This change causes issues with distributed build systems, which may only
have compiler binaries without any runtime libraries. See discussion
about this on https://reviews.llvm.org/D15225.
llvm-svn: 338444
Summary:
See binutils-gdb/bfd/elf.c, GNU objcopy also strips .stab* (STABS)
.line* (DWARF 1) .gnu.linkonce.wi.* (linkonce section for .debug_info) but
I'm not sure we need to be compatible with it.
Reviewers: dblaikie, alexshap, jakehehrlich, jhenderson
Reviewed By: alexshap, jakehehrlich
Subscribers: aprantl, JDevlieghere, jakehehrlich, llvm-commits
Differential Revision: https://reviews.llvm.org/D50100
llvm-svn: 338443
Because of incomplete support for CXXDefaultArgExpr, we cannot yet commit to
asserting that the same destructor won't be elided twice.
Suppress the assertion failure for now. Proper support is still an open problem.
Differential Revision: https://reviews.llvm.org/D49213
llvm-svn: 338441
Some test setups run tests in a read-only path, which means that opening the
default output path (a.out) for write will fail. This change adds appropriate -o
flags so the tests will not fail spuriously.
llvm-svn: 338440
This is a refactoring patch; no functional change intended.
The common part of ConstructionContextLayer and ConstructedObjectKey is
factored out into a new structure, ConstructionContextItem.
Various sub-kinds of ConstructionContextItem are enumerated in order to
provide richer information about construction contexts.
Differential Revision: https://reviews.llvm.org/D49210.
llvm-svn: 338439
parameters can have default arguments.
At least for function templates and class template partial
specializations, it's possible for a template parameter with a default
argument to be followed by a non-pack template parameter with no default
argument, and this case was not properly handled here.
Testcase by Steve O'Brien!
llvm-svn: 338438
In r330377 and r338425 we have already identified what constitutes function
argument constructors and added stubs in order to prevent confusing them
with other temporary object constructors.
Now we implement a ConstructionContext sub-class to carry all the necessary
information about the construction site, namely call expression and argument
index.
On the analyzer side, the patch interacts with the recently implemented
pre-C++17 copy elision support in an interesting manner. If on the CFG side we
didn't find a construction context for the elidable constructor, we build
the CFG as if the elidable constructor is not elided, and the non-elided
constructor within it is a simple temporary. But the same problem may occur
in the analyzer: if the elidable constructor has a construction context but
the analyzer doesn't implement such context yet, the analyzer should also
try to skip copy elision and still inline the non-elided temporary constructor.
This was implemented by adding a "roll back" mechanism: when elision fails,
roll back the changes and proceed as if it's a simple temporary. The approach
is wonky, but i'm fine with that as long as it's merely a defensive mechanism
that should eventually go away once all construction contexts become supported.
Differential Revision: https://reviews.llvm.org/D48681.
llvm-svn: 338436
If any of our inputs are bitcode files, the LTO code generator may create
references to certain library functions that might not be explicit in the
bitcode file's symbol table. If any of those library functions are defined
in a bitcode file in an archive member, we need to arrange to use LTO to
compile those archive members by adding them to the link beforehand.
Differential Revision: https://reviews.llvm.org/D50017
llvm-svn: 338434
Summary:
Size_t, intptr_t, uintptr_t and ptrdiff_t cannot be used as kernel
arguments, according to OpenCL Specification s6.9k:
The size in bytes of these types are implementation-defined and in
addition can also be different for the OpenCL device and the host
processor making it difficult to allocate buffer objects to be passed
as arguments to a kernel declared as pointer to these types.
Patch by: Andrew Savonichev
Reviewers: Anastasia, yaxunl
Subscribers: yaxunl, Anastasia, cfe-commits
Differential Revision: https://reviews.llvm.org/D49725
llvm-svn: 338432
Correct the address space for the inserted argument
stack slot.
AMDGPU seems to not do anything with this information,
so I don't think this was breaking anything.
llvm-svn: 338428
Summary:
OpenCL specification forbids use of several types as kernel arguments.
This patch improves existing diagnostic to look through arrays.
