Remove target features crypto for Cortex-R82, because it doesn't have any, and
add LSE which was missing while we are at it.
This also removes crypto from the v8-R architecture description because that
aligns better with GCC and so far none of the R-cores have implemented crypto,
so is probably a more sensible default.
Differential Revision: https://reviews.llvm.org/D91994
... and give more guidance to users.
If specifying -msve-vector-bits on a non-SVE target, clang would say:
error: '-msve-vector-bits' is not supported without SVE enabled
1. The driver lacks logic for "implied features".
This would result in this error being raised for -march=...+sve2,
even though +sve2 implies +sve.
2. Feature implication is well modelled in LLVM, so push the error down
the stack.
3. Hint to the user what flag they need to consider setting.
Now clang fails later, when the feature is used, saying:
aarch64-sve-vector-bits.c:42:41: error: 'arm_sve_vector_bits' attribute is not supported on targets missing 'sve'; specify an appropriate -march= or -mcpu=
typedef svint32_t noflag __attribute__((arm_sve_vector_bits(256)));
Move clang/test/Sema/{neon => arm}-vector-types-support.c and put tests for
this warning together in one place.
Reviewed By: sdesmalen
Differential Revision: https://reviews.llvm.org/D92487
When the evaluator encounters an error-dependent returnstmt, before this patch
it returned a ESR_Returned without setting the result, the callsides think this
is a successful execution, and try to access the Result which causes the crash.
The fix is to always return failed as we don't know the result of the
error-dependent return stmt.
Differential Revision: https://reviews.llvm.org/D92969
This patch implements amx programming model that discussed in llvm-dev
(http://lists.llvm.org/pipermail/llvm-dev/2020-August/144302.html).
Thank Hal for the good suggestion in the RA. The fast RA is not in the patch yet.
This patch implemeted 7 components.
1. The c interface to end user.
2. The AMX intrinsics in LLVM IR.
3. Transform load/store <256 x i32> to AMX intrinsics or split the
type into two <128 x i32>.
4. The Lowering from AMX intrinsics to AMX pseudo instruction.
5. Insert psuedo ldtilecfg and build the def-use between ldtilecfg to amx
intruction.
6. The register allocation for tile register.
7. Morph AMX pseudo instruction to AMX real instruction.
Change-Id: I935e1080916ffcb72af54c2c83faa8b2e97d5cb0
Differential Revision: https://reviews.llvm.org/D87981
It should specify --sysroot to test the paths of crt1.o/crti.o/crtbegin.o.
For a user who enable VE but do not actually have VE sysroot,
the "nld" command line will have bare "crt1.o" "crti.o" ... "crtbegin.o"
The new PM is considered stable and many downstream groups have adopted it (some
have adopted it for more than two years). Add -f[no-]legacy-pass-manager to reflect the
fact that it is no longer experimental and the legacy pass manager is something we strive to retire.
In the future, when the legacy PM eventually goes away,
-fno-experimental-new-pass-manager and -flegacy-pass-manager will be removed.
This patch also changes -f[no-]legacy-pass-manager to pass `-plugin-opt={new,legacy}-pass-manager` to the linker (supported by both ld.lld and LLVMgold.so) when -flto/-flto=thin is specified
Reviewed By: aeubanks, rsmith
Differential Revision: https://reviews.llvm.org/D92915
This is needed for CUDA compilation where NVPTX back-end only supports DWARF2,
but host compilation should be allowed to use newer DWARF versions.
Differential Revision: https://reviews.llvm.org/D92617
This attribute permits a typedef to be associated with a class template
specialization as a preferred way of naming that class template
specialization. This permits us to specify that (for example) the
preferred way to express 'std::basic_string<char>' is as 'std::string'.
The attribute is applied to the various class templates in libc++ that have
corresponding well-known typedef names.
This is a re-commit. The previous commit was reverted because it exposed
a pre-existing bug that has since been fixed / worked around; see
PR48434.
Differential Revision: https://reviews.llvm.org/D91311
Ensure that we can deserialize a TypedefType even while in the middle of
deserializing its TypedefDecl, by removing the need to look at the
TypedefDecl while constructing the TypedefType.
This fixes all the currently-known failures for PR48434, but it's not a
complete fix, because we can still trigger deserialization cycles, which
are not supposed to happen.
Swiftcall does it's own target-independent argument type classification,
since it is not designed to be ABI compatible with anything local on the
target that isn't LLVM-based. This means it never uses inalloca.
However, we have duplicate logic for checking for inalloca parameters
that runs before call argument setup. This logic needs to know ahead of
time if inalloca will be used later, and we can't move the
CGFunctionInfo calculation earlier.
This change gets the calling convention from either the
FunctionProtoType or ObjCMethodDecl, checks if it is swift, and if so
skips the stackbase setup.
Depends on D92883.
Differential Revision: https://reviews.llvm.org/D92944
Currently when -gsplit-dwarf is specified (could be buried in a build system),
there is no convenient way to cancel debug fission without affecting the debug
information amount (all of -g0, -g1 -fsplit-dwarf-inlining and -gline-directives-only
can, but they affect the debug information amount).
