Summary:
The range checks performed for the vqrdmulh_lane and vqrdmulh_lane Neon
intrinsics were incorrectly using their return type as the base type for
the range check performed on their 'lane' argument.
This patch updates those intrisics to use the type of the proper reference
argument to perform the range checks.
Reviewers: jmolloy, t.p.northover, rsmith, olista01, dnsampaio
Reviewed By: dnsampaio
Subscribers: dnsampaio, kristof.beyls, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74766
Summary:
Range checks were not properly performed in the lane arguments of Neon
intrinsics implemented based on splat operations. Calls to those
intrinsics where translated to `__builtin__shufflevector` calls directly
by the pre-processor through the arm_neon.h macros, missing the chance
for the proper range checks.
This patch enables the range check by introducing an auxiliary splat
instruction in arm_neon.td, delaying the translation to shufflevector
calls to CGBuiltin.cpp in clang after the checks were performed.
Reviewers: jmolloy, t.p.northover, rsmith, olista01, ostannard
Reviewed By: ostannard
Subscribers: ostannard, dnsampaio, danielkiss, kristof.beyls, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74619
Summary:
As multiple versions of the same Neon intrinsic can be created through
the same TableGen definition with the same argument types, the existing
`call` operator is not always able to properly perform overload
resolutions.
As these different intrinsic versions are differentiated later on by the
NeonEmitter through name mangling, this patch introduces a new
`call_mangled` operator to the TableGen definitions, which allows a call
for an otherwise ambiguous intrinsic by matching its mangled name with
the mangled variation of the caller.
Reviewers: jmolloy, t.p.northover, rsmith, olista01, dnsampaio
Reviewed By: dnsampaio
Subscribers: dnsampaio, kristof.beyls, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74618
Summary:
Some of the `*_laneq` intrinsics defined in arm_neon.td were missing the
setting of the `isLaneQ` attribute. This patch sets the attribute on the
related definitions, as they will be required to properly perform range
checks on their lane arguments.
Reviewers: jmolloy, t.p.northover, rsmith, olista01, dnsampaio
Reviewed By: dnsampaio
Subscribers: dnsampaio, kristof.beyls, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74616
Summary:
Some of the `*_laneq` intrinsics defined in arm_neon.td were missing the
setting of the `isLaneQ` attribute. This patch sets the attribute on the
related definitions, as they will be required to properly perform range
checks on their lane arguments.
Reviewers: jmolloy, t.p.northover, rsmith, olista01, dnsampaio
Reviewed By: dnsampaio
Subscribers: dnsampaio, kristof.beyls, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74616
The SVE ACLE allows using a short-form for the intrinsics, e.g.
the following two declarations generate the same code:
svuint32_t svld1(svbool_t, uint32_t const *);
svuint32_t svld1_u32(svbool_t, uint32_t const *);
using the attribute:
__clang_arm_builtin_alias
so that any call to svld1(svbool_t, uint32_t const *) will
map to __builtin_sve_svld1_u32.
Reviewers: SjoerdMeijer, miyuki, efriedma, simon_tatham, rengolin
Reviewed By: SjoerdMeijer
Tags: #clang
Differential Revision: https://reviews.llvm.org/D75861
The only subexpression that is considered an error now is TypoExpr, but
we plan to add expressions with errors to improve editor tooling on broken
code. We intend to use the same mechanism to guard against spurious
diagnostics on those as well.
See the follow-up revision for an actual usage of the flag.
Original patch from Ilya.
Reviewers: sammccall
Reviewed By: sammccall
Tags: #clang
Differential Revision: https://reviews.llvm.org/D65591
Add a test for UsedDeclVisitor
This test is reduced from mlir/lib/Transforms/AffineDataCopyGeneration.cpp
to make sure there is no assertion due to UsedDeclVisitor.
If the ancestor device modifier is used and the value of the device
clause is evaluated to 1, the ancestor device shall be used for the
execution.
Since the reverse offloading is not supported yet, the target construct
execution is always initiated from the host, not from the device. So, if
the ancestor modifier is specified, just execute target region on the
host.
The existence of the class is more confusing than helpful, I think; the
commonality is mostly just "GEP is legal", which can be queried using
APIs on GetElementPtrInst.
Differential Revision: https://reviews.llvm.org/D75660
HIPToolChain::TranslateArgs call TranslateArgs of host toolchain with
the input args to get a list of derived args called DAL, then
go through the input args by itself and append them to DAL.
This assumes that the host toolchain should not append any unchanged
args to DAL, otherwise there will be duplicates since
HIPToolChain will append it again.
