[tsan] compile-time instrumentation: do not instrument a read if
a write to the same temp follows in the same BB.
Also add stats printing.
On Spec CPU2006 this optimization saves roughly 4% of instrumented reads
(which is 3% of all instrumented accesses):
Writes : 161216
Reads : 446458
Reads-before-write: 18295
Eric Christopher [Tue, 10 Apr 2012 18:18:10 +0000 (18:18 +0000)]
To ensure that we have more accurate line information for a block
don't elide the branch instruction if it's the only one in the block,
otherwise it's ok.
Jim Grosbach [Tue, 10 Apr 2012 17:31:55 +0000 (17:31 +0000)]
ARM fix cc_out operand handling for t2SUBrr instructions.
We were incorrectly conflating some add variants which don't have a
cc_out operand with the mirroring sub encodings, which do. Part of the
awesome non-orthogonality legacy of thumb1. Similarly, handling of
add/sub of an immediate was sometimes incorrectly removing the cc_out
operand for add/sub register variants.
Fix a dagcombine optimization which assumes that the vsetcc result type is always
of the same size as the compared values. This is ture for SSE/AVX/NEON but not
for all targets.
Modify the code that lowers shuffles to blends from using blendvXX to vblendXX.
blendv uses a register for the selection while vblend uses an immediate.
On sandybridge they still have the same latency and execute on the same execution ports.
Make a somewhat subtle change in the logic of block placement. Sometimes
the loop header has a non-loop predecessor which has been pre-fused into
its chain due to unanalyzable branches. In this case, rotating the
header into the body of the loop in order to place a loop exit at the
bottom of the loop is a Very Bad Idea as it makes the loop
non-contiguous.
I'm working on a good test case for this, but it's a bit annoynig to
craft. I should get one shortly, but I'm submitting this now so I can
begin the (lengthy) performance analysis process. An initial run of LNT
looks really, really good, but there is too much noise there for me to
trust it much.
Andrew Trick [Tue, 10 Apr 2012 05:14:42 +0000 (05:14 +0000)]
Fix 12513: Loop unrolling breaks with indirect branches.
Take this opportunity to generalize the indirectbr bailout logic for
loop transformations. CFG transformations will never get indirectbr
right, and there's no point trying.
Andrew Trick [Tue, 10 Apr 2012 02:25:24 +0000 (02:25 +0000)]
Added register unit sets to the target description.
This is a new algorithm that finds sets of register units that can be
used to model registers pressure. This handles arbitrary, overlapping
register classes. Each register class is associated with a (small)
list of pressure sets. These are the dimensions of pressure affected
by the register class's liveness.
Andrew Trick [Tue, 10 Apr 2012 02:25:21 +0000 (02:25 +0000)]
Added register unit weights to the target description.
This is a new algorithm that associates registers with weighted
register units to accuretely model their effect on register
pressure. This handles registers with multiple overlapping
subregisters. It is possible, but almost inconceivable that the
algorithm fails to find an exact solution for a target description. If
an exact solution cannot be found, an inexact, but reasonable solution
will be chosen.
Fix a long standing tail call optimization bug. When a libcall is emitted
legalizer always use the DAG entry node. This is wrong when the libcall is
emitted as a tail call since it effectively folds the return node. If
the return node's input chain is not the entry (i.e. call, load, or store)
use that as the tail call input chain.
Chad Rosier [Mon, 9 Apr 2012 20:32:02 +0000 (20:32 +0000)]
When performing a truncating store, it's possible to rearrange the data
in-register, such that we can use a single vector store rather then a
series of scalar stores.
Lang Hames [Mon, 9 Apr 2012 20:17:30 +0000 (20:17 +0000)]
Patch r153892 for PR11861 apparently broke an external project (see PR12493).
This patch restores TwoAddressInstructionPass's pre-r153892 behaviour when
rescheduling instructions in TryInstructionTransform. Hopefully this will fix
PR12493. To refix PR11861, lowering of INSERT_SUBREGS is deferred until after
the copy that unties the operands is emitted (this seems to be a more
appropriate fix for that issue anyway).
David Blaikie [Mon, 9 Apr 2012 16:29:35 +0000 (16:29 +0000)]
Fix accidentally constant conditions found by uncommitted improvements to -Wconstant-conversion.
A couple of cases where we were accidentally creating constant conditions by
something like "x == a || b" instead of "x == a || x == b". In one case a
conditional & then unreachable was used - I transformed this into a direct
assert instead.
