Summary:
This patch generalizes the existing code to support CDE intrinsics
which will share some properties with existing MVE intrinsics
(some of the intrinsics will be polymorphic and accept/return values
of MVE vector types).
Specifically the patch:
* Adds new tablegen backends -gen-arm-cde-builtin-def,
-gen-arm-cde-builtin-codegen, -gen-arm-cde-builtin-sema,
-gen-arm-cde-builtin-aliases, -gen-arm-cde-builtin-header based on
existing MVE backends.
* Renames the '__clang_arm_mve_alias' attribute into
'__clang_arm_builtin_alias' (it will be used with CDE intrinsics as
well as MVE intrinsics)
* Implements semantic checks for the coprocessor argument of the CDE
intrinsics as well as the existing coprocessor intrinsics.
* Adds one CDE intrinsic __arm_cx1 to test the above changes
Reviewers: simon_tatham, MarkMurrayARM, ostannard, dmgreen
Reviewed By: simon_tatham
Subscribers: sdesmalen, mgorny, kristof.beyls, danielkiss, cfe-commits, llvm-commits
Tags: #clang, #llvm
Differential Revision: https://reviews.llvm.org/D75850
Add fixits for messaging self in MRR or using super, as the intent is
clear, and it turns out people do that a lot more than expected.
Allow for objc_direct_members on main interfaces, it's extremely useful
for internal only classes, and proves to be quite annoying for adoption.
Add some better warnings around properties direct/non-direct clashes (it
was done for methods but properties were a miss).
Add some errors when direct properties are marked @dynamic.
Radar-Id: rdar://problem/58355212
Signed-off-by: Pierre Habouzit <phabouzit@apple.com>
Differential Revision: https://reviews.llvm.org/D73755
Add fixits for messaging self in MRR or using super, as the intent is
clear, and it turns out people do that a lot more than expected.
Allow for objc_direct_members on main interfaces, it's extremely useful
for internal only classes, and proves to be quite annoying for adoption.
Add some better warnings around properties direct/non-direct clashes (it
was done for methods but properties were a miss).
Radar-Id: rdar://problem/58355212
Signed-off-by: Pierre Habouzit <phabouzit@apple.com>
This feature is generic. Make it applicable for AArch64 and X86 because
the backend has only implemented NOP insertion for AArch64 and X86.
Reviewed By: nickdesaulniers, aaron.ballman
Differential Revision: https://reviews.llvm.org/D72221
These annotations will be used in an upcomming static analyzer check
that finds handle leaks, use after releases, and double releases.
Differential Revision: https://reviews.llvm.org/D70469
This is equivalent to the existing `import_name` and `import_module`
attributes which control the import names in the final wasm binary
produced by lld.
This maps the existing
This attribute currently requires a string rather than using the
symbol name for a couple of reasons:
1. Avoid confusion with static and dynamic linking which is
based on symbol name. Exporting a function from a wasm module using
this directive is orthogonal to both static and dynamic linking.
2. Avoids name mangling.
Differential Revision: https://reviews.llvm.org/D70520
__attribute__((objc_direct)) is an attribute on methods declaration, and
__attribute__((objc_direct_members)) on implementation, categories or
extensions.
A `direct` property specifier is added (@property(direct) type name)
These attributes / specifiers cause the method to have no associated
Objective-C metadata (for the property or the method itself), and the
calling convention to be a direct C function call.
The symbol for the method has enforced hidden visibility and such direct
calls are hence unreachable cross image. An explicit C function must be
made if so desired to wrap them.
The implicit `self` and `_cmd` arguments are preserved, however to
maintain compatibility with the usual `objc_msgSend` semantics,
3 fundamental precautions are taken:
1) for instance methods, `self` is nil-checked. On arm64 backends this
typically adds a single instruction (cbz x0, <closest-ret>) to the
codegen, for the vast majority of the cases when the return type is a
scalar.
2) for class methods, because the class may not be realized/initialized
yet, a call to `[self self]` is emitted. When the proper deployment
target is used, this is optimized to `objc_opt_self(self)`.
