New testcase for PR1047

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32528 91177308-0d34-0410-b5e6-96231b3b80d8

Expand (f64 extload f32) to (f64 fp_ext (load f32)) if f64 type action is expand.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32527 91177308-0d34-0410-b5e6-96231b3b80d8

Expand fsqrt, fsin, and fcos to libcalls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32526 91177308-0d34-0410-b5e6-96231b3b80d8

Add #ifdef switch toggle between old and new pass manager.  However,
continue to use old pass manager at the moment. To use new manager
remove #define USE_OLD_PASSMANAGER 1 from Pass.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32525 91177308-0d34-0410-b5e6-96231b3b80d8

revert my recent int<->fp and vector union promotion changes, they expose
obscure bugs affecting the X86 code generator.  I will reenable this
when fixed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32524 91177308-0d34-0410-b5e6-96231b3b80d8

Expand f32 / f64 to i32 / i64 conversion to soft-fp library calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32523 91177308-0d34-0410-b5e6-96231b3b80d8

Fix a regression that completely broke make install.
Instead of installing into /usr/local/bin it would install into /usr/localbin

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32522 91177308-0d34-0410-b5e6-96231b3b80d8

Replace CastInst::createInferredCast calls with more accurate cast
creation calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32521 91177308-0d34-0410-b5e6-96231b3b80d8

Change createInferredCast calls to more accurate cast creation calls.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32520 91177308-0d34-0410-b5e6-96231b3b80d8

FunctionPassManager does not support runOnModule().

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32519 91177308-0d34-0410-b5e6-96231b3b80d8

Move getAnalysis() and getAnalysisID() definitions from Pass.h to
PassAnalysisSupport.h

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32518 91177308-0d34-0410-b5e6-96231b3b80d8

Implement PassManager_New destructors.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32517 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused constructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32516 91177308-0d34-0410-b5e6-96231b3b80d8

Replace inferred getCast(V,Ty) calls with more strict variants.
Rename getZeroExtend and getSignExtend to getZExt and getSExt to match
the the casting mnemonics in the rest of LLVM.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32514 91177308-0d34-0410-b5e6-96231b3b80d8

Add routines to dump pass manager queue.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32513 91177308-0d34-0410-b5e6-96231b3b80d8

FunctionPassManager()

Set AnalysisResolver_New and add FPM to PassManagers list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32512 91177308-0d34-0410-b5e6-96231b3b80d8

Do not runOnFunction on external functions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32510 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize AnalysisImpls for each pass before executing the pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32509 91177308-0d34-0410-b5e6-96231b3b80d8

collectRequiredAnalysisPasses().

Include RequiredTrainsitiveSet also.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32508 91177308-0d34-0410-b5e6-96231b3b80d8

removeNotPreservedAnalysis().

Do not remove ImmutablePass from the list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32507 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize activeManager.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32506 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unused constructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32505 91177308-0d34-0410-b5e6-96231b3b80d8

Maintain ImmutablePasses list at top level only. Do not make them
directly available to individual managers.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32504 91177308-0d34-0410-b5e6-96231b3b80d8

findAnalysisPass().

First search all available passes before searching ImmutablePasses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32503 91177308-0d34-0410-b5e6-96231b3b80d8

Fix thinko.
While searching for a analysis in a pass manager, do not search it into
pass manager's manager.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32501 91177308-0d34-0410-b5e6-96231b3b80d8

o s/OtherPassManagers/IndirectPassManagers
o Make IndirectPassManagers vector of PMDataManager *
o Move PMTopLevelManager implementation below all class declarations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32499 91177308-0d34-0410-b5e6-96231b3b80d8

Initialize AnalysisImpls for ImmutablePass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32498 91177308-0d34-0410-b5e6-96231b3b80d8

Expand FP constant to integers if FP types are not legal.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32497 91177308-0d34-0410-b5e6-96231b3b80d8

Set top level manager.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32496 91177308-0d34-0410-b5e6-96231b3b80d8

Soft fp FNEG, SINT_TO_FP, UINT_TO_FP libcall expansion.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32495 91177308-0d34-0410-b5e6-96231b3b80d8

Expand ConstantFP to load from CP if float types are being expanded.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32494 91177308-0d34-0410-b5e6-96231b3b80d8

Expand i32/i64 CopyToReg f32/f64 to BIT_CONVERT + CopyToReg.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32493 91177308-0d34-0410-b5e6-96231b3b80d8

