This is intended to be the minimal change needed to fix PR20354 ( http://llvm.org/bugs/show_bug.cgi?id=20354 ). The check for a vector operation was wrong; we need to check that the fabs itself is not a vector operation.
This patch will not generate the optimal code. A constant pool load and 'and' op will be generated instead of just returning a value that we can calculate in advance (as we do for the scalar case). I've put a 'TODO' comment for that here and expect to have that patch ready soon.
There is a very similar optimization that we can do in visitFNEG, so I've put another 'TODO' there and expect to have another patch for that too.
llvm-svn: 214670
sequence - AArch64 target support
This patch turns off madd/msub generation in the DAGCombiner and generates
them in the MachineCombiner instead. It replaces the original code sequence
with the combined sequence when it is beneficial to do so.
When there is no machine model support it always generates the madd/msub
instruction. This is true also when the objective is to optimize for code
size: when the combined sequence is shorter is always chosen and does not
get evaluated.
When there is a machine model the combined instruction sequence
is evaluated for critical path and resource length using machine
trace metrics and the original code sequence is replaced when it is
determined to be faster.
rdar://16319955
llvm-svn: 214669
sequence - target independent framework
When the DAGcombiner selects instruction sequences
it could increase the critical path or resource len.
For example, on arm64 there are multiply-accumulate instructions (madd,
msub). If e.g. the equivalent multiply-add sequence is not on the
crictial path it makes sense to select it instead of the combined,
single accumulate instruction (madd/msub). The reason is that the
conversion from add+mul to the madd could lengthen the critical path
by the latency of the multiply.
But the DAGCombiner would always combine and select the madd/msub
instruction.
This patch uses machine trace metrics to estimate critical path length
and resource length of an original instruction sequence vs a combined
instruction sequence and picks the faster code based on its estimates.
This patch only commits the target independent framework that evaluates
and selects code sequences. The machine instruction combiner is turned
off for all targets and expected to evolve over time by gradually
handling DAGCombiner pattern in the target specific code.
This framework lays the groundwork for fixing
rdar://16319955
llvm-svn: 214666
This makes EmitWindowsUnwindTables a virtual function and lowers the
implementation of the function to the X86WinCOFFStreamer. This method is a
target specific operation. This enables making the behaviour target dependent
by isolating it entirely to the target specific streamer.
llvm-svn: 214664
The frame information stored in this structure is driven by the requirements for
Windows NT unwinding rather than Windows 64 specifically. As a result, this
type can be shared across multiple architectures (ARM, AXP, MIPS, PPC, SH).
Rename this class in preparation for adding support for supporting unwinding
information for Windows on ARM.
Take the opportunity to constify the members as everything except the
ChainedParent is read-only. This required some adjustment to the label
handling.
llvm-svn: 214663
This slipped in in r214467, so something like
V_MOV_B32_e32 v0, ... is now printed with 2 spaces
between the instruction name and first operand.
llvm-svn: 214660
when let can do the same thing. Keep the 64bit variants as codegen-only.
While they have a different register class, the encoding is the same for
32bit and 64bit mode. Having both present would otherwise confuse the
disassembler.
llvm-svn: 214636
Darwin x86 asm comment prefix designed to work around GAS on that
platform. That makes the comment-matching of the test much more stable.
llvm-svn: 214629
lowering with a small addition to it and adding PSHUFB combining.
There is one obvious place in the new vector shuffle lowering where we
should form PSHUFBs directly: when without them we will unpack a vector
of i8s across two different registers and do a potentially 4-way blend
as i16s only to re-pack them into i8s afterward. This is the crazy
expensive fallback path for i8 shuffles and we can just directly use
pshufb here as it will always be cheaper (the unpack and pack are
two instructions so even a single shuffle between them hits our
three instruction limit for forming PSHUFB).
However, this doesn't generate very good code in many cases, and it
leaves a bunch of common patterns not using PSHUFB. So this patch also
adds support for extracting a shuffle mask from PSHUFB in the X86
lowering code, and uses it to handle PSHUFBs in the recursive shuffle
combining. This allows us to combine through them, combine multiple ones
together, and generally produce sufficiently high quality code.
Extracting the PSHUFB mask is annoyingly complex because it could be
either pre-legalization or post-legalization. At least this doesn't have
to deal with re-materialized constants. =] I've added decode routines to
handle the different patterns that show up at this level and we dispatch
through them as appropriate.
The two primary test cases are updated. For the v16 test case there is
still a lot of room for improvement. Since I was going through it
systematically I left behind a bunch of FIXME lines that I'm hoping to
turn into ALL lines by the end of this.
llvm-svn: 214628
of normally binary shuffle instructions like PUNPCKL and MOVLHPS.
This detects cases where a single register is used for both operands
making the shuffle behave in a unary way. We detect this and adjust the
mask to use the unary form which allows the existing DAG combine for
shuffle instructions to actually work at all.
