Summary:
This patch is a fix for following issue:
https://bugs.llvm.org/show_bug.cgi?id=31362 The problem was caused by front end
lowering C calling conventions without taking into account calling conventions
enforced by attribute. In this case win64cc was no correctly lowered on targets
other than Windows.
Reviewed By: rnk (Reid Kleckner)
Differential Revision: https://reviews.llvm.org/D43016
Author: belickim <mateusz.belicki@intel.com>
llvm-svn: 324594
I found this while looking at the ppc failures caused by the dso_local
change.
The issue was that the patch would produce the wrong answer for
available_externally. Having ForDefinition_t available in places where
the code can just check the linkage is a bit of a foot gun.
This patch removes the ForDefiniton_t argument in places where the
linkage is already know.
llvm-svn: 324499
When trying to track down a different bug, we discovered
that calling __builtin_va_arg on a vec3f type caused
the SROA pass to issue a warning that there was an illegal
access.
Further research showed that the vec3f type is
alloca'ed as size '12', but the _builtin_va_arg code
on x86_64 was always loading this out of registers as
{double, double}. Thus, the 2nd store into the vec3f
was storing in bytes 12-15!
This patch alters the original implementation which always
assumed {double, double} to use the actual coerced type
instead, so the LLVM-IR generated is a load/GEP/store of
a <2 x float> and a float, rather than a double and a double.
Tests were added for all combinations I could think of that
would fit in 2 FP registers, and all work exactly as expected.
Differential Revision: https://reviews.llvm.org/D42811
llvm-svn: 324098
Pass and return _Float16 as if it were an int or float for ARM, but with the
top 16 bits unspecified, similarly like we already do for __fp16.
We will implement proper half-precision function argument lowering in the ARM
backend soon, but want to use this workaround in the mean time.
Differential Revision: https://reviews.llvm.org/D42318
llvm-svn: 323185
RISCVABIInfo is implemented in terms of XLen, supporting both RV32 and RV64.
Unfortunately we need to count argument registers in the frontend in order to
determine when to emit signext and zeroext attributes. Integer scalars are
extended according to their type up to 32-bits and then sign-extended to XLen
when passed in registers, but are anyext when passed on the stack. This patch
only implements the base integer (soft float) ABIs.
For more information on the RISC-V ABI, see [the ABI
doc](https://github.com/riscv/riscv-elf-psabi-doc/blob/master/riscv-elf.md),
my [golden model](https://github.com/lowRISC/riscv-calling-conv-model), and
the [LLVM RISC-V calling convention
patch](https://reviews.llvm.org/D39898#2d1595b4) (specifically the comment
documenting frontend expectations).
Differential Revision: https://reviews.llvm.org/D40023
llvm-svn: 322494
As @rjmccall suggested in D40023, we can get rid of
ABIInfo::shouldSignExtUnsignedType (used to handle cases like the Mips calling
convention where 32-bit integers are always sign extended regardless of the
sign of the type) by adding a SignExt field to ABIArgInfo. In the common case,
this new field is set automatically by ABIArgInfo::getExtend based on the sign
of the type. For targets that want greater control, they can use
ABIArgInfo::getSignExtend or ABIArgInfo::getZeroExtend when necessary. This
change also cleans up logic in CGCall.cpp.
There is no functional change intended in this patch, and all tests pass
unchanged. As noted in D40023, Mips might want to sign-extend unsigned 32-bit
integer return types. A future patch might modify
MipsABIInfo::classifyReturnType to use MipsABIInfo::extendType.
Differential Revision: https://reviews.llvm.org/D41999
llvm-svn: 322396
Darwin uses char * for the variadic list type (va_list). We use the PPC
SVR4 ABI for PPC, which uses a structure type for the va_list. When
constructing the GEP, we would fail due to the incorrect handling for
the va_list. Correct this to use the right type.
llvm-svn: 316599
Summary:
Convert clang::LangAS to a strongly typed enum
Currently both clang AST address spaces and target specific address spaces
are represented as unsigned which can lead to subtle errors if the wrong
type is passed. It is especially confusing in the CodeGen files as it is
not possible to see what kind of address space should be passed to a
function without looking at the implementation.
I originally made this change for our LLVM fork for the CHERI architecture
where we make extensive use of address spaces to differentiate between
capabilities and pointers. When merging the upstream changes I usually
run into some test failures or runtime crashes because the wrong kind of
address space is passed to a function. By converting the LangAS enum to a
C++11 we can catch these errors at compile time. Additionally, it is now
obvious from the function signature which kind of address space it expects.
I found the following errors while writing this patch:
- ItaniumRecordLayoutBuilder::LayoutField was passing a clang AST address
space to TargetInfo::getPointer{Width,Align}()
- TypePrinter::printAttributedAfter() prints the numeric value of the
clang AST address space instead of the target address space.
