Summary:
When apps or other libraries link against a library with symbol
versions, the version string is recorded in the import table, and used
at runtime to resolve the symbol back to a library that provides that
version (vaguely like how two-level namespaces work in Mach-O).  ld's
--default-symver flag tags every exported symbol with a symbol version
string equal to the library's soname.  Using --default-symver means
multiple versions of libLLVM can coexist within the same process, at
least to the extent that they don't try to pass data between each
other's llvms.

As an example, imagine a language like Rust using llvm for CPU codegen,
binding to OpenGL, with Mesa as the OpenGL implementation using llvm for
R600 codegen.  With --default-symver Rust and Mesa will resolve their
llvm usage to the version each was linked against, which need not match.

(Other ELF platforms like BSD and Solaris might have similar semantics,
I've not checked.)

This is based on an autoconf version of this patch by Adam Jackson.

This new option can be used to add --default-symver to the linker flags
for libLLVM.so.

Reviewers: beanz

Reviewed By: beanz

Subscribers: mgorny, llvm-commits

Differential Revision: https://reviews.llvm.org/D30997

llvm-svn: 302026

Reviewers: chandlerc, beanz, mgorny

Reviewed By: beanz

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D31773

llvm-svn: 302025

VFPCLASS is for vector types and not scalar, so it cannot get here.

Differential Revision: https://reviews.llvm.org/D32694

llvm-svn: 302023

It is needed to check that the number of operands are 2 when
finding the case of a logic combination, e.g. 'and' of two compares.

Review: Ulrich Weigand
llvm-svn: 302022

This patch implements the LLVM part for no_caller_saved_registers attribute as appears here: https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=5ed3cc7b66af4758f7849ed6f65f4365be8223be.
In order to implement the attribute, we use the dynamic CSR mechanism to remove returned/passed arguments from the function regmask/CSR list.

Differential Revision: https://reviews.llvm.org/D31876

llvm-svn: 302020

Fixes PR31789 - When loop-vectorize tries to use these intrinsics for a
non-default address space pointer we fail with a "Calling a function with a
bad singature!" assertion. This patch solves this by adding the 'vector of
pointers' argument as an overloaded type which will determine the address
space.

Differential revision: https://reviews.llvm.org/D31490

llvm-svn: 302018

llvm-svn: 302017

Summary:
Currently, loop deletion deletes loop where the only values
that are used outside the loop are loop-invariant.
This patch adds logic to delete loops where the loop is proven to be
never executed (i.e. the only predecessor of the loop preheader has a
constant conditional branch as terminator, and the preheader is not the
taken target). This will remove loops that become dead after
loop-unswitching generates constant conditional branches.

The next steps are:
1. moving the loop deletion implementation to LoopUtils.
2. Add logic in loop-simplifyCFG which will support changing conditional
constant branches to unconditional branches. If loops become unreachable in this
process, they can be removed using `deleteDeadLoop` function.

Reviewers: chandlerc, efriedma, sanjoy, reames

Reviewed by: sanjoy

Subscribers: mzolotukhin, llvm-commits

Differential Revision: https://reviews.llvm.org/D32494

llvm-svn: 302015

This reverts commit 358ad02d999e88853d2cfc954bd2f668308a51f7.

llvm-svn: 302014

The "macosx" OS type is still the canonical type. In the future "macos" will
become the canonical OS type (but we will still support "macosx").

rdar://27043820

Differential Revision: https://reviews.llvm.org/D32748

llvm-svn: 302011

Pre-commit as requested in D32769.

llvm-svn: 302010

The patch is failing to add StringTableStreamBuilder.h, but that isn't
even discovered because the corresponding StringTableStreamBuilder.cpp
isn't added to any CMakeLists.txt file and thus never built. I think
this patch is just incomplete.

llvm-svn: 302002

This was reported by the ASAN bot, and it turned out to be
a fairly fundamental problem with the design of VarStreamArray
and the way it passes context information to the extractor.

The fix was cumbersome, and I'm not entirely pleased with it,
so I plan to revisit this design in the future when I'm not
pressed to get the bots green again.  For now, this fixes
the issue by storing the context information by value instead
of by reference, and introduces some impossibly-confusing
template magic to make things "work".

llvm-svn: 301999

No change in which intrinsics should be speculated.

llvm-svn: 301995

We should always expect values to be named before running the module summary
analysis (see NameAnonGlobals pass), so it's fine if we crash in that case.

llvm-svn: 301991

Summary:
This is the corresponding llvm change to D28037 to ensure no performance
regression.

Reviewers: bogner, kbarton, hfinkel, iteratee, echristo

Subscribers: nemanjai, llvm-commits

Differential Revision: https://reviews.llvm.org/D28329

llvm-svn: 301990

llvm-svn: 301987

Previously we had knowledge of how to serialize and deserialize
a string table inside of DebugInfo/PDB, but the string table
that it serializes contains a piece that is actually considered
CodeView and can appear outside of a PDB.  We already have logic
in llvm-readobj and MCCodeView to read and write this format,
so it doesn't make sense to duplicate the logic in DebugInfoPDB
as well.

This patch makes codeview::StringTable (for writing) and
codeview::StringTableRef (for reading), updates DebugInfoPDB
to use these classes for its own writing, and updates llvm-readobj
to additionally use StringTableRef for reading.

It's a bit more difficult to get MCCodeView to use this for
writing, but it's a logical next step.

llvm-svn: 301986

llvm-svn: 301985

This patch verifies the .debug_line:
- verify all addresses in a line table sequence have ascending addresses
- verify that all line table file indexes are valid

Unit tests added for both cases.

Differential Revision: https://reviews.llvm.org/D32765

llvm-svn: 301984

When we replaced the multiplicand the destination node might already exist.
When that happens the original gets CSEd and deleted. However, it's actually
used as the offset so nonsense is produced.