As a consequence, this uncovered a number of obvious bugs in the
existing DAG combine which are fixed. It also now canonicalizes several
shuffles even with the existing lowering. These typically are trying to
match the shuffle to the domain of the input where before we only really
modeled them with the floating point variants. All of the cases which
change to an integer shuffle here have something in the integer domain, so
there are no more or fewer domain crosses here AFAICT. Technically, it
might be better to go from a GPR directly to the floating point domain,
but detecting floating point *outputs* despite integer inputs is a lot
more code and seems unlikely to be worthwhile in practice. If folks are
seeing domain-crossing regressions here though, let me know and I can
hack something up to fix it.
Also as a consequence, a bunch of missed opportunities to form pshufb
now can be formed. Notably, splats of i8s now form pshufb.
Interestingly, this improves the existing splat lowering too. We go from
3 instructions to 1. Yes, we may tie up a register, but it seems very
likely to be worth it, especially if splatting the 0th byte (the
common case) as then we can use a zeroed register as the mask.
llvm-svn: 214625
so using a single helper which adds operands back onto the worklist.
Several places didn't rigorously do this but a couple already did.
Factoring them together and doing it rigorously is important to delete
things recursively early on in the combiner and get a chance to see
accurate hasOneUse values. While no existing test cases change, an
upcoming patch to add DAG combining logic for PSHUFB requires this to
work correctly.
llvm-svn: 214623
during DAGCombine in certain circumstances. Unfortunately, the circumstances required
to trigger the issue seem to require a pretty specific interaction of DAGCombines,
and I haven't been able to find a testcase that reproduces on X86, ARM, or AArch64.
The functionality added here is replicated in essentially every other DAG combine,
so it seems pretty obviously correct.
llvm-svn: 214622
expanding pseudo LOAD_STATCK_GUARD using instructions that are normally used
in pic mode. This patch fixes the bug.
<rdar://problem/17886592>
llvm-svn: 214614
This is something that I have found to be very useful in my work and I
wanted to contribute it back to the community since several people in
the past have asked me for something along these lines. (Jakob, I know
this has been a while coming ; )]
The way you use this is you create a script that takes in as its first
argument a count. The script passes into LLVM the count via a command
line flag that disables a pass after LLVM has run after the pass has
run for count number of times. Then the script invokes a test of some
sort and indicates whether LLVM successfully compiled the test via the
scripts exit status. Then you invoke bisect as follows:
bisect --start=<start_num> --end=<end_num> ./script.sh "%(count)s"
And bisect will continually call ./script.sh with various counts using
the exit status to determine success and failure.
llvm-svn: 214610
Summary:
This patch add a --show-xfail flag. If this flag is specified then each xfail test will be printed to output.
When it is not given xfail tests are ignored. Ignoring xfail tests is the current behavior.
This flag is meant to mirror the --show-unsupported flag that was recently added.
Reviewers: ddunbar, EricWF
Reviewed By: EricWF
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D4750
llvm-svn: 214609
makes a mess of the lit output when they ultimately fail.
The 2012-10-02-DAGCycle test is really frustrating because the *only*
explanation for what it is testing is a rdar link. I would really rather
that rdar links (which are not public or part of the open source
project) were not committed to the source code. Regardless, the actual
problem *must* be described as the rdar link is completely opaque. The
fact that this test didn't check for any particular output further
exacerbates the inability of any other developer to debug failures.
The mem-promote-integers test has nice comments and *seems* to be
a great test for our lowering... except that we don't actually check
that any of the generated code is correct or matches some pattern. We
just avoid crashing. It would be great to go back and populate this test
with the actual expectations.
llvm-svn: 214605
Instead of creating global variables for source locations and global names,
just create metadata nodes and strings. They will be transformed into actual
globals in the instrumentation pass (if necessary). This approach is more
flexible:
1) we don't have to ensure that our custom globals survive all the optimizations
2) if globals are discarded for some reason, we will simply ignore metadata for them
and won't have to erase corresponding globals
3) metadata for source locations can be reused for other purposes: e.g. we may
attach source location metadata to alloca instructions and provide better descriptions
for stack variables in ASan error reports.
No functionality change.
llvm-svn: 214604
introduced during legalization. This pattern is based on other patterns
in the legalizer that I changed in the same way. Now, the legalizer
eagerly collects its garbage when necessary so that we can survive
leaving such nodes around for it.
Instead, we add an assert to make sure the node will be correctly
handled by that layer.
llvm-svn: 214602
When the cost model determines vectorization is not possible/profitable these remarks print an analysis of that decision.
Note that in selectVectorizationFactor() we can assume that OptForSize and ForceVectorization are mutually exclusive.
Reviewed by Arnold Schwaighofer
llvm-svn: 214599
Updated `verify-uselistorder` to more than double the number of use-list
orders it checks.
- Every time it verifies an order, it then reverses the order and
verifies again.
- It now verifies the initial order, before running any shuffles.
Changed the default to `-num-shuffles=1`, since this is already four
checks, and after r214584 shuffling is guaranteed to make a new order.