Reviewed By: #debug-info, dblaikie
Differential Revision: https://reviews.llvm.org/D92809
RFC: http://lists.llvm.org/pipermail/cfe-dev/2020-May/065430.html
Agreement from GCC: https://sourceware.org/pipermail/gcc-patches/2020-May/545688.html
g_flags_Group options generally don't affect the amount of debugging
information. -gsplit-dwarf is an exception. Its order dependency with
other gN_Group options make it inconvenient in a build system:
* -g0 -gsplit-dwarf -> level 2
-gsplit-dwarf "upgrades" the amount of debugging information despite
the previous intention (-g0) to drop debugging information
* -g1 -gsplit-dwarf -> level 2
-gsplit-dwarf "upgrades" the amount of debugging information.
* If we have a higher-level -gN, -gN -gsplit-dwarf will supposedly decrease the
amount of debugging information. This happens with GCC -g3.
The non-orthogonality has confused many users. GCC 11 will change the semantics
(-gsplit-dwarf no longer implies -g2) despite the backwards compatibility break.
This patch matches its behavior.
New semantics:
* If there is a g_Group, allow split DWARF if useful
(none of: -g0, -gline-directives-only, -g1 -fno-split-dwarf-inlining)
* Otherwise, no-op.
To restore the original behavior, replace -gsplit-dwarf with -gsplit-dwarf -g.
Reviewed By: dblaikie
Differential Revision: https://reviews.llvm.org/D80391
Currently, -ftime-report + new pass manager emits one line of report for each
pass run. This potentially causes huge output text especially with regular LTO
or large single file (Obeserved in private tests and was reported in D51276).
The behaviour of -ftime-report + legacy pass manager is
emitting one line of report for each pass object which has relatively reasonable
text output size. This patch adds a flag `-ftime-report=` to control time report
aggregation for new pass manager.
The flag is for new pass manager only. Using it with legacy pass manager gives
an error. It is a driver and cc1 flag. `per-pass` is the new default so
`-ftime-report` is aliased to `-ftime-report=per-pass`. Before this patch,
functionality-wise `-ftime-report` is aliased to `-ftime-report=per-pass-run`.
* Adds an boolean variable TimePassesHandler::PerRun to control per-pass vs per-pass-run.
* Adds a new clang CodeGen flag CodeGenOptions::TimePassesPerRun to work with the existing CodeGenOptions::TimePasses.
* Remove FrontendOptions::ShowTimers, its uses are replaced by the existing CodeGenOptions::TimePasses.
* Remove FrontendTimesIsEnabled (It was introduced in D45619 which was largely reverted.)
Differential Revision: https://reviews.llvm.org/D92436
Function Parser::ParseAvailabilityAttribute checks that the message string of
an availability attribute is not a wide string literal. Test case
clang/test/Parser/attr-availability.c specifies that a string literal is
expected.
The code checked that the first token in a string concatenation is a string
literal, and then that the concatenated string consists of 1-byte characters.
On a target where wide character is 1 byte, a string concatenation "a" L"b"
passes both those checks, but L"b" alone is rejected. More generally, "a" u8"b"
passes the checks, but u8"b" alone is rejected.
So check isAscii() instead of character size.
This test shows we're in some cases not getting strictfp information from
the AST. Correct that.
Differential Revision: https://reviews.llvm.org/D92596
Sometimes people get minimal crash reports after a UBSAN incident. This change
tags each trap with an integer representing the kind of failure encountered,
which can aid in tracking down the root cause of the problem.
This patch adds tests that showcase a behavior that is currently buggy.
Fix in a follow-up patch.
Differential Revision: https://reviews.llvm.org/D91269
This change exposed a pre-existing issue with deserialization cycles
caused by a combination of attributes and template instantiations
violating the deserialization ordering restrictions; see PR48434 for
details.
A previous commit attempted to work around PR48434, but appears to have
only been a partial fix, and fixing this properly seems non-trivial.
Backing out for now to unblock things.
This reverts commit 98f76adf4e and
commit a64c26a47a.
Instruction darn was introduced in ISA 3.0. It means 'Deliver A Random
Number'. The immediate number L means:
- L=0, the number is 32-bit (higher 32-bits are all-zero)
- L=1, the number is 'conditioned' (processed by hardware to reduce bias)
- L=2, the number is not conditioned, directly from noise source
GCC implements them in three separate intrinsics: __builtin_darn,
__builtin_darn_32 and __builtin_darn_raw. This patch implements the
same intrinsics. And this change also addresses Bugzilla PR39800.
Reviewed By: steven.zhang
Differential Revision: https://reviews.llvm.org/D92465
These lit tests now requires amdgpu-registered-target since they
use clang driver and clang driver passes an LLVM option which
is available only if amdgpu target is registered.
Change-Id: I2df31967409f1627fc6d342d1ab5cc8aa17c9c0c
This is really just a workaround for a more fundamental issue in the way
we deserialize attributes. See PR48434 for details.
Also fix tablegen code generator to produce more correct indentation to
resolve buildbot issues with -Werror=misleading-indentation firing
inside the generated code.
Committing on behalf of thejh (Jann Horn).
As part of this change, one existing test case has to be adjusted
because it accidentally stripped the NoDeref attribute without
getting caught.
