Sanjay Patel [Sun, 19 Aug 2018 16:50:30 +0000 (16:50 +0000)]
[CodeGen] add rotate builtins that map to LLVM funnel shift
This is a partial retry of rL340137 (reverted at rL340138 because of gcc host compiler crashing)
with 1 change:
Remove the changes to make microsoft builtins also use the LLVM intrinsics.
This exposes the LLVM funnel shift intrinsics as more familiar bit rotation functions in clang
(when both halves of a funnel shift are the same value, it's a rotate).
We're free to name these as we want because we're not copying gcc, but if there's some other
existing art (eg, the microsoft ops) that we want to replicate, we can change the names.
The funnel shift intrinsics were added here:
https://reviews.llvm.org/D49242
With improved codegen in:
https://reviews.llvm.org/rL337966
https://reviews.llvm.org/rL339359
And basic IR optimization added in:
https://reviews.llvm.org/rL338218
https://reviews.llvm.org/rL340022
...so these are expected to produce asm output that's equal or better to the multi-instruction
alternatives using primitive C/IR ops.
In the motivating loop example from PR37387:
https://bugs.llvm.org/show_bug.cgi?id=37387#c7
...we get the expected 'rolq' x86 instructions if we substitute the rotate builtin into the source.
Ivan A. Kosarev [Sun, 19 Aug 2018 16:30:57 +0000 (16:30 +0000)]
[NEON] Define fp16 vld and vst intrinsics conditionally
This patch fixes definitions of vld and vst NEON intrinsics so
that we only define them if half-precision arithmetic is
supported on the target platform, as prescribed in ACLE 2.0.
Sanjay Patel [Sun, 19 Aug 2018 14:44:47 +0000 (14:44 +0000)]
[CodeGen] add/fix rotate builtins that map to LLVM funnel shift (retry)
This is a retry of rL340135 (reverted at rL340136 because of gcc host compiler crashing)
with 2 changes:
1. Move the code into a helper to reduce code duplication (and hopefully work-around the crash).
2. The original commit had a formatting bug in the docs (missing an underscore).
Original commit message:
This exposes the LLVM funnel shift intrinsics as more familiar bit rotation functions in clang
(when both halves of a funnel shift are the same value, it's a rotate).
We're free to name these as we want because we're not copying gcc, but if there's some other
existing art (eg, the microsoft ops that are modified in this patch) that we want to replicate,
we can change the names.
The funnel shift intrinsics were added here:
https://reviews.llvm.org/D49242
With improved codegen in:
https://reviews.llvm.org/rL337966
https://reviews.llvm.org/rL339359
And basic IR optimization added in:
https://reviews.llvm.org/rL338218
https://reviews.llvm.org/rL340022
...so these are expected to produce asm output that's equal or better to the multi-instruction
alternatives using primitive C/IR ops.
In the motivating loop example from PR37387:
https://bugs.llvm.org/show_bug.cgi?id=37387#c7
...we get the expected 'rolq' x86 instructions if we substitute the rotate builtin into the source.
Sanjay Patel [Sun, 19 Aug 2018 13:48:06 +0000 (13:48 +0000)]
revert r340135: [CodeGen] add rotate builtins
At least a couple of bots (PPC only?) are showing the compiler dying while trying to compile:
http://lab.llvm.org:8011/builders/clang-ppc64be-linux-multistage/builds/11065/steps/build%20stage%201/logs/stdio
http://lab.llvm.org:8011/builders/clang-ppc64be-linux-lnt/builds/18267/steps/build%20stage%201/logs/stdio
Sanjay Patel [Sun, 19 Aug 2018 13:12:40 +0000 (13:12 +0000)]
[CodeGen] add rotate builtins
This exposes the LLVM funnel shift intrinsics as more familiar bit rotation functions in clang
(when both halves of a funnel shift are the same value, it's a rotate).
We're free to name these as we want because we're not copying gcc, but if there's some other
existing art (eg, the microsoft ops that are modified in this patch) that we want to replicate,
we can change the names.
The funnel shift intrinsics were added here:
D49242
With improved codegen in:
rL337966
rL339359
And basic IR optimization added in:
rL338218
rL340022
...so these are expected to produce asm output that's equal or better to the multi-instruction
alternatives using primitive C/IR ops.
In the motivating loop example from PR37387:
https://bugs.llvm.org/show_bug.cgi?id=37387#c7
...we get the expected 'rolq' x86 instructions if we substitute the rotate builtin into the source.
