Alex Lorenz [Wed, 8 Jul 2015 20:22:20 +0000 (20:22 +0000)]
MIR Parser: Use source locations for MBB naming errors.
This commit changes the type of the field 'Name' in the struct
'yaml::MachineBasicBlock' from 'std::string' to 'yaml::StringValue'. This change
allows the MIR parser to report errors related to the MBB name with the proper
source locations.
The inferred output file name is based on the first input file, not the
first one with extension .obj. The output file was also being written to
the wrong directory; it needs to be written to whichever directory on the
libpath it was found in. This change fixes both issues.
Adam Nemet [Wed, 8 Jul 2015 18:47:38 +0000 (18:47 +0000)]
[LAA] Add missing debug output after r239285
r239285 ([LoopAccessAnalysis] Teach LAA to check the memory dependence
between strided accesses.) introduced a new case under
MemoryDepChecker::isDependent. We normally have debug output for each
case.
[SEH] Ensure that empty __except blocks have their own BB
The 32-bit lowering assumed that WinEHPrepare had this invariant.
WinEHPrepare did it for C++, but not SEH. The result was that we would
insert calls to llvm.x86.seh.restoreframe in normal basic blocks, which
corrupted the frame pointer.
James Y Knight [Wed, 8 Jul 2015 18:08:36 +0000 (18:08 +0000)]
Expand LangRef.html's documentation on LLVM's inline assembly.
While trying to figure out how this was all supposed to work, I
figured I'd start writing down some documentation, since it was
basically completely missing.
James Y Knight [Wed, 8 Jul 2015 16:25:12 +0000 (16:25 +0000)]
[SPARC] Cleanup handling of the Y/ASR registers.
- Implement copying ASR to/from GPR regs.
- Mark ASRs as non-allocatable, so it won't try to arbitrarily use
them inappropriately.
- Instead of inserting explicit WRASR/RDASR nodes in the MUL/DIV
routines, just do normal register copies.
- Also...mark div as using Y, not just writing it.
Added a test case with some code which previously died with an
assertion failure (with -O0), or produced wrong code (otherwise).
Simon Atanasyan [Wed, 8 Jul 2015 10:12:40 +0000 (10:12 +0000)]
[yaml2obj] Align section content using AddressAlign field's value
Use AddressAlign field's value to properly align sections content in the
yaml2obj tool. Before this change the yaml2obj ignored AddressAlign and
always aligned section on 16 bytes boundary.
[LAA] Merge memchecks for accesses separated by a constant offset
Summary:
Often filter-like loops will do memory accesses that are
separated by constant offsets. In these cases it is
common that we will exceed the threshold for the
allowable number of checks.
However, it should be possible to merge such checks,
sice a check of any interval againt two other intervals separated
by a constant offset (a,b), (a+c, b+c) will be equivalent with
a check againt (a, b+c), as long as (a,b) and (a+c, b+c) overlap.
Assuming the loop will be executed for a sufficient number of
iterations, this will be true. If not true, checking against
(a, b+c) is still safe (although not equivalent).
As long as there are no dependencies between two accesses,
we can merge their checks into a single one. We use this
technique to construct groups of accesses, and then check
the intervals associated with the groups instead of
checking the accesses directly.
Adrian Prantl [Wed, 8 Jul 2015 02:04:15 +0000 (02:04 +0000)]
llvm-objdump: Replace the -macho -raw option with a generic -raw-clang-ast
option that works with all object container formats.
Now that clang modules/PCH are object containers this option is useful to
to construct pipes like
Mehdi Amini [Wed, 8 Jul 2015 01:00:38 +0000 (01:00 +0000)]
Remove IsLittleEndian from TargetLowering and redirect to DataLayout
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Wed, 8 Jul 2015 01:00:31 +0000 (01:00 +0000)]
Redirect DataLayout from TargetMachine to Module in SjLjEHPrepare
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
[WinEH] Make llvm.x86.seh.restoreframe work for stack realignment prologues
The incoming EBP value points to the end of a local stack allocation, so
we can use that to restore ESI, the base pointer. Once we do that, we
can use local stack allocations. If we know we need stack realignment,
spill the original frame pointer in the prologue and reload it after
restoring ESI.