Depends on D92140
Differential Review: https://reviews.llvm.org/D92141
Committing on behalf of thejh (Jann Horn).
Given an attribute((noderef)) pointer "p" to the struct
struct s { int a[2]; };
ensure that the following expressions are treated the same way by the
noderef logic:
p->a
(*p).a
Until now, the first expression would be treated correctly (nothing is
added to PossibleDerefs because CheckMemberAccessOfNoDeref() bails out
on array members), but the second expression would incorrectly warn
because "*p" creates a PossibleDerefs entry.
Handle this case the same way as for the AddrOf operator.
Differential Revision: https://reviews.llvm.org/D92140
This attributes specifies how (or if) a given function or method will be
imported into a swift async method. rdar://70111252
Differential revision: https://reviews.llvm.org/D92742
_Nullable_result generally like _Nullable, except when being imported into a
swift async method. rdar://70106409
Differential revision: https://reviews.llvm.org/D92495
Such fields will likely have offset zero making
__sanitizer_dtor_callback poisoning wrong regions.
E.g. it can poison base class member from derived class constructor.
Differential Revision: https://reviews.llvm.org/D92727
This attribute permits a typedef to be associated with a class template
specialization as a preferred way of naming that class template
specialization. This permits us to specify that (for example) the
preferred way to express 'std::basic_string<char>' is as 'std::string'.
The attribute is applied to the various class templates in libc++ that have
corresponding well-known typedef names.
Differential Revision: https://reviews.llvm.org/D91311
As reported in PR48177, the type-deduction extraction ends up going into
an infinite loop when the type referred to has a recursive definition.
This stops recursing and just substitutes the type-source-info the
TypeLocBuilder identified when transforming the base.
When we annotating a function header so that it could be used by other
TU, we also need to make sure the function is parsed correctly within
the same TU. So if we can find the function's implementation,
ignore the annotations, otherwise, false positive would occur.
Move the escape by value case to post call and do not escape the handle
if the function is inlined and we have analyzed the handle.
Differential Revision: https://reviews.llvm.org/D91902
Emit error for use of 128-bit integer inside device code had been
already implemented in https://reviews.llvm.org/D74387. However,
the error is not emitted for SPIR64, because for SPIR64, hasInt128Type
return true.
hasInt128Type: is also used to control generation of certain 128-bit
predefined macros, initializer predefined 128-bit integer types and
build 128-bit ArithmeticTypes. Except predefined macros, only the
device target is considered, since error only emit when 128-bit
integer is used inside device code, the host target (auxtarget) also
needs to be considered.
The change address:
1. (SPIR.h) Correct hasInt128Type() for SPIR targets.
2. Sema.cpp and SemaOverload.cpp: Add additional check to consider host
target(auxtarget) when call to hasInt128Type. So that __int128_t
and __int128() are allowed to avoid error when they used outside
device code.
3. SemaType.cpp: add check for SYCLIsDevice to delay the error message.
The error will be emitted if the use of 128-bit integer in the device
code.
Reviewed By: Johannes Doerfert and Aaron Ballman
Differential Revision: https://reviews.llvm.org/D92439
As Power9 introduced hardware support for IEEE quad-precision FP type,
the feature should be enabled by default on Power9 or newer targets.
Reviewed By: steven.zhang
Differential Revision: https://reviews.llvm.org/D90213
Currently, Baremetal toolchain requires user to pass a sysroot location
using a --sysroot flag. This is not very convenient for the user. It also
creates problem for toolchain vendors who don't have a fixed location to
put the sysroot bits.
Clang does provide 'DEFAULT_SYSROOT' which can be used by the toolchain
builder to provide the default location. But it does not work if toolchain
is targeting multiple targets e.g. arm-none-eabi/riscv64-unknown-elf which
clang is capable of doing.
This patch tries to solve this problem by providing a default location of
the toolchain if user does not explicitly provides --sysroot. The exact
location and name can be different but it should fulfill these conditions:
1. The sysroot path should have a target triple element so that multi-target
toolchain problem (as I described above) could be addressed.
2. The location should not be $TOP/$Triple as this is used by gcc generally
and will be a problem for installing both gcc and clang based toolchain at
the same location.
Reviewed By: jroelofs
Differential Revision: https://reviews.llvm.org/D92677
in their corresponding class interfaces
Categories that add protocol conformances to classes with direct members should prohibit protocol
conformances when the methods/properties that the protocol expects are actually declared as 'direct' in the class.
Differential Revision: https://reviews.llvm.org/D92602
The swift_async_name attribute provides a name for a function/method that can be used
to call the async overload of this method from Swift. This name specified in this attribute
assumes that the last parameter in the function/method its applied to is removed when
Swift invokes it, as the the Swift's await/async transformation implicitly constructs the callback.
Differential Revision: https://reviews.llvm.org/D92355
The swift_attr attribute is a generic annotation attribute that's not used by clang,
but is used by the Swift compiler. The Swift compiler can use these annotations to provide
various syntactic and semantic sugars for the imported Objective-C API declarations.
