCall())
-    return false;
-
   // Ok, now that we know we have a pseudo-entry block WITH all of the
   // required PHI nodes, add entries into the PHI node for the actual
   // parameters passed into the tail-recursive call.
   for (unsigned i = 0, e = CI->getNumArgOperands(); i != e; ++i)
     ArgumentPHIs[i]->addIncoming(CI->getArgOperand(i), BB);
 
   if (AccRecInstr) {
     insertAccumulator(AccRecInstr);
bool TailRecursionEliminator::eliminateCall(CallInst *CI) {
 
   BB->getInstList().erase(Ret);  // Remove return.
   BB->getInstList().erase(CI);   // Remove call.
   DTU.applyUpdates({{DominatorTree::Insert, BB, HeaderBB}});
   ++NumEliminated;
   return true;
 }
 
-bool TailRecursionEliminator::foldReturnAndProcessPred(
-    ReturnInst *Ret, bool CannotTailCallElimCallsMarkedTail) {
+bool TailRecursionEliminator::foldReturnAndProcessPred(ReturnInst *Ret) {
   BasicBlock *BB = Ret->getParent();
 
   bool Change = false;
 
   // Make sure this block is a trivial return block.
   assert(BB->getFirstNonPHIOrDbg() == Ret &&
          "Trying to fold non-trivial return block");
 
   // If the return block contains nothing but the return and PHI's,
   // there might be an opportunity to duplicate the return in its
   // predecessors and perform TRE there. Look for predecessors that end
   // in unconditional branch and recursive call(s).
   SmallVector<BranchInst*, 8> UncondBranchPreds;
   for (pred_iterator PI = pred_begin(BB), E = pred_end(BB); PI != E; ++PI) {
     BasicBlock *Pred = *PI;
     Instruction *PTI = Pred->getTerminator();
     if (BranchInst *BI = dyn_cast<BranchInst>(PTI))
       if (BI->isUnconditional())
         UncondBranchPreds.push_back(BI);
   }
 
   while (!UncondBranchPreds.empty()) {
     BranchInst *BI = UncondBranchPreds.pop_back_val();
     BasicBlock *Pred = BI->getParent();
-    if (CallInst *CI =
-            findTRECandidate(BI, CannotTailCallElimCallsMarkedTail)) {
+    if (CallInst *CI = findTRECandidate(BI)) {
       LLVM_DEBUG(dbgs() << "FOLDING: " << *BB
                         << "INTO UNCOND BRANCH PRED: " << *Pred);
       FoldReturnIntoUncondBranch(Ret, BB, Pred, &DTU);
 
       // Cleanup: if all predecessors of BB have been eliminated by
       // FoldReturnIntoUncondBranch, delete it.  It is important to empty it,
       // because the ret instruction in there is still using a value which
       // eliminateRecursiveTailCall will attempt to remove.
       if (!BB->hasAddressTaken() && pred_begin(BB) == pred_end(BB))
         DTU.deleteBB(BB);
 
       eliminateCall(CI);
       ++NumRetDuped;
       Change = true;
     }
   }
 
   return Change;
 }
 
-bool TailRecursionEliminator::processReturningBlock(
-    ReturnInst *Ret, bool CannotTailCallElimCallsMarkedTail) {
-  CallInst *CI = findTRECandidate(Ret, CannotTailCallElimCallsMarkedTail);
+bool TailRecursionEliminator::processReturningBlock(ReturnInst *Ret) {
+  CallInst *CI = findTRECandidate(Ret);
   if (!CI)
     return false;
 
   return eliminateCall(CI);
 }
 
 void TailRecursionEliminator::cleanupAndFinalize() {
   // If we eliminated any tail recursions, it's possible that we inserted some
bool TailRecursionEliminator::eliminate(Function &F,
                                         const TargetTransformInfo *TTI,
                                         AliasAnalysis *AA,
                                         OptimizationRemarkEmitter *ORE,
                                         DomTreeUpdater &DTU) {
   if (F.getFnAttribute("disable-tail-calls").getValueAsString() == "true")
     return false;
 
   bool MadeChange = false;
-  bool AllCallsAreTailCalls = false;
-  MadeChange |= markTails(F, AllCallsAreTailCalls, ORE);
-  if (!AllCallsAreTailCalls)
-    return MadeChange;
+  MadeChange |= markTails(F, ORE);
 
   // If this function is a varargs function, we won't be able to PHI the args
   // right, so don't even try to convert it...
   if (F.getFunctionType()->isVarArg())
     return MadeChange;
 
-  // If false, we cannot perform TRE on tail calls marked with the 'tail'
-  // attribute, because doing so would cause the stack size to increase (real
-  // TRE would deallocate variable sized allocas, TRE doesn't).
-  bool CanTRETailMarkedCall = canTRE(F);
+  if (!canTRE(F))
+    return MadeChange;
 