Mehdi Amini [Tue, 7 Jul 2015 23:38:49 +0000 (23:38 +0000)]
Redirect DataLayout from TargetMachine to Module in StackProtector
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Alex Lorenz [Tue, 7 Jul 2015 23:27:53 +0000 (23:27 +0000)]
MIR Printer: Use a module slot tracker to print global address operands. NFC.
This commit adopts the 'ModuleSlotTracker' class, which was surfaced in r240842,
to print the global address operands. This change ensures that the slot tracker
won't have to be recreated every time a global address operand is printed,
making the MIR printing more efficient.
Rename llvm.frameescape and llvm.framerecover to localescape and localrecover
Summary:
Initially, these intrinsics seemed like part of a family of "frame"
related intrinsics, but now I think that's more confusing than helpful.
Initially, the LangRef specified that this would create a new kind of
allocation that would be allocated at a fixed offset from the frame
pointer (EBP/RBP). We ended up dropping that design, and leaving the
stack frame layout alone.
These intrinsics are really about sharing local stack allocations, not
frame pointers. I intend to go further and add an `llvm.localaddress()`
intrinsic that returns whatever register (EBP, ESI, ESP, RBX) is being
used to address locals, which should not be confused with the frame
pointer.
Naming suggestions at this point are welcome, I'm happy to re-run sed.
Chris Bieneman [Tue, 7 Jul 2015 20:24:55 +0000 (20:24 +0000)]
[CMake] Cleanup tools/CMakeLists.txt to take advantage of the auto-registration that was already partially working.
Summary:
The tools CMakeLists file already had implicit tool registration, but there were a few things off about it that needed to be altered to make it work. This change addresses all that. The changes in this patch are:
* factored out canonicalizing tool names from paths to CMake variables
* removed the LLVM_IMPLICIT_PROJECT_IGNORE mechanism in favor of LLVM_EXTERNAL_${nameUPPER}_BUILD which I renamed to LLVM_TOOL_${nameUPPER}_BUILD because it applies to internal and external tools
* removed ignore_llvm_tool_subdirectory() in favor of just setting LLVM_TOOL_${nameUPPER}_BUILD to Off
* Added create_llvm_tool_options() to resolve a bug in add_llvm_external_project() - the old LLVM_EXTERNAL_${nameUPPER}_BUILD would not work on a clean CMake directory because the option could be created after it was set in code.
* Removed all but the minimum required calls to add_llvm_external_project from tools/CMakeLists.txt
Mehdi Amini [Tue, 7 Jul 2015 19:07:19 +0000 (19:07 +0000)]
Redirect DataLayout from TargetMachine to Module in SelectionDAG
Summary:
SelectionDAG itself is not invoking directly the DataLayout in the
TargetMachine, but the "TargetLowering" class is still using it. I'll
address it in a following commit.
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
David Majnemer [Tue, 7 Jul 2015 18:49:41 +0000 (18:49 +0000)]
Revert "Revert r241570, it caused PR24053"
This reverts commit r241602. We had a latent bug in SCCP where we would
make a basic block empty and then proceed to ask questions about it's
terminator.
Mehdi Amini [Tue, 7 Jul 2015 18:49:25 +0000 (18:49 +0000)]
Redirect DataLayout from TargetMachine to Module in GlobalMerge
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Tue, 7 Jul 2015 18:45:17 +0000 (18:45 +0000)]
Redirect DataLayout from TargetMachine to Module in CodeGen Prepare
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Tue, 7 Jul 2015 18:39:02 +0000 (18:39 +0000)]
Redirect DataLayout from TargetMachine to Module in FastISel
Summary:
This change is part of a series of commits dedicated to have a single
DataLayout during compilation by using always the one owned by the
module.
Mehdi Amini [Tue, 7 Jul 2015 18:20:57 +0000 (18:20 +0000)]
Redirect DataLayout from TargetMachine to Module in MachineFunction
Summary:
This change is part of a series of commits dedicated to have a
single DataLayout during compilation by using always the one owned by the
module.
