Owen Anderson [Fri, 27 Feb 2015 17:57:01 +0000 (17:57 +0000)]
Switch a std::map to a DenseMap in CodeGenRegisters.
The keys of the map are unique by pointer address, so there's no need
to use the llvm::less comparator. This allows us to use DenseMap
instead, which reduces tblgen time by 20% on my stress test.
Chandler Carruth [Fri, 27 Feb 2015 12:13:14 +0000 (12:13 +0000)]
[x86] Run most of the rest of the shuffle combining over non-128-bit
vectors. This lets us fix the rest of the v16 lowering problems when
pshufb is clearly better.
We might still be able to improve some of the lowerings by enabling the
other combine-based rewriting to fire for non-128-bit vectors, but this
at least should remove any regressions from using the fancy v16i16
lowering strategy.
Chandler Carruth [Fri, 27 Feb 2015 11:45:13 +0000 (11:45 +0000)]
[x86] Teach a bunch of the x86-specific shuffle combining to work with
256-bit vectors as well as 128-bit vectors. Fixes some of the redundant
shuffles for v16i16.
Chandler Carruth [Fri, 27 Feb 2015 11:33:46 +0000 (11:33 +0000)]
[x86] Make the v8i16 clever single-input shuffle lowering usable for
repeated 128-bit lane shuffles of wider vector types and use it to lower
256-bit v16i16 vector shuffles where applicable.
This should let us perfectly lowering the pattern of pshuflw and pshufhw
even for AVX2 256-bit patterns.
I've not added AVX-512 support, but it should be trivial for someone
working on that to wire up.
Note that currently this generates bad, long shuffle chains because we
don't combine 256-bit target shuffles. The subsequent patches will fix
that.
Chandler Carruth [Fri, 27 Feb 2015 11:25:10 +0000 (11:25 +0000)]
[x86] Add a bunch more tests for v16i16 shuffles. All of these are taken
by mirroring v8i16 test cases across both 128-bit lanes. This should
highlight problems where we aren't correctly using 128-bit shuffles to
implement things.
[mips] Account for constant-zero operands in ADDE nodes.
Summary:
We identify the cases where the operand to an ADDE node is a constant
zero. In such cases, we can avoid generating an extra ADDu instruction
disguised as an identity move alias (ie. addu $r, $r, 0 --> move $r, $r).
Anna Zaks [Fri, 27 Feb 2015 03:12:36 +0000 (03:12 +0000)]
[asan] Skip promotable allocas to improve performance at -O0
Currently, the ASan executables built with -O0 are unnecessarily slow.
The main reason is that ASan instrumentation pass inserts redundant
checks around promotable allocas. These allocas do not get instrumented
under -O1 because they get converted to virtual registered by mem2reg.
With this patch, ASan instrumentation pass will only instrument non
promotable allocas, giving us a speedup of 39% on a collection of
benchmarks with -O0. (There is no measurable speedup at -O1.)
Sanjoy Das [Fri, 27 Feb 2015 02:19:11 +0000 (02:19 +0000)]
Fix a use-iterator-after-invalidate error
AnalysisResult::getResultImpl reuses an iterator into a DenseMap after
inserting elements into it. This change adds code to recompute the
iterator before the second use.
Eric Christopher [Fri, 27 Feb 2015 00:11:34 +0000 (00:11 +0000)]
Rewrite MachineOperand::print and MachineInstr::print to avoid
uses of TM->getSubtargetImpl and propagate to all calls.
This could be a debugging regression in places where we had a
TargetMachine and/or MachineFunction but don't have it as part
of the MachineInstr. Fixing this would require passing a
MachineFunction/Function down through the print operator, but
none of the existing uses in tree seem to do this.
Zachary Turner [Thu, 26 Feb 2015 23:49:23 +0000 (23:49 +0000)]
[llvm-pdbdump] Fix dumping of function pointers and basic types.