Fix a bug in the lowering of broadcasts: ConstantPools need to use the target pointer type.
Move NormalizeVectorShuffle and LowerVectorBroadcast into X86TargetLowering.
Cleanup and relax a restriction on the matching of global offsets into
x86 addressing modes. This allows PIE-based TLS offsets to fit directly
into an addressing mode immediate offset, which is the last remaining
code quality issue from PR12380. With this patch, that PR is completely
fixed.
To understand why this patch is correct to match these offsets into
addressing mode immediates, break it down by cases:
1) 32-bit is trivially correct, and unmodified here.
2) 64-bit non-small mode is unchanged and never matches.
3) 64-bit small PIC code which is RIP-relative is handled specially in
the match to try to fit RIP into the base register. If it fails, it
now early exits. This behavior is unchanged by the patch.
4) 64-bit small non-PIC code which is not RIP-relative continues to work
as it did before. The reason these immediates are safe is because the
ABI ensures they fit in small mode. This behavior is unchanged.
5) 64-bit small PIC code which is *not* using RIP-relative addressing.
This is the only case changed by the patch, and the primary place you
see it is in TLS, either the win64 section offset TLS or Linux
local-exec TLS model in a PIC compilation. Here the ABI again ensures
that the immediates fit because we are in small mode, and any other
operations required due to the PIC relocation model have been handled
externally to the Wrapper node (extra loads etc are made around the
wrapper node in ISelLowering).
I've tested this as much as I can comparing it with GCC's output, and
everything appears safe. I discussed this with Anton and it made sense
to him at least at face value. That said, if there are issues with PIC
code after this patch, yell and we can revert it.
Fold 15 tiny test cases into a single file that implements the
comprehensive testing of TLS codegen for x86. Convert all of the ones
that were still using grep to use FileCheck. Remove some redundancies
between them.
Perhaps most interestingly expand the test cases so that they actually
fully list the instruction snippet being tested. TLS operations are
*very* narrowly defined, and so these seem reasonably stable. More
importantly, the existing test cases already were crazy fine grained,
expecting specific registers to be allocated. This just clarifies that
no *other* instructions are expected, and fills in some crucial gaps
that weren't being tested at all.
This will make any subsequent changes to TLS much more clear during
review.
Only have codegen turn fdiv by a constant into fmul by the reciprocal
when -ffast-math, i.e. don't just always do it if the reciprocal can
be formed exactly. There is already an IR level transform that does
that, and it does it more carefully.
Simplify code that tries to do vector extracts for shuffles when the mask width and the input vector widths don't match. No need to check the min and max are in range before calculating the start index. The range check after having the start index is sufficient. Also no need to check for an extract from the beginning differently.
Teach LLVM about a PIE option which, when enabled on top of PIC, makes
optimizations which are valid for position independent code being linked
into a single executable, but not for such code being linked into
a shared library.
I discussed the design of this with Eric Christopher, and the decision
was to support an optional bit rather than a completely separate
relocation model. Fundamentally, this is still PIC relocation, its just
that certain optimizations are only valid under a PIC relocation model
when the resulting code won't be in a shared library. The simplest path
to here is to expose a single bit option in the TargetOptions. If folks
have different/better designs, I'm all ears. =]
I've included the first optimization based upon this: changing TLS
models to the *Exec models when PIE is enabled. This is the LLVM
component of PR12380 and is all of the hard work.
Move the TLSModel information into the TargetMachine rather than hiding
in TargetLowering. There was already a FIXME about this location being
odd. The interface is simplified as a consequence. This will also make
it easier to change TLS models when compiling with PIE.
Benjamin Kramer [Sun, 8 Apr 2012 14:53:14 +0000 (14:53 +0000)]
EngineBuilder::create is expected to take ownership of the TargetMachine passed to it. Delete it on error or when we create an interpreter that doesn't need it.
Remove an over zealous assert. The assert was trying to catch places
where a chain outside of the loop block-set ended up in the worklist for
scheduling as part of the contiguous loop. However, asserting the first
block in the chain is in the loop-set isn't a valid check -- we may be
forced to drag a chain into the worklist due to one block in the chain
being part of the loop even though the first block is *not* in the loop.
This occurs when we have been forced to form a chain early due to
un-analyzable branches.
