Summary:
This patch adds CustomRenderer which renders the matched
operands to the specified instruction.
Targets can enable the matching of SDNodeXForm by adding
a definition that inherits from GICustomOperandRenderer and
GISDNodeXFormEquiv as follows.
def gi_imm8 : GICustomOperandRenderer<"renderImm8”>,
GISDNodeXFormEquiv<imm8_xform>;
Custom renderer functions should be of the form:
void render(MachineInstrBuilder &MIB, const MachineInstr &I);
Reviewers: dsanders, ab, rovka
Reviewed By: dsanders
Subscribers: kristof.beyls, javed.absar, llvm-commits, mgrang, qcolombet
Differential Revision: https://reviews.llvm.org/D42012
llvm-svn: 322582
Summary: Not sure this needs a review or not. Erring on the safe side.
Reviewers: dblaikie
Differential Revision: https://reviews.llvm.org/D41666
llvm-svn: 322538
The ELF specification says that all ELF data structures are aligned to
their natural alignments both in memory and file. That means when we
access mmap'ed ELF files, we could assume that all data structures are
aligned properly.
However, in reality, we assume that the data structures are aligned only
to two bytes because .a files only guarantee that their member files are
aligned to two bytes in archive files. So the data access is already
unaligned.
This patch relaxes the alignment requirement even more, so that we
accept unaligned access to all ELF data structures.
This patch in particular makes lld bug-compatible with icc. Intel C
compiler doesn't seem to care about data alignment and generates unaligned
relocation sections (https://bugs.llvm.org/show_bug.cgi?id=35854).
I also saw another instance of compatibility issues with our internal tool
which creates unaligned section headers.
Because GNU linkers are not picky about alignment, looks like it is
not uncommon that ELF-generating tools create unaligned files.
There is a performance penalty with this patch on host machines on which
unaligned access is expensive. x86 and AArch64 are fine. ARMv6 is a
problem, but I don't think using ARMv6 machines as hosts is common, so I
believe it's not a real problem.
Differential Revision: https://reviews.llvm.org/D41978
llvm-svn: 322407
Summary:
In preparation for https://reviews.llvm.org/D41675 this NFC changes this
prototype of MemIntrinsicInst::setAlignment() to accept an unsigned instead
of a Constant.
llvm-svn: 322403
Summary:
See D37528 for a previous (non-deferred) version of this
patch and its description.
Preserves dominance in a deferred manner using a new class
DeferredDominance. This reduces the performance impact of
updating the DominatorTree at every edge insertion and
deletion. A user may call DDT->flush() within JumpThreading
for an up-to-date DT. This patch currently has one flush()
at the end of runImpl() to ensure DT is preserved across
the pass.
LVI is also preserved to help subsequent passes such as
CorrelatedValuePropagation. LVI is simpler to maintain and
is done immediately (not deferred). The code to perform the
preversation was minimally altered and simply marked as
preserved for the PassManager to be informed.
This extends the analysis available to JumpThreading for
future enhancements such as threading across loop headers.
Reviewers: dberlin, kuhar, sebpop
Reviewed By: kuhar, sebpop
Subscribers: mgorny, dmgreen, kuba, rnk, rsmith, hiraditya, llvm-commits
Differential Revision: https://reviews.llvm.org/D40146
llvm-svn: 322401
We can probably take this a step further since the only
user of the isUsed flag is AsmParser it should probably
be doing this explicitly. For now this is a step in the
right direction though.
Differential Revision: https://reviews.llvm.org/D41971
llvm-svn: 322386
It was never fully disallowed. We were rejecting it in the asm parser,
but not in the verifier.
Currently TargetMachine::shouldAssumeDSOLocal returns true for hidden
ifuncs. I considered changing it and moving the check from the asm
parser to the verifier.
The reason for deciding to allow it instead is that all linkers handle
a direct reference just fine. They use the plt address as the address
of the function. In fact doing that means that clang doesn't have the
same bug as gcc: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=83782.
This patch then removes the check from the asm parser and updates the
bitcode reader and writer.
llvm-svn: 322378
ExecutionSession will represent a running JIT program.
