Summary:
When X = 0 and Y = inf, the original code produces inf, but the transformed
code produces nan. So this transform (and its relatives) should only be
used when the no-infs-fp-math flag is explicitly enabled.
Also disable the transform using fmad (intermediate rounding) when unsafe-math
is not enabled, since it can reduce the precision of the result; consider this
example with binary floating point numbers with two bits of mantissa:
x = 1.01
y = 111
x * (y + 1) = 1.01 * 1000 = 1010 (this is the exact result; no rounding occurs at any step)
x * y + x = 1000.11 + 1.01 =r 1000 + 1.01 = 1001.01 =r 1000 (with rounding towards zero)
The example relies on rounding towards zero at least in the second step.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98578
Reviewers: RKSimon, tstellarAMD, spatel, arsenm
Subscribers: wdng, llvm-commits
Differential Revision: https://reviews.llvm.org/D26602
llvm-svn: 288506
Summary:
Unfortunately, there is no way to emit an llvm masked load/store in
clang without optimizations, and AVX enabled. Unsure how we should go
about making sure this test only runs if it's possible to execute AVX
code.
Reviewers: kcc, RKSimon, pgousseau
Subscribers: kubabrecka, dberris, llvm-commits
Differential Revision: https://reviews.llvm.org/D26506
llvm-svn: 288504
Summary: They are currently being parsed as %f14, %f16, and %f18.
Reviewers: venkatra, jyknight
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D27342
llvm-svn: 288503
Summary: Virtual method overrides of dependent types cannot be recognized unless
they are marked as override or final.
Exclude methods marked as final from check and add test.
Reviewers: sbenza, hokein, alexfh
Subscribers: malcolm.parsons, JDevlieghere, cfe-commits
Differential Revision: https://reviews.llvm.org/D27248
llvm-svn: 288502
When trying to vectorize trees that start at insertelement instructions
function tryToVectorizeList() uses vectorization factor calculated as
MinVecRegSize/ScalarTypeSize. But sometimes it does not work as tree
cost for this fixed vectorization factor is too high.
Patch tries to improve the situation. It tries different vectorization
factors from max(PowerOf2Floor(NumberOfVectorizedValues),
MinVecRegSize/ScalarTypeSize) to MinVecRegSize/ScalarTypeSize and tries
to choose the best one.
Differential Revision: https://reviews.llvm.org/D27215
llvm-svn: 288497
getTargetConstantBitsFromNode currently only extracts constant pool vector data, but it will need to be generalized to support broadcast and scalar constant pool data as well.
Converted Constant bit extraction and Bitset splitting to helper lambda functions.
llvm-svn: 288496
Summary:
This replaces all the uses of the __ANDROID_NDK__ define with __ANDROID__. This
is a preparatory step to remove our custom android toolchain file and rely on
the standard android NDK one instead, which does not provide this define.
Instead I rely, on __ANDROID__, which is set by the compiler.
I haven't yet removed the cmake variable with the same name, as we will need to
do something completely different there -- NDK toolchain defines
CMAKE_SYSTEM_NAME to Android, while our current one pretends it's linux.
Reviewers: tberghammer, zturner
Subscribers: danalbert, srhines, mgorny, lldb-commits
Differential Revision: https://reviews.llvm.org/D27305
llvm-svn: 288494
removed as a duplicate header search path
The commit r126167 started passing the First index into RemoveDuplicates, but
forgot to update 0 to First in the loop that looks for the duplicate. This
resulted in a bug where an -iquoted search path was incorrectly removed if you
passed in the same path into -iquote and more than one time into -isystem.
rdar://23991350
Differential Revision: https://reviews.llvm.org/D27298
llvm-svn: 288491
This allows us to delinearize code such as the one below, where the array
sizes are A[][2 * n] as there are n times two elements in the innermost
dimension. Alternatively, we could try to generate another dimension for the
struct in the innermost dimension, but as the struct has constant size,
recovering this dimension is easy.
struct com {
double Real;
double Img;
};
void foo(long n, struct com A[][n]) {
for (long i = 0; i < 100; i++)
for (long j = 0; j < 1000; j++)
A[i][j].Real += A[i][j].Img;
}
int main() {
struct com A[100][1000];
foo(1000, A);
llvm-svn: 288489
LLD used to take 11.73 seconds to link Clang. Now it is 6.94 seconds.
MSVC link takes 83.02 seconds. Note that ICF is enabled by default on
Windows, so a low latency ICF is more important than in ELF.
llvm-svn: 288487
It looks like the way dtrace works is
* The user creates .o files that reference magical symbol names.
* dtrace reads those files, collecs the info it needs and changes the
relocation to R_X86_64_NONE expecting the linker to ignore them.
llvm-svn: 288485
Associative sections are sections that need to be linked if their associated
sections are linked. Associative sections are used to append auxiliary data
such as debug info.
Previously, we compared all associative sections when comparing two comdat
sections. Because usually assocative sections are not mergeable sections,
we missed a lot of mergeable sections. MSVC linker doesn't seem to check
the identity of associative sections.
This patch makes LLD to ignore associative sections when doing ICF.
llvm-svn: 288483
r288228 seems to have regressed ICF performance in some cases in which
a lot of sections are actually mergeable. In r288228, I made a change
to create a Range object for each new color group. So every time we
split a group, we allocated and added a new group to a list of groups.
This patch essentially reverted r288228 with an improvement to
parallelize the original algorithm.
Now the ICF main loop is entirely allocation-free and lock-free.
Just like pre-r288228, we search for group boundaries by linear scan
instead of managing the information using Range class. r288228 was
neutral in performance-wise, and so is this patch.