However, long term we might want to emit something better that the
optimizer can reason about. When inlining kicks in, these calls
aren't optimized away as the optimizer has no idea that a single call
is really necessary.
3) the calling convention for the `_cmd` argument is changed: the caller
leaves the second argument to the call undefined, and the selector is
loaded inside the body when it's referenced only.
As far as error reporting goes, the compiler refuses:
- making any overloads direct,
- making an overload of a direct method,
- implementations marked as direct when the declaration in the
interface isn't (the other way around is allowed, as the direct
attribute is inherited from the declaration),
- marking methods required for protocol conformance as direct,
- messaging an unqualified `id` with a direct method,
- forming any @selector() expression with only direct selectors.
As warnings:
- any inconsistency of direct-related calling convention when
@selector() or messaging is used,
- forming any @selector() expression with a possibly direct selector.
Lastly an `objc_direct_members` attribute is added that can decorate
`@implementation` blocks and causes methods only declared there (and in
no `@interface`) to be automatically direct. When decorating an
`@interface` then all methods and properties declared in this block are
marked direct.
Radar-ID: rdar://problem/2684889
Differential Revision: https://reviews.llvm.org/D69991
Reviewed-By: John McCall
Summary:
This is a follow up on https://reviews.llvm.org/D61634
This patch is simpler and only adds the no_builtin attribute.
Reviewers: tejohnson, courbet, theraven, t.p.northover, jdoerfert
Subscribers: mgrang, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D68028
This is a re-submit after it got reverted in https://reviews.llvm.org/rGbd8791610948 since the breakage doesn't seem to come from this patch.
Summary:
This is a follow up on https://reviews.llvm.org/D61634
This patch is simpler and only adds the no_builtin attribute.
Reviewers: tejohnson, courbet, theraven, t.p.northover, jdoerfert
Subscribers: mgrang, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D68028
This allows you to declare a function with a name of your choice (say
`foo`), but have clang treat it as if it were a builtin function (say
`__builtin_foo`), by writing
static __inline__ __attribute__((__clang_arm_mve_alias(__builtin_foo)))
int foo(args);
I'm intending to use this for the ACLE intrinsics for MVE, which have
to be polymorphic on their argument types and also need to be
implemented by builtins. To avoid having to implement the polymorphism
with several layers of nested _Generic and make error reporting
hideous, I want to make all the user-facing intrinsics correspond
directly to clang builtins, so that after clang resolves
__attribute__((overloadable)) polymorphism it's already holding the
right BuiltinID for the intrinsic it selected.
However, this commit itself just introduces the new attribute, and
doesn't use it for anything.
To avoid unanticipated side effects if this attribute is used to make
aliases to other builtins, there's a restriction mechanism: only
(BuiltinID, alias) pairs that are approved by the function
ArmMveAliasValid() will be permitted. At present, that function
doesn't permit anything, because the Tablegen that will generate its
list of valid pairs isn't yet implemented. So the only test of this
facility is one that checks that an unapproved builtin _can't_ be
aliased.
Reviewers: dmgreen, miyuki, ostannard
Subscribers: cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D67159
This is mostly the same as the
[[clang::require_constant_initialization]] attribute, but has a couple
of additional syntactic and semantic restrictions.
In passing, I added a warning for the attribute form being added after
we have already seen the initialization of the variable (but before we
see the definition); that case previously slipped between the cracks and
the attribute was silently ignored.
llvm-svn: 370972
The default behavior of Clang's indirect function call checker will replace
the address of each CFI-checked function in the output file's symbol table
with the address of a jump table entry which will pass CFI checks. We refer
to this as making the jump table `canonical`. This property allows code that
was not compiled with ``-fsanitize=cfi-icall`` to take a CFI-valid address
of a function, but it comes with a couple of caveats that are especially
relevant for users of cross-DSO CFI:
- There is a performance and code size overhead associated with each
exported function, because each such function must have an associated
jump table entry, which must be emitted even in the common case where the
function is never address-taken anywhere in the program, and must be used
even for direct calls between DSOs, in addition to the PLT overhead.