Differential Revision: https://reviews.llvm.org/D92354
Currently we have a diagnostic that catches the other storage class specifies for the range based for loop declaration but we miss the thread_local case. This changes adds a diagnostic for that case as well.
Differential Revision: https://reviews.llvm.org/D92671
The variables used in atomic construct should be captured in outer
task-based regions implicitly. Otherwise, the compiler will crash trying
to find the address of the local variable.
Differential Revision: https://reviews.llvm.org/D92682
Baremetal toolchain add Driver.SysRoot/include to the system include
paths without checking if Driver.SysRoot is empty. This resulted in
"-internal-isystem" "include" in the command. This patch adds check for
empty sysroot.
Reviewed By: jroelofs
Differential Revision: https://reviews.llvm.org/D92176
Compiler needs to convert some of the loop iteration
variables/conditions to different types for better codegen and it may
lead to spurious warning messages about implicit signed/unsigned
conversions.
Differential Revision: https://reviews.llvm.org/D92655
552c6c2 removed support for promoting VLAs to constant arrays when the bounds
isn't an ICE, since this can result in miscompiling a conforming program that
assumes that the array is a VLA. Promoting VLAs for fields is still supported,
since clang doesn't support VLAs in fields, so no conforming program could have
a field VLA.
This change is really disruptive, so this commit carves out two more cases
where we promote VLAs which can't miscompile a conforming program:
- When the VLA appears in an ivar -- this seems like a corollary to the field thing
- When the VLA has an initializer -- VLAs can't have an initializer
Differential revision: https://reviews.llvm.org/D90871
The static variable causes it only initialized once and take
the same value for different GPU archs, whereas they
may be different for different GPU archs, e.g. when
there are both gfx900 and gfx1010.
Removing static fixes that.
Differential Revision: https://reviews.llvm.org/D92628
template-parameter-list in a lambda.
This implements one of the missing parts of P0857R0. Mark it as not done
on the cxx_status page given that it's still incomplete.
This also teaches MachO writers/readers about the MachO cpu subtype,
beyond the minimal subtype reader support present at the moment.
This also defines a preprocessor macro to allow users to distinguish
__arm64__ from __arm64e__.
arm64e defaults to an "apple-a12" CPU, which supports v8.3a, allowing
pointer-authentication codegen.
It also currently defaults to ios14 and macos11.
Differential Revision: https://reviews.llvm.org/D87095
shouldRTTIBeUnique() returns false for iOS64CXXABI, which causes
RTTI objects to be emitted hidden. Update two tests that didn't
expect this to happen for the default triple.
Also rename iOS64CXXABI to AppleARM64CXXABI, since it's used for
arm64-apple-macos triples too.
Part of PR46644.
Differential Revision: https://reviews.llvm.org/D91904
We have a plan to add libcxx and libcxxabi for VE. In order to do so,
we need to compile cxx source code with bootstarapped header files.
This patch adds such expected path to make clang++ work, at least
not crash at the startup. Add regression test for that, also.
Reviewed By: simoll
Differential Revision: https://reviews.llvm.org/D92386
Fix bogus diagnostics that would get confused and think a "no viable
fuctions" case was an "undeclared identifiers" case, resulting in an
incorrect diagnostic preceding the correct one. Use overload resolution
to determine which function we should select when we can find call
candidates from a dependent base class. Make the diagnostics for a call
that could call a function from a dependent base class more specific,
and use a different diagnostic message for the case where the call
target is instead declared later in the same class. Plus some minor
diagnostic wording improvements.
This patch implements correct hostness based overloading resolution
in isBetterOverloadCandidate.
Based on hostness, if one candidate is emittable whereas the other
candidate is not emittable, the emittable candidate is better.
If both candidates are emittable, or neither is emittable based on hostness, then
other rules should be used to determine which is better. This is because
hostness based overloading resolution is mostly for determining
viability of a function. If two functions are both viable, other factors
should take precedence in preference.
If other rules cannot determine which is better, CUDA preference will be
used again to determine which is better.
However, correct hostness based overloading resolution
requires overloading resolution diagnostics to be deferred,
which is not on by default. The rationale is that deferring
overloading resolution diagnostics may hide overloading reslolutions
issues in header files.
An option -fgpu-exclude-wrong-side-overloads is added, which is off by
default.
When -fgpu-exclude-wrong-side-overloads is off, keep the original behavior,
that is, exclude wrong side overloads only if there are same side overloads.
This may result in incorrect overloading resolution when there are no
same side candates, but is sufficient for most CUDA/HIP applications.
When -fgpu-exclude-wrong-side-overloads is on, enable deferring
overloading resolution diagnostics and enable correct hostness
based overloading resolution, i.e., always exclude wrong side overloads.
Differential Revision: https://reviews.llvm.org/D80450
Android has a handful of API levels relevant to developers described
here: https://developer.android.com/studio/build#module-level.
`__ANDROID_API__` is too vague and confuses a lot of people. Introduce
a new macro name that is explicit about which one it represents. Keep
the old name around because code has been using it for a decade.
Summary:
AIX uses the existing EH infrastructure in clang and llvm.
The major differences would be
1. AIX do not have CFI instructions.
2. AIX uses a new personality routine, named __xlcxx_personality_v1.
It doesn't use the GCC personality rountine, because the
interoperability is not there yet on AIX.
