Summary:
This at least saves compile time. I also encountered a case where
ephemeral values affect whether other variables are promoted, causing
performance issues. It may be a bug in LSR, but I didn't manage to
reduce it yet. Anyhow, I believe it's in general not worth considering
ephemeral values in LSR.
Reviewers: atrick, hfinkel
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D11115
llvm-svn: 242011
Three things:
- The atomic intrinsics mandate memory barriers, let's start emitting
some.
- We don't need to manually create RMW operations, we can just do
__atomic_fetch_foo instead of performing __atomic_foo_fetch and
undoing foo.
- Don't use inline assembly, we don't need it for these intrinsics.
This fixes PR24101.
llvm-svn: 242009
clang-cl doesn't compile std::atomic_flag correctly (PR24101). Since the COFF
linker doesn't use threads yet, just revert r241420 and r241481 for now to
work around this clang-cl bug.
llvm-svn: 242006
With LLVM_ENABLE_THREADS disabled, all the llvm code assumes that it runs on
a single thread and doesn't use any mutexes. lld still spawned lots of threads
in that case and called into llvm, assuming that llvm is thread-safe.
As fix, let lld use only a single thread if LLVM_ENABLE_THREADS is disabled.
I left in all the mutexes in lld. That means lld is a bit slower than
necessary in single-thread mode, but that's probably worth the simpler code.
llvm-svn: 242004
before the first imported declaration.
We don't need to track all formerly-canonical declarations of an entity; it's sufficient to track those ones for which no other formerly-canonical declaration was imported into the same module. We call those ones "key declarations", and use them as our starting points for collecting redeclarations and performing namespace lookups.
llvm-svn: 241999
In the test, y1 is not reference compatible to y2 and we currently assume
the cast is ill-formed so we emit a diagnostic. Instead, in order to honour
the standard, if y1 it's not reference-compatible to y2 then it can't be
converted using a static_cast, and a reinterpret_cast should be tried instead.
Richard Smith provided the correct interpretation of the standard and
explanation about the subtle difference between "can't be cast" and "the cast
is ill-formed". The former applies in this case.
PR: 23802
llvm-svn: 241998
Register r12 ('ip') is used by GCC for this purpose
and hence is used here. As discussed on the GCC mailing
list, the register choice is an ABI issue and so
choosing the same register as GCC means
__builtin_call_with_static_chain is compatible.
A similar patch has just gone in the AArch64 backend,
so this is just the ARM counterpart, following the same
discussion.
Patch by Stephen Cross.
llvm-svn: 241996
While the v4i32 shl operation is already vectorized using a cvttps2dq/pmulld pattern, the lshr/ashr opeations are still scalarized.
This patch adds vectorization support for non-uniform v4i32 shift operations - it splats constant shift amounts to allow them to use the immediate sse shift instructions, or extracts/zero-extends non-constant shift amounts. The individual results are then blended together.
Differential Revision: http://reviews.llvm.org/D11063
llvm-svn: 241989
The function uses parallel_for() and then writes error messages from the
parallel loop's body. This produces nondetermistic error messages. Instead,
copy error messages to a vector and sort it by the atom's file offsets before
printing all error messages after the parallel_for(). This results in a few
string copies, but only in the error case. (And passing tests seem more
important than performance.)
This makes tests elf/AArch64/rel-prel16-overflow.test and
elf/AArch64/rel-prel32-overflow.test pass on Windows: Both tests check that
atom error messages are emitted in a certain order, and on Windows they
happened to be emitted in a different order before this patch.
llvm-svn: 241988
There is no suitable basic block to sink instructions in loops without
exits. The only way an instruction in a loop without exits can be used
is as an incoming value to a PHI. In such cases, the incoming block for
the corresponding value is unreachable.
This fixes PR24013.
Differential Revision: http://reviews.llvm.org/D10903
llvm-svn: 241987
r238842 added the TargetRecip system for controlling use of reciprocal
estimates for sqrt and division using a set of parameters that can be set by
the frontend. Clang now supports a sophisticated -mrecip option, and this will
allow that option to effectively control the relevant code-generation
functionality of the PPC backend.
llvm-svn: 241985
This adds support for the 'nest' attribute, which allows the static chain
register to be set for functions calls under non-Darwin PPC/PPC64 targets. r11
is the chain register (which the PPC64 ELF ABI calls the "environment
pointer"). For indirect calls under PPC64 ELFv1, this would normally be loaded
from the function descriptor, but providing an explicit 'nest' parameter will
override that process and use the value provided.
This allows __builtin_call_with_static_chain to work as expected on PowerPC.
llvm-svn: 241984
This reverts commit r241962, as it was breaking all ARM buildbots.
