__attribute__((amdgpu_flat_work_group_size(<min>, <max>))) - request minimum and maximum flat work group size
__attribute__((amdgpu_waves_per_eu(<min>[, <max>]))) - request minimum and/or maximum waves per execution unit
Differential Revision: https://reviews.llvm.org/D24513
llvm-svn: 282371
Move the definition of `getTriple()` into the header. It would just call
`getTarget().getTriple()`. Inline the definition to allow the compiler to see
the same amount of the layout as previously. Remove the more verbose
`getTarget().getTriple()` in favour of `getTriple()`.
llvm-svn: 281487
The PPC64 DWARF register-size table did not match the ABI specification (or
GCC, for that matter). Fix that, and add a regression test.
Fixes PR27931.
llvm-svn: 280053
Structs are currently handled as pointer + byval, which makes AMDGPU
LLVM backend generate incorrect code when structs are used. This patch
changes struct argument to be handled directly and without flattening,
which Clover (Mesa 3D Gallium OpenCL state tracker) will be able to
handle. Flattening would expand the struct to individual elements and
pass each as a separate argument, which Clover can not
handle. Furthermore, such expansion does not fit the OpenCL
programming model which requires to explicitely specify each argument
index, size and memory location.
Patch by Vedran Miletić
llvm-svn: 279463
We processed unnamed bitfields after our logic for non-vector field
elements in records larger than 128 bits. The vector logic would
determine that the bit-field disqualifies the record from occupying a
register despite the unnamed bit-field not participating in the record
size nor its alignment.
N.B. This behavior matches GCC and ICC.
llvm-svn: 278656
An __m512 vector type wrapped in a structure should be passed in a
vector register.
Our prior implementation was based on a draft version of the psABI.
This fixes PR28975.
N.B. The update to the ABI was made here:
https://github.com/hjl-tools/x86-psABI/commit/30f9c9
llvm-svn: 278655
Summary:
Based on a patch by Michael Mueller.
This attribute specifies that a function can be hooked or patched. This
mechanism was originally devised by Microsoft for hotpatching their
binaries (which they're constantly updating to stay ahead of crackers,
script kiddies, and other ne'er-do-wells on the Internet), but it's now
commonly abused by Windows programs that want to hook API functions. It
is for this reason that this attribute was added to GCC--hence the name,
`ms_hook_prologue`.
Depends on D19908.
Reviewers: rnk, aaron.ballman
Subscribers: cfe-commits
Differential Revision: https://reviews.llvm.org/D19909
llvm-svn: 278050
The size of image type is reported incorrectly as size of a pointer to address space 0, which causes error when casting image type to pointers by __builtin_astype.
The fix is to get image address space from TargetInfo then report the size accordingly.
Differential Revision: https://reviews.llvm.org/D22927
llvm-svn: 277647
Summary:
In RenderScript, the size of the argument or return value emitted in the
IR is expected to be the same as the size of corresponding qualified
type. For ARM and AArch64, the coercion performed by Clang can
change the parameter or return value to a type whose size is different
(usually larger) than the original aggregate type. Specifically, this
can happen in the following cases:
- Aggregate parameters of size <= 64 bytes and return values smaller
than 4 bytes on ARM
- Aggregate parameters and return values smaller than bytes on
AArch64
This patch coerces the cases above to an integer array that is the same
size and alignment as the original aggregate. A new field is added to
TargetInfo to detect a RenderScript target and limit this coercion just
to that case.
Tests added to test/CodeGen/renderscript.c
Reviewers: rsmith
Subscribers: aemerson, srhines, llvm-commits
Differential Revision: https://reviews.llvm.org/D22822
llvm-svn: 276904
Allows AMDGCN target to generate images (such as %opencl.image2d_t) in constant address space.
Images will still be generated in global address space by default.
Added tests to existing opencl-types.cl in test\CodeGenOpenCL.
