Evgeniy Stepanov [Tue, 23 Jun 2015 20:57:26 +0000 (20:57 +0000)]
Pass 2 more variables to lit tests.
Pass ADB and ADB_SERIAL environment variables to lit tests.
This would allow running Android tests in compiler-rt when
there is more than one device attached to the host.
Mark Heffernan [Tue, 23 Jun 2015 18:26:50 +0000 (18:26 +0000)]
This change fixes three bugs in loop unswitching. This change causes an 81% speed-up on a benchmark that is based on EigenConvolutionKernel2D from Eigen3, where the lack of loop unswitching blocks hoisting of loads out of a nested loop (see bug 23816 for how loop unswitching and load hoisting are related).
Change 1: Unswitching on trivial conditions should always happen regardless of the computed unswitching cost, as really the cost is zero. While there is code to make that happen, the logic that checks the unswitching cost against a threshold was moved to an earlier point (revision 147935) than the point where trivial unswitching is detected, so trivial unswitching is currently blocked by the cost threshold. This change fixes that.
Change 2: Before revision 147935 (from 2012-01-11), the threshold parameter was a per-loop threshold. So an unswitching happened only if the cost of the unswitching was less than the threshold. In an indirect way (and I believe unintentionally), the logic for this since then has been that the threshold is an over-all budget across all loops for all loop unswitching done by a given LoopUnswitch loop pass object. So if an unswitching with cost 100 happens in one function, that in effect reduces the threshold from 100 to 0 for the loops even in another function. This persists for the lifetime of that loop pass object. This makes no difference for most small examples but it is important for large examples. This revision fixes that.
Change 3: The cost is currently calculated as std::min(NumInstructions, 5 * NumBlocks). So a loop with 2 blocks and a million instructions will have an unswitching cost of 10. I changed this to just NumInstructions, as it were before revision 147935, though I'm open to e.g. instead replacing std::min with std::max.
I've tried to make the change minimally invasive while staying with what I think was the original intent of the code.
Submitted on behalf of broune@.
Alex Lorenz [Tue, 23 Jun 2015 18:22:10 +0000 (18:22 +0000)]
ADT: Add a string APSInt constructor.
This commit moves the APSInt initialization code that's used by
the LLLexer class into a new APSInt constructor that constructs
APSInts from strings.
This change is useful for MIR Serialization, as it would allow
the MILexer class to use the same APSInt initialization as
LLexer when parsing immediate machine operands.
Alex Lorenz [Tue, 23 Jun 2015 17:10:10 +0000 (17:10 +0000)]
AsmParser: Extend the API to make the global value and metadata node slot mappings publicly accessible.
This commit creates a new structure called 'SlotMapping' in the AsmParser library.
This structure can be passed into the public parsing APIs from the AsmParser library
in order to extract the data structures that map from slot numbers to unnamed global
values and metadata nodes.
This change is useful for MIR Serialization, as the MIR Parser has to lookup the
unnamed global values and metadata nodes by their slot numbers.
John Brawn [Tue, 23 Jun 2015 16:02:11 +0000 (16:02 +0000)]
[ARM] ARMLoadStoreOpt::UpdateBaseRegUses should stop on def
When UpdateBaseRegUses sees an instruction that defines the base
register it must stop, as the base register value it is updating is no
longer live. Ideally we would already have seen the register be killed
(which is already checked for), but the kill flags may be inaccurate
and we have to account for this.
Benjamin Kramer [Tue, 23 Jun 2015 15:28:10 +0000 (15:28 +0000)]
[Option] Plug a leak when move-assigning an InputArgList.
The class has a non-trivial dtor so we have to clean up before we move
in new members. Remove misleading comment as a default move assignment
operator will never be synthesized for this class.
Toma Tabacu [Tue, 23 Jun 2015 14:39:42 +0000 (14:39 +0000)]
[mips] [IAS] Add partial support for the ULHU pseudo-instruction.
Summary:
This only adds support for ULHU of an immediate address with/without a source register.
It does not include support for ULHU of the address of a symbol.
Rafael Espindola [Tue, 23 Jun 2015 12:21:54 +0000 (12:21 +0000)]
Use MCSymbols for FastISel.
The summary is that it moves the mangling earlier and replaces a few
calls to .addExternalSymbol with addSym.
I originally wanted to replace all the uses of addExternalSymbol with
addSym, but noticed it was a lot of work and doesn't need to be done
all at once.
Daniel Jasper [Tue, 23 Jun 2015 11:31:32 +0000 (11:31 +0000)]
Revert r240302 ("Bring r240130 back.").
This causes errors like:
ld: error: blah.o: requires dynamic R_X86_64_PC32 reloc against '' which
may overflow at runtime; recompile with -fPIC
blah.cc:function f(): error: undefined reference to ''
blah.o:g(): error: undefined reference to ''
I have not yet come up with an appropriate reproduction.