- There is no good way to take a CFI-valid address of a function written in
assembly or a language not supported by Clang. The reason is that the code
generator would need to insert a jump table in order to form a CFI-valid
address for assembly functions, but there is no way in general for the
code generator to determine the language of the function. This may be
possible with LTO in the intra-DSO case, but in the cross-DSO case the only
information available is the function declaration. One possible solution
is to add a C wrapper for each assembly function, but these wrappers can
present a significant maintenance burden for heavy users of assembly in
addition to adding runtime overhead.
For these reasons, we provide the option of making the jump table non-canonical
with the flag ``-fno-sanitize-cfi-canonical-jump-tables``. When the jump
table is made non-canonical, symbol table entries point directly to the
function body. Any instances of a function's address being taken in C will
be replaced with a jump table address.
This scheme does have its own caveats, however. It does end up breaking
function address equality more aggressively than the default behavior,
especially in cross-DSO mode which normally preserves function address
equality entirely.
Furthermore, it is occasionally necessary for code not compiled with
``-fsanitize=cfi-icall`` to take a function address that is valid
for CFI. For example, this is necessary when a function's address
is taken by assembly code and then called by CFI-checking C code. The
``__attribute__((cfi_jump_table_canonical))`` attribute may be used to make
the jump table entry of a specific function canonical so that the external
code will end up taking a address for the function that will pass CFI checks.
Fixes PR41972.
Differential Revision: https://reviews.llvm.org/D65629
llvm-svn: 368495
Summary:
This is the first part of work announced in
"[RFC] Adding lifetime analysis to clang" [0],
i.e. the addition of the [[gsl::Owner(T)]] and
[[gsl::Pointer(T)]] attributes, which
will enable user-defined types to participate in
the lifetime analysis (which will be part of the
next PR).
The type `T` here is called "DerefType" in the paper,
and denotes the type that an Owner owns and a Pointer
points to. E.g. `std::vector<int>` should be annotated
with `[[gsl::Owner(int)]]` and
a `std::vector<int>::iterator` with `[[gsl::Pointer(int)]]`.
[0] http://lists.llvm.org/pipermail/cfe-dev/2018-November/060355.html
Reviewers: gribozavr
Subscribers: xazax.hun, cfe-commits
Tags: #clang
Differential Revision: https://reviews.llvm.org/D63954
llvm-svn: 367040
This patch introduces support of hip_pinned_shadow variable for HIP.
A hip_pinned_shadow variable is a global variable with attribute hip_pinned_shadow.
It has external linkage on device side and has no initializer. It has internal
linkage on host side and has initializer or static constructor. It can be accessed
in both device code and host code.
This allows HIP runtime to implement support of HIP texture reference.
Differential Revision: https://reviews.llvm.org/D62738
llvm-svn: 364381
Seems like a logical extension to me - and of interest because it might
help reduce the debug info size of libc++ by applying this attribute to
type traits that have a disproportionate debug info cost compared to the
benefit (& possibly harm/confusion) they cause users.
llvm-svn: 362856
prettyprint
__declspec(nothrow) should work on function pointers as well as function
references, so this changes it to FunctionLike. Additionally,
FunctionLike needed to be modified to permit function references.
Finally, the TypePrinter didn't properly print the NoThrow exception
specifier, so make sure we get that right as well.
llvm-svn: 362435
Swift requires certain classes to be not just initialized lazily on first
use, but actually allocated lazily using information that is only available
at runtime. This is incompatible with ObjC class initialization, or at least
not efficiently compatible, because there is no meaningful class symbol
that can be put in a class-ref variable at load time. This leaves ObjC
code unable to access such classes, which is undesirable.
objc_class_stub says that class references should be resolved by calling
a new ObjC runtime function with a pointer to a new "class stub" structure.
Non-ObjC compilers (like Swift) can simply emit this structure when ObjC
interop is required for a class that cannot be statically allocated,
then apply this attribute to the `@interface` in the generated ObjC header
for the class.
This attribute can be thought of as a generalization of the existing
`objc_runtime_visible` attribute which permits more efficient class
resolution as well as supporting the additon of categories to the class.
