AArch64: Make getNextNode() cleanup in r249764 more clear
After r249764, if you didn't see the full context, it looked like
`std::next(I)` would get the same result as
`++MachineBasicBlock::iterator(I)`. However, `I` is a `MachineInstr*`
(not a `MachineBasicBlock::iterator`).
Use the `getIterator()` helper I added later (r249782) to make this code
more clear.
Revert "Support: Partially revert r249782 to unbreak clang build"
This reverts commit r249783, fully reinstating r249782. I've fixed the
bug in clang: it was a non-const iterator that dereferenced to const
(but had an implicit conversion to non-const).
Dehao Chen [Fri, 9 Oct 2015 16:50:16 +0000 (16:50 +0000)]
Make HeaderLineno a local variable.
http://reviews.llvm.org/D13576
As we are using hierarchical profile, there is no need to keep HeaderLineno a member variable. This is because each level of the inline stack will have its own header lineno. One should use the head lineno of its own inline stack level instead of the actual symbol.
Kevin Enderby [Fri, 9 Oct 2015 16:48:44 +0000 (16:48 +0000)]
Fixed two bugs in llvm-objdump’s printing of Objective-C meta data
from malformed Mach-O files that caused crashes. The first because the
offset in a dyld bind table entry was out of range. The second because their
was no image info section and the routine printing it did not have the
need check to see the section did not exist.
Vector element extraction without stack operations on Power 8
This patch corresponds to review:
http://reviews.llvm.org/D12032
This patch builds onto the patch that provided scalar to vector conversions
without stack operations (D11471).
Included in this patch:
- Vector element extraction for all vector types with constant element number
- Vector element extraction for v16i8 and v8i16 with variable element number
- Removal of some unnecessary COPY_TO_REGCLASS operations that ended up
unnecessarily moving things around between registers
Not included in this patch (will be in upcoming patch):
- Vector element extraction for v4i32, v4f32, v2i64 and v2f64 with
variable element number
- Vector element insertion for variable/constant element number
Testing is provided for all extractions. The extractions that are not
implemented yet are just placeholders.
[MemCpyOpt] Fix wrong merging adjacent nontemporal stores into memset calls.
Pass MemCpyOpt doesn't check if a store instruction is nontemporal.
As a consequence, adjacent nontemporal stores are always merged into a
memset call.
Example:
;;;
define void @foo(<4 x float>* nocapture %p) {
entry:
store <4 x float> zeroinitializer, <4 x float>* %p, align 16, !nontemporal !0
%p1 = getelementptr inbounds <4 x float>, <4 x float>* %dst, i64 1
store <4 x float> zeroinitializer, <4 x float>* %p1, align 16, !nontemporal !0
ret void
}
!0 = !{i32 1}
;;;
In this example, the two nontemporal stores are combined to a memset of zero
which does not preserve the nontemporal hint. Later on the backend (tested on a
x86-64 corei7) expands that memset call into a sequence of two normal 16-byte
aligned vector stores.
With this patch, we no longer merge nontemporal stores into calls to memset.
In this example, llc correctly expands the two stores into two movntps:
xorps %xmm0, %xmm0
movntps %xmm0, 16(%rdi)
movntps %xmm0, (%rdi)
In theory, we could extend the usage of !nontemporal metadata to memcpy/memset
calls. However a change like that would only have the effect of forcing the
backend to expand !nontemporal memsets back to sequences of store instructions.
A memset library call would not have exactly the same semantic of a builtin
!nontemporal memset call. So, SelectionDAG will have to conservatively expand
it back to a sequence of !nontemporal stores (effectively undoing the merging).
Jonas Paulsson [Fri, 9 Oct 2015 07:19:16 +0000 (07:19 +0000)]
[SystemZ] Minor bugfixes.
LLCH, LLHH and CLIH had the wrong register classes for the def-operand.
Tie operands if changing opcode to an instruction with tied ops.
Comment typo fix.
These fixes were needed in order to make regression test case
SystemZ/asm-18.ll pass with -verify-machineinstrs (not used by
default).
Jonas Paulsson [Fri, 9 Oct 2015 07:19:12 +0000 (07:19 +0000)]
[SystemZ] Bugfix in SystemZAsmParser.cpp.
Let parseRegister() allow RegFP Group if expecting RegV Group, since the
%f register prefix yields the FP group even while used with vector instructions.
Accept r11 when targeting Windows on ARM rather than just low registers.
Because we are in a thumb-2 only mode, this may be slightly more expensive in
code size, but results in better code for the environment since it spills the
frame register, which is generally desired for fast stack walking as per the
ABI.