Justin Bogner [Tue, 23 Jun 2015 07:32:55 +0000 (07:32 +0000)]
MCExpr: Avoid UB by evaluating this shift as unsigned
We hit undefined behaviour in some MCExpr tests when the LHS of a left
shift is -1. Twos-complement semantics are completely reasonable here,
so we should just do the shift in unsigned.
Justin Bogner [Tue, 23 Jun 2015 06:46:54 +0000 (06:46 +0000)]
test: Move target dependent test in their own folder for c API test
Dissasembly tests depends on target. The problem is that it disable
all tests if all targets are not compiled. This moves things around in
order to get target specific code in a target specific folder.
Sanjoy Das [Tue, 23 Jun 2015 01:05:26 +0000 (01:05 +0000)]
[FaultMaps] Move FaultMapParser to Object/
Summary:
That way llvm-objdump can rely on it without adding an extra dependency
on CodeGen.
This change duplicates the FaultKind enum and the code that serializes
it to a string. I could not figure out a way to get around this without
adding a new dependency to Object
Sanjay Patel [Tue, 23 Jun 2015 00:39:40 +0000 (00:39 +0000)]
[x86] generalize reassociation optimization in machine combiner to 2 instructions
Currently ( D10321, http://reviews.llvm.org/rL239486 ), we can use the machine combiner pass
to reassociate the following sequence to reduce the critical path:
A = ? op ?
B = A op X
C = B op Y
-->
A = ? op ?
B = X op Y
C = A op B
'op' is currently limited to x86 AVX scalar FP adds (with fast-math on), but in theory, it could
be any associative math/logic op (see TODO in code comment).
This patch generalizes the pattern match to ignore the instruction that defines 'A'. So instead of
a sequence of 3 adds, we now only need to find 2 dependent adds and decide if it's worth
reassociating them.
This generalization has a compile-time cost because we can now match more instruction sequences
and we rely more heavily on the machine combiner to discard sequences where reassociation doesn't
improve the critical path.
For example, in the new test case:
A = M div N
B = A add X
C = B add Y
We'll match 2 reassociation patterns, but this transform doesn't reduce the critical path:
A = M div N
B = A add Y
C = B add X
We need the combiner to reject that pattern but select this:
Logan Chien [Mon, 22 Jun 2015 23:16:02 +0000 (23:16 +0000)]
Code cleanup: Remove std::move() around xvalue (NFC)
Remove std::move() around xvalue so that copy elision is eligible.
In case that copy elision is not appliable, the c++ standard also
guarantees the move semantics on xvalue. Thus, it is not necessary
to wrap Args with std::move.
Adam Nemet [Mon, 22 Jun 2015 22:59:40 +0000 (22:59 +0000)]
[LoopDist] Improve variable names and comments in LoopVersioning class, NFC
As with the previous patch, the goal is to turn the class into a general
loop-versioning class. This patch removes any references to loop
distribution.
David Blaikie [Mon, 22 Jun 2015 22:06:37 +0000 (22:06 +0000)]
Modify ParseArgs to return the InputArgList by value - there's no need for dynamic allocation/ownership here
The one caller that does anything other than keep this variable on the
stack is the single use of DerivedArgList in Clang, which is a bit more
interesting but can probably be cleaned up/simplified a bit further
(have DerivedArgList take ownership of the InputArgList rather than
needing to reference its Args indirectly) which I'll try to after this.
Chris Bieneman [Mon, 22 Jun 2015 21:58:02 +0000 (21:58 +0000)]
[CMake] Treating LLVM_INCLUDE_TOOLS, which is a bool, as a string to change behaviors of the build is a dirty hack. We shouldn't do it.
Summary: I don't think anyone is relying on this behavior for bootstrapping (because I don't think it works), but if you do need it, speak now or forever hold your peace.
Frederic Riss [Mon, 22 Jun 2015 21:33:24 +0000 (21:33 +0000)]
[Object] Search for architecures by name in MachOUniversalBinary::getObjectForArch()
The reason we need to search by name rather than by Triple::ArchType
is to handle subarchitecture correclty. There is no different ArchType
for the x86_64h architecture (it identifies itself as x86_64), or for
the various ARM subarches. The only way to get to the subarch slice
in an universal binary is to search by name.
This issue led to hard to debug and transient symbolication failures
in Asan tests (it mostly works, because the files are very similar).
This also affects the Profiling infrastucture as it is the other user
of that API.
Ahmed Bougacha [Mon, 22 Jun 2015 20:51:51 +0000 (20:51 +0000)]
[X86] Teach load folding to accept scalar _Int users of MOVSS/MOVSD.
The _Int instructions are special, in that they operate on the full
VR128 instead of FR32. The load folding then looks at MOVSS, at the
user, and bails out when it sees a size mismatch.