No test case here as I have no idea how to even begin reducing one, and
it will be hopelessly fragile. We have to somehow end up with a loop
header of an inner loop which is a successor of a basic block with an
unanalyzable pair of branch instructions. Ow. Self-host triggers it so
it is unlikely it will regress.
This at least gets block placement back to passing selfhost and the test
suite. There are still a lot of slowdown that I don't like coming out of
block placement, although there are now also a lot of speedups. =[ I'm
seeing swings in both directions up to 10%. I'm going to try to find
time to dig into this and see if we can turn this on for 3.1 as it does
a really good job of cleaning up after some loops that degraded with the
inliner changes.
Teach InstCombine to nuke a common alloca pattern -- an alloca which has
GEPs, bit casts, and stores reaching it but no other instructions. These
often show up during the iterative processing of the inliner, SROA, and
DCE. Once we hit this point, we can completely remove the alloca. These
were actually showing up in the final, fully optimized code in a bunch
of inliner tests I've been working on, and notably they show up after
LLVM finishes optimizing away all function calls involved in
hash_combine(a, b).
AVX2: Build splat vectors by broadcasting a scalar from the constant pool.
Previously we used three instructions to broadcast an immediate value into a
vector register.
On Sandybridge we continue to load the broadcasted value from the constant pool.
Bill Wendling [Sun, 8 Apr 2012 10:20:49 +0000 (10:20 +0000)]
Remove the 'Parent' pointer from the MDNodeOperand class.
An MDNode has a list of MDNodeOperands allocated directly after it as part of
its allocation. Therefore, the Parent of the MDNodeOperands can be found by
walking back through the operands to the beginning of that list. Mark the first
operand's value pointer as being the 'first' operand so that we know where the
beginning of said list is.
This saves a *lot* of space during LTO with -O0 -g flags.
Bill Wendling [Sun, 8 Apr 2012 10:16:43 +0000 (10:16 +0000)]
Allow subclasses of the ValueHandleBase to store information as part of the
value pointer by making the value pointer into a pointer-int pair with 2 bits
available for flags.
Turn avx2 vinserti128 intrinsic calls into INSERT_SUBVECTOR DAG nodes and remove patterns for selecting the intrinsic. Similar was already done for avx1.
Move vinsertf128 patterns near the instruction definitions. Add AddedComplexity to AVX2 vextracti128 patterns to give them priority over the integer versions of vextractf128 patterns.
1. Remove the part of r153848 which optimizes shuffle-of-shuffle into a new
shuffle node because it could introduce new shuffle nodes that were not
supported efficiently by the target.
2. Add a more restrictive shuffle-of-shuffle optimization for cases where the
second shuffle reverses the transformation of the first shuffle.
Convert floating point division by a constant into multiplication by the
reciprocal if converting to the reciprocal is exact. Do it even if inexact
if -ffast-math. This substantially speeds up ac.f90 from the polyhedron
benchmarks.
Perform partial SROA on the helper hashing structure. I really wish the
optimizers could do this for us, but expecting partial SROA of classes
with template methods through cloning is probably expecting too much
heroics. With this change, the begin/end pointer pairs which indicate
the status of each loop iteration are actually passed directly into each
layer of the combine_data calls, and the inliner has a chance to see
when most of the combine_data function could be deleted by inlining.
Similarly for 'length'.
We have to be careful to limit the places where in/out reference
parameters are used as those will also defeat the inliner / optimizers
from properly propagating constants.
With this change, LLVM is able to fully inline and unroll the hash
computation of small sets of values, such as two or three pointers.
These now decompose into essentially straight-line code with no loops or
function calls.
There is still one code quality problem to be solved with the hashing --
LLVM is failing to nuke the alloca. It removes all loads from the
alloca, leaving only lifetime intrinsics and dead(!!) stores to the
alloca. =/ Very unfortunate.
Fix ValueTracking to conclude that debug intrinsics are safe to
speculate. Without this, loop rotate (among many other places) would
suddenly stop working in the presence of debug info. I found this
looking at loop rotate, and have augmented its tests with a reduction
out of a very hot loop in yacr2 where failing to do this rotation costs
sometimes more than 10% in runtime performance, perturbing numerous
downstream optimizations.
This should have no impact on performance without debug info, but the
change in performance when debug info is enabled can be extreme. As
a consequence (and this how I got to this yak) any profiling of
performance problems should be treated with deep suspicion -- they may
have been wildly innacurate of debug info was enabled for profiling. =/
Just a heads up.