   TailRecursionEliminator TRE(F, TTI, AA, ORE, DTU);
 
   // Change any tail recursive calls to loops.
-  //
-  // FIXME: The code generator produces really bad code when an 'escaping
-  // alloca' is changed from being a static alloca to being a dynamic alloca.
-  // Until this is resolved, disable this transformation if that would ever
-  // happen.  This bug is PR962.
   for (Function::iterator BBI = F.begin(), E = F.end(); BBI != E; /*in loop*/) {
     BasicBlock *BB = &*BBI++; // foldReturnAndProcessPred may delete BB.
     if (ReturnInst *Ret = dyn_cast<ReturnInst>(BB->getTerminator())) {
-      bool Change = TRE.processReturningBlock(Ret, !CanTRETailMarkedCall);
+      bool Change = TRE.processReturningBlock(Ret);
       if (!Change && BB->getFirstNonPHIOrDbg() == Ret)
-        Change = TRE.foldReturnAndProcessPred(Ret, !CanTRETailMarkedCall);
+        Change = TRE.foldReturnAndProcessPred(Ret);
       MadeChange |= Change;
     }
   }
 
   TRE.cleanupAndFinalize();
 
   return MadeChange;
 }
 
 namespace {
 struct TailCallElim : public FunctionPass {
   static char ID; // Pass identification, replacement for typeid
   TailCallElim() : FunctionPass(ID) {
     initializeTailCallElimPass(*PassRegistry::getPassRegistry());
   }
 
   void getAnalysisUsage(AnalysisUsage &AU) const override {
     AU.addRequired<TargetTransformInfoWrapperPass>();
     AU.addRequired<AAResultsWrapperPass>();
     AU.addRequired<OptimizationRemarkEmitterWrapperPass>();
     AU.addPreserved<GlobalsAAWrapperPass>();
     AU.addPreserved<DominatorTreeWrapperPass>();
     AU.addPreserved<PostDominatorTreeWrapperPass>();
   }
 
   bool runOnFunction(Function &F) override {
     if (skipFunction(F))
       return false;
 
     auto *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>();
     auto *DT = DTWP ? &DTWP->getDomTree() : nullptr;
     auto *PDTWP = getAnalysisIfAvailable<PostDominatorTreeWrapperPass>();
     auto *PDT = PDTWP ? &PDTWP->getPostDomTree() : nullptr;
     // There is no noticable performance difference here between Lazy and Eager
     // UpdateStrategy based on some test results. It is feasible to switch the
     // UpdateStrategy to Lazy if we find it profitable later.
     DomTreeUpdater DTU(DT, PDT, DomTreeUpdater::UpdateStrategy::Eager);
 
     return TailRecursionEliminator::eliminate(
         F, &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F),
         &getAnalysis<AAResultsWrapperPass>().getAAResults(),
         &getAnalysis<OptimizationRemarkEmitterWrapperPass>().getORE(), DTU);
   }
 };
 }
 
 char TailCallElim::ID = 0;
 INITIALIZE_PASS_BEGIN(TailCallElim, "tailcallelim", "Tail Call Elimination",

llvm/test/Transforms/TailCallElim/basic.ll

 ; RUN: opt < %s -tailcallelim -verify-dom-info -S | FileCheck %s
 
 declare void @noarg()
 declare void @use(i32*)
 declare void @use_nocapture(i32* nocapture)
 declare void @use2_nocapture(i32* nocapture, i32* nocapture)
 
 ; Trivial case. Mark @noarg with tail call.
 define void @test0() {
 ; CHECK: tail call void @noarg()
   call void @noarg()
   ret void
 }
 
-; PR615. Make sure that we do not move the alloca so that it interferes with the tail call.
+; Make sure that we do not do TRE if pointer to local stack
+; escapes through function call.
 define i32 @test1() {
 ; CHECK: i32 @test1()
 ; CHECK-NEXT: alloca
   %A = alloca i32   ; <i32*> [#uses=2]
   store i32 5, i32* %A
   call void @use(i32* %A)
-; CHECK: tail call i32 @test1
-  %X = tail call i32 @test1()    ; <i32> [#uses=1]
+; CHECK: call i32 @test1
+  %X = call i32 @test1()    ; <i32> [#uses=1]
   ret i32 %X
 }
 