What we really know is that the rm_Int instructions don't load the
higher lanes, so folding is fine.
This happens for the straightforward intrinsic code, e.g.:
Alex Lorenz [Mon, 22 Jun 2015 20:37:46 +0000 (20:37 +0000)]
MIR Serialization: Introduce a lexer for machine instructions.
This commit adds a function that tokenizes the string containing
the machine instruction. This commit also adds a struct called
'MIToken' which is used to represent the lexer's tokens.
Pete Cooper [Mon, 22 Jun 2015 19:57:33 +0000 (19:57 +0000)]
Move MCSymbol Value in to the union of Offset and CommonSize.
This is a reapplication of r239440 which was reverted in r239441.
There are no changes to this patch from then, but this had instead exposed
a bug in .thumb_set which was fixed in r240318. Having fixed that bug, it
is now safe to re-apply this code.
Original commit message below:
It wasn't possible to have a variable Symbol with offset or 'isCommon' so
this just enables better packing of the MCSymbol class.
Pete Cooper [Mon, 22 Jun 2015 19:35:57 +0000 (19:35 +0000)]
Change .thumb_set to have the same error checks as .set.
According to the documentation, .thumb_set is 'the equivalent of a .set directive'.
We didn't have equivalent behaviour in terms of all the errors we could throw, for
example, when a symbol is redefined.
This change refactors parseAssignment so that it can be used by .set and .thumb_set
and implements tests for .thumb_set for all the errors thrown by that method.
Sanjay Patel [Mon, 22 Jun 2015 18:29:44 +0000 (18:29 +0000)]
[x86] set default reciprocal (division and square root) codegen to match GCC
D8982 ( checked in at http://reviews.llvm.org/rL239001 ) added command-line
options to allow reciprocal estimate instructions to be used in place of
divisions and square roots.
This patch changes the default settings for x86 targets to allow that recip
codegen (except for scalar division because that breaks too much code) when
using -ffast-math or its equivalent.
This matches GCC behavior for this kind of codegen.
Sanjoy Das [Mon, 22 Jun 2015 18:20:10 +0000 (18:20 +0000)]
Fix MSVC build.
I had some unnecessary `typename`s left in after addressing review.
This compiled successfully with clang++ but MSVC reported an error. Fix
the build error by removing the redundant `typename`s.
Sanjoy Das [Mon, 22 Jun 2015 18:03:02 +0000 (18:03 +0000)]
[FaultMaps] Add a parser for the __llvm__faultmaps section.
Summary:
The parser is exercised by llvm-objdump using -print-fault-maps. As is
probably obvious, the code itself was "heavily inspired" by
http://reviews.llvm.org/D10434.
Rafael Espindola [Mon, 22 Jun 2015 17:46:53 +0000 (17:46 +0000)]
Avoid a Symbol -> Name -> Symbol conversion.
Before this we were producing a TargetExternalSymbol from a MCSymbol.
That meant extracting the symbol name and fetching the symbol again
down the pipeline.
This patch adds a DAG.getMCSymbol that lets the MCSymbol pass unchanged on the
DAG.
Doing so removes the need for MO_NOPREFIX and fixes the root cause of pr23900,
allowing r240130 to be committed again.
Alex Lorenz [Mon, 22 Jun 2015 17:02:30 +0000 (17:02 +0000)]
MIR Serialization: Serialize machine instruction names.
This commit implements initial machine instruction serialization. It
serializes machine instruction names. The instructions are represented
using a YAML sequence of string literals and are a part of machine
basic block YAML mapping.
This commit introduces a class called 'MIParser' which will be used to
parse the machine instructions and operands.
Pawel Bylica [Mon, 22 Jun 2015 15:58:11 +0000 (15:58 +0000)]
Fix shl folding in DAG combiner.
Summary: The code responsible for shl folding in the DAGCombiner was assuming incorrectly that all constants are less than 64 bits. This patch simply changes the way values are compared.
Joseph Tremoulet [Mon, 22 Jun 2015 15:27:58 +0000 (15:27 +0000)]
[ORC] Add NullResolver
Summary:
This is an implementation of RuntimeDyld::SymbolResolver that simply
rejects all resolution requests; useful for clients that do not have any
cross-object symbol references.
Toma Tabacu [Mon, 22 Jun 2015 12:08:39 +0000 (12:08 +0000)]
[mips] [IAS] Add support for LASym with identical source and destination register operands.
Summary:
In this case, we're supposed to load the address of the symbol in AT and then ADDu it with the source register and
put it in the destination register.
[X86] Allow more call sequences to use push instructions for argument passing
This allows more call sequences to use pushes instead of movs when optimizing for size.
In particular, calling conventions that pass some parameters in registers (e.g. thiscall) are now supported.
projects / llvm / log