I confirmed that this produces the exact same result as before
using chromium and clang as tests.
llvm-svn: 288480
After having built memory accesses we perform some additional transformations
on them to increase the chances that our delinearization guesses the right
shape. Only after these transformations, we take the assumptions that the
array shape we predict is such that no out-of-bounds memory accesses arise.
Before this change, the construction of the memory access, the access folding
that improves the represenation for certain parametric subscripts, and taking
the assumption was all done right after a memory access was created. In this
change we split this now into three separate iterations over all memory
accesses. This means only after all memory accesses have been built, we start
to canonicalize accesses, and to take assumptions. This split prepares for
future canonicalizations that must consider all memory accesses for deriving
additional beneficial transformations.
llvm-svn: 288479
Now that PointerType is no longer a SequentialType, all SequentialTypes
have an associated number of elements, so we can move that information to
the base class, allowing for a number of simplifications.
Differential Revision: https://reviews.llvm.org/D27122
llvm-svn: 288464
As proposed on llvm-dev:
http://lists.llvm.org/pipermail/llvm-dev/2016-October/106640.html
This is for a couple of reasons:
- Values of type PointerType are unlike the other SequentialTypes (arrays
and vectors) in that they do not hold values of the element type. By moving
PointerType we can unify certain aspects of how the other SequentialTypes
are handled.
- PointerType will have no place in the SequentialType hierarchy once
pointee types are removed, so this is a necessary step towards removing
pointee types.
Differential Revision: https://reviews.llvm.org/D26595
llvm-svn: 288462
This is a fairly reasonable bfd extension since there is one obvious value.
dtrace depends on this feature as it creates multiple absolute
symbols with the same value.
llvm-svn: 288461
Instead, expose whether the current type is an array or a struct, if an array
what the upper bound is, and if a struct the struct type itself. This is
in preparation for a later change which will make PointerType derive from
Type rather than SequentialType.
Differential Revision: https://reviews.llvm.org/D26594
llvm-svn: 288458
Summary:
The test introduced by rL288448 is currently failing because
unimportant but unexpected errors appear as output from a test compile
line. This patch looks for a more specific error message, in order to
avoid false positives.
Reviewers: jlebar
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D27328
Switch to more specific error
llvm-svn: 288453
In r266692, we made it possible to emit linkage names for just inlined
functions, putting the attribute on the abstract origin. Make sure we
don't think the linkage-name was already emitted on a declaration.
Differential Revision: http://reviews.llvm.org/D27320
llvm-svn: 288450
We try to include the headers of the module textually in this case, still
enforcing the modules semantic rules. In order to make that work, we need to
still track that we're entering and leaving the module. Also, if the module was
also marked as unavailable (perhaps because it was missing a file), we
shouldn't mark the module unavailable -- we don't need the module to be
complete if we're going to enter it textually.
llvm-svn: 288449
This fixes a bug that was introduced in rL287285. The bug made it
illegal to pass -fsanitize=address during CUDA compilation because the
CudaToolChain class was switched from deriving from the Linux toolchain
class to deriving directly from the ToolChain toolchain class. When
CudaToolChain derived from Linux, it used Linux's getSupportedSanitizers
method, and that method allowed ASAN, but when it switched to deriving
directly from ToolChain, it inherited a getSupportedSanitizers method
that didn't allow for ASAN.
This patch fixes that bug by creating a getSupportedSanitizers method
for CudaToolChain that supports ASAN.
This patch also fixes the test that checks that -fsanitize=address is
passed correctly for CUDA builds. That test didn't used to notice if an
error message was emitted, and that's why it didn't catch this bug when
it was first introduced. With the fix from this patch, that test will
now catch any similar bug in the future.
llvm-svn: 288448
Summary:
We were doing an optimization in the ThinLTO backends of importing
constant unnamed_addr globals unconditionally as a local copy (regardless
of whether the thin link decided to import them). This should be done in
the thin link instead, so that resulting exported references are marked
and promoted appropriately, but will need a summary enhancement to mark
these variables as constant unnamed_addr.
The function import logic during the thin link was trying to handle
this proactively, by conservatively marking all values referenced in
the initializer lists of exported global variables as also exported.
However, this only handled values referenced directly from the
initializer list of an exported global variable. If the value is itself
a constant unnamed_addr variable, we could end up exporting its
references as well. This caused multiple issues. The first is that the
transitively exported references weren't promoted. Secondly, some could
not be promoted/renamed (e.g. they had a section or other constraint).
recursively, instead of just adding the first level of initializer list
references to the ExportList directly.
Remove this optimization and the associated handling in the function
import backend. SPEC measurements indicate we weren't getting much
from it in any case.
Fixes PR31052.
Reviewers: mehdi_amini
Subscribers: krasin, llvm-commits
Differential Revision: https://reviews.llvm.org/D26880
llvm-svn: 288446
Since the spill is for the whole wave, these
don't have the swizzling problems that vector stores do
and a single 4-byte allocation is enough to spill a 64 element
register. This should reduce the number of spill instructions and
put all the spills for a register in the same cacheline.
This should save allocated private size, but for now it doesn't.
The extra slots are allocated for each component, but never used
because the frame layout is essentially finalized before frame
indices are replaced. For always using the scalar store path,
this should probably be moved into processFunctionBeforeFrameFinalized.
llvm-svn: 288445
After r256463, both the LHS and RHS now refer to the same variable. Before,
they referred to the member, the parameter respectively. Now GCC6's
-Wtautological-compare complains.
llvm-svn: 288444
Filipe Cabecinhas