3. AIX do not use eh_frame sections. Instead, it would use a eh_info
section (compat unwind section) to store the information about
personality routine and LSDA data address.
Reviewed By: daltenty, hubert.reinterpretcast
Differential Revision: https://reviews.llvm.org/D91455
This patch diagnoses invalid references of global host variables in device,
global, or host device functions.
Differential Revision: https://reviews.llvm.org/D91281
In C++ when a reference variable is captured by copy, the lambda
is supposed to make a copy of the referenced variable in the captures
and refer to the copy in the lambda. Therefore, it is valid to capture
a reference to a host global variable in a device lambda since the
device lambda will refer to the copy of the host global variable instead
of access the host global variable directly.
However, clang tries to avoid capturing of reference to a host global variable
if it determines the use of the reference variable in the lambda function is
not odr-use. Clang also tries to emit load of the reference to a global variable
as load of the global variable if it determines that the reference variable is
a compile-time constant.
For a device lambda to capture a reference variable to host global variable
and use the captured value, clang needs to be taught that in such cases the use of the reference
variable is odr-use and the reference variable is not compile-time constant.
This patch fixes that.
Differential Revision: https://reviews.llvm.org/D91088
OpenMPIRBuilder::createParallel outlines the body region of the parallel
construct into a new function that accepts any value previously defined outside
the region as a function argument. This function is called back by OpenMP
runtime function __kmpc_fork_call, which expects trailing arguments to be
pointers. If the region uses a value that is not of a pointer type, e.g. a
struct, the produced code would be invalid. In such cases, make createParallel
emit IR that stores the value on stack and pass the pointer to the outlined
function instead. The outlined function then loads the value back and uses as
normal.
Reviewed By: jdoerfert, llitchev
Differential Revision: https://reviews.llvm.org/D92189
Commit 6b1341eb fixed alignment for 128-bit FP types on PowerPC.
However, the quadword alignment adjustment shouldn't be applied to IBM
extended double (ppc_fp128 in IR) values.
Reviewed By: jsji
Differential Revision: https://reviews.llvm.org/D92278
The static_assert in "libcxx/include/memory" was the main offender here,
but then I figured I might as well `git grep -i instantat` and fix all
the instances I found. One was in user-facing HTML documentation;
the rest were in comments or tests.
This allows us to use its value everywhere, rather than just clang. Some
other places, like opt and lld, will use its value soon.
Rename it internally to LLVM_ENABLE_NEW_PASS_MANAGER.
The #define for it is now in llvm-config.h.
The initial land accidentally set the value of
LLVM_ENABLE_NEW_PASS_MANAGER to the string
ENABLE_EXPERIMENTAL_NEW_PASS_MANAGER instead of its value.
Reviewed By: rnk, hans
Differential Revision: https://reviews.llvm.org/D92072
This allows us to use its value everywhere, rather than just clang. Some
other places, like opt and lld, will use its value soon.
The #define for it is now in llvm-config.h.
Reviewed By: rnk, hans
Differential Revision: https://reviews.llvm.org/D92072
Fix hip test failures that were introduced by
previous changes to hip-toolchain-rdc-static-lib.hip
test. The .*lld.* is matching a longer string than
expected.
Differential Revision: https://reviews.llvm.org/D92342
The restriction on pointer-to-pointer kernel arguments has been
relaxed in OpenCL 2.0. Apply the same address space restrictions for
pointer argument types to the inner pointer types.
Differential Revision: https://reviews.llvm.org/D92091
PreferedType were not set when parsing compound literals, hence
designated initializers were not available as code completion suggestions.
This patch sets the preferedtype to parsed type for the following initializer
list.
Fixes https://github.com/clangd/clangd/issues/142.
Differential Revision: https://reviews.llvm.org/D92370
Methods synthesized from declared properties were being added to the
method lists twice. This came from the change to list them in the
class's method list, which missed removing the place in CGObjCGNU that
added them again.
Reviewed By: lanza
Differential Revision: https://reviews.llvm.org/D91874
This is the #2 of 2 changes that make remarks hotness threshold option
available in more tools. The changes also allow the threshold to sync with
hotness threshold from profile summary with special value 'auto'.
This change expands remarks hotness threshold option
-fdiagnostics-hotness-threshold in clang and *-remarks-hotness-threshold in
other tools to utilize hotness threshold from profile summary.
Remarks hotness filtering relies on several driver options. Table below lists
how different options are correlated and affect final remarks outputs:
| profile | hotness | threshold | remarks printed |
|---------|---------|-----------|-----------------|
| No | No | No | All |
| No | No | Yes | None |
| No | Yes | No | All |
| No | Yes | Yes | None |
| Yes | No | No | All |
| Yes | No | Yes | None |
| Yes | Yes | No | All |
| Yes | Yes | Yes | >=threshold |
In the presence of profile summary, it is often more desirable to directly use
the hotness threshold from profile summary. The new argument value 'auto'
indicates threshold will be synced with hotness threshold from profile summary
during compilation. The "auto" threshold relies on the availability of profile
summary. In case of missing such information, no remarks will be generated.