George Karpenkov [Sat, 18 Aug 2018 01:45:50 +0000 (01:45 +0000)]
[analyzer] [NFC] Split up RetainSummaryManager from RetainCountChecker
ARCMigrator is using code from RetainCountChecker, which is a layering
violation (and it also does it badly, by using a different header, and
then relying on implementation being present in a header file).
This change splits up RetainSummaryManager into a separate library in
lib/Analysis, which can be used independently of a checker.
Alex Lorenz [Fri, 17 Aug 2018 22:18:08 +0000 (22:18 +0000)]
[ObjC] Error out when using forward-declared protocol in a @protocol
expression
Clang emits invalid protocol metadata when a @protocol expression is used with a
forward-declared protocol. The protocol metadata is missing protocol conformance
list of the protocol since we don't have access to the definition of it in the
compiled translation unit. The linker then might end up picking the invalid
metadata when linking which will lead to incorrect runtime protocol conformance
checks.
This commit makes sure that Clang fails to compile code that uses a @protocol
expression with a forward-declared protocol. This ensures that Clang does not
emit invalid protocol metadata. I added an extra assert in CodeGen to ensure
that this kind of issue won't happen in other places.
Reid Kleckner [Fri, 17 Aug 2018 22:11:31 +0000 (22:11 +0000)]
Don't warn on returning the address of a label from a statement expression
Summary:
There isn't anything inherently wrong with returning a label from a
statement expression. In practice, the Linux kernel uses this pattern to
materialize PCs.
George Karpenkov [Fri, 17 Aug 2018 21:40:38 +0000 (21:40 +0000)]
[analyzer] Drop support for GC mode in RetainCountChecker
A lot of code in RetainCountChecker deals with GC mode.
Given that GC mode is deprecated, Apple does not ship runtime for it,
and modern compiler toolchain does not support it, it makes sense to
remove the code dealing with it in order to aid understanding of
RetainCountChecker.
Reid Kleckner [Fri, 17 Aug 2018 20:59:27 +0000 (20:59 +0000)]
[MS] Mangle a hash of the main file path into anonymous namespaces
Summary:
This is needed to avoid conflicts in mangled names for codeview types in
anonymous namespaces. In CodeView, types refer to each other typically
through forward declarations, which contain mangled names. These names
have to be unique, otherwise the debugger will look up the mangled name
and find the wrong definition.
Furthermore, ThinLTO will deduplicate the types, and debug info
verification can fail when the types have the wrong sizes. This is
PR38608.
Richard Smith [Fri, 17 Aug 2018 19:43:40 +0000 (19:43 +0000)]
Improve diagnostic for missing comma in template parameter list.
Given 'typename T typename U', we would correctly diagnose the missing
comma, but incorrectly disambiguate the first parameter as being a
non-type parameter and complain that the 'T' is not a qualified-id.
Yaxun Liu [Fri, 17 Aug 2018 17:47:31 +0000 (17:47 +0000)]
[HIP] Make __hip_gpubin_handle hidden to avoid being merged across different shared libraries
Different shared libraries contain different fat binary, which is stored in a global variable
__hip_gpubin_handle. Since different compilation units share the same fat binary, this
variable has linkonce linkage. However, it should not be merged across different shared
libraries.
This patch set the visibility of the global variable to be hidden, which will make it invisible
in the shared library, therefore preventing it from being merged.
Nico Weber [Fri, 17 Aug 2018 17:19:06 +0000 (17:19 +0000)]
Make __shiftleft128 / __shiftright128 real compiler built-ins.
r337619 added __shiftleft128 / __shiftright128 as functions in intrin.h.
Microsoft's STL plans on using these functions, and they're using intrin0.h
which just has declarations of built-ins to not pull in the huge intrin.h
header in the standard library headers. That requires that these functions are
real built-ins.
Akira Hatanaka [Fri, 17 Aug 2018 15:46:07 +0000 (15:46 +0000)]
[CodeGen] Merge identical block descriptor global variables.
Currently, clang generates a new block descriptor global variable for
each new block literal. This commit merges block descriptors that are
identical inside and across translation units using the same approach
taken in r339438.
To enable merging identical block descriptors, the size and signature of
the block and information about the captures are encoded into the name
of the block descriptor variable. Also, the block descriptor variable is
marked as linkonce_odr and unnamed_addr.
Teresa Johnson [Fri, 17 Aug 2018 13:19:36 +0000 (13:19 +0000)]
[ThinLTO] Correct documentation on default number of threads
Summary:
The number of threads used for ThinLTO backend parallelism was
dropped to the number of cores in r284618 to avoid oversubscribing
physical cores due to hyperthreading. This updates the documentation
to reflect that change.