Should fix PR32726.

llvm-svn: 301983

This time, I fixed, built, and tested clang.

This reverts r301712.

llvm-svn: 301981

Remove "_NC" suffix and semantics from TLSDESC_LD{64,32}_LO12 and
  TLSDESC_ADD_LO12 relocations
Rearrange ordering in AArch64.def to follow relocation encoding
Fix name:
  R_AARCH64_P32_LD64_GOT_LO12_NC => R_AARCH64_P32_LD32_GOT_LO12_NC
Add support for several "TLS", "TLSGD", and "TLSLD" relocations for
  ILP32
Fix return values from isNonILP32reloc
Add implementations for
  R_AARCH64_ADR_PREL_PG_HI21_NC, R_AARCH64_P32_LD32_GOT_LO12_NC,
  R_AARCH64_P32_TLSIE_LD32_GOTTPREL_LO12_NC,
  R_AARCH64_P32_TLSDESC_LD32_LO12, R_AARCH64_LD64_GOT_LO12_NC,
  *TLSLD_LDST128_DTPREL_LO12, *TLSLD_LDST128_DTPREL_LO12_NC,
  *TLSLE_LDST128_TPREL_LO12, *TLSLE_LDST128_TPREL_LO12_NC
Modify error messages to give name of equivalent relocation in the
  ABI not being used, along with better checking for non-existent
  requested relocations.
Added assembler support for "pg_hi21_nc"
Relocation definitions added without implementations:
  R_AARCH64_P32_TLSDESC_ADR_PREL21, R_AARCH64_P32_TLSGD_ADR_PREL21,
  R_AARCH64_P32_TLSGD_ADD_LO12_NC, R_AARCH64_P32_TLSLD_ADR_PREL21, 
  R_AARCH64_P32_TLSLD_ADR_PAGE21, R_AARCH64_P32_TLSLD_ADD_LO12_NC,
  R_AARCH64_P32_TLSLD_LD_PREL19, R_AARCH64_P32_TLSDESC_LD_PREL19,
  R_AARCH64_P32_TLSGD_ADR_PAGE21, R_AARCH64_P32_TLS_DTPREL,
  R_AARCH64_P32_TLS_DTPMOD, R_AARCH64_P32_TLS_TPREL,
  R_AARCH64_P32_TLSDESC
Fix encoding:
  R_AARCH64_P32_TLSDESC_ADR_PAGE21

Reviewers: Peter Smith

Patch by: Joel Jones (jjones@cavium.com)

Differential Revision: https://reviews.llvm.org/D32072

llvm-svn: 301980

The directory and file tables now have form-based content descriptors.
Parse these and extract the per-directory/file records based on the
descriptors.  For now we support only DW_FORM_string (inline) for the
path names; follow-up work will add support for indirect forms (i.e.,
DW_FORM_strp, strx<N>, and line_strp).

Differential Revision: http://reviews.llvm.org/D32713

llvm-svn: 301978

Turns out this wasn't NFC-ish at all because there's a bug processing shuffles
that change the size of their input vectors (that case always seems to trip us
up). 

This should fix PR32872 while we investigate how it failed and reduce a testcase:
https://bugs.llvm.org/show_bug.cgi?id=32872
 

llvm-svn: 301977

llvm-svn: 301974

LTO and other fancy linking previously led to DWARF that contained invalid references. We already validate that CU relative references fall into the CU, and the DW_FORM_ref_addr references fall inside the .debug_info section, but we didn't validate that the references pointed to correct DIE offsets. This new verification will ensure that all references refer to actual DIEs and not an offset in between.

This caught a bug in DWARFUnit::getDIEForOffset() where if you gave it any offset, it would match the DIE that mathes the offset _or_ the next DIE. This has been fixed.

Differential Revision: https://reviews.llvm.org/D32722

llvm-svn: 301971

I doubt anyone actually uses it, and I'm not even entirely convinced it exists
myself; but it is our default for "clang -arch armv6". Functionally, if it does
exist it's identical to the arm1176jz-f from LLVM's point of view (the
difference is apparently in the "Security Extensions").

llvm-svn: 301962

On AMDGPU if an SGPR is spilled to a VGPR, the frame index
is deleted. If there were any CSR SGPRs, this woudl
assert when setting the offset.

llvm-svn: 301961

This is a follow up to the previous
inline cost patch for quicker filtering.

llvm-svn: 301959

LDS use in leaf functions not currently handled.

llvm-svn: 301958

Patch by Colin LeMahieu.

llvm-svn: 301957

Patch by Colin LeMahieu.

llvm-svn: 301956

Patch by Sid Manning.

llvm-svn: 301955

The compiler was generating code that ends up ignoring a multiple
latency dependence between two instructions by scheduling the
intructions in back-to-back packets. 

The packetizer needs to end a packet if the latency of the current
current insruction and the source in the previous packet is
greater than 1 cycle. This case occurs when there is still room in
the current packet, but scheduling the instruction causes a stall.
Instead, the packetizer should start a new packet. Also, if the
current packet already contains a stall, then it is okay to add
another instruction to the packet that also causes a stall. This
occurs when there are no instructions that can be scheduled in
between the producer and consumer instructions.

This patch changes the latency for loads to 2 cycles from 3 cycles.
This change refects that a load only needs to be separated by
one extra packet to eliminate the stall.

Patch by Ikhlas Ajbar.

llvm-svn: 301954

llvm-svn: 301953

Patch by Colin LeMahieu.

llvm-svn: 301952

Patch by Colin LeMahieu.

llvm-svn: 301951

Just let TTI's cost do this instead of arbitrarily restricting
this.

llvm-svn: 301950

Patch by Colin LeMahieu.

llvm-svn: 301949