It also reverts the two subsequent related commits:
r241974: "[ExecutionEngine] Add a static cast to the unittest for r241962 to suppress a warning."
r241973: "[ExecutionEngine] Remove cruft and fix a couple of warnings in the test case for r241962."
llvm-svn: 241983
r236894 caused PR23626 (Clang miscompiles webkit's base64 decoder), and was
reverted in r237984. This reapplies the patch with an additional test case for
PR23626 and the associated fix (both scales and offsets in the
BasicAliasAnalysis::constantOffsetHeuristic should initially be zero).
Patch by Nick White, thanks!
llvm-svn: 241981
The following functions are moved from the LoopVectorizer to VectorUtils:
- getGEPInductionOperand
- stripGetElementPtr
- getUniqueCastUse
- getStrideFromPointer
These used to be static functions in LoopVectorize, but will also be used by
the upcoming loop versioning LICM transformation.
Patch by Ashutosh Nema!
llvm-svn: 241980
This change adds new attribute called "argmemonly". Function marked with this attribute can only access memory through it's argument pointers. This attribute directly corresponds to the "OnlyAccessesArgumentPointees" ModRef behaviour in alias analysis.
Differential Revision: http://reviews.llvm.org/D10398
llvm-svn: 241979
This is used the canonicalize floating point values, which is useful for
implementing certain numeric primitives. See the LangRef changes for
the full details of its semantics.
llvm-svn: 241977
No in-tree alias analysis used this facility, and it was not called in
any particularly rigorous way, so it seems unlikely to be correct.
Note that one of the only stateful AA implementations in-tree,
GlobalsModRef is completely broken currently (and any AA passes like it
are equally broken) because Module AA passes are not effectively
invalidated when a function pass that fails to update the AA stack runs.
Ultimately, it doesn't seem like we know how we want to build stateful
AA, and until then trying to support and maintain correctness for an
untested API is essentially impossible. To that end, I'm planning to rip
out all of the update API. It can return if and when we need it and know
how to build it on top of the new pass manager and as part of *tested*
stateful AA implementations in the tree.
Differential Revision: http://reviews.llvm.org/D10889
llvm-svn: 241975
Drop 8 bytes off of `MCDwarfLoc` by restricting the `Isa`, `Column`, and
`Flags` members to appropriate sizes (from `DWARFDebugLine::Row`).
Saves a little over 0.5% off the heap of llc with no real functionality
change.
(I'm looking at `llc` memory usage on `verify-uselistorder.lto.opt.bc`;
see r236629 for details.)
llvm-svn: 241970
jGetLoadedDynamicLibrariesInfos. This packet is similar to
qXfer:libraries:read except that lldb supplies the number of solibs
that should be reported about, and the start address for the list
of them. At the initial process launch we'll read the full list
of solibs linked by the process -- at this point we could be using
qXfer:libraries:read -- but on subsequence solib-loaded notifications,
we'll be fetching a smaller number of solibs, often only one or two.
A typical Mac/iOS GUI app may have a couple hundred different
solibs loaded - doing all of the loads via memory reads takes
a couple of megabytes of traffic between lldb and debugserver.
Having debugserver summarize the load addresses of all the solibs
and sending it in JSON requires a couple of hundred kilobytes
of traffic. It's a significant performance improvement when
communicating over a slower channel.
This patch leaves all of the logic for loading the libraries
in DynamicLoaderMacOSXDYLD -- it only call over ot ProcesGDBRemote
to get the JSON result.
If the jGetLoadedDynamicLibrariesInfos packet is not implemented,
the normal technique of using memory read packets to get all of
the details from the target will be used.
<rdar://problem/21007465>
llvm-svn: 241964
It's possible for TagRedeclarations to involve decls without a name,
ie, anonymous enums. We hit some undefined behaviour if we bind these
null names to the reference here.
We never dereference the name, so it's harmless if it's null - make it
a pointer to allow that.
Fixes the Modules/submodules-merge-defs.cpp test under ubsan.
llvm-svn: 241963
Disallow all mutation of `MCSubtargetInfo` expect the feature bits.
Besides deleting the assignment operators -- which were dead "code" --
this restricts `InitMCProcessorInfo()` to subclass initialization
sequences, and exposes a new more limited function called
`setDefaultFeatures()` for use by the ARMAsmParser `.cpu` directive.
There's a small functional change here: ARMAsmParser used to adjust
`MCSubtargetInfo::CPUSchedModel` as a side effect of calling
`InitMCProcessorInfo()`, but I've removed that suspicious behaviour.
Since the AsmParser shouldn't be doing any scheduling, there shouldn't
be any observable change...
llvm-svn: 241961
Justin Bogner