However, this code is not used so I kept the current behaviour
- initializeForBlockHeader() in CGBlocks.cpp was passing
LangAS::opencl_generic to TargetInfo::getPointer{Width,Align}()
- CodeGenFunction::EmitBlockLiteral() was passing a AST address space to
TargetInfo::getPointerWidth()
- CGOpenMPRuntimeNVPTX::translateParameter() passed a target address space
to Qualifiers::addAddressSpace()
- CGOpenMPRuntimeNVPTX::getParameterAddress() was using
llvm::Type::getPointerTo() with a AST address space
- clang_getAddressSpace() returns either a LangAS or a target address
space. As this is exposed to C I have kept the current behaviour and
added a comment stating that it is probably not correct.
Other than this the patch should not cause any functional changes.
Reviewers: yaxunl, pcc, bader
Reviewed By: yaxunl, bader
Subscribers: jlebar, jholewinski, nhaehnle, Anastasia, cfe-commits
Differential Revision: https://reviews.llvm.org/D38816
llvm-svn: 315871
In OpenCL the kernel function and non-kernel function has different calling conventions.
For certain targets they have different argument ABIs. Also kernels have special function
attributes and metadata for runtime to launch them.
The blocks passed to enqueue_kernel is supposed to be executed as kernels. As such,
the block invoke function should be emitted as kernel with proper calling convention and
argument ABI.
This patch emits enqueued block as kernel. If a block is both called directly and passed
to enqueue_kernel, separate functions will be generated.
Differential Revision: https://reviews.llvm.org/D38134
llvm-svn: 315804
This change will make it possible to use -fsanitize=function on Darwin and
possibly on other platforms. It fixes an issue with the way RTTI is stored into
function prologue data.
On Darwin, addresses stored in prologue data can't require run-time fixups and
must be PC-relative. Run-time fixups are undesirable because they necessitate
writable text segments, which can lead to security issues. And absolute
addresses are undesirable because they break PIE mode.
The fix is to create a private global which points to the RTTI, and then to
encode a PC-relative reference to the global into prologue data.
Differential Revision: https://reviews.llvm.org/D37597
llvm-svn: 313096
This patch adds a flag -fclang-abi-compat that can be used to request that
Clang attempts to be ABI-compatible with some older version of itself.
This is provided on a best-effort basis; right now, this can be used to undo
the ABI change in r310401, reverting Clang to its prior C++ ABI for pass/return
by value of class types affected by that change, and to undo the ABI change in
r262688, reverting Clang to using integer registers rather than SSE registers
for passing <1 x long long> vectors. The intent is that we will maintain this
backwards compatibility path as we make ABI-breaking fixes in future.
The reversion to the old behavior for r310401 is also applied to the PS4 target
since that change is not part of its platform ABI (which is essentially to do
whatever Clang 3.2 did).
llvm-svn: 311823
This patch is intended to enable the use of basic double letter constraints used in GCC extended inline asm {Yi Y2 Yz Y0 Ym Yt}.
Supersedes D35205
llvm counterpart: D36369
Differential Revision: https://reviews.llvm.org/D36371
llvm-svn: 311643
The comment markers accepted by the assembler vary between different targets,
but '//' is always accepted, so we should use that for consistency.
Differential revision: https://reviews.llvm.org/D36666
llvm-svn: 311325
This is causing failures when compiling clang with -O3
as one of the structures used by clang is passed by
value and uses the fastcc calling convention.
Faliures manifest for stage2 mips build.
llvm-svn: 310704
This is an improvement over always using byval for
structs.
This will use registers until ~16 are used, and then
switch back to byval. This needs more work, since I'm
not sure it ever really makes sense to use byval. If
the register limit is exceeded, the arguments still
end up passed on the stack, but with a different ABI.
It also may make sense to base this on number of
registers used for non-struct arguments, rather than
just arguments that appear first in the argument list.
llvm-svn: 310527
OpenCL 2.0 atomic builtin functions have a scope argument which is ideally
represented as synchronization scope argument in LLVM atomic instructions.
Clang supports translating Clang atomic builtin functions to LLVM atomic
instructions. However it currently does not support synchronization scope
of LLVM atomic instructions. Without this, users have to use LLVM assembly
code to implement OpenCL atomic builtin functions.
This patch adds OpenCL 2.0 atomic builtin functions as Clang builtin
functions, which supports generating LLVM atomic instructions with
synchronization scope operand.
Currently only constant memory scope argument is supported. Support of
non-constant memory scope argument will be added later.
Differential Revision: https://reviews.llvm.org/D28691
llvm-svn: 310082
This option when combined with -mgpopt and -membedded-data places all
uninitialized constant variables in the read-only section.
Reviewers: atanasyan, nitesh.jain
Differential Revision: https://reviews.llvm.org/D35917
llvm-svn: 309940
The ARM Runtime ABI document (IHI0043) defines the AEABI floating point
helper functions in 4.1.2 The floating-point helper functions. These
functions always use the base PCS (soft-fp). However helper functions
defined outside of this document such as the complex-number multiply and
divide helpers are not covered by this requirement and should use
hard-float PCS if the target is hard-float as both compiler-rt and libgcc
for a hard-float sysroot implement these functions with a hard-float PCS.
All of the floating point helper functions that are explicitly soft float
are expanded in the llvm ARM backend. This change makes clang not force the
BuiltinCC to AAPCS for AAPCS_VFP. With this change the ARM compiler-rt
tests involving _Complex pass with both hard-fp and soft-fp targets.