Sanjoy Das [Fri, 9 Oct 2015 02:44:45 +0000 (02:44 +0000)]
[SCEV] Call `StrengthenNoWrapFlags` after `GroupByComplexity`; NFCI
The current implementation of `StrengthenNoWrapFlags` is agnostic to the
order of `Ops`, so this commit should not change anything semantic. An
upcoming change will make `StrengthenNoWrapFlags` sensitive to the order
of `Ops`.
Summary:
- Recurse from cleanupendpads to their cleanuppads, to make sure the
cleanuppad is visited if it has a cleanupendpad but no cleanupret.
- Check for and avoid double-processing cleanuppads, to allow for them to
have multiple cleanuprets (plus cleanupendpads).
- Update Cxx state numbering to visit toplevel cleanupendpads and to
recurse from cleanupendpads to their preds, to ensure we number any
funclets in inlined cleanups. SEH state numbering already did this.
Reid Kleckner [Fri, 9 Oct 2015 00:15:01 +0000 (00:15 +0000)]
[llvm-symbolizer] Make --relative-address work with DWARF contexts
Summary:
Previously the relative address flag only affected PDB debug info. Now
both DIContext implementations always expect to be passed virtual
addresses. llvm-symbolizer is now responsible for adding ImageBase to
module offsets when --relative-offset is passed.
Support: Partially revert r249782 to unbreak clang build
Apparently the iterators in `clang::CFGBlock` have an auto-conversion to
`CFGBlock *`, but the dereference operator gives `const CFGBlock &`.
Until I have a moment to fix that, revert the GenericDomTree chagnes
from r249782.
IR: Remove implicit iterator conversions from lib/IR, NFC
Stop converting implicitly between iterators and pointers/references in
lib/IR. For convenience, I've added a `getIterator()` accessor to
`ilist_node` so that callers don't need to know how to spell the
iterator class (i.e., they can use `X.getIterator()` instead of
`Function::iterator(X)`).
I'll eventually disallow these implicit conversions entirely, but
there's a lot of code, so it doesn't make sense to do it all in one
patch. One library or so at a time.
Why? To root out cases of `getNextNode()` and `getPrevNode()` being
used in iterator logic. The design of `ilist` makes that invalid when
the current node could be at the back of the list, but it happens to
"work" right now because of a bug where those functions never return
`nullptr` if you're using a half-node sentinel. Before I can fix the
function, I have to remove uses of it that rely on it misbehaving.
(Maybe the function should just be deleted anyway? But I don't want
deleting it -- potentially a huge project -- to block fixing
ilist/iplist.)
Sanjoy Das [Thu, 8 Oct 2015 23:18:33 +0000 (23:18 +0000)]
[IRBuilder] Change the `gc.statepoint` creation interface
This is to enable me to address review for D13491 -- `Flags` is a
bitfield of `StatepointFlags`, not an individual item out of the enum,
so it should be represented as an `uint32_t`.
Kevin Enderby [Thu, 8 Oct 2015 22:50:55 +0000 (22:50 +0000)]
Fix a bug in llvm-objdump’s printing of Objective-C meta data
from malformed Mach-O files that caused a crash because of a
section header had a size that extended past the end of the file.
Support: Stop relying on iterator auto-conversion, NFC
Stop relying on ilist implicit conversions from `value_type&` to
`iterator` in YAMLParser.cpp.
I eventually want to outlaw this entirely. It encourages
`getNextNode()` and `getPrevNode()` in iterator logic, which is
extremely fragile (and relies on them never returning `nullptr`).
FTR, there's nothing nefarious going on in this case, it was just easy
to clean up since the callers really wanted iterators to begin with.
Stop using `getNextNode()` to get an insertion point (at least, in this
one place). Instead, use iterator logic directly.
The `getNextNode()` interface isn't actually supposed to work for
creating iterators; it's supposed to return `nullptr` (not a real
iterator) if this is the last node. It's currently broken and will
"happen" to work, but if we ever fix the function, we'll get some
strange failures in places like this.
Stop using `getNextNode()` to get an iterator to a fragment (at least,
in this one place). Instead, use iterator logic directly.
The `getNextNode()` interface isn't actually supposed to work for
creating iterators; it's supposed to return `nullptr` (not a real
iterator) if this is the last node. It's currently broken and will
"happen" to work, but if we ever fix the function, we'll get some
strange failures in places like this.
Frederic Riss [Thu, 8 Oct 2015 22:35:53 +0000 (22:35 +0000)]
[dsymutil] Try to find lipo first besides dsymutil before looking up the PATH.