Subclassing these classes from ObjC is currently not allowed.
Patch by Slava Pestov!
llvm-svn: 362054
We want to make objc_nonlazy_class apply to implementations, but ran into this.
There doesn't seem to be any reason that this isn't supported.
Differential revision: https://reviews.llvm.org/D60542
llvm-svn: 358200
This reverts commit r353765. After talking with our c stdlib folks, we decided
to use the existing pass_object_size attribute to implement _FORTIFY_SOURCE
wrappers, like Bionic does (I didn't realize that pass_object_size could be used
for this purpose). Sorry for the flip/flop, and thanks to James Y. Knight for
pointing this out to me.
llvm-svn: 356103
The new __attribute__ ((mig_server_routine)) is going to be used for annotating
Mach Interface Generator (MIG) callback functions as such, so that additional
static analysis could be applied to their implementations. It can also be
applied to regular functions behavior of which is supposed to be identical to
that of a MIG server routine.
Differential Revision: https://reviews.llvm.org/D58365
llvm-svn: 354530
This fixes a regression that was caused by r335084, which reversed
the order that attributes are applied. objc_method_family can change
whether a method is an init method, so the order that these
attributes are applied matters. The commit fixes this by delaying the
init check until after all attributes have been applied.
rdar://47829358
Differential revision: https://reviews.llvm.org/D58152
llvm-svn: 353976
This attribute applies to declarations of C stdlib functions
(sprintf, memcpy...) that have known fortified variants
(__sprintf_chk, __memcpy_chk, ...). When applied, clang will emit
calls to the fortified variant functions instead of calls to the
defaults.
In GCC, this is done by adding gnu_inline-style wrapper functions,
but that doesn't work for us for variadic functions because we don't
support __builtin_va_arg_pack (and have no intention to).
This attribute takes two arguments, the first is 'type' argument
passed through to __builtin_object_size, and the second is a flag
argument that gets passed through to the variadic checking variants.
rdar://47905754
Differential revision: https://reviews.llvm.org/D57918
llvm-svn: 353765
A non-lazy class will be initialized eagerly when the Objective-C runtime is
loaded. This is required for certain system classes which have instances allocated in
non-standard ways, such as the classes for blocks and constant strings.
Adding this attribute is essentially equivalent to providing a trivial
+load method but avoids the (fairly small) load-time overheads associated
with defining and calling such a method.
Differential Revision: https://reviews.llvm.org/D56555
llvm-svn: 353116
This is similar to import_module, but sets the import field name
instead.
By default, the import field name is the same as the C/asm/.o symbol
name. However, there are situations where it's useful to have it be
different. For example, suppose I have a wasm API with a module named
"pwsix" and a field named "read". There's no risk of namespace
collisions with user code at the wasm level because the generic name
"read" is qualified by the module name "pwsix". However in the C/asm/.o
namespaces, the module name is not used, so if I have a global function
named "read", it is intruding on the user's namespace.
With the import_field module, I can declare my function (in libc) to be
"__read", and then set the wasm import module to be "pwsix" and the wasm
import field to be "read". So at the C/asm/.o levels, my symbol is
outside the user namespace.
Differential Revision: https://reviews.llvm.org/D57602
llvm-svn: 352930
With commit r351627, LLVM gained the ability to apply (existing) IPO
optimizations on indirections through callbacks, or transitive calls.
The general idea is that we use an abstraction to hide the middle man
and represent the callback call in the context of the initial caller.
It is described in more detail in the commit message of the LLVM patch
r351627, the llvm::AbstractCallSite class description, and the
language reference section on callback-metadata.
This commit enables clang to emit !callback metadata that is
understood by LLVM. It does so in three different cases:
1) For known broker functions declarations that are directly
generated, e.g., __kmpc_fork_call for the OpenMP pragma parallel.
2) For known broker functions that are identified by their name and
source location through the builtin detection, e.g.,
pthread_create from the POSIX thread API.
3) For user annotated functions that carry the "callback(callee, ...)"
attribute. The attribute has to include the name, or index, of
the callback callee and how the passed arguments can be
identified (as many as the callback callee has). See the callback
attribute documentation for detailed information.