Bob Wilson [Sat, 7 Apr 2012 16:51:59 +0000 (16:51 +0000)]
Fix Thumb __builtin_longjmp with integrated assembler. <rdar://problem/11203543>
The tLDRr instruction with the last register operand set to the zero register
prints in assembly as if no register was specified, and the assembler encodes
it as a tLDRi instruction with a zero immediate. With the integrated assembler,
that zero register gets emitted as "r0", so we get "ldr rx, [ry, r0]" which
is broken. Emit the instruction as tLDRi with a zero immediate. I don't
know if there's a good way to write a testcase for this. Suggestions welcome.
Opportunities for follow-up work:
1) The asm printer should complain if a non-optional register operand is set
to the zero register, instead of silently dropping it.
2) The integrated assembler should complain in the same situation, instead of
silently emitting the operand as "r0".
Target/X86/MCTargetDesc/X86MCAsmInfo.cpp: Enable DwarfCFI (aka DW2) on Cygming.
Cygwin-1.7 supports dw2. Some recent mingw distros support one, too.
I have confirmed test-suite/SingleSource/Benchmarks/Shootout-C++/except.cpp can pass on Cygwin.
Sean Hunt [Sat, 7 Apr 2012 00:37:53 +0000 (00:37 +0000)]
Output UTF-8-encoded characters as identifier characters into assembly
by default.
This is a behaviour configurable in the MCAsmInfo. I've decided to turn
it on by default in (possibly optimistic) hopes that most assemblers are
reasonably sane. If this proves a problem, switching to default seems
reasonable.
I'm not sure if this is the opportune place to test, but it seemed good
to make sure it was tested somewhere.
Sean Callanan [Fri, 6 Apr 2012 18:21:09 +0000 (18:21 +0000)]
Fixed two leaks in the MC disassembler. The MC
disassembler requires a MCSubtargetInfo and a
MCInstrInfo to exist in order to initialize the
instruction printer and disassembler; however,
although the printer and disassembler keep
references to these objects they do not own them.
Previously, the MCSubtargetInfo and MCInstrInfo
objects were just leaked.
I have extended LLVMDisasmContext to own these
objects and delete them when it is destroyed.
Sink the collection of return instructions until after *all*
simplification has been performed. This is a bit less efficient
(requires another ilist walk of the basic blocks) but shouldn't matter
in practice. More importantly, it's just too much work to keep track of
all the various ways the return instructions can be mutated while
simplifying them. This fixes yet another crasher, reported by Daniel
Dunbar.
Make GVN's propagateEquality non-recursive. No intended functionality change.
The modifications are a lot more trivial than they appear to be in the diff!
Allow 256-bit shuffles to be split if a 128-bit lane contains elements from a single source. This is a rewrite of the 256-bit shuffle splitting code based on similar code from legalize types. Fixes PR12413.
Sink the return instruction collection until after we're done deleting
dead code, including dead return instructions in some cases. Otherwise,
we end up having a bogus poniter to a return instruction that blows up
much further down the road.
It turns out that this pattern is both simpler to code, easier to update
in the face of enhancements to the inliner cleanup, and likely cheaper
given that it won't add dead instructions to the list.
Thanks to John Regehr's numerous test cases for teasing this out.
We had special instructions for iOS because r9 is call-clobbered, but
that is represented dynamically by the register mask operands now, so
there is no need for the pseudo-instructions.
Jim Grosbach [Thu, 5 Apr 2012 23:51:24 +0000 (23:51 +0000)]
ARM: Don't form a t2LDRi8 or t2STRi8 with an offset of zero.
The load/store optimizer splits LDRD/STRD into two instructions when the
register pairing doesn't work out. For negative offsets in Thumb2, it uses
t2STRi8 to do that. That's fine, except for the case when the offset is in
the range [-4,-1]. In that case, we'll also form a second t2STRi8 with
the original offset plus 4, resulting in a t2STRi8 with a non-negative
offset, which ends up as if it were an STRT, which is completely bogus.
Similarly for loads.
No testcase, unfortunately, as any I've been able to construct is both large
and extremely fragile.
The empty 1-argument operator delete is for the benefit of the
destructor. A couple of spot checks of running yaml-bench under
valgrind against a few of the files under test/YAMLParser did
not reveal any leaks introduced by this change.