 ; This function contains intervening instructions which should be moved out of the way
 define i32 @test2(i32 %X) {
 ; CHECK: i32 @test2
 ; CHECK-NOT: call
 ; CHECK: ret i32

llvm/test/Transforms/TailCallElim/tre-multiple-exits.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -tailcallelim -verify-dom-info -S | FileCheck %s
+
+; This test checks that TRE would be done for only one recursive call.
+; The test_multiple_exits function has three recursive calls.
+; First recursive call could not be eliminated because there is
+; escaped pointer to local variable. Second recursive call could
+; be eliminated. Thrid recursive call could not be eliminated since
+; this is not last call. Thus, test checks that TRE would be done
+; for only second recursive call.
+
+; IR for that test was generated from the following C++ source:
+;
+; void capture_arg (int*);
+; void test_multiple_exits (int param);
+;   if (param >= 0 && param < 10) {
+;     int temp;
+;     capture_arg(&temp);
+;     // TRE could not be done because pointer to local
+;     // variable "temp" is escaped.
+;     test_multiple_exits(param + 1);
+;   } else if (param >=10 && param < 20) {
+;     // TRE should be done.
+;     test_multiple_exits(param + 1);
+;   } else if (param >= 20 && param < 22) {
+;     // TRE could not be done since recursive
+;     // call is not last call.
+;     test_multiple_exits(param + 1);
+;     func();
+;   }
+;
+;   return;
+; }
+
+; Function Attrs: noinline optnone uwtable
+declare void @_Z11capture_argPi(i32* %param) #0
+
+; Function Attrs: noinline optnone uwtable
+declare void @_Z4funcv() #0
+
+; Function Attrs: noinline nounwind uwtable
+define dso_local void @_Z19test_multiple_exitsi(i32 %param) local_unnamed_addr #2 {
+; CHECK-LABEL: @_Z19test_multiple_exitsi(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TEMP:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    br label [[TAILRECURSE:%.*]]
+; CHECK:       tailrecurse:
+; CHECK-NEXT:    [[PARAM_TR:%.*]] = phi i32 [ [[PARAM:%.*]], [[ENTRY:%.*]] ], [ [[ADD6:%.*]], [[IF_THEN5:%.*]] ]
+; CHECK-NEXT:    [[TMP0:%.*]] = icmp ult i32 [[PARAM_TR]], 10
+; CHECK-NEXT:    br i1 [[TMP0]], label [[IF_THEN:%.*]], label [[IF_ELSE:%.*]]
+; CHECK:       if.then:
+; CHECK-NEXT:    [[TMP1:%.*]] = bitcast i32* [[TEMP]] to i8*
+; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull [[TMP1]]) #1
+; CHECK-NEXT:    call void @_Z11capture_argPi(i32* nonnull [[TEMP]])
+; CHECK-NEXT:    [[ADD:%.*]] = add nuw nsw i32 [[PARAM_TR]], 1
+; CHECK-NEXT:    call void @_Z19test_multiple_exitsi(i32 [[ADD]])
+; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull [[TMP1]]) #1
+; CHECK-NEXT:    br label [[IF_END14:%.*]]
+; CHECK:       if.else:
+; CHECK-NEXT:    [[PARAM_OFF:%.*]] = add i32 [[PARAM_TR]], -10
+; CHECK-NEXT:    [[TMP2:%.*]] = icmp ult i32 [[PARAM_OFF]], 10
+; CHECK-NEXT:    br i1 [[TMP2]], label [[IF_THEN5]], label [[IF_ELSE7:%.*]]
+; CHECK:       if.then5:
+; CHECK-NEXT:    [[ADD6]] = add nuw nsw i32 [[PARAM_TR]], 1
+; CHECK-NEXT:    br label [[TAILRECURSE]]
+; CHECK:       if.else7:
+; CHECK-NEXT:    [[TMP3:%.*]] = and i32 [[PARAM_TR]], -2
+; CHECK-NEXT:    [[TMP4:%.*]] = icmp eq i32 [[TMP3]], 20
+; CHECK-NEXT:    br i1 [[TMP4]], label [[IF_THEN11:%.*]], label [[IF_END14]]
+; CHECK:       if.then11:
+; CHECK-NEXT:    [[ADD12:%.*]] = add nsw i32 [[PARAM_TR]], 1
+; CHECK-NEXT:    tail call void @_Z19test_multiple_exitsi(i32 [[ADD12]])
+; CHECK-NEXT:    tail call void @_Z4funcv()
+; CHECK-NEXT:    ret void
+; CHECK:       if.end14:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %temp = alloca i32, align 4
+  %0 = icmp ult i32 %param, 10
+  br i1 %0, label %if.then, label %if.else
+
+if.then:                                          ; preds = %entry
+  %1 = bitcast i32* %temp to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %1) #2
+  call void @_Z11capture_argPi(i32* nonnull %temp)
+  %add = add nuw nsw i32 %param, 1
+  call void @_Z19test_multiple_exitsi(i32 %add)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %1) #2
+  br label %if.end14
+
+if.else:                                          ; preds = %entry
+  %param.off = add i32 %param, -10
+  %2 = icmp ult i32 %param.off, 10
+  br i1 %2, label %if.then5, label %if.else7
+
+if.then5:                                         ; preds = %if.else
+  %add6 = add nuw nsw i32 %param, 1
+  call void @_Z19test_multiple_exitsi(i32 %add6)
+  br label %if.end14
+
+if.else7:                                         ; preds = %if.else
+  %3 = and i32 %param, -2
+  %4 = icmp eq i32 %3, 20
+  br i1 %4, label %if.then11, label %if.end14
+
+if.then11:                                        ; preds = %if.else7
+  %add12 = add nsw i32 %param, 1
+  call void @_Z19test_multiple_exitsi(i32 %add12)
+  call void @_Z4funcv()
+  br label %if.end14
+
+if.end14:                                         ; preds = %if.then5, %if.then11, %if.else7, %if.then
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #2
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #2
+
+attributes #0 = { nofree noinline norecurse nounwind uwtable }
+attributes #1 = { nounwind uwtable }
+attributes #2 = { argmemonly nounwind willreturn }