This is part of PR5680.
llvm-svn: 214596
`shuffleUseLists()` is only used in `verify-uselistorder`, so move it
there to avoid bloating other executables. As a drive-by, update some
of the header docs.
This is part of PR5680.
llvm-svn: 214592
This updates the instrumentation based profiling format so that when
we have multiple functions with the same name (but different function
hashes) we keep all of them instead of rejecting the later ones.
There are a number of scenarios where this can come up where it's more
useful to keep multiple function profiles:
* Name collisions in unrelated libraries that are profiled together.
* Multiple "main" functions from multiple tools built against a common
library.
* Combining profiles from different build configurations (ie, asserts
and no-asserts)
The profile format now stores the number of counters between the hash
and the counts themselves, so that multiple sets of counts can be
stored. Since this is backwards incompatible, I've bumped the format
version and added some trivial logic to skip this when reading the old
format.
llvm-svn: 214585
`parseBitcodeFile()` uses the generic `getLazyBitcodeFile()` function as
a helper. Since `parseBitcodeFile()` isn't actually lazy -- it calls
`MaterializeAllPermanently()` -- bypass the unnecessary call to
`materializeForwardReferencedFunctions()` by extracting out a common
helper function. This removes the last of the use-list churn caused by
blockaddresses.
This highlights that we can't reproduce use-list order of globals and
constants when parsing lazily -- but that's necessarily out of scope.
When we're parsing lazily, we never have all the functions in memory, so
the use-lists of globals (and constants that reference globals) are
always incomplete.
This is part of PR5680.
llvm-svn: 214581
Stop using ST registers for function returns and inline-asm instructions and use
FP registers instead. This allows removing a large amount of code in the
stackifier pass that was needed to track register liveness and handle copies
between ST and FP registers and function calls returning floating point values.
It also fixes a bug which manifests when an ST register defined by an
inline-asm instruction was live across another inline-asm instruction, as shown
in the following sequence of machine instructions:
1. INLINEASM <es:frndint> $0:[regdef], %ST0<imp-def,tied5>
2. INLINEASM <es:fldcw $0>
3. %FP0<def> = COPY %ST0
<rdar://problem/16952634>
llvm-svn: 214580
variables (for example, by-value struct arguments passed in registers, or
large integer values split across several smaller registers).
On the IR level, this adds a new type of complex address operation OpPiece
to DIVariable that describes size and offset of a variable fragment.
On the DWARF emitter level, all pieces describing the same variable are
collected, sorted and emitted as DWARF expressions using the DW_OP_piece
and DW_OP_bit_piece operators.
http://reviews.llvm.org/D3373
rdar://problem/15928306
What this patch doesn't do / Future work:
- This patch only adds the backend machinery to make this work, patches
that change SROA and SelectionDAG's type legalizer to actually create
such debug info will follow. (http://reviews.llvm.org/D2680)
- Making the DIVariable complex expressions into an argument of dbg.value
will reduce the memory footprint of the debug metadata.
- The sorting/uniquing of pieces should be moved into DebugLocEntry,
to facilitate the merging of multi-piece entries.
llvm-svn: 214576
fromulation of the node, which isn't really the desired behavior from
within the combiner or legalizer, but is necessary within ISel. I've
added a hopefully helpful comment and fixed the only two places where
this took place.
Yet another step toward the combiner and legalizer not needing to use
update listeners with virtual calls to manage the worklists behind
legalization and combining.
llvm-svn: 214574
Now that we can reliably handle forward references to `BlockAddress`
(r214563), change the mechanics to simplify predicting use-list order.
Previously, we created dummy `GlobalVariable`s to represent block
addresses. After every function was materialized, we'd go through any
forward references to its blocks and RAUW them with a proper
`BlockAddress` constant. This causes some (potentially a lot of)
unnecessary use-list churn, since any constant expression that it's a
part of will need to be rematerialized as well.
Instead, pre-construct a `BasicBlock` immediately -- without attaching
it to its (empty) `Function` -- and use that to construct a
`BlockAddress`. This constant will not have to be regenerated. When
the function body is parsed, hook this pre-constructed basic block up
in the right place using `BasicBlock::insertInto()`.
Both before and after this change, the IR is temporarily in an invalid
state that gets resolved when `materializeForwardReferencedFunctions()`
gets called.
This is a prep commit that's part of PR5680, but the only functionality
change is the reduction of churn in the constant pool.
llvm-svn: 214570
Users keep emailing us about the difficulties of getting LD_LIBRARY_PATH
into their environment, which should be completely unecessary. Try to
strengthen the rpath recommentation by putting in an example cmake
invocation.
Speaking of which, we might want to make CMake the recommended build
system in GettingStarted.html.
llvm-svn: 214565
Although unlinked `BasicBlock`s can be created, there's currently no way
to insert them into `Function`s after the fact. In particular,
`moveAfter()` and `moveBefore()` require that the basic block is already
linked.
Extract the logic for initially linking a `BasicBlock` out of the
constructor and into a member function that can be used for lazy
insertion.
- Asserts that the basic block is currently unlinked.
- Matches the logic of the constructor.