Even if we don't have it in PATH, lipo should usually exist in the same directory
as dsymutil. Keep the fallback looking up the PATH, it's very useful when
testing a non-installed executable.
PowerPC: Don't use getNextNode() for insertion point
Stop using `getNextNode()` to create an insertion point for machine
instructions (at least, in this one place). Instead, use an iterator.
As a drive-by, clean up dump statements to use iterator logic.
The `getNextNode()` interface isn't actually supposed to work for
insertion points; it's supposed to return `nullptr` if this is the last
node. It's currently broken and will "happen" to work, but if we ever
fix the function, we'll get some strange failures.
This is an implementation of
https://github.com/google/sanitizers/issues/579
It has a number of advantages over the current mapping:
* Works for non-PIE executables.
* Does not require ASLR; as a consequence, debugging MSan programs in
gdb no longer requires "set disable-randomization off".
* Supports linux kernels >=4.1.2.
* The code is marginally faster and smaller.
This is an ABI break. We never really promised ABI stability, but
this patch includes a courtesy escape hatch: a compile-time macro
that reverts back to the old mapping layout.
Move the MMX subtarget feature out of the SSE set of features and into
its own variable.
This is needed so that we can explicitly turn off MMX without turning
off SSE and also so that we can diagnose feature set incompatibilities
that involve MMX without SSE.
Rationale:
// sse3
__m128d test_mm_addsub_pd(__m128d A, __m128d B) {
return _mm_addsub_pd(A, B);
}
// mmx
void shift(__m64 a, __m64 b, int c) {
_mm_slli_pi16(a, c);
_mm_slli_pi32(a, c);
_mm_slli_si64(a, c);
_mm_srli_pi16(a, c);
_mm_srli_pi32(a, c);
_mm_srli_si64(a, c);
_mm_srai_pi16(a, c);
_mm_srai_pi32(a, c);
}
clang -msse3 -mno-mmx file.c -c
For this code we should be able to explicitly turn off MMX
without affecting the compilation of the SSE3 function and then
diagnose and error on compiling the MMX function.
This matches the existing gcc behavior and follows the spirit of
the SSE/MMX separation in llvm where we can (and do) turn off
MMX code generation except in the presence of intrinsics.
Updated a couple of tests, but primarily tested with a couple of tests
for turning on only mmx and only sse.
This is paired with a patch to clang to take advantage of this behavior.
Diego Novillo [Thu, 8 Oct 2015 19:40:37 +0000 (19:40 +0000)]
Re-apply r249644: Handle inline stacks in gcov-encoded sample profiles.
This fixes memory allocation problems by making the merge operation keep
the profile readers around until the merged profile has been emitted.
This is needed to prevent the inlined function names to disappear from
the function profiles. Since all the names are kept as references, once
the reader disappears, the names are also deallocated.
Additionally, XFAIL on big-endian architectures. The test case uses a
gcov file generated on a little-endian system.
[bpf] Do not expand UNDEF SDNode during insn selection lowering
o Before this patch, BPF backend will expand UNDEF node
to i64 constant 0.
o For second pass of dag combiner, legalizer will run through
each to-be-processed dag node.
o If any new SDNode is generated and has an undef operand,
dag combiner will put undef node, newly-generated constant-0 node,
and any node which uses these nodes in the working list.
o During this process, it is possible undef operand is
generated again, and this will form an infinite loop
for dag combiner pass2.
o This patch allows UNDEF to be a legal type.
Sanjoy Das [Thu, 8 Oct 2015 18:28:36 +0000 (18:28 +0000)]
[SCEV] Pick backedge values for phi nodes correctly
Summary:
`getConstantEvolutionLoopExitValue` and `ComputeExitCountExhaustively`
assumed all phi nodes in the loop header have the same order of incoming
values. This is not correct, and this commit changes
`getConstantEvolutionLoopExitValue` and `ComputeExitCountExhaustively`
to lookup the backedge value of a phi node using the loop's latch block.
Unfortunately, there is still some code duplication
`getConstantEvolutionLoopExitValue` and `ComputeExitCountExhaustively`.
At some point in the future we should extract out a helper class /
method that can evolve constant evolution phi nodes across iterations.
Fixes 25060. Thanks to Mattias Eriksson for the spot-on analysis!
Ulrich Weigand [Thu, 8 Oct 2015 17:46:59 +0000 (17:46 +0000)]
[SystemZ] Fix another assertion failure in tryBuildVectorShuffle
This fixes yet another scenario where tryBuildVectorShuffle would
attempt to create a BUILD_VECTOR node with an invalid combination
of types. This can happen if the incoming BUILD_VECTOR has elements
of a type different from the vector element type, which is allowed
in certain cases as long as they are all the same type.