Alex Lorenz [Tue, 7 Jul 2015 17:46:43 +0000 (17:46 +0000)]
MIR Parser: wrap 'MBBSlots' from the MI parsing functions in a struct. NFC.
This commit modifies the interface for the machine instruction parsing
functions by wrapping the parameter 'MBBSlots' in a new structure called
'PerFunctionMIParsingState'. This change is useful as in the future I will be
able to pass new parameters to the machine instruction parser just by modifying
the 'PerFunctionMIParsingState' structure instead of adding a new parameter to
each function.
Delete UnknownAddress. It is a perfectly valid symbol value.
getSymbolValue now returns a value that in convenient for most callers:
* 0 for undefined
* symbol size for common symbols
* offset/address for symbols the rest
Code that needs something more specific can check getSymbolFlags.
Dan Liew [Tue, 7 Jul 2015 15:50:33 +0000 (15:50 +0000)]
Fix bug in test-release.sh where the script would not exit if any
of the build stages that are sent through a pipe (e.g. tee) failed.
This potentially allowed builds and/or tests to fail without anyone
noticing. It appears that for the LLVM 3.6.[01] releases this actually
happened for the Ubuntu 14.04LTS binary releases. The essence of the
issue is that without ``set -o pipefail`` the following command in bash
has a zero exit code.
[WinEH] Add a report_fatal_error for 32-bit stack realignment
This type of prologue isn't supported yet. Implementing it should be a
matter of copying the adjusted incoming EBP into ESI (the base pointer)
instead of EBP. The original EBP can be saved and restored from other
memory afterwards.
This commit changes the target arch to fix the test case commited in r241566
that was failing on ninja-x64-msvc-RA-centos6. Also add checks to make sure
the callee's address is loaded to blx's operand.
In these two contexts we really just want the raw n_value. No need to use
getSymbolValue which checks for special cases where, semantically, the symbol
has no value.
David Majnemer [Tue, 7 Jul 2015 09:15:29 +0000 (09:15 +0000)]
[IR] Make getFirstNonPHI return null if the BB is empty
getFirstNonPHI's documentation states that it returns null if there is
no non-PHI instruction. However, it instead returns a pointer to the
end iterator. The implementation of getFirstNonPHI claims that
dereferencing the iterator will result in an assertion failure but this
doesn't occur. Instead, machinery like getFirstInsertionPt will attempt
to isa<> this invalid memory which results in unpredictable behavior.
Instead, make getFirst* return null if no such instruction exists.
[ARM] Define a subtarget feature and use it to decide whether long calls should
be emitted.
This is needed to enable ARM long calls for LTO and enable and disable it on a
per-function basis.
Out-of-tree projects currently using EnableARMLongCalls to emit long calls
should start passing "+long-calls" to the feature string (see the changes made
to clang in r241565).
Alex Lorenz [Tue, 7 Jul 2015 02:08:46 +0000 (02:08 +0000)]
MIR Parser: Verify the implicit machine register operands.
This commit verifies that the parsed machine instructions contain the implicit
register operands as specified by the MCInstrDesc. Variadic and call
instructions aren't verified.
[StackMap Liveness] Calling the base class' getAnalysisUsage method. NFCI.
Calling into the base class' getAnalysisUsage method after we did our pass
specific modifications. This shouldn't really matter since this is the last
pass in the pipeline anyways.
Alex Lorenz [Mon, 6 Jul 2015 23:07:26 +0000 (23:07 +0000)]
MIR Serialization: Serialize the implicit register flag.
This commit serializes the implicit flag for the register machine operands. It
introduces two new keywords into the machine instruction syntax: 'implicit' and
'implicit-def'. The 'implicit' keyword is used for the implicit register
operands, and the 'implicit-def' keyword is used for the register operands that
have both the implicit and the define flags set.