- Changed the constructor to use it since the logic matches.
This is needed in a follow-up commit for PR5680.
llvm-svn: 214563
so that we can use it to get the old-style JIT out of the subtarget.
This code should be removed when the old-style JIT is removed
(imminently).
llvm-svn: 214560
`BlockAddress`es are interesting in that they can reference basic blocks
from *outside* the block's function. Since basic blocks are not global
values, this presents particular challenges for lazy parsing.
One corner case was found in PR11677 and fixed in r147425. In that
case, a global variable references a block address. It's necessary to
load the relevant function to resolve the forward reference before doing
anything with the module.
By inspection, I found (and have fixed here) two other cases:
- An instruction from one function references a block address from
another function, and only the first function is lazily loaded.
I fixed this the same way as PR11677: by eagerly loading the
referenced function.
- A function whose block address is taken is dematerialized, leaving
invalid references to it.
I fixed this by refusing to dematerialize functions whose block
addresses are taken (if you have to load it, you can't unload it).
llvm-svn: 214559
Rewrite the single unit test in `BitReaderTest` so that it's easier to
add more tests.
- Parse from an assembly string rather than using API.
- Use more helper functions.
- Use a separate context for the module on the other side.
Aside from relying on the assembly parser, there's no functionality
change intended.
llvm-svn: 214556
This is consistent with how we parse them in a standalone .s file, and
inline assembly shouldn't differ.
This fixes errors about requiring more registers than available in
cases like this:
void f();
void __declspec(naked) g() {
__asm pusha
__asm call f
__asm popa
__asm ret
}
There are no registers available to pass the address of 'f' into the asm
blob. The asm should now directly call 'f'.
Tests will land in Clang shortly.
llvm-svn: 214550
This lifts the (very few) places the legalizer would delete dead nodes
into the outer loop around the legalizer. This is significantly simpler
because it doesn't require the legalizer itself to manage the iterator
validity, and it doesn't require the legalizer to be a DAG update
listener in order to remove things from the legalized set. It also makes
the interface much less contrived for the case of the legalizer running
inside the last phase of DAG combining.
I'm working on centralizing the deletion of nodes during both legalizing
and combining as much as possible. My hope is to remove the need for DAG
update listeners from the combiner next, which would remove a costly
virtual dispatch chain on every deletion. This in turn should allow us
to more aggressively delete DAG nodes during combining which will in
turn allow us to combine more aggressively by exposing the actual nodes
which have single users to the combine phases.
llvm-svn: 214546
This patch adds code to emits the StackMap section on ELF systems. This section is required to support llvm.experimental.stackmap and llvm.experimental.patchpoint intrinsics.
Reviewers: ributzka, echristo
Differential Revision: http://reviews.llvm.org/D4574
llvm-svn: 214538
Add branch weights to branch instructions, so that the following passes can
optimize based on it (i.e. basic block ordering).
Fixes <rdar://problem/17887137>.
llvm-svn: 214537
This change adds code to explicitly mark a function which requires runtime stack realignment as not having a fixed frame size in the StackMap section. As it happens, this is not actually a functional change. The size that would be reported without the check is also "-1", but as far as I can tell, that's an accident. The code change makes this explicit.
Note: There's a separate bug in handling of stackmaps and patchpoints in functions which need dynamic frame realignment. The current code assumes that offsets can be calculated from RBP, but realigned frames must use RSP. (There's a variable gap between RBP and the spill slots.) This change set does not address that issue.
Reviewers: atrick, ributzka
Differential Revision: http://reviews.llvm.org/D4572
llvm-svn: 214534
This is a followup patch for r214366, which added the same behavior to the
AArch64 and X86 FastISel code. This fix reproduces the already existing
behavior of SelectionDAG in FastISel.
llvm-svn: 214531
The tbz/tbnz checks the sign bit to convert
op w1, w1, w10
cmp w1, #0
b.lt .LBB0_0
to
op w1, w1, w10
tbnz w1, #31, .LBB0_0
Differential Revision: http://reviews.llvm.org/D4440
llvm-svn: 214518
Found by inspection while looking at PR20280: code would mark slots
in the parameter save area where a byval parameter is passed as
"immutable". This is not correct since code is allowed to modify
byval parameters in place in the parameter save area.
llvm-svn: 214517
Note: The current code in DecodeMSRMask() rejects the unpredictable A/R MSR mask '0000' with Fail. The code in the patch follows this style and rejects unpredictable M-class MSR masks also with Fail (instead of SoftFail). If SoftFail is preferred in this case then additional changes to ARMInstPrinter (to print non-symbolic masks) and ARMAsmParser (to parse non-symbolic masks) will be needed.