Patch by Aaron En Ye Shi.
Differential Revision: https://reviews.llvm.org/D22523
llvm-svn: 276161
Added the opencl.ocl.version metadata to be emitted with amdgcn. Created a static function emitOCLVerMD which is shared between triple spir and target amdgcn.
Also added new testcases to existing test file, spir_version.cl inside test/CodeGenOpenCL.
Patch by Aaron En Ye Shi.
Differential Revision: https://reviews.llvm.org/D22424
llvm-svn: 276010
Summary:
Summary:
Change Clang calling convention SpirKernel to OpenCLKernel.
Set calling convention OpenCLKernel for amdgcn as well.
Add virtual method .getOpenCLKernelCallingConv() to TargetCodeGenInfo
and use it to set target calling convention for AMDGPU and SPIR.
Update tests.
Reviewers: rsmith, tstellarAMD, Anastasia, yaxunl
Subscribers: kzhuravl, cfe-commits
Differential Revision: http://reviews.llvm.org/D21367
llvm-svn: 274220
smaller than register as argument in variadic functions on
big endian architectures.
Differential Revision: http://reviews.llvm.org/D21611
llvm-svn: 273665
We would incorrectly emit the directive sections due to the missing overridden
methods. We now emit the expected "/DEFAULTLIB" rather than "-l" options for
requested linkage
llvm-svn: 273558
Summary:
Clang does not detect `aapcs-vfp` for the EABIHF environment. The reason is that only GNUEABIHF is considered while choosing calling convention, EABIHF is ignored.
This causes clang to use `aapcs` for EABIHF and add the `arm_aapcscc` specifier to functions in generated IR.
The modified `arm-cc.c` test checks that no calling convention specifier is added to functions for EABIHF, which means the default one is used (`CallingConv::ARM_AAPCS_VFP`).
Reviewers: rengolin, compnerd, t.p.northover
Subscribers: aemerson, rengolin, asl, cfe-commits
Differential Revision: http://reviews.llvm.org/D20219
llvm-svn: 269419
Use a utility function to check whether the number of elements is a power of 2
and drop the redundant upper limit (a 128-bit vector with more than 16 elements
would have each element < 8 bits, not possible).
llvm-svn: 268422
Before this change, we would pass all non-HFA record arguments on
Windows with byval. Byval often blocks optimizations and results in bad
code generation. Windows now uses the existing workaround that other
x86_32 platforms use.
I also expanded the workaround to handle C++ records with constructors
on Windows. On non-Windows platforms, we have to keep generating the
same LLVM IR prototypes if we want our bitcode to be ABI compatible.
Otherwise we will encounter mismatch issues like PR21573.
Essentially fixes PR27522 in Clang instead of LLVM.
Reviewers: hans
Differential Revision: http://reviews.llvm.org/D19756
llvm-svn: 268261
Summary:
Port rL265324 to SystemZ to allow using the 'swiftcall' attribute on that architecture.
Depends on D19414.
Reviewers: kbarton, rjmccall, uweigand
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D19432
llvm-svn: 267879
SPIR target can be used for C/C++ inputs too (i.e. in OpenCL compatible mode for the libs creation).
Patch by Neil Henning!
Review: http://reviews.llvm.org/D19478
llvm-svn: 267561
Currently, for the ppc64--gnu and aarch64 ABIs, we recognize:
typedef __attribute__((__ext_vector_type__(3))) float v3f32;
typedef __attribute__((__ext_vector_type__(16))) char v16i8;
struct HFA {
v3f32 a;
v16i8 b;
};
as an HFA. Since the first type encountered is used as the base type,
we pass the HFA as:
[2 x <3 x float>]
Which leads to incorrect IR (relying on padding values) when the
second field is used.
Instead, explicitly widen the vector (after size rounding) in
isHomogeneousAggregate.
