x86 scalar select-of-constants (Cond ? C1 : C2) combining/lowering is a mess
with missing optimizations. We handle some patterns, but miss logical variants.
To clean that up, we should convert all select-of-constants to logic/math and
enhance the combining for the expected patterns from that. DAGCombiner already
has the foundation to allow the transforms, so we just need to fill in the holes
for x86 math op lowering. Selecting 0 or -1 needs extra attention to produce the
optimal code as shown here.
Attempt to verify that all of these IR forms are logically equivalent:
http://rise4fun.com/Alive/plxs
Earlier steps in this series:
rL306040
rL306072
Differential Revision: https://reviews.llvm.org/D34652
llvm-svn: 307404
- Check that we have invariant accesses.
- Use `-polly-use-llvm-names` for better names in the test.
- Rename test function to `f` for brevity.
llvm-svn: 307401
Prior to this commit both of the added test cases were passing. However, in the
latter case (test7) we were doing a lot more work to arrive at the same answer
(i.e., we were using isImpliedCondMatchingOperands() to determine the
implication.).
llvm-svn: 307400
- This already works, but add this to ensure that there is no
regressions when I expand the invariant load hoisting ability of
`PPCGCodeGeneration`.
llvm-svn: 307398
Today the safepoint IR verifier catches some unrelocated uses of base
pointers that are actually valid.
With this change, we narrow down the set of false positives.
Specifically, the verifier knows about compares to null and compares
between 2 unrelocated pointers.
Reviewed by: skatkov
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D35057
llvm-svn: 307392
Summary:
This replaces the static functions used for creating
NativeProcessProtocol instances with a factory pattern, and modernizes
the interface of the new class in the process -- I use llvm::Expected
instead of the Status+value combo. I also move some of the common code
(like the Delegate registration into the base class). The new
arrangement has multiple benefits:
- it removes the NativeProcess*** dependency from Process/gdb-remote
(which for example means that liblldb no longer pulls in this code).
- it enables unit testing of the GDBRemoteCommunicationServerLLGS class
(by providing a mock Native Process).
- serves as another example on how to use the llvm::Expected class (I
couldn't get rid of the Initialize-type functions completely here
because of the use of shared_from_this, but that's the next thing on
my list here)
Tests still pass on Linux and I've made sure NetBSD compiles after this.
Reviewers: zturner, eugene, krytarowski
Subscribers: srhines, lldb-commits, mgorny
Differential Revision: https://reviews.llvm.org/D33778
llvm-svn: 307390
Summary:
This change gives a 0.89% speed on execution time, a 0.94% improvement
in benchmark scores and a 0.62% increase in binary size on a Cortex-A57.
These numbers are the geomean results on a wide range of benchmarks from
the test-suite, SPEC2000, SPEC2006 and a range of proprietary suites.
The software optimization guide for the Cortex-A57 recommends 16 byte
branch alignment.
Reviewers: t.p.northover, mcrosier, javed.absar, kristof.beyls, sbaranga
Reviewed By: kristof.beyls
Subscribers: aemerson, rengolin, llvm-commits
Differential Revision: https://reviews.llvm.org/D34954
llvm-svn: 307389
deployment target if the SDK is newer than the system
This commit reverts the revert commit r305891. Now the change from r305678
should be correct because `llvm::sys::getProcessTriple` now returns the correct
macOS version of the system after the LLVM change r307372.
Original commit message:
This commit improves the driver by making sure that it picks the system version
for the deployment target when the version of the macOS SDK is newer than the
system version.
rdar://29449467
Differential Revision: https://reviews.llvm.org/D34175
llvm-svn: 307388
Check the `-mlong-calls` command line option and pass the `long-calls`
feature flag to the backend. Handling of this feature flag in the backend
needs to be implemented by a separate commit.
llvm-svn: 307386
the serialised diagnostics
Clang should avoid calling report_fatal_error when the file with the serialised
diagnostics is empty. This commit changes Clang's serialised diagnostic reader,
now it reports an appropriate error instead of crashing.
rdar://31939877
Differential Revision: https://reviews.llvm.org/D35069
llvm-svn: 307384
Summary:
This change gives a 0.34% speed on execution time, a 0.61% improvement
in benchmark scores and a 0.57% increase in binary size on a Cortex-A72.
These numbers are the geomean results on a wide range of benchmarks from
the test-suite, SPEC2000, SPEC2006 and a range of proprietary suites.
The software optimization guide for the Cortex-A72 recommends 16 byte
branch alignment.
Reviewers: t.p.northover, kristof.beyls, rengolin, sbaranga, mcrosier, javed.absar
Reviewed By: kristof.beyls
Subscribers: llvm-commits, aemerson
Differential Revision: https://reviews.llvm.org/D34961
llvm-svn: 307380
Summary:
For the following code: `modernize-use-override` generates a replacement with incorrect location.
```
struct IntPair
{
int first, second;
};
struct A
{
virtual void il(IntPair);
};
struct B : A
{
void il(IntPair p = {1, (2 + 3)}) {};
// Generated Fixit: void il(IntPair p = override {1, (2 + 3)}) {};
// Should be: void il(IntPair p = {1, (2 + 3)}) override {};
};
```
This fixes that and adds a unit test.
Reviewers: alexfh, aaron.ballman, hokein
Reviewed By: alexfh
Subscribers: JDevlieghere, xazax.hun, cfe-commits
Tags: #clang-tools-extra
Patch by Victor Gao!