Differential Revision: https://reviews.llvm.org/D85808
This adds multilibs for Fuchsia that is built with the relative vtables ABI,
one with and another without exceptions.
Differential Revision: https://reviews.llvm.org/D85576
By explicitly requesting the system linker with `-fuse-ld=`, the
tests are able to CHECK for the system linker even with
CLANG_DEFAULT_LINKER=lld.
Alternative to D74704.
Differential Revision: https://reviews.llvm.org/D92291
Thanks to D77248, we can bypass the use of stubs altogether and use PLT
relocations if they are available for the target. LLVM and LLD support the
R_AARCH64_PLT32 relocation, so we can also guarantee a static PLT relocation on AArch64.
Not emitting these stubs saves a lot of extra binary size.
Differential Revision: https://reviews.llvm.org/D83812
After D17993, with -fno-delete-null-pointer-checks we add the dereferenceable attribute to the `this` pointer.
We have observed that one internal target which worked before fails even with -fno-delete-null-pointer-checks.
Switching to dereferenceable_or_null fixes the problem.
dereferenceable currently does not always respect NullPointerIsValid and may
imply nonnull and lead to aggressive optimization. The optimization may be
related to `CallBase::isReturnNonNull`, `Argument::hasNonNullAttr`, or
`Value::getPointerDereferenceableBytes`. See D66664 and D66618 for some discussions.
Reviewed By: bkramer, rsmith
Differential Revision: https://reviews.llvm.org/D92297
Enable performing mandatory inlinings upfront, by reusing the same logic
as the full inliner, instead of the AlwaysInliner. This has the
following benefits:
- reduce code duplication - one inliner codebase
- open the opportunity to help the full inliner by performing additional
function passes after the mandatory inlinings, but before th full
inliner. Performing the mandatory inlinings first simplifies the problem
the full inliner needs to solve: less call sites, more contextualization, and,
depending on the additional function optimization passes run between the
2 inliners, higher accuracy of cost models / decision policies.
Note that this patch does not yet enable much in terms of post-always
inline function optimization.
Differential Revision: https://reviews.llvm.org/D91567
This change introduces a new clang switch `-fpseudo-probe-for-profiling` to enable AutoFDO with pseudo instrumentation. Please refer to https://reviews.llvm.org/D86193 for the whole story.
One implication from pseudo-probe instrumentation is that the profile is now sensitive to CFG changes. We perform the pseudo instrumentation very early in the pre-LTO pipeline, before any CFG transformation. This ensures that the CFG instrumented and annotated is stable and optimization-resilient.
The early instrumentation also allows the inliner to duplicate probes for inlined instances. When a probe along with the other instructions of a callee function are inlined into its caller function, the GUID of the callee function goes with the probe. This allows samples collected on inlined probes to be reported for the original callee function.
Reviewed By: wmi
Differential Revision: https://reviews.llvm.org/D86502
The fd parameter of
```
void *mmap(void *addr, size_t length, int prot, int flags, int fd, off_t offset)
```
should be constrained to the range [0, IntMax] as that is of type int.
Constraining to the range [0, Off_tMax] would result in a crash as that is
of a signed type with the value of 0xff..f (-1).
The crash would happen when we try to apply the arg constraints.
At line 583: assert(Min <= Max), as 0 <= -1 is not satisfied
The mmap64 is fixed for the same reason.
Reviewed By: martong, vsavchenko
Differential Revision: https://reviews.llvm.org/D92307
Currently clang is not correctly retrieving from the AST the metadata for
constrained FP builtins. This patch fixes that for the non-target specific
builtins.
Differential Revision: https://reviews.llvm.org/D92122
Similar to Windows Itanium, PS4 is also an Itanium C++ ABI variant
which shares the goal of semantic compatibility with Microsoft C++
code that uses dllimport/export.
This change introduces a new function to determine from the triple
if an environment aims for compatibility with MS C++ code w.r.t to
these attributes and guards the relevant code paths using that
function.
Differential Revision: https://reviews.llvm.org/D90299
Given the following case:
```
auto k() {
return undef();
return 1;
}
```
Prior to the patch, clang emits an `cannot initialize return object of type
'auto' with an rvalue of type 'int'` diagnostic on the second return
(because the return type of the function cannot be deduced from the first contain-errors return).
This patch suppresses this error.
Differential Revision: https://reviews.llvm.org/D92211
class to the declaring class in a class member access.
This check does not appear to be backed by any rule in the standard (the
rule in question was likely removed over the years), and only ever
produces duplicate diagnostics. (It's also not meaningful because there
isn't a unique declaring class after the resolution of core issue 39.)
On platforms where the integrated as isn't called by default this
test fails since the output is not what it expects. Adding this
option generates the expected output on those platforms as well.
This patch implements the definition of __ARM_FEATURE_ATOMICS and fixes the
missing definition of __ARM_FEATURE_CRC32 for Armv8.1-A.
Differential Revision: https://reviews.llvm.org/D91438
This patch enables vector type arguments on AIX. All non-aggregate Altivec vector types are 16bytes in size and are 16byte aligned.