Luke Cheeseman [Fri, 17 Aug 2018 12:55:05 +0000 (12:55 +0000)]
[AArch64] - return address signing
- Add a command line options -msign-return-address to enable return address
signing
- Armv8.3a added instructions to sign the return address to help mitigate
against ROP attacks
- This patch adds command line options to generate function attributes that
signal to the back whether return address signing instructions should be
added
To suppress TSan diagnostics about racy counter updates, make the
counter updates atomic when TSan is enabled. If there's general interest
in this mode it can be surfaced as a clang/swift driver option.
Emmett Neyman [Thu, 16 Aug 2018 20:13:40 +0000 (20:13 +0000)]
Update README and Dockerfile to include llvm-proto-fuzzer
Summary: Added commands to Dockerfile to build llvm-proto-fuzzer and the other related tools. Also added a section to the bottom of the README describing what llvm-proto-fuzzer does and how to run it.
Bruno Ricci [Thu, 16 Aug 2018 10:48:16 +0000 (10:48 +0000)]
[AST] Store the OwnedTagDecl as a trailing object in ElaboratedType.
The TagDecl *OwnedTagDecl in ElaboratedType is quite commonly
null (at least when parsing all of Boost, it is non-null for only about 600
of the 66k ElaboratedType). Therefore we can save a pointer in the
common case by storing it as a trailing object, and storing a bit in the
bit-fields of Type indicating when the pointer is null.
Emmett Neyman [Wed, 15 Aug 2018 23:05:48 +0000 (23:05 +0000)]
Implementation of nested loops in cxx_loop_proto
Summary: Extended `cxx_loop_proto` to have neste for loops. Modified `loop_proto_to_llvm` and `loop_proto_to_cxx` to handle the new protos. All protos have a set of statements designated as "inner loop" statements and a set of statements designated as "outer loop" statements.
Chris Bieneman [Wed, 15 Aug 2018 22:50:06 +0000 (22:50 +0000)]
[Darwin Driver] Fix Simulator builtins and test cases
In r339807, I broke linking the builtins libraries for simulator targets, which itself was bad, but turns out it was all completely untested and marked with FIXME in the test suite.
This fixes all the test cases so they actually work, and fixes the bug I introduced in r339807.
Chris Bieneman [Wed, 15 Aug 2018 20:09:38 +0000 (20:09 +0000)]
Refactor Darwin driver to refer to runtimes by component
Summary:
In r335809, Petr Hosek lays out support for what he calls the multiarch
runtimes layout. This new way of laying out the directories for runtime
libraries is workable for all platforms. Petr did some of the common
infrastructure and made it work for Linux and Fuscia. This patch is a
cleanup to the Darwin and MachO drivers to serve as a step toward
supporting it in Darwin.
This patch does primarily two things:
(1) Changes the APIs for how the Darwin driver refers to compiler-rt
libraries to use the component names, similar to how Linux and Fuscia do
(2) Removes some legacy functionality for supporting macOS versions
before 10.6. This functionality is effectively dead code because in
r339277, the support was removed from compiler-rt for generating the 10.4
runtime support library, and Xcode 10 (currently in beta) removes
libgcc_s.10.4 and libgcc_s.10.5 from the macOS SDK.
With this patch landed a subsequent patch can modify
MachO::AddLinkRuntimeLib to support the multiarch runtimes layout.
Worth noting: None of the removed functionality was actually covered in
the test suite. So no test case updates are required.
Alexey Bataev [Wed, 15 Aug 2018 19:45:12 +0000 (19:45 +0000)]
[OPENMP] FIx processing of declare target variables.
The compiler may produce unexpected error messages/crashes when declare
target variables were used. Patch fixes problems with the declarations
marked as declare target to or link.
Momchil Velikov [Wed, 15 Aug 2018 12:22:08 +0000 (12:22 +0000)]
Use .cpp extension for certain tests instead of .cc
The tests `CodeGen/aapcs[64]-align.cc` are not run since files with a `.cc`
suffix aren't recognisze as tests. This patch renames the above two files to
`.cpp`.
David Green [Wed, 15 Aug 2018 10:39:43 +0000 (10:39 +0000)]
Fix ASTMatchersTraversalTest testcase compile on older compilers
Some versions of gcc, especially when invoked through ccache (-E), can have
trouble with raw string literals inside macros. This moves the string out of
the macro.