When one of these elements is used in the residual vector, and
UNDEF elements are added to fill up the residual vector, those
UNDEFs then have to use the type of the original element, not
the vector element type, or else the resulting BUILD_VECTOR
will have an invalid type combination.
Sanjay Patel [Thu, 8 Oct 2015 17:09:31 +0000 (17:09 +0000)]
[InstCombine] transform masking off of an FP sign bit into a fabs() intrinsic call (PR24886)
This is a partial fix for PR24886:
https://llvm.org/bugs/show_bug.cgi?id=24886
Without this IR transform, the backend (x86 at least) was producing inefficient code.
This patch is making 2 assumptions:
1. The canonical form of a fabs() operation is, in fact, the LLVM fabs() intrinsic.
2. The high bit of an FP value is always the sign bit; as noted in the bug report, this isn't specified by the LangRef.
Sanjay Patel [Thu, 8 Oct 2015 16:56:55 +0000 (16:56 +0000)]
[ValueTracking] teach computeKnownBits that a fabs() clears sign bits
This was requested in D13076: if we're going to canonicalize to fabs(), ValueTracking
should know that fabs() clears sign bits.
In this patch (as in D13076), we're not handling vectors yet even though computeKnownBits'
fabs() case itself should be vector-ready via the splat in this patch.
Fixing this will require follow-on patches to correct other logic that uses 'getScalarType'.
George Rimar [Thu, 8 Oct 2015 16:03:19 +0000 (16:03 +0000)]
Windows: Fixed sys::findProgramByName to work with files containing dot in their name.
Problem was in SearchPathW function that does not attach an extension if file already has one.
That does not work for executables like ld.lld2 for example which require to have .exe extension but SearchPath thinks that its "lld2".
Solution was to add the extension manually.
Teresa Johnson [Thu, 8 Oct 2015 15:56:24 +0000 (15:56 +0000)]
Support for llvm-bcanalyzer dumping of record array strings.
Summary:
Adds support for automatically detecting and printing strings
represented by Array abbrev operands, analogous to the string dumping
performed for Blob abbrev operands.
Enhanced the ThinLTO combined index test to check for the appropriate
module and function strings.
Frederic Riss [Thu, 8 Oct 2015 15:45:08 +0000 (15:45 +0000)]
[X86] Disable X86CallFrameOptimization on Darwin in presence of EH
We emit 1 compact unwind encoding per function, and this can’t represent
the varying stack pointer that will be generated by X86CallFrameOptimization.
Disable the optimization on Darwin.
(It might be possible to split the function into multiple ranges
and emit 1 compact unwind info per range. The compact unwind emission
code isn’t ready for that and this kind of info certainly isn’t
tested/used anywhere. It might be worth exploring this path if we want
to get the space savings at some point though)
Teresa Johnson [Thu, 8 Oct 2015 13:52:56 +0000 (13:52 +0000)]
Fix combined function index abbrev (NFC)
Removed an unused abbrev op in the VST_CODE_COMBINED_FNENTRY abbrev.
I noticed while writing/testing an array string dumper for
llvm-bcanalyze that the combined function's VST entry abbrevs contained
an old field that I am not using. Everything was working fine since the
bitcode writer and reader were in sync on how the record fields were
actually being set up and interpreted.
James Molloy [Thu, 8 Oct 2015 12:40:06 +0000 (12:40 +0000)]
Compute demanded bits for icmp instructions
Instead of bailing out when we see an icmp, we can instead at least
say that if the upper bits of both operands are known zero, they are
not demanded. This doesn't help with signed comparisons, but it's at
least better than bailing out.
James Molloy [Thu, 8 Oct 2015 12:39:50 +0000 (12:39 +0000)]
Make demanded bits lazy
The algorithm itself is still eager, but it doesn't get run until a
query function is called. This greatly reduces the compile-time impact
of requiring DemandedBits when at runtime it is not often used.
[X86] Fix wrong treatment of multi-lane blends in BUILD_VECTORtoBlendMask()
This fixes two separate bugs:
1) The mask for the high lane was not set correctly. That fixes PR24532.
2) The transformation should bail out if it believes it involves more than
2 lanes, as it does not currently do anything sensible in this case.
Jonas Paulsson [Thu, 8 Oct 2015 07:40:16 +0000 (07:40 +0000)]
[SystemZ] Use load-and-test for fp compare with 0 if vector support is present.
Since the LTxBRCompare instructions can't be used with vector registers, a
normal load-and-test instruction (with a modelled def operand) is used instead.