Patch by Petr Pavlu!
llvm-svn: 214505
The ARM ARM prohibits LDRB/LDRSB instructions with writeback into the destination register. With this commit this constraint is now enforced and we stop assembling LDRH/LDRSH instructions with unpredictable behavior.
llvm-svn: 214500
The ARM ARM prohibits LDRH/LDRSH instructions with writeback into the source register. With this commit this constraint is now enforced and we stop assembling LDRH/LDRSH instructions with unpredictable behavior.
llvm-svn: 214499
The ARM ARM prohibits LDR instructions with writeback into the destination register. With this commit this constraint is now enforced and we stop assembling LDR instructions with unpredictable behavior.
llvm-svn: 214498
Summary:
Big-endian mode was not correctly adjusting the offset for types smaller
than an ABI slot.
Fixes PR19612
Reviewers: dsanders
Reviewed By: dsanders
Subscribers: sstankovic, llvm-commits
Differential Revision: http://reviews.llvm.org/D4556
llvm-svn: 214493
"Create a default symver on Linux like ELF OSes."
Fails the build under Debian with ld.gold:
/usr/bin/ld.gold: --default-symver: unknown option
llvm-svn: 214482
Altivec vector loads on PowerPC have an interesting property: They always load
from an aligned address (by rounding down the address actually provided if
necessary). In order to generate an actual unaligned load, you can generate two
load instructions, one with the original address, one offset by one vector
length, and use a special permutation to extract the bytes desired.
When this was originally implemented, I generated these two loads using regular
ISD::LOAD nodes, now marked as aligned. Unfortunately, there is a problem with
this:
The alignment of a load does not contribute to its identity, and SDNodes
are uniqued. So, imagine that we have some unaligned load, L1, that is not
aligned. The routine will create two loads, L1(aligned) and (L1+16)(aligned).
Further imagine that there had already existed a load (L1+16)(unaligned) with
the same chain operand as the load L1. When (L1+16)(aligned) is created as part
of the lowering of L1, this load *is* also the (L1+16)(unaligned) node, just
now marked as aligned (because the new alignment overwrites the old). But the
original users of (L1+16)(unaligned) now get the data intended for the
permutation yielding the data for L1, and (L1+16)(unaligned) no longer exists
to get its own permutation-based expansion. This was PR19991.
A second potential problem has to do with the MMOs on these loads, which can be
used by AA during instruction scheduling to break chain-based dependencies. If
the new "aligned" loads get the MMO from the original unaligned load, this does
not represent the fact that it will load data from below the original address.
Normally, this would not matter, but this load might be combined with another
load pair for a previous vector, and then the dependency on the otherwise-
ignored lower bytes can matter.
To fix both problems, instead of generating the necessary loads using regular
ISD::LOAD instructions, ppc_altivec_lvx intrinsics are used instead. These are
provided with MMOs with a conservative address range.
Unfortunately, I no longer have a failing test case (since PR19991 was
reported, other changes in CodeGen have forced this bug back into hiding it
again). Nevertheless, this should fix the underlying problem.
llvm-svn: 214481
ADDS and SUBS cannot encode negative immediates or immediates larger than 12bit.
This fix checks if the immediate version can be used under this constraints and
if we can convert ADDS to SUBS or vice versa to support negative immediates.
Also update the test cases to test the immediate versions.
llvm-svn: 214470
When generating unaligned vector loads, we need to search for other loads or
stores nearby offset by one vector width. If we find one, then we know that we
can safely generate another aligned load at that address. Otherwise, we must
generate the next load using an offset of the vector width minus one byte (so
we don't read off the end of the allocation if the base unaligned address
happened to be aligned at runtime). We had previously done this using only
other vector loads and stores, but did not consider the PowerPC-specific vector
load/store intrinsics. Now we'll also consider vector intrinsics. By itself,
this change is a feature enhancement, but is a necessary step toward fixing the
underlying problem behind PR19991.
llvm-svn: 214469
This improves the diagnostics from the regular assembler, but more
importantly it fixes an assertion when parsing inline assembly. Test
landing in Clang.
llvm-svn: 214468
Abs/neg folding has moved out of foldOperands and into the instruction
selection phase using complex patterns. As a consequence of this
change, we now prefer to select the 64-bit encoding for most
instructions and the modifier operands have been dropped from
integer VOP3 instructions.
llvm-svn: 214467
This is useful for cases when stand-alone patterns are preferred to the
patterns included in the instruction definitions. Instead of requiring
that stand-alone patterns set a larger AddedComplexity value, which
can be confusing to new developers, the allows us to reduce the
complexity of the included patterns to achieve the same result.
There will be test cases for this added to the R600 backend in a
future commit.
llvm-svn: 214466
We were incorrectly assuming that all VOP2 instructions can read SGPRs
in Src0, but this is not true for instructions that read carry-in from
VCC.
The old logic has been replaced with new logic which checks the defined
register classes of the VOP2 instruction to determine whether or not to
legalize the operands.
llvm-svn: 214465
This allows assembling the two new instructions, encls and enclu for the
SKX processor model.
Note the diffs are a bigger than what might think, but to fit the new
MRM_CF and MRM_D7 in things in the right places things had to be
renumbered and shuffled down causing a bit more diffs.
rdar://16228228
llvm-svn: 214460
If INTRINSIC_W_CHAIN and INTRINSIC_VOID are MemIntrinsicSDNodes, and a
MemIntrinsicSDNode is a MemSDNode, then INTRINSIC_W_CHAIN and INTRINSIC_VOID
must be MemSDNodes too.