Differential Revision: https://reviews.llvm.org/D35078
llvm-svn: 307379
In preparation for the addition of rangeThunks() calculate the addresses
of all the inputSections so that ThunkSections can be inserted at the right
place.
Differential Revision: https://reviews.llvm.org/D34688
llvm-svn: 307373
the system's version of macOS
sys::getProcessTriple returns LLVM_HOST_TRIPLE, whose system version might not
be the actual version of the system on which the compiler running. This commit
ensures that, for macOS, sys::getProcessTriple returns a triple with the
system's macOS version.
rdar://33177551
Differential Revision: https://reviews.llvm.org/D34446
llvm-svn: 307372
This is relative to PR33173,
Previously if something wrong happened on DWARF parsers side during parsing
object for building gdb index (like was in PR: unsupported relocation)
then LLD continued and finished the link. DWARF parsers sure showed error
message on their side, but that is all.
Patch changes behavior to fail the link in this case and show more
detailed message.
Differential revision: https://reviews.llvm.org/D34814
llvm-svn: 307370
requirements in protocol/class/category declarations
The unguarded availability warnings in the protocol requirements of a protocol
/class/category declaration can be avoided. This matches the behaviour of
Swift's diagnostics. The warnings for deprecated/unavailable protocols are
preserved.
rdar://33156429
Differential Revision: https://reviews.llvm.org/D35061
llvm-svn: 307368
The assignAddresses() function accumulates state in the LinkerScript that
prevents it from being called multiple times. This change moves the state
into a separate structure AddressState that is created at the start of the
function and disposed of at the end.
CurAddressState is used rather than passing a reference to the state as a
parameter to the functions used by assignAddresses(). This is because the
getSymbolValue function needs to be executed in the context of AddressState
but it is stored in ScriptParser when AddressState is not available.
The AddressState is also used in a limited context by processCommands()
Differential Revision: https://reviews.llvm.org/D34345
llvm-svn: 307367
We lower to a sequence consisting of:
- MOVi 0 into a register
- VCMPS to do the actual comparison and set the VFP flags
- FMSTAT to move the flags out of the VFP unit
- MOVCCi to either use the "zero register" that we have previously set
with the MOVi, or move 1 into the result register, based on the values
of the flags
As was the case with soft-float, for some predicates (one, ueq) we
actually need two comparisons instead of just one. When that happens, we
generate two VCMPS-FMSTAT-MOVCCi sequences and chain them by means of
using the result of the first MOVCCi as the "zero register" for the
second one. This is a bit overkill, since one comparison followed by
two non-flag-setting conditional moves should be enough. In any case,
the backend manages to CSE one of the comparisons away so it doesn't
matter much.
Note that unlike SelectionDAG and FastISel, we always use VCMPS, and not
VCMPES. This makes the code a lot simpler, and it also seems correct
since the LLVM Lang Ref defines simple true/false returns if the
operands are QNaN's. For SNaN's, even VCMPS throws an Invalid Operand
exception, so they won't be slipping through unnoticed.
Implementation-wise, this introduces a template so we can share the same
code that we use for handling integer comparisons, since the only
differences are in the details (exact opcodes to be used etc). Hopefully
this will be easy to extend to s64 G_FCMP.
llvm-svn: 307365
This fixes last testcase provided in PR28414.
In short issue is next: when we had X@@Version symbol in object A,
we did not resolve it to X early. Then when in another object B
we had reference to undefined X, symbol X from archive was fetched.
Since both archive and object A contains another symbol Z, duplicate
symbol definition was triggered as a result.
Correct behavior is to use X@@Version from object A instead and do not fetch
any symbols from archive.
Differential revision: https://reviews.llvm.org/D35059
llvm-svn: 307364
Some of our emitters were using the name of the Target to reference things that were created by others emitters using Namespace.
Apparently all targets have the same Target name as their instruction and register Namespace field?
Someone on IRC had a target that didn't do this and was getting build errors. This patch is a necessary, but maybe not sufficient fix.
llvm-svn: 307358
r283051 added some functions to cmath (in namespace std) that have the
same name as functions in math.h (in the global namespace). Clang's
limited support for `-fdelayed-template-parsing` chokes on this. Rename
the ones in `cmath` and their uses in `complex` and the test.
rdar://problem/32848355
llvm-svn: 307357
Based strictly on the name, this seems to have something to do
width edit & continue. The goal of this patch has nothing to do
with supporting edit and continue though. msvc link.exe writes
very basic information into this area even when *not* compiling
with support for E&C, and so the goal here is to bring lld-link
to parity. Since we cannot know what assumptions standard tools
make about the content of PDB files, we need to be as close as
possible.
This ECNames data structure is a standard PDB string hash table.
link.exe puts a single string into this hash table, which is the
full path to the PDB file on disk. It then references this string
from the module descriptor for the compiler generated `* Linker *`
module.
With this patch, lld-link will generate the exact same sequence of
bytes as MSVC link for this subsection for a given object file
input (as reported by `llvm-pdbutil bytes -ec`).
llvm-svn: 307356
- Instead of running with -O0, we enable the highest optimization level, but
then disable optimizations. This ensures that possibly important metadata
is still emitted.
- Update the code for attribute removal to work with latest LLVM
- Do not cut an arbitrary number of lines from the LL file. It is undocumented
why this was needed at the first place, and such a feature is likely to
break with trivial IR changes that may come in the future.
llvm-svn: 307355
Tom Stellard