VModuleKey is a unique key assigned to each module added as part of
an ExecutionSession. The Layer concept will be updated in future to
require a VModuleKey when a module is added.
llvm-svn: 322336
The PeepholeOptimizer would fail for vregs without a definition. If this
was caused by an undef operand abort to keep the code simple (so we
don't need to add logic everywhere to replicate the undef flag).
Differential Revision: https://reviews.llvm.org/D40763
llvm-svn: 322319
While updating clang tests for having clang set dso_local I noticed
that:
- There are *a lot* of tests to update.
- Many of the updates are redundant.
They are redundant because a GV is "obviously dso_local". This patch
starts formalizing that a bit by requiring that internal and private
GVs be dso_local too. Since they all are, we don't have to print
dso_local to the textual representation, making it a bit more compact
and easier to read.
llvm-svn: 322317
Summary:
References to the source operand within class MemTransferInst are currently
by a constant 1. Abstract this out into a named constant.
llvm-svn: 322289
Summary:
Addresses issue: https://bugs.llvm.org/show_bug.cgi?id=34595
The topmost class, `SmallVector`, has internal storage for some
elements; `N - 1` elements' bytes worth of space. Meanwhile a base
class `SmallVectorTemplateCommon` has room for one element as well,
totaling `N` elements' worth of space.
The space for the N elements is contiguous and straddles
`SmallVectorTemplateCommon` and `SmallVector`.
A class "between" those two owning the storage, `SmallVectorImpl`, in
its destructor, calls the destructor for elements contained in the
vector, if any. It uses `destroy_range(begin, end)` and deletes all
items in sequence, starting from the end.
By the time the destructor for `SmallVectorImpl` is running, though, the
memory for elements `[1, N)` is already poisoned, due to `SmallVector`'s
destructor having done its thing already.
So if the element type `T` has a nontrivial destructor that accesses any
members of the `T` instance being destroyed, we'll run into a
user-after-poison bug.
This patch moves the destruction loop into `SmallVector`'s destructor,
so any memory being accessed while dtors are running is not yet
poisoned.
Confirmed this broke before (and now works with this patch) with these
compiler flags:
-fsanitize=memory
-fsanitize-memory-use-after-dtor
-fsanitize-memory-track-origins
and with the cmake flag
`-DLLVM_USE_SANITIZER='MemoryWithOrigins;Undefined'` as well as
`MSAN_OPTIONS=poison_in_dtor=1`.
Patch By: elsteveogrande
Reviewers: eugenis, morehouse, dblaikie
Reviewed By: eugenis, dblaikie
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D41916
llvm-svn: 322241
Revert for now as the testcase is hitting a pre-existing verifier error
that manifest as a failure when expensive checks are enabled (or
-verify-machineinstrs) is used.
This reverts commit r322200.
llvm-svn: 322231
Simplify the code slightly: Instead of creating empty subranges in one
case and immediately removing them, do not create them in the first
place.
llvm-svn: 322226
Currently we infer the scale at isel time by analyzing whether the base is a constant 0 or not. If it is we assume scale is 1, else we take it from the element size of the pass thru or stored value. This seems a little weird and I think it makes more sense to make it explicit in the DAG rather than doing tricky things in the backend.
Most of this patch is just making sure we copy the scale around everywhere.
Differential Revision: https://reviews.llvm.org/D40055
llvm-svn: 322210
Large callframes (calls with several hundreds or thousands or
parameters) could lead to situations in which the emergency spillslot is
out of range to be addressed relative to the stack pointer.
This commit forces the use of a frame pointer in the presence of large
callframes.
This commit does several things:
- Compute max callframe size at the end of instruction selection.
- Add mirFileLoaded target callback. Use it to compute the max callframe size
after loading a .mir file when the size wasn't specified in the file.
- Let TargetFrameLowering::hasFP() return true if there exists a
callframe > 255 bytes.
- Always place the emergency spillslot close to FP if we have a frame
pointer.
- Note that `useFPForScavengingIndex()` would previously return false
when a base pointer was available leading to the emergency spillslot
getting allocated late (that's the whole effect of this callback).
Which made no sense to me so I took this case out: Even though the
emergency spillslot is technically not referenced by FP in this case
we still want it allocated early.