Function pointers were not correctly handled by the dumper, and
they would print as "* name". They now print as
"int (__cdecl *name)(int arg1, int arg2)" as they should.
Also, doubles were being printed as floats. This fixes that bug
as well, and adds tests for all builtin types. as well as a test
for function pointers.
Eric Christopher [Thu, 26 Feb 2015 22:38:43 +0000 (22:38 +0000)]
getRegForInlineAsmConstraint wants to use TargetRegisterInfo for
a lookup, pass that in rather than use a naked call to getSubtargetImpl.
This involved passing down and around either a TargetMachine or
TargetRegisterInfo. Update all callers/definitions around the targets
and SelectionDAG.
Eric Christopher [Thu, 26 Feb 2015 22:38:34 +0000 (22:38 +0000)]
Add a TargetMachine argument to the AddressingModeMatcher, we'll
need this shortly to get a TargetRegisterInfo from the subtarget
for TargetLowering routines.
Chandler Carruth [Thu, 26 Feb 2015 22:15:34 +0000 (22:15 +0000)]
[x86] Fix PR22706 where we would incorrectly try lower a v32i8 dynamic
blend as legal.
We made the same mistake in two different places. Whenever we are custom
lowering a v32i8 blend we need to check whether we are custom lowering
it only for constant conditions that can be shuffled, or whether we
actually have AVX2 and full dynamic blending support on bytes. Both are
fixed, with comments added to make it clear what is going on and a new
test case.
Chandler Carruth [Thu, 26 Feb 2015 21:29:06 +0000 (21:29 +0000)]
[x86] Restructure the comments and the conditions for handling
dynamic blends.
This makes it much more clear what is going on. The case we're handling
is that of dynamic conditions, and we're bailing when the nature of the
vector types and subtarget preclude lowering the dynamic condition
vselect as an actual blend.
No functionality changed here, but this will make a subsequent bug-fix
to this code much more clear.
Chandler Carruth [Thu, 26 Feb 2015 21:21:36 +0000 (21:21 +0000)]
[x86] Re-order the combines of select in the X86 backend. This doesn't
change functionality, but makes it more clear that the dynamic case and
the shuffle case don't overlap in any interesting way.
Justin Bogner [Thu, 26 Feb 2015 20:06:28 +0000 (20:06 +0000)]
InstrProf: Simplify the construction of BinaryCoverageReader
Creating BinaryCoverageReader is a strange and complicated dance where
the constructor sets error codes that member functions will later
read, and the object is in an invalid state if readHeader isn't
immediately called after construction.
Instead, make the constructor private and add a static create method
to do the construction properly. This also has the benefit of removing
readHeader completely and simplifying the interface of the object.
Sanjoy Das [Thu, 26 Feb 2015 19:51:35 +0000 (19:51 +0000)]
SCEVExpander incorrectly marks generated subtractions as nuw/nsw
It is not sound to mark the increment operation as `nuw` or `nsw`
based on a proof off of the add recurrence if the increment operation
we emit happens to be a `sub` instruction.
I could not come up with a test case for this -- the cases where
SCEVExpander decides to emit a `sub` instruction is quite small, and I
cannot think of a way I'd be able to get SCEV to prove that the
increment does not overflow in those cases.
Frederic Riss [Thu, 26 Feb 2015 19:48:07 +0000 (19:48 +0000)]
[MC] Use the non-EH register mapping in the debug_frame section.
On 32bits x86 Darwin, the register mappings for the eh_frane and
debug_frame sections are different. Thus the same CFI instructions
should result in different registers in the object file. The
problem isn't target specific though, but it requires that the
mappings for EH register numbers be different from the standard
Dwarf one.
The patch looks a bit clumsy. LLVM uses the EH mapping as
canonical for everything frame related. Thus we need to do a
double conversion EH -> LLVM -> Non-EH, when emitting the
debug_frame section.