Simon Pilgrim [Mon, 6 Jul 2015 22:46:46 +0000 (22:46 +0000)]
[X86][AVX] Add support for shuffle decoding of vperm2f128/vperm2i128 with zero'd lanes
The vperm2f128/vperm2i128 shuffle mask decoding was not attempting to deal with shuffles that give zero lanes. This patch fixes this so that the assembly printer can provide shuffle comments.
As this decoder is also used in X86ISelLowering for shuffle combining, I've added an early-out to match existing behaviour. The hope is that we can add zero support in the future, this would allow other ops' decodes (e.g. insertps) to be combined as well.
Simon Pilgrim [Mon, 6 Jul 2015 20:46:41 +0000 (20:46 +0000)]
[X86][SSE4A] Shuffle lowering using SSE4A EXTRQ/INSERTQ instructions
This patch adds support for v8i16 and v16i8 shuffle lowering using the immediate versions of the SSE4A EXTRQ and INSERTQ instructions. Although rather limited (they can only act on the lower 64-bits of the source vectors, leave the upper 64-bits of the result vector undefined and don't have VEX encoded variants), the instructions are still useful for the zero extension of any lane (EXTRQ) or inserting a lane into another vector (INSERTQ). Testing demonstrated that it wasn't typically worth it to use these instructions for v2i64 or v4i32 vector shuffles although they are capable of it.
As well as adding specific pattern matching for the shuffles, the patch uses EXTRQ for zero extension cases where SSE41 isn't available and its more efficient than the SSE2 'unpack' default approach. It also adds shuffle decode support for the EXTRQ / INSERTQ cases when the instructions are handling full byte-sized extractions / insertions.
From this foundation, future patches will be able to make use of the instructions for situations that use their ability to extract/insert at the bit level.
Simon Pilgrim [Mon, 6 Jul 2015 20:30:47 +0000 (20:30 +0000)]
[X86][SSE] Use the general SMAX/SMIN/UMAX/UMIN opcodes and remove the X86 implementation
With the completion of D9746 there is now a common implementation of integer signed/unsigned min/max nodes, removing the need for the equivalent X86 specific implementations.
This patch removes the old X86ISD nodes, legalizes the relevant SSE2/SSE41/AVX2/AVX512 instructions for the ISD versions and converts the small amount of existing X86 code.
Alex Lorenz [Mon, 6 Jul 2015 17:44:26 +0000 (17:44 +0000)]
llc: Add a 'run-pass' option.
This commit adds a 'run-pass' option to llc, which instructs the compiler to run
one specific code generation pass only.
Llc already has the 'start-after' and the 'stop-after' options, and this new
option complements the other two by making it easier to write tests that want
to invoke a single pass only.
Daniel Sanders [Mon, 6 Jul 2015 16:56:07 +0000 (16:56 +0000)]
Change the last few internal StringRef triples into Triple objects.
Summary:
This concludes the patch series to eliminate StringRef forms of GNU triples
from the internals of LLVM that began in r239036.
At this point, the StringRef-form of GNU Triples should only be used in the
public API (including IR serialization) and a couple objects that directly
interact with the API (most notably the Module class). The next step is to
replace these Triple objects with the TargetTuple object that will represent
our authoratative/unambiguous internal equivalent to GNU Triples.
Teresa Johnson [Mon, 6 Jul 2015 16:22:42 +0000 (16:22 +0000)]
Resubmit "Add new EliminateAvailableExternally module pass" (r239480)
This change includes a fix for https://code.google.com/p/chromium/issues/detail?id=499508#c3,
which required updating the visibility for symbols with eliminated definitions.
--Original Commit Message--
Add new EliminateAvailableExternally module pass, which is performed in
O2 compiles just before GlobalDCE, unless we are preparing for LTO.
This pass eliminates available externally globals (turning them into
declarations), regardless of whether they are dead/unreferenced, since
we are guaranteed to have a copy available elsewhere at link time.
This enables additional opportunities for GlobalDCE.
If we are preparing for LTO (e.g. a -flto -c compile), the pass is not
included as we want to preserve available externally functions for possible
link time inlining. The FE indicates whether we are doing an -flto compile
via the new PrepareForLTO flag on the PassManagerBuilder.
projects / llvm / log