Summary:
New flag causes crash reports to be written in the specified directory
rather than the temp directory.
Patch by Chijioke Kamanu.
Reviewers: hans, inglorion, rnk
Reviewed By: hans
Subscribers: zturner, hiraditya, llvm-commits, cfe-commits
Differential Revision: https://reviews.llvm.org/D48601
llvm-svn: 336604
When the parsing of the functions happens inside of the declare target
region, we may erroneously mark local variables as declare target
thought they are not. This attribute can be applied only to global
variables.
llvm-svn: 336592
This is part of an ongoing attempt at making 512 bit vectors illegal in the X86 backend type legalizer due to CPU frequency penalties associated with wide vectors on Skylake Server CPUs. We want the loop vectorizer to be able to emit IR containing wide vectors as intermediate operations in vectorized code and allow these wide vectors to be legalized to 256 bits by the X86 backend even though we are targetting a CPU that supports 512 bit vectors. This is similar to what happens with an AVX2 CPU, the vectorizer can emit wide vectors and the backend will split them. We want this splitting behavior, but still be able to use new Skylake instructions that work on 256-bit vectors and support things like masking and gather/scatter.
Of course if the user uses explicit vector code in their source code we need to not split those operations. Especially if they have used any of the 512-bit vector intrinsics from immintrin.h. And we need to make it so that merely using the intrinsics produces the expected code in order to be backwards compatible.
To support this goal, this patch adds a new IR function attribute "min-legal-vector-width" that can indicate the need for a minimum vector width to be legal in the backend. We need to ensure this attribute is set to the largest vector width needed by any intrinsics from immintrin.h that the function uses. The inliner will be reponsible for merging this attribute when a function is inlined. We may also need a way to limit inlining in the future as well, but we can discuss that in the future.
To make things more complicated, there are two different ways intrinsics are implemented in immintrin.h. Either as an always_inline function containing calls to builtins(can be target specific or target independent) or vector extension code. Or as a macro wrapper around a taget specific builtin. I believe I've removed all cases where the macro was around a target independent builtin.
To support the always_inline function case this patch adds attribute((min_vector_width(128))) that can be used to tag these functions with their vector width. All x86 intrinsic functions that operate on vectors have been tagged with this attribute.
To support the macro case, all x86 specific builtins have also been tagged with the vector width that they require. Use of any builtin with this property will implicitly increase the min_vector_width of the function that calls it. I've done this as a new property in the attribute string for the builtin rather than basing it on the type string so that we can opt into it on a per builtin basis and avoid any impact to target independent builtins.
There will be future work to support vectors passed as function arguments and supporting inline assembly. And whatever else we can find that isn't covered by this patch.
Special thanks to Chandler who suggested this direction and reviewed a preview version of this patch. And thanks to Eric Christopher who has had many conversations with me about this issue.
Differential Revision: https://reviews.llvm.org/D48617
llvm-svn: 336583
Add a number of builtins for __float128 Round To Odd.
This is the Clang portion of the builtins work.
Differential Revision: https://reviews.llvm.org/D47548
llvm-svn: 336579
In generic data-sharing mode we are allowed to not globalize local
variables that escape their declaration context iff they are declared
inside of the parallel region. We can do this because L2 parallel
regions are executed sequentially and, thus, we do not need to put
shared local variables in the global memory.
llvm-svn: 336567
I believe these have been broken since their introduction into clang.
I've enhanced the tests for these intrinsics to using a real rounding mode and checking all the intrinsic arguments instead of just the name.
llvm-svn: 336498
DanglingInternalBufferChecker now tracks use-after-free problems related
to the incorrect usage of std::basic_string::data().
Differential Revision: https://reviews.llvm.org/D48532
llvm-svn: 336497
Add a bug visitor to DanglingInternalBufferChecker that places a note
at the point where the dangling pointer was obtained. The visitor is
handed over to MallocChecker and attached to the report there.
Differential Revision: https://reviews.llvm.org/D48522
llvm-svn: 336495
Extend MallocBugVisitor to place a note at the point where objects with
AF_InternalBuffer allocation family are destroyed.
Differential Revision: https://reviews.llvm.org/D48521
llvm-svn: 336489
deprecated.
Add a -Wdeprecated warning for this in C++2a onwards. (In C++17 and
before, there isn't a reasonable alternative because [=,this] is
ill-formed.)
llvm-svn: 336480
This case occurs in the intrinsic headers so we should avoid emitting the mask in those cases.
Factor the code into a helper function to make this easy.
llvm-svn: 336472
This moves the LTO-specific code for outlining from ToolChains/Clang.cpp to
ToolChains/Darwin.cpp. Passing -mllvm flags isn't sufficient for making sure
that the specified pass will actually run in LTO. This makes sure that when
-moutline is passed, the MachineOutliner will actually be added to the LTO
pass pipeline as expected.
llvm-svn: 336471
We had the mask versions of the rounding intrinsics, but not one without masking.