Honor cpu directive, take two.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32492 91177308-0d34-0410-b5e6-96231b3b80d8

Ignore entries with blank names.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32491 91177308-0d34-0410-b5e6-96231b3b80d8

- When expanding a bit_convert whose src operand is also to be expanded and
its expansion result type is equal to the result type of the bit_convert,
e.g. (i64 bit_convert (f64 op)) if FP is not legal
returns the result of the expanded source operand.
- Store f32 / f64 may be expanded to a single store i32/i64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32490 91177308-0d34-0410-b5e6-96231b3b80d8

Remove unneeded include.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32489 91177308-0d34-0410-b5e6-96231b3b80d8

Rollback changes to take a different tack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32488 91177308-0d34-0410-b5e6-96231b3b80d8

this can be trunc or bitcast, per line 3092.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32487 91177308-0d34-0410-b5e6-96231b3b80d8

Fix regression on 400.perlbench last night.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32486 91177308-0d34-0410-b5e6-96231b3b80d8

more general matching of the MVN instruction

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32484 91177308-0d34-0410-b5e6-96231b3b80d8

Honor the command line specification for machine type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32483 91177308-0d34-0410-b5e6-96231b3b80d8

don't use "ordinary" addressing mode 1 when mvn is appropriate

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32482 91177308-0d34-0410-b5e6-96231b3b80d8

Reduce number of instructions to load 64-bit constants.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32481 91177308-0d34-0410-b5e6-96231b3b80d8

Fix numerous inferred casts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32479 91177308-0d34-0410-b5e6-96231b3b80d8

Get even more accurate on the casting.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32478 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the casting for the computation of the Malloc size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32477 91177308-0d34-0410-b5e6-96231b3b80d8

Expand formal arguments and call arguments recursively: e.g. f64 -> i64 -> 2 x i32.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32476 91177308-0d34-0410-b5e6-96231b3b80d8

Don't create usless casts for same-bith-width floating point casts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32475 91177308-0d34-0410-b5e6-96231b3b80d8

fit in 80 cols

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32474 91177308-0d34-0410-b5e6-96231b3b80d8

this can only be fptrunc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32473 91177308-0d34-0410-b5e6-96231b3b80d8

This case isn't needed with recent changes to ConstantInt::get

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32472 91177308-0d34-0410-b5e6-96231b3b80d8

split up inttoptr from ptrtoint handling, the cases aren't similar at all.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32471 91177308-0d34-0410-b5e6-96231b3b80d8

the operand of a bitcast is always the right size, just emit it in place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32470 91177308-0d34-0410-b5e6-96231b3b80d8

Change inferred getCast into specific getCast. Passes all tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32469 91177308-0d34-0410-b5e6-96231b3b80d8

Patch for PR1045 and Transforms/ScalarRepl/2006-12-11-SROA-Crash.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32468 91177308-0d34-0410-b5e6-96231b3b80d8

new testcase for pr1045

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32467 91177308-0d34-0410-b5e6-96231b3b80d8

Revert Nate's patch to fix X86/store-fp-constant.ll.  With the dag combiner
and legalizer separated like they currently are, I don't see a way to handle
this xform.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32466 91177308-0d34-0410-b5e6-96231b3b80d8

make this code more aggressive about turning store fpimm into store int imm.
This is not sufficient to fix X86/store-fp-constant.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32465 91177308-0d34-0410-b5e6-96231b3b80d8

This should check for darwin literal sections as well.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32464 91177308-0d34-0410-b5e6-96231b3b80d8

teach scev to analyze X*4|1 like X*4+c.  This allows us to produce:

LBB1_1: #bb
        movdqa (%esi), %xmm2
        movaps %xmm2, %xmm3
        punpcklbw %xmm0, %xmm3
        movaps %xmm3, %xmm4
        punpcklwd %xmm0, %xmm4
        cvtdq2ps %xmm4, %xmm4
        mulps %xmm1, %xmm4
        movaps %xmm4, (%edi)
        leal 1(,%eax,4), %ebx
        shll $4, %ebx
        punpckhwd %xmm0, %xmm3
        cvtdq2ps %xmm3, %xmm3
        mulps %xmm1, %xmm3
        movaps %xmm3, (%edx,%ebx)
        leal 2(,%eax,4), %ebx
        shll $4, %ebx
        punpckhbw %xmm0, %xmm2
        movaps %xmm2, %xmm3
        punpcklwd %xmm0, %xmm3
        cvtdq2ps %xmm3, %xmm3
        mulps %xmm1, %xmm3
        movaps %xmm3, (%edx,%ebx)
        leal 3(,%eax,4), %ebx
        shll $4, %ebx
        punpckhwd %xmm0, %xmm2
        cvtdq2ps %xmm2, %xmm2
        mulps %xmm1, %xmm2
        movaps %xmm2, (%edx,%ebx)
        addl $64, %edi
        incl %eax
        addl $16, %esi
        cmpl %ecx, %eax
        jne LBB1_1      #bb

instead of:

LBB1_1: #bb
        movdqa (%esi), %xmm2
        movaps %xmm2, %xmm3
        punpcklbw %xmm0, %xmm3
        movaps %xmm3, %xmm4
        punpcklwd %xmm0, %xmm4
        cvtdq2ps %xmm4, %xmm4
        mulps %xmm1, %xmm4
        movaps %xmm4, (%edi)
        leal 1(,%eax,4), %ebx
        shll $4, %ebx
        punpckhwd %xmm0, %xmm3
        cvtdq2ps %xmm3, %xmm3
        mulps %xmm1, %xmm3
        movaps %xmm3, (%edx,%ebx)
        leal 2(,%eax,4), %ebx
        shll $4, %ebx
        punpckhbw %xmm0, %xmm2
        movaps %xmm2, %xmm3
        punpcklwd %xmm0, %xmm3
        cvtdq2ps %xmm3, %xmm3
        mulps %xmm1, %xmm3
        movaps %xmm3, (%edx,%ebx)
        leal 3(,%eax,4), %ebx
        shll $4, %ebx
        punpckhwd %xmm0, %xmm2
        cvtdq2ps %xmm2, %xmm2
        mulps %xmm1, %xmm2
        movaps %xmm2, (%edx,%ebx)
        addl $64, %edi
        incl %eax
        addl $16, %esi
        cmpl %ecx, %eax
        jne LBB1_1      #bb

for a testcase.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32463 91177308-0d34-0410-b5e6-96231b3b80d8

Tidy up a bit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32462 91177308-0d34-0410-b5e6-96231b3b80d8

Generate the correct cast opcode for constant expressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32461 91177308-0d34-0410-b5e6-96231b3b80d8

Change inferred cast creation calls to more specific cast creations.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32460 91177308-0d34-0410-b5e6-96231b3b80d8

use MVN to handle small negative constants

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32459 91177308-0d34-0410-b5e6-96231b3b80d8

Don't bother upgrading this.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32458 91177308-0d34-0410-b5e6-96231b3b80d8

Implement getIntegerCast and getFPCast for ConstantExpr. These are similar
to the createIntegerCast and createFPCast for CastInst instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32457 91177308-0d34-0410-b5e6-96231b3b80d8

Implement createIntegerCast and createFPCast factory methods for handling
integer and floating point cast creation. createIntegerCast generates
ZExt/SExt, BitCast or Trunc. createFPCast generates FPExt, Bitcast, or
FPTrunc.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32456 91177308-0d34-0410-b5e6-96231b3b80d8

Fix PR1040:
Don't rebuild llvm-config if none of the library dependencies changed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32455 91177308-0d34-0410-b5e6-96231b3b80d8

add mvn

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32454 91177308-0d34-0410-b5e6-96231b3b80d8

Fix the BitCastUnion type for 32-bit targets.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32453 91177308-0d34-0410-b5e6-96231b3b80d8

add note

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32452 91177308-0d34-0410-b5e6-96231b3b80d8

Another step forward in PPC64 JIT support: we now no-longer need stubs
emitted for external globals in PPC64-JIT-PIC mode (which is good because
we didn't handle them before!).