Diego Novillo [Thu, 8 Oct 2015 00:39:11 +0000 (00:39 +0000)]
Handle inline stacks in gcov-encoded sample profiles.
This patch adds support for reading sample profiles with inline stacks.
Inline stacks in a profile are generated when the sampled binary has
samples in inlined functions.
For instance, if main() calls foo() and foo() calls bar(), and bar() is
inlined into foo() and foo() inlined into main(), the profile may look
something like:
Justin Bogner [Thu, 8 Oct 2015 00:36:22 +0000 (00:36 +0000)]
CodeGen: print and verify after TargetPassConfig::insertPass by default
In r224059, we started verifying after addPass, but missed doing so on
insertPass. There isn't a good reason for the discrepancy, and
skipping the verifier in these cases causes bugs.
This also exposes a verifier error that was introduced in r249087, but
the verifier doesn't run until after the register coalescer, when the
issue happens to have been resolved. I've skipped the verifier after
SIFixSGPRLiveRangesID to avoid the failures for now and will follow up
with Matt for a proper fix.
Reid Kleckner [Wed, 7 Oct 2015 23:55:01 +0000 (23:55 +0000)]
[WinEH] Fix 32-bit funclet epilogues in the presence of dynamic allocas
In particular, passing non-trivially copyable objects by value on win32
uses a dynamic alloca (inalloca). We would clobber ESP in the epilogue
and end up returning to outer space.
Lang Hames [Wed, 7 Oct 2015 21:53:41 +0000 (21:53 +0000)]
[Orc] Enable user supplied partitioning functors in the CompileOnDemand layer.
Previously the CompileOnDemand layer always created single-function partitions.
In theory this new API allows for more interesting partitions, though this has
not been well tested yet.
David Majnemer [Wed, 7 Oct 2015 21:34:00 +0000 (21:34 +0000)]
[WinEH] Refer to filter funclets using their symbol-table symbol
The relocation for the filter funclet will be against a symbol table
entry for a function instead of the section, making it easier to
understand what is going on.
Reid Kleckner [Wed, 7 Oct 2015 21:13:15 +0000 (21:13 +0000)]
[WinEH] Undo the effect of r249578 for 32-bit
The __CxxFrameHandler3 tables for 32-bit are supposed to hold stack
offsets relative to EBP, not ESP. I blindly updated the win-catchpad.ll
test case, and immediately noticed that 32-bit catching stopped working.
While I'm at it, move the frame index to frame offset WinEH table logic
out of PEI. PEI shouldn't have to know about WinEHFuncInfo. I realized
we can calculate frame index offsets just fine from the table printer.
David Majnemer [Wed, 7 Oct 2015 21:08:25 +0000 (21:08 +0000)]
[WinEH] Remove unreachable blocks before preparation
We remove unreachable blocks because it is pointless to consider them
for coloring. However, we still had stale pointers to these blocks in
some data structures after we removed them from the function.
Instead, remove the unreachable blocks before attempting to do anything
with the function.
Recycler just needs a singly-linked list, and it takes less (and
simpler) code to hand-roll one of those than to build up the equivalent
`iplist_traits`. In theory, this should speed things up a bit too, but
this is really just a drive-by cleanup so I haven't measured.
IR: Create SymbolTableList wrapper around iplist, NFC
Create `SymbolTableList`, a wrapper around `iplist` for lists that
automatically manage a symbol table. This commit reduces a ton of code
duplication between the six traits classes that were used previously.
As a drive by, reduce the number of template parameters from 2 to 1 by
using a SymbolTableListParentType metafunction (I originally had this as
a separate commit, but it touched most of the same lines so I squashed
them).
I'm in the process of trying to remove the UB in `createSentinel()` (see
the FIXMEs I added for `ilist_embedded_sentinel_traits` and
`ilist_half_embedded_sentinel_traits`). My eventual goal is to separate
the list logic into a base class layer that knows nothing about (and
isn't templated on) the downcasted nodes -- removing the need to invoke
UB -- but for now I'm just trying to get a handle on all the current use
cases (and cleaning things up as I see them).
Besides these six SymbolTable lists, there are two others that use the
addNode/removeNode/transferNodes() hooks: the `MachineInstruction` and
`MachineBasicBlock` lists. Ideally there'll be a way to factor these
hooks out of the low-level API entirely, but I'm not quite there yet.
Summary:
This is necessary to keep the cloner from making bogus copies of debug
metadata attached to the IR it is cloning.
Also, avoid running RemapInstruction over all instructions in the common
case that no cloning was performed.
projects / llvm / log