Also change the rounding tests to not use the CUR_DIRECTION rounding mode.
llvm-svn: 336470
For some of the clauses the closing location erroneously points to the
beginning of the next clause rather than on the location of the closing
bracket of the clause.
llvm-svn: 336460
This patches adds support for passing -mcpu=native for AArch64. It will
get turned into the host CPU name, before we get the target features.
CPU = native is handled in a similar fashion in
getAArch64MicroArchFetauresFromMtune and getAArch64TargetCPU already.
Having a good test case for this is hard, as it depends on the host CPU
of the machine running the test. But we can check that native has been
replaced with something else.
When cross-compiling, we will get a CPU name from the host architecture
and get ` the clang compiler does not support '-mcpu=native'` as error
message, which seems reasonable to me.
Reviewers: rengolin, peter.smith, dlj, javed.absar, t.p.northover
Reviewed By: peter.smith
Tags: #clang
Differential Revision: https://reviews.llvm.org/D48931
llvm-svn: 336429
A Chromium developer reported a bug which turned out to be a mangling
collision between these two literals:
char s[] = "foo";
char t[32] = "foo";
They may look the same, but for the initialization of t we will (under
some circumstances) use a literal that's extended with zeros, and
both the length and those zeros should be accounted for by the mangling.
This actually makes the mangling code simpler: where it previously had
special logic for null terminators, which are not part of the
StringLiteral, that is now covered by the general algorithm.
(The problem was reported at https://crbug.com/857442)
Differential Revision: https://reviews.llvm.org/D48928
llvm-svn: 336415
The '%tu'/'%td' as formatting specifiers have been used to print out the
NSInteger/NSUInteger values for a long time. Typically their ABI matches, but that's
not the case on watchOS. The ABI difference boils down to the following:
- Regular 32-bit darwin targets (like armv7) use 'ptrdiff_t' of type 'int',
which matches 'NSInteger'.
- WatchOS arm target (armv7k) uses 'ptrdiff_t' of type 'long', which doesn't
match 'NSInteger' of type 'int'.
Because of this ABI difference these specifiers trigger -Wformat warnings only
for watchOS builds, which is really inconvenient for cross-platform code.
This patch avoids this -Wformat warning for '%tu'/'%td' and NS[U]Integer only,
and instead uses the new -Wformat-pedantic warning that JF introduced in
https://reviews.llvm.org/D47290. This is acceptable because Darwin guarantees that,
despite the watchOS ABI differences, sizeof(ptrdiff_t) == sizeof(NS[U]Integer),
and alignof(ptrdiff_t) == alignof(NS[U]Integer) so the warning is therefore noisy
for pedantic reasons.
I'll update public documentation to ensure that this behaviour is properly
communicated.
rdar://41739204
Differential Revision: https://reviews.llvm.org/D48852
llvm-svn: 336396
Shufflevector is easier to generate and matches what the backend pattern matches without relying on constant selects being turned into shuffles.
While I was there I also made the IR regular expressions a little stricter to ensure operand order on the shuffle.
llvm-svn: 336388
Implement support for MS-style PCH through headers.
This enables support for /Yc and /Yu where the through header is either
on the command line or included in the source. It replaces the current
support the requires the header also be specified with /FI.
This change adds a -cc1 option -pch-through-header that is used to either
start or stop compilation during PCH create or use.
When creating a PCH, the compilation ends after compilation of the through
header.
When using a PCH, tokens are skipped until after the through header is seen.
Patch By: mikerice
Differential Revision: https://reviews.llvm.org/D46652
llvm-svn: 336379
As brought up on cfe-commits[1], r334650 causes the dependency of the
out parameter to the __builtin_*_overflow functions to be ignored. The result
was a usage that was otherwise constexpr (both operands to the operation were
constexpr) would be evaluated, but the out parameter wouldn't be modified, so
it would still be 'undef'.
This patch correctly handles the return value of handleAssignment to ensure that
this value is properly considered/evaluated.
[1] http://lists.llvm.org/pipermail/cfe-commits/Week-of-Mon-20180702/233667.html
llvm-svn: 336364
This patch removes on optimization used with the TRUE/FALSE
predicates, as was suggested in https://reviews.llvm.org/D45616
for r335339.
The optimization was buggy, since r335339 used it also
for *_mask builtins, without actually applying the mask -- the
mask argument was just ignored.
