Ahmed Bougacha [Tue, 2 Dec 2014 18:09:51 +0000 (18:09 +0000)]
[MachineCSE] Clear kill-flag on registers imp-def'd by the CSE'd instruction.
Go through implicit defs of CSMI and MI, and clear the kill flags on
their uses in all the instructions between CSMI and MI.
We might have made some of the kill flags redundant, consider:
subs ... %NZCV<imp-def> <- CSMI
csinc ... %NZCV<imp-use,kill> <- this kill flag isn't valid anymore
subs ... %NZCV<imp-def> <- MI, to be eliminated
csinc ... %NZCV<imp-use,kill>
Since we eliminated MI, and reused a register imp-def'd by CSMI
(here %NZCV), that register, if it was killed before MI, should have
that kill flag removed, because it's lifetime was extended.
Also, add an exhaustive testcase for the motivating example.
Tim Northover [Tue, 2 Dec 2014 17:15:22 +0000 (17:15 +0000)]
AArch64: make register block rules apply to vector types too.
The blocking code originated in ARM, which is more aggressive about casting
types to a canonical representative before doing anything else, so I missed out
most vector HFAs and broke the ABI. This should fix it.
Tom Stellard [Tue, 2 Dec 2014 16:45:47 +0000 (16:45 +0000)]
Triple: Add AMDHSA operating system type
This operating system type represents the AMD HSA runtime,
and will be required by the R600 backend in order to generate
correct code for this runtime.
[LICM] Avoind store sinking if no preheader is available
Load instructions are inserted into loop preheaders when sinking stores
and later removed if not used by the SSA updater. Avoid sinking if the
loop has no preheader and avoid crashes. This fixes one more side effect
of not handling indirectbr instructions properly on LoopSimplify.
Removing an unused function which is causing one of the build bots to fail.
This was introduced in the commit r223113. A proper cleanup of the so_imm
tblgen defintion (made redundant by the mod_imm definition) needs to happen
soon.
Add support for ARM modified-immediate assembly syntax.
Certain ARM instructions accept 32-bit immediate operands encoded as a 8-bit
integer value (0-255) and a 4-bit rotation (0-30, even). Current ARM assembly
syntax support in LLVM allows the decoded (32-bit) immediate to be specified
as a single immediate operand for such instructions:
The ARMARM defines an extended assembly syntax allowing the encoding to be made
more explicit, as in:
mov r0, #255, #8 ; (same 32-bit value as above)
The behaviour of the two instructions can be different w.r.t flags, which is
documented under "Modified immediate constants" in ARMARM. This patch enables
support for this extended syntax at the MC layer.
Charlie Turner [Tue, 2 Dec 2014 08:22:29 +0000 (08:22 +0000)]
Emit Tag_ABI_FP_denormal correctly in fast-math mode.
The default ARM floating-point mode does not support IEEE 754 mode exactly. Of
relevance to this patch is that input denormals are flushed to zero. The way in
which they're flushed to zero depends on the architecture,
* For VFPv2, it is implementation defined as to whether the sign of zero is
preserved.
* For VFPv3 and above, the sign of zero is always preserved when a denormal
is flushed to zero.
When FP support has been disabled, the strategy taken by this patch is to
assume the software support will mirror the behaviour of the hardware support
for the target *if it existed*. That is, for architectures which can only have
VFPv2, it is assumed the software will flush to positive zero. For later
architectures it is assumed the software will flush to zero preserving sign.
Fix several bugs in r221220's new program finding code.
In both the Unix and Windows variants, std::getenv was called and the
result passed directly to a function accepting a StringRef. This isn't
OK because it might return a null pointer and that causes the StringRef
constructor to assert (and generally produces crash-prone code if
asserts are disabled). Fix this by independently testing the result as
non-null prior to splitting things.
This in turn uncovered another bug in the Unix variant where it would
infinitely recurse if PATH="", or after this fix if PATH isn't set.
There is no need to recurse at all. Slightly re-arrange the code to make
it clear that we can just fixup the Paths argument based on the
environment if we find anything.
I don't know of a particularly useful way to test these routines in
LLVM. I'll commit a test to Clang that ensures that its driver correctly
handles various settings of PATH. However, I have no idea how to
correctly write a Windows test for the PATHEXT change. Any Windows
developers who could provide such a test, please have at. =D
Many thanks to Nick Lewycky and others for helping debug this. =/ It was
quite nasty for us to track down.