Patch by: Andrew Savonichev
Reviewers: Anastasia, yaxunl
Subscribers: yaxunl, Anastasia, cfe-commits
Differential Revision: https://reviews.llvm.org/D49723
llvm-svn: 338427
Like any normal funciton, Objective-C message can return a C++ object
in Objective-C++. Such object would require a construction context.
This patch, therefore, is an extension of r327343 onto Objective-C++.
Differential Revision: https://reviews.llvm.org/D48608
llvm-svn: 338426
CFG now correctly identifies construction context for temporaries constructed
for the purpose of passing into a function as an argument.
Such context is still not fully implemented because the information it provides
is not rich enough: it doens't contain information about argument index.
It will be addresssed later.
This patch is an extension of r330377 to C++ construct-expressions and
Objective-C message expressions which aren't call-expressions but require
similar handling. C++ new-expressions with placement arguments still remain to
be handled.
Differential Revision: https://reviews.llvm.org/D49826
llvm-svn: 338425
This fix is similar to r337769 and addresses a regression caused by r337167.
When an operation between a nonloc::LocAsInteger and a non-pointer symbol
is performed, the LocAsInteger-specific part of information is lost.
When the non-pointer symbol is collapsing into a constant, we cannot easily
re-evaluate the result, because we need to recover the missing
LocAsInteger-specific information (eg., integer type, or the very fact that
this pointer was at some point converted to an integer).
Add one more defensive check to prevent crashes on trying to simplify a
SymSymExpr with different Loc-ness of operands.
Differential Revision:
llvm-svn: 338420
When lowering calling conventions, prefer to decompose vectors
into the constitute register types. This avoids artifical constraints
to satisfy a wide super-register.
This improves code quality because now optimizations don't need to
deal with the super-register constraint. For example the immediate
folding code doesn't deal with 4 component reg_sequences, so by
breaking the register down earlier the existing immediate folding
code is able to work.
This also avoids the need for the shader input processing code
to manually split vector types.
llvm-svn: 338416
offload targets.
Changed the linkage of omp_offloading.img_start.<triple> and omp_offloading.img_end.<triple> symbols from external to external weak to allow dropping of some targets during linking.
llvm-svn: 338413
Don't declare them as X86SchedWritePair when the folded class will never be used.
Note: MOVBE (load/store endian conversion) instructions tend to have a very different behaviour to BSWAP.
llvm-svn: 338412
As was done for vector rotations, we can efficiently use ISD::MULHU for vXi8/vXi16 ISD::SRL lowering.
Shift-by-zero cases are still problematic (mainly on v32i8 due to extra AND/ANDN/OR or VPBLENDVB blend masks but v8i16/v16i16 aren't great either if PBLENDW fails) so I've limited this first patch to known non-zero cases if we can't easily use PBLENDW.
Differential Revision: https://reviews.llvm.org/D49562
llvm-svn: 338407
This patch does the same thing as r338153 for COFF.
Note that this patch affects only the order of log messages.
The output file is already deterministic.
Differential Revision: https://reviews.llvm.org/D50023
llvm-svn: 338406
These aren't exhaustive, but cover some instructions that are only available in 32-bit mode (where would we be without good BCD math performance?).
llvm-svn: 338404
Summary:
Similar to D49636, but for PMADDUBSW. This instruction has the additional complexity that the addition of the two products saturates to 16-bits rather than wrapping around. And one operand is treated as signed and the other as unsigned.
A C example that triggers this pattern
```
static const int N = 128;
int8_t A[2*N];
uint8_t B[2*N];
int16_t C[N];
void foo() {
for (int i = 0; i != N; ++i)
C[i] = MIN(MAX((int16_t)A[2*i]*(int16_t)B[2*i] + (int16_t)A[2*i+1]*(int16_t)B[2*i+1], -32768), 32767);
}
```
Reviewers: RKSimon, spatel, zvi
Reviewed By: RKSimon, zvi
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D49829
llvm-svn: 338402
This commit fixes two issues with the liveness information after the
call:
1) The code always spills RCX and RDX if InProlog == true, which results
in an use of undefined phys reg.
2) FinalReg, JoinReg, RoundedReg, SizeReg are not added as live-ins to
the basic blocks that use them, therefore they are seen undefined.
https://llvm.org/PR38376
Differential Revision: https://reviews.llvm.org/D50020
llvm-svn: 338400
Michael Kruse