This works for GNU toolchain since it returns an empty list for DAL.
However, MSVC toolchain will append unchanged args to DAL, which
causes duplicate args.
This patch let MSVC toolchain not append unchanged args for HIP
offloading kind, which fixes this issue.
Differential Revision: https://reviews.llvm.org/D76032
Summary:
This is another set of instructions too complicated to be sensibly
expressed in IR by anything short of a target-specific intrinsic.
Given input vectors a,b, the instruction generates intermediate values
2*(a[0]*b[0]+a[1]+b[1]), 2*(a[2]*b[2]+a[3]+b[3]), etc; takes the high
half of each double-width values, and overwrites half the lanes in the
output vector c, which you therefore have to provide the input value
of. Optionally you can swap the elements of b so that the are things
like a[0]*b[1]+a[1]*b[0]; optionally you can round to nearest when
taking the high half; and optionally you can take the difference
rather than sum of the two products. Finally, saturation is applied
when converting back to a single-width vector lane.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76359
Summary:
Copy of https://reviews.llvm.org/D72334, submitting with Ilya's permission.
Handles template declaration of all kinds.
Also builds template declaration nodes for specializations and explicit
instantiations of classes.
Some missing things will be addressed in the follow-up patches:
specializations of functions and variables,
template parameters.
Reviewers: gribozavr2
Reviewed By: gribozavr2
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76346
Summary:
- https://reviews.llvm.org/D68578 revises the `GlobalDecl` constructors
to ensure all GPU kernels have `ReferenceKenelKind` initialized
properly with an explicit constructor and static one. But, there are
lots of places using the implicit constructor triggering the assertion
on non-GPU kernels. That's found in compilation of many tests and
workloads.
- Fixing all of them may change more code and, more importantly, all of
them assumes the default kernel reference kind. This patch changes
that constructor to tell `CUDAGlobalAttr` and construct `GlobalDecl`
properly.
Reviewers: yaxunl
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76344
Reworked the patch to avoid sharing a header (SVETypeFlags.h) between
include/clang/Basic and utils/TableGen/SveEmitter.cpp. Now the patch
generates the enum/flags which is included in TargetBuiltins.h.
Also renamed one of the SveEmitter options to be in line with MVE.
Summary:
This is a first patch in a series for the SveEmitter to generate the arm_sve.h
header file and builtins.
I've tried my best to strip down this patch as best as I could, but there
are still a few changes that are not necessarily exercised by the load intrinsics
in this patch, mostly around the SVEType class which has some common logic to
represent types from a type and prototype string. I thought it didn't make
much sense to remove that from this patch and split it up.
Summary:
These are complicated integer multiply+add instructions with extra
saturation, taking the high half of a double-width product, and
optional rounding. There's no sensible way to represent that in
standard IR, so I've converted the clang builtins directly to
target-specific intrinsics.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: miyuki
Subscribers: kristof.beyls, hiraditya, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76123
Summary:
These instructions compute multiply+add in integers, with one of the
operands being a splat of a scalar. (VMLA and VMLAS differ in whether
the splat operand is a multiplier or the addend.)
I've represented these in IR using existing standard IR operations for
the unpredicated forms. The predicated forms are done with target-
specific intrinsics, as usual.
When operating on n-bit vector lanes, only the bottom n bits of the
i32 scalar operand are used. So we have to tell that to isel lowering,
to allow it to remove a pointless sign- or zero-extension instruction
on that input register. That's done in `PerformIntrinsicCombine`, but
first I had to enable `PerformIntrinsicCombine` for MVE targets
(previously all the intrinsics it handled were for NEON), and make it
a method of `ARMTargetLowering` so that it can get at
`SimplifyDemandedBits`.
Reviewers: dmgreen, MarkMurrayARM, miyuki, ostannard
Reviewed By: dmgreen
Subscribers: kristof.beyls, hiraditya, danielkiss, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D76122
- After https://reviews.llvm.org/D68578, the implicit conversion from
`FunctionDecl` to `GlobalDecl` needs replacing with `getGlobalDecl`;
otherwise, assertion is triggered.
Summary:
[Clang] Attribute to allow defining undef global variables
Initializing global variables is very cheap on hosted implementations. The
C semantics of zero initializing globals work very well there. It is not
necessarily cheap on freestanding implementations. Where there is no loader
available, code must be emitted near the start point to write the appropriate
values into memory.
At present, external variables can be declared in C++ and definitions provided
in assembly (or IR) to achive this effect. This patch provides an attribute in
order to remove this reason for writing assembly for performance sensitive
freestanding implementations.