Hal Finkel [Thu, 26 Feb 2015 18:56:03 +0000 (18:56 +0000)]
[InstCombine/PowerPC] Convert aligned QPX load/store intrinsics into loads/stores
InstCombine has long had logic to convert aligned Altivec load/store intrinsics
into regular loads and stores. This mirrors that functionality for QPX vector
load/store intrinsics.
Paul Robinson [Thu, 26 Feb 2015 18:47:57 +0000 (18:47 +0000)]
When the source has a series of assignments, users reasonably want to
have the debugger step through each one individually. Turn off the
combine for adjacent stores at -O0 so we get this behavior.
Possibly, DAGCombine shouldn't run at all at -O0, but that's for
another day; see PR22346.
Petar Jovanovic [Thu, 26 Feb 2015 18:35:15 +0000 (18:35 +0000)]
Fix justify error for small structures in varargs for MIPS64BE
There was a problem when passing structures as variable arguments.
The structures smaller than 64 bit were not left justified on MIPS64
big endian. This is now fixed by shifting the value to make it left-
justified when appropriate.
This fixes the bug http://llvm.org/bugs/show_bug.cgi?id=21608
Use ".arch_extension" ARM directive to support hwdiv on krait
In case of "krait" CPU, asm printer doesn't emit any ".cpu" so the
features bits are not computed. This patch lets the asm printer
emit ".cpu cortex-a9" directive for krait and the hwdiv feature is
enabled through ".arch_extension". In short, krait is treated
as "cortex-a9" with hwdiv. We can not emit ".krait" as CPU since
it is not supported bu GNU GAS yet
Use ".arch_extension" ARM directive to specify the additional CPU features
This patch is in response to r223147 where the avaiable features are
computed based on ".cpu" directive. This will work clean for the standard
variants like cortex-a9. For custom variants which rely on standard cpu names
for assembly, the additional features of a CPU should be propagated. This can be
done via ".arch_extension" as long as the assembler supports it. The
implementation for krait along with unit test will be submitted in next patch.
[X86][MMX] Remove widening experimental flag from MMX tests.
Turns out that after the past MMX commits, we don't need to rely on this
flag to get better codegen for MMX. Also update the tests to become
triple neutral.
The latency for the WriteMULm class was set to 4, which is actually lower than the latency for WriteMULr (5).
A better estimate would be 4 added to WriteMULr, that is, 9.
Hal Finkel [Thu, 26 Feb 2015 14:22:41 +0000 (14:22 +0000)]
[InstCombine] Add a test for altivec load/store intrinsic simplification
InstCombine has logic to convert aligned Altivec load/store intrinsics into
regular loads and stores. Unfortunately, there seems to be no regression test
covering this behavior. Adding one...
Add `CHECK-SAME`, which requires that the pattern matches on the *same*
line as the previous `CHECK`/`CHECK-NEXT` -- in other words, no newline
is allowed in the skipped region. This is similar to `CHECK-NEXT`,
which requires exactly 1 newline in the skipped region.
My motivation is to simplify checking the long lines of LLVM assembly
for the new debug info hierarchy. This allows CHECK sequences like the
following:
Adam Nemet [Thu, 26 Feb 2015 04:39:09 +0000 (04:39 +0000)]
[LoopAccesses] Add command-line option for RuntimeMemoryCheckThreshold
Also remove the somewhat misleading initializers from
VectorizationFactor and VectorizationInterleave. They will get
initialized with the default ctor since no cl::init is provided.
Use the IRBuilder helpers for gc.statepoint and gc.result, instead of
coding the construction by hand. Note that the gc.statepoint IRBuilder
handles only CallInst, not InvokeInst; retain that part of hand-coding.
Eric Christopher [Thu, 26 Feb 2015 00:00:35 +0000 (00:00 +0000)]
Remove a FIXME.
Explanation: This function is in TargetLowering because it uses
RegClassForVT which would need to be moved to TargetRegisterInfo
and would necessitate moving isTypeLegal over as well - a massive
change that would just require TargetLowering having a TargetRegisterInfo
class member that it would use.