System memory allocation functions, which are identified at the IR level by the
noalias attribute on the return value, must return a pointer into a memory region
disjoint from any other memory accessible to the caller. We can use this
property to simplify pointer comparisons between allocated memory and local
stack addresses and the addresses of global variables. Neither the stack nor
global variables can overlap with the region used by the memory allocator.
Philip Reames [Mon, 1 Dec 2014 22:52:56 +0000 (22:52 +0000)]
[Statepoints 2/4] Statepoint infrastructure for garbage collection: MI & x86-64 Backend
This is the second patch in a small series. This patch contains the MachineInstruction and x86-64 backend pieces required to lower Statepoints. It does not include the code to actually generate the STATEPOINT machine instruction and as a result, the entire patch is currently dead code. I will be submitting the SelectionDAG parts within the next 24-48 hours. Since those pieces are by far the most complicated, I wanted to minimize the size of that patch. That patch will include the tests which exercise the functionality in this patch. The entire series can be seen as one combined whole in http://reviews.llvm.org/D5683.
The STATEPOINT psuedo node is generated after all gc values are explicitly spilled to stack slots. The purpose of this node is to wrap an actual call instruction while recording the spill locations of the meta arguments used for garbage collection and other purposes. The STATEPOINT is modeled as modifing all of those locations to prevent backend optimizations from forwarding the value from before the STATEPOINT to after the STATEPOINT. (Doing so would break relocation semantics for collectors which wish to relocate roots.)
The implementation of STATEPOINT is closely modeled on PATCHPOINT. Eventually, much of the code in this patch will be removed. The long term plan is to merge the functionality provided by statepoints and patchpoints. Merging their implementations in the backend is likely to be a good starting point.
Philip Reames [Mon, 1 Dec 2014 21:18:12 +0000 (21:18 +0000)]
[Statepoints 1/4] Statepoint infrastructure for garbage collection: IR Intrinsics
The statepoint intrinsics are intended to enable precise root tracking through the compiler as to support garbage collectors of all types. The addition of the statepoint intrinsics to LLVM should have no impact on the compilation of any program which does not contain them. There are no side tables created, no extra metadata, and no inhibited optimizations.
A statepoint works by transforming a call site (or safepoint poll site) into an explicit relocation operation. It is the frontend's responsibility (or eventually the safepoint insertion pass we've developed, but that's not part of this patch series) to ensure that any live pointer to a GC object is correctly added to the statepoint and explicitly relocated. The relocated value is just a normal SSA value (as seen by the optimizer), so merges of relocated and unrelocated values are just normal phis. The explicit relocation operation, the fact the statepoint is assumed to clobber all memory, and the optimizers standard semantics ensure that the relocations flow through IR optimizations correctly.
This is the first patch in a small series. This patch contains only the IR parts; the documentation and backend support will be following separately. The entire series can be seen as one combined whole in http://reviews.llvm.org/D5683.
Jingyue Wu [Mon, 1 Dec 2014 21:16:17 +0000 (21:16 +0000)]
[NVPTX] Do not emit .weak symbols for NVPTX
Summary:
".weak" symbols cannot be consumed by ptxas (PR21685). This patch makes the
weak directive in MCAsmPrinter customizable, and disables emitting ".weak"
symbols for NVPTX.
r208210 introduced an optimization that improves the vector select
codegen by doing the setcc on vectors directly.
This is a problem they the setcc operands are i1s, because the
optimization would create vectors of i1, which aren't legal.
Ahmed Bougacha [Mon, 1 Dec 2014 20:52:32 +0000 (20:52 +0000)]
[AArch64] Fix v2i8->i16 bitcast legalization.
r213378 improved f16 bitcasts, so that they go directly through subregs,
instead of through the stack. That code now causes an assertion failure
for bitcasts from other 16-bits types (most importantly v2i8).
Correct that by doing the custom lowering for i16 bitcasts only when the
input is an f16.
Ahmed Bougacha [Mon, 1 Dec 2014 18:43:53 +0000 (18:43 +0000)]
[MachineVerifier] Accept a MBB with a single landing pad successor.
The MachineVerifier used to check that there was always exactly one
unconditional branch to a non-landingpad (normal) successor.
If that normal successor to an invoke BB is unreachable, it seems
reasonable to only have one successor, the landing pad.