A close analogue in tree is LDS memory for amdgcn, where the kernel is
responsible for initializing the memory after it starts executing on the gpu.
Uninitalized variables in LDS are observably cheaper than zero initialized.
Patch is loosely based on the cuda __shared__ and opencl __local variable
implementation which also produces undef global variables.
Reviewers: kcc, rjmccall, rsmith, glider, vitalybuka, pcc, eugenis, vlad.tsyrklevich, jdoerfert, gregrodgers, jfb, aaron.ballman
Reviewed By: rjmccall, aaron.ballman
Subscribers: Anastasia, aaron.ballman, davidb, Quuxplusone, dexonsmith, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74361
Discovered in a downstream, it is often useful to have slightly
different semantics for an attribute based on its namespace, however our
spelling infrastructure doesn't consider namespace when deciding to
elide the enum list. The result is that the solution for a case where
an attribute has slightly different semantics based on a namespace
requires checking against the integer value, which is fragile.
This patch makes us always emit the spelling enum if there is more than
1 and we're generating the header.
Differential Revision: https://reviews.llvm.org/D76289
This excludes some builtins that take argument types not yet handled
by the `-fdeclare-opencl-builtins` machinery, and also excludes
builtins that are already defined in `Builtins.def`.
MCTargetOptionsCommandFlags.inc and CommandFlags.inc are headers which contain
cl::opt with static storage.
These headers are meant to be incuded by tools to make it easier to parametrize
codegen/mc.
However, these headers are also included in at least two libraries: lldCommon
and handle-llvm. As a result, when creating DYLIB, clang-cpp holds a reference
to the options, and lldCommon holds another reference. Linking the two in a
single executable, as zig does[0], results in a double registration.
This patch explores an other approach: the .inc files are moved to regular
files, and the registration happens on-demand through static declaration of
options in the constructor of a static object.
[0] https://bugzilla.redhat.com/show_bug.cgi?id=1756977#c5
Differential Revision: https://reviews.llvm.org/D75579
Sizeless types can't be used with "new", so it doesn't make sense
to use them with "delete" either. The SVE ACLE therefore doesn't
allow that.
This is slightly stronger than for normal incomplete types, since:
struct S;
void f(S *s) { delete s; }
is (by necessity) just a default-on warning rather than an error.
Differential Revision: https://reviews.llvm.org/D76219
Summary:
Currently, ValueRange is very hard to extend with new kind of constraints.
For instance, it forcibly encapsulates relations between arguments and the
return value (ComparesToArgument) besides handling the regular value
ranges (OutOfRange, WithinRange).
ValueRange in this form is not suitable to add new constraints on
arguments like "not-null".
This refactor introduces a new base class ValueConstraint with an
abstract apply function. Descendants must override this. There are 2
descendants: RangeConstraint and ComparisonConstraint. In the following
patches I am planning to add the NotNullConstraint, and additional
virtual functions like `negate()` and `warning()`.
Reviewers: NoQ, Szelethus, balazske, gamesh411, baloghadamsoftware, steakhal
Subscribers: whisperity, xazax.hun, szepet, rnkovacs, a.sidorin, mikhail.ramalho, donat.nagy, dkrupp, Charusso, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D74973
new-expressions for a type T require sizeof(T) to be computable,
so the SVE ACLE does not allow them for sizeless types. At the moment:
auto f() { return new __SVInt8_t; }
creates a call to operator new with a zero size:
%call = call noalias nonnull i8* @_Znwm(i64 0)
This patch reports an appropriate error instead.
Differential Revision: https://reviews.llvm.org/D76218
This flag is used by avr-gcc (starting with v10) to set the width of the
double type. The double type is by default interpreted as a 32-bit
floating point number in avr-gcc instead of a 64-bit floating point
number as is common on other architectures. Starting with GCC 10, a new
option has been added to control this behavior:
https://gcc.gnu.org/wiki/avr-gcc#Deviations_from_the_Standard
This commit keeps the default double at 32 bits but adds support for the
-mdouble flag (-mdouble=32 and -mdouble=64) to control this behavior.
Differential Revision: https://reviews.llvm.org/D76181
In the current SVE ACLE spec, the usual rules for throwing and
catching incomplete types also apply to sizeless types. However,
throwing pointers to sizeless types should not pose any real difficulty,
so as an extension, the clang implementation allows that.
This patch enforces these rules for catch statements.
Differential Revision: https://reviews.llvm.org/D76090
Lucas Prates