Eric Christopher [Thu, 26 Feb 2015 00:00:24 +0000 (00:00 +0000)]
Remove an argument-less call to getSubtargetImpl from TargetLoweringBase.
This required plumbing a TargetRegisterInfo through computeRegisterProperties
and into findRepresentativeClass which uses it for register class
iteration. This required passing a subtarget into a few target specific
initializations of TargetLowering.
Philip Reames [Wed, 25 Feb 2015 23:45:20 +0000 (23:45 +0000)]
[GC Docs] Update LangRef to link to Statepoint docs
Add a brief section linking to the experimental statepoint intrinsics analogous to the one we have linking to patchpoint.
While I'm here, cleanup some wording about what the gc "name" attribute actually means. It's not the name of a *collector* it's the name of the *strategy* which may be compatible with multiple collectors.
Justin Bogner [Wed, 25 Feb 2015 22:52:20 +0000 (22:52 +0000)]
InstrProf: Make the __llvm_profile_runtime_user symbol hidden
This symbol exists only to pull in the required pieces of the runtime,
so nothing ever needs to refer to it. Making it hidden avoids the
potential for issues with duplicate symbols when linking profiled
libraries together.
IR: Drop newline from AssemblyWriter::printMDNodeBody()
Remove a newline from `AssemblyWriter::printMDNodeBody()`, and add one
to `AssemblyWriter::writeMDNode()`. NFCI for assembly output.
However, this drops an inconsistent newline from `Metadata::print()`
when `this` is an `MDNode`. Now the newline added by `Metadata::dump()`
won't look so verbose.
Sanjay Patel [Wed, 25 Feb 2015 22:46:08 +0000 (22:46 +0000)]
only propagate equality comparisons of FP values that we are certain are non-zero
This is a follow-on to r227491 which tightens the check for propagating FP
values. If a non-constant value happens to be a zero, we would hit the same
bug as before.
Eric Christopher [Wed, 25 Feb 2015 22:41:30 +0000 (22:41 +0000)]
Move TargetLoweringBase::getTypeConversion to the .cpp file from
the .h file. It's used in only one place (other than recursively)
and there's no need to include it everywhere.
Saves almost 900k from total llvm object file size.
JF Bastien [Wed, 25 Feb 2015 22:30:51 +0000 (22:30 +0000)]
InstCombine: extract instead of shuffle when performing vector/array type punning
Summary: SROA generates code that isn't quite as easy to optimize and contains unusual-sized shuffles, but that code is generally correct. As discussed in D7487 the right place to clean things up is InstCombine, which will pick up the type-punning pattern and transform it into a more obvious bitcast+extractelement, while leaving the other patterns SROA encounters as-is.
It turns out we have a macro to ensure that debuggers can access
`dump()` methods. Use it. Hopefully this will prevent me (and others)
from committing crimes like in r223802 (search for /10000/, or just see
the fix in r224407).
Hal Finkel [Wed, 25 Feb 2015 21:36:59 +0000 (21:36 +0000)]
[PowerPC] Make LDtocL and friends invariant loads
LDtocL, and other loads that roughly correspond to the TOC_ENTRY SDAG node,
represent loads from the TOC, which is invariant. As a result, these loads can
be hoisted out of loops, etc. In order to do this, we need to generate
GOT-style MMOs for TOC_ENTRY, which requires treating it as a legitimate memory
intrinsic node type. Once this is done, the MMO transfer is automatically
handled for TableGen-driven instruction selection, and for nodes generated
directly in PPCISelDAGToDAG, we need to transfer the MMOs manually.
Also, we were not transferring MMOs associated with pre-increment loads, so do
that too.
Lastly, this fixes an exposed bug where R30 was not added as a defined operand of
UpdateGBR.
This problem was highlighted by an example (used to generate the test case)
posted to llvmdev by Francois Pichet.