Differential Revision: https://reviews.llvm.org/D40876
llvm-svn: 322200
Summary:
This extends TableGen's AsmMatcherEmitter with code that generates
a table with tied-operand constraints. The constraints are checked
when parsing the instruction. If an operand is not equal to its tied operand,
the assembler will give an error.
Patch [2/3] in a series to add operand constraint checks for SVE's predicated ADD/SUB.
Reviewers: olista01, rengolin, mcrosier, fhahn, craig.topper, evandro, echristo
Reviewed By: fhahn
Subscribers: javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D41446
llvm-svn: 322166
There were a few places where outs() was being used
directly rather than the ScopedPrinter object.
Differential Revision: https://reviews.llvm.org/D41370
llvm-svn: 322141
version being used on some of the green dragon builders (plus a clang-format).
Workaround: AsynchronousSymbolQuery and VSO want to work with
JITEvaluatedSymbols anyway, so just use them (instead of JITSymbol, which
happens to tickle the bug).
The libcxx bug being worked around was fixed in r276003, and there are plans to
update the offending builders.
llvm-svn: 322140
Adds option /guard:cf to clang-cl and -cfguard to cc1 to emit function IDs
of functions that have their address taken into a section named .gfids$y for
compatibility with Microsoft's Control Flow Guard feature.
The original patch didn't have the lit.local.cfg file that restricts the new
test to x86, thus the new test was failing on the non-x86 bots.
Differential Revision: https://reviews.llvm.org/D40531
The reverts r322008, which was a revert of r322005.
This reverts commit a05b89f9aca70597dc79fe97bc49b50b51f525ba.
llvm-svn: 322136
Summary:
When performing constant propagation for call instructions we have historically replaced all uses of the return from a call, but not removed the call itself. This is required for correctness if the calls have side effects, however the compiler should be able to safely remove calls that don't have side effects.
This allows the compiler to completely fold away calls to functions that have no side effects if the inputs are constant and the output can be determined at compile time.
Reviewers: davide, sanjoy, bruno, dberlin
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D38856
llvm-svn: 322125
These indexes are useful because they are not always zero based and
functions and globals are referenced elsewhere by their index.
This matches what we already do for the type index space.
Differential Revision: https://reviews.llvm.org/D41877
llvm-svn: 322121
Patch by Takuto Ikuta.
This patch reduces lld link time of chromium's blink_core.dll in
component build.
Total size of input argument in .directives become nearly 300MB in the
build and almost all its content are /EXPORT.
To reduce time of parsing too many /EXPORT option in the build, I
introduce fastpath for /EXPORT in ArgParser::parseDirectives.
On my desktop machine, 4 times stats of the link time are like below.
Improved around 20%.
This patch
TotalSeconds : 8.6217627
TotalSeconds : 8.5402175
TotalSeconds : 8.6855853
TotalSeconds : 8.3624441
Ave : 8.5525024
master
TotalSeconds : 10.9975031
TotalSeconds : 11.3409428
TotalSeconds : 10.6332897
TotalSeconds : 10.7650687
Ave : 10.934201075
llvm-svn: 322117
Summary:
This pass synthesizes function entry counts by traversing the callgraph
and using the relative block frequencies of the callsites. The intended
use of these counts is in inlining to determine hot/cold callsites in
the absence of profile information.
The pass is split into two files with the code that propagates the
counts in a callgraph in a Utils file. I plan to add support for
propagation in the thinlto link phase and the propagation code will be
shared and hence this split. I did not add support to the old PM since
hot callsite determination in inlining is not possible in old PM
(although we could use hot callee heuristic with synthetic counts in the
old PM it is not worth the effort tuning it)
Reviewers: davidxl, silvas
Subscribers: mgorny, mehdi_amini, llvm-commits
Differential Revision: https://reviews.llvm.org/D41604
llvm-svn: 322110
llc, opt, and clang can all autodetect the CPU and supported features. lli cannot as far as I could tell.
This patch uses the getCPUStr() and introduces a new getCPUFeatureList() and uses those in lli in place of MCPU and MAttrs.
Ideally, we would merge getCPUFeatureList and getCPUFeatureStr, but opt and llc need a string and lli wanted a list. Maybe we should just return the SubtargetFeature object and let the caller decide what it needs?
Differential Revision: https://reviews.llvm.org/D41833
llvm-svn: 322100