Reviewed By: Xiangling_L
Differential Revision: https://reviews.llvm.org/D92117
This patch adds tests for things that happened to be fixed by previous
patches, but that should continue working if we do decide to treat
sizeless types as incomplete types.
Differential Revision: https://reviews.llvm.org/D79584
I am working on a baremetal riscv toolchain using LLVM runtime and
LLD linker. Baremetal.cpp provides most of the things needed for such
toolchain. So I have modified it to also handle riscv64/32-unknown-elf
targets alongside arm-none-eabi.
Currently, targets like riscv64-unknown-elf are handled by RISCVToolChain
which mostly expects a gcc toolchain to be present. If you dont
want the dependency on gcc-toolchain/libgloss or want to use LLD, then
RISCVToolChain is not a good fit.
So in the toolchain selection code, I have made this dependency of
RISCVToolChain on gcc toolchain explicit. It is created if gcc-toolchain
option is present. Otherwise Baremetal toolchain is created. I will be
happy to hear if there is a better way to choose between these two
toolchains.
Reviewed By: jroelofs
Differential Revision: https://reviews.llvm.org/D91442
same type in multiple base classes.
Not even if the type is introduced by distinct declarations (for
example, two typedef declarations, or a typedef and a class definition).
This is partly in preparation for an upcoming change that can change the
order in which DeclContext lookup results are presented.
In passing, fix some obvious errors where name lookup's notion of a
"static member function" missed static member function templates, and
where its notion of "same set of declarations" was confused by the same
declarations appearing in a different order.
This code got quite twisted because we consider some MSVC builtins to be
target agnostic, and some to be target specific. Target specific
intrinsics have a pattern of doing up-front argument evaluation, while
general intrinsics do not evaluate their arguments up front. As we tried
to share codepaths between the target-specific and target-agnostic
handling, we ended up doing double evaluation.
Instead, have each target handle MSVC intrinsics consistently before up
front argument evaluation. This requires passing less data around and is
more consistent with target independent intrinsic handling.
See D50979 for past examples of this bug. I noticed this while looking
into adding some more intrinsics.
Differential Revision: https://reviews.llvm.org/D92061
The macro is emitted when wargeting SVE code generation with the additional command line option `-msve-vector-bits=<N>`.
The behavior implied by the macro is described in sections "3.7.3.3. Behavior specific to SVE vectors" of the SVE ACLE (Version 00bet6) that can be found at https://developer.arm.com/documentation/100987/latest
Reviewed By: rengolin, rsandifo-arm
Differential Revision: https://reviews.llvm.org/D90956
Previously we only considered using a substitution for a template-name
after already having mangled its prefix, so we'd produce nonsense
manglings like NS3_S4_IiEE where we should simply produce NS4_IiEE.
This is not ABI-compatible with previous Clang versions, and the old
behavior is restored by -fclang-abi-compat=11.0 or earlier.
Added support for the options mabi=vec-extabi and mabi=vec-default which are analogous to qvecnvol and qnovecnvol when using XL on AIX.
The extended Altivec ABI on AIX is enabled using mabi=vec-extabi in clang and vec-extabi in llc.
Reviewed By: Xiangling_L, DiggerLin
Differential Revision: https://reviews.llvm.org/D89684
In GCC, `aarch64-*-linux` and `aarch64-*-freebsd` made the switch in 2018
(https://gcc.gnu.org/pipermail/gcc-patches/2018-March/495549.html).
In Clang, FreeBSD/Fuchsia/NetBSD/MinGW aarch64 default to -fasynchronous-unwind-tables.
This patch defaults Generic_GCC aarch64 (which affects Linux) to use -fasynchronous-unwind-tables.
Reviewed By: nickdesaulniers
Differential Revision: https://reviews.llvm.org/D91760
Previously this option could be used to skip devirtualizations of the
given functions in regular LTO and in the ThinLTO indexing step. This
change allows them to be skipped in the backend as well, which is useful
when debugging WPD in a distributed ThinLTO backend.
Differential Revision: https://reviews.llvm.org/D91812
After commit 2482648a79, a GNU grep option
is just passed unconditionally to `grep` in general. This patch fixes
the test for platforms where `grep` is not GNU grep.
New MachO LLD doesn't implement the old -macos_version_min (etc)
flags, but it understands the modern platform_version flag.
So make the clang driver pass that when using new MachO LLD.
Also, while here, don't pass -lto_library to LLD, since it
links in LTO libraries statically (which it can because it's
versioned alongside clang).
Differential Revision: https://reviews.llvm.org/D92037
The test failed silently if lld wasn't built alongside clang.
But the test uses -###, so the "invalid linker name in -fuse-ld=lld"
diag didn't make clang fail, and something else happened to match
"-demangle", so the test passed.
To fix, pass -B to a directory with two empty +x files (which works
on non-Windows), and look for `"-demangle"` instead of just `-demangle`.
Also force linker_version to 0 and pass a darwin triple.
Differential Revision: https://reviews.llvm.org/D92028
This patch:
- adds an ld64.lld.darwinnew symlink for lld, to go with f2710d4b57,
so that `clang -fuse-ld=lld.darwinnew` can be used to test new
Mach-O lld while it's in bring-up. (The expectation is that we'll
remove this again once new Mach-O lld is the defauld and only Mach-O
lld.)