Reviewers: craig.topper, uriel.k, RKSimon, andrew.w.kaylor, spatel, scanon, efriedma
Reviewed By: spatel
Differential Revision: https://reviews.llvm.org/D48715
llvm-svn: 336355
Add test cases with each predicate using the following
intrinsics:
_mm_cmp_pd
_mm_cmp_ps
_mm256_cmp_pd
_mm256_cmp_ps
_mm_cmp_pd_mask
_mm_cmp_ps_mask
_mm256_cmp_pd_mask
_mm256_cmp_ps_mask
_mm512_cmp_pd_mask
_mm512_cmp_ps_mask
_mm_mask_cmp_pd_mask
_mm_mask_cmp_ps_mask
_mm256_mask_cmp_pd_mask
_mm256_mask_cmp_ps_mask
_mm512_mask_cmp_pd_mask
_mm512_mask_cmp_ps_mask
Some of these are marked with FIXME, as there is bug in lowering
e.g. _mm512_mask_cmp_ps_mask.
llvm-svn: 336346
Update clang to treat fp128 as a valid base type for homogeneous aggregate
passing and returning.
Differential Revision: https://reviews.llvm.org/D48044
llvm-svn: 336308
Summary: In the provided test case the PathDiagnostic compare function was not able to find a difference.
Reviewers: xazax.hun, NoQ, dcoughlin, george.karpenkov
Reviewed By: george.karpenkov
Subscribers: a_sidorin, szepet, rnkovacs, a.sidorin, mikhail.ramalho, cfe-commits
Differential Revision: https://reviews.llvm.org/D48474
llvm-svn: 336275
Summary:
Can happen when getConstructorName is called on invalid decls,
specifically the ones that do not have the injected class name.
Reviewers: bkramer, rsmith
Reviewed By: rsmith
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D48880
llvm-svn: 336244
merged function definitions; also merge functions with deduced return
types.
This seems like two independent fixes, but unfortunately they are hard
to separate because it's challenging to reliably test either one of them
without also testing the other.
A complication arises with deduced return type support: we need the type
of the function in order to know how to merge it, but we can't load the
actual type of the function because it might reference an entity
declared within the function (and we need to have already merged the
function to correctly merge that entity, which we would need to do to
determine if the function types match). So we instead compare the
declared function type when merging functions, and defer loading the
actual type of a function with a deduced type until we've finished
loading and merging the function.
This reverts r336175, reinstating r336021, with one change (for PR38015):
we look at the TypeSourceInfo of the first-so-far declaration of each
function when considering whether to merge two functions. This works
around a problem where the calling convention in the TypeSourceInfo for
subsequent redeclarations may not match if it was implicitly adjusted.
llvm-svn: 336240
If a function has multiple format_arg attributes, clang only considers
the first it finds (because AttributeLists are in reverse order, not
necessarily the textually first) and ignores all others.
Loop over all FormatArgAttr to print warnings for all declared
format_arg attributes.
For instance, libintl's ngettext (select plural or singular version of
format string) has two __format_arg__ attributes.
Differential Revision: https://reviews.llvm.org/D48734
llvm-svn: 336239
The constexpr evaluator was erroring out because these templates weren't
defined. Despite being used in a discarded statement, we still need to constexpr
evaluate them, which means that we need to instantiate them. Fixes PR37585.
Differential revision: https://reviews.llvm.org/D48322
llvm-svn: 336233
Summary:
Scudo works on PPC64 as is, so mark the architecture as supported for it. This
will also require a change to config-ix.cmake on the compiler-rt side.
Update the tests accordingly.
Reviewers: eugenis, alekseyshl
Reviewed By: alekseyshl
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D48833
llvm-svn: 336202
This caused test failures in 32-bit builds (PR38015).
> merged function definitions; also merge functions with deduced return
> types.
>
> This seems like two independent fixes, but unfortunately they are hard
> to separate because it's challenging to reliably test either one of them
> without also testing the other.
>
> A complication arises with deduced return type support: we need the type
> of the function in order to know how to merge it, but we can't load the
> actual type of the function because it might reference an entity
> declared within the function (and we need to have already merged the
> function to correctly merge that entity, which we would need to do to
> determine if the function types match). So we instead compare the
> declared function type when merging functions, and defer loading the
> actual type of a function with a deduced type until we've finished
> loading and merging the function.
llvm-svn: 336175
Summary:
When clang required to infer target os version from --target option and
the os version is not specified in targets, check the host triple. If the
host and target are both macOS, use host triple to infer target os
version.
rdar://problem/41651999
Reviewers: arphaman, dexonsmith
Reviewed By: arphaman
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D48849
llvm-svn: 336168
Summary:
Emmiting new intrinsic that strips invariant.groups to make
devirtulization sound, as described in RFC: Devirtualization v2.
Reviewers: rjmccall, rsmith, amharc, kuhar
Subscribers: llvm-commits, cfe-commits
Differential Revision: https://reviews.llvm.org/D47299
Co-authored-by: Krzysztof Pszeniczny <krzysztof.pszeniczny@gmail.com>
llvm-svn: 336137
MSVC limits char16_t and char32_t string literal names to 32 bytes of character
data, not to 32 characters. wchar_t string literal names on the other hand can
get up to 64 bytes of character data.
https://reviews.llvm.org/D48781
llvm-svn: 336097
Bob Haarman