On targets other than AArch64 (and on AArch64 with a different testcase),
the branch folder turns the branch to the landing pad into a fallthrough.
The MachineVerifier, which relies on AnalyzeBranch, is unable to check
the condition, and doesn't complain. However, it does in this specific
testcase, where the branch to the landing pad remained.
Make the MachineVerifier accept it.
Hans Wennborg [Mon, 1 Dec 2014 17:08:32 +0000 (17:08 +0000)]
SelectionDAG switch lowering: Replace unreachable default with most popular case.
This can significantly reduce the size of the switch, allowing for more
efficient lowering.
I also worked with the idea of exploiting unreachable defaults by
omitting the range check for jump tables, but always ended up with a
non-neglible binary size increase. It might be worth looking into some more.
Matt Arsenault [Mon, 1 Dec 2014 15:52:46 +0000 (15:52 +0000)]
R600/SI: Various instruction format bit test cleanups
- Fix missing SALU format bits
- Remove unused isSALUInstr
- Add isVALU
- Switch isDS to use a bit like the others
- Move SIInstrInfo::is* functions to header
- Reorder so they are approximately sorted by type (SALU, VALU, memory)
Vladimir Medic [Mon, 1 Dec 2014 11:12:04 +0000 (11:12 +0000)]
The andi16, addiusp and jraddiusp micromips instructions were missing dedicated decoder methods in MipsDisassembler.cpp to properly decode immediate operands. These methods are added together with corresponding tests.
Jay Foad [Mon, 1 Dec 2014 09:42:32 +0000 (09:42 +0000)]
[PowerPC] Fix unwind info with dynamic stack realignment
Summary:
PowerPC DWARF unwind info defined CFA as SP + offset even in a function
where the stack had been dynamically realigned. This clearly doesn't
work because the offset from SP to CFA is not a constant. Fix it by
defining CFA as BP instead.
This was causing the AddressSanitizer null_deref test to fail 50% of
the time, depending on whether SP happened to be 32-byte aligned on
entry to a particular function or not.
Charlie Turner [Mon, 1 Dec 2014 08:50:27 +0000 (08:50 +0000)]
Add post-decode checking of HVC instruction.
Add checkDecodedInstruction for post-decode checking of instructions, to catch
the corner cases like HVC that don't fit into the general pattern. Needed to
check for an invalid condition field in instruction encoding despite HVC not
taking a predicate.
Craig Topper [Sun, 30 Nov 2014 01:20:17 +0000 (01:20 +0000)]
Revert r222957 "Replace std::map<K, V*> with std::map<K, V> to handle ownership and deletion of the values."
Upon further review I think the MultiClass is being copied into the map instead of being moved due to the copy constructor on the nested Record type. This ultimately got exposed when the vector in DefPrototype vector was changed to hold unique_ptrs in another commit. This caused gcc 4.7 to fail due to the use of the copy constructor on unique_ptr with the error pointing back to one of the insert calls from this commit. Not sure why clang was able to build.
Hans Wennborg [Sun, 30 Nov 2014 00:24:43 +0000 (00:24 +0000)]
Speculatively qualify some llvm::make_unique calls trying to please MSVC
It was failing with this kind of error:
C:\b\build\slave\CrWinClang\build\src\third_party\llvm\lib\TableGen\TGParser.cpp(1243) : error C2668: 'llvm::make_unique' : ambiguous call to overloaded function
C:\b\build\slave\CrWinClang\build\src\third_party\llvm\include\llvm/ADT/STLExtras.h(408): could be 'std::unique_ptr<llvm::Record,std::default_delete<_Ty>> llvm::make_unique<llvm::Record,std::string,llvm::SMLoc&,llvm::RecordKeeper&,bool>(std::string &&,llvm::SMLoc &,llvm::RecordKeeper &,bool &&)'
with
[
_Ty=llvm::Record
]
C:\b\depot_tools\win_toolchain\vs2013_files\win8sdk\bin\..\..\VC\include\memory(1637): or 'std::unique_ptr<llvm::Record,std::default_delete<_Ty>> std::make_unique<llvm::Record,std::string,llvm::SMLoc&,llvm::RecordKeeper&,bool>(std::string &&,llvm::SMLoc &,llvm::RecordKeeper &,bool &&)' [found using argument-dependent lookup]
with
[
_Ty=llvm::Record
]
while trying to match the argument list '(std::string, llvm::SMLoc, llvm::RecordKeeper, bool)'
David Blaikie [Sat, 29 Nov 2014 07:04:51 +0000 (07:04 +0000)]
Use deque<T> rather than vector<T*> since it provides the same invalidation semantics (at least when removal is not needed) without the extra indirection/ownership complexity
Order matters for this container, it seems (using a forward_list and
replacing the original push_backs with emplace_fronts caused test
failures). I didn't look too deeply into why.