This also fixes a bug handling the picbase delta, which we would get wrong
in some cases.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32451 91177308-0d34-0410-b5e6-96231b3b80d8

Create the cast constant expression that was read instead of attempting
to infer the cast from its operand and type. This fixes:
test/Regression/Bytecode/2006-12-11-Cast-ConstExpr.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32450 91177308-0d34-0410-b5e6-96231b3b80d8

Test case for ensuring correct reading back of cast constant expressions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32449 91177308-0d34-0410-b5e6-96231b3b80d8

getInstrItineraryData shouldn't copy the itineraries

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32448 91177308-0d34-0410-b5e6-96231b3b80d8

Fix constant folding of FP->int due to cut & paste error in last commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32447 91177308-0d34-0410-b5e6-96231b3b80d8

Add a test for constant folding of bitcast to/from floating point.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32446 91177308-0d34-0410-b5e6-96231b3b80d8

Provide a test for int<->float and long<->double bitcasts.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32445 91177308-0d34-0410-b5e6-96231b3b80d8

Implement correct bitcast of int<->float and long<->double by using a
union to perform the bitcast.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32444 91177308-0d34-0410-b5e6-96231b3b80d8

Re-apply changes that were backed out and fix a naughty typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32442 91177308-0d34-0410-b5e6-96231b3b80d8

It seems the llvm::OStream class does not handle stream manipulators.
For now, just grab the stream and perform the output on it directly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32441 91177308-0d34-0410-b5e6-96231b3b80d8

Revert changes that broke oggenc on ppc

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32440 91177308-0d34-0410-b5e6-96231b3b80d8

Missing opcode.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32439 91177308-0d34-0410-b5e6-96231b3b80d8

Implement correct constant folding of bitcast.  This implements
Transforms/ConstProp/bitcast.ll and fixes
SingleSource/Regression/C/2003-10-12-GlobalVarInitializers

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32438 91177308-0d34-0410-b5e6-96231b3b80d8

constant folding of bitcast is broken.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32437 91177308-0d34-0410-b5e6-96231b3b80d8

Layout proper frame for ppc64.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32436 91177308-0d34-0410-b5e6-96231b3b80d8

Provide a script that can track down which optimization pass causes
different code to be produced between two llvm builds that differe slightly.
This is useful in tracking down mis-optimization bugs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32435 91177308-0d34-0410-b5e6-96231b3b80d8

Don't convert store double C, Ptr to store long C, Ptr if i64 is not a legal type.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32434 91177308-0d34-0410-b5e6-96231b3b80d8

f32 / f64 node is expanded to one i32 / i64 node.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32433 91177308-0d34-0410-b5e6-96231b3b80d8

Clean up some bad code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32432 91177308-0d34-0410-b5e6-96231b3b80d8

Move something that should be in the dag combiner from the legalizer to the
dag combiner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32431 91177308-0d34-0410-b5e6-96231b3b80d8

Properly mangles symbol table names
Supports constant pools
Supports relocations to jump tables
Supports relocations within the data segment (global = address of global)
Allocates memory in a non-hacky for all non-code objects.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32430 91177308-0d34-0410-b5e6-96231b3b80d8

Add a default relocation type, and allow clients to set the ConstantVal
field after creation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32429 91177308-0d34-0410-b5e6-96231b3b80d8

fix typo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32428 91177308-0d34-0410-b5e6-96231b3b80d8

Update note, with the SROA change, we now produce:

_pairtest:
        movl 8(%esp), %eax
        movl 4(%esp), %ecx
        movd %eax, %xmm0
        movd %ecx, %xmm1
        addss %xmm0, %xmm1
        movl 12(%esp), %eax
        movss %xmm1, (%eax)
        ret

instead of:

_pairtest:
        subl $12, %esp
        movl 20(%esp), %eax
        movl %eax, 4(%esp)
        movl 16(%esp), %eax
        movl %eax, (%esp)
        movss (%esp), %xmm0
        addss 4(%esp), %xmm0
        movl 24(%esp), %eax
        movss %xmm0, (%eax)
        addl $12, %esp
        ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32427 91177308-0d34-0410-b5e6-96231b3b80d8

trunc to integer, not to FP.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32426 91177308-0d34-0410-b5e6-96231b3b80d8

the xform is done by scalarrepl not the code generator.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32425 91177308-0d34-0410-b5e6-96231b3b80d8

this is done

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32424 91177308-0d34-0410-b5e6-96231b3b80d8

Evan implemented the machineinstr improvements.
The rot example works if the type is cast to ubyte.

Move sroa examples together and upgrade them to HEAD syntax.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32423 91177308-0d34-0410-b5e6-96231b3b80d8

implement promotion of unions containing two packed types of the same width.
This implements Transforms/ScalarRepl/union-packed.ll

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@32422 91177308-0d34-0410-b5e6-96231b3b80d8