Artem Dergachev [Wed, 15 Aug 2018 00:33:55 +0000 (00:33 +0000)]
[analyzer] Add support for constructors of arguments.
Once CFG-side support for argument construction contexts landed in r338436,
the analyzer could make use of them to evaluate argument constructors properly.
When evaluated as calls, constructors of arguments now use the variable region
of the parameter as their target. The corresponding stack frame does not yet
exist when the parameter is constructed, and this stack frame is created
eagerly.
Construction of functions whose body is unavailable and of virtual functions
is not yet supported. Part of the reason is the analyzer doesn't consistently
use canonical declarations o identify the function in these cases, and every
re-declaration or potential override comes with its own set of parameter
declarations. Also it is less important because if the function is not
inlined, there's usually no benefit in inlining the argument constructor.
Artem Dergachev [Tue, 14 Aug 2018 21:10:46 +0000 (21:10 +0000)]
[CFG] [analyzer] Find argument constructors in CXXTemporaryObjectExprs.
CXXTemporaryObjectExpr is a sub-class of CXXConstructExpr. If it has arguments
that are structures passed by value, their respective constructors need to be
handled by providing a ConstructionContext, like for regular function calls and
for regular constructors.
Craig Topper [Tue, 14 Aug 2018 20:21:05 +0000 (20:21 +0000)]
[InlineAsm] Update the min-legal-vector-width function attribute based on inputs and outputs to inline assembly
Summary:
Another piece of my ongoing to work for prefer-vector-width.
min-legal-vector-width will eventually be used by the X86 backend to know whether it needs to make 512 bits type legal when prefer-vector-width=256. If the user used inline assembly that passed in/out a 512-bit register, we need to make sure 512 bits are considered legal. Otherwise we'll get an assert failure when we try to wire up the inline assembly to the rest of the code.
This patch just checks the LLVM IR types to see if they are vectors and then updates the attribute based on their total width. I'm not sure if this is the best way to do this or if there's any subtlety I might have missed. So if anyone has other opinions on how to do this I'm open to suggestions.
David Chisnall [Tue, 14 Aug 2018 10:04:36 +0000 (10:04 +0000)]
Add a stub mangling for ObjC selectors in the Microsoft ABI.
This mangling is used only for outlined SEH finally blocks, which have
internal linkage.
This fixes the failure of CodeGenObjC/2007-04-03-ObjcEH.m on builds with
expensive checks enabled, on Windows. This test should probably be
specifying a triple: it currently picks up whatever the host environment
is using. Unfortunately, I have no idea what it is trying to test,
because it contains no comments and predates Clang having working
Objective-C IR generation.
[XRay][clang] Add more test cases of -fxray-modes= (NFC)
This confirms expectations for multiple values provided through the
driver when selecting specific modes and the order of appearance of
individual values for the `-fxray-modes=` flag.
This change just adds more test cases to an existing test file.
David Carlier [Tue, 14 Aug 2018 05:12:53 +0000 (05:12 +0000)]
[CStringSyntaxChecker] Check strlcat sizeof check
- Assuming strlcat is used with strlcpy we check as we can if the last argument does not equal os not larger than the buffer.
- Advising the proper usual pattern.
Reid Kleckner [Tue, 14 Aug 2018 01:55:37 +0000 (01:55 +0000)]
Revert r339623 "Model type attributes as regular Attrs."
This breaks compiling atlwin.h in Chromium. I'm sure the code is invalid
in some way, but we put a lot of work into accepting it, and I'm sure
rejecting it was not an intended consequence of this refactoring. :)
George Karpenkov [Mon, 13 Aug 2018 23:12:43 +0000 (23:12 +0000)]
[analyzer] [NFC] Introduce separate targets for testing the analyzer: check-clang-analyzer and check-clang-analyzer-z3
Current testing setup for analyzer tests with Z3 is rather inconvenient:
There's no way to run the analyzer tests separately (I use
LIT_FILTER=Analysis ninja check-clang, but a direct target is nicer).
When Clang is built with Z3 support, there's no way to *not* run tests
with Z3 solver, and this is often desired, as tests with Z3 solver take
a very long time.
Richard Smith [Mon, 13 Aug 2018 22:07:09 +0000 (22:07 +0000)]
Model type attributes as regular Attrs.
Specifically, AttributedType now tracks a regular attr::Kind rather than
having its own parallel Kind enumeration, and AttributedTypeLoc now
holds an Attr* instead of holding an ad-hoc collection of Attr fields.
projects / clang / log