(& in retrospect, I might go back & change some of the forward_lists I
introduced to deques anyway - since most don't require removal, deque is
a more memory-friendly data structure (moderate locality while not
invalidating pointers))
Craig Topper [Sat, 29 Nov 2014 05:31:10 +0000 (05:31 +0000)]
Use unique_ptr to remove some explicit deletes on some error case returns. At least one spot of weird ownership passing that needs some future cleanup.
Revert "Simplify some more ownership using forward_list<T> rather than vector<unique_ptr<T>>"
This reverts commit r222935 and its follow-up r222938 ("Push unique_ptr
a bit further through some APIs and simplify some cleanup"), since it
causes bot failures (at least on Darwin):
David Blaikie [Fri, 28 Nov 2014 21:37:54 +0000 (21:37 +0000)]
Use std::map<K, V> rather than std::map<K, std::unique_ptr<V>>
Pointers and references to map elements are never invalidated (except on
removal, which isn't used here) so there's no need for the indirection
unless there's polymorphism at work.
A little const correctness had to be fixed, since the indirection
allowed some benign const violations.
This reverts commit r222632 (and follow-up r222636), which caused a host
of LNT failures on an internal bot. I'll respond to the commit on the
list with a reproduction of one of the failures.
David Blaikie [Fri, 28 Nov 2014 20:35:57 +0000 (20:35 +0000)]
Simplify ownership by using forward_list<T> rather than vector<unique_ptr<T>>
Since the elements were not polymorphic, the unique_ptr was only used to
avoid pointer invalidation on container resizes - might as well skip the
indirection and use a container with suitable invalidation semantics.
David Majnemer [Fri, 28 Nov 2014 19:58:29 +0000 (19:58 +0000)]
InstCombine: FoldOrOfICmps harder
We may be in a situation where the icmps might not be near each other in
a tree of or instructions. Try to dig out related compare instructions
and see if they combine.
N.B. This won't fire on deep trees of compares because rewritting the
tree might end up creating a net increase of IR. We may have to resort
to something more sophisticated if this is a real problem.
Loop simplify skips exit-block insertion when exits contain indirectbr
instructions. This leads to an assertion in LICM when trying to sink
stores out of non-dedicated loop exits containing indirectbr
instructions. This patch fix this issue by re-checking for dedicated
exits in LICM prior to store sink attempts.
[SwitchLowering] Handle multiple destinations on condensed case stmts
Switch cases statements with sequential values that branch to the same
destination BB may often be handled together in a single new source BB.
In this scenario we need to remove remaining incoming values from PHI
instructions in the destination BB, as to match the number of source
branches.
Rafael Espindola [Fri, 28 Nov 2014 16:41:24 +0000 (16:41 +0000)]
Add back r222727 with a fix.
The original patch would fail when:
* A dst opaque type (%A) is matched with a src type (%A).
* A src opaque (%E) type is then speculatively matched with %A and the
speculation fails afterward.
* When rolling back the speculation we would cancel the source %A to dest
%A mapping.
The fix is to keep an explicit list of which resolutions are speculative.
Original message:
Fix overly aggressive type merging.
If we find out that two types are *not* isomorphic, we learn nothing about
opaque sub types in both the source and destination.
Evgeniy Stepanov [Fri, 28 Nov 2014 11:17:58 +0000 (11:17 +0000)]
[msan] Fix origin propagation for select of floats.
MSan does not assign origin for instrumentation temps (i.e. the ones that do
not come from the application code), but "select" instrumentation erroneously
tried to use one of those.
Charlie Turner [Fri, 28 Nov 2014 11:14:47 +0000 (11:14 +0000)]
Test all <build attribute, value> pairs.
Add more tests to make sure the encoding/decoding of build attributes works
correctly for all permissible values of build attributes. For cases where there
are an infinite number of such values, a representative subset has been settled
for.