Differential Revision: http://reviews.llvm.org/D18998
llvm-svn: 266784
Non-owning pointers that cache LLVM types and constants can use
'nullptr' default member initializers so that we don't need to mention
them in the constructor initializer list.
Owning pointers should use std::unique_ptr so that we don't need to
manually delete them in the destructor. They also don't need to be
mentioned in the constructor at that point.
NFC
llvm-svn: 266263
Revert the two changes to thread CodeGenOptions into the TargetInfo allocation
and to fix the layering violation by moving CodeGenOptions into Basic.
Code Generation is arguably not particularly "basic". This addresses Richard's
post-commit review comments. This change purely does the mechanical revert and
will be followed up with an alternate approach to thread the desired information
into TargetInfo.
llvm-svn: 265806
This is a mechanical move of CodeGenOptions from libFrontend to libBasic. This
fixes the layering violation introduced earlier by threading CodeGenOptions into
TargetInfo. It should also fix the modules based self-hosting builds. NFC.
llvm-svn: 265702
Summary:
r246764 handled __fp16 arguments and returns for AAPCS, but skipped this
handling for OpenCL. Simlar to OpenCL, RenderScript also handles __fp16
type natively.
This patch adds the -fnative-half-arguments-and-returns command line
flag to allow such languages to skip this coercion of __fp16.
Reviewers: srhines, olista01
Subscribers: cfe-commits
Differential Revision: http://reviews.llvm.org/D18138
llvm-svn: 263795
For compatibility with GCC, classify __m64 as SSE.
However, clang is a platform compiler for certain targets; retain our
old behavior on those targets: classify __m64 as integer.
This fixes PR26832.
llvm-svn: 262688
Fixes PR11517 for SPARC.
On most targets, clang lowers va_arg itself, eschewing the use of the
llvm vaarg instruction. This is necessary (at least for now) as the type
argument to the vaarg instruction cannot represent all the ABI
information that is needed to support complex calling conventions.
However, on targets with a simpler varrags ABIs, the LLVM instruction
can work just fine, and clang can simply lower to it. Unfortunately,
even on such targets, vaarg with a struct argument would fail, because
the default lowering to vaarg was naive: it didn't take into account the
ABI attribute computed by classifyArgumentType. In particular, for the
DefaultABIInfo, structs are supposed to be passed indirectly and so
llvm's vaarg instruction should be emitted with a pointer argument.
Now, vaarg instruction emission is able to use computed ABIArgInfo for
the provided argument type, which allows the default ABI support to work
for structs too.
I haven't touched the EmitVAArg implementation for PPC32_SVR4 or XCore,
although I believe both are now redundant, and could be switched over to
use the default implementation as well.
Differential Revision: http://reviews.llvm.org/D16154
llvm-svn: 261717
This uses the general emitVoidPtrVAArg lowering logic for everything, since
this supports all types, and we don't have any special requirements.
llvm-svn: 261557
This modification applies the following Android commit when we have an
Android environment. This is the sole non-renderscript in the Android repo
commit 9212d4fb30a3ca2f4ee966dd2748c35573d9682c
Author: Tim Murray <timmurray@google.com>
Date: Fri Aug 15 16:00:15 2014 -0700
Update vector calling convention for AArch64.
bug 16846318
Change-Id: I3cfd167758b4bd634d8480ee6ba6bb55d61f82a7
Reviewers: srhines, jyknight
Subscribers: mcrosier, aemerson, rengolin, tberghammer, danalbert, srhines
Differential Revision: http://reviews.llvm.org/D17448
llvm-svn: 261533
It can happen that when we only have 1 more register left in the regsave
area we need to store a value bigger than 1 register and therefore we
go to the overflow area. In this case we have to leave the last slot
in the regsave area unused and keep using overflow area. Do this
by storing a limit value to the used register counter in the overflow block.
Issue diagnosed by and solution tested by Mark Millard!
llvm-svn: 261422
Konstantin Zhuravlyov