Differential Revision: https://reviews.llvm.org/D55483
llvm-svn: 351629
Summary:
This attribute will allow users to opt specific functions out of
speculative load hardening. This compliments the Clang attribute
named speculative_load_hardening. When this attribute or the attribute
speculative_load_hardening is used in combination with the flags
-mno-speculative-load-hardening or -mspeculative-load-hardening,
the function level attribute will override the default during LLVM IR
generation. For example, in the case, where the flag opposes the
function attribute, the function attribute will take precendence.
The sticky inlining behavior of the speculative_load_hardening attribute
may cause a function with the no_speculative_load_hardening attribute
to be tagged with the speculative_load_hardening tag in
subsequent compiler phases which is desired behavior since the
speculative_load_hardening LLVM attribute is designed to be maximally
conservative.
If both attributes are specified for a function, then an error will be
thrown.
Reviewers: chandlerc, echristo, kristof.beyls, aaron.ballman
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54909
llvm-svn: 351565
When applied to out-parameters, the attributes specify the expected lifetime of the written-into object.
Additionally, introduce OSReturnsRetainedOn(Non)Zero attributes, which
specify that an ownership transfer happens depending on a return code.
Differential Revision: https://reviews.llvm.org/D56292
llvm-svn: 350942
This attribute, called "objc_externally_retained", exposes clang's
notion of pseudo-__strong variables in ARC. Pseudo-strong variables
"borrow" their initializer, meaning that they don't retain/release
it, instead assuming that someone else is keeping their value alive.
If a function is annotated with this attribute, implicitly strong
parameters of that function aren't implicitly retained/released in
the function body, and are implicitly const. This is useful to expose
for performance reasons, most functions don't need the extra safety
of the retain/release, so programmers can opt out as needed.
This attribute can also apply to declarations of local variables,
with similar effect.
Differential revision: https://reviews.llvm.org/D55865
llvm-svn: 350422
Downstream forks that have their own attributes often run into this
test failing when a new attribute is added to clang because the
number of supported attributes no longer match. This is redundant
information for this test, so we can get by without it.
rdar://46288577
llvm-svn: 348218
The addition adds three attributes for communicating ownership,
analogous to existing NS_ and CF_ attributes.
The attributes are meant to be used for communicating ownership of all
objects in XNU (Darwin kernel) and all of the kernel modules.
The ownership model there is very similar, but still different from the
Foundation model, so we think that introducing a new family of
attributes is appropriate.
The addition required a sizeable refactoring of the existing code for
CF_ and NS_ ownership attributes, due to tight coupling and the fact
that differentiating between the types was previously done using a
boolean.
Differential Revision: https://reviews.llvm.org/D54912
llvm-svn: 347947
Summary:
Resubmit this with no changes because I think the build was broken
by a different diff.
-----
The prior diff had to be reverted because there were two tests
that failed. I updated the two tests in this diff
clang/test/Misc/pragma-attribute-supported-attributes-list.test
clang/test/SemaCXX/attr-speculative-load-hardening.cpp
----- Summary from Previous Diff (Still Accurate) -----
LLVM IR already has an attribute for speculative_load_hardening. Before
this commit, when a user passed the -mspeculative-load-hardening flag to
Clang, every function would have this attribute added to it. This Clang
attribute will allow users to opt into SLH on a function by function basis.
This can be applied to functions and Objective C methods.
Reviewers: chandlerc, echristo, kristof.beyls, aaron.ballman
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54915
llvm-svn: 347701
until I figure out why the build is failing or timing out
***************************
Summary:
The prior diff had to be reverted because there were two tests
that failed. I updated the two tests in this diff
clang/test/Misc/pragma-attribute-supported-attributes-list.test
clang/test/SemaCXX/attr-speculative-load-hardening.cpp
LLVM IR already has an attribute for speculative_load_hardening. Before
this commit, when a user passed the -mspeculative-load-hardening flag to
Clang, every function would have this attribute added to it. This Clang
attribute will allow users to opt into SLH on a function by function
basis.