llvm/test/Transforms/TailCallElim/tre-noncapturing-alloca-calls.ll

+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -tailcallelim -verify-dom-info -S | FileCheck %s
+
+; IR for that test was generated from the following C++ source:
+;
+;int count;
+;__attribute__((noinline)) void globalIncrement(const int* param) { count += *param; }
+;
+;void test(int recurseCount)
+;{
+;    if (recurseCount == 0) return;
+;    int temp = 10;
+;    globalIncrement(&temp);
+;    test(recurseCount - 1);
+;}
+;
+
+@count = dso_local local_unnamed_addr global i32 0, align 4
+
+; Function Attrs: nofree noinline norecurse nounwind uwtable
+declare void @_Z15globalIncrementPKi(i32* nocapture readonly %param) #0
+
+; Test that TRE could be done for recursive tail routine containing
+; call to function receiving a pointer to local stack.
+
+; Function Attrs: nounwind uwtable
+define dso_local void @_Z4testi(i32 %recurseCount) local_unnamed_addr #1 {
+; CHECK-LABEL: @_Z4testi(
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TEMP:%.*]] = alloca i32, align 4
+; CHECK-NEXT:    br label [[TAILRECURSE:%.*]]
+; CHECK:       tailrecurse:
+; CHECK-NEXT:    [[RECURSECOUNT_TR:%.*]] = phi i32 [ [[RECURSECOUNT:%.*]], [[ENTRY:%.*]] ], [ [[SUB:%.*]], [[IF_END:%.*]] ]
+; CHECK-NEXT:    [[CMP:%.*]] = icmp eq i32 [[RECURSECOUNT_TR]], 0
+; CHECK-NEXT:    br i1 [[CMP]], label [[RETURN:%.*]], label [[IF_END]]
+; CHECK:       if.end:
+; CHECK-NEXT:    [[TMP0:%.*]] = bitcast i32* [[TEMP]] to i8*
+; CHECK-NEXT:    call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull [[TMP0]])
+; CHECK-NEXT:    store i32 10, i32* [[TEMP]], align 4
+; CHECK-NEXT:    call void @_Z15globalIncrementPKi(i32* nonnull [[TEMP]])
+; CHECK-NEXT:    [[SUB]] = add nsw i32 [[RECURSECOUNT_TR]], -1
+; CHECK-NEXT:    call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull [[TMP0]])
+; CHECK-NEXT:    br label [[TAILRECURSE]]
+; CHECK:       return:
+; CHECK-NEXT:    ret void
+;
+entry:
+  %temp = alloca i32, align 4
+  %cmp = icmp eq i32 %recurseCount, 0
+  br i1 %cmp, label %return, label %if.end
+
+if.end:                                           ; preds = %entry
+  %0 = bitcast i32* %temp to i8*
+  call void @llvm.lifetime.start.p0i8(i64 4, i8* nonnull %0) #6
+  store i32 10, i32* %temp, align 4
+  call void @_Z15globalIncrementPKi(i32* nonnull %temp)
+  %sub = add nsw i32 %recurseCount, -1
+  call void @_Z4testi(i32 %sub)
+  call void @llvm.lifetime.end.p0i8(i64 4, i8* nonnull %0) #6
+  br label %return
+
+return:                                           ; preds = %entry, %if.end
+  ret void
+}
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.start.p0i8(i64 immarg, i8* nocapture) #2
+
+; Function Attrs: argmemonly nounwind willreturn
+declare void @llvm.lifetime.end.p0i8(i64 immarg, i8* nocapture) #2
+
+attributes #0 = { nofree noinline norecurse nounwind uwtable }
+attributes #1 = { nounwind uwtable }
+attributes #2 = { argmemonly nounwind willreturn }