Noticed by inspection.
llvm-svn: 214452
Currently when DAGCombine converts loads feeding a switch into a switch of
addresses feeding a load the new load inherits the isInvariant flag of the left
side. This is incorrect since invariant loads can be reordered in cases where it
is illegal to reoarder normal loads.
This patch adds an isInvariant parameter to getExtLoad() and updates all call
sites to pass in the data if they have it or false if they don't. It also
changes the DAGCombine to use that data to make the right decision when
creating the new load.
llvm-svn: 214449
The current remark is ambiguous and makes it sounds like explicitly specifying vectorization will allow the loop to be vectorized. This is not the case. The improved remark directs the user to -Rpass-analysis=loop-vectorize to determine the cause of the pass-miss.
Reviewed by Arnold Schwaighofer`
llvm-svn: 214445
This is a follow-up to the activity in the bug at
http://llvm.org/bugs/show_bug.cgi?id=18663 . The underlying issue has
to do with how the KILL pseudo-instruction is handled. I defer to
Hal/Jakob/Uli for additional details and background.
This will disable the (bad?) assert, add an associated fixme comment,
and add a pair of tests.
The code change and the pr18663-2.ll test are copied from the referenced
bug. That test does not immediately fail in my environment, but I have
added the pr18663.ll test which does.
(Comment from Hal)
to provide everyone else with some context, this assert was not bad when
it was written. At that time, we only generated KILL pseudo instructions
around subregister copies. This logic, unfortunately, had its own problems.
In r199797, the relevant logic in MachineCopyPropagation was replaced to
generate KILLs for other kinds of copies too. This change in semantics broke
this now-problematic assumption in AggressiveAntiDepBreaker. The
AggressiveAntiDepBreaker really needs a proper cleanup to deal with the
change, but removing the assert (which just allows the function to return
false) is a safe conservative behavior, and should do for the time being.
llvm-svn: 214429
It seems that when I fixed this, almost exactly a year ago, I did not quite do
it correctly. When we have duplicate block predecessors, we can indeed not have
different incoming values for the same block, but we *must* have duplicate
entries. So, instead of skipping the duplicates, we explicitly add the
duplicate incoming values.
Fixes PR20442.
llvm-svn: 214423
Correctly sort self-users (such as PHI nodes). I added a targeted test
in `test/Bitcode/use-list-order.ll` and the final missing RUN line to
tests in `test/Assembly`.
This is part of PR5680.
llvm-svn: 214417
Switch array type shadow from a single integer to
an array of integers (i.e. make it per-element).
This simplifies instrumentation of extractvalue and fixes PR20493.
llvm-svn: 214398
We can only propagate the nsw bits if both subtraction instructions are
marked with the appropriate bit.
N.B. We only propagate the nsw bit in InstCombine because the nuw case
is already handled in InstSimplify.
This fixes PR20189.
llvm-svn: 214385
If the NUW bit is set for 0 - Y, we know that all values for Y other
than 0 would produce a poison value. This allows us to replace (0 - Y)
with 0 in the expression (X - (0 - Y)) which will ultimately leave us
with X.
This partially fixes PR20189.
llvm-svn: 214384
This commit updates the existing SelectionDAG tests for the stackmap and patchpoint
intrinsics and enables FastISel testing. It also splits up the tests into separate
files, due to different codegen between SelectionDAG and FastISel.
llvm-svn: 214382
Currently the large code model for MachO uses the GOT to make function calls.
Emit the required adrp and ldr instructions to load the address from the GOT.
Related to <rdar://problem/17733076>.
llvm-svn: 214381
Instead allow the variable to be declared, but don't attach an initializer. This allows more than a single error to be emitted before we exit.
Test case to follow soon in another patch.
llvm-svn: 214375
This is currently for assigning from one bit init to another. It can easily be extended to other types.
Test to follow soon in another patch.
llvm-svn: 214374
Since initializers of GlobalValues are being assigned IDs before
GlobalValues themselves, explicitly exclude GlobalValues from the
constant pool. Added targeted test in `test/Bitcode/use-list-order.ll`
and added two more RUN lines in `test/Assembly`.
This is part of PR5680.
llvm-svn: 214368
This fixes a mistake where I accidentially dropped the upper 32bit of a
64bit pointer during FastISel lowering of the patchpoint intrinsic.
llvm-svn: 214367
UNDEF arguments are not ment to be touched - especially for the webkit_js
calling convention. This fix reproduces the already existing behavior of
SelectionDAG in FastISel.
llvm-svn: 214366
Add RUN line for `verify-uselistorder` to every test in `test/Assembly`,
unless it's a negative check (assembler rejects it) or verification
fails.
There are three files that verification fails on (so I've left out the
RUN lines):
- 2002-08-22-DominanceProblem.ll
- ConstantExprFold.ll
- ConstantExprFoldCast.ll
This is part of PR5680.
llvm-svn: 214365
In some places we've been using different suffixes for the different
file formats involved in instrprof, but in others we've just
ambiguously used .profdata. Update the test files to indicate the
types of file more obviously.