This can be applied to functions and Objective C methods.
Reviewers: chandlerc, echristo, kristof.beyls, aaron.ballman
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54915
This reverts commit a5b3c232d1e3613f23efbc3960f8e23ea70f2a79.
(r347617)
llvm-svn: 347628
Summary:
The prior diff had to be reverted because there were two tests
that failed. I updated the two tests in this diff
clang/test/Misc/pragma-attribute-supported-attributes-list.test
clang/test/SemaCXX/attr-speculative-load-hardening.cpp
----- Summary from Previous Diff (Still Accurate) -----
LLVM IR already has an attribute for speculative_load_hardening. Before
this commit, when a user passed the -mspeculative-load-hardening flag to
Clang, every function would have this attribute added to it. This Clang
attribute will allow users to opt into SLH on a function by function basis.
This can be applied to functions and Objective C methods.
Reviewers: chandlerc, echristo, kristof.beyls, aaron.ballman
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D54915
llvm-svn: 347617
Summary:
This attribute allows excluding a member of a class template from being part
of an explicit template instantiation of that class template. This also makes
sure that code using such a member will not take for granted that an external
instantiation exists in another translation unit. The attribute was discussed
on cfe-dev at [1] and is primarily motivated by the removal of always_inline
in libc++ to control what's part of the ABI (see links in [1]).
[1]: http://lists.llvm.org/pipermail/cfe-dev/2018-August/059024.html
rdar://problem/43428125
Reviewers: rsmith
Subscribers: dexonsmith, cfe-commits
Differential Revision: https://reviews.llvm.org/D51789
llvm-svn: 343790
Summary:
We previously disallowed use of undocumented attributes with #pragma clang
attribute, but the justification for doing so was weak and it prevented many
reasonable use cases.
Reviewers: aaron.ballman, arphaman
Subscribers: cfe-commits, rnk, benlangmuir, dexonsmith, erik.pilkington
Differential Revision: https://reviews.llvm.org/D51507
llvm-svn: 341437
This commit adds the flag -fno-c++-static-destructors and the attributes
[[clang::no_destroy]] and [[clang::always_destroy]]. no_destroy specifies that a
specific static or thread duration variable shouldn't have it's destructor
registered, and is the default in -fno-c++-static-destructors mode.
always_destroy is the opposite, and is the default in -fc++-static-destructors
mode.
A variable whose destructor is disabled (either because of
-fno-c++-static-destructors or [[clang::no_destroy]]) doesn't count as a use of
the destructor, so we don't do any access checking or mark it referenced. We
also don't emit -Wexit-time-destructors for these variables.
rdar://21734598
Differential revision: https://reviews.llvm.org/D50994
llvm-svn: 340306
As documented here: https://software.intel.com/en-us/node/682969 and
https://software.intel.com/en-us/node/523346. cpu_dispatch multiversioning
is an ICC feature that provides for function multiversioning.
This feature is implemented with two attributes: First, cpu_specific,
which specifies the individual function versions. Second, cpu_dispatch,
which specifies the location of the resolver function and the list of
resolvable functions.
This is valuable since it provides a mechanism where the resolver's TU
can be specified in one location, and the individual implementions
each in their own translation units.
The goal of this patch is to be source-compatible with ICC, so this
implementation diverges from the ICC implementation in a few ways:
1- Linux x86/64 only: This implementation uses ifuncs in order to
properly dispatch functions. This is is a valuable performance benefit
over the ICC implementation. A future patch will be provided to enable
this feature on Windows, but it will obviously more closely fit ICC's
implementation.
2- CPU Identification functions: ICC uses a set of custom functions to identify
the feature list of the host processor. This patch uses the cpu_supports
functionality in order to better align with 'target' multiversioning.
1- cpu_dispatch function def/decl: ICC's cpu_dispatch requires that the function
marked cpu_dispatch be an empty definition. This patch supports that as well,
however declarations are also permitted, since the linker will solve the
issue of multiple emissions.
Differential Revision: https://reviews.llvm.org/D47474
llvm-svn: 337552
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
Mikhail Maltsev