- lets the clang driver know if the linker is lld (currently
only triggered if `-fuse-ld=lld` or `-fuse-ld=lld.darwinnew` is
passed). Currently only used for the next point, but could be used
to implement other features that need close coordination between
compiler and linker, e.g. having a diag for calling `clang++` instead
of `clang` when link errors are caused by a missing C++ stdlib.
- lets the clang driver pass `-demangle` to Mach-O lld (both old and
new), in addition to ld64
- implements -demangle for new Mach-O lld
- changes demangleItanium() to accept _Z, __Z, ___Z, ____Z prefixes
(and updates one test added in D68014). Mach-O has an extra
underscore for symbols, and the three (or, on Mach-O, four)
underscores are used for block names.
Differential Revision: https://reviews.llvm.org/D91884
Recently HIP toolchain made a change to use clang instead of opt/llc to do compilation
(https://reviews.llvm.org/D81861). The intention is to make HIP toolchain canonical like
other toolchains.
However, this change introduced an unintentional change regarding backend fp fuse
option, which caused regressions in some HIP applications.
Basically before the change, HIP toolchain used clang to generate bitcode, then use
opt/llc to optimize bitcode and generate ISA. As such, the amdgpu backend takes
the default fp fuse mode which is 'Standard'. This mode respect contract flag of
fmul/fadd instructions and do not fuse fmul/fadd instructions without contract flag.
However, after the change, HIP toolchain now use clang to generate IR, do optimization,
and generate ISA as one process. Now amdgpu backend fp fuse option is determined
by -ffp-contract option, which is 'fast' by default. And this -ffp-contract=fast language option
is translated to 'Fast' fp fuse option in backend. Suddenly backend starts to fuse fmul/fadd
instructions without contract flag.
This causes wrong result for some device library functions, e.g. tan(-1e20), which should
return 0.8446, now returns -0.933. What is worse is that since backend with 'Fast' fp fuse
option does not respect contract flag, there is no way to use #pragma clang fp contract
directive to enforce fp contract requirements.
This patch fixes the regression by introducing a new value 'fast-honor-pragmas' for -ffp-contract
and use it for HIP by default. 'fast-honor-pragmas' is equivalent to 'fast' in frontend but
let the backend to use 'Standard' fp fuse option. 'fast-honor-pragmas' is useful since 'Fast'
fp fuse option in backend does not honor contract flag, it is of little use to HIP
applications since all code with #pragma STDC FP_CONTRACT or any IR from a
source compiled with -ffp-contract=on is broken.
Differential Revision: https://reviews.llvm.org/D90174
This fixes the Builtins-sparc-linux testsuite failures on Linux
SPARC which occur because clang cannot find the 32-bit runtime
libraries when -m32 is passed on the command line. The same
workaround is already being used on X86 and PPC.
Also, switch the CHECK-DEBIAN-SPARC tests to use debian_multiarch_tree
as both sparc and sparc64 are using the MultiArch mechanism on modern Debian
systems the same way as x86_64, powerpc64el and others. Thus, switch the
CHECK-DEBIAN-SPARC32 and CHECK-DEBIAN-SPARC64 tests to use the files from
the debian_multiarch_tree directory for the header and linker path tests.
Finally, rename CHECK-DEBIAN-SPARC32 to CHECK-DEBIAN-SPARC to match the naming
scheme of the Debian MultiArch checks for the other Debian architectures.
Reviewed By: MaskRay, phosek
Differential Revision: https://reviews.llvm.org/D90524
Ensure that the DSO Locality of the globals in the IR is derived from
their final visibility when using -fvisibility-from-dllstorageclass.
To accomplish this we reset the DSO locality of globals (before
setting their visibility from their dllstorageclass) at the end of
IRGen in Clang. This removes any effects that visibility options or
annotations may have had on the DSO locality.
The resulting DSO locality of the globals will be pessimistic
w.r.t. to the normal compiler IRGen.
Differential Revision: https://reviews.llvm.org/D91779
These tests invoke opt and llc even though they are in the frontend.
We now do a better job of generating commuted patterns for fma so
these tests now form fmls instead of fmla+fneg.
The test needs an object file, which it currenty gets with
`-target x86_64-apple-darwin10`. Rather than adding `REQUIRES: X86`, create
the object file via yaml2obj. This way, the test runs and passes even if the
host arch isn't x86 and only the host arch is built.
Part of PR46644.
Previously, lax conversions were only allowed between SVE vector-length
agnostic types and vector-length specific types. This meant that code
such as the following:
#include <arm_sve.h>
#define N __ARM_FEATURE_SVE_BITS
#define FIXED_ATTR __attribute__ ((vector_size (N/8)))
typedef float fixed_float32_t FIXED_ATTR;
void foo() {
fixed_float32_t fs32;
svfloat64_t s64;
fs32 = s64;
}
was not allowed.
This patch makes a minor change to areLaxCompatibleSveTypes to allow for
lax conversions to be performed between SVE vector-length agnostic types
and GNU vectors.
Differential Revision: https://reviews.llvm.org/D91696
Sjoerd Meijer