No functional change.
llvm-svn: 214357
Before this patch we had
@a = weak global ...
but
@b = alias weak ...
The patch changes aliases to look more like global variables.
Looking at some really old code suggests that the reason was that the old
bison based parser had a reduction for alias linkages and another one for
global variable linkages. Putting the alias first avoided the reduce/reduce
conflict.
The days of the old .ll parser are long gone. The new one parses just "linkage"
and a later check is responsible for deciding if a linkage is valid in a
given context.
llvm-svn: 214355
This improves the code generation for the XALU intrinsics when the
condition is feeding a select instruction.
This also updates and enables the XALU unit tests for FastISel.
This fixes <rdar://problem/17831117>.
llvm-svn: 214350
This improves the code generation for the XALU intrinsics when the
condition is feeding a branch instruction.
This is related to <rdar://problem/17831117>.
llvm-svn: 214349
This commit adds support for the {s|u}{add|sub|mul}.with.overflow intrinsics.
The unit tests for FastISel will be enabled in a later commit, once there is
also branch and select folding support.
This is related to <rdar://problem/17831117>.
llvm-svn: 214348
Currently the shift-immediate versions are not supported by tblgen and
hopefully this can be later removed, once the required support has been
added to tblgen.
llvm-svn: 214345
While we can already transform A | (A ^ B) into A | B, things get bad
once we have (A ^ B) | (A ^ B ^ Cst) because reassociation will morph
this into (A ^ B) | ((A ^ Cst) ^ B). Our existing patterns fail once
this happens.
To fix this, we add a new pattern which looks through the tree of xor
binary operators to see that, in fact, there exists a redundant xor
operation.
What follows bellow is a correctness proof of the transform using CVC3.
$ cat t.cvc
A, B, C : BITVECTOR(64);
QUERY BVXOR(A, B) | BVXOR(BVXOR(B, C), A) = BVXOR(A, B) | C;
QUERY BVXOR(BVXOR(A, C), B) | BVXOR(A, B) = BVXOR(A, B) | C;
QUERY BVXOR(A, B) & BVXOR(BVXOR(B, C), A) = BVXOR(A, B) & ~C;
QUERY BVXOR(BVXOR(A, C), B) & BVXOR(A, B) = BVXOR(A, B) & ~C;
$ cvc3 < t.cvc
Valid.
Valid.
Valid.
Valid.
llvm-svn: 214342
The lifetime intrinsics need some work in order to make it clear which
optimizations are or are not valid.
For now dropping this optimization avoids a miscompilation.
Patch by Björn Steinbrink.
llvm-svn: 214336
This patch adds an explicit triple to the test case introduced by r214322. This
should fix build failueres that are occuring on bots that are cross building.
llvm-svn: 214330
Someone asked about this on IRC the other day, and I couldn't
find the magic prefix documented anywhere.
Differential Revision: http://reviews.llvm.org/D4728
llvm-svn: 214329
DAGCombine may choose to rewrite graphs where two loads feed a select into
graphs where a select of two addresses feed a load. While it sanity checks the
loads to make sure they are broadly equivalent it currently just uses the
alignment restriction of the left node. In cases where the right node has
stronger alignment requiresment this may lead to bad codegen, such as generating
an aligned load where an unaligned load is required. This patch makes the
combine generate a load with an alignment that is the same as whichever is more
restrictive of the two alignments.
Tests included.
rdar://17762530
llvm-svn: 214322
When predicting use-list order, we visit functions in reverse order
followed by `GlobalValue`s and write out use-lists at the first
opportunity. In the reader, this will translate to *after* the last use
has been added.
For this to work, we actually need to descend into `GlobalValue`s.
Added a targeted test in `use-list-order.ll` and `RUN` lines to the
newly passing tests in `test/Bitcode`.
There are two remaining failures in `test/Bitcode`:
- blockaddress.ll: I haven't thought through how to model the way
block addresses change the order of use-lists (or how to work around
it).
- metadata-2.ll: There's an old-style `@llvm.used` global array here
that I suspect the .ll parser isn't upgrading properly. When it
round-trips through bitcode, the .bc reader *does* upgrade it, so
the extra variable (`i8* null`) has an extra use, and the shuffle
vector doesn't match.
I think the fix is to upgrade old-style global arrays (or reject
them?) in the .ll parser.
This is part of PR5680.
llvm-svn: 214321
Turns out `parseBitcodeFile()` does *not* take ownership of the buffer.
This was already clear in the header docs, but I obviously didn't read
them (having noticed that it gets stored in a `unique_ptr<>`).
llvm-svn: 214313
the jump instruction table pass. First, the verifier is already built
into all the tools. The test case is adapted to just run llvm-as
demonstrating that we still catch the broken module. Second, the
verifier is *extremely* slow. This was responsible for very significant
compile time regressions.