Differential Revision: https://reviews.llvm.org/D35538
llvm-svn: 309257
This change is part of the RegCall calling convention support for LLVM.
Existing RegCall implementation was extended to include correct handling of
Complex Long Double type. Complex long double types should be returned/passed
in memory and not register stack. This patch implements this behavior.
Patch by: eandrews
Differential Revision: https://reviews.llvm.org/D35259
llvm-svn: 308769
This patch adds support for the `long_call`, `far`, and `near` attributes
for MIPS targets. The `long_call` and `far` attributes are synonyms. All
these attributes override `-mlong-calls` / `-mno-long-calls` command
line options for particular function.
Differential revision: https://reviews.llvm.org/D35479
llvm-svn: 308667
Certain targets (e.g. amdgcn) require global variable to stay in global or constant address
space. In C or C++ global variables are emitted in the default (generic) address space.
This patch introduces virtual functions TargetCodeGenInfo::getGlobalVarAddressSpace
and TargetInfo::getConstantAddressSpace to handle this in a general approach.
It only affects IR generated for amdgcn target.
Differential Revision: https://reviews.llvm.org/D33842
llvm-svn: 307470
In running some internal vectorcall tests in 32 bit mode, we discovered that the
behavior I'd previously implemented for x64 (and applied to x32) regarding the
assignment of SSE registers was incorrect. See spec here:
https://msdn.microsoft.com/en-us/library/dn375768.aspx
My previous implementation applied register argument position from the x64
version to both. This isn't correct for x86, so this removes and refactors that
section. Additionally, it corrects the integer/int-pointer assignments. Unlike
x64, x86 permits integers to be assigned independent of position.
Finally, the code for 32 bit was cleaned up a little to clarify the intent,
as well as given a descriptive comment.
Differential Revision: https://reviews.llvm.org/D34455
llvm-svn: 305928
Summary: OpenCL and SPIR version metadata must be generated once per module instead of once per mangled global value.
Reviewers: Anastasia, yaxunl
Reviewed By: Anastasia
Subscribers: ahatanak, cfe-commits
Differential Revision: https://reviews.llvm.org/D34235
llvm-svn: 305796
Rationale: OpenCL kernels are called via an explicit runtime API
with arguments set with clSetKernelArg(), not as normal sub-functions.
Return SPIR_KERNEL by default as the kernel calling convention to ensure
the fingerprint is fixed such way that each OpenCL argument gets one
matching argument in the produced kernel function argument list to enable
feasible implementation of clSetKernelArg() with aggregates etc. In case
we would use the default C calling conv here, clSetKernelArg() might
break depending on the target-specific conventions; different targets
might split structs passed as values to multiple function arguments etc.
https://reviews.llvm.org/D33639
llvm-svn: 304389
This patch adds support for the `micromips` and `nomicromips` attributes
for MIPS targets.
Differential revision: https://reviews.llvm.org/D33363
llvm-svn: 303546
Alloca always returns a pointer in alloca address space, which may
be different from the type defined by the language. For example,
in C++ the auto variables are in the default address space. Therefore
cast alloca to the expected address space when necessary.
Differential Revision: https://reviews.llvm.org/D32248
llvm-svn: 303370
Modified MipsABIInfo::classifyArgumentType so that it now coerces
aggregate structures only if the size of said aggregate is less than
16/64 bytes, depending on the ABI.
Patch by Stefan Maksimovic.
Differential Revision: https://reviews.llvm.org/D32900
with minor changes (use regexp instead of the hardcoded values) to the test.
llvm-svn: 302670
Use variadic templates instead of relying on <cstdarg> + sentinel.
This enforces better type checking and makes code more readable.
Differential revision: https://reviews.llvm.org/D32550
llvm-svn: 302572
Reverting
Modified MipsABIInfo::classifyArgumentType so that it now coerces
aggregate structures only if the size of said aggregate is less than 16/64
bytes, depending on the ABI.
as it broke clang-with-lto-ubuntu builder.
llvm-svn: 302555
Modified MipsABIInfo::classifyArgumentType so that it now coerces aggregate
structures only if the size of said aggregate is less than 16/64 bytes,
depending on the ABI.
Patch by Stefan Maksimovic.
Differential Revision: https://reviews.llvm.org/D32900
llvm-svn: 302547
It turns out there are some sort-of-but-not-quite empty structs that break all
the rules. For example:
struct SuperEmpty { int arr[0]; };
struct SortOfEmpty { struct SuperEmpty e; };
Both of these have sizeof == 0, even in C++ mode, for GCC compatibility. The
first one also doesn't occupy a register when passed by value in GNU C++ mode,
unlike everything else.
On Darwin, we want to ignore the lot (and especially don't want to try to use
an i0 as we were).
llvm-svn: 302313
These two attributes specify the same info in a different way.
AMGPU BE only checks the latter as a target specific attribute
as opposed to language specific reqd_work_group_size.
This change produces amdgpu_flat_work_group_size out of
reqd_work_group_size if specified.
Differential Revision: https://reviews.llvm.org/D31728
llvm-svn: 299678