If you have deployed a Clang binary anywhere from r210280 to this
commit, you really want to re-deploy.
llvm-svn: 214287
We now (1) correctly decode the branch immediate, (2) modify the immediate to
corretly treat it as PC-rel, and (3) properly populate the stub entry.
Previously we had been doing each of these wrong.
<rdar://problem/17750739>
llvm-svn: 214285
neverHasSideEffects is deprecated, and hasSideEffects = 0 is already
set on the base classes of the basic ALU instruction classes. The
base classes also already set mayLoad = 0 and mayStore = 0
llvm-svn: 214283
This commit fixes undefined behaviour that caused the revert in r214249.
The problem was two unsequenced operations on a `DenseMap<>`, giving
different behaviour in GCC and Clang. This:
DenseMap<T*, unsigned> DM;
for (auto &X : ...)
DM[&X] = DM.size() + 1;
should have been:
DenseMap<T*, unsigned> DM;
for (auto &X : ...) {
unsigned Size = DM.size();
DM[&X] = Size + 1;
}
Until r214242, this difference between compilers didn't matter. In
r214242, `OrderMap::LastGlobalValueID` was introduced and compared
against IDs, which in GCC were off-by-one my expectations.
llvm-svn: 214270
We can treat ds_read2_* as a single offset if the offsets are adjacent.
No test since emission of read2 instructions for partially
aligned loads isn't implemented yet.
llvm-svn: 214269
Fix the missing case in ScalarizeVectorResult() that was exposed with
libclcore.bc in Android.
Differential Revision: http://reviews.llvm.org/D4645
llvm-svn: 214266
use in -verify mode.
This patch adds three hidden command line options to llvm-rtdyld:
-target-addr-start <start-addr> : Specify the start of the virtual address
space on the phony target.
-target-addr-end <end-addr> : Specify the end of the virtual address space
on the phony target.
-target-section-sep <sep> : Specify the separation (in bytes) between the
end of one section and the start of the next.
These options automatically default to sane values for the target platform. In
particular, they allow narrow (e.g. 32-bit, 16-bit) targets to be tested from
wider (e.g. 64-bit, 32-bit) hosts without overflowing pointers.
The section separation option defaults to zero, but can be set to a large number
(e.g. 1 << 32) to force large separations between sections in order to
stress-test large-code-model code.
llvm-svn: 214255
r214242 was subtle enough it really deserves a targeted test with
comments. This adds some global variables that trigger the relevant
code path. Sorry this wasn't committed with the fix.
llvm-svn: 214243
To avoid unnecessary forward references, the reader doesn't process
initializers of `GlobalValue`s until after the constant pool has been
processed, and then in reverse order. Model this when predicting
use-list order. This gets two more Bitcode tests passing with
`llvm-uselistorder`.
Part of PR5680.
llvm-svn: 214242
This moves some tests around to make it clearer what's being tested,
and adds very rudimentary comment syntax to the text input format to
make specifying this kind of test a little bit simpler.
llvm-svn: 214235
MSVC [1] thinks `UseListShuffleVector` needs a copy constructor, but I
don't. Let's see if being explicit about `UseListOrder` is convincing.
[1]: http://lab.llvm.org:8011/builders/lld-x86_64-win7/builds/11664/steps/build_Lld/logs/stdio
Here's the failure:
C:/lld-x86_64_win7/lld-x86_64-win7/llvm.src/include\llvm/IR/UseListOrder.h(92): error C2248: 'llvm::UseListShuffleVector::operator =' : cannot access private member declared in class 'llvm::UseListShuffleVector' (C:\lld-x86_64_win7\lld-x86_64-win7\llvm.src\lib\Bitcode\Writer\ValueEnumerator.cpp) [C:\lld-x86_64_win7\lld-x86_64-win7\llvm.obj\lib\Bitcode\Writer\LLVMBitWriter.vcxproj]
C:/lld-x86_64_win7/lld-x86_64-win7/llvm.src/include\llvm/IR/UseListOrder.h(56) : see declaration of 'llvm::UseListShuffleVector::operator ='
C:/lld-x86_64_win7/lld-x86_64-win7/llvm.src/include\llvm/IR/UseListOrder.h(32) : see declaration of 'llvm::UseListShuffleVector'
This diagnostic occurred in the compiler generated function 'llvm::UseListOrder &llvm::UseListOrder::operator =(const llvm::UseListOrder &)'
llvm-svn: 214224
Remove the copy constructor added in r214178 to appease MSVC17 since it
shouldn't be called at all. My guess is that explicitly deleting it
will make the compiler happy. To round out the operations I've also
deleted copy assignment and added move assignment. Otherwise no
functionality change.
llvm-svn: 214213
full paths for its first argument.
This allows us to remove the annoying sed lines in the test cases, and write
direct references to file names in stub_addr calls (rather than <filename>
placeholders).
llvm-svn: 214211
This will let users in other libraries know which error occurred. In particular,
it will be possible to check if the parsing failed or if the file is not
bitcode.
